Current Technology Developments of WiMax Systems

Current Technology Developments of WiMax Systems Maode Ma Editor Current Technology Developments of WiMax Systems 1...

Author: Maode Ma

48 downloads 1058 Views 11MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Current Technology Developments of WiMax Systems

Maode Ma Editor

Current Technology Developments of WiMax Systems

13

Editor Dr. Maode Ma Nanyang Technological University School of Electrical & Electronic Engineering 50 Nanyang Avenue Singapore 639798 Singapore

ISBN: 978-1-4020-9299-2

e-ISBN: 978-1-4020-9300-5

DOI 10.1007/978-1-4020-9300-5 Library of Congress Control Number: 2008936718 Springer Science+Business Media B.V. 2009 No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.

c

Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

Preface

Recent developments on wireless communication technology have resulted in tremendous innovations to make broadband wireless networks able to compete with 3G cellular network. IEEE 802.16X standards have not only specified WiMax wireless access networks but also designed a framework of wireless metropolitan area networks with mobility functionality. It is obvious that with further development of various WiMax technologies, wide range of high-quality, flexible wireless mobile applications and services could be provided, which will revolutionarily improve our modern life to achieve the goal of accessing the global information at any place and at any time by any mobile devices in the near future. WiMax technology is the most promising global telecommunication technology recently. WiMax technology and various WiMax networks are specified by the IEEE 802.16X standards, which define the Medium Access Control (MAC) layer and the Physical (PHY) layer of fixed and mobile broadband wireless access systems. This edited book has been produced by many contributors who have much knowledge of the standards and rich teaching and/or research experience of the WiMax technology. This edited book is intended to be a comprehensive reference book to address the recent developments of WiMax technologies for both academia and industry. It can serve as an introduction book for beginners to get the fundamental knowledge of various aspects of WiMax systems. It is also expected to be a good reference for researchers and engineers to understand the recent developments of the technology in order to promote further development of WiMax technologies and systems. The book consists of 13 chapters. Each of the chapters is either a technical overview or literature survey on a particular topic or a proposed solution to a research issue of Wimax technology. The 13 chapters can be roughly classified into 3 parts. The first part is major on the fundamental issues in WiMax point-to-multipoint (PMP) topology, consisting of Chapter 1 to Chapter 7. Chapter 1 addresses the deployment of multi-antenna base stations in WiMax systems and corresponding design of signal processing algorithms for interference mitigation involved in the deployment. Two main solutions have been proposed for interference management at the physical layer of WiMax systems. Chapter 2 presents a general background introduction on various dynamic bandwidth allocation schemes and introduces a Two-Phase Proportionating (TPP) algorithm as a solution to achieve a feasible dyv

vi

Preface

namic bandwidth allocation in order to increase the utilization of precious bandwidth and provide service differentiation. Chapter 3 is to address the frame allocation issue for the Quality of Service (QoS) provisioning, particularly based on the Orthogonal Frequency Division Multiple Access (OFDMA) PHY layer technology. A modular framework to handle frame allocation problems has been proposed to decouple the constraints of data region allocation into the MAC frames. Chapter 4 provides a detailed overview of the state-of-the-art scheduling techniques employed in WiMax systems and various issues related to the implementation of some of the scheduling algorithms. Chapter 5 presents an introduction and literature review on the QoS provisioning architecture and various technologies of traffic management at different levels including call admission control (CAC) and traffic scheduling to provide QoS guarantee in WiMax systems. In Chapter 6, a load-balancing approach to handle radio resource management in the mobile WiMax networks has been presented. This approach is a set of algorithms including call admission control, adaptive transmission, horizontal handover, and dynamic bandwidth allocation algorithms, which jointly maximize the network capacity and guarantee QoS requirements from different types of applications. Chapter 7 is a comparison study on the random access technologies in various wireless systems including the third generation (3G) of cellular systems like Wideband Code Division Multiple Access (WCDMA) and CDMA2000, mobile WiMax, and 3G Long Term Evolution (LTE). The second part is major on the mobility issues in WiMax cellular networks, consisting of Chapter 8 and Chapter 9. Chapter 8 is a proposed improvement on a handover mechanism, which is the Fast Base Station Switching (FBSS), used in mobile WiMax systems. The proposed FBSS with Reuse Partitioning Cell Structure scheme is to enhance the performance of the traditional FBSS. Chapter 9 is a literature review on various handover schemes recently proposed to improve the standard one specified in the IEEE 802.16e. It has presented the recent research efforts to reduce the latency introduced in the handover process with aim to provide QoS to different types of traffic during the handovers. The third part is major on other topologies of WiMax networks and the integration of WiMax networks with other wireless/wired networks, consisting of Chapter 10 to Chapter 13. Chapter 10 is a comprehensive study on WiMax systems with a proposal of a multiservice CAC mechanism, which can significantly improve the performance of an existing CAC scheme, and a proposal on the architecture of an interworking wireless network composed of WiMax and Wireless Local Area Network (WLAN) systems. Chapter 11 examines the power consumption performance in WiMax relay networks specified by IEEE 802.16j. Chapter 12 studies the issue to provide large-scale reliable multicast and broadcast services by social psychology principles and game theory. Finally, Chapter 13 addresses physical layer technologies such as Multiple Input Multiple Output (MIMO) antennas and Adaptive Modulation and Coding (AMC), the operations of WiMax mesh networks, and the integration of wireless and wired/optical MANs. It is obvious that without the great contributions and profound, excellent knowledge of WiMax technologies from the authors of each chapter, this book could not be published to serve as a reference book to the world. I wish to thank each contributor

Preface

vii

of the book for his/her time, huge efforts, and great enthusiasm to the publication of the book. I would also thank the publisher of the book and the representatives, Mr. Mark de Jongh, Mrs. Cindy Zitter, and Ms. Deivanai Loganathan, Integra for their patience and great helps in the publication process. Singapore

Dr. Maode Ma

Contents

1 Deployment and Design of Multi-Antenna WiMax Systems in a Non-Stationary Interference Environment . . . . . . . . . . . . . . . . . . . . . . . . M. Nicoli, S. Savazzi, O. Simeone, R. Bosisio, G. Primolevo, L. Sampietro and C. Santacesaria

1

2 Dynamic Bandwidth Allocation for 802.16E-2005 MAC . . . . . . . . . . . . . 17 Yi-Neng Lin, Shih-Hsin Chien, Ying-Dar Lin, Yuan-Cheng Lai and Mingshou Liu 3 A Downlink MAC Frame Allocation Framework in IEEE 802.16e OFDMA: Design and Performance Evaluation . . . . . . . . . . . . . . . . . . . . 31 Andrea Bacioccola, Claudio Cicconetti, Alessandro Erta, Luciano Lenzini, Enzo Mingozzi and Jani Moilanen 4 Scheduling Techniques for WiMax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Aymen Belghith and Loutfi Nuaymi 5 QoS Provision Mechanisms in WiMax . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Maode Ma and Jinchang Lu 6 Mobile WiMax Performance Optimization . . . . . . . . . . . . . . . . . . . . . . . . 115 Stanislav Filin, Sergey Moiseev and Mikhail Kondakov 7 A Comparative Study on Random Access Technologies of 3G and B3G Mobile Communications Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Jungchae Shin and Ho-Shin Cho 8 An Improved Fast Base Station Switching for IEEE 802.16e with Reuse Partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 I-Kang Fu, Hsiang-Jung Chiu and Wern-Ho Sheen ix

x

Contents

9 Fast Handover Schemes in IEEE 802.16E Broadband Wireless Access System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Qi Lu, Maode Ma and Hui Ming Liew 10 Addressing Multiservice Classes and Hybrid Architecture in WiMax Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 Kamal Gakhar, Mounir Achir, Alain Leroy and Annie Gravey 11 Energy-Efficient Multimedia Delivery in WMAN Using User Cooperation Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Ki-Dong Lee, Byung K. Yi and Victor C.M. Leung 12 Game Theory Modeling of Social Psychology Principle for Reliable Multicast Services in WiMax Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 Markos P. Anastasopoulos, Athanasios V. Vasilakos and Panayotis G. Cottis 13 IEEE 802.16: Enhanced Modes of Operation and Integration with Wired MANs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Isabella Cerutti, Luca Valcarenghi, Piero Castoldi, Dania Marabissi, Filippo Meucci, Laura Pierucci, Enrico Del Re, Luca Simone Ronga, Ramzi Tka and Farouk Kamoun

Biography of the Editor

Dr. Maode Ma received his BE degree in Computer Engineering from Tsinghua University in 1982, ME degree in Computer Engineering from Tianjin University in 1991 and PhD degree in Computer Science from Hong Kong University of Science and Technology in 1999. He is an Associate Professor in the School of Electrical and Electronic Engineering at Nanyang Technological University in Singapore. He has extensive research interests including wireless networking, optical networking, and so forth. He has been a member of the technical program committee for more than 80 international conferences. He has been a technical track chair, tutorial chair, publication chair, and session chair for more than 40 international conferences. Dr. Ma has published more than 130 international academic research papers on wireless networks and optical networks. He currently serves as an Associate Editor for IEEE Communications Letters, an Editor for IEEE Communications Surveys and Tutorials, an Associate Editor for International Journal of Wireless Communications and Mobile Computing, an Associate Editor for International Journal of Security and Communication Networks, and an Associate Editor for International Journal of Vehicular Technology.

xi

Contributors

M. Achir TELECOM Bretagne, France M. P. Anastasopoulos Wireless & Satellite Communications Group, School of Electrical & Computer Engineering, National Technical University of Athens, Greece A. Bacioccola Dipartimento di Ingegneria dell’Informazione, University of Pisa, Italy A. Belghith TELECOM Bretagne R. Bosisio Dip. di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32, 20133 Milano, Italy P. Castoldi Scuola Superiore Sant’Anna, Pisa, Italy, [email protected] I. Cerutti Scuola Superiore Sant’Anna, Pisa, Italy, [email protected] S.-H. Chien National Chiao Tung University, University Road, Hsinchu, Taiwan H.-J. Chiu Department of Communication Engineering, National Chiao Tung University, Hsinchu, Taiwan H.-S. Cho School of Electrical Engineering and Computer Science, Kyungpook National University, Daegu, Korea C. Cicconetti Dipartimento di Ingegneria dell’Informazione, University of Pisa, Italy P. G. Cottis Wireless & Satellite Communications Group, School of Electrical & Computer Engineering, National Technical University of Athens, Greece E. Del Re Università degli Studi di Firenze, Italy, [email protected] xiii

xiv

Contributors

A. Erta IMT Lucca Institute for Advanced Studies, Lucca, Italy S. Filin National Institute of Information and Communications Technology, 3-4, Hikarino-oka, Yokosuka, 239-0847, Japan I.-K. Fu Department of Communication Engineering, National Chiao Tung University, Hsinchu, Taiwan K. Gakhar 68 Rue Gallieni; 92100 Boulogne Billancourt, France, [email protected] A. Gravey Department of Computer Science, TELECOM Bretagne, France ´ F. Kamoun Ecole Nationale des Sciences de l’Informatique, Manouba, Tunisia, [email protected] M. Kondakov Kodofon, 97, Moskovsky Prospekt, Voronezh, Russia Y.-C. Lai National Taipei University of Science and Technology, Taipei, Taiwan K.-D. Lee LD Electronics Mobile Research, San Diego, CA 92131, USA, [email protected] L. Lenzini Dipartimento di Ingegneria dell’Informazione, University of Pisa, Italy A. Leroy TELECOM Bretagne, France V. C. M. Leung LD Electronics Mobile Research, San Diego, CA 92131, USA H. M. Liew School of Electrical and Electronic Engineering, Nanyang Technological Unversity, Singapore Y.-N. Lin National Chiao Tung University, University Road, Hsinchu, Taiwan Y.-D. Lin National Chiao Tung University, University Road, Hsinchu, Taiwan M. Liu Intel Innovation Center, Taiwan J. Lu School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore Q. Lu School of Electrical and Electronic Engineering, Nanyang Technological Unversity, Singapore M. Ma School of Electrical and Electronic Engineering, Nanyang Technological Unversity, Singapore

Contributors

xv

D. Marabissi Università degli Studi di Firenze, Italy, [email protected] F. Meucci Università degli Studi di Firenze, Italy, [email protected] E. Mingozzi Dipartimento di Ingegneria dell’Informazione, University of Pisa, Italy, [email protected] J. Moilanen Nokia Siemens Networks, Helsinki, Finland S. Moiseev Kodofon, 97, Moskovsky Prospekt, Voronezh, Russia M. Nicoli Dip. di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32, 20133 Milano, Italy, [email protected] L. Nuaymi TELECOM Bretagne L. Pierucci Università degli Studi di Firenze, Italy, [email protected] G. Primolevo WISYTech, Via Cadore, 21, 20035 Lissone, Milano Italy L. S. Ronga CNIT, Firenze, Italy, [email protected] L. Sampietro Nokia Siemens Networks S.p.A. Com CRD MW, S.S. 11 Padana Superiore Km. 158, 20060 Cassina de’ Pecchi (Milano), Italy C. Santacesaria Nokia Siemens Networks S.p.A. Com CRD MW, S.S. 11 Padana Superiore Km. 158, 20060 Cassina de’ Pecchi (Milano), Italy S. Savazzi Dip. di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32, 20133 Milano, Italy W.-H. Sheen Department of Communication Engineering, National Chiao Tung University, Hsinchu, Taiwan J. Shin School of Electrical Engineering and Computer Science, Kyungpook National University, Daegu, Korea O. Simeone CCSPR, New Jersey Institute of Technology (NJIT), University Heights 07102, Newark, USA U. Spagnolini Dip. di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32, 20133 Milano, Italy

xvi

Contributors

´ R. Tka Ecole Nationale des Sciences de l’Informatique, Manouba, Tunisia, [email protected] L. Valcarenghi Scuola Superiore Sant’Anna, Pisa, Italy, [email protected] A. V. Vasilakos Wireless & Satellite Communications Group, School of Electrical & Computer Engineering, National Technical University of Athens, Greece B. K. Yi LG Electronics Mobile Research, San Diego, CA 92131, USA

Chapter 1

Deployment and Design of Multi-Antenna WiMax Systems in a Non-Stationary Interference Environment M. Nicoli, S. Savazzi, O. Simeone, R. Bosisio, G. Primolevo, L. Sampietro and C. Santacesaria

Abstract WiMax has already been adopted worldwide by operators attracted by promises of large throughput and coverage for broadband wireless access. However, towards the goal of an efficient deployment of the technology, a thorough analysis of its performance in presence of frequency reuse under realistic traffic conditions is mandatory. In particular, in both fixed and mobile WiMax applications, an important performance limiting factor is inter-cell interference, which has strong time-varying and non-stationary features. Two main solutions have been proposed for interference management at the physical layer of WiMax systems, namely, multi-antenna technology and random subcarrier permutation (as in the latest version of the standard, IEEE 802.16-2005). In this chapter, system deployment of multi-antenna base stations, and related design of signal processing algorithms for interference mitigation, are discussed. Extensive numerical results for realistic interference models show the advantages of the optimized multi-antenna deployment and design in combination with subcarrier permutation.

1.1 Introduction WiMax (Worldwide Interoperability for Microwave Access) is a standard-based technology that provides last mile broadband wireless access. Operators worldwide have already embraced this solution as either a complement or an alternative to existing wired and wireless technologies, such as cable, Digital Subscriber Line (DSL) or second/third generation (2G/3G) cellular systems [1]. Applications range from the provision of wireless services for rural or developing areas, to intensive and real-time applications on notebooks and other mobile devices. A first version of the standard, IEEE 802.16-2004 [2], was designed to provide broadband access M. Nicoli (B) Dip. di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32, 20133 Milano, Italy e-mail: [email protected]

M. Ma (ed.), Current Technology Developments of WiMax Systems, C Springer Science+Business Media B.V. 2009 DOI 10.1007/978-1-4020-9300-5 1,

1

2

M. Nicoli et al.

to fixed subscriber stations, while the recently approved IEEE 802.16-2005 [3] supports both fixed and mobile access. Among the various options proposed in the IEEE 802.16 physical (PHY) layer specifications, the Orthogonal Frequency Division Multiplexing mode with 256 subcarriers in [2] and the Scalable OFDMA mode in [3] (here referred to as, respectively, 802.16-OFDM-256 and 802.16-SOFDMA) have sparkled the most interest as access solutions for the deployment of a cellular Wireless Metropolitan Area Network (WMAN). While analysis of single WiMax links for fixed applications is by now well studied (see, e.g., the survey in [1]), the impact of a deployment of WiMax (either fixed or mobile) over a given geographical area is currently under investigation. WiMax access points distributed over the coverage area form a cellular structure with given frequency reuse factor (see Fig. 1.1(a) for square cells with reuse factor

(a)

(c) Δθ =

33 7

SINR [dB]

4

45

deg 3.5

40

SS1

SS2

BS1

Linear array

Δ e [λ]

2.5

SS3 BS3

BS0

Δi

35

BS2 SS0

Δe

3

30

2 1.5

25

1

20

1000 m

Δe

0.5

15

(d)

(b)

SINR [dB]

4

50

620 m 3.5

45

Δθ =4 SS2

Δ e [λ]

deg

SS0

40

6.1

SS1 BS0

3

Δe

2.5 35 2 30 1.5

Δi

25

Δe

1

SS3 Linear array

20

0.5

No-SD Max-FD 0.5

1

1.5

2

2.5

3

3.5

4

15

Δ i [λ]

Fig. 1.1 Deployment of a non-uniform linear antenna array at the base station of a WiMax multicell system. Uplink cellular layouts are considered with either square (a) or hexagonal (b) cells. The average SINR at the output of the MVDR filter is shown versus the antenna spacings ⌬e and ⌬i , for the square (c) and hexagonal (d) layouts

1

Deployment and Design of Multi-Antenna

3

4, and Fig. 1.1(b) for hexagonal cells with reuse factor 3). In such a scenario, the main technological challenge is the mitigation of inter-cell interference. This issue is even more relevant in WiMax than in existing 2G/3G cellular systems due to the larger data rates that WiMax promises to offer, which call for more sophisticated interference-reduction techniques. The task of designing a WiMax system robust to inter-cell interference is made even more challenging by the non-stationary nature of interference in both fixed and mobile applications. In a fixed system, based on IEEE 802.16-OFDM-256, asynchronous transmission in interfering cells causes the inter-cell interference to vary within a given communication session. In fact, in a typical scenario, out-of-cell subscriber stations (SS) are expected to start and end their transmission on a time scale that cannot be controlled by the interfered cells. On the other hand, mobile WiMax based on IEEE 802.16-SOFDMA adopts random subcarrier permutation within a time-frequency grid (as in the Partially Used Subchannelization, PUSC, mode [3, 4]) in order to provide interference diversity. After the permutation, subcarriers allocated to a given user are subject to the interference of different out-of-cell subscribers, thus leading to a non-stationary interference scenario. This chapter presents an overview of existing solutions to the problem of (nonstationary) interference-mitigation techniques at the PHY layer of a WiMax-compliant system for both IEEE 802.16-OFDM-256 and SOFDMA. As recognized at academic and industrial levels, a satisfactory interference management hinges on multi-antenna technology (see, e.g., [5, 6]). Therefore, in this chapter we focus on deployment and design of WiMax systems in presence of multi-antenna at the base station (BS). As a case study, we investigate the uplink (UL), i.e. the communication from SSs to BS, as illustrated in Fig. 1.1(a, b). We first study optimal deployment of an antenna array at WiMax base stations as a trade-off between diversity and interference-rejection capability of the antenna array (Section 1.3). The advantages of an optimized array in terms of coverage and average throughput with respect to conventional antenna array deployments are shown through extensive numerical simulations in Section 1.4. Then, signal processing techniques at the BS that allow to cope efficiently with non-stationary interference are addressed. These techniques are based on the exploitation of the pilot subcarriers prescribed by the IEEE 802.16-OFDM-256 standard for adaptive estimation of channel/interference parameters (Section 5.1). Finally, subcarrier permutation as defined by the PUSC mode of IEEE 802.16-SOFDMA is considered, in combination with multi-antenna technology, in order to cope with non-stationary interference. Performance of such a solution is discussed to validate its suitability (Section 5.2).

1.2 WiMax System In this section we present a brief overview of the PHY layer specified by the standards IEEE 802.16-OFDM-256 and IEEE 802.16-SOFDMA. We focus on the UL side of the radio link, where complex receiver algorithms at the BS side can give

4

M. Nicoli et al.

the most relevant gains in cell coverage and link quality. Moreover, we introduce the basic model of a multi-cell WiMax system, that will be used for performance assessment.

1.2.1 Overview of IEEE 802.16 PHY Layer Both the OFDM-256 [2] and the SOFDMA [3] modes of IEEE 802.16 are targeted to bandwidths in the low 2-11 GHz range (for performance evaluation purposes, in the remainder of the chapter we will consider 3.5 GHz as the carrier frequency) and employ OFDM modulation for their basic symbol structure. Apart from these fundamental similarities, the two PHY layers exhibit profound differences in both the chosen multiple access scheme and the system parameters, and require a separate introduction. A summary of the basic system parameters of IEEE 802.16 is given in Table 1.1. Table 1.1 Basic parameter for IEEE 802.16-OFDM-256 and IEEE 802.16-SOFDMA Overall subcarriers Guard subcarriers

IEEE 802.16 OFDM-256

IEEE 802.16 SOFDMA

256 (irrespective of BW)

128

512

1024

2048

43

91

183

367

55

Active subcarriers

192 data+ 8 pilots

N. of slots in the BW

–

Mandatory coding scheme

Concatenated RS-CC

48 data + 24 pilots (6 tiles × 3 symbols) 3

15

30

60

Tail Biting CC (also CTC)

1.2.1.1 IEEE 802.16-OFDM-256 The OFDM mode has been tailored for a deployment with fixed users, as a lastmile access solution. Users are accommodated on the UL frame by a Time Division Multiple Access (TDMA) scheme: different users transmit in physically separate bursts, each including a long preamble (the first OFDM symbol) used for channel estimation and synchronization, followed by a sequence of OFDM symbols carrying coded data. Eight pilot subcarriers are also embedded in each OFDM data symbol (see Table 1.1). Since there is at most one active interferer per adjacent cell on a given OFDM symbol, and the interferers can be reasonably expected to transmit continuously for a number of OFDM symbols, interference rejection in the OFDM mode is a feasible task. However, bursts are not necessarily synchronized among different cells and thus the transmission from the desired user might experience different interferers switching on and off. Specific techniques are therefore needed so as to gather data about all interferer’s spatial signature within each transmitted burst, track it for the burst duration and reject its transmission (if needed). To this end, the receiver can exploit both the burst preamble and the pilots embedded in the subsequent OFDM data symbols (see Section 1.5.1).

1


5

1.2.1.2 IEEE 802.16-SOFDMA In the SOFDMA mode, a hybrid Time/Frequency Division Multiple Access (TDMA/ FDMA) scheme is utilized. The logical structure of the UL frame can be visualized as a time/frequency grid, where each chunk, dubbed slot in the standard, represents a set of subcarriers observed over a certain number of OFDM symbols. All user allocations are then defined as contiguous sets of slots in the frame. The standard provides various options for grouping the subcarriers into slots. The basic option, and the most likely to be utilized in the first generation devices, is the UL-PUSC [4]. In this case, each slot spans 24 subcarriers, observed over 3 OFDM symbols (see Table I). However, the subcarriers assigned to a slot are not contiguous in frequency, as each slot selects 6 groups of 4 contiguous subcarriers by means of a pseudo-random permutation. When observed over 3 OFDM symbols each group is called tile (spanning 12 subcarriers, of which 8 for data and 4 pilots). The permutation depends on both a cell-specific identifier and the time index of the slot in the frame. This results in a pseudo-random spreading of the signal over the frequency domain that improves the system diversity in the following way: 1 The transmission of any user is spread over the available bandwidth, so that subcarriers exhibiting bad channel conditions impact only on a portion of the transmitted data (frequency diversity); 2 The transmission of a strong interferer affects only part of the signals transmitted by a given user. In fact, the interferer’s tiles are likely assigned to different slots and thus to different terminals in the cell of the considered user. Furthermore, since data is coded, the corrupted portion of the received signal might be recovered. Random permutation can thus yield a significant performance improvement with respect to fixed channel assignments where a strong interference affects the entire codeword (interference diversity). The effects of random subcarrier permutation will be investigated in Section V-B.

1.2.2 System Model 1.2.2.1 Multi-cell Layout We consider the UL of a IEEE 802.16 compliant system [2, 3]. Figure 1.1 exemplifies the scenario of interest for, respectively: (a) square layout with cell side r = 1 km and frequency reuse factor F = 4; (b) hexagonal layout with cell side r = 620 m and frequency reuse factor F = 3. In these examples, the transmission by the subscriber station SS0 to its own base station BS0 at a given time instant and frequency is impaired by the interference from at most N I = 3 out-of-cell subscriber NI (see the shaded cells in the Fig. 1.1(a,b) representing the first ring stations {SSi }i=1 of interferers). Base station BS0 is equipped with a linear symmetric array of M antennas, while SS’s have a single antenna.

6

M. Nicoli et al.

1.2.2.2 Modeling the Radio Environment Let us consider the antenna-array receiver at the base station BS0 . The (base-band) signal received on the t-th subcarrier on a given OFDM symbol can be written as yt = ht xt + nt ,

(1.1)

where xt denotes the symbol transmitted by the desired station SS0 , while ht is a vector gathering the M (complex) channel gains between the transmitter SS0 and the M antennas of BS0 . These gains account for path-loss, shadowing and fast-fading effects due to the propagation from SS0 to BS0 [7]. Notice that long-term fading effects due to shadowing can be, to a certain extent, mitigated through power control. Generally, the propagation channel ht is the superposition of the contributions of several paths, each characterized by a complex amplitude, a time of arrival and a direction of arrival (DOA). In our performance analysis, the multipath components are modelled according to the Stanford University Interim (SUI) channel models [8], and DOAs are considered as Gaussian distributed around the main direction SS0 BS0 , with a moderate angular spread. The baseline case of signal coming from a single direction that might be different from the line-of-sight (LOS) one (i.e., with null angular spread) will also be considered and referred to as a no-spatial-diversity (No-SD) channel. It is perfectly understood that this particular case has limited applicability either in the fixed and in the mobile wireless environment, although it might model, in some cases, propagation scenarios where the BS0 and the SS0 are marginally surrounded by scattering. As further performance references, we consider two simplified fading models that deviate from the SUI one and can be seen as extreme cases of frequency selectivity:

r r

No frequency diversity (No-FD): the channel gains are constant over the subcarriers (as for a null delay spread or, equivalently, a frequency-flat channel); Maximum frequency diversity (Max-FD): the channel gains are uncorrelated over the subcarriers (as for the ideal case of a maximum delay spread).

The focus of this chapter is on the effect of the noise vector nt , that is given by the sum of the background noise and inter-cell interference. The latter is generated by the set It of users, {SSi }i∈It , that are active in the nearby cells on the same subcarrier and the same OFDM symbol as the desired transmission. Propagation from interferers to BS0 is modelled similarly to the user SS0 . The main difference is that shadowing effects on the interfering channels cannot be compensated by power control and lead to fluctuations of the interference level up to 20-30dB. 1.2.2.3 Characterization of the Inter-cell Interference According to the discussion above, inter-cell interference is characterized at the base station BS0 by the multipath channels corresponding to the propagation from the users {SSi }i∈It to BS0 . This information is summarized by the spatial covariance of the interference-plus-noise signal nt , defined as Qt = E[nt nH t ], that collects the

1


7

noise correlation between any pair of antennas [6]. This quantity is fundamental when treating the interference as Gaussian; it depends not only on the propagation environment but also on the inter-element spacing used at the antenna array (see Section 1.3). Accurate estimation and tracking algorithms for the covariance matrix Qt are necessary tools in order to design interference mitigation algorithms at the BS. As explained above, while this task is feasible in IEEE 802.16-OFDM-256 systems (see Section 1.5.1), this is highly impractical in the UL-PUSC mode of the SOFDMA standard (see Section 1.5.2).

1.3 Antenna Array Design In this section, we tackle the problem of optimal antenna deployment at the base station BS0 . In particular, as firstly proposed in [9], we investigate the optimal antenna spacings for a non-uniform linear antenna array. Herein, we focus on IEEE 802.16-OFDM-256 systems with interference scenarios sketched in Fig. 1.1(a) and 1.1(b). According to the standard specifications, only one user is active within each cell in the bandwidth of interest. Thereby, in both cases up to three interferers impair the transmission from SS0 to BS0 . To reduce the effects of this interference, BS0 applies a spatial filter (beamforming) to the received signal (1.1). The optimal beamforming technique is the minimum variance distortionless (MVDR) filter [6], which minimizes the output power subject to the constraint of unitary gain in the steering direction SS0 -BS0 . This leads to effective interference-rejection capabilities, as nulls are steered in directions of strong interferers. The interference-rejection capability may be quantified in terms of signal-tointerference-plus-noise ratio (SINR) at the output of the spatial filter. This depends on the channel response (ht ), the spatial pattern of the interference (Qt ) and the antenna spacing. The first two quantities are determined by the cellular layout geometry, the SS positions and the propagation environment, while the antenna spacings are free parameters that can be designed for a specific layout/environment so as to maximize the SINR performance. 3 placed at the center of their Let us at first consider the three interferers {SSi }i=1 respective cells and focus on the simplified propagation model with no shadowing, path-loss simulated according to the Hata-Okamura model [7] (with path-loss exponent 4), maximum frequency-diversity (Max-FD) and null angular spread (No-SD). As shown in Fig. 1.1, a non-uniform antenna array with M = 4 elements is considered at the BS: the array structure consists of two clusters of two antenna each that are positioned at a distance ⌬i among each other, antennas in each cluster are ⌬e spaced. In Fig. 1.1(c) and 1.1(d) the average SINR at the output of the MVDR filter, for M = 4 antennas, is plotted in gray scale versus the external (⌬e ) and internal (⌬i ) spacings of the antenna array, for the two plannings in Fig. 1.1(a) and 1.1(b), respectively. For each pair (⌬e , ⌬i ), the SINR value is obtained by averaging over fading, noise and the position of the desired user SS0 (uniformly distributed within the cell). The results obtained for both the layouts show that the minimum-length array that

8

M. Nicoli et al.

maximizes the average SINR is a uniform linear array (ULA) with ⌬i = ⌬e = ⌬opt , being ⌬opt = 1.8λ the optimal spacing for the layout (a) and ⌬opt = 1.4λ for the layout (b) (λ denotes the carrier wavelength). This confirms the analytical results of [10], where the optimal spacing is found to be ⌬opt = nλ/ sin(⌬θ ) where n is a non-zero integer and ⌬θ the angular separation between interferers (n = 1 for minimum-length array). We recall that the inter-element spacing normally used for beamforming purposes is the one that maximizes the DOA resolution under the non-alias constraint, i.e., ⌬m = λ/[2 sin(θmax )] where θmax is the largest DOA admissible for the considered cellular layout. In particular, this equals the usual spacing ⌬m = λ/2 when the antenna array covers the whole sector of 180deg (θmax = π/2), while for the plannings in Fig. 1.1 it is: θmax = π/4 and ⌬m = 0.71λ for the square layout; (b) θmax = π/3 and ⌬m = 0.58λ for the hexagonal layout. With respect to this standard antenna deployment with spacing ⌬m , the optimal antenna array is wider (⌬opt > ⌬m ) and provides a larger SINR at the output. From Fig. 1.1(c,d), the SINR gain with respect to the first solution is around 5dB. Such a gain can be justified by noticing that the optimal spacing introduces a certain degree of angular equivocation in the directivity function of the array, so that the three interferers with DOAs [θ1 , θ2 , θ3 ] = [−⌬θ, 0, +⌬θ ] are grouped together along the unique direction θ = 0. The spatial wave numbers associated to the DOAs of the interferers SS1 and ⌬ SS3 are indeed ω1 = ω3 = ±2π λopt sin(⌬θ ) = ± 2πn and coincide with that of the broadside interferer SS2 (ω2 = 0). This effect renders interference mitigation more effective, as one null of the directivity function on the broadside is enough to virtually reject three interferers, thus leaving other degrees of freedom to increase the spatial diversity. When the position of the interferers is not known a-priori at the time of the antenna deployment, or it varies due to terminal mobility as prescribed in IEEE 802.16-SOFDMA [3], these concepts have to be adapted to randomness of SS’s positions. In such scenarios, the SINR has to be averaged over the expected positions of both user and interferers within their cells before being optimized. Interestingly, simulation results show that the optimal spacing remains essentially the same as in the static scenario considered above (for n = 1), due to the symmetry of the problem at hand. The performance gain with respect to standard antenna deployment reduces to 2dB (the reader may refer to [11] for a numerical validation).

1.4 Coverage Analysis In this section we compare the average cell throughput provided by the optimized ULA with inter-element spacing ⌬opt and MVDR processing, as derived in the previous section, with that obtained by a conventional ULA with spacing ⌬m . The performance gain of the optimized array is evaluated for an IEEE 802.16-OFDM256 system with bandwidth 4MHz. As prescribed in [2], seven possible transmis7 , can be used, with throughput ranging from 1.2 Mbit/s (T1 ) to sion modes, {Ti }i=1

1


9

11.9 Mbits/s (T7 ) for the selected bandwidth. According to the adaptive modulation and coding approach, the transmission mode is selected based on specific channel measurements so as to guarantee a fixed bit error rate. The average throughput for each position of the user SS0 in the cell is obtained for fixed interferers placed at the center of their respective cells (as indicated in Fig. 1.1(a,b)), as follows. For each position of SS0 , the average BER (averaged with respect to the channel, the noise and the interference) at the output of the decoder 7 . The best transmission mode is then is evaluated for all transmission modes {Ti }i=1 selected as the one that satisfies the constraint BER≤ 10−6 (e.g., to model applications that have stringent reliability requirements) and provides the largest bit-rate. This allows to obtain a coverage map, detailed for all transmission modes, as those exemplified for the square layout in Fig. 1.2 and for the hexagonal layout in Fig. 1.3, for the channel model Max-FD with null angular spread (No-SD). In these examples the number of receiving antennas M ranges from 1 (left figures) to 4 (right figures), while the inter-element spacing is the conventional one (⌬m ) used for beamforming (top) or the optimized one (⌬opt ) for throughput maximization (bottom). Once the coverage maps have been obtained, the average throughput R¯ b [bit/s] for the overall cell can be evaluated through a weighted average of the throughputs associated to the different transmission modes, using as weighting factors the (normalized) areas where the modes are supported. The results are summarized in Fig. 1.4 for the square layout and in Fig. 1.5 for the hexagonal layout. The number of antennas ranges between M = 1 and 4, both the case of omnidirectional (on the top) and directional antennas (on the bottom) at the BS are considered; performances are evaluated for the two antenna spacings ⌬opt and ⌬m . Each column refers to a different channel model: no frequency diversity and null angular spread (No-FD No-SD, column (1); SUI-3 with null angular spread (SUI-3 No-SD, column (2); M=2

M=3

M=4

No-SD Max-FD

Δ = Δ m = 0.7λ

M=1

5.94 Mbps

4.61 Mbps

6.80 Mbps

7.64 Mbps

T6 T5 T4

λ

Δ = Δ opt = 1.8

T7

Δθ

SS1

T3

SS 2

4.61 Mbps

SS 3

T2

6.70 Mbps

8.05 Mbps

8.90 Mbps T1

7 Fig. 1.2 Coverage for all transmission modes {Ti }i=1 represented in gray-scale for the square planning shown on the right. The arrows indicate the directions of arrival of the interferers. The BS antenna array has a number of elements ranging from M = 1 (left) to M = 4 (right), and inter-element spacing optimized for beamforming ⌬m (top) or for coverage ⌬opt (bottom)

10

M. Nicoli et al. M=2

M=3

No-SD Max-FD

M=4

Δ = Δ m = 0.6λ

M=1

4.47 Mbps

5.56 Mbps

7.55 Mbps

6.70 Mbps

T7

Δ = Δ opt = 1.4 λ

T6 SS SS

T5

Δθ

T4

SS

4.47 Mbps

T3

6.60 Mbps

T2

9.06 Mbps

8.00 Mbps

T1

7 Fig. 1.3 Coverage for all transmission modes {Ti }i=1 represented in gray-scale for the hexagonal planning shown on the right. The arrows indicate the directions of arrival of the interferers. The BS antenna array has a number of elements ranging from M = 1 (left) to M = 4 (right), and inter-element spacing optimized for beamforming ⌬m (top) or for coverage ⌬opt (bottom)

SUI-4 with null angular spread (SUI-4 No-SD, column (3); maximum frequency diversity with null angular spread (Max-FD No-SD, column (4); no frequency diversity with angular spread 5deg (No-FD SD, column (5). For instance, the results obtained in Fig. 1.2 (and in Fig. 1.3) can be easily found in the top section of the tables in Fig. 1.4 (and in Fig. 1.5) in the columns referring to No-FD No-SD channel. As expected, from the exploitation of diversity either in space or in frequency, the overall system throughput may be considerably enhanced. In addition, for the square cell planning, an optimized array is shown to provide substantial throughput improvements with respect to the conventional beamforming-oriented array. On the

l ne an as Ch nn te An

No-FD No-SD

SUI 3 No-SD

SUI 4 No-SD

Δ opt

Max-FD No-SD

No-FD SD

Δm

Δ opt

Δm

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

M=1

0.19

0.19

0.96

0.96

1.15

1.15

4.61

4.61

0.19

0.19

M=2

0.32

0.42

1.50

1.98

1.80

2.36

5.94

6.70

1.14

1.81

M=3

0.39

0.61

1.89

2.93

2.26

3.41

6.80

8.05

1.94

3.75

M=4

0.53

0.77

2.53

3.47

2.99

4.06

7.64

8.90

2.93

5.27

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

M=1

0.27

0.27

1.31

1.31

1.57

1.57

5.50

5.50

0.27

0.27

M=2

0.39

0.58

1.92

2.71

2.29

3.18

6.87

7.81

1.50

2.45

M=3

0.51

1.09

2.47

4.50

2.96

5.21

7.67

10.12

2.55

4.85

M=4

0.70

1.49

3.25

5.52

3.81

6.27

8.67

10.63

3.76

6.51

Fig. 1.4 Average throughput [Mbit/s] for the UL of the square cellular planning shown on the right, with an antenna array of M elements at the BS and a single antenna at each SS. We consider both omnidirectional antennas (on top) and directional antennas (on the bottom) cases

1

Deployment and Design of Multi-Antenna Ch

l ne an as nn te An

No-FD No-SD

SUI 3 No-SD

11

SUI 4 No-SD

Max-FD No-SD

No-FD SD

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

M=1

0.21

0.21

0.95

0.95

1.13

1.13

4.47

4.47

0.20

0.20

M=2

0.29

0.39

1.36

1.88

1.61

2.23

5.56

6.60

0.96

1.60

M=3

0.39

0.61

1.86

2.87

2.23

3.37

6.70

8.00

1.60

3.14

M=4

0.52

0.82

2.45

3.63

2.91

4.22

7.55

9.06

2.37

4.54

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

Δm

Δ opt

M=1

0.30

0.20

1.38

1.38

1.64

1.64

5.63

5.63

0.29

0.29

M=2

1.16

1.57

4.34

5.53

5.00

6.28

9.74

10.66

2.23

2.56

M=3

2.01

2.18

6.60

6.98

7.31

7.68

11.08

11.22

4.26

5.14

M=4

2.53

2.63

7.73

7.78

8.42

8.49

11.36

11.40

5.94

7.07

Fig. 1.5 Average throughput [Mbit/s] for the UL of the hexagonal cellular planning shown on the right, with an antenna array of M elements at the BS and a single antenna at each SS. We consider both omnidirectional antennas (on top) and directional antennas (on the bottom) cases

other hand, for the hexagonal cell planning, due to the reduced frequency reuse factor that increases the co-channel average interference level, a substantially lower throughput improvement is observed. In this case, directional antennas can be used for optimizing performances.

1.5 Impact of Non-Stationary Inter-Cell Interference In this section we analyze the impact of non-stationary inter-cell interference on both fixed [2] and mobile [3] WiMax. As discussed in Section 1.2, in IEEE 802.16-OFDM-256 systems [2], estimation and tracking of the interferers’ spatial properties along with the use of an optimized array for beamforming is a viable solution to mitigate time-varying interference. On the other hand, in IEEE 802.16SOFDMA systems [3], the fast variability of the interference and the limited number of pilot symbols in each tile makes the previous approach unfeasible; as a consequence, a robust exploitation of the interference diversity is mandatory at the receiver.

1.5.1 Time-Varying Interference Mitigation in IEEE 802.16-OFDM-256 The design of an effective beamforming, as discussed in Section III-IV, requires an estimate of the channel gains ht (defined by the multipath channel of user SS0 ) and an estimate of the interference covariance matrix Qt (which defines the power and spatial features of interferers). Such channel/interference parameters have to be evaluated for each subcarrier and OFDM symbol (i.e., for each index value t).

12

M. Nicoli et al.

The estimation of the channel/interference parameters may be obtained from pilots through the least-squares (LS) method, then followed by interpolation to extend the estimate over the whole time-frequency bandwidth [12]. Here, we consider a IEEE 802.16-OFDM-256 fixed access scenario, where the channel coherence time is large enough to make the channel gains invariant over the whole frame interval. The whole bandwidth is assigned to one user at a time, thus the set of interferers It and the corresponding covariance matrix Qt do not vary over the frequency domain. However, due to the asynchronous access of users in neighboring cells, even in this fixed scenario the set It of active interferers can vary over the time (within the burst interval), generating abrupt changes in the signal interfering on user SS0 and thus in its covariance matrix Qt . In [13] a method was proposed to estimate the channel and to track the power/ spatial features of the interference by exploiting both the preambles and the pilots included in each OFDM symbol of the frame. At first, the signals measured in several preambles of the frame are jointly processed to obtain an estimate of the channel ht (which is constant for the whole frame) and a first estimate of the interference covariance matrix Qt in each preamble. The channel ht is evaluated through a weighted average of the LS estimates obtained separately from the different preambles: the average accounts for the stationarity of the channel, while the weighting accounts for the possible variations of the interference scenario. The interference covariance matrix Qt is then updated within each burst, by using the embedded pilots. Abrupt variations of the interference are detected by comparing the covariance estimate obtained from the current pilots with the one extracted from the previous OFDM symbol, in order to decide whether the spatial structure of the interference has changed or not: if the correlation ρ between the two subsequent covariance estimates is larger than a given threshold ρ, ¯ the interference covariance estimate is refined by averaging, otherwise is re-initialized according to the new estimate value. An example is shown in Fig. 1.6 for the square cellular layout in Fig. 1.1(a). The optimized ULA with M = 4 elements and inter-element spacing ⌬opt is adopted by BS0 . The receiver consists of MVDR filtering, soft demodulation and convolutional/Reed-Solomon (CC/RS) decoding. The user SS0 transmits with power 27dBm and transmission mode T2 . Interferers {SSi }i∈It are uniformly distributed in their cells; their power (subject to log-normal shadowing with standard deviation 8dB) and transmission mode are adaptively selected based on the channel state so as to guarantee a BER≤10−3 . Multipath channels are modelled according to the SUI-3 model, DOAs of both user and interferers are drawn from a Gaussian distribution with standard deviation 5 deg. We consider the transmission of 3 bursts of 10 symbols each, with the user SS0 placed in broadside at a distance d = 0.8 km from BS0 . The interference scenario changes at the third and seventh symbol of each burst, with positions of the three interferers selected uniformly within their cell. Figure 1.6 shows the BER (top) and the interference correlation (bottom) over the OFDM symbols. The estimation of the interference matrix Qt is obtained using three different approaches: estimation only from the preamble of the current burst (thick line); re-estimation within each OFDM symbol without memory (dashed line); tracking in


Fig. 1.6 Performance of IEEE 802.16-OFDM-256 in presence of non-stationary inter-cell interferece: BER (top) and interferencecovariance-matrix correlation ρ (bottom) as a function of the time index over the frame. The positions of preambles are indicated by vertical thick line, while vertical thin lines denote changes (in power and DOA) of interference scenario

13

10–1 Interference estimation: from preamble only from current pilots (no memory) with detection/tracking

BER

1

10–2

10–3

5

10

15 20 Symbol index

25

30

5

10

15 20 Symbol index

25

30

Correlation ρ

1

ρ

0.5

each OFDM symbol with interference-change detection (thin line). The BER results confirm that the proposed tracking method with interference-change detection is an effective approach for time-varying interference mitigation.

1.5.2 Non-Stationary Interference in IEEE 802.16-SOFDMA As shown in Section 1.2, the IEEE 802.16-SOFDMA standard adopts a random subcarrier permutation that artificially induces fluctuations on the interference scenario along the transmission of the reference user. This non-stationarity enhances the robustness of the system against strong interferers, as the interfered slots are spread over different SSs in the cell (interference diversity). On the contrary, such a behavior of the interference becomes difficult to predict. The problem is particularly relevant in multi-antenna systems, where the non-stationarity of the interference and the limited number of pilots per tile makes the estimation/tracking of the interference spatial covariance (and thus beamforming for interference-mitigation) unfeasible. A more suited multi-antenna solution is obtained by largely spaced antennas so as to get signal uncorrelation and thus spatial diversity. Since the interference is spatially uncorrelated (i.e., uncorrelated over the antennas), the optimal MVDR receiver in this case reduces to a maximum ratio combiner (MRC) [6].

14

M. Nicoli et al.

A first analysis of the impact of inter-cell interference on the throughput of IEEE 802.16-SOFDMA has been derived in [14]. BER performance evaluation after decoding in UL-PUSC systems [3] shows that the performance heavily depends on the amount of interference-state information available at the decoder [15]. More specifically, optimum decoding can be performed when the detector knows the instantaneous interference power (i.e., the interference power in each tile), while a conventional receiver approximates the interference as stationary along the coded data packet with remarkable performance degradation. This means that the interference diversity can be fully exploited only when some knowledge of the interference statistics is available at the BS. A practical solution [15] consists in a joint interference-power estimation and decoding. Differently from spatial covariance, the estimation of a single scalar parameter (the interference power) is feasible even from a limited set of pilots. For performance assessment, we assume that the user SS0 requests two slots for the transmission of a data-packet of 192 bits, by using the rate-1/2 convolutional (CC) code for QPSK modulation [3]. BS0 is equipped with M = 2 antennas and the simplified channel model No-FD is employed (the reader might refer to [16] for the frequency selective case). We assume a slow fading scenario so that the channel can be assumed as stationary over three consecutive OFDM symbols. We consider N I = 3 interfering cells as in Fig. 1.1(b), with interferers’ power subject to lognormal shadowing fluctuations. The system load υ (fraction of slots allocated out of the whole number of available slots) is assumed to be the same for all the cells. Figure 1.7(a) shows the BER versus the signal-to-interference-ratio (SIR) for system load υ = {0.9; 0.5}. The figure compares the simulation results (markers) and the analytical analysis (dashed/dotted lines) obtained by truncating to the first term the union bound approximation for the bit error probability [15]. The optimum receiver offers a considerable performance gain with respect to conventional scheme for moderate load (about 2 dB when υ = 0.5). The motivation is that for limited

(b)

(a) 10–2

100

SIR = 15db

10–1

BER

υ = 0.5

υ = 0.9

10–3 10

–4

10–5 –5

BER

10–3

10–2

10–4

Conventional receiver Practical receiver Optimum receiver Analytic Conventional Analytic Optimum

0

5

10

SIR [dB]

SIR = 20db

Conventional receiver Practical receiver Optimum receiver

15

20

25

10–5

1

0.9

0.8

0.7

0.6

0.5

0.4

υ

Fig. 1.7 Analytic (dashed/dotted lines) and simulated (markers) BER at the output of the CC decoder in IEEE 802.16-SOFDMA: transmission mode with code rate r = 1/2 and QPSK modulation, N I = 3 inter-cell interferers, M = 2 receiving antennas, SNR=30dB. (a) BER versus SIR for load υ ∈ {0.5, 0.9}; (b) BER vs. load υ for SIR∈ {15, 20}dB

1


15

load (υ → 0) the number of interfering users sharing the same subcarrier abruptly changes from tile to tile, thus making the interference to heavily fluctuate along the coded packet. On the other hand, for large load (υ → 1) the number of collisions with the interferers is almost constant (N I ) and the influence of the interference non-stationarity is mitigated. As a consequence, the conventional receiver is mainly effective in systems with large load. The results are corroborated by Fig. 1.7(b) that shows the BER versus the system load υ for SIR= {15; 20}dB.

1.6 Conclusions This chapter has focused on technological solutions at the PHY layer for management of inter-cell interference in WiMax-compliant systems. The study has assumed components and features as defined in the standards IEEE 802.16-OFDM-256 [2] and SOFDMA [3]. The main conclusion is that an appropriate system design (deployment and signal processing) allows to harness relevant performance gains in terms of transmission quality-of-service. From an evolutionary perspective, further enhancements in the interference rejection capabilities of the PHY layer could be achieved by: (1) introducing multi-cell cooperation: decoding at different cell-sites is performed jointly by capitalizing on the existing high-capacity backbone: (2) cross-layer optimization of PHY layer and higher layers functionalities, such as scheduling. Acknowledgments The authors would like to acknowledge the former students D. Archetti, A. Bonfanti, M. Sala and A. Villarosa for their contribution to the development of the simulator for IEEE 802.16 systems. This work was supported by Nokia Siemens Networks S.p.A. Com CRD MW, Cassina de’ Pecchi, Italy and by MIUR-FIRB Integrated System for Emergency (InSyEme) project under the grant RBIP063BPH.

References 1. A. Ghosh, D. R. Wolter, J. G. Andrews, R. Chen, “Broadband wireless access with WiMax/802.16: current performance benchmarks and future potential,” IEEE Commun. Mag., Vol. 43, No. 2, pp. 129–136, Feb. 2005. 2. IEEE Std 802.16TM -2004, “802.16TM IEEE standard for local and metropolitan area networks Part 16: Air interface for fixed broadband wireless access systems,” Oct. 2004. 3. IEEE Std 802.16eTM -2005 and IEEE Std 802.16TM -2004/Cor 1-2005 5, “IEEE standard for local and metropolitan area networks Part 16: Air interface for fixed and mobile broadband wireless access systems. Amendment 2: Physical and medium access control layers for combined fixed and mobile operation in licensed bands and Corrigendum 1,” Sep. 2005. 4. H. Yaghoobi, “Scalable OFDMA physical layer in IEEE 802.16 WirelessMAN”, Intel Tech. J., Vol. 8, No. 3, pp. 201–212, Aug. 2004. 5. A. Salvekar, S. Sandhu, Q. Li, M. Vuong, X. Qian, “Multiple-antenna technology in WiMax Systems,” Intel Technology Journal, Vol. 8, No. 3, pp. 229–240, Aug. 2004. 6. H. L. Van Trees, Optimum Array Processing, Wiley, 2002.

16

M. Nicoli et al.

7. A. Goldsmith, Wireless Communications, Cambridge University Press, 2005. 8. IEEE 802.16.3c-01/53, IEEE 802.16 Broadband Wireless Access Working Group, “Simulating the SUI channel models,” April 2004. 9. R. Jana and S. Dey, “3G wireless capacity optimization for widely spaced antenna arrays,” IEEE Pers. Commun., Vol. 7, No. 6, pp. 32–35, Dec. 2000. 10. S. Savazzi, O. Simeone, and U. Spagnolini, “Optimal design of linear arrays in a TDMA cellular system with Gaussian interference,” EURASIP Journ. on Wireless Comm. and Networking, Smart Antennas for Next Generation Wireless Systems, 2006. 11. M. Nicoli, L. Sampietro, C. Santacesaria, S. Savazzi, O. Simeone, U. Spagnolini, “Throughput optimization for non-uniform linear antenna arrays in multicell WiMax systems,” Int’l. ITGIEEE Workshop on Smart Antennas, March 2006. 12. Y. Li, L. J. Cimini and N. R. Sollenberger, “Robust channel estimation for OFDM systems with rapid dispersive fading,” IEEE Trans. Commun., Vol. 46, No. 7, pp. 902–915, July 1998. 13. M. Nicoli, M. Sala, L. Sampietro, C. Santacesaria, O. Simeone, “Adaptive array processing for time-varying interference mitigation in IEEE 802.16,” Proc. IEEE Int’l. Symp. on Pers. Indoor and Mobile Radio Commun. (PIMRC’06), Helsinki, Sep. 2006. 14. S-E. Elayoubi, B. Fourestiè, and X. Auffret, “On the capacity of OFDMA 802.16 systems,” Proc. IEEE Int’l. Conf. on Commun. (ICC’06), June 2006. 15. R. Bosisio and U. Spagnolini, “Collision model for the bit error rate analysis of multicell multiantenna OFDMA systems,” Proc. IEEE Int’l. Conf. on Commun. (ICC’07), June 2007. 16. D. Molteni, M. Nicoli, R. Bosisio, L. Sampietro, “Performance analysis of multiantenna WiMax systems over frequency selective fading channels,” Proc. IEEE Int’l. Symp. on Pers. Indoor and Mobile Radio Commun. (PIMRC’07), Athens, Sep. 2007.

Chapter 2

Dynamic Bandwidth Allocation for 802.16E-2005 MAC Yi-Neng Lin, Shih-Hsin Chien, Ying-Dar Lin, Yuan-Cheng Lai and Mingshou Liu

Abstract The IEEE 802.16e-2005 is designed to support high bandwidth for the wireless metropolitan area network. However, the link quality is likely to degrade drastically due to the unstable wireless links, bringing ordeals to the real-time applications. Therefore, a feasible bandwidth allocation algorithm is required to utilize the precious bandwidth and to provide service differentiation. This article presents the general background of allocation schemes and introduces a Two-Phase Proportionating (TPP) algorithm to tackle the above challenges. The first phase dynamically determines the subframe sizes while the second phase further differentiates service classes and prevents from bandwidth waste. Performance comparison with other algorithms confirms that TPP achieves the highest bandwidth utilization and the most appropriate differentiation. Keywords WiMax · Dynamic · Bandwidth Allocation · Proportion · Service Differentiation

2.1 Introduction General broadband technologies have been used to provide multimedia applications with stable connectivity. However, for a growing volume of hand-held devices running these applications, those technologies are unable to meet the requirements such as ubiquitous access, low deployment cost, and mobility support. Broadband wireless access (BWA), standardized as 802.16e-20051 [1] and known as WiMax, has emerged to be a potential candidate to meet these criteria. The standard defines signaling mechanisms [2] between base stations (BSs) and subscriber stations (SSs) considering both fixed and mobile wireless broadband. It supports not only seamless handover at vehicle speeds but also an extra service class compared to the previous version, 802.16-2004 [3]. Y.-N. Lin (B) National Chiao Tung University, University Road, Hsinchu, Taiwan 1

In the following contexts we use 802.16 to represent 802.16e-2005.


17

18

Y.-N. Lin et al.

However, the nature of wireless communication makes it difficult to provide stable signal quality, and could lead to much degraded bandwidth. For example, signal gradually fades as the transmission distance stretches, and channels are usually interfered with each other. Furthermore, though 802.16 defines service classes for differentiation, no mechanism is specified to fulfill the QoS guarantees. Therefore, a feasible algorithm is required to utilize and fairly allocate the bandwidth considering the following issues. First, the Grant Per SS (GPSS) scheme specified in the standard needs to be adhered to. In this scheme, the BS grants requested bandwidth to each SS rather than to each connection, so that the SS can flexibly respond to different QoS requirements of the connections. Second, in order to make the best use of the link, the separation between uplink and downlink subframes and the number of physical-layer slots needed given a certain amount of requested bytes, have to be carefully determined. Similar situations to design allocation algorithm in 802.16 can be seen in systems such as Wi-Fi (Wireless Fidelity) [4] and DOCSIS (Data over Cable System Interface Specifications) [5] because of the similar point-to-multipoint architectures. However, Wi-Fi adopts arbitrary contention for transmission opportunities in any time and is thus not appropriate in the WiMax environment having lengthy roundtrip delay. Also little can be referenced from works regarding the DOCSIS since it follows the Grant Per Connection (GPC) scheme [6] which is not flexible for SSs to be adaptive to connections of real-time applications and is not supported by the standard. Several works [7–10] investigating allocation algorithms over 802.16 are proposed, but again only the GPC scheme is supported. The solution researched by [11] is based on GPSS, but the separation of the uplink and downlink channels is fixed so that bandwidth is usually not properly utilized. In this article, a novel bandwidth allocation algorithm, Two-Phase Proportionating (TPP), is introduced to maximize the bandwidth utilization as well as to meet the QoS requirements under the Time Division Duplexing (TDD) mode. TDD, compared to the Frequency Division Duplexing (FDD), is frequently favored because of the flexibility to divide a time frame into adequate uplink and downlink subframes so that bandwidth waste could be minimized. Employing the concept of proportionate allocation, the algorithm dynamically adjusts the uplink and downlink subframes considering different slot definitions, and fairly allocates each subframe to queues of different classes. Simulation results further validate the efficiency of bandwidth utilization and service differentiation. The rest of this article is organized as follows. We brief the IEEE 802.16 MAC and review the related works to justify our problems. Then we introduce the TPP algorithm and exemplify the operations, followed by the simulation setup and results. Some conclusive remarks are given finally, outlining some future directions.

2.2 Background Unlike Wi-Fi which is used for small range communications, WiMax is mainly applied to metropolitan area networks and therefore must master all data transmission decisions to/from SSs to avoid synchronization problems. In this section, we

2

Dynamic Bandwidth Allocation

19

brief the WiMax frame structure under TDD mode, describe the five service classes whose connections fill up the frame, and detail the packet flow in the BS MAC. The bandwidth allocation module as well as its input and output is identified according to the flow. Some related researches investigating the allocation problem are discussed.

2.2.1 Overview of the MAC Protocol 2.2.1.1 TDD Subframe The frame structure under TDD includes (1) UL-MAP and DL-MAP for control messages, and (2) downlink and uplink data bursts whose scheduled time is determined by the bandwidth allocation algorithm and is indicated in the MAP messages. All UL-MAP/DL-MAP and data bursts are composed of a number of OFDMA (Orthogonal Frequency Division Multiplexing Access) slots, in which a slot is one subchannel by three OFDMA symbols in uplink and one subchannel by two OFDMA symbols in downlink. This mode is named PUSC (Partial Usage of Subchannels), the mandatory mode in 802.16, and is considered throughout the work. 2.2.1.2 Uplink Scheduling Classes The 802.16 currently supports five uplink scheduling classes, namely the Unsolicited Grant Service (UGS), Real-time Polling Service (rtPS), Non-real-time polling Service (nrtPS), Best Effort (BE), and the lately proposed Extended Realtime Polling Service (ertPS). Each service class defines different data handling mechanisms to carry out service differentiation. The UGS has the highest priority and reserves a fixed amount of slots at each interval for bandwidth guarantee. rtPS, nrtPS, and BE rely on the periodic polling to gain transmission opportunities from BS, while the ertPS reserves a fixed number of slots as UGS does and notifies the BS in the contention period of possible reservation changes. nrtPS and BE also contend, according to their pre-configured priority, for transmission opportunities if they fail to get enough bandwidth from polling. An nrtPS service flow is always superior to that of BE. 2.2.1.3 Detailed Packet Flow in the MAC Layer The complete packet flow in the uplink and downlink of a BS MAC is illustrated as follows. For the downlink processing flow, both IP and ATM packets in the network layer are transformed from/to the MAC Convergence Sublayer (CS) by en/de-capsulating the MAC headers. According to the addresses and ports, packets are classified to the corresponding connection ID of a service flow which further determines the QoS parameters. Fragmentation and packing are then performed to form a basic MAC Protocol Data Unit (PDU), whose size frequently adapts to the channel quality, followed by the allocation of resulting PDUs into queues. Once the allocation starts, the bandwidth management unit arranges the data burst

20

Y.-N. Lin et al.

transmissions to fill up the frame. The MAP builder then writes the arrangement, namely the allocation results, into the MAP messages to notify the PHY interface when to send/receive the scheduled data in the time frame. Encryption, header checksum and frame CRC calculations are carried out to the PDUs before they are finally sent to the PHY. The uplink processing flow is similar to that of the downlink except that the BS also receives standalone or piggybacked bandwidth requests. Among the above operations, it is obvious that the bandwidth management, and thus the bandwidth allocation algorithm, are critical and need to be carefully designed in order to improve the performance of the system.

2.2.2 Related Work A number of studies regarding the bandwidth allocation over 802.16 can be found. Hawa and Petr [7] propose a QoS architecture applicable for both DOCSIS and 802.16 using semi-preemptive priority for scheduling UGS traffic while priorityenhanced Weighted Fair Queuing (WFQ) for others. Chu et al. [8] employ the Multi-class Priority Fair Queuing (MPFQ) for the SS scheduler and the Weighted Round Robin (WRR) for that of the BS. Though innovative in the architectural design, both of them do not present experiment results validating the architecture. Wongthavarawat and Ganz [9] introduce the Uplink Packet Scheduling (UPS) for service differentiation. It applies the Strict Priority to the selection among service classes, in which the UL and DL have same capacity, and each service class adopts a certain scheduling algorithm for queues within it. However, this scheme deals with only uplink channel so that overall bandwidth utilization suffers. The Deficient Fair Priority Queue (DFPQ) [10], which uses the maximum sustained rate as the deficit counter to specify the transmission quantum, dynamically adjusts the uplink and downlink proportion. Nonetheless, this method is suitable only for GPC rather than GPSS. Maheshwari et al. [11] support GPSS using proportion, though the proportion is not alterable in run-time. Furthermore, the above schemes do not consider the slot definition when translating data bytes requested by SSs into OFDMA slots to practically determine the allocation of a time frame.

2.2.3 Goals To solve the allocation problem which could lead to long latency and serious jittering, a well-designed bandwidth allocation algorithm shall possess three merits. First and obviously, the algorithm must implement GPSS to comply with the standard as well as to provide flexible packet scheduling in SSs. Second, service classes should adhere to the corresponding QoS requirements such as Maximum Sustained Traffic Rate (MSTR) and Minimum Reserved Traffic Rate (MRTR) for differentiated guarantees. The former prevents a certain class from consuming too much bandwidth while the latter sustains a service class with least feeds. Third, in order to achieve

2


21

high throughput, the proportion of the uplink and downlink subframes should be able to be dynamically adjusted. The separator was previously fixed and failed to adapt to situations in which uplink and downlink bandwidth needs vary.

2.3 Two-Phase Proportionating This section details the concept and procedure of the proposed Two-Phase Proportionating (TPP) algorithm. Each phase manipulates different levels of allocation to achieve high bandwidth utilization and QoS guarantees. An example is presented finally.

2.3.1 Overview of the Algorithm The goal of bandwidth allocation in 802.16 is actually to fill up the whole TDD time frame, in which the proportions of the uplink and downlink subframes can be dynamically adjusted. Every subframe is further allocated to service classes/queues of different QoS requirements. Observing these two targets, the Two-Phase Proportionating (TPP) is proposed in this work to well utilize the bandwidth. The first phase decides the subframe sizes according to the requested sizes of both downlink and uplink, while the second phase distributes the bandwidth to each queue based on the corresponding QoS parameter represented as weight, and an adjustment factor reflecting the practical demand. Finally the TPP adheres to the GPSS by granting SSs the allocated bandwidth of each queue. The operations of the algorithm are depicted in Fig. 2.1 and elaborated in the following subsections.

Convergence Sublayer

Translator

Downlink Queues with/o latency

Downlink data

Uplink

Assign slots for queues Determine UL/DL subframe First phase proportionating

UGS

Second phase proportionating

Second phase proportionating

Assign slots to SSs

Assign slots to SSs

Write in DL-MAP

Write in UL-MAP

ertPS

Translator Bandwidth requests

rtPS nrtPS

Two-Phase Proportionating

BE

PHY Layer

Fig. 2.1 Architecture of the Two-Phase Proportionating (TPP)

Downlink frame

Uplink frame

22

Y.-N. Lin et al.

2.3.2 Detailed Operations of TPP 2.3.2.1 Bandwidth Translation and Slot Dispatching A service flow in an SS issues a bandwidth request whenever necessary. After the BS receives the data traffic from the backbone network or the uplink bandwidth requests from SSs, the TPP translates them from data bytes into the OFDMA slots, which are the basic transmission unit in PHY. This can be done by dividing the data bytes by the OFDMA slot size, in which the OFDMA slot size is derived by multiplying the number of bits that can be encoded over a subchannel by the number of symbols in a slot. Notably the number of symbols in a slot is three for UL while two in DL, and the data bytes should include the requested bandwidth from a SS, MAC headers, and PHY overhead such as the Forward Error Correction (FEC), preamble, and guardtime. These slots are then dispatched to the corresponding service queues comprising the five uplink classes as well as the two downlink classes with/o the latency guarantee. Each queue employs three variables, the bandwidth request slots (BRQ), Rmax , and Rmin , to accumulate the number of requested slots, MSTR and MRTR, respectively. All of them are translated from data rate to number of slots per frame duration. 2.3.2.2 First Phase: Dividing a Frame into Downlink and Uplink Subframes To fit the traffic data into the time frame, TPP determines the proportion of the uplink and downlink subframes according to their accumulated BRQs in each frame. However, this is not trivial because of different slot definitions of the uplink and downlink, and could result in unused symbols. For example, if the uplink is proportionally allocated 19 symbols, only 18 of them will be used to form 18/3 = 6 slot columns, where a slot column contains three consecutive symbols. This problem is solved as follows. Depicted in Fig. 2.2, the most appropriate placement of the separator dividing uplink and downlink subframes is assumed to be x steps from the right, in which one step is considered 6 symbols, the least common multiple of the uplink and downlink slots. This is to ensure that all symbols are used up after the division. Two cases need to be discussed here, namely when S, the number of symbols in a frame, is odd and when S is even. If S is odd, the scheme starts with an initial condition in which a slot column exists in the uplink subframe so that the number of remaining symbols, S-3, is dividable by 2 in the downlink, leaving no unused symbols. Then the separator moves x steps toward left, which is supposed to be the correct position, resulting in 1 + 2x slot columns for the uplink and (S − 3)/2 − 3x slot columns in the downlink. The ratio should be the same as the ratio of the uplink and downlink requested slots, namely 1 + 2x UR = , S−3 DR − 3x 2

(2.1)


23

Fig. 2.2 The placement of the separator in the first phase

S symbols 1 slot comprises 2 symbols

1 slot comprises 3 symbols

UR

S −3 − 3x slot col. 2 UR 1 + 2x = DR S − 3 − 3 x 2

Separator (finial placement)

Control messages

Subchannel index

DR

S −3 slot col. 2 Downlink Subframe

TTG

1+2x slot col.

moving x steps, namely 6x symbols

Separator (initial placement)

2

Slot col.

Uplink Subframe

where UR and DR represents the BRQ of the uplink and downlink, respectively. Similar concept can be applied to the case when S is even, except that in the initial condition no slot column exists in the uplink whereas S/2 slot columns are derived in the downlink, 2x UR . = S DR − 3x 2

(2.2)

The x can be obtained after solving the equation and notably is rounded off if it has a fraction. 2.3.2.3 Second Phase: Allocating Subframes to Queues After properly dividing the frame into uplink and downlink subframes, in the second phase we start to allocate them to service queues. In this phase, the Rmin of all queues are firstly satisfied for minimum slots guarantee, followed by the proportionating of the remaining slots to queues except the UGS and ertPS whose requested slots are already served. Since higher service classes typically have higher Rmax values, we take the Rmax as the weight for proportion. However, only referring to Rmax may cause bandwidth waste or starvation of some queues. An example for the former case is a high class queue having a BRQ very close to Rmin . The additional number of slots assigned will be excessive because of the large Rmax , leading to unnecessary bandwidth waste. Similarly, a low class queue yet having a BRQ close to Rmax may not get enough feed. We use an adjustment factor, B R Q − Rmin Rmax − Rmin

24

Y.-N. Lin et al.

referred to as the A-Factor, for the Rmax of each queue to fix this problem so that a high class queue requiring less bandwidth (BRQ) will be reflected and offset while a low class queue demanding much will be compensated. The remaining slots are therefore allocated according to the following proportion rt P S nr t P S BE B R Q r t P S − Rmin B R Q nr t P S − Rmin B R Q B E − Rmin rt P S nr t P S BE Rmax : Rmax : Rmax . rt P S nr t P S BE B E r t P S nr t P S Rmax − Rmin Rmzx − Rmin Rmax − Rmin (2.3)

2.3.2.4 Per-SS Bandwidth Grant within Each Queue The slots allocated to each queue are finally distributed to SSs in the fashion of GPSS. Similar to the second phase, the minimum number of requested slots of each SS is satisfied first. Nevertheless, the remaining slots of each queue are evenly assigned to SSs since there is no priority among them. 2.3.2.5 Example This section gives an example of the TPP, in which UR and DR are 60 and 40, respectively. Suppose S is 26, then the separator should be moved toward left with number of steps x = 3 according to Eq. (2.1), indicating 6x/3 = 6 slot columns for uplink while (26−6x)/2 = 4 slot columns for downlink. If we use direct proportion, however, the number of symbols for uplink is 26 × [60/(60 + 40)] ∼ = 16, in which only 15 symbols are effective. The uplink is adopted as an example for the second phase. Assuming 6 subchannels in a symbol, 6 × 6 = 36 slots are thus allocated to the uplink after the first phase. Rmin , BRQ, and Rmax of the five service classes are as in Table 2.1. The scheduler allocates the guaranteed minimum number of slots to each queue, and later proportionate the remaining slots to queues of the lower three classes according to Eq. (2.2) since the UGS and ertPS are already satisfied. As we can see in the table, using Rmax as the weight without the A-Factor causes three slots to be unnecessarily assigned to the rtPS.

Table 2.1 Parameters and allocation results of the second phase Item

UGS

ertPS

rtPS

nrtPS

BE

Rmax

8

8

16

8

12

BRQ

8

8

6

8

12

Rmin

8

8

6

4

2

BRQ-Rmin

n/a

n/a

0

4

10

Rmax with A-Factor

n/a

n/a

0

2

6

Rmax without A-Factor

n/a

n/a

3

3

2

2


25

2.4 Simulation Through OPNET simulation we evaluate the TPP algorithm, focusing on the bandwidth utilization and the differentiated guarantee among service classes.

2.4.1 Simulation Setup We have made several modifications on the original DOCSIS module of OPNET to adapt to the IEEE 802.16 requirements. The topology consists of one BS serving 20 SSs, and two remote stations including an FTP server and a voice endpoint. Five service classes are supported and each class involves four SSs. The UL and DL channel capacity is 10.24 Mbps and the frame duration is 5ms. All classes in Figs. 2.4, 2.5 and 2.6 run voice applications with G.711 codec and 64Kbps bit rate. The Rmax of rtPS, nrtPS and BE are 8, 6, and 4, respectively, while Rmin are 4, 2, and 1, respectively.

2.4.2 Numerical Results 2.4.2.1 Subframe Allocation: Static vs. Dynamic The first-phase of TPP is advantageous in utilizing the bandwidth when the load of the uplink and downlink are different, as Fig. 2.3 proves. The FTP traffic load of the downlink is three times of the uplink. In Fig. 2.3a the downlink utilization is bound to 50% because of the static subframe allocation. However, by stealing the unused uplink slot columns for the downlink, TPP improves the overall link utilization from 75 to 96%. (b) 100 90 80 70 60 50 40 30 20 10 0

Total

0

10

Downlink

20

30

Uplink

40 50 Time (sec)

Bandwidth utilization (%)

Bandiwidth utilization (%)

(a)

60

70

80

100 90 80 70 60 50 40 30 20 10 0

Total

0

10

20

30

Downlink

40 50 Time (sec)

60

Uplink

70

80

Fig. 2.3 Bandwidth utilization: (a) static subframe allocation with UL:DL = 1:1; (b) dynamic subframe allocation with UL:DL = 1:3

2.4.2.2 Effectiveness of the A-Factor As introduced previously, the A-Factor helps avoid bandwidth waste by reflecting the requested amount of classes. To understand the effectiveness, we compare it with

26 1.4

Rmax with A-Factor Rmin Rmax BRQ BRQ-Rmin

1.2 Grant Ratio.

Fig. 2.4 Effectiveness of the A-Factor. Four schemes with a simple weight such as Rmin , Rmax , BRQ, and BRQ-Rmin are involved for comparison

Y.-N. Lin et al.

1 0.8 0.6 0.4 0.2 0

rtPS

nrtPS

BE

four schemes which simply use a weight such as Rmin , Rmax , BRQ, and BRQ–Rmin , for each class. A term named Grant Ratio is defined as the ratio of number of allocated slots to the number of requested ones. A grant ratio larger than 1 means that the service class is allocated more than requested, resulting in bandwidth waste. As presented in Fig. 2.4, the Grant Ratios of rtPS using Rmin , Rmax and BRQ are about 1.2, implying excessive allocations, while appropriate amounts are provided when using the A-Factor and BRQ–Rmin . The nrtPS using Rmax with A-Factor obtains more slots than those in other schemes. In BE, though the one using BRQ–Rmin has the highest Grant Ratio, this scheme is not feasible because it tends to favor classes with a small Rmin which oftentimes is BE, and therefore violates the spirit of service differentiation. Service Differentiation — Figure 2.5a displays the number of granted and minimally reserved slots, respectively, as well as the average delay for each class under different numbers of SSs. As we can see in the figure, the UGS and ertPS sustain the number of reserved slots even when the number of SSs advances 60. For other classes, the system guarantees the differentiated Rmin , namely 4:2:1, until the number of SSs exceeds 50. For the average delay depicted in Fig. 2.5b, only minor difference is observed among classes initially until the number of SSs reaches 40, rather than 50. This is because not enough additional slots can be allocated but only the minimum requirement is satisfied. Again, the delay of the UGS and ertPS are always kept under 10 ms. Performance — The performance of TPP is compared with the Deficit Fair Priority Queue (DFPQ) and Strict Priority (SP) in terms of bandwidth utilization, as depicted in Fig. 2.6a. From the figure we can learn that the bandwidth utilizations of the three algorithms increase linearly but start to decrease when hitting a certain level: 85.5% for TPP, 80.6% for DFPQ and 68.4% for SP. The reason why they are not fully utilized is explored by looking into the average frame occupation of service classes, as presented in Fig. 2.6b. Each class has an unused portion, which occurs during the translation from requested bytes to slots. Since the calculation, namely dividing the requested bytes by slot size, always rounds up, the resulted assignment is often larger than expected. As an example shown in Fig. 2.6c, assuming that a slot contains 64 bytes, which is one of the supported sizes, and the amount requested by service flow (SF) #1 is 213 bytes, the number of requested slots is thus four, causing a 256 − 213 = 43


Fig. 2.5 Service differentiation: (a) the variation of minimally reserved slots and granted slots of each request under each class; (b) average delay between service classes. To make the differences recognizable, in (a) the numbers of allocated slots per request for (rtPS-total, nrtPS-total, BE), which is (47.2, 52.7, 53.7) for 10 SSs and (22.8, 23,2, 21,7) for 20 SSs are omitted

27 UGS nrtPS-min nrtPS-total

(a) Number of allocated slots per request

2

ertPS BE-min BE-total

rtPS-min rtPS-total

10 8 6 4 2 0

10

20

30 40 Number of SSs

50

60

(b) Average delay (ms)

10000 UGS ertPS rtPS nrtPS BE

1000 100 10 1

10

20

30 40 Number of SSs

50

60

bytes waste. However, TPP alleviates this effect by reserving minimum required slots first, rather than paying up all requested slots at once for an SF. Take Fig. 2.6c for instance and assume that the number of available slots is nine and the MRTR of each SF is three, TPP breadth-firstly allocates every SF three slots which are slightly insufficient whereas the allocated slots are not wasted; nonetheless, the DFPQ depth-firstly tries to satisfy all SFs’ requested slots but results in the waste for the first two SFs and the starvation of the third which has the lowest priority. The SP has a largest waste also because of its static allocation. Besides, the UGS contributes to the relatively more amount of unused portion than other classes, revealing the drawback of unnecessary slot reservation. Finally, aside the high efficiency in bandwidth consumption, TPP is advantageous in service differentiation. As depicted in Fig. 2.6b, the ratio of allocated bandwidth for rtPS, nrtPS and BE is very close to 4:2:1, compared to other two algorithms.

2.5 Conclusions and Future Work This work considers the problem of bandwidth allocation for 802.16 in order to well utilize the precious wireless link and to support service differentiation. Among others, an allocation scheme called TPP is presented. The uplink and downlink

Y.-N. Lin et al. Bandwidth utilization (%)

28 100 90 80 70 60 50 40 30 20 10 0

TPP DFPQ SP

5

10

15

(a)

20 25 30 35 Number of SSs

40

45

50

Frame occupation (%)

35 UGS ertPS rtPS nrtPS BE unused

30 25 20 15 10 5 0

(b)

TPP

DFPQ

Unused

SP

213 bytes

4 slots 256 bytes

SF1

SF1

Requested 64 bytes per slot

Allocated Allocated but unused

Requested

(c)

SF1

SF2

SF3

TPP

SF1

SF2

DFPQ

SF3

SF1

SF2

SF3

Fig. 2.6 Performance comparison with SP and DFPQ: (a) bandwidth utilization; (b) frame occupation under three schemes with 48 SSs; (c) example of allocations by TPP and DFPQ, in which number of slots to be allocated to three service flows is 9

bandwidth allocations are considered at the same time so that the allocation can be dynamically adjusted. Simulation results confirm that bandwidth utilization increases 20% by applying the first phase proportionating; differentiation among classes is appropriately achieved in the second phase.

2


29

Though service differentiation is carried out in BS, the SSs should also be capable of providing similar support in order to meet the QoS requirement of various applications. Therefore, the future work will be focusing on designing a sophisticated allocation algorithm for the SS to manipulate the per-SS grant. The ultimate target will be implementing both algorithms in real BSs and SSs for performance validation.

References 1. IEEE 802.16 Working Group, “Air Interface for Fixed and Mobile Broadband Wireless Access Systems – Amendment for Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands,” Feb. 2006. 2. G. Nair et al., “IEEE 802.16 Medium Access Control and Service Provisioning,” Intel Technology Journal, Vol 8, Issue 3, Aug. 2004. 3. IEEE 802.16 Working Group, “Air Interface for Fixed Broadband Wireless Access Systems,” Oct. 2004. 4. IEEE 802.11 Working Group, “Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications,” Sep. 1999. 5. Cable Television Laboratories Inc., “Data-Over-Cable Service Interface Specifications - Radio Frequency Interface Specification v1.1,” July 1999. 6. W. M. Yin, C. J. Wu and Y. D. Lin, “Two-phase Minislot Scheduling Algorithm for HFC QoS Services Provisioning,” GLOBECOM, Nov. 2001. 7. M. Hawa and D. W. Petr, “Quality of Service Scheduling in Cable and Broadband Wireless Access Systems,” IWQoS, May 2002. 8. G. S. Chu, D. Wang and S. Mei, “A QoS Architecture for the MAC Protocol of IEEE 802.16 BWA System,” Communications, Circuits and Systems and West Sino Expositions, IEEE, July 2002. 9. K. Wongthavarawat and A. Ganz, “IEEE 802.16 Based Last Mile Broadband Wireless Military Networks with Quality of Service Support,” MILCOM, Oct. 2003. 10. J. Chen, W. Jiao and H. Wang, “A Service Flow Management Strategy for IEEE802.16 Broadband Wireless Access Systems in TDD Mode,” ICC, May 2005. 11. S. Maheshwari, S. lyer and K. Paul, “An Efficient QoS Scheduling Architecture for IEEE 802.16 Wireless MANs,” Asian International Mobile Computing Conference, Jan. 2006.

Chapter 3

A Downlink MAC Frame Allocation Framework in IEEE 802.16e OFDMA: Design and Performance Evaluation Andrea Bacioccola, Claudio Cicconetti, Alessandro Erta, Luciano Lenzini, Enzo Mingozzi and Jani Moilanen

Abstract The IEEE 802.16e standard specifies a connection-oriented centralized Medium Access Control (MAC) protocol, based on Time Division Multiple Access (TDMA), which adds mobility support defined by the IEEE 802.16 standard for fixed broadband wireless access. To this end, Orthogonal Frequency Division Multiple Access (OFDMA) is specified as the air interface. In OFDMA, the MAC frame extends over two dimensions: time, in units of OFDMA symbols, and frequency, in units of logical sub-channels. The Base Station (BS) is responsible for allocating data into the frames so as to meet the Quality of Service (QoS) guarantees of the Mobile Stations’ (MSs) admitted connections. This is done on a frame-by-frame basis by defining the content of map messages, which advertise the position and shape of data regions reserved for transmission to/from MSs. We refer to the process of defining the content of map messages as frame allocation. Through a detailed analysis of the standard, we show that the latter is an overly complex task. We then propose a modular framework to solve the frame allocation problem, which decouples the constraints of data region allocation into the MAC frame, i.e. the definition of the position and shape of the data regions according to a set of scheduled grants, from the QoS requirements of connections. Allocation is carried out by means of the Sample Data Region Allocation algorithm (SDRA), which also supports Hybrid Automatic Repeat Request (H-ARQ), an optional feature of IEEE 802.16e. Finally, we evaluate the effectiveness of SDRA by means of Monte Carlo analysis in several scenarios, involving mixed Voice over IP (VoIP) and Best Effort (BE) MSs with varied modulations, with different sub-carrier permutations and frequency re-use plans. Keywords IEEE 802.16 · Medium access control protocols · Resource allocation

E. Mingozzi (B) Dipartimento di Ingegneria dell’Informazione, University of Pisa, Italy e-mail: [email protected]


31

32

A. Bacioccola et al.

3.1 Introduction During the last few years, we have witnessed a daunting increase of the use of electronic communication devices in everyday life. This is due to the spread of sophisticated handheld equipments, such as mobile phones and palmtops, which are available at an increasingly lower cost. This has boosted a technology advance in the area of mobile Broadband Wireless Access (BWA), since these devices, by necessity, cannot rely on the use of wired connections. A by-product of these two factors is that people are becoming more and more accustomed to portable communication devices, which, in turn, produces unforeseen needs and requirements. For instance, the well-known GSM technology, primarily targeted at traditional voice applications, has recently moved towards the GPRS/EDGE architectures for packet-based data transmission. Packet access is also implemented for multimedia services in UMTS, by means of High-Speed Downlink/Uplink Packet Access (HSDPA/HSUPA), in addition to the legacy code-division multiple access (CDMA) based circuit-switched mode. In the context of mobile BWA, a novel standard has been published recently by the IEEE, namely IEEE 802.16e [1], which extends the 2004 version of IEEE 802.16 for fixed BWA [2,3], so that high transmission efficiency for mobile users is coupled with Quality of Service (QoS) support to enable multimedia services. A non-profit association, the Worldwide Interoperability for Microwave Access (WiMax) forum, was formed to define the specifications for compatibility and interoperability of the IEEE 802.16 wireless equipments. According to the WiMax forum, Orthogonal Frequency Division Multiple Access (OFDMA) is the target air interface for mobile BWA with IEEE 802.16e. In fact, OFDMA has been shown to provide mobile users with an improved resilience against multi-path fading in non-line-of-sight environments, with respect to alternative technologies, like FDM/TDM and CDMA, which are used in competing mobile wireless standards, respectively GPRS, EDGE, and UMTS [4]. Furthermore, a multiple access technique based on OFDMA has been proposed by the 3GPP consortium as the downlink air interface in the context of the Long Term Evolution (LTE) project, which is an ongoing effort to lead the current UMTS standard towards 4G wireless technologies [5]. An IEEE 802.16e cell consists of a number of Mobile Stations (MSs) served by a Base Station (BS), which controls the access to the wireless medium in a centralized manner. Before transmitting to/receiving from the BS, any MS must request the admission of a new connection. If the connection is accepted, the BS is then responsible for meeting the requested QoS guarantees. The access to the medium is scheduled on a frame basis. MAC frames extend over two dimensions: time, in units of OFDMA symbols, and frequency, in units of logical sub-channels [4]. Data packets are thus conveyed into bi-dimensional (i.e. time and frequency) data regions, which are advertised by the BS via specific in-band control messages, which share the same resources as data. In the following, the process of defining the content of maps is referred to as frame allocation, which is left unspecified by the standard. Since the frame allocation problem is a complex task, which can significantly impact on the overall performance of the system, we propose a modular framework

3

A Downlink MAC Frame Allocation Framework

33

based on a pipeline approach. Specifically, our framework decouples the process of scheduling data to the admitted connections from the activity of allocating the data regions into the MAC frames. The SDRA algorithm is then provided as a simple solution to the latter, i.e. allocation, which does not depend on the specific scheduling algorithms adopted by the BS to meet the QoS requirements of the admitted connections. SDRA has been originally proposed in [6] and is extended here to support Hybrid Automatic Repeat Request (H-ARQ), which is an optional feature of the IEEE 802.16e to enhance coverage and capacity in mobile applications. An extensive analysis of the proposed algorithm is carried out through Monte Carlo analysis to evaluate its performance under varied system configurations.

3.2 IEEE 802.16e The IEEE 802.16e MAC protocol is connection-oriented and explicitly supports QoS by defining five different QoS scheduling services: namely, Unsolicited Grant Service (UGS), real-time Polling Service (rtPS), extended real-time Polling Service (ertPS), non real-time Polling Service (nrtPS), and Best Effort (BE). Each scheduling service is designed to support a specific class of applications and is therefore characterized by a different set of QoS requirements (see [7] for further details). Connections are uni-directional, i.e. either uplink or downlink, while MSs can establish multiple connections with the BS. MAC Service Data Units (SDUs), which are used to convey data from the upper layers, e.g. Internet Protocol version 4 (IPv4) datagrams or Ethernet frames, are encapsulated into MAC Protocol Data Units (PDUs), which are then transmitted to the peer MAC layer through the physical layer. Furthermore, time is partitioned into frames of fixed duration, which are, in turn, divided into downlink and uplink sub-frames1 . The former is used by the BS to transmit data to the MSs, whereas the MSs transmit to the BS in the latter. Fig. 3.1 reports the frame structure in case of Time Division Duplex (TDD). OFDMA is a multiplexing technique for Non-Line-of-Sight (NLOS) operations which subdivides the bandwidth into multiple frequency sub-carriers then grouped into subsets, called sub-channels [8, 9]. The IEEE 802.16e standard specifies a few sub-carrier permutations, i.e. the mapping of logical sub-channels onto physical subcarriers. Different mappings are tailored to different transmission environments and user characteristics, as described in [10]. Two sub-carrier permutations are mandatory for downlink transmission: Partial Usage of Sub-channels (PUSC) and Full Usage of Sub-channels (FUSC). For uplink transmission, PUSC is the only mandatory sub-carrier permutation. A zone is a portion of the frame in which one of the above sub-carrier permutations is applied. As shown in Fig. 3.1, multiple zones within the same downlink or uplink sub-frame, employing different sub-channels mapping schemes, may exist. 1 Frame partition into sub-frames can occur both in the time domain, i.e. TDD, and in the frequency domain, i.e. FDD. The latter is supported by the standard, but TDD is the option currently deployed in the first mobile WiMax releases.

34


Fig. 3.1 IEEE 802.16e OFDMA sample TDD frame structure. The number of OFDMA symbols and sub-channels reported in the figure are not realistic

With a slight abuse of notation, we can consider a frame as a matrix where rows and columns are sub-channels and OFDMA symbols, respectively (see Fig. 3.1). The number of OFDMA symbols and sub-channels per frame depends on the frame duration, the channel bandwidth, the Fast Fourier Transform (FFT) size, the Cyclic Prefix length, the direction, and the sub-carrier permutation. In any case, the minimum resource allocation unit is the OFDMA slot (hereafter slot) and consists of one or more sub-channels by one or more OFDMA symbols. With regard to the mandatory sub-carrier permutations, a downlink PUSC slot is two OFDMA symbols by one sub-channel large, a downlink FUSC slot is one OFDMA symbol by two sub-channels large, and an uplink PUSC slot is three OFDMA symbols by one sub-channel large. Furthermore, the standard specifies several Modulation and Coding Schemes (MCSs) which can be used by the BS and MSs to adapt the transmission rate to the channel conditions. Therefore, while the dimensions of a slot are determined by the physical layer parameters and sub-carrier permutation only, the amount of data that can be conveyed into a single slot depends on the robustness of the MCS. The more robust the MCS is, the less the number of data bytes per slot are. The set of available MCSs, both for uplink and downlink transmission, is advertised by the BS on a regular basis through dedicated control messages, called Downlink Channel Descriptor (DCD) and Uplink Channel Descriptor (UCD). As illustrated in Fig. 3.1, the downlink sub-frame begins with a physical preamble needed by the MSs for synchronization and channel quality estimation. The preamble, which consists of a known sequence of modulated pilots, lasts one OFDMA symbol in the time domain and covers all the available sub-channels in the frequency domain. After the preamble, the BS transmits the Frame Control Header (FCH), which defines the structure of the first downlink PUSC zone and the length in slots of the downlink map (DL-MAP) message. The DL-MAP message

3


35

is thus allocated within the downlink sub-frame in column-wise order from the end of the FCH. If the DL-MAP exceeds the number of rows in a column, the allocation continues from the beginning of the next column. The remaining part of the downlink sub-frame is allocated as a number of data regions, which consist of twodimensional portions of the downlink sub-frame formed by a group of contiguous logical sub-channels in a group of contiguous OFDMA symbols. A data region may be visualized as a rectangle of OFDMA slots as shown in Fig. 3.1, and cannot span over two zones. However, MAC SDUs can be fragmented into many MAC PDUs, which in turn can be conveyed in different data regions2 . In other words, data that are intended to be conveyed in a data region can actually span over multiple data regions, provided that the appropriate slot boundaries are taken into account. If the size of the MAC PDUs of a data region is not a multiple of the number of bytes contained in the data region slots, the remaining portion is padded with stuff bytes. The size of each data region and its coordinates, with respect to the upper-left corner of the downlink sub-frame, depend on the frame allocation process and are specified by the BS in the downlink map (DL-MAP) message, which is advertised at the beginning of the downlink sub-frame in a PUSC zone. The standard allows that data directed to MSs employing the same MCS can be grouped into a single data region to reduce the overhead. In fact, each data region is specified through an Information Element (IE) into the DL-MAP, which can optionally include the list of connections to which the data region is addressed, namely the data region recipients. With regard to the uplink direction, MSs access the medium in accordance with the uplink data regions advertised by the BS. Unlike downlink, uplink data regions are explicitly defined by the standard as a number of contiguous slots, starting from the upper-left corner of the uplink sub-frame, allocated in row-wise order as shown in Fig. 3.1. The end of each uplink data region is the beginning of the next one. Like downlink, each uplink data region incurs the overhead of one IE in the uplink map (UL-MAP) message. The latter is transmitted by the BS in the PUSC zone of the downlink sub-frame. In the following we call maps the DL-MAP and UL-MAP messages. Since maps have to be decoded by all the MSs in the cell, the BS must employ the most robust MCS among those currently in use by the MSs. Therefore, even though the number of bits required to encode a single IE is rather limited, the total amount of overhead, in slots, required for the transmission of maps can be significant. To increase the range and robustness of data transmission, the IEEE 802.16e standard specifies several advanced techniques. For instance, H-ARQ can be enabled on a per-connection basis for fast recovering from channel errors at an intermediate level between the MAC and physical layers [10]. A H-ARQ data region consists of a set of H-ARQ sub-bursts, each containing one or more MAC PDUs followed by a Cyclic Redundancy Check (CRC). The latter is used by the recipient to determine 2

This is only true provided that variable-length fragmentation of MAC SDUs into multiple MAC PDUs is allowed. While the latter is an optional feature of the IEEE 802.16 MAC, it can be effectively employed to reduce the MAC overhead due to padding the unused portion of data regions, and is thus specified as a mandatory feature by the WiMax compliance datasheets.

36


whether the MAC PDUs have been received correctly, in which case it sends a positive indication to the sender. Otherwise, the H-ARQ sub-burst is retransmitted, employing the same MCS, until either it is received correctly or the maximum number of retransmissions is exceeded. Unlike classical ARQ schemes employed at the MAC layer of wireless MAC protocols (e.g. IEEE 802.11), the H-ARQ recipient keeps a copy of all the failed transmissions, and it combines them together to infer the correct sequence of bytes transmitted. Furthermore, acknowledgments of the correct/incorrect reception of the H-ARQ sub-bursts are sent via a dedicated logical channel in the MAC frame, which ensures fast convergence of the H-ARQ process, and hence low transmission latencies. From the frame allocation point of view, H-ARQ adds to major changes with respect to non-H-ARQ data transmissions. First, H-ARQ sub-bursts must be transmitted as a whole, that is, they cannot span over two data regions. Second, in the downlink direction only, H-ARQ sub-bursts that are addressed to MSs with different MCSs can be packed into a single downlink H-ARQ data region. In fact, each H-ARQ downlink data region is advertised in the DL-MAP by means of a specific IE, which specifies the starting point (top-left corner), width (number of OFDMA symbols), height (number of logical sub-channels), and number of H-ARQ subbursts. Each H-ARQ sub-burst is then advertised within the downlink H-ARQ data region IE by means of the length (in number of slots), the MCS, and the recipient.

3.3 OFDMA Frame Allocation An IEEE 802.16e BS is in charge of allocating capacity in both downlink and uplink, by advertising maps on a frame-by-frame basis. We call this process frame allocation. Since the IEEE 802.16e standard specifies neither a mandatory nor a reference algorithm to perform such task, it is assumed to be manufacturer-specific. However, the standard specifications impose a number of constraints and/or requirements on the candidate algorithm to be implemented, which makes its definition an extremely complex task. More specifically, three main issues may be identified as follows. First, the BS has to ensure that admitted connections are provided with the negotiated QoS guarantees [7]. MAC frame allocation could significantly impact the QoS of a given connection. Therefore, the issue of determining which MSs are granted capacity in the next frame, and how many bytes they are allowed to transmit or receive, must be based on the QoS requirements of every admitted connection. Second, the capacity available for data transmission in a frame cannot be granted arbitrarily, but rather it must obey to a number of constraints derived straightforwardly from the standard specification of the OFDMA MAC frame structure. In fact, grants must be organized into a set of data regions in order to transmit (receive) data to (from) the scheduled MSs. Furthermore, downlink data regions must have rectangular shape and must be allocated in the sub-frame without overlapping with each other and without spanning over multiple zones. Finally, use of H-ARQ introduces an additional constraint to data regions, in that they cannot be split into

3


37

smaller data regions. It is easy to see that to satisfy all these constraints is a very challenging task. Note that, although in principle one could think of granting capacity on a slot-by-slot basis, this immediately proves to be unfeasible in practice in an IEEE 802.16 OFDMA system, due to the unbearable cost in terms of the resulting map overhead. Third, signaling in IEEE 802.16 is in-band, i.e. the overall capacity of the downlink sub-frame is shared between control messages, including maps, and data messages, i.e. MAC PDUs transporting user data. As mentioned above, each downlink (uplink) data region is advertised by a different IE in the DL-MAP (UL-MAP). Therefore, when performing frame allocation, one desirable objective would be that of minimizing the MAC control overhead (map overhead hereafter) so as to maximize the capacity available to users. The complexity of the frame allocation task comes from the fact that the above mentioned issues are often in contrast one to each other. On the one hand, to satisfy the QoS constraints that a given user may require, e.g. real-time traffic with strict deadlines, the BS may be forced to grant bandwidth to any MS in certain frames. Yet, should each MS receive a small amount of capacity in every frame, for the sake of service timeliness, this would very likely cause an unacceptable loss of capacity due to the map overhead. On the other hand, one could follow the simplistic approach of scheduling only one MS per sub-frame, which minimizes the map overhead. However, this would lead to an increased latency between two consecutive capacity grants to the same MS. While such an approach would be fairly acceptable for elastic best-effort applications, such as File Transfer Protocol (FTP), it is very unlikely to fit the maximum delay requirements of real-time interactive applications envisaged for IEEE 802.16 networks, such as Voice over IP (VoIP). Finally, given a number of PDUs to be transmitted in a given sub-frame, not all their possible arrangements into a set of data regions are equivalent from the point of view of the corresponding map overhead, thus calling for data region arrangement strategies which aim at minimizing the map overhead. Matters are further complicated by the fact that a data region arrangement which would minimize the map overhead can possibly entail no feasible frame allocations complying with the frame structure constraints.

3.3.1 A Modular Framework In order to tackle the complexity of the frame allocation problem, a general framework can be assumed, which is based on a pipeline approach: each of the above issues is basically solved independently of each other, i.e. forcing when necessary the breaking of mutual dependencies, and the corresponding solutions are then combined in sequence in order to provide the final allocation.3

3 A similar approach has been followed in [18], where frame allocation constraints came from the Subscriber Stations being half-duplex transmission capable only, while the system works in Frequency Division Duplex mode.

38


In particular, we envisage the frame allocation task as being split into three separate and sequential sub-tasks, namely grant scheduling, data region arrangement, and data region allocation. The grant scheduling sub-task takes as input the connections’ QoS requirements, their current traffic load and the available frame capacity, and determines how many MAC PDUs must be transmitted to each MS to satisfy the requirements of its admitted connections. Provided that a given MCS is associated to each MS, this turns into an amount of capacity, in slots, to be granted to each MS, irrespective of its assignment to a specific data region and frequency and time domain allocation in the MAC frame. The data region arrangement sub-task takes as input the output of the grant scheduling sub-task, and arranges the selected grants into different groups according to specific arrangement strategies, which aim at, e.g., minimizing the map overhead or providing better service in terms of packet delay to the MS. Such groups are tentatively targeted to being mapped one-to-one to as many data regions to be allocated in the frame. These data regions are then fed to the last sub-task, i.e. data region allocation, which is responsible for defining the final contents of maps, possibly re-arranging data regions to fit them into the OFDMA frequency and time domains, and re-sizing scheduled grants to make the allocation feasible. So far we have not distinguished between uplink and downlink in our discussion. However, in the uplink sub-frame, the IEEE 802.16e standard explicitly defines the procedures for the data region arrangement and the data region allocation sub-tasks, whereas grant scheduling is left up to manufacturers. Conversely, in the downlink sub-frame, all the three sub-tasks are left unspecified. Therefore, in the following, we will mainly refer to the downlink sub-frame allocation procedure unless otherwise specified. A schematic representation of the modular framework with a numerical example is reported in Fig. 3.2. Grant scheduling selects a list of MSs and a number of slots for each MS that will be transmitted in the forthcoming downlink sub-frame. The data region arrangement module groups the list of grants into a set of data regions, according to a data region arrangement strategy (described in Section 3.3.2). In the example, we assume that data addressed to the MSs employing the same MCSs are grouped together so as to form a single data region (e.g. grants addressed to MS 1 and MS 8 are grouped together into a data region with area equal to 30 slots). Finally, the data region allocation module is in charge of defining the final layout

Fig. 3.2 Modular pipelined approach

3


39

of the downlink sub-frame. Note that, in the example of Fig. 3.2, the data region allocation module is not able to allocate all the data regions as they are received from the data region arrangement module. Thus, the area of data region 4 is allocated into two distinct data regions. It is evident that such a modular approach may result in sub-optimal solutions, even though the optimal solution to each sub-problem is found, because constraints are considered separately. However, we argue that it has substantial advantages, which are outlined in the following. On the one hand, with this approach, QoS provisioning only depends on the algorithm which is implemented by the grant scheduling sub-task. Therefore, the task of scheduling bandwidth for QoS support can be confined to a well-identified functional sub-module, and is naturally abstracted from the details originating from grant allocation within the MAC frame. On the other hand, such isolation allows for implementing scheduling algorithms as a result of simple adaptation of well-known algorithms already proposed for wired networks, where this discipline has been extensively studied [11–13]. Assuming that the available literature on scheduling algorithms is able to provide adequate solutions to the issue of supporting QoS, we dedicate the next two subsections to discuss in detail those issues which are more specific to IEEE 802.16e, i.e. the data region arrangement and allocation sub-tasks, respectively.

3.3.2 Data Region Arrangement As mentioned above, different data region arrangement strategies, both in downlink and uplink sub-frames, may impact differently on the MAC control overhead. Since the control signaling is in-band, reducing the size of maps is beneficial as it allows more data to be conveyed into the same downlink sub-frame, thus possibly increasing the overall throughput while preserving QoS guarantees. We investigate separately how the different data region arrangement strategies affect the MAC control overhead in the downlink and the uplink sub-frame, respectively. We focus on the downlink direction first. IEEE 802.16e provides standard mechanisms for the purpose of reducing the DL-MAP overhead. First, data addressed to MSs with the same MCS can be packed into a single data region. In fact, as mentioned in Section 3.2, for each IE, the BS is allowed to omit the list of MSs to which the PDUs are addressed, provided that the latter are all transmitted with the same MCS. While this may result in a considerable reduction of the DL-MAP size, it has the drawback that each MS is forced to decode all data regions transmitted with its own MCS4 , due to the lack of information with regard to the actual receivers of the data contained into the data region. Therefore, on average, an MS incurs greater energy consumption than needed. This might be a problem especially with mobile terminals, due to their limited energy capabilities. 4 Note that this does not violate the security functions of the IEEE 802.16 to ensure privacy of data transmitted over-the-air, since the payload of MAC PDUs can be encrypted so that only the intended recipient is able to retrieve the original data sent by the BS.

40


On the other hand, energy wastage can be mitigated by explicitly specifying the lists of data region recipients in the IEs, which reduces the chance that an MS unnecessarily decodes the data region, at the cost of increasing the overall map overhead. In fact, in this case, an MS only decodes those data regions actually containing data addressed to it. However, an MS can still incur some energy wastage, due to the fact that a data region may also convey data addressed to other MSs with the same MCS. Yet, having exactly one MS for each data region completely avoids waste of energy, though at the cost of the highest overhead. To evaluate the relevance of such overhead, we define the DL-MAP size theoretical lower bound (TLB), with respect to a set of data regions, as the size of the minimum DL-MAP needed to allocate them. Specifically, TLB is computed assuming that each data region never needs to be split. However, this assumption is in general over-optimistic, since there might not be any feasible allocation of all data regions without splitting one or more of them. In Fig. 3.3 we show the DL-MAP overhead, in terms of the ratio between TLB and the downlink sub-frame duration against the total number of MSs served in the downlink and uplink sub-frames. We assume that each MS is provided with one connection. Numerical results have been obtained with the following system parameters: the frame duration is 10 ms with an FFT size of 1024 and a channel bandwidth of 10 MHz; the downlink and uplink sub-frames consist of 28 and 18 OFDMA symbols, respectively; the number of sub-channels is 30; and MSs employ four different MCSs. The curves reported in Fig. 3.3 are related to three different strategies for the definition of the DL-MAP: (i) MSs are grouped per MCS without

Fig. 3.3 DL and UL map overhead

3


41

specifying the list of the downlink data region recipients; (ii) same as (i), but the list of the downlink data region recipients5 is specified; (iii) each MS is assigned to only one downlink data region. As shown in Fig. 3.3, with strategy (i) the overhead remains constant with the number of MSs. In fact, the size of the DL-MAP does not depend on the number of MSs served in the downlink sub-frame, but it only depends on the number of different downlink MCSs currently used, i.e. four in this scenario. On the other hand, strategy (iii) exhibits an increasing overhead with the number of MSs per frame which severely affects the downlink sub-frame capacity by consuming 34% of the sub-frame with 23 MSs. Intermediate results have been obtained with strategy (ii), where the list of the downlink data region recipients is also advertised within the IE relative to a specific MCS. With regard to the uplink direction, the IEEE 802.16e standard explicitly requires the BS to add an IE to the UL-MAP for each MS served in the uplink sub-frame, regardless of the MCS of the MS. In other words, the UL-MAP size only depends on the number of MSs served in the uplink sub-frame. Therefore, unlike downlink, the BS cannot count on any mechanism of the standard to reduce the control messages overhead. Note that the UL-MAP overhead affects the capacity available for transmission in the downlink sub-frame, although it depends on the number of MSs served in the uplink sub-frame. For the sake of completeness, we also reported the UL-MAP overhead in Fig. 3.3, which confirms that the UL-MAP overhead increases linearly with the number of MSs due to the presence of one IE for each MS. We stress the fact that the frame allocation procedure of the uplink sub-frame is explicitly defined by the standard. Therefore, the UL-MAP curve only depends on the number of MSs actually scheduled to transmit data in the uplink sub-frame since no data region arrangements are permitted, except that specified by the standard. We can conclude that, in an IEEE 802.16e system, the overhead due to maps can significantly reduce the capacity available for transmitting data. This problem can be mitigated by employing a data region arrangement strategy which limits the number of IEs to be advertised.

3.3.3 Downlink Data Region Allocation We now describe the downlink data region allocation sub-task, which consists of selecting both the shape, i.e. width and height, and the position, i.e. the OFDMA symbol and sub-channel offsets, of each data region. These two problems are discussed separately. With regard to data region shaping, in general, there are several ways to shape the same number of scheduled slots into a data region. For example, assume that the BS has to transmit eight slots of data to an MS. If only one data region is used,

5 Note that since we assume that each MS has only one connection, the number of the data region recipients is equal to the number of MSs per frame.

42


there are four shapes available: 2×4, 4×2, 1×8 or 8×1. Still, it is possible for the BS to employ multiple data regions, e.g. two data regions of four slots each. Finally, the BS can inflate the data regions so that more than eight slots are used, e.g. a single 3 × 3 data region, in which case part of the inflated data regions remains empty. However, not all alternatives are equivalent, since the MAC overhead varies depending on the fragmentation of MAC PDUs over various data regions, and the padding of slots that are not completely used for user data. As far as positioning of data regions is concerned, this problem can be viewed as a bi-dimensional bin packing problem, where a number of items (i.e. data regions) with a specified width and height have to be packed into a fixed-size bin (i.e. the downlink sub-frame). This problem is well known in the literature of operations research and has been proved to be NP-hard [14]. Therefore it is not feasible to implement exact methods at the BS, since positioning of data regions is a hard real-time task, with a deadline comparable to the frame duration. Additionally, no efficient heuristics to solve general bin packing problems are known. We thus argue that simple, yet effective, heuristics should be envisaged, by taking into account the inherent properties of IEEE 802.16. For instance, a joint approach between shaping and placing data regions might be employed, so that the shape of items to be allocated is modified to broaden the space of solutions. Furthermore, the use of H-ARQ introduces an additional constraint when shaping data regions because H-ARQ data regions cannot be split at arbitrary boundaries. For example, assume that the BS decides to split an H-ARQ data region consisting of two H-ARQ sub-bursts each consisting of four slots into two data regions. The only feasible configuration is {4, 4}, while {5, 3}, {6, 2}, {7, 1} are not allowed since they would require H-ARQ sub-bursts to be fragmented. 3.3.3.1 A Sample Data Region Allocation Algorithm In this section, we propose an algorithm, that we name Sample Data Region Allocation (SDRA) algorithm, as a solution to the data region allocation problem. According to our modular framework, SDRA works independently of the specific grant scheduling algorithm and the data region arrangement strategy adopted. Without loss of generality, we describe our algorithm in the case where data addressed to MSs which employ a common MCS are grouped into a single data region (strategy (i) in Section 3.3.2). The SDRA algorithm is then extended to support H-ARQ as described at the end of this section. The performance of SDRA is assessed by means of Monte Carlo analysis in Section 3.4. For the ease of readability, we describe the SDRA algorithm in the case of data transmitted in the FUSC zone only. Since the IEEE 802.16e standard specifies that the DL-MAP is always transmitted in the PUSC zone, the slots that lie “below” it, if any, are left unallocated. The extension to the more general case of combined allocation of data into the PUSC and FUSC zones is straightforward. Likewise, we omit the procedure to allocate the UL-MAP, which must be transmitted as a single data region in the PUSC zone, because its size does not depend on the process of downlink data region allocation.

3


43

In the following a slot is said to be allocated if the DL-MAP defines a data region which covers it. More formally, slot (i, j) is allocated if there is at least one IE such that x ≤ i ≤ x + w ∧ y ≤ j ≤ y + h, where x, y are the time, sub-channel upper-left coordinates of the data region, respectively, and w, h are the width, height data region dimensions, respectively. Note that an allocated slot might not contain any data actually. Moreover, we define the number of slots in a row of the downlink sub-frame as W; likewise, the number of slots in a column of the downlink subframe is H. Lastly, we assume without loss of generality that columns and rows are numbered starting from 1. SDRA is based on the following key concepts: (i) the order in which the data regions are passed to SDRA is preserved when they are allocated into the downlink sub-frame; (ii) allocation proceeds backwards in column-wise order, i.e. any slot of column i is not allocated until all the slots of columns j, i + 1 < j ≤ W are allocated. These design choices are motivated as follows. Because of (i) the scheduler is free to decide the priority of the downlink MAC PDUs that have to be transmitted, which can be exploited to assign to certain MAC PDUs a higher priority than others. Examples of MAC PDUs which may need a higher priority than others include: the UL-MAP (in the PUSC zone) and other MAC management messages; data that have not been allocated in previous frames; data belonging to admitted connections with strict QoS requirements, e.g. UGS connections. Additionally, a (partially) opportunistic approach could also be devised, where MAC PDUs of MSs enjoying higher transmission efficiency (i.e. less robust MCS) are more likely to be placed than the others. The impact of this choice on the network utilization is investigated in the performance analysis at the end of this section. Furthermore, choice (ii) is motivated by the implementation concern that the DL-MAP grows in column-wise order starting from the beginning of the downlink sub-frame, according to the standard specifications. Therefore, by letting data filling up the sub-frame from the opposite direction, any grant needs only be allocated once and for all by the allocation procedure, as described below. We first describe the procedure to allocate a set of non-H-ARQ data regions, which are called pending data regions, while those actually placed into the downlink sub-frame are called allocated data regions. The procedure for H-ARQ data regions is described afterwards. Note that these procedures are described separately for the clarity of illustration, but they can be seamlessly integrated into a single allocation procedure to jointly allocate both non-H-ARQ and H-ARQ data regions. As already discussed in Section 3.2, we assume that any non-H-ARQ data region can be split into smaller data regions arbitrarily, though at slot boundaries. Thus, since data regions are allocated in column-wise order, each data region can be split at most into three smaller data regions. The scheduler, thus, must overprovision the amount of slots requested for each data region so as to include the MAC overhead due to (up to) two fragmented MAC SDUs. However, the quantitative impact on the network utilization is negligible. An example of the resulting allocation of four data regions, labeled A, B, C, and D, is depicted in Fig. 3.4. As can be seen, the four data regions input produces eight data regions: A is split into two data regions, C into three, D into two, while B is allocated as a whole. Under

44


Fig. 3.4 Example of non-H-ARQ allocation via SDRA, with FUSC

the assumption of employing FUSC as the sub-carrier permutation, no more data can be allocated into the downlink sub-frame, because the free room that lies below the DL-MAP (column 3 in Fig. 3.4) cannot be exploited. The detailed allocation procedure is described by means of the pseudo-code in Appendix A of this paper. With H-ARQ, the allocation procedure described above is modified according to the fact that H-ARQ data regions can only be split at H-ARQ sub-bursts boundaries. To achieve this, each data region is always allocated as a rectangle with height equal to the number of rows available. The number of columns is then set to the minimum value such that the sum of the sizes of all the H-ARQ sub-bursts contained into the data region is smaller than or equal to the data region size. This way, up to H – 1 slots can remain unused in each data region. A sample allocation of three H-ARQ data regions, A with three sub-bursts, and B and C with two sub-bursts, is depicted in Fig. 3.5.

Fig. 3.5 Example of H-ARQ allocation via SDRA

3


45

There is a major difference between the non-H-ARQ and the H-ARQ versions of SDRA. With H-ARQ, each input data region produces at most one allocated data region, while non-H-ARQ data regions can be split into up to three data regions. However, the latter can always be allocated provided that there is free room in the downlink sub-frame left by the DL-MAP. On the other hand, a H-ARQ data region can contain a sub-burst that does not fit into the remaining space, and thus needs to remain unallocated, while some sub-bursts of other H-ARQ data regions may be successfully allocated. Therefore, the allocation procedure with H-ARQ must proceed even when one or more sub-bursts are discarded, which is never the case with non-H-ARQ. The detailed allocation procedure is described by means of the pseudo-code in Appendix A of this paper.

3.4 Performance Evaluation The performance evaluation is carried out by means of Monte Carlo analysis. Specifically a set of input numerical instances is defined based on the system and network configuration parameters reported below. An input numerical instance is defined as a list of data regions, each with a specified size and MCS; with H-ARQ, the data region also contains the list of H-ARQ sub-bursts. The procedure used to generate a numerical instance is described in Appendix B. The parameters are classified into static and dynamic parameters: static parameters are set in a deterministic manner and are used to derive the dynamic parameters; dynamic parameters are sampled from a random distribution during the numerical instance generation, based on the set of static parameters. Any combination of static parameters is defined as a snapshot. The allocation procedure is then fed with several variations of the snapshot, by initializing the random number generator functions with different seeds. Performance metrics, defined in Section 3.4.1, are estimated for a given snapshot by averaging their respective values over all the numerical instances of that snapshot. No weighting is applied; therefore all numerical instances are assumed as equally probable states of the system, and confidence intervals can be derived using the standard method of independent replications [15]. Confidence intervals are however not drawn whenever negligible with respect to the estimated average. The system and network configuration parameters are reported in Table 3.1. As outlined in Section 3.2, a slot consists of a two-dimensional time/sub-channel rectangle, whose exact dimensions depend on the sub-carrier permutation, i.e. PUSC or FUSC. This difference accounts for the varying size of the downlink sub-frames reported in Table 3.1. Additionally, the greater downlink sub-frame sizes, i.e. 17×30 and 34×16, correspond to the case of 10 MHz physical bandwidth, with a frame duration equal to 5 ms, and ratio between the downlink and uplink sub-frames set to 35/12. The same profile is used for the cases 17×10 and 34×5, respectively for PUSC and FUSC, where a physical re-use of three is assumed. In any case, the downlink sub-frame size includes the slots used to transmit the DL-MAP, which is computed according to the standard specifications as a function

46

A. Bacioccola et al. Table 3.1 Numerical instance parameters

Parameter name

Symbol

Type

Possible values

Sub-carrier permutation H-ARQ support Downlink sub-frame size (slots)

h s

static static static

PUSC, FUSC enabled, disabled 17 × 30, 17 × 10 (PUSC) 34 × 16, 34 × 5 (FUSC) seven, three random (RND), More Robust First (MRF), Less Robust First (LRF) [0, 1] 0.2, 0.4, 0.6, 0.8 72 192 Seven MCSs QPSK-1/2 (6), QPSK-3/4 (9), 16QAM-1/2 (12), 16QAM-3/4 (18), 64QAM-1/2 (18), 64QAM-2/3 (24), 64QAM-3/4 (27) Three MCSs QPSK-1/2 (6), 16QAM-1/2 (12), 64QAM-1/2 (18) VoIP, BE 1, 2, 4

Number of MCSs Data region order

static static

Target offered load Target percentage of VoIP users Size of VoIP PDUs, bytes Size of BE PDUs, bytes MCS (number of bytes/slot)

tol v Sv Sb mcs

static static static static dynamic

User type Average number of BE PDUs per user

u 1/ pbe

dynamic dynamic

of the number of data regions (both non-H-ARQ and H-ARQ) and H-ARQ subbursts (H-ARQ only). For the purpose of analysis, we define the offered load (in slots, ol S ) as the fraction of the overall downlink sub-frame size s required to be allocated as user data: n ols =

j=1

s

aj

,

where a j is the size of the j-th data region provided as input to the allocation procedure, and n is the number of data regions. On the other hand, the target offered load for a given numerical instance is the reference value used as a stopping criterion during the numerical instance generation (see below). It is worth noting that the offered load associated to a given instance may differ from the target offered load, due to the integer nature of the numerical instance generation process. Finally, the offered load (in bytes, ol B ) is computed as the sum of the bytes conveyed by the data regions provided as input to the allocation procedure. As far as data region arrangement is concerned, we assume that all MAC PDUs directed to users with the same MCS are combined into the same data region. The number of MCSs employed depends on the static parameter specified in Table 3.1.

3


47

Specifically, if seven MCSs are used, this means that each user is served with the MCS that best fits its current channel conditions, as reported by the MSs [16]. On the other hand, with three MCSs we simulate a system where the BS can transmit data to some MSs, i.e. the ones that would be served with the MCSs that are unavailable in the three MCSs case, using a more robust MCS. This approach leads to a slightly lower transmission efficiency for some MSs, but reduces the number of IEs that are advertised on average in the DL-MAP, and hence the map overhead. Lastly, the order in which data regions are fed to the allocation algorithm is specified as a static parameter. Three possible approaches are defined:

r r r

Random (RND): Data regions, each related to a different MCS, are shuffled in a random fashion. More Robust First (MRF): Data regions are sorted in decreasing robustness order. In other words, data that have the least transmission efficiency, in terms of the number of bytes conveyed per slot, have the highest chance of being allocated. Less Robust First (LRF): This is opposite approach than MRF, i.e. data regions are sorted in increasing robustness order.

The impact on the performance of both the ordering approach selected and the number of employed MCSs is evaluated in the performance analysis in Section 3.4.2.

3.4.1 Metrics The following performance metrics are defined. The carried load (in slots, cl S ) is defined as the fraction of the downlink sub-frame size s that is allocated as user data by the allocation algorithm. Note that the slots formally allocated, but not used for data transmission, are not included in this computation. On the other hand, the carried load (in bytes, cl B ) is defined as the number of bytes that can be conveyed by the data regions resulting from the output of the allocation algorithm, depending on their MCSs. Then, the success probability is defined as the probability that the carried load is equal to the offered load. It measures the capability of the algorithm to find a solution that satisfies all the specified constraints. The map overhead is defined as the fraction of the downlink sub-frame size s that is allocated to transmit the DL-MAP message. The unused slots per frame is defined as the fraction of the downlink sub-frame size s that is not allocated for any transmission. The padding overhead is defined as the fraction of the downlink sub-frame size s that is allocated in any data region, as a result of map definition, but is not actually used for data transmission. This accounts for slots that are allocated so as to define data regions as rectangles, but are not needed to accommodate transmission foreseen in that data region. This metric is only meaningful when H-ARQ is enabled, as discussed in Section 3.3.3.1.

48


3.4.2 Results In order to evaluate the impact of the system and network configuration parameters on the performance of the downlink MAC frame allocation alone, the results in this section have been obtained without taking into account the MAC overhead due to the management messages other than the DL-MAP, including the UL-MAP. Furthermore, in all the evaluated scenarios, we verified that the performance improvement with PUSC is negligible with respect to FUSC. In other words, the gain due to the number of slots that lie below the DL-MAP, which cannot be allocated in FUSC mode, is quantitatively small in terms of the metrics defined in Section 3.4.1. Therefore, we only report the results with FUSC for clarity of illustration. We start with the analysis of the downlink MAC frame allocation procedure without H-ARQ support, by evaluating the carried load (in slots, cl S ), the map overhead, and the success probability when the offered load (in slots, ol S ) increases from 0.2 to 1. For the ease of presentation, the carried load plotted also includes the map overhead. In Fig. 3.6 we first compare the results with different downlink sub-frame sizes, i.e. 34 × 16 and 34 × 5. As can be seen, a large portion of the downlink sub-frame (i.e. about 20% with a 34×16 sub-frame) is consumed by the transmission of the DL-MAP. Furthermore, this overhead does not depend significantly on the offered load. In fact, the size of the DL-MAP only depends on the number of IEs contained, i.e. data regions allocated, which varies in a narrow interval, i.e. between 1 and 7,

Fig. 3.6 FUSC mode without H-ARQ. Carried load (in slots), map overhead, and success probability vs. the offered load (in slots), with different downlink sub-frame size, i.e. 34 × 16 and 34 × 5, seven MCSs and data regions randomly passed to the allocation algorithm

3


49

which is the maximum number of MCSs. Note that the map overhead does not decrease linearly with the sub-frame size, because of the fixed amount of information that needs to be conveyed in the DL-MAP, regardless of the number of data regions allocated. As far as the carried load is concerned, both the 34 × 16 and the 34 × 5 curves increase when the offered load increases until there is enough room in the downlink sub-frame to allocate the input data completely. On the other hand, when the success probability becomes smaller than one, the carried load saturates to a constant value. In other words, after some value of the offered load that depends on the size of the sub-frame, the latter is always completely allocated (i.e. the sum of the carried load and the map overhead is one), but some input bursts are still discarded (i.e. the offered load is greater than the carried load). Since this effect is due to the DL-MAP transmission overhead, the actual value of the offered load that saturates the sub-frame depends on the sub-frame size: the offered load with a 34×16 sub-frame saturates earlier than that with a 34×5 sub-frame. Note that there is a transient interval of the success probability where most, but not all, the numerical instances fail to be allocated. In the results above, all seven MCSs have been considered. This case is compared to that with three MCSs only in Fig. 3.7, with a 34 × 16 sub-frame. As can be seen, having a smaller number of MCSs reduces the map overhead, which in turn improves the success probability. In fact, with three MCSs the latter drops below one at a value of the offered load about 6% greater than that with seven MCSs.

Fig. 3.7 FUSC mode without H-ARQ. Carried load (in slots), map overhead, and success probability vs. the offered load (in slots), with different number of MCSs employed, i.e. seven and three, 34 × 16 downlink sub-frame size, and data regions randomly passed to the allocation algorithm

50


Fig. 3.8 FUSC mode without H-ARQ. Carried load (in bytes) vs. offered load (in bytes), with different number of MCSs employed, i.e. seven and three, 34 × 16 downlink sub-frame size, and data regions randomly passed to the allocation algorithm

However, the lower overhead due to maps does not improve significantly the performance, in terms of the carried load (in bytes) conveyed in the allocated regions, which is reported in Fig. 3.8 against the offered load (in bytes). This can be explained as follows. With three MCSs only, some PDUs that can be transmitted with a less robust MCS (e.g. 16-QAM-2/3) are allocated instead by means of a less efficient MCS (e.g. 16-QAM-1/2). While this reduces the number of data regions allocated on average, and hence the map overhead, the average transmission efficiency is also lessened. Specifically, the plot in Fig. 3.8 exhibits three phases. First, when the offered load can be entirely allocated (i.e. when it is smaller than ∼3500 bytes), the carried load with both seven and three MCSs coincide with the former. Then, there is a phase (i.e. with the offered load between ∼3500 bytes and ∼4500 bytes) where the carried load does not increase in a linear manner anymore. In this interval having seven MCSs yields a higher carried load, in terms of bytes, due to the higher transmission efficiency. Finally, when the offered load is greater than ∼4500 bytes, the effect due to the map overhead reduction with three MCSs overruns that of the transmission efficiency, which leads to improved performance with three MCSs only. Thus, while the number of slots allocated per sub-frame, reported in Fig. 3.7, is always greater with a smaller number of MCSs, the performance in terms of bytes carried shows a trade-off depending on the input rate, in bytes, to the allocation algorithm. We now investigate the performance in the cases where the data regions, before being fed to the allocation algorithm, are sorted according to the MRF and the LRF strategies described above. In this scenario, we consider seven MCSs.

3


51

Fig. 3.9 FUSC mode without H-ARQ. Carried load (in slots), map overhead, and success probability vs. the offered load (in slots), with different ordering of data regions depending on their MCSs, i.e. MRF and LRF, 34 × 16 downlink sub-frame size, and seven MCSs employed by the BS

As can be seen in Fig. 3.9, the specific ordering strategy does not affect significantly the performance in terms of carried load (in slots). In fact, with both the MRF and the LRF strategies all the curves (almost) overlap. However, beyond the value of about 0.8 of the offered load, the map overhead curves exhibit a divergent trend. Specifically, the MRF strategy entails an increasing smaller map overhead than the LRF one. This can be explained as follows. When the offered load is such that the success probability is smaller than one, i.e. the allocation algorithm fails to allocate all the data regions, the number of data regions allocated on average, and thus the map overhead, is greater with LRF than with MRF. In fact, the greater the transmission efficiency of an MCS, the smaller the size of the data region to convey the same amount of data. Therefore, allocating the data regions in decreasing robustness of MCS produces, on average, a smaller number of data regions which are actually allocated. This accounts for the overhead reduction in the MRF case when the value of the offered load is greater than 0.8. Conversely, the higher the transmission efficiency, the higher the carried load, in terms of bytes, which can be achieved. This result is shown in Fig. 3.10 where the carried load (in bytes) versus the offered load is reported. With an offered load greater than ∼3500 bytes the MRF and LRF curves no longer overlap. The higher transmission efficiency obtained by the LRF strategy yields a carried load up to ∼5300 bytes while the MRF strategy reaches ∼4200 bytes at most. So far we have analyzed scenarios where the H-ARQ is not considered. We now extend our analysis to the case of FUSC sub-carrier permutation with H-ARQ.

52


Fig. 3.10 FUSC mode without H-ARQ. Carried load (in bytes) vs. offered load (in bytes), with different ordering of data regions depending on their MCSs, i.e. MRF and LRF, 34×16 downlink sub-frame size, and seven MCSs employed by the BS

Again, in Fig. 3.11, we report the carried load (in slots), the map overhead and the success probability versus the offered load (in slots) with 34×16 downlink subframes and seven MCSs. The padding overhead is also reported as a measure of the capacity wasted due to the H-ARQ allocation algorithm. The same considerations of the results described earlier without H-ARQ still hold. However, remarkable is the impact of the padding overhead. The latter is comparable with the map overhead, reaching the value of about 0.3. As an effect of the padding overhead, note that the point in which the success probability starts dropping to zero is when the offered load reaches the value of about 0.5 which is actually much less than the value of about 0.75 obtained when H-ARQ is disabled (see Fig. 3.6). The padding overhead also impacts on the carried load (in bytes), which is plotted in Fig. 3.12 against the offered load (in bytes). The results with both seven MCSs and three MCSs are reported. While the curves exhibit a similar trend to the case with H-ARQ disabled (see Fig. 3.8), the quantitative results are much different. Specifically, the maximum value of carried load obtained with H-ARQ is ∼4000 bytes whereas it is ∼5000 bytes when the H-ARQ is disabled. Furthermore, as the offered load increases beyond the value of ∼3000 bytes, the combined effect of the transmission efficiency and the map overhead described in Fig. 3.8 is smoothed by the presence of the padding overhead. Therefore, it is no longer possible to precisely identify the phases in which the curve with seven MCSs lies above the one with three MCSs and vice-versa. We can thus conclude that H-ARQ yields poorer performance, in terms of the carried load, due to the padding overhead.

3


53

Fig. 3.11 FUSC mode with H-ARQ. Carried load (in slots), map overhead, padding overhead, and success probability vs. the offered load (in slots), with 34 × 16 downlink sub-frame size, seven MCSs and data regions randomly passed to the allocation algorithm

Fig. 3.12 FUSC mode with H-ARQ. Carried load (in bytes) vs. offered load (in bytes), with different number of MCSs employed, i.e. seven and three, 34 × 16 downlink sub-frame size, and data regions randomly passed to the allocation algorithm

54


Fig. 3.13 FUSC mode with H-ARQ. Carried load (in bytes) vs. offered load (in bytes), with different ordering of data regions depending on their MCSs, i.e. MRF and LRF, 34 × 16 downlink sub-frame size, and seven MCSs employed by the BS

Finally, in Fig. 3.13 we compare the carried load (in bytes) with the MRF and LRF strategies versus the offered load (in bytes). The results obtained are similar to those without H-ARQ, which have been reported in Fig. 3.10. In particular, note that the LRF curve lies above the MRF curve as soon as the offered load becomes greater than ∼3000 bytes.

3.5 Conclusions In this work we have studied the frame allocation problem in IEEE 802.16e with the OFDMA air interface. Through a detailed analysis of the standard we have identified the constraints and requirements that need to be met by the BS, which are related to both complying with the QoS guarantees of the admitted connections and satisfying the MAC and physical layer specifications. Since addressing all the issues together is an overly complex task, the frame allocation problem has been split into three separated sub-tasks, namely grant scheduling, data region arrangement, and data region allocation. A solution for data region arrangement and allocation, called SDRA, is proposed. SDRA is designed to work with both non-H-ARQ and H-ARQ data regions. An extensive performance evaluation of SDRA has been carried out by means of Monte Carlo analysis under varied system and network configuration parameters,

3


55

with both VoIP and BE users. The following conclusions can be drawn from the results obtained. First, the smaller the MAC frame, the smaller the impact of the map overhead on the overall frame capacity, which anyway consumes a significant amount of the available bandwidth (i.e. up to 20%), and therefore the higher the probability that SDRA completely allocates the set of input data regions. Second, employing a reduced set of MCSs, i.e. three instead of the full set of seven MCSs, greatly reduces the map overhead, as well. However, there is a trade-off between the number of slots that are allocated per frame and the number of bytes that can be actually conveyed by them. In fact, lessening the set of available MCSs reduces the average transmission efficiency. Finally, the performance of SDRA has been shown to greatly depend on the order in which data regions are fed to the algorithm. For instance, ordering the data regions in increasing robustness order achieves the highest amount of data actually carried per frame, while ordering the data regions in decreasing robustness order produces the least amount of map overhead. While the same conclusions hold both for non-H-ARQ and H-ARQ data regions, the latter exacerbates the trade-off between transmission efficiency and map overhead.

Appendix A: SDRA Algorithm Pseudo-Codes In this section we report and describe the pseudo-codes of the SDRA algorithm both in the non-H-ARQ and the H-ARQ cases which have been informally introduced in Section 3.3.3.1. Figure 3.14 reports the pseudo-code of SDRA without H-ARQ support. Data structures are initialized by creating an empty map (1) and a list of allocated data regions (2), and by setting the working variables w and h to the dimensions of the downlink sub-frame, respectively width and height (3–4). These variables will keep, at each step of the procedure, the coordinates of the next slot to be allocated. The allocation procedure is performed by placing one of the data regions in list pending at each iteration of the main loop (7). The iteration terminates when either there are no more data regions to be allocated (7) or adding a new IE to the DL-MAP would make the latter overlap with already allocated data regions (12, 23, 32), whichever comes first. The size of the DL-MAP is updated via the add harq region() function. Newly allocated data regions are added to the list allocated data structures, which consists of a list of data region coordinates, expressed in terms of upper-left corner (i.e. starting column and row, respectively, in slots) and dimensions (i.e. width and height, respectively, in slots). After the data region currently under allocation is extracted from the pending list (9), the allocation consists of (up to) three steps:

r

The first step (11–19) allocates a data region upwards starting from h into the remaining portion of the current column, i.e. w. The number of slots allocated is thus equal to h, provided that there are enough slots in the data region, otherwise the whole data region is allocated (e.g. B in Fig. 3.4). This step is skipped if the next slot to be allocated is at the bottom of downlink sub-frame, i.e. h is equal to H.

56


Fig. 3.14 Pseudo-code of SDRA, without H-ARQ support

r r

The second step (22–29) allocates a data region that spans over multiple columns (cols) (e.g. the second data region resulting from the allocation of C in Fig. 3.4). This step is skipped if the amount of remaining slots to be allocated for the data region is smaller than a full-height column. The third step (32–36) allocates the remaining portion of the data region, which is by construction smaller than a full-height column (e.g. the third portion of C

3


57

in Fig. 3.4). This step is skipped only if the number of slots of the data region before the second step is a multiple of H. The detailed allocation procedure of H-ARQ data regions is described by means of the pseudo-code in Fig. 3.15. Data structures are initialized by creating an empty map (1) and a list of allocated data regions (2), and by setting the working variable width to zero (3). This variable will keep, at each step, the number of columns allocated so far. Unlike with non-H-ARQ data regions, there is no need to store the number of rows allocated so far, since the size of H-ARQ data regions is enforced to be a multiple of the downlink sub-frame height. The allocation procedure is performed by placing one of the data regions in list pending at each iteration of the main loop (5). The iteration terminates when there are no more data regions to be allocated. The size of the DL-MAP is updated via the add harq region() function (8). As for the non-H-ARQ case, newly allocated data regions are added to the list allocated data structure (12), which consists of a list of data region coordinates, expressed in terms of upper-left corner (i.e. starting column and row, respectively, in slots) and dimensions (i.e. width and height, respectively, in slots). After the data region currently under allocation is extracted from the pending list (6), the number of columns needed to fully allocate it are computed and stored into new w (7). Then, there are two cases:

r

The data region fits into the remaining space, after the size of the DL-MAP has been updated (8). In this case the variable width is updated (11), a new data region is added to list allocated (12), and the current data region is removed from the pending list (13). Allocation then restarts with the next data region in list pending.

Fig. 3.15 Pseudo-code of SDRA, with H-ARQ support

58

r


Otherwise, a sub-burst is removed from the current data region (18) and the allocation procedure restarts. If the sub-burst removed is the only one of the data region, then the latter is removed, as well (19). Note that the size of the DL-MAP must be also restored to its previous value (17) since no data region was allocated in this step.

To conclude this section we now compute the computational complexity of SDRA. With both non-H-ARQ and H-ARQ, the procedure is run as a single iteration, each of which only includes simple constant time operations (e.g. addition to/removal from the head of a list, elementary mathematical operations with integer values). Therefore, we can state that the worst-case computational complexity is O(n), where n is the number of data regions or the number of sub-bursts, respectively in the non-H-ARQ and H-ARQ cases.

Appendix B: Numerical Instance Generation A numerical instance is generated as follows. Starting from an empty list, the input list of data regions is produced iteratively according to the following steps: 1. Initialize the offered load in slots (ol S ) and bytes (ol B ) to 0. 2. Consider a new SDU to be transmitted to a new user. 3. Determine the associated MCS mcs according to a probability distribution reflecting a typical MSs’ deployment scenario [17]. 4. Determine the SDU type (i.e. application) according to the corresponding probability distribution: the SDU is VoIP with probability v, BE with probability 1 – v. 5. Determine the number and size (in bytes) of PDUs used to transmit the SDU: 6. If VoIP was selected at point 3), then only one PDU is needed, of size v; 7. Otherwise, if BE was selected, the number n be of PDUs is drawn from a geometric distribution with average 1/ pbe ; 8. Determine the number and size (in slots) to transmit the above PDUs according to mcs. Each PDU is added to the data region reserved for MCS mcs. Moreover, in H-ARQ enabled snapshots, the PDUs are appended to the list of H-ARQ sub-bursts of data region mcs. 9. Update the offered load, both in bytes and slots. If ol S < tol restart from step 1. 10. As the final step, the set of data regions is sorted according to the ordering approach described above. There are three cases: a) RND: Data regions are permuted according to a random uniform distribution. b) MRF: Data regions are sorted deterministically in decreasing robustness order. c) LRF: Data regions are sorted deterministically in increasing robustness order.

3


59

References 1. IEEE 802.16e-2005 (February 2006), IEEE Standard for Local and metropolitan area networks – Part 16: Air Interface for Fixed Broadband Wireless Access Systems – Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum 1. 2. IEEE 802.16-2004 (October 2004), IEEE Standard for Local and metropolitan area networks – Part 16: Air Interface for Fixed Broadband Wireless Access Systems. 3. A. Ghosh, D. R. Wolter, J. G. Andrews, and R. Chen, Broadband Wireless Access with WiMax/802.16: Current Performance Benchmarks and Future Potential, IEEE Comm. Mag. 43(2), 129–136 (2005). 4. WiMax forum (June 2006), Mobile WiMax: A technical overview and performance evaluation. 5. H. Holma and A. Toskala, HSDPA/HSUPA for UMTS: High Speed Radio Access for Mobile Communications (John Wiley & Sons, Hoboken, NJ, 2006). 6. A. Bacioccola, C. Cicconetti, A. Erta, L. Lenzini, E. Mingozzi, A Downlink Data Region Allocation Algorithm for IEEE 802.16e OFDMA, Proc. Information, Communications & Signal Processing (ICICS), Singapore (China), Dec. 10–13, 2007. 7. C. Cicconetti, L. Lenzini, E. Mingozzi, and C. Eklund, Quality of Service Support in IEEE 802.16 Networks, IEEE Network (20)2, 50–55 (2006). 8. L. J. Cimini, Analysis and Simulation of a Digital Mobile Channel Using Orthogonal Frequency Division Multiplexing, IEEE Trans. Comm. (33)7, 665–675 (1985). 9. R. Van Nee and R. Prasad, OFDM for Wireless Multimedia Communications (Artech House, Norbrook, MA, 2000). 10. H. Yaghoobi, Scalable OFDMA Physical Layer in IEEE 802.16 WirelessMAN, Intel Tech. J. (8)3, 201–212 (2004). 11. C. Cicconetti, A. Erta, L. Lenzini, and E. Mingozzi, Performance Evaluation of the IEEE 802.16 MAC for QoS Support, IEEE Trans. Mobile Comput. (6)1, 26–38 (2007). 12. M. Shreedhar and G. Varghese, Efficient Fair Queueing Using Deficit Round Robin, IEEE/ACM Trans. Networking, (4)3, 375–385 (1996). 13. D. Stiliadis and A. Varma, Latency-rate Servers: A General Model for Analysis of Traffic Scheduling Algorithms, IEEE/ACM Trans. Networking, (6)5, 675–689 (1998). 14. S. Martello and P. Toth, Knapsack Problems: Algorithms and Computer Implementations (John Wiley & Sons, Hoboken, NJ, 1990). 15. A. M. Law and W. D. Kelton, Simulation Modeling and Analysis (McGraw-Hill, Columbus, OH, 2000). 16. C. Eklund, R. B. Marks, S. Ponnuswamy, K. L. Stanwood, and N. J. M. Van Waes, WirelessMAN: Inside the IEEE 802.16 Standard for Wireless Metropolitan Area Networks (IEEE Press, 2006). 17. J. Moilanen, OFDMA Allocation Numerical Guidelines, Technical Report, 2006 (unpublished). 18. A. Bacioccola, C. Cicconetti, A. Erta, L. Lenzini, and E. Mingozzi, Half Duplex Station Scheduling in IEEE 802.16 Wireless Networks, IEEE Trans. Mobile Comput. (6)12, 1384–1397 (2007).

Chapter 4

Scheduling Techniques for WiMax Aymen Belghith and Loutfi Nuaymi

Abstract This chapter proposes a state-of-the-art of scheduling techniques for WiMax. We first summarize the practical considerations of a WiMax scheduling algorithm in order to make a link between a scheduling algorithm and its implementation in WiMax. Then, we analyze the proposed use of some known algorithms for WiMax and then some scheduling algorithms specifically proposed for WiMax. Finally, we draw a comparison between the different possible scheduling methods and highlight the main points of each of them. Keywords Scheduling · WiMax

4.1 Introduction The Worldwide Interoperability for Microwave Access (WiMax) [1] system is based on the IEEE 802.16-2004 standard [2] and its amendment IEEE 802.16e [3]. The IEEE 802.16 standard defines the physical (PHY) and medium access control (MAC) layer of fixed and mobile broadband wireless access system. The use of the (2–11 GHz) frequency band for WiMax allows this technology to perform a non line of sight (NLOS) propagation. WiMax is a technology that promises high throughput and spectrum efficiency and provides powerful Quality of Service (QoS) support. The QoS support in wireless networks is a difficult task due to the characteristics of the wireless link and, in the case of a multimedia system such as WiMax, the high variability of the traffic. The IEEE 802.16-2004 MAC specifies four scheduling services, also known as QoS classes, in order to fulfil QoS requirements: Unsolicited Grant Service (UGS), real-time Polling Service (rtPS), non-real-time Polling Service (nrtPS), and Best Effort (BE). The IEEE 802.16e MAC added a fifth service class: extended real-time Polling Service (ertPS). The radio resources have to be scheduled according to the QoS requirements. The WiMax/IEEE 802.16 standard does not define a mandatory scheduling algorithm. Only the framework is given in the standard. Therefore, the choice of the A. Belghith (B) TELECOM Bretagne


61

62

A. Belghith and L. Nuaymi

Fig. 4.1 Packets Scheduling in BS and SS. The uplink scheduler may have different scheduling classes depending on the service type

algorithm is left to the vendor or the operator. The choice of a scheduling algorithm for WiMax/IEEE 802.16 is an open question. There are many known scheduling techniques. There are also scheduling techniques that are specifically proposed for WiMax. The scheduling must be applied to the downlink or the uplink direction in the Base Station (BS). Only uplink scheduling is applied in subscribers stations (SS) (see Fig. 4.1). Since the scheduling is a very active field, we cannot describe all the algorithms proposed for WiMax and then we have selected some of them. The rest of the chapter is organized as follows. Section 4.4.2 presents the considerations to take into account in the design of a scheduling technique. Section 4.4.3 describes some known scheduling methods and the performance evaluation of the deployment of some schedulers in WiMax. Section 4.4.4 presents schedulers specifically proposed for WiMax as well as their performance evaluations. Section 4.5 presents a synthesis of the deployment of different schedulers in WiMax. The conclusion is in Section 4.4.6.

4.2 WiMax practical Scheduling Considerations 4.2.1 The BS Announces its Scheduling Decisions In WiMax, the MAC architecture is centralized at the BS. The BS scheduler is responsible for the whole control access for the different wireless subscribers. In order to indicate the assignment of the downlink and uplink transmission intervals

4

Scheduling Techniques for WiMax

63

Fig. 4.2 Time Division Duplexing (TDD) frame structure in IEEE 802.16

(or bursts) in each frame, the BS transmits the downlink map (DL-MAP) and uplink map (UL-MAP) MAC management messages, respectively. These messages are transmitted at the beginning of the downlink subframe (see Fig. 4.2). When it receives an UL-MAP management message, the SS determines if it can access to the uplink channel during the current frame. Since the SS may have different connections, an uplink scheduler is required in each SS.

4.2.2 The WiMax Scheduling Classes The scheduling algorithm is not specified by the standard. However, the IEEE 802.16-2004 document [2] defines four scheduling service classes in order to fulfil QoS requirements:

r r

r

r

Unsolicited Grant Service (UGS): designed to support real-time applications. The packets have a fixed size and are generated periodically. An UGS connection never requests bandwidth. real-time Polling Service (rtPS): designed to support real-time applications. The packets have a variable size and are generated periodically. An rtPS connection requests bandwidth by responding to unicast polls which are transmitted periodically by the BS. The most important QoS parameters are the minimum reserved traffic rate and the maximum latency. non-real-time Polling Service (nrtPS): designed to support applications that do not have delay requirements. An nrtPS connection requests bandwidth by responding to broadcast polls which are transmitted periodically by the BS. Each nrtPS connection has a minimum reserved traffic rate parameter. This parameter determines the minimum amount of bandwidth to reserve. Best Effort (BE): designed to support applications that have not delay requirements. A BE connection requests bandwidth by responding to broadcast polls which are transmitted periodically by the BS.

The 802.16e [3] added a fifth scheduling service class, called extended real-time Polling Service (ertPS). An ertPS connection benefits from the advantages of both UGS and rtPS connections. This kind of connection is designed to support real-time

64


applications. The packets have a variable size and are generated periodically like in an rtPS connection. Unicast grants are provided in an unsolicited manner like in an UGS connection. However, the ertPS allocations are dynamic and the BS must not change the size of the allocations until receiving a new bandwidth change request from the subscribers.

4.2.3 Link Adaptation and Scheduling In the IEEE 802.16 standard, the frame has a fixed number of OFDM symbols. The number of symbols depends on some parameters such as the frequency bandwidth and Cyclic Prefix (CP). However, the number of useful data bits is variable and depends on the used Modulation Coding Scheme (MCS). The MCS to be used is defined by the link adaptation procedure. The choice of the appropriate MCS depends on the value of the receiver Signal-to-Noise Ratio (SNR). The IEEE 802.16 standard proposes some thresholds of the SNR (see Table 4.1) only as indicative values. For example, when the SNR is equal to 7.0 dB, the station uses QPSK 1/2. The different MCS that are defined by the standard are the following: BPSK 1/2, QPSK 1/2, QPSK 3/4, 16QAM 1/2, 16QAM 3/4, 64QAM 1/2, and 64QAM 3/4. In general, the subscriber switches to a more energy efficient MCS if the SNR is good. The SS can also switch to a more robust MCS if the SNR is poor. Once the MCS is defined, the number of bits per symbol, and then the useful number of bits per frame, can be computed. Therefore, the BS must take into account the link adaptation in its scheduling considerations. Table 4.1 Receiver SNR thresholds (values proposed by the IEEE 802.16e standard) Modulation

Coding

Receiver SNR (dB)

BPSK QPSK QPSK 16-QAM 16-QAM 64-QAM 64-QAM

1/2 1/2 3/4 1/2 3/4 2/3 3/4

3.0 6.0 8.5 11.5 15.0 19.0 21.0

4.3 Well-Known Scheduling Methods Some well-know scheduling methods are presented in this section. The list of the presented schedulers is not exhaustive. It contains the most used schedulers such as the Round Robin (RR), Maximum Signal-to-Interference (mSIR), Prorate, Weighted Round Robin (WRR), and Deficit Round Robin (DRR) schedulers.

4.3.1 Round Robin (RR) Scheduler The Round Robin scheduler, also called cyclic scheduler, equitably distributes the channel resources to the multiplexed packet data calls. This technique is suitable if

4


65

the subscribers have the same traffic and radio characteristics. Indeed, the radio characteristics permit the determination of the Modulation and Coding Scheme (MCS) to use. Therefore, when all the subscribers use the same MCS and have the same traffic, they need the same resources and then the RR scheduler is suitable in these conditions. Nevertheless, these conditions are generally not applicable in a WiMax context.

4.3.2 Maximum Signal-to-Interference (mSIR) Scheduler The maximum Signal-to-Interference (mSIR) scheduler allocates the radio resources to subscriber stations (SS) having the highest Signal-to-Interference Rate (SIR). Then, this scheduler offers high spectrum efficiency. Nevertheless, subscribers having a SIR that is always small may never be served.

4.3.3 Prorate Scheduler The resources allocation in the Prorate scheduler depends on the number of symbols that are required by the different connections. The allocated portion of symbols (of the total number of available symbols) for the connection i is equal to the number of required symbols by the connection i divided by the total number of symbols required by all the connections. The higher the demand of a subscriber is, the more symbols are allocated to this subscriber. However, there are no considerations for the SIR or the MCS used.

4.3.4 Weighted Round Robin (WRR) Scheduler The weighted round robin scheduler is an extension of Round Robin scheduler based on the static weight. An example of WRR algorithm execution is represented in Fig. 4.3. In this example, there are:

r r

Three queues: A, B, and C. The weight of the queues A, B, and C are equal to 2, 1, and 3, respectively.

Fig. 4.3 Packets queues for Weighted Round Robin (WRR) scheduling algorithm

66


In WiMax, the connections have different QoS parameters and the subscribers use different MCS. In another hand, the subscribers have not generally the same traffic. Therefore, the connections do not need the same resources. The use of the WRR scheduler can be suitable for WiMax because different values of weights can be assigned to the different queues to take into account the different requested resources.

4.3.5 Deficit Round Robin (DRR) Scheduler The Deficit Round Robin scheduler associates a fixed quantum (Q i) and a deficit counter (DC i) with each flow i. At the start of each round, DC i is incremented by Q i for each flow i. The head of the queue Queue i is eligible to be dequeued if DC i is greater than the length of the packet waiting to be sent (L i). In this case, DC i is decremented by L i. At each round, one packet at most can be dequeued for each flow. An example of DRR algorithm execution is represented in Fig. 4.4. In this example, there are three queues: A, B, and C:

r r r

A contains three packets: a1 (200), a2 (750), and a3 (280). B contains two packets: b1 (500), and b2 (300). C contains four packets: c1 (100), c2 (900), c3 (300), c4 (250), and c5 (900).

Fig. 4.4 Packets queues for Deficit Round Robin (DRR) scheduling algorithm

4


67

The DDR scheduler requires a minimum rate to be reserved for each packet flow before being scheduled. This characteristic can be useful in WiMax because the subscribers usually require an allocation of a minimum of resources.

4.3.6 DRR and WRR Schedulers Evaluated for WiMax 4.3.6.1 Choice of Schedulers for the BS and the SS Three schedulers must be chosen for the BS and SS; an uplink and downlink schedulers for the BS and an uplink scheduler for the SS. Authors in [4] make the following choice. To use the DRR scheduler, a station must have the knowledge of the packet size of each queue. Since the BS and SS have all the information about their downlink and uplink queues, respectively, they can use the DRR scheduler as a downlink and uplink schedulers, respectively. It remains to choose the uplink scheduler in the BS because the BS does not have the knowledge of the packet length of the different subscribers. However, the BS can estimate the amount of backlog of each connection through the bandwidth requests. Then, the WRR scheduler can be selected as the uplink scheduler in the BS (see Fig. 4.5).

Fig. 4.5 Use of DRR and WRR schedulers

4.3.6.2 Performance Evaluation The performance of IEEE 802.16 is analyzed by simulation in [4]. The main parameters of the simulation are the following: the frequency band is 2–11 GHz, the air interference is the WirelessMAN-OFDM (using orthogonal frequency-division multiplexing), the channel bandwidth is 7 MHz, the duplexing mode is Frequency Division Duplexing (FDD), and the frame duration is 10 ms.

68

r r r


The metrics to evaluate are: Maximum achievable throughput: represents the maximum amount of data that a station can send successfully. Packet-transfer delay: represents the time between the arrival of the packet at the MAC layer of the source of the traffic and the time of the arrival of this packet at the upper layer of the destination of the traffic. Delay variation: represents the difference between the maximum packet-transfer delay and the packet transmission delay.

Two scenarios are considered: residential scenario and Small and Medium-sized Enterprises (SME) scenario. In the first scenario, the BS provides Internet access to SS by means of BE (Best Effort) connections. The results of the average delay and maximum achievable throughput depending on the offered load and number of subscribers are represented in Table 4.2. In the second scenario, the BS provides three types of services:

r r r

Voice over Internet Protocol (VoIP) service: each station has four VoIP sources multiplexed into an rtPS connection. Videoconference service: each station has two videoconference sources multiplexed into an nrtPS connection. Data service: each station has a data source provided by a BE connection. Table 4.2 Performance evaluation of the residential and SME scenario

Metric/Results Residential scenario: Average delay

Results for downlink traffic

r Constant at low cell load r Sharply increase at high cell load

r Downlink delay < uplink delay

Residential scenario: Maximum achievable throughput

Small and Medium-sized Enterprises scenario: Average delay

Small and Medium-sized Enterprises scenario: Delay variation

r Decrease when number of SS increases (because BS transmits higher MAC overhead)

r Constant for voIP service r Smoothly increase at low

Results for uplink traffic

r Constant at low cell load. r Sharply increase at high cell load.

r Uplink delay > downlink delay

r Decrease when number of SS increases (because SSs transmit higher number of physical preambles)

r Constant for voIP service r Smoothly increase at low

cell load for videoconfercell load for videoconference and data service ence and data service r Sharply increase at high r Sharply increase at high cell load for videoconfercell load for videoconference and data service ence and data service

r Delay variation of VoIP

service = delay variation of videoconference service

Downlink scheduler is DRR and uplink scheduler is WRR.

r Delay variation of VoIP

service < delay variation of videoconference service

4


69

The BS, periodically, grants a unicast poll to each VoIP and videoconference services in order to allow SS to send bandwidth request. The results of the average delay and delay variation depending on the offered load, number of subscribers, and service types are represented in Table 4.2.

4.4 Schedulers Specially Proposed for WiMax 4.4.1 Temporary Removal Scheduler (TRS) The temporary removal scheduler (TRS) [5] is described as follows. The TRS identifies the packet calls under power radio conditions. These packet calls are temporarily removed from the scheduling list for a certain adjustable time period TR . The scheduling list contains all the SSs that can be served at the next frame. If TR expires, the temporarily removed packet is checked again. If the radio conditions are still poor, this packet is temporarily removed for another time period TR . The whole process is repeated up to L times. When the packet is removed for a period of L × TR , it is included in the scheduling list independently of the current radio conditions. Then, a penalty time TP prevents the packet call from being immediately selected once more. If an improvement is observed in the radio channel, the packet could be topped up in the scheduling list again. The temporary removal scheduling can be combined with a common scheduler. It can be combined with the RR, and mSIR schedulers. 4.4.1.1 Temporary Removal Scheduler + Round Robin The temporary TRS can be combined with the RR scheduler. The combined scheduler is called TRS+RR. For example, if there are k packet calls and only one of them is temporary removed, each packet call has a portion, equal to 1/(k − 1), of the whole channel resources. 4.4.1.2 Temporary Removal Scheduler + Maximum Signal to Interference Ratio The TRS can be combined with the mSIR scheduler. The combined scheduler is called TRS+mSIR. This scheduler assigns the whole channel resources to the packet call that has the maximum value of the Signal to Noise Ration (SNR). The station to be served has to belong to the scheduling list. 4.4.1.3 Performance Comparison The performance analysis TRS applied to WiMax is performed by simulation. The main parameters of the simulation are the following: the frequency band is 3.5 GHz, the air interference is the WirelessMAN-OFDM, the channel bandwidth is 3.5 MHz, and the frame duration is 2 ms. The schedulers to evaluate are the following: the RR,

70


mSIR, TRS+RR, and TRS+mSIR schedulers. The results for File Transfer Protocol (FTP) 300 kByte download are represented below:

r r r r

The mSIR and TRS+mSIR schedulers provide the highest throughput whereas RR provides the worst results. The mSIR, TRS+mSIR, and TRS+RR schedulers provide the lowest download time whereas the RR scheduler provides the worst results. The TRS+mSIR and TRS+RR schedulers provide the lowest channel utilization whereas the RR and mSIR schedulers provide the worst results. The TRS+mSIR and TRS+RR schedulers provide the lowest packet call blocking whereas the RR and mSIR schedulers provide the worst results.

4.4.2 Opportunistic Deficit Round Robin (O-DRR) Scheduler 4.4.2.1 O-DRR Scheduler Description In [6], the Opportunistic Deficit Round Robin (O-DRR) scheduler is used as an uplink scheduler. The O-DRR scheduler works as follows. The BS polls all the subscribers periodically, every k frames. After each period, called a scheduling epoch, the BS determines the set of subscribers that are eligible to transmit as well as their bandwidth requirements. This set is called eligible set. A subscriber is eligible to transmit when:

r r

the subscriber has a non empty queue, and, the signal-to-interfernece-plus-noise ratio (SINR) of its wireless link is above a minimum threshold, called SINRth . A scheduled subscriber is a subscriber that:

r r

the subscriber is eligible at the start of the current scheduling epoch, and, the subscriber is eligible during a given frame of the current scheduling epoch.

The scheduled set is changed dynamically. This changing depends on the wireless link state of each eligible subscriber. At the beginning of a new scheduling epoch, the BS resets the eligible and scheduled set and performs the above process again. 4.4.2.2 Determination of the Polling Interval k The BS polls all the subscribers every k frames. A low value of k causes a polling overhead; thus, the efficiency will be low. Furthermore, a high value of k causes an unfair traffic and a non satisfaction of QoS requirements. Therefore, the BS objective is to minimize the worst-case relative fairness in bandwidth and the normalized delay.

4


71

4.4.2.3 Bandwidth Assignment Using the O-DRR Scheduler It is considered that the BS knows the SINR of each channel. During a scheduling epoch, if the SINR of the wireless link of the subscriber i is above SINRth , then:

r r r

The quantum Qi of the subscriber i is distributed among the scheduled subscribers. The lead/lag counter of the subscriber i is incremented by Qi . The lead/lag counter of the scheduled subscriber j is decremented by the amount that the subscriber j receives above its quantum Qj .

4.4.2.4 Performance Evaluation of the O-DRR Scheduler The results of the number of slots assigned depending on the number of subscribers and k value show that the number of slots assignment increases when k increases. In fact, when k increases, the SINR becomes more variable for the different subscribers and the lead/lag counter has more influence on the bandwidth assignment. The results of the fairness in bandwidth using Jain’s Fairness Index [7] depending on the number of subscribers and k value show that Jain’s Fairness Index remains above 90%. This gives more choice to the provider to choose an appropriate value of k at which the fairness and the bandwidth requirements are both satisfied.

4.4.3 Uplink Packet Scheduler with Call Admission Control (CAC) Mechanism In [8], an uplink packet scheduler with Call Admission Control (CAC) is proposed. The CAC mechanism is based on the token bucket principle. The token bucket is a mechanism used to control network traffic rate. The uplink packet scheduler algorithm works as follows (see Fig. 4.6). First, All the UGS connections are granted. Then, the CAC is applied to the rtPS packets. The different deadlines of these packets are computed. Once the deadlines are determined, the Earliest Deadline First (EDF) scheduler is applied for the attribution of the priorities for the rtPS connections. The EDF scheduler attributes priorities to different packets according to their deadlines. The closer is the deadline of a packet; the higher is its priority. After the allocation of resources for the UGS and the the rtPS connections, resources for the nrtPS connections are allocated if there are remaining bandwidth and the bandwidth requirements of these connections are below a threshold, called TnrtPS . Then, resources for the BE connections are allocated if there are remaining bandwidth and the bandwidth requirements of the BE connections are below a threshold, called TBE . If there are still remaining bandwidth, the nrtPS connections and then BE connections are granted until the use of the whole available bandwidth. The simulation is used to validate the CAC. The main parameters are the following: the frame duration is 1 ms, the size of the bandwidth request is 48 bits, the size

72


Fig. 4.6 Main steps of the uplink packet scheduler with CAC

of the UGS, rtPS, nrtPS, and BE packets are 64 bits, 256 bits, 256 bits, and 128 bits, respectively, and the number of flows are 1000. The simulation results show that the proposed uplink packet scheduler can receive a high number of rtPS connections and guarantee their delay requirements.

4


73

4.4.4 Cross-layer Scheduling Algorithm with QoS Support 4.4.4.1 Cross-Layer Scheduler Description In [9], a novel scheduling algorithm is proposed for WiMax networks. This scheduler is based on affecting a priority to each connection. For the UGS connections, the scheduler must guarantee a fixed quantum of radio resources. The UGS connections are characterized by a constant number of time slots allocated. Therefore, the transmission mode is selected and remains the same during the whole service time. For the rtPS and nrtPS connections, the scheduler must guarantee the latency and the minimum reserved rate respectively. For the BE connections, there is no QoS guarantee but a Packet Error Rate (PER) should be maintained. After serving all the UGS connections, the scheduler allocates all the residual time slots to the rtPS, nrtPS, then BE connections that have the maximum value of a defined Priority Function (PFR). The PRFs, for the rtPS, nrtPS, and BE connections consequently depends on the delay satisfaction indicator, the rate of the average transmission rate over the minimum reserved rate, and the normalized channel quality. The PRFs details are presented in [9]. 4.4.4.2 Performance Evaluation When there is sufficient available bandwidth, the simulation results show that the delay outage probabilities of the rtPS connections are always below 5%. Therefore the latency constraints are guaranteed. The results also show that the average reserved rate of each nrtPS connection is greater than its minimum reserved rate. However, the average transmission rates of the BE connections have large variations and sometimes are null. This behaviour is expected since there is no guarantees for the BE connections. When the residual slots decrease, the performance of the BE connections (then the nrtPS connections) degrades. This is due to the insufficient available bandwidth and the fact that the nrtPS connections have higher priority than the BE connections.

4.4.5 Hybrid Scheduling Algorithm 4.4.5.1 Hybrid Scheduler Description The hybrid scheduling algorithm for QoS in WiMax in [10] works as follows. The Earliest Due Date (EDD) scheduler is used for the real time services while the Weighted Fair Queue (WFQ) scheduler is used for the non-real time services. This is then a hybrid scheduling algorithm. The EDD scheduler is based on dynamic priority. In an EDD queue, the packets are classified in order of their deadline values. The expected deadline time of a packet is calculated by adding the packet arrival time and maximum service time of this packet.

74


The WFQ scheduler provides a required throughput rate for each service. The delay in WFQ for a service is computed as follows: DWFQ =

i=1..n

wi

(R ∗ wi )

Where:

r r r

n: represents the number of services. wi : represents the weight given to the queue i. R: represents the link transmission rate.

In [11], the same hybrid scheduler is proposed; the EDD scheduler is used for the UGS QoS class and the WFQ scheduler is used for nrtPS and BE QoS classes. The only difference is that the nrtPS QoS class has more priority than that of the BE QoS class. 4.4.5.2 Performance Evaluation In [10], the hybrid scheduler is compared with the EDD scheduler employed for the real and non-real time services. The simulation results show that the hybrid scheduler provides, for real time services, less delay than the EDD scheduler. This is due to the competition of the non-real time packets for channel access when the EDD scheduler is used. However, the non-real packets wait for more time when the hybrid scheduler is used. Since the real time services have delay requirements, it is recommended to use the hybrid scheduler in WiMax. In [11], the number of contention slots is investigated using the same hybrid scheduler. The contention slots are used by the SSs to send their bandwidth requests through contention. The simulation results show that the throughput increases when the number of contention slots increases without exceeding the half BE connections. This is due to the decrease of the probability of bandwidth request collisions. If the number of contention slots is grater than the half of the BE connections, the throughput decreases when the number of contention slots increases. This is because less radio resources are reserved for the data transmission.

4.4.6 Frame Registry Tree Scheduler (FRTS) 4.4.6.1 Frame Registry Tree Scheduler (FRTS) Description The Frame Registry Tree Scheduler (FRTS) scheduler [12] contains three operations: packet/request arrival, frame creation, and subscriber’s modulation type change or connection QoS service change. The basic idea of the packet/request operation is to distribute packet transmissions in time frames, based on their deadline. For UGS and rtPS services, the packet deadline is equal to the arrival time plus

4


75

the latency of this packet. The subtree of the last time frame where this packet can be transmitted is updated, if it exists. Otherwise, it is created. For nrtPS and BE services, the packet deadline does not need to be calculated. Then, the subtree of the last existing time frame is updated. The frame creation procedure decides on the frame contents. There are three cases:

r r r

If the subtree of the first time frame contains a number of packets equal to one time frame, all these packets fill up the frame content. If the subtree of the first time frame contains a number of packets less than one time frame, the empty slots are occupied by packets from the next time frame subtrees and/or will be left for contention. If the subtree of the first time frame contains a number of packets more than one time frame, packets for BE service are moved to the next time frame subtree. If there are still excess packets to transmit, first nrtPS packets, then rtPS packets and finally UGS packets are deleted until the number of packets fit exactly into one time frame.

A change in a subscriber’s modulation type or connection QoS service causes a moving of the corresponding subtree to the right modulation substructure or service substructure. 4.4.6.2 Performance Evaluation The FRTS scheduler is compared with a simple scheduler that serves higher priority before lower priority traffic (UGS has the highest priority, then rtPS, nrtPS, and finally BE). Simulation results show that the FRTS scheduler provides better throughput. This is due to the less lost packets. Indeed, FRTS takes into account the deadline of the real time packets (UGS and rtPS). Simulations results also show that the FRTS scheduler can serve nrtPS and BE connections even if the load is high. This is because this scheduler profits from the latency tolerance of the real time packets.

4.4.7 Adaptive rtPS Scheduler 4.4.7.1 Adaptive rtPS Scheduler Description The adaptive rtPS scheduler [13] is used only for the rtPS QoS class. It is based on the prediction of the rtPS packets arrival. As defined in the IEEE 802.16 standard, the BS allocates bandwidth for rtPS traffic after receiving a bandwidth request. When the request is granted by the BS, the subscriber may receive from upper layers new rtPS packets. These packets will wait for the next grant to be sent and, therefore, suffer from extra delay. The basic idea of the adaptive rtPS scheduler is to propose an rtPS bandwidth request process in which the subscriber requests time slots for the data present in the rtPS queue and also for the data which will arrive. The authors of [13] define a stochastic prediction algorithm in order to estimate the data arrival.

76


4.4.7.2 Performance Evaluation The adaptive rtPS scheduler is compared with the weighted scheduler. The simulation results show that the adaptive rtPS scheduler provides better average delay in low and medium load. This is because this scheduler also considers the data generated between the time of the bandwidth request sending and the time of the bandwidth allocation by the BS. Therefore, the adaptive rtPs scheduler requires less buffer size for the rtPS data queue. In high load, the adaptive rtPS and weighted scheduler have the same performance. This is due to the saturation of the network.

4.4.8 Scheduler Ensuring QoS Requirements 4.4.8.1 Description of the Scheduling Proposal In [14], a simple scheduler that ensures the QoS requirements for the different QoS service classes is proposed. This scheduler consists of the allocation of the minimum bandwidth requirements to all the connections and then the allocation of free slots to the rtPS, nrtPS, and BE connections without exceeding their maximum bandwidth requirements (see Fig. 4.7). The number of slots to allocate to the UGS and ertPS is constant since the minimum and maximum bandwidth requirements of these connections are the same. The minimum and maximum number of slots to allocate to the rtPS and nrtPS connexions depends on the minimum and maximum bandwidth requirements, respectively. Since BE connection has no QoS requirements, its minimum number of slots to allocate is null and its maximum number depends on its bandwidth request. After satisfying the minimum bandwidth requirements of all the connections, the unused slots are allocated to the rtPS and nrtPS connections and then distributed between the BE connections.

4.4.8.2 Performance Evaluation When the system contains different type of QoS service classes, the simulations results show that the proposed scheduler allocates the whole requested bandwidth of the UGS and erTPS connections, provides a bandwidth greater than the minimum bandwidth requirements of the rtPS and nrtPS connections and the remaining bandwidth is allocated to the BE connections. The simulations results show that the proposed scheduler fairly distributes the bandwidth between the different BE connections. The simulation results also show that the throughput decreases when the number of the served connections increases. This is due to the increase of the overhead size. Despite of the change of the MCS used and therefore the number of slots to reserve, the proposed scheduler provides the bandwidth requirements.

4


77

Fig. 4.7 Main steps of a scheduler ensuring QoS requirements

4.4.9 Adaptive Bandwidth Allocation Scheme (ABAS) 4.4.9.1 Adaptive Bandwidth Allocation Scheme (ABAS) Description An Adaptive Bandwidth Allocation Scheme (ABAS) for 802.16 TDD systems is defined in [15]. It aims to dynamically determine the suitable downlink-to-uplink bandwidth ratio. In fact, the IEEE 802.16 standard specifies the frame structure but it is up to operators to choose the downlink-to-uplink bandwidth ratio. ABAS

78


performs as follows. The BS determines the different information of the downlink and uplink connections such as their bandwidth requests and the number of downlink and uplink connections. Then, the BS determines the number of slots allocated to the downlink and uplink subframe and adjusts the split between the two parts of the frame. Finally, the BS informs the different SSs about its decision using DLMAP and UL-MAP MAC management messages. The mechanism is repeated at the beginning of each frame. 4.4.9.2 Performance Evaluation ABAS is compared with a static downlink to uplink bandwidth ratio mechanism. Only Transmission Control Protocol (TCP) transfers and BE QoS service class are considered. The results of the throughput of the downlink connections depending on the number of TCP transfers show that ABAS provides better throughput. This is because that ABAS determines the most appropriate value of the downlink to uplink bandwidth ratio. On the other hand, the static downlink to uplink bandwidth ratio mechanism can degrade the throughput if the initial static ratio is not suitable with the traffic characteristics. Moreover, the number of subscribers in the system as well as their connections characteristics can change every time. The results of the throughput of the downlink and uplink connections depending on the ratio of downloading to uploading TCP transfers show that ABAS also provides better throughput. This is due to the tacking into account the traffic characteristics in the determination of the suitable value of the downlink to uplink bandwidth ratio.

4.4.10 Adaptive Polling Service (aPS) 4.4.10.1 Adaptive Polling Service (aPS) Description A novel adaptive Polling Service (aPS) for WiMax system is defined in [16]. The main idea of the proposed mechanism is to adjust the polling period based on the reception of bandwidth request. The BS initializes the polling period with Tmin . Tmin is determined using the average packet arrival rate. If the BS does not receive bandwidth request after N polls, it exponentially increases the polling period until reaching Tmax . Tmax is determined using the tolerable delay of the connection since the aPS mechanism is defined for real-time applications. When the BS receives a bandwidth request, it resets the polling period to Tmin . 4.4.10.2 Performance Evaluation The aPS mechanism is compared with the rtPS mechanism defined in the IEEE 802.16 standard. The simulation results of an ON/OFF TCP traffic show that the aPS mechanism reduces the signalling overhead by 50% to 66%. This reduction comes from the increase of the polling period during the OFF periods. On the other

4


79

hand, the aPS mechanism provides higher delay than the rtPS mechanism. However, the delay is still acceptable for almost of the applications. The simulations results of a TCP-based application show that the aPS mechanism reduces the signalling overhead by 66%. Like for the ON/OFF traffic, the increase of the delay is still acceptable. The simulation results of an online game application working over User Datagram Protocol (UDP) shows that the aPS mechanism reduces the signalling overhead by 50–75%. However, the delay increases and becomes not suitable with all the applications. The increase of delay depends on the parameters of the aPS mechanism. In the different scenarios, the authors investigates only Tmax and show that when Tmax increases, the reduction of the signalling overhead and delay increase. Then, we can choose the suitable value of Tmax depending on the tolerate delay.

4.4.11 Modified Maximum Signal-to-Interference Ratio (mmSIR) Scheduler 4.4.11.1 Modified Maximum Signal-to-Interference Ratio (mmSIR) Scheduler Description In [17], a problem that may exist with the rtPS QoS class is highlighted. If the BS allocates unicast request opportunities and resource grants for rtPS connections in the same frame, the BS cannot immediately take into account the new length of the uplink data connection of the subscriber. The reason is that the BS allocates symbols for rtPS connections before receiving the latest unicast bandwidth request (see Fig. 4.8). Moreover, the mSIR scheduler serves those subscribers having the highest SIR at each frame. So, subscribers having a slightly smaller SIR may be not served and then the mean delay to deliver data increases. The proposed scheduler is based on the

Fig. 4.8 Allocation of symbols for rtPS connection

80


Fig. 4.9 Main steps of the proposed mmSIR scheduler

modification the mSIR scheduler in order to decrease the mean time of sojourn and called modified maximum Signal-to-Interference Ratio (mmSIR) scheduler. The main idea of the mmSIR scheduler is that the BS only serves the subscribers that do not have unicast request opportunities in the same frame. The main steps of this proposed scheduler are shown in Fig. 4.9. 4.4.11.2 Performance Evaluation The mmSIR scheduler is compared with the mSIR scheduler. Simulation results show that the mmSIR provides a decrease in the mean sojourn time. This is mainly due to the non freezing of the SSs having a small SIR. Indeed, the BS serves these SSs when it has already allocated unicast request opportunities to SSs having a higher SIR. The simulation results also show that the mmSIR scheduler provides better throughput. Indeed, the mmSIR scheduler, like the mSIR scheduler, favors those

4


81

SSs having the highest SIR. If it does not serve an SS having unicast request opportunities, it gives priority to other SSs having higher SIR. Furthermore, the mSIR scheduler cannot immediately benefit from the unicast request opportunities of SSs since it has already reserved resources for rtPS connections before receiving the bandwidth requests. Moreover, the mmSIR scheduler serves fewer SSs than the mSIR scheduler. As a preamble is added to each uplink burst (see Section 8.3.5.1 of Ref. [2]), the BS schedules more useful symbols when it serves fewer SSs per frame.

4.5 Synthesis of Different Schedulers Deployed in WiMax After presenting different scheduling methods as well as their performance evaluations, a synthesis of deployment of different schedulers is presented in Table 4.3. If the BS can use a scheduler in downlink traffic, the SS can use the same kind of scheduler in the uplink direction between its connections. In fact, in the two situations, the queues contents are well known. Therefore, the synthesis table contains only the downlink and uplink schedulers in the BS (and not the uplink scheduler in the SS). Further to this synthesis, we can identify some schedulers that can be used for the downlink or/and the uplink traffic. The RR, mSIR, and Prorate scheduler are not suitable to use in a WiMax context for most of the cases. Indeed, the RR scheduler is only suitable for subscribers that have the same characteristics such as the SNR and the traffic load. The mSIR and Prorate schedulers may block the traffic that is generated by subscribers having a poor SIR value or less data, respectively. The DRR scheduler can be used only in the downlink traffic because the BS does not know the queue length of the subscribers. This prevents the BS to perform the DRR scheduler for the uplink traffic. The CAC is performed to control the connections that are transmitted by the subscribers. So, the uplink scheduler with CAC mechanism is only proposed for the uplink traffic. The WRR, TRS, O-DRR, cross-layer, hybrid, and FRTS schedulers can be used in the downlink and uplink traffic. An important research topic is the performance evaluation intended to determine adequate values to the different parameters of these schedulers. A use of the WRR or the hybrid (EDD+WFQ) schedulers requires the determination of the suitable weights. The removal time, penalty time, and maximum number of repetitions are the main parameters of the TRS scheduler. Different values of these parameters completely change the behaviour and performance of the TRS scheduler. For the O-DRR scheduler, the polling interval and minimum threshold of the SINR parameters have an immediate impact on the delay and throughput, respectively. The adaptive rtPS and mmSIR are proposed for the rtPS QoS class in the uplink direction. The parameters of the stochastic algorithm defined for the adaptive rtPS scheduler.

82

A. Belghith and L. Nuaymi Table 4.3 Synthesis of some WiMax schedulers

Scheduler

Possibility of use for WiMax

Comments

Algorithm parameters

Round Robin

No

–

Maximum SIR

Usually not

Prorate

Usually not

Not suitable. Subscribers do not have the same traffic, radio characteristics, and QoS requirements. Subscribers having a poor SIR may be scheduled after an excessive delay. Subscribers, having less data, may be scheduled after an excessive delay. Can be used for the downlink and uplink traffic. Can be used only for the downlink traffic; BS does not know the packet sizes at the SS queues. Can be used for the downlink and uplink traffic.

Weighted Round Yes Robin Deficit Round Robin Yes

Temporary Removal Scheduler

Yes

Opportunistic Deficit Yes Round Robin

Can be used for the downlink and uplink traffic.

Uplink scheduler with CAC mechanism

Yes

Was proposed for the uplink traffic.

Cross-layer

Yes

Hybrid (EDD+WFQ) Frame Registry Tree Scheduler Adaptive rtPS

Yes

Yes

Modified mSIR

Yes

Can be used for the downlink and uplink traffic. Can be used for the downlink and uplink traffic. Can be used for the downlink and uplink traffic. Was proposed for rtPS QoS class in the uplink direction. Was proposed for rtPS QoS class in the uplink direction.

Yes

–

–

Static weights. Fixed quantum.

Removal time (TR ), number of repetitions (L), and penalty time (TP ). Polling interval, and minimum threshold of the SINR (SINRth ). Parameters of token bucket, thresholds of the AC used for the nrtPS and BE connections (TnrtPS and TBE ). – Weights for WFQ scheduler. – Stochastic prediction algorithm. –

4.6 Conclusion In this chapter, we propose a state-of-the-art of WiMax Scheduling. We first present the scheduling framework of WiMax and then describe different proposed trends for WiMax scheduling algorithms. The choice of the scheduling algorithm is highly dependent on the transmission service type and the traffic shape in addition to other

4


83

QoS requirements. We propose a synthesis table for some of the proposed scheduling algorithms. Evidently, all the proposed WiMax scheduling algorithms could not be studied in this chapter. The efficiency of a scheduling algorithm can be estimated through simulations. We propose an NS-2 (Network Simulator) Module for WiMax scheduling in [18]. The details of our WiMax module are presented in [19].

References 1. WiMax forum, http://www.wimaxforum.org/home/, last visited in 17-07-2008. 2. IEEE Std 802.16-2004, Part 16: Air interface for Fixed Broadband Wireless Access Systems, 1 October 2004. 3. IEEE Std 802.16e, Part 16: Air interface for Fixed and Mobile Broadband Wireless Access Systems, 28 February 2006. 4. C. Cicconetti, L. Lenzini, E. Mingozzi, C. Eklund, “Quality of Service in IEEE 802.16 Networks”, IEEE Network – Special Issue on Multimedia Over Broadband Wireless Network, Vol. 20, No. 2, pp. 50–55, March/April 2006. 5. C.F. Ball, F. Treml, X. Gaube, A. Klein, “Performance Analysis of Temporary Removal Scheduling applied to mobile WiMax Scenarios in Tight Frequency Reuse”, The 16th Annual IEEE International Symposium on Personal Indoor and Mobile Radio Communications, PIMRC’2005, Berlin, 11–14 September 2005. 6. H. K. Rath, A. Bhorkar, V. Sharma, “An Opportunistic DRR (O-DRR) Uplink Scheduling Scheme for IEEE 802.16-based Broadband Wireless Networks”, IETE, International Conference on Next Generation Networks (ICNGN), Mumbai, 9 February 2006. 7. R. Jain, D. Chiu, W. Hawe, “A quantitative Measure Of Fairness And Discrimination For Resource Allocation In Shared Computer Systems”, DEC Research Report TR-301, September 1984. 8. T. Tsai, C. Jiang, C. Wang, “CAC and Packet scheduling Using Token Bucket for IEEE 802.16 Networks”, Journal of Communications, Vol. 1, No. 2, May 2006. 9. Q. Liu, X. Wang, G. B. Giannakis, A. Ramamoorthly, “A Cross-Layer Scheduling Algorithm With QoS Support in Wireless Networks”, IEEE Transactions on Vehicular Technology, Vol. 55, No. 3, May 2006. 10. K. Vinay, N. Sreenivasulu, D. Jayaram, D. Das, “Performance Evaluation of End-to-end Delay by Hybrid Scheduling Algorithm for QoS in IEEE 802.16 Network”, Wireless and Optical Communications Networks, 2006 IFIP International Conference on, 11–13 April 2006. 11. D. Tarchi, R. Fantacci, and M. Bardazzi, “Quality of Service Management in IEEE 802.16 Wireless Metropolitan Area Network”, International Conference Communications, ICC’06, Istanbul, Turkey, 11–15 June 2006. 12. S. A. Xergias, N. Passas, and L. Marekos, “Flexible Resource Allocation in IEEE 802.16 Wireless Metropolitan Area Networks”, the 14th IEEE Workshop on Local and Metropolitan Area Networks, LANMAN 2005, Chania, Greece, 18–21 September 2005. 13. R. Mukul, P. Singh, D. Jayaram, D. Das, N. Sreenivasulu, K. Vinay, and A. Ramamoorthly, “An Adaptive Bandwidth Request Mechanism for QoS Enhancement in WiMax Real Time Communication”, Wireless and Optical Communications Networks, 2006 IFIP International Conference on, Bangalore, India, 11–13 April 2006. 14. A. Sayenko, O. Alanen, J. Karhila, and T. Hämäläinen, “Ensuring the QoS Requirements in 802.16 Scheduling”, the 9th ACM international symposium on Modeling analysis and simulation of wireless and mobile systems, MSWiM 06, Malaga, Spain, 2–6 October 2006. 15. C. Chiang, W. Liao, and T. Liu, “Adaptive Downlink/Uplink Bandwidth Allocation in IEEE 802.16 (WiMax) Wireless Networks: A Cross-Layer Approach”, Global Telecommunications Conference, 2007, GLOBECOM 07, Washington, USA, 26–30 November 2007.

84


16. C. Nie, M. Venkatachalam, X. Yang, “Adaptive Polling Service for Next-Generation IEEE 802.16 WiMax Networks”, Global Telecommunications Conference, 2007, GLOBECOM 07, Washington, USA, 26–30 November 2007. 17. A. Belghith, L. Nuaymi, “Comparison of WiMax scheduling algorithms and proposals for the rtPS QoS class”, 14th European Wireless 2008, EW2008, Prague, Czech Republic, 22–25 June 2008. 18. Design and implementation of a QoS included WiMax module for NS-2 simulator: https://perso.enst-bretagne.fr/aymenbelghith/tools/, last visited in 17-07-2008. 19. A. Belghith and L. Nuaymi, “Design and implementation of a QoS included WiMax module for NS-2 simulator”, First International Conference on Simulation Tools and Techniques for Communications, Networks and Systems, SimuTools 2008, Marseille, France, 3–7 March 2008.

Chapter 5

QoS Provision Mechanisms in WiMax Maode Ma and Jinchang Lu

Abstract This chapter presents QoS support mechanisms in WiMax networks. Existing proposals with the state-of-the-art technology have been classified into three main categories: QoS support architecture, Bandwidth management mechanism and Traffic management mechanism. Representative schemes from each of the categories have been evaluated with respect to major distinguishing characteristics of the WiMax MAC layer and PHY layer as specified in the IEEE 802.16d standard. Future research issues and trends are also highlighted. Keywords QoS Provisioning · WiMax · Traffic Scheduling · Admission Control

5.1 Background Broadband Wireless Access (BWA) systems, e.g. IEEE 802.16d standard [1], provide fixed-wireless access between the subscriber station and the Internet service provider (ISP) through the base station. BWA systems have been deployed not only to be complement and expansion of existing last mile wired networks such as cable modem and xDSL but also to be competitor to wired broadband access networks. Due to the upcoming air interface technologies, which promise to deliver high transmission data rates, BWA systems become an attractive alternative. The MAC Layer of IEEE 802.16d was designed for PMP broadband wireless access applications. It is designed to meet the requirements of very-high-data-rate applications with a variety of quality of service (QoS) requirements. The MAC layer is composed of three sub-layers. From bottom to top: the Security Sub-layer (PS), the MAC Common Part Sub-layer (CPS), and the Service Specific Convergence Sub-layer (CS). The former deals with security and network access authentication procedures. CPS carries out the key MAC functions. The CS sub-layer provides the interface to the upper layer; decides the MAC service class for the specific connection and initializes the resource allocation requests of the CPS. M. Ma (B) Nanyang Technological University, Singapore


85

86

M. Ma and J. Lu

The bandwidth request and grant mechanism has been designed to be scalable, efficient, and self-correcting. The 802.16d does not lose efficiency when presented with multiple connections per terminal, multiple QoS levels per terminal, and a large number of statistically multiplexed users. It takes advantage of a wide variety of request mechanisms, balancing the stability of contentionless access with the efficiency of contention-oriented access. The IEEE 802.16d PHY specifies 4 different PHY specifications, namely, WirelessMAN-SC PHY specification, WirelessMAN-SCa PHY specification, WirelessMAN-OFDM PHY specification, WirelessMAN-OFDMA PHY specification. IEEE 802.16d supports both frequency division duplex (FDD) and time division duplex (TDD) PHYs. The standard defines three different modulation schemes. On the uplink, support for QPSK is mandatory, while 16-QAM and 64-QAM are optional. The downlink supports QPSK and 16-QAM, while 64-QAM is optional. In addition to these modulation schemes, the 802.16d PHY also defines various forward error correction (FEC) schemes on the uplink as well as the downlink. These include Reed-Solomon (RS) codes, RS concatenated with inner Block Convolution Codes (BCC), and turbo codes. Support for such a wide variety of modulation and coding schemes permits vendors to tradeoff efficiency for robustness depending on the channel conditions. The advanced technology of the 802.16d PHY requires equally advanced radio link control (RLC), particularly the capability of the PHY to transition from one burst profile to another. The RLC must control this capability as well as the traditional RLC functions of power control and ranging. Figure 5.1 [2, 3] shows the existing QoS architecture of IEEE 802.16d. Uplink Bandwidth Allocation scheduling resides in the BS to control all the uplink packet transmissions. The communication path between SS and BS has two directions: uplink channel (from SS to BS) and downlink channel (from BS to SS). On the downlink (from BS to SS), the transmission is relatively simple because the BS is the only one that transmits packets during the downlink sub-frame. The data packets are broadcasted to all SSs and an SS only picks up the packets destined to it. The uplink channel is shared by SSs. Time in the uplink channel is usually slotted (mini-slots) called by time-division multiple access (TDMA), whereas on the downlink channel BS uses a continuous time-division multiplexing (TDM) scheme. The BS dynamically determines the duration of these sub-frames. Each sub-frame consists of a number of time slots. SSs and BS have to be synchronized and transmit data into predetermined time slots. On the uplink (from SS to BS), the BS determines the number of time slots that each SS will be allowed to transmit in an uplink sub-frame. This information is broadcasted by the BS through the uplink map message (UL-MAP) at the beginning of each frame. UL-MAP contains information element (IE), which includes the transmission opportunities, i.e. the time slots in which the SS can transmit during the uplink sub-frame. After receiving the UL-MAP, each SS will transmit data in the predefined time slots as indicated in IE. The BS uplink-scheduling module determines the IEs using bandwidth request PDU (BW-request) sent from SSs to BS.

5

QoS Provision Mechanisms in WiMax

87

Fig. 5.1 Existing QoS architecture of IEEE 802.16d

The 802.16d MAC provides QoS differentiation for different types of applications that might operate over 802.16d networks. The 802.16d standard defines the following types of services: Unsolicited Grant Services (UGS): UGS is designed to support Constant Bit Rate (CBR) services, such as T1/E1 emulation, and Voice Over IP (VoIP) without silence suppression. Real-Time Polling Services (rtPS): rtPS is designed to support real-time services that generate variable size data packets on a periodic basis, such as MPEG video or VoIP with silence suppression. Non-Real-Time Polling Services (nrtPS): nrtPS is designed to support nonreal-time services that require variable size data grant burst types on a regular basis. Best Effort (BE) Services: BE services are typically provided by the Internet today for Web surfing. Since IEEE 802.16d MAC protocol is connection oriented, the application first establishes the connection with the BS as well as the associated service flow (UGS, rtPS, nrtPS or BE). BS will assign the connection with a unique connection ID (CID). The connection can represent either an individual application or a group of applications (all in one SS) sending data with the same CID. All packets from the application layer in the SS are classified by the connection classifier based on CID and are forwarded to the appropriate queue. At the SS, the Scheduler will retrieve the packets from the queues and transmit them to the network in the appropriate time slots as defined by the UL-MAP sent by the BS. The UL-MAP is determined by the Uplink Bandwidth Allocation Scheduling module based on the BW-request messages that report the current queue size of each connection in SS.

88

M. Ma and J. Lu

IEEE 802.16d defines the required QoS signaling mechanisms such as BWRequest and UL-MAP, but it does not define the Uplink Scheduler, i.e. the mechanism that determines the IEs in the UL-MAP. IEEE 802.16d defines the connection signaling (connection request, response) between SS and BS but it does not define the admission control process. IEEE 802.16d medium access control, which is based on the concepts of connections and service flows, specifies QoS signaling mechanisms (per connection or per station) such as bandwidth requests and bandwidth allocation. However, IEEE 802.16d standard left the details of the QoS based packet-scheduling algorithms and reservation management that determine the uplink and downlink bandwidth allocation, undefined. IEEE 802.16d PHY gives AMC and the conceptually power control. It also left details of AMC and adaptive power control algorithm, undefined.

5.2 Overview of the Quality of Service in WiMax Network The QoS term can be interpreted in different ways. In general, QoS can be described from two perspectives: user perspective and network perspective. In user perspective, QoS refers to the application quality as perspective by the user. In network perspective, QoS refers the service quality that the network offers to applications or users in term of network QoS parameters that include: latency or delay of packets traveling across the network, reliability of packet transmission and throughput. From the network perspective, the networks’ goal is to provide the QoS services that adequately to meet the users’s needs while maximizing the network resources’ utilization. To achive this goal, the networks analyze the application requirements, manage the network resources and deploy various network QoS mechanisms. QoS parameters quantitatively represent the applications’s QoS requirements. They are: 1. 2. 3. 4. 5.

Throughput Delay Delay jitter Error rate Packet loss rate

Networks may use a combination of QoS services, i.e., per-flow and quantitative, per-class and quantitative. Some networks may include multiple types of QoS services in order to support a wide range of applications. In recently years, QoS support architecture and QoS support algorithms for WiMax system have been proposed [4–36]. They can be classified according to the following taxonomy as shown in Fig. 5.2.

5


89

QoS Support Mechanisms in WiMAX QoS Support Architecture

QoS Support Schemes

Traffic Handling

Bandwidth management

Cross Layer Approach

MAC - Higher layer

Bandwidth/ Resource Reservation

Admission Control

Cross layer APA

Holonomic

Link Adaptation

Congestion Avoidance

Scheduling

Buffer tuning

Traffic Policing

Cross layer

Hierarchical

Channel access

Classification

Explicit congestion notification

Tail-drop

Fig. 5.2 Hierarchical taxonomy of QoS mechanisms that support different level Quality of Service in IEEE 802.16d

5.3 QoS Support Architecture in WiMax Networks

Downstream

Paper [4] proposed an inclusive architecture to support QoS mechanisms in IEEE 802.16 standard as shown in Fig. 5.3. Authors developed some compatible methods for specific modules such as Scheduler, Traffic Shaper, and Request and Grant Manager to optimize Delay, Throughput and Bandwidth Utilization metrics. In paper [5], authors proposed MAC layer cross to network upper layer QoS framework in the downlink mode and uplink mode to provide QoS support in WiMax networks as shown in Figs. 5.4 and 5.5. The proposed cross-layer QoS framework integrated L3 and L2 QoS in the IEEE 802.16 network. Main functional blocks in the framework include: QoS mapping from L3 to L2, Admission control, Fragment Control, and Remapping. Fragment Control handles the data frames from the same IP datagram as a group in L2 operations to reduce useless transmission. Remapping is designed for more flexible use of L2 buffers by changing the mapping rules from IP QoS to L2 service type under congested situation of the rtPS queue.

Base Station

Upstream Upstream Analyzer

Downstream Analyzer Connection Establishment Request

Data

Service Specification

Total Grant

UpLink Service Flow Data Base

Data

Subscriber Station

Call Admission Control

Classifier

Bandwidth Request

Shaper and Policer

UpLink Service Flow Data Base

DownLink Service Flow Data Base

Grant Allocator UGS

rPS

nrtPS

Classifier

Shaper and Policer

UGS

BE

rPS

nrtPS

BE

Polling Manager Downlink Scheduler Uplink Scheduler

Request Generator Connection Requests

Request Selector

UP Stream Generator

Upstream

Fig. 5.3 The proposed architecture to provide QoS in IEEE 802.16 standards

Downstream Generator

Downstream

90

M. Ma and J. Lu

Fig. 5.4 Cross-layer QoS framework in the downlink mode in WiMax

Fig. 5.5 Cross-layer QoS framework in the uplink mode in WiMax

Paper [6] proposed cross-layer design frameworks for 802.16e OFDMA systems that are compatible with WiBro based on various kinds of cross-layer protocols for performance improvement: a cross-layer adaptation framework and a design example of primitives for cross-layer operation between its MAC and PHY layers as shown in Fig. 5.6. In the proposed model, the MAC layer contains a user grouper, scheduler, and resource controller. Each functional entity exploits physical layer information to increase system throughput. The physical layer consists of a diversity channel PPDU controller, AMC channel PPDU controller, control information controller, and HARQ functional blocks. AMC subchannel users and diversity subchannel users are classified by the user grouper. Since the properties of AMC subchannels and diversity subchannels are quite different, the grouping of users into two channel types is essential if system throughput is to be increased.

5


91

Fig. 5.6 Cross layer adaptation scheme for efficient resource allocation in WiMax

The scheduler determines the scheduling of users and the quantity of packets that should be scheduled in the current frame. For cross-layer optimization, the scheduler should be designed to exploit not only PHY information but also application layer information.

5.4 Bandwidth Management QoS Mechanisms in WiMax Networks Bandwidth management mechanisms are mechanisms that manage the network resources by coordinating and configuring network devices’ traffic handling mechanism. The main mechanisms are: 1. Resource reservation 2. Connection admission control 3. Cross layer approach bandwidth management mechanisms

92

M. Ma and J. Lu

5.4.1 Resource Reservation Mechanisms Resource reservation mechanisms inform the network entities on the QoS requirements of the applications using the network resources. The network devices will use this information to manage the network resources in order to meet such requirements. The resource reservation mechanisms include the following functions: 1. provision of resource reservation signaling that notifies all devices along the communication path on the mulitimedia application’s QoS requirements. 2. Delivery of QoS requirements to the connection admission control mechanism that decides if there are available resoureces to meet the new request QoS requirements. 3. Notification of the application regarding the admission result. The representive proposal tailored for WiMax is Dynamic Resource Reservation (DRR) scheme [7] as shown in Fig. 5.7. The basic principle of DRR [7] is: the reserved bandwidth will vary between a minimum and a maximum value as per the bandwidth utilized by the clients. The proposal is able to optimize reservation and utilization of bandwidth for Committed Bandwidth (CB) type traffic. However, it is very difficult to select parameters Cm and T. And the fluctuations of the reserved bandwidth from CM to Cm will increases signaling costs. Fig. 5.7 The amount of reserved bandwidth varies according to the number of active flows

5.4.2 The Connection Admission Control Mechanism Admission control is a network Quality of Service (QoS) procedure. Admission control determines how bandwidth and latency are allocated therefore need to be implemented between network edges and core to control the traffic entering the network. The role of CAC is to control the number of connection flows into the network. A new connection request is progressed only when sufficient resources are available at each successive network element to establish the connection through the whole network based on its service category, traffic contract, and QoS, while

5


93

the agreed QoS of all existing connections are still maintained. Admission control is useful in situations where a certain number of connections may all share a link, while an even greater number of connections causes significant degradation in all connections to the point of making them all useless such as in congestive collapse. The proposed CAC schemes [8–13, 30] can be classified mainly into two strategies. The first one based on service degradation. It would consist in gracefully degrading existing connections to make room for the new one [8–13]. The second strategy is without degradation strategy. It would maintain the QoS provided for ongoing connections and simply reject the new service flow if no sufficient resources are available [30]. This first category of CAC schemes include all the CAC algorithms based on service degradation [8], bandwidth borrowing [9–12], or bandwidth stealing [13] strategies. The main idea of these policies is to decrease—when necessary and possible—the resources provided to ongoing connections in order to be able to accept a new service flow. The strategy could be combined by a threshold-based capacity sharing approach in order to avoid starvation [13] or a guard channel strategy that reserves a dedicated amount of bandwidth for more bandwidth-sensitive flows [12]. The second category [30] has no graceful service degradation of existing connections to accept a new flow. Thus, a new connection is accepted only if (1) it will receive QoS guarantees in terms of both bandwidth and delay—for real-time flows—and (2) the QoS of existing connections is maintained. Paper [11] proposed R-CAC which is characterized by defining thresholds of allocated bandwidth at BS for each class of service based on the priority of the service. The proposal introduced two parameters, denoted as Cu and Cr as shown in Fig. 5.8. Cu is the bandwidth exclusively reserved for UGS service, which is allocated with the highest priority in the four service classes supported in IEEE 802.16d networks. Cr is the bandwidth exclusively reserved for UGS and rtPS to mitigate the bandwidth competition coming from the other two types of service. The residual bandwidth (i.e., C – Cu – Cr) is the only part that a BS can assign to the nrtPS connection requests to meet their minimum bandwidth requirements, while it can also be assigned to UGS and rtPS services.

C: total bandwidth of the system Cu: Bandwidth that only can be used by UGS connections Cr: Bandwidth that only can be used by UGS and rtPS connections C – Cu – Cr: bandwidth that can used by UGS, rtPS and nrtPS connections

Fig. 5.8 Proposed R-CAC bandwidth allocation scheme

94

M. Ma and J. Lu

D-CAC [12] is a Priority Support-based CAC. The proposed scheme can give the highest priority for UGS flows and maximizes the bandwidth utilization by bandwidth borrowing and degradation. Token bucket based CAC [13] is to control each connection. The connection is controlled by two token bucket parameters: token rate ri (bps) and bucket size bi (bits). When a traffic flow wants to establish a connection with BS, it sends these two parameters to BS and waits for response from BS. An extra parameter, delay requirement di , will be sent by rtPS flow. The proposed call admission control algorithm is detailed as followings: Step 1. Calculate the remainder uplink capacity Cremain : Cremain = Cuplink -CUGS CrtPS -CnrtPS -CBE . Step 2. Compare Cremain to the bandwidth requirement of the new connection. If there is enough capacity, accept it. If not, go to Step 3. Step 3. First look at the connections that belong to lower classes than this new connection. If there is a class that uses more capacity than its threshold, calculate CL, which means how much capacity can be stolen from it. If the sum of CL and Cremain is greater than or equal to the bandwidth requirement of this new connection, accept it. If not, look at if the capacity occupied by the class of the new connection is less than its threshold. If not, deny it. In order to avoid starvation of some classes, the proposal suggested that a threshold is set for each class. Combining with a proper scheduler, Token bucket based CAC can reserve bandwidth needed by real-time flows and thus delay requirements of rtPS flows can be promised. Other CAC schemes like Binary search approach fairly allocating bandwidth CAC [14] used Gaussian model for aggregated traffic in large network and Chernoff bound method to obtain upper bound blocking probability. Based on the analysis result, binary search approach was applied to solve the problem that given a total bandwidth, fairly allocating bandwidth to each class of multimedia traffic in 802.16d MAN. The total bandwidth (C) is completely portioned for four classes of traffic, and the partition value Ci(i = 1, 2, 3, 4) is calculated by above binary search algorithm using Chernoff bound. If a new connection of UGS and BE arrives at SS, it send request to BS for bandwidth. It is granted bandwidth if the new aggregated bandwidth including this connection is less than Ci(i = 1, 2, 3, 4). Else, it is blocked. If a new connection of rtPS and nrtPS arrives at SS, the connection is always admitted, but the burst within the connection will be blocked when the total used bandwidth larger than Ci(i = 2, 3). This mechanism can guarantee the pre-required upper bound blocking probability for UGS and BE connections, and the burst blocking probability for rtPS and nrtPS.

5.4.3 Cross Layer Approach Bandwidth Management Mechanism There is a few prior works that deals with MAC cross to the higher layers scheduling, i.e. the MAC to application and transport layers. Paper [15] aims at providing

5


95

end to end QoS guarantee using IntServ and DiffServ in connection oriented WiMax PMP and mesh networks. The work maps the RSVP in the IP layer and DSA/DSC/DSD in MAC layer. The author proposed that message exchange for DSA and DSC can be deployed to carry QoS parameters of IntServ services for end-to-end resource (bandwidth/buffer) reservation. For DiffServ services, on the other hand, a number of per-hop behaviors (PHBs) for different classes of aggregated traffic can be mapped into different connections directly. When a new application flow arrives in IP layer, it is firstly parsed according to the definition in PATH message (for InteServ) or Differentiated Services Code Point (DSCP for DiffServ); then classified and mapped into one of four types of services (UGS, rtPS, nrtPS or BE). The proposed dynamic service model in SS sends a request message to the BS, where the admission control determines whether this request is approved or not. If it is not approved, the service module informs the upper layer to deny this service; else the admission control notifies the scheduling module to make a provision based on the parameter values in the request message. At the same time the accepted service is transferred into a traffic grooming module. According to the grooming result, the SS will send Bandwidth Request message to BS. The centralized scheduling module in BS will retrieve the requests and generate UL-MAP message carrying the bandwidth allocation results. Finally, the SS will package SDUs from IP layer into PDUs and upload them in its allocated uplink slots to BS. Figure 5.9 shows the PATH, RESERV messages for IntServ. Here the sender sends a PATH message including traffic specification (TSpec) information. Parameters such as max/min bounds for bandwidth, delay and jitter is mapped into parameters of the DSA message such as Maximum Sustained Traffic Rate, Minimum Reserved Traffic Rate, Tolerated Jitter and Maximum Latency. According to the response received for the DSA message, the provisioned bandwidth can also be mapped into Reserved Specification (RSpec) in the RESV message. These parameters can be used at the MAC scheduler to allocate resources according to the QoS requirements of flows along their path.

Fig. 5.9 Traffic classification and mapping for IntServ services

96

M. Ma and J. Lu

5.5 Traffic Handling Mechanisms QoS Support in WiMax Networks Traffic handling mechanisms are mechnism that classsify, handle, police and monitor the traffic across the network. The main mechanisms are: 1. 2. 3. 4. 5. 6. 7.

Classification Channel access Traffic policing Buffer management Congestion avoidance Packet scheduling Cross layer APC

5.5.1 Traffic Classification The classification mechanism identifies and separates different traffic into flows or group of flows. Therefore each flow or group of flows can be handled differently. Application traffic is identified by the classification mechanism and is forwarded to the appropriate queue awaiting service from other mechanism such as traffic shaping and packet scheduling. The granularity level of the classification mechanism can be per-user, per-flow or per-class depending on the type of QoS services provided. To identify and classify the traffic, the traffic classification mechanism requires some form of tagging or marking of packets. Paper [16] proposed two- stage packet classification algorithm. Authors suggested that prefix-based fields and range-based fields should be processed separately in two stages. In the first stage, packets are classified by a matching scheme with their prefix-based fields, while other range-based fields are processed by a rangebased scheme in the second stage. The Prefix-Matching-Tree (PMT) used in the first stage to handle prefix-based fields. The PMT is constructed by a prefix-based matching scheme that can speedup the searching process by simultaneously processing multiple prefix-based fields. Each tree node of the first stage is connected to the second stage according to protocol types (TCP or UDP). The Range-MatchingTree(RMT) is employed to deal with range-based fields in the second stage.

5.5.2 Channel Access Mechansim In wireless networks, all hosts communicate through a shared wireless medium. When multiple hosts try to transmit packets on the shared communication channel, collisions can occur. Therefore, wireless networks need a channel access mechanism which controls the access to the shared channel.collision-based chanael such as Random access and collision-free channel access such as TDMA or Polling channel access mechanism can provide different QoS services.

5


97

In IEEE802.16d, there are two access protocols based on multi-channel slotted Aloha and periodic polling. The former is a contention based access protocol, while the latter provides periodic polling services without contention, such as unsolicited grant service (UGS), e.g., Voice over IP and real time or non-real time services. In addition, IEEE802.16d allows unicast-polling for a single SS or multicast-polling for groups of SSs as periodic polling service. In paper [17], the performance of IEEE802.16d Random Access Protocol is evaluated by using Transient Queueing Analysis. The paper derived the random access success probability from the system equilibrium. Retransmission probability is also derived by including a binary exponential backoff algorithm. In paper [18], authors considered a capacity allocation scheme of periodic polling services for a multimedia traffic in an IEEE802.16d system. Considering a base station assigns a subscriber station contiguous M uplink subframes for uplink traffic transmission and v vacation frames for saving power and opportunities for other subscriber stations, they proposed a capacity allocation scheme for a multimedia traffic in WiMax network. The bandwidth allocated to an SS will be returned to a BS, when the SS’s queue is empty during M contiguous uplink subframes. The returned bandwidth will be allocated to other types of services requested from the multichannel slotted Aloha.

5.5.3 Traffic Policing The traffic policing is the mechanism that monitors the admitted sessions’ traffic so that the sessions do not violate their QoS contract. The traffic policing mechanism makes sure that all traffic that passes through it will confirm to agreed traffic parameters. In case vialation is found, a traffic poling mechanism is enforced by shaping the traffic and dropping traffic to enforce compliance with that contract. Traffic sources which are aware of a traffic contract sometimes apply Traffic Shaping in order to ensure their output stays within the contract and is thus not dropped. Due to fact that traffic policing shapes the traffic based on some known quantative traffic parameters, multimetedia (real-time) application are naturally compatible to traffic policing. Most multimedia application traffic (Voice, video) is generated by a standard codec which generally provides certain knowledge of the quantitative traffic parameters. Traffic policing can be applied to individual multimedia flows. Non-real time traffic does not provide quantitative traffic parameters and usaually demands bandwidth as much as possible. Therefore, traffic policing enforces non-real time traffic based on network policy. Such policing is usually enforced on aggregated non-real time traffic flows. Traffic policing, in cooperation with other QoS mechanisms usually, can provide QoS support. In [19], a new traffic shaping based on concept of “Fair Marker”(FM) was proposed to enforce fairness among distinct flows. FM controls token distribution from the token bucket to the flows originating from the same subscriber network. The FM explores the duality between packet queueing and token bucket utilization. Fairness in token distribution is a function of the fair allocation algorithm used by FM. In

98

M. Ma and J. Lu

order to reach this purpose, it records information regarding the consumption of tokens by the monitored flows.

5.5.4 Buffer Management Buffer management refers to any particular discipline used to regulate the occupancy of a particular queue where packets may be held (or dropped). Buffer is set to improve link utilization and system performance, but it also increases packet’s queue delay. With the increase of user demands for service quality, providing stable and low delay has been the primary requirement of real-time services. The most important and easy to control part of total delay is queue delay. So how to set the capacity of the buffer, how to efficiently control buffer length while network circumstance is dynamic, and how to achieve the tradeoff between throughput and queue delay are the important problems to be solved in buffer management and QoS control of whole networks. At present, support [20] for wired and wireless network is included for drop-tail (FIFO) queueing, Random Early Detection (RED) buffer management, class based queueing (CBQ) (including a priority and round-robin scheduler), and variants of Fair Queueing including Fair Queueing (FQ), Stochastic Fair Queueing (SFQ), and Deficit Round-Robin (DRR). RED is the most intensive researched class of AQM (Active Queue Management), it selectively discards packets of some flows based on probability determine. In WiMax networks, the base station (BS) is a likely bottleneck for downlink (DL) TCP connections due to difference in available bandwidth between the fixed network and the wireless link. This may result in buffer overflows or excessive delays at the BS, as these buffers are connection-specific. In order to avoid buffer overflows, different Active Queue Management (AQM) methods may be applied at the BS. Paper [21] analyzed of RED; Packet Discard Prevention Counter (PDPC) and time-to-live based RED AQM mechanisms and proved that they are indeed very useful: AQM reduces considerably DL delays at the WiMax BS without sacrificing TCP throughput.

5.5.5 Congestion Control Congestion control concerns controlling traffic entry into a network, so as to avoid congestive collapse by attempting to avoid oversubscription of any of the processing or link capabilities of the intermediate nodes and networks and taking resource reducing steps, such as reducing the rate of sending packets. In congestion control, the packet loss information can serve as an index of network congestion for effective rate adjustment, but in a wireless network environment, common channel errors due to multipath fading, shadowing, and attenuation may cause bit errors and packet loss, which are quite different from the packet loss caused by network

5


99

congestion. Therefore wireless packet loss can mistakenly lead to dramatic performance degradation. If the link capacity in one system is temporarily degraded due to a high traffic load in the WLAN system (congestion) or due to interference, two effects, leading to inefficiency on the IEEE802.16d link, could occur:

r r

Loss of data due to buffer overflow and therewith unnecessary retransmissions. Waste of bandwidth due to unused reserved transmission opportunities.

To avoid the above-mentioned effects, a congestion control mechanism needs to be worked out to dynamically adapt the QoS demands of a connection during runtime for a specifically defined period of time. Paper [22] proposed Dynamic Service Change (DSC) congestion control mechanism to support the Explicit Congestion Notification mechanism in future deployment of TCP. The paper contributed to IEEE 802.16-REVd D1 standard by adding MAC DSC TEMP.request and MAC DSC TEMP.indication message in MAC entity to provide the temporary traffic reduction mechanism to overcome congestion effect. Since network congestion is directly related to the congestion packet loss, packet loss can be caused by either congestion loss or wireless channel errors, resulting from mulitipath fading, shadowing, or attenuation. New approach of congestion control over wireless network is to perform packet loss classification so that congestion control algorithms can more effectively adapt the sending rate based on congestion loss instead of from wireless loss. Paper [23] considered two packet loss classes, congestion loss and wireless loss and proposed packet loss classification (PLC) method based on the trend of ROTT (relative one-way trip time) to assist packet loss classification in the ambiguous area of ROTT distribution. By taking advantage of the QoS features offered by one of the four proposed WiMax service flow arrangement, paper [24] aimed at more flexible layer constructing and subscription while reliable in diverse channel conditions and fitting users’ demand. Through effective integration of packet loss classification (PLC) [23], endto-end available bandwidth probing, congestion control via layered structure and packet level FEC, for layered multicast applications over WiMax for disseminating scalable extension of H.264/AVC compressed video is proposed. The optimality comes from the best tradeoff of number of video layers subscription with number of additional FEC packets insertion simultaneously to satisfy the estimated available bandwidth and wireless channel error condition.

5.5.6 Packet Scheduling Algorithm Packet scheduling refers to the decision process used to choose which packets should be serviced or dropped. Packet scheduling is the process of resolving contention for bandwidth. A scheduling algorithm has to determine the allocation of bandwidth among the users and their transmission order. One of the most important

100

M. Ma and J. Lu

tasks of a scheduling scheme is to satisfy the Quality of Service (QoS) requirements of its users while efficiently utilizing the available bandwidth. Many legacy scheduling algorithms, capable of providing certain guaranteed QoS, have been developed for wireline networks. However, these existing service disciplines, such as fair queueing scheduling, virtual clock, and EDD, are not directly applicable in wireless networks because they do not consider the varying wireless link capacity and the location-dependent channel state. The characteristics of wireless communication pose special problems that do not exist in wireline networks. These include: 1) 2) 3) 4) 5)

high error rate and bursty errors; location-dependent and time-varying wireless link capacity; scarce bandwidth; user mobility; and power constraint of the mobile hosts.

All of the above characteristics make developing efficient and effective scheduling algorithms for wireless networks very challenging. WiMax networks provide services for heterogeneous classes of traffic with different quality of service (QoS) requirements. Currently, there is an urgent need to develop new technologies for providing QoS differentiation and guarantees in WiMax networks. Among all the technical issues that need to be resolved, packet scheduling in WiMax networks is one of the most important. In this sub-section, we assess proposed scheduling algorithms for QoS support in WiMax networks thoroughly with respect to the characteristics of the IEEE 802.16d MAC layer and PHY layer. We classify those scheduling algorithms in WiMax into three categories: holonomic approach [25–28]; hierarchical approach [29–34] and cross-layer approach [35–38] with respect to the nature of scheduling algorithm mechanism as holonomic approach uses single layer scheduling scheme, in contrast, hierarchical approach uses several layers or stages scheduling schemes and cross layer approach uses information from several layers in OSI model. Each of three categories can be classified as per–flow, per-class, per-packet and hybrid scheduling algorithms. Representative schemes in each of these categories will be discussed next. 5.5.6.1 Holonomic Packet Scheduling Algorithm Paper [25] applied holonomic approach and proposed one layer hybrid scheduling scheme which combines per-flow and per-class scheduler termed as “Frame Registry Tree Scheduler” (FRTS). The proposal aims at providing differentiated treatment to data connections, based on their QoS characteristics. This approach focuses on properly preparing future transmitted frames by using a tree based approach. The tree consists of six levels; root, time frame, modulation, subscriber, QoS service and connection level. First level is taken to be the root. The second level represents time frames immediately after the current time frame. The third level represents the available modulation types. The fourth level organizes all the connections according

5


101

to the SS each SS has one uplink node and one downlink node at this level. The fifth level organizes the connections according to their QoS. The last level consists of leaves for each active connection queue. The data structure presented achieves time frame creation and reduces the processing needs at the beginning of each frame. The algorithm schedules each packet at the last time frame before its deadline. This allows more packets to be transmitted and hence an increased throughput. This method also avoids fragmentation of transmissions to/from the same SS or same modulation. Another good feature is its ability to handle changes in the connection characteristics like modulation type or service type of the channel. In [26], the authors proposed a Token Bank Fair Queueing (TBFQ) scheduling algorithm at packet level for BWA systems as shown in Fig. 5.10. TBFQ uses the priority index Ei/ri to keep track of the normalized service received by backlogged flows. Ei is the number of tokens exchanged between the ‘bank’ and the flow i. Ei is negative if the flow continues to borrow tokens from the bank to serve the traffic that exceeds its average rate. Ei is positive if the traffic is below its average rate. The flows are first served based on their token generation rate to guarantee the throughput and latency, and then the remaining bandwidth is distributed based on their priority index. The parameters (debt limit, credit burst, and creditable threshold) determine the dynamic behavior of the algorithm. TBFQ can be adapted to operate under varying channel error conditions. It has demonstrated its effectiveness in achieving fairness, maximizing channel utilization, and fast convergence to guaranteed throughput. Its ability to serve and isolate real-time traffic and data traffic that is under severe error conditions makes it a suitable candidate as the wireless packet scheduling algorithm for BWA systems.

ri P1

λ1

Ei

D1 r2

Token Denk Size B

P2

λ2

E2

D2 rn Wireless Terminals

Pn λ3

En

D3

Fig. 5.10 TBFQ Algorithm

102

M. Ma and J. Lu

In order to maximize throughput of non-real-time traffics with satisfying QoS requirements of real-time traffics, paper [27] proposed urgency and efficiency based packet scheduling (UEPS) algorithm which was designed not only to support multiple users simultaneously but also to offer real-time and non-real-time services to a user at the same time. The UEPS algorithm uses the time-utility function as a scheduling urgency factor and the relative status of the current channel to the average one as an efficiency indicator of radio resource usage. The proposed packet scheduler assigns priorities to the packets to be transmitted, based on the channel status reported by the user equipments as well as the QoS statistics maintained by the BS. Since the scheduler works in a global timeline, a time utility function (TUF) is used for the scheduling. Two scheduling factors, the urgency of scheduling and the efficiency of radio resource usage, are used to schedule RT and NRT traffic packets at the same time. The TUF is used to represent the urgency of scheduling while the channel state is used to indicate the efficiency of radio resource usage. There are three steps in the UEPS scheduling algorithm. In the first step, the incoming packets are stored in the buffer corresponding to the SS and the traffic type. The QoS profiles of each arrived packet, such as the arrival time, deadline, packet type, head of line (HOL) delay, and packet size are also maintained. In the next step, at each scheduling instance, the urgency factor of each HOL packet of each buffer is calculated from the TUF. The TUF of a real-time traffic is a straight line up till a threshold with a dead drop. Because it is a hard and discontinuous function in delay, the unit change of utility can not be obtained directly at its delay time. Thus a continuous TUF having straight line up till a threshold with a gradual z-shaped drop is used instead. The gradual decay occurs in the marginal scheduling time interval (MSTI) which is a small time window of delay jitters around the deadline. This relaxed z-shaped function is given by U RT (t) = e−a(t−c) /(1+e−a(t−c) ) where a and c are the parameters which determine the slope and location of the point of inflection. The unit change of utility of a real time traffic at the inflection point (t = c) is given by |U ’ RT (t = c)| = a/4, and this value forms the urgency factor. For non real time traffics, the TUFs are monotonic decreasing functions in time (delay). TUF of a non-real time traffic is f N RT (t) = 1 − ex p(at)/ex p(D) where D is the maximum time used for normalization purpose. The resulting urgency function is |U ’N RT (t)| = aex p(at)/ex p(D). Once the urgency has been calculated, the highest urgency factor for an SS out of the four buffers for that SS is taken as the representative for that user. The efficiency factor of each SS is then calculated. It is a moving average of the channel state of the SS, given by R(t) = (1 − 1/W )R’(t − 1) + (1/W )R(t), where R(t) is the channel state at time t, R’(t) is the scheduling priority value of each SS is calculated by p(t) = U ’(t)∗ (R(t)/R’(t)). As a last step, based on the priorities calculated for each SS, the top n number of packets having the highest priorities are transmitted, where n is the number of different streams the OFDMA can send simultaneously. Paper [28] proposed a new per-flow based scheduling algorithm, referred as Service Criticality (SC) based scheduling scheme. SC is based on a dual of buffer occupancies at nodes and allowable latencies for a particular service. In the proposed

5


103

scheme, a flow would receive the service through bandwidth allocation depending on degree of service criticality (SCindex ) computed at SS and conveyed to BS during bandwidth request burst of UL. The core of the proposed SC scheme is the way SS computes the SCindex for active flows existing at node. The computation of SCindex depends on both, the buffer occupancy and latency experienced by flow. Proposed scheme employs linear dependence on buffer occupancy and tunable sigmoid like relation with experienced latency to compute SCindex . SCindex encompasses two service parameters, maximum allowable latency and permissible packet loss and it is derived from two elements, referred to as Service Desperation (SD) and Buffer Occupancy (BO). SD denotes urgency of bandwidth allocation or inversely denotes the amount of time a flow (service) can tolerate not being provisioned. BO denotes current level up to which flow buffer is full with respect to allocated buffer. Both of these elements are normalized over all the different flows in the system. The normalization process ensures fairness amongst multiple flows. Normalization also implies that SD and BO denote a ratio between 0 and 1. The maximum of SD and BO is chosen as SCindex such that when SCindex is close to 1 the flow must be provisioned, else it will be timed out. 5.5.6.2 Hierarchical Scheduling Scheme In paper [29], the issue of differentiated service provisioning will be addressed with the non-real-time polling service in WiMax systems. The proposed solution has been designed to have an ability to accommodate integrated traffic in the networks with effective scheduling schemes. A hierarchical scheduling algorithm to provide service differentiation to enhance the nrtPS service in WiMax systems was proposed. In order to meet the time constraints of real-time messages as much as possible and avoid scheduling starvation of the non-real-time messages, the data transmission scheduling has been divided into two levels, inter-class scheduling and intra-class scheduling as shown in Fig. 5.11. With the objective to provide service differentiation between the real-time and non-real-time classes of traffic, the proportional delay differentiation (PDD) model was proposed as inter-class scheduling algorithm. Intra-class scheduling scheme was designed as a priority assignment scheme based on tardy rate, the message with less transmission time will have higher priority to be served. The message which

Fig. 5.11 Structure of two levels hieratical scheduling scheme

104

M. Ma and J. Lu

Fig. 5.12 Hierarchical structure of bandwidth allocation

has been delayed a longer time in the queue will also have higher priority to be transmitted. With this scheme, the total time for messages to stay in the network could be reduced and the starvation of the messages with longer transmission time could be avoided. Paper [30] proposed a hybrid scheduling algorithm that combines EDF, WFQ and FIFO scheduling algorithms. The overall allocation of bandwidth is done in a strict priority manner. EDF scheduling algorithm is used for SSs of the rtPS class, WFQ is used for SSs of the nrtPS class and FIFO for SSs of the BE class. In paper [31], authors enhanced the proposal scheduling architecture in [30], the scheduling architecture is similarly divided into two layers as shown in Fig. 5.12. The first layer is for bandwidth requests. The authors suggested a Deficit Fair Priority Queue (DFPQ) for scheduling at this layer. The second layer scheduling is for the data traffic. UGS are not scheduled because they already have a reserved bandwidth. For the other three traffic classes, a hybrid of scheduling algorithm is proposed. The authors suggested Earliest Deadline First (EDF) for rtPS traffic, where the packets with earliest deadline are scheduled first. For nrtPS, WFQ is proposed. The bandwidth left is allocated to each BE traffic in a round robin (RR) manner. Compared with fixed bandwidth allocation, the proposed solution [31] improves the performance of throughput under unbalanced uplink and downlink traffic. And better performance in fairness can be achieved by the proposed DFPQ algorithm than strict PQ scheduling. Paper [32] extended the scheduling architecture in paper [31]. It proposed a preemptive DFPQ scheduling algorithm that enhances the DFPQ algorithm proposed in [31], and improves the performance of the rtPS service flow class. The Preemptive DFPQ defines for each non-preemptive queue a Quantum Critical (Q crit ) to give the queue another chance to service critical packets. Q crit value is a percentage of the original value of the queue’s quantum. Queues are allowed to use Q crit to serve critical packets only. The processing of critical packets continues until the Q crit of the non-preemptive queue becomes less than or equal to zero. Q crit is initialized

5


105

only once per frame and not at every round like the quantum Q. This algorithm gives more chances to rtPS packets to get serviced before the expiration of their deadlines. The authors in [33] proposed an architecture consisting of three schedulers. The first scheduler is concerned with UGS and rtPS flow, as well as rtPS and nrtPS polling flow. An EDF scheduling is applied in this scheduler. The second scheduler is concerned with flows requiring a minimum bandwidth mainly nrtPS. WFQ scheduling is used here where the weight is the size of the requested bandwidth. The third scheduler is used for BE traffic and here too a WFQ scheduling is employed where the weight is the traffic priority. Among the schedulers, the first level has the highest priority, and only after all the packets have been served, is the second scheduler considered. The third scheduler comes when the first two have become free. The delay and delay jitter character for UGS, rtPS, nrtPS and BE can be improved simultaneously using the proposed architecture. In Paper [34] authors considered a scheduling algorithm should possess the following features: efficient link utilization, bounded delay, enough fairness, high throughput, low implementation complexity, graceful performance degradation, strong isolation, more delay/bandwidth decoupling, and flexible scalability and proposed solution follows the framework specified in the IEEE 802.16d standard with the following unique features as shown in Fig. 5.13. (1) The Pre-scale Dynamic Resource Reservation (PDRR) scheme is proposed to allocate bandwidth to UL subframe and DL subframe dynamically. (2) The Priority-based Queue Length Weighted (PQLW) scheduling algorithm was proposed for inter-class scheduling and the Max-Min Fair Sharing (MMFS) scheduling was introduced for inter-SS scheduling within each class of service at the BS as Tier 1 scheduling. (3) The Self-Clocked Fair Queuing (SCFQ) and Weighted Round Robin (WRR) scheduling schemes have been applied to inter-connection scheduling within each class of service at each SS as Tier 2 scheduling. (4) Earliest Deadline First (EDF) and Shortest Packet Length First (SPLF) scheduling have been applied to packet scheduling within each of the connections carrying burst traffic as Tier 3 scheduling. 5.5.6.3 Cross-Layer QoS Scheduling for WiMax Networks Wireless communication systems have unique characteristics – namely, timingvarying channel conditions and multi-user diversity. New MAC design and new scheduling solutions need to be developed that are specifically tailored for this Wireless communication environment [35]. Opportunistic MAC (OMAC) is the modern view of communicating over spatiotemporally varying wireless link. The cross-layer nature embeds OMAC with the potential to revolutionize the design of wireless data networks from physical to data link layers. The wireless resources (bandwidth and power) are more scarce and expensive than their wired counterparts, because the overall system performance degrades dramatically due to multi-path fading, Doppler, and time-dispersive effects caused by the wireless air interface. Unlike wired networks, even if large bandwidth/power is allocated to a certain wireless connection, the loss and delay requirements may not

106

M. Ma and J. Lu

Fig. 5.13 Proposed 3-tier scheduling framework

be satisfied when the channel experiences deep fades. In this scenario, the scheduler plays extremely important role. An OMAC seeks to pick among competing users the one who is currently experiencing the relatively best channel conditions in each scheduling instant. The judicious schemes should be developed to support prioritization and resource reservation in wireless networks, in order to enable guaranteed QoS with efficient resource utilization. Cross-layer MAC designs tailored for WiMax have been proposed [36–38]. Some researchers have considered cross layer scheduling using the MAC scheduler and the PHY resource allocator. In paper [36], authors proposed a cross layer design of packet scheduling and resource allocation in OFDMA wireless networks which concentrates on downlink scheduling in the BS. In OFDMA systems, each carrier is subdivided into a number of sub-carriers which can be controlled or allocated separately making the PHY very robust. In the proposed system, the sub carriers are grouped into allocation units (AU) for allocation so that overheads are minimized. The BS estimates the sub channel condition of each user and allocated resources to the users on a frame-by-frame basis. The authors considered a system where some SS have real time traffic and other systems have non-real time traffic, i.e. real time and non-real time traffic do not co-exist in the same SS. The BS can estimate the channel gain of each user on a sub channel and an AU can be independently allocated to a particular user. The aim is to maximize the overall utilization while satisfying the rate requirements of individual users. To do this, author in [36] formulates a linear programming method and subsequently obtain a sub-optimal solution which reduces the computational time. The scheme also proposes a packet scheduler

5


107

at the BS MAC layer which provides equal priority to both real time and non-real time traffic except if the ratio of the waiting time of a real time packet in the MAC queue and the delay constraint of the packet exceeds a certain threshold. Associated with cross-layer packet scheduler, the paper [36] proposed a 3-stage resource allocation algorithm where the first stage schedules the urgent packets. The second stage schedules the non-real time packets and the real-time packets which are not urgent, higher priority is given to users whose channel quality is better regardless of the traffic type. The third stage allocates the unallocated AU’s if any. After the packet scheduler decides the rate requirements of each user, the actual resources are allocated in the PHY layer. This is carried out by using a sub-optimal heuristic algorithm which first allocates the sub-channels (or subcarriers) constraints in order to maximize the utilization (or minimize the transmission power). After that, the algorithm relocates (or swaps) sub-channels so as to satisfy the rate requirements of each user. Though this kind of cross layer scheme can be followed in the downlink at the BS, it cannot be adopted for uplink scheduling since individual nodes cannot decide the channel condition and place request for those particular channels. A priority-based scheduler [37] was proposed at the medium access control (MAC) layer for multiple connections with diverse QoS requirements, where each connection employs adaptive modulation and coding (AMC) scheme at the physical (PHY) layer. The priority-based scheduler is termed as priority function (PRF) for each connection admitted in the system and update it dynamically depending on the wireless channel quality, QoS satisfaction, and service priority across layers. Thus, the connection with the highest priority is scheduled each time. The connection with the highest priority is scheduled each time. Efficient bandwidth utilization for a prescribed PER performance at the PHY can be accomplished with AMC schemes, which match transmission parameters to the time-varying wireless channel conditions adaptively and have been adopted by WiMax networks. Authors defined Transmission Mode(TM) through the information of adaptive modulation and coding scheme according to wireless channel condition abstracted from IEEE802.16d standard as shown in Table 5.1. To simplify the AMC design, authors approximated the PER expression in AWGN channels as P E Rn (γ ) ≈

1, i f γ ≤ γ pn an exp(−gn γ ), i f

γ ≥ γ pn

(5.1)

where n is the mode index and γ is the received SNR. Parameters an , gn are modedependent. With packet length N b = 128 bytes/packet, the fitting parameters for transmission modes in TM are provided in Table 5.1. AMC design guarantees that the PER is less than or equal to Minimum PER by determining minimum SNR required boundary and updating the transmission mode as in Table 5.1. Authors further developed a cross-layer opportunistic scheduling algorithm termed as priority function (PRF) to schedule rtPS, nrtPS, and BE connections.

108

M. Ma and J. Lu Table 5.1 Transmission modes in the IEEE 802.16D standard

Mode n

1

2

3

4

5

6

Modulation RS Code CC Code Rate Coding Rate Rc Rn (bits/symbol) an (dB) gn γ pn (dB)

QPSK (32,24,4) 2/3 1/2 1.0 232.9242 22.7925 3.7164

QPSK (40,36,2) 5/6 3/4 1.5 140.7922 8.2425 5.9474

16QAM (64,48,8) 2/3 1/2 2.0 264.0330 6.5750 9.6598

16QAM (80,72,4) 5/6 3/4 3.0 208.5741 2.7885 12.3610

64QAM (108,96,6) 3/4 2/3 4.0 216.8218 1.0675 16.6996

64QAM (102,108,6) 5/6 3/4 4.5 220.7515 0.8125 17.9629

Each connection i, where i denotes the connection identification (CID) of rtPS, nrtPS, and BE services, adopts AMC at the PHY. Given a prescribed PER ξ i, the SNR thresholds γ pn (1), for connection i are determined by setting P0 = ξ i. Thus, the possible transmission rate (capacity), i.e., the number of packets that could be carried by Nr time slots for connection i at time t (frame index), can be expressed as Ci(t) = Nr Ri(t)

(5.2)

where Ri(t) is the number of packets that can be carried by one time slot and is determined by the channel quality of connection i via AMC as in Table 5.1. Either Ri(t) or Ci(t) indicates the channel quality or capacity, which is accounted for by the scheduler. At the MAC, the scheduler simply allocates all Nr time slots per frame to the connection i = arg max ϕi(t)

(5.3)

where ϕi(t) is the PRF for connection i at time t, which is specified differently for rtPS and nrtPS and BE service classes. If multiple connections have the same value max {ϕi(t)}, the scheduler will randomly select one of them with even opportunity. For each rtPS connection, the scheduler timestamps each arriving packet according to its arrival time and defines its timeout if the waiting time of such a packet in queue is over the maximum latency (deadline) Ti. The PRF for a rtPS connection i at time t is defined as: ⎧ R (t) 1 i ⎪ ⎨βr t R N Fi (t) , If Fi (t) ≥ 1, Ri (t) = 0 ϕi(t) = βr t , If Fi (t) < 1, Ri (t) = 0 ⎪ ⎩ 0 Ri (t) = 0

(5.4)

ε where βr t ε[0, 1] is the rtPS-class coefficient and Fi(t) is the delay satisfaction indicator, which is defined as: Fi (t) = Ti − ⌬Ti − Wi (t) + 1

(5.5)

5


109

With ⌬Ti ε[0, Ti ] denoting the guard time region ahead of the deadline Ti and W i(t)ε[0, Ti ] denoting the longest packet waiting time, i.e., the head of the line (HOL) delay. If Fi(t) ≥ 1, i.e., W i(t)ε[0, T i − ⌬T i], the delay requirement is satisfied, and the effect on priority is quantified as 1/Fi(i)ε[0, 1]: Large values of Fi(t) indicate high degree of satisfaction, which leads to low priority. If Fi(t) < 1, i.e., W i(t)ε[T i − ⌬T i, T i], the packets of connection i should be sent immediately to avoid packet drop due to delay outage, so that the highest value of PRF βr t is set. Parameter R N = max{Ri(t)}, and the factor Ri(t)/R N ε[0, 1] quantifies the normalized channel quality because high received SNR induces high capacity, which results in high priority. When Ri(t) = 0, the channel is in deep fade and the capacity is zero, so that connection i should not be served regardless of delay performance. The value of ϕi(t) for rtPS connection i lies in [0, βr t ]. For each nrtPS connection, guaranteeing the minimum reserved rate ηi means that the average transmission rate should be greater than ηi. In practice, if data of connection i are always available in queue, the average transmission rate at time t is usually estimated over a window size tc based on (5.3) and (5.4) as: ηˆ i (t + 1) =

ηˆ i (t)/(1 − 1/tc ) ηˆ i (t)/(1 − 1/tc ) + Ci (t)/tc

If i = i ∗ If i = i ∗

(5.6)

In order to guarantee ηi(t) ˆ ≥ ηi during the entire service period. The PRF for an nrtPS connection i at time t is defined as: ⎧ Ri (t) 1 ⎪ ⎪ , If Fi (t) ≥ 1, ⎪βnr t ⎪ ⎪ R N Fi (t) ⎨ Ri (t) = 0 ϕi(t) = (5.7) ⎪ ⎪ , If F (t) < 1, R (t) = 0 β ⎪ nr t i i ⎪ ⎪ ⎩0 Ri (t) = 0 where βnr t ε [0, 1] is the nrtPS-class coefficient, and Fi(t) is the ratio of the average transmission rate over the minimum reserved rate. Fi(t) = ηi(t)/ηi. ˆ

(5.8)

Quantity Fi(t) here is the rate satisfaction indicator. If Fi(t) ≥ 1, the rate requirement is satisfied, and its effect on priority is quantified as 1/Fi(t)ε [0, 1]. If Fi(t) < 1, the packets of connection i should be sent as soon as possible to meet the rate requirement; in this case, the upper-bound value βnrt is set for ϕi(t). Once again, the value of ϕi(t) lies in [0, βnrt]. For BE connections, there is no QoS guarantee. The PRF for a BE connection i at time t is φi (t) = β B E

Ri (t) RN

(5.9)

110

M. Ma and J. Lu

where βBEε [0, 1] is the BE-class coefficient. Notice that ϕi(t) varies in [0, βBE], which only depends on the normalized channel quality regardless of delay or rate performance. The role of βrtPS, βnrtPS, and βBE is to provide different priorities for different QoS classes. For example, if the priority order for different QoS classes is rtPS > nrtPS > BE, the coefficients can be set under the constraint βrtPS > βnrtPS > βBE; e.g., βrtPS = 1.0 > βnrtPS = 0.8 > βBE = 0.6. Thus, the QoS of connections in a high-priority QoS class can be satisfied prior to those of a lowpriority QoS class because the value of ϕi(t) for QoS unsatisfied connections will equal the upper-bound βrtPS, βnrtPS, and βBEforrtPS, nrtPS, and BE connections, respectively. The purpose of normalizing ϕi(t) in [0, βrtPS], [0, βnrtPS], and [0, βBE], respectively, is to provide comparable priorities among connections with different kinds of services, which enable exploiting multiuser diversity among all connections with rtPS, nrtPS, and BE services. In paper [38], authors enhanced the proposal in [6] by proposing an optimal resource allocation scheme and an opportunistic scheduling rule similar as in paper [37] that can satisfy the QoS requirements of the application layer and optimize MAC-PHY cross-layer performance. The main concept of the proposed Optimal Resource Allocation is to solve optimization problem. Assume that a downlink OFDM system with N subchannels and M time slots. There are K users, J connections and L packets in the system. The objective of the resource allocation is to maximize the overall system throughput while guaranteeing the provision of QoS, which is formulated into the following constrained optimization problem:

arg

max

Ci (m,n),g( j)

M N L

Ci (m, n)Ri (m, n)

(5.10)

i=1 m=1 n=1

subject to

Ci (m, n) − 1 ≤ 0,Ci (m, n) ∈ {0, 1}∀m, n

(5.11)

i

Wi ≤ T j ∀i, i → j

Cmin ( j) , P j (t) , ∀ j g( j) ≥ d( j)

(5.12) (5.13)

where Ci (m, n) identifies whether packet i is allocated to slot (m, n) and Ci (m, n) is equivalent data rate packet i can obtain on this slot. Since CSI is usually assumed to keep constant per frame, it simplifies as Ri (n). (5.11) is to ensure one slot can only be allocated to one packet while (5.12) and (5.13) correspond to the QoS requirements in terms of delay and throughput. Here Tj and Cmin ( j) are maximum latency of rtPS connection and minimum reserved data rate of nrtPS connection respectively with d(j) being connection j’s packet length and Pj(t) being number of

5


111

packets present in connection j. Moreover, Wi denotes waiting time of packet i and g(j) is number of connection j’s scheduled packets.

5.5.7 MAC-PHY Cross Layer Approaches with APA and CAC Appropriate power control can not only reduce power consumption, but also improve system capacity by adjusting coverage ranges and improve spatial reuse efficiency. Higher transmission rates require higher SINR to maintain the same bit error rate. Due to this close relationship between rate and channel condition, incorporating transmission rate selection into MAC design is another promising way to increase the system performance. Paper [39] considered Adaptive Power Allocation (APA) emphasizes how to share the limited power resource of base station among different WiMax subscribers and further influences the access bandwidth of each subscriber; CAC highlights how to assign a subscriber’s access bandwidth to different types of applications. Authors suggested that APA and CAC have to work cooperatively to provide cross-layer resource management to build a WiMax access network. A power allocation scheme that produces optimal revenue and this is known as the optimal revenue criterion was studied. In order to investigate the APA revenue of a certain scheme, the revenue rate of each type of service as the revenue generated by a bandwidth unit was defined. Let rerUGS , rerrtPS , rernrtPS , and rerBE be the revenue rates of the following, respectively:

r r r r

Unsolicited Grant Service Real-Time Polling Service Non-Real-Time Polling Service Best-Effort Service

Obviously, different services have different prices due to their specific QoS requirements. As a result, r er U G S , r er r t P S , r er nr t P S , and r er B E always take distinct values. To maximize the APA revenue, the potential revenue of each subscriber, which is defined as the revenue that could be achieved if all arriving traffic is served, must be investigated. The potential revenue of a given subscriber k is determined by the amount of UGS, rtPS, nrtPS, and BE traffic load in its local network and the price of service (i.e., r er U G S , r er r t P S , r er nr t P S , and r er B E ). Let T Lk D denote the arriving downlink traffic load in subscriber k’s local network, and suppose this traffic load can generate potential revenue R Dk . Then, the revenue-to-bandwidth ratio of the kth subscriber is defined as RBRkD = RDk /TLk D . Since different WiMax subscribers can have different amount of UGS, rtPS, nrtPS, and BE traffics in their local networks, they hold distinct revenue-to-bandwidth ratios. The optimal revenue-criterion-based APA optimization has inherent preference to allocate more power resource to the sub-carriers that belong to the subscriber of high revenue-to-bandwidth ratio.

112

M. Ma and J. Lu

5.6 Summary This chapter presents various QoS support mechanisms in WiMax networks. Existing proposals with state-of-art technology have been classified into three main categories: QoS support architecture; Bandwidth management mechanism and Traffic handling mechanism. Representative schemes from each of the categories have been evaluated with respect to major distinguishing characteristics of the WiMax MAC layer and PHY layer as specified in the IEEE 802.16d standard. Since scheduling algorithms provide mechanisms for bandwidth allocation and multiplexing at the packet level. Admission control and congestion control policies are all dependent on the specific scheduling disciplines used. For the uplink traffic, the scheduling algorithm has to work in tandem with Call Admission Control (CAC) to satisfy the QoS requirements. The CAC algorithm ensures that a connection is accepted into the network only if its QoS requirements can be satisfied as well as the performance of existing connections in the network is not deteriorated. The computational complexity of a proposed algorithm strongly influences its scalability. We suggest that the QoS support framework in WiMax network could be hierarchical structure. Since opportunistic MAC (OMAC) is the modern view of communicating over spatiotemporally varying wireless link whereby the multi-user diversity is exploited rather than combated to maximize band width efficiency or system throughput. The cross-layer nature embeds OMAC with the potential to revolutionize the design of wireless data networks from physical to data link layers. We suggest that cross-layer opportunistic scheduling can combine with adaptive power control scheme to provide QoS support in WiMax. Different mechanisms can address different issues in QoS support in WiMax network. How to combine Bandwidth management mechanisms; Traffic handling mechanisms in a cross-layer struture to provide new QoS support scheme in WiMax are open issues.

References 1. “IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems,” IEEE Std 802.16–2004 (Revision of IEEE Std 802.16-2001), pp:1–857, 2004. 2. A. Ghosh, David R. Wolter J. G. Andrews and Runhua Chen, “Broadband Wireless Access with WiMax/802.16: Current Performance Benchmarks and Future Potential,” Communications Magazine, IEEE Volume 43, Issue 2, Feb. 2005, pp:129–136. 3. Dong-Hoon Cho, Jung-Hoon Song, Min-Su Kim, and Ki-Jun Han, “Performance Analysis of the IEEE 802.16 Wireless Metropolitan Area Network,” First International Conference on Distributed Frameworks for Multimedia Applications, 2005, DFMA ’05, pp:130–136, Feb. 2005. 4. Alavi, H.S.; Mojdeh, M., Yazdani, N. “A Quality of Service Architecture for IEEE 802.16Standards,” Asia-Pacific Conference on Communications, pp:249–253, Oct. 2005. 5. Yi-Ting Mai; Chun-Chuan Yang; Yu-Hsuan Lin, “Cross-Layer QoS Framework in the IEEE 802.16 Network,” The 9th International Conference on Advanced Communication Technology, Volume 3, pp:2090–2095, Feb. 2007.

5


113

6. T. Kwon, H. Lee, S. Choi, J. Kim, and D. Cho, “Design and Implementation of a Simulator Based on a Cross-Layer Protocol between MAC and PHY layers in a WiBro Compatible IEEE802.16e OFDMA System,” IEEE Communication Magazine, Dec. 2005, Vol. 43, no.12, pp:136–146. 7. K. Gakhar, M. Achir, A. Gravey, “Dynamic Resource Reservation in IEEE 802.16 Broadband Wireless Networks,” IWQoS 2006, Jun. 2006, pp:140–148. 8. Y. Ge and G.-S. Kuo, “An Efficient Admission Control Scheme for Adaptive Multimedia Services in IEEE 802.16e Networks,” In IEEE 64th Vehicular Technology Conference, VTC-2006, pp: 1–5, Sept. 2006. 9. D. Niyato and E. Hossain, “Radio Resource Management Games in Wireless Networks: An Approach to Bandwidth Allocation and Admission Control for Polling Service in IEEE 802.16,” IEEE Wireless Communications, 14(1), Feb. 2007. 10. L. Wang, F. Liu, Y. Ji, and N. Ruangchaijatupon, “Admission Control for Non-preprovisioned Service Flow in Wireless Metropolitan Area Networks,” In Fourth European Conference on Universal Multiservice Networks, ECUMN ’07, pp:243–249, Feb. 2007. 11. H. Fen; H. Pin-Han; S. Xuemin; “WLC17-1: Performance Analysis of a Reservation Based Connection Admission Scheme in 802.16 Networks,” IEEE GLOBECOM’06, Nov. 2006, pp:1–5. 12. H. Wang; W. Li; D.P. Agrawal, “Dynamic admission control and QoS for 802.16 wireless MAN,” Wireless Telecommunications Symposium, April, 2005, pp:60–66. 13. C.-H. Jiang; T.-C. Tsai, “Token bucket based CAC and packet scheduling for IEEE 802.16 broadband wireless access networks,” Consumer Communications and Networking Conference, CCNC 2006, Volume 1, pp:183–187, 2006. 14. H. Wang; B. He; D.P. Agrawal, “Admission control and bandwidth allocation above packet level for IEEE 802.16 wireless MAN,” Parallel and Distributed Systems, 2006. ICPADS 2006, Jul. 2006. 15. J. Chen, W. Jiao, and Q. Guo, “An integrated QoS control architecture for IEEE 802.16 broadband wireless access systems,” in Proc. Global Telecommunication Conf., (Globecom) 2005, Vol. 6 pp:3330–3335, 2005. 16. W.T. Chen; S.B. Shih; J.L. Chiang, “A Two-Stage Packet Classification Algorithm,” Advanced Information Networking and Applications, 2003. AINA 2003, Mar. 2003, pp:762–767. 17. S. Jun-Bae; L. Hyong-Woo; C. Choong-Ho, “Performance of IEEE802.16 Random Access Protocol - Transient Queueing Analysis,” IEEE Global Telecommunications Conference, 2006, GLOBECOM’06, pp:1–6, Nov. 2006. 18. J.-B. Seo; S.-J. Kim; H.-W. Lee; C.-H. Cho, “An Efficient Capacity Allocation Scheme of Periodic Polling Services for a Multimedia Traffic in an IEEE802.16 System,” Mobile Adhoc and Sensor Systems (MASS) 2006, pp:11–20, Oct. 2006. 19. Lu´ıs Felipe M. de Moraes and Paulo Ditarso Maciel Jr,” An Alternative QoS Architecture for the IEEE 802.16 Standard,” www ravel.ufrj.br/arquivosPublicacoes/conext06.pdf. 20. Y. Chen; L. Li “A Random Early Expiration Detection Based Buffer Management Algorithm for Real-time Traffic over Wireless Networks,” Computer and Information Technology, 2005, CIT 2005, pp:507–511, Sept. 2005. 21. J. Lakkakorpi; A. Sayenko; J. Karhula; O. Alanen; J. Moilanen. “Active Queue Management for Reducing Downlink Delays in WiMax,” IEEE Vehicular Technology Conference, 2007, VTC-2007, 66th Volume, pp:326–330 , Fall 2007. 22. M. Engels, P. Coenen (IMEC). Christian Hoymann, “Congestion control mechanism for interworking between WLAN and WMAN,” IEEE C802.16d-03/83. Congestion Control. wirelessman.org/tgd/contrib/C80216d-03 83.pdf. 23. H.-F. Hsiao; A. Chindapol; J.A. Ritcey; Yaw-Chung Chen; Jenq-Neng Hwang, “A new multimedia packet loss classification algorithm for congestion control over wired/wireless channels,” Acoustics, Speech, and Signal Processing, 2005. ICASSP ’05, Volume 2, pp:ii/1105 - ii/1108, Mar. 2005.

114

M. Ma and J. Lu

24. C.-W. Huang; J.-N. Hwang; D.C.W. Chang, “Congestion and error control for layered scalable video multicast over WiMax,” IEEE Mobile WiMax Symposium, 2007, pp:114–119, Mar. 2007. 25. S.A. Xergias; N. Passas; L. Merakos “Flexible resource allocation in IEEE 802.16 wireless metropolitan area networks,” Local and Metropolitan Area Networks, 2005. LANMAN 2005, pp:6pp, Sept. 2005. 26. W.K. Wong; H. Tang; Shanzeng Guo; Leung, “Scheduling algorithm in a point-to-multipoint broadband wireless access network,” Vehicular Technology Conference, 2003. VTC 2003, Volume 3, pp:1593–1597. Fall 2003. 27. S. Ryu; B. Ryu; H. Seo; M. Shi, “Urgency and efficiency based wireless downlink packet scheduling algorithm in OFDMA system,” IEEE Vehicular Technology Conference, 2005, 61st Volume 3, pp:1456–1462, May. 2005. 28. A. Shejwal; A. Parhar, “Service Criticality Based Scheduling for IEEE 802.16 WirelessMAN,” The 2nd International Conference on Wireless Broadband and Ultra Wideband Communications (AusWireless 2007), pp:12–12, Aug. 2007. 29. M. Ma, B.C. Ng, “Supporting Differentiated Services in Wireless Access Networks,” Communication systems, ICCS 2006, pp:1–5. 2006. 30. K. Wongthavarawat, A. Ganz, “Packet scheduling for QoS support in IEEE 802.16 broadband wireless access systems”, International Journal of Communication Systems, vol. 16, issue 1, pp:81–96, Feb.2003. 31. J. Chen, W. Jiao, H. Wang, “A service flow management strategy for IEEE 802.16 broadband wireless access systems in TDD mode,” IEEE International Conference, ICC 2005. Volume: 5, pp:3422–3426, 2005. 32. H. Safa.; H. Artail.; M. Karam.; R. Soudan.; S. Khayat, “New Scheduling Architecture for IEEE 802.16 Wireless Metropolitan Area Network,” IEEE/ACS International Conference on Computer Systems and Applications, 2007. AICCSA07, pp:203–210, May. 2007. 33. N. Liu; X. Li; C. Pei; B. Yang, “Delay Character of a Novel Architecture for IEEE 802.16 Systems,” Proceedings of the Sixth International Conference on Parallel and Distributed Computing, Applications and Technologies (PDCAT’05), pp:293–296, Dec. 2005. 34. M. Ma, J. Lu, Sanjay Kumar Bose, and Boon Chong Ng. “A Three-Tier Framework and scheduling to Support QoS Service in WiMax,” Information, Communications & Signal Processing, 2007, 6th International Conference, pp:1–5, Dec. 2007. 35. X. Liu, “Optimal Opportunistic Scheduling in Wireless Networks”. IEEE 58th Vehicular Technology Conference, 2003, Vol. 3, pp:1417–1421, Oct. 2003. 36. J. Sang, D. Jeong, and W. Jeon, “Cross-layer Design of Packet Scheduling and Resource Allocation in OFDMA Wireless Multimedia Networks,” in Proc. 63r d Vehicular Technology Conf. (VTC) 2006, pp:309–313, 2006. 37. Q. Liu, X. Wang and G. B. Giannakis, “A Cross-Layer Scheduling Algorithm With QoS Support in Wireless Networks,” IEEE Transactions on Vehicular Technology, May. 2006, Vol. 55, no.3, pp:839–847. 38. L. Wan, W. Ma, Z. Guo, “A Cross-layer Packet Scheduling and Subchannel Allocation Scheme in 802.16e OFDMA System,” Wireless Communications and Networking Conference, 2007, WCNC 2007, pp:1865–1870, Mar. 2007. 39. B. Rong Y. Qian Hsiao-Hwa Chen , “Adaptive power allocation and call admission control in multiservice WiMax access networks,” IEEE Wireless Communications, Feb. 2007, Volume: 14, Issue: 1, pp:: 14–19.

Chapter 6

Mobile WiMax Performance Optimization Stanislav Filin, Sergey Moiseev and Mikhail Kondakov

Abstract The Mobile WiMax system is a promising solution for delivering broadband wireless access services to mobile users. Radio resource management (RRM) algorithms play a key role in the Mobile WiMax network. The Mobile WiMax network has a number of distinct features that complicate the use of conventional RRM algorithms. We propose load-balancing approach to RRM in the Mobile WiMax network. To illustrate the advantages of the load-balancing approach, we present RRM algorithms, including call admission control, adaptive transmission, horizontal handover, and dynamic bandwidth allocation algorithms. These algorithms jointly maximize the network capacity and guarantee users quality-of-service requirements. Keywords Mobile WiMax · OFDMA · Performance optimization · Loadbalancing · System load · Quality-of-service · Radio resource management · Call admission control · Adaptive transmission · Handover · Dynamic bandwidth allocation

6.1 Introduction IEEE standards 802.16 [1] and 802.16e [2] specify the requirements for the medium access control (MAC) and physical (PHY) layers of the WiMax and Mobile WiMax systems. These systems are attracting huge interest as a promising solution for delivering fixed and mobile broadband wireless access services. The standards have incorporated such key technologies as quality-of-service (QoS) mechanisms, adaptive coding and modulation, power control, selective and hybrid automatic repeat request, orthogonal frequency division multiplexing (OFDM) and orthogonal frequency division multiple access (OFDMA), support of adaptive antenna systems and multiple-input multiple-output transmission. This provides great potential for satisfying users and operators needs.

S. Filin (B) National Institute of Information and Communications Technology, 3-4, Hikarino-oka, Yokosuka, 239-0847, Japan


115

116

S. Filin et al.

From a user’s perspective, a wireless system should deliver his/her data with the required QoS level, while an operator aims at maximizing network capacity and revenue [3]. These two goals are achieved by radio resource management (RRM) algorithms. These algorithms play a key role in the Mobile WiMax system but their complexity is high due to a lot of degrees of freedom and optimization opportunities. RRM in the OFDMA-based Mobile WiMax network includes call admission control, adaptive transmission, handover, and dynamic bandwidth allocation algorithms. Call admission control algorithm handles system overloading and satisfies users QoS by limiting the number of users in the system [4]. Adaptive transmission algorithms enable QoS-guaranteed opportunistic data transmission over a wireless medium [5]. They include scheduling, adaptive coding and modulation, power control, and time-frequency resource allocation. Seamless horizontal handover guarantees continuous service by assigning new serving base stations to a user during his/her mobility and system load variations. Dynamic bandwidth allocation algorithm distributes signal bandwidth within a group of sectors to balance their load. The Mobile WiMax network has a number of distinct features that complicate the use of conventional RRM algorithms. Users may have multiple service flows with different traffic and QoS requirements, be in different receiving conditions, and use different coding and modulation schemes and transmission power values. Consequently, it is difficult to determine the time-frequency and power resources required for the given number of users. This is the main challenge of the call admission control algorithm. In adaptive transmission algorithms, the key problems are computational intensity due to a large number of degrees of freedom, complicated frame structure, and complex MAC and PHY layers processing procedures. Traditional horizontal handover algorithms are based on the received signal level or signal-to-interference-plus-noise ratio (SINR). However, such handover algorithms neither guarantee QoS requirements nor maximize the capacity in the Mobile WiMax OFDMA network. In the Mobile WiMax network several sectors may form a group of sectors sharing the same signal bandwidth by using different groups of subcarriers. This provides additional degree of freedom for optimizing network capacity and QoS. Therefore, RRM algorithms taking into account the distinct features of the Mobile WiMax OFDMA network should be provided. We describe load-balancing approach to RRM in the Mobile WiMax OFDMA network, which is based on our system load model [6]. This model takes into account different traffic, QoS requirements, and receiving conditions of the users and efficiently combines time-frequency and power resources, as well as downlink and uplink resources in the expression for the system load. To illustrate the advantages of the load-balancing approach, we present RRM algorithms, including call admission control, adaptive transmission, horizontal handover, and dynamic bandwidth allocation algorithms. We demonstrate the advantages of our algorithms over the conventional ones by means of system level simulation. Each algorithm individually and all algorithms as a whole satisfy QoS requirements and maximize the network capacity. We define the network capacity as the maximum achievable network throughput when QoS requirements are satisfied for all the users served.

6

Mobile WiMax Performance Optimization

117

First, we describe our system load model for the Mobile WiMax OFDMA network. Then, based on this system load model we present call admission control, adaptive transmission, horizontal handover, and dynamic bandwidth allocation algorithms for the Mobile WiMax OFDMA network.

6.2 System Load Model A system load model is traditionally used in call admission control algorithms. System load characterizes the degree of system resources consumption. When all system resources have been consumed, new users are not admitted to the system. We have proposed a system load model for the Mobile WiMax OFDMA network in [6]. We consider the Mobile WiMax network comprising some sectors and some users. The sectors transmit data to the users in the downlink and the users transmit data to the sectors in the uplink. Each user may have several downlink service flows and several uplink service flows, where a service flow is a flow of data packets from an application. Different service flows may have different traffic arrival rates. The network uses the OFDM technology, the OFDMA multiple access, and the time division duplex. Each sector uses frames for the downlink and uplink data transmission, where a frame comprises a downlink subframe and an uplink subframe (see Fig. 6.1). The frame boundary between the downlink and uplink subframes may be adaptively adjusted. In the time domain the frame comprises OFDM symbols, while in the frequency domain it comprises subcarriers. In the frame, the time resource is equal to the number of OFDM symbols and the size of the frequency resource is defined by the number of subcarriers in one OFDM symbol. In the OFDMA, each subcarrier can be assigned to any user. When a subcarrier is assigned to a user, a coding and modulation scheme and a transmission power value are selected for this user on this subcarrier. In addition, sectors in the downlink and users in the uplink have the maximum transmission power constraints. The adaptation parameters available in the Mobile WiMax network are frame boundary

Fig. 6.1 OFDMA frame structure in Mobile WiMax network

118

S. Filin et al.

position, subcarriers assignment, coding and modulation schemes, and transmission power values. Data packets of a service flow should be transmitted with the required QoS. The set of QoS requirements includes the minimum average data rate, the maximum data block reception error probability, and the maximum average data block transmission delay. All these QoS requirements can be satisfied by selecting an appropriate transmission power value for the given coding and modulation scheme and the given receiving conditions [7–9]. The system load model of the Mobile WiMax network should satisfy the following requirements:

r r

r

The Mobile WiMax network has shared and individual resources. For example, transmission power of a sector is a shared system resource, because it is concurrently used by several users. Transmission power of a user is its individual resource. The system load model should include shared system resources only. When adaptive coding and modulation and power control are employed, different adaptive transmission algorithms having different target functions may be used, which leads to different amount of the consumed system resource. Consequently, the amount of currently consumed system resource cannot characterize the system load. The system load should be equal to the minimum amount of the system resources needed to satisfy QoS requirements for all the users served. The amount of the available system resource may be different in different sectors. To compare system load values of different sectors, the system load should be normalized to the available system resource. In addition, normalization simplifies RRM algorithms.

Our system load model includes uplink load, downlink load, sector load, and network load. To calculate each system load, we use the following approach. First, we write an expression for the amount of the normalized shared system resources, consumed by all users, as a function of adaptation parameters. Then, we find the system load by minimizing this expression over adaptation parameters under the constraint on the individual system resources, while satisfying the QoS requirements for all users. In the uplink, the shared system resource is the time-frequency resource of the uplink subframe. Transmission power of each user is an individual system resource. Adaptation parameters are the set of the assigned uplink subcarriers, coding and modulation schemes, and transmission power values. In the downlink, the shared system resources are the time-frequency resource of the downlink subframe and the downlink transmission power. Adaptation parameters are the set of the assigned downlink subcarriers, coding and modulation schemes and transmission power values. In a sector, the shared system resources are the uplink resources and the downlink resources. A new adaptation parameter, that is, the frame boundary position between the downlink and uplink subframes is added to the uplink and downlink adaptation parameters.

6


119

The proposed system load model takes into account all distinct features of the Mobile WiMax network, including time-frequency and power resources, QoS requirements, adaptive coding and modulation and power control, time division duplex and adaptation of the frame boundary between the downlink and uplink subframes, different users with different traffics and receiving conditions.

6.3 Call Admission Control Any wireless network has constrained resources and can serve a limited number of users with a given QoS level. Hence, a call admission control algorithm that decides whether new users should be admitted to the network is required. The admission criteria may be different. The known call admission control schemes are based on SINR, interference, bandwidth, load, or system capacity [10]. In the Mobile WiMax network the most suitable scheme is the one maximizing network capacity while satisfying QoS requirements for all admitted users. Such scheme maximizes operator’s revenue and guarantees user’s satisfaction. This call admission control algorithm admits a new user in the sector, if the sector load remains less than one when a new user and all previous users are served by the sector. Since the system load is equal to the minimum required system resources in our model, this call admission control algorithm maximizes the sector capacity. More details on the call admission control algorithm in the Mobile WiMax OFDMA network can be found in [11].

6.4 Adaptive Transmission Adaptive transmission algorithms include scheduling, adaptive coding and modulation, power control, and adaptive resource allocation. A scheduler guarantees QoS and fairness and can also handle user priorities by making a decision, how much data and of which service flows will be transmitted in the current frame. To satisfy the required QoS level, different pairs of coding and modulation scheme number and transmission power value can be used. The selection of a pair for data transmission is based on the target function. For example, when the total transmission power is minimized, a coding and modulation scheme with the minimum transmission rate and a corresponding transmission power are selected. However, when the total allocated time-frequency resource is minimized, a coding and modulation scheme with the maximum transmission rate and a corresponding transmission power are selected. In the OFDMA, adaptive resource allocation algorithm plays an important role. Receiving conditions are different for different users on the same subcarrier. Moreover, they are different for the same user on different subcarriers. Users can be assigned to the subcarriers with the best receiving conditions, thereby multi-user diversity gain is obtained. To enable opportunistic data transmission over a wireless medium a cross-layer approach should be used [5]. In this case, scheduling, adaptive

120

S. Filin et al.

coding and modulation, power control, and adaptive resource allocation algorithms should have the same target function and should be jointly optimized. Optimization of the OFDMA systems is a subject of a considerable literature. Time-frequency resource minimization, transmission power minimization, throughput maximization, and utility function optimization are traditionally performed. However, most of the known algorithms do not consider the distinct features of the Mobile WiMax network. First, MAC and PHY layers processing is not taken into account. Data blocks of a service flow arrive from the upper layers at the MAC layer, where they are converted into data packets using fragmentation and packing operations (see Fig. 6.2). Also, cyclic redundancy check and automatic repeat request mechanisms can be used. The set of data packets of the service flow arrives at the PHY layer, where it is converted into coding blocks. Each coding block is coded and decoded independently. QoS requirements are specified for the data blocks, while the coding blocks are transmitted and received. Hence, MAC and PHY layers processing should be taken into account to enable QoS-guaranteed data transmission [7–9]. Moreover, most of the known algorithms do not perform joint downlink and uplink optimization. A common problem for all the adaptive transmission algorithms in the OFDMA networks is their computational complexity due to a large number of degrees of freedom. Using our system load model results in an efficient and fast adaptive transmission algorithm. We proposed the adaptive transmission algorithm maximizing the sector capacity while satisfying the QoS requirements for all service flows scheduled for transmission in the current frame in [12]. In our algorithm, we consider such adaptation parameters as a position of the frame boundary between the downlink and uplink subframes, coding and modulation schemes, transmission power values, and positions of the service flows within the frame. Our algorithm includes selecting the optimal position of the frame boundary and maximizing the downlink and uplink capacity. The initial position of the frame boundary is selected in such a way that the available downlink and uplink resources are proportional to the downlink and uplink loads. In most cases, the initial position is very close to the optimal one. Then, we search for the optimal position in the vicinity of the initial position. When we maximize the downlink and uplink capacities, we place the service flows into the frame in a load-balancing manner. We

Fig. 6.2 MAC and PHY layers processing in Mobile WiMax network

6


121

sequentially place the service flows starting from the ones with the best receiving conditions. When a service flow is placed, we minimize the consumed shared system resources. Consequently, using our algorithm we maximize the sector capacity. The system level simulation results show the efficiency of the load-balancing adaptive transmission algorithm. The simulation topology includes seven cells each having three sectors, where six cells surround the central cell. We collect the statistics for the sectors of the central cell. The surrounding sectors are the sources of interference. Each cell has three sectors and frequency reuse factor is three. Cell radius is 1000 m. Sector bandwidth is 10 MHz, carrier frequency is 2.3 GHz. Maximum transmission power of each sector is 10 W, while maximum transmission power of each terminal is 1 W. Each sector has 120-degree antenna, each terminal has omnidirectional antenna. We use Vehicular B propagation channel model [13]. Propagation channel components are transmit and receive antenna gains, median path loss, and fast fading. In each interfering sector we pseudo-randomly distribute 10 users. We select traffic load of each interfering user such that frame load in each interfering sector is approximately 75%. In each central sector we pseudo-randomly distribute 5, 10, . . ., 100 users. Each central user has one downlink and one uplink service flow, each carrying Video traffic [14]. QoS requirements for these service flows of central users are the same. Minimum average data rate is 32 kb/s, maximum average data block delay is 200 ms, and maximum data block reception error probability is 0.001. Figure 6.3 shows the sector throughput as a function of the traffic load for the load-balancing algorithm and for two known algorithms, that is, total consumed time-frequency resource minimization and total transmission power minimization [7–9]. The known algorithms also take into account the distinct features of the Mo-

Fig. 6.3 Sector throughput as a function of traffic load for load-balancing and known total consumed time-frequency resource minimization and total transmission power minimization adaptive transmission algorithms in Mobile WiMax network

122

S. Filin et al.

Fig. 6.4 Simulation time as a function of traffic load for load-balancing and known total consumed time-frequency resource minimization and total transmission power minimization adaptive transmission algorithms in Mobile WiMax network

bile WiMax network, but they are not load-balancing. The load-balancing algorithm has a 0.3 b/s/Hz spectral efficiency gain when the sector is almost fully loaded. Figure 6.4 shows the simulation time as a function of the traffic load for three algorithms considered. The load-balancing algorithm has a considerable computational efficiency gain of several orders. Although the system load model is not traditionally employed in the adaptive transmission algorithms its usage results in a very efficient algorithm that provides a spectral efficiency gain and is considerably less computationally intensive.

6.5 Horizontal Handover Handover algorithms first appeared in cellular networks with mobile users. When moving, a user passes from the serving sector’s coverage area to the coverage area of another sector. As the receiving conditions of this user in the serving sector degrade, we come to a point when the user can no longer maintain a connection in his/her serving sector. Therefore it appears reasonable to hand over this user to the sector, to the coverage area of which he/she currently belongs. The receiving conditions are characterized by the received signal level or SINR. Consequently, traditional handover algorithms are based on the received signal level or SINR. The user is handed over to the sector with the maximum signal level or SINR value. This scheme may be expanded by adding thresholds to decrease the number of ping-pong events and signaling load and to keep the call dropping probability low. [15] In the Mobile WiMax network the horizontal handover should be seamless. Seamless horizontal handover is a handover that continuously guarantees the required QoS for all user’s service flows while he/she is active in the network. To guarantee QoS requirements, downlink and uplink receiving conditions and the sector load of the serving sector should be taken into account. The horizontal handover

6


123

algorithm guarantees QoS requirements by assigning a new serving sector to the user when the receiving conditions or the sector load change. Traditional horizontal handover algorithms do not take the load into account. Hence, they cannot guarantee QoS requirements in the Mobile WiMax network. We have proposed the load-balancing QoS-guaranteed horizontal handover algorithm that maximizes the capacity of the Mobile WiMax network in. [16] Our algorithm provides the capacity maximization by distributing the load of the overloaded sectors among other sectors and by balancing the load of the sectors that are not overloaded. In other words, we minimize the maximum sector load in all sets of sectors of the Mobile WiMax network. We use the optimization procedure consisting of K-1 steps, where K is the number of sectors in the Mobile WiMax network. During the first step we select the serving sectors for all network users to minimize the maximum sector load among all K sectors. Then, for the users of the sector with the maximum sector load this sector becomes a new serving sector. This sector and all its users are excluded from further optimization. During the second step the remaining users and K-1 sectors are optimized in the same way. During the last step we minimize the maximum sector load for two remaining sectors. After this optimization procedure we initiate a horizontal handover procedure for the users, whose serving sector number has been changed. The proposed horizontal handover algorithm maximizes the network capacity and guarantees QoS requirements when the network is not overloaded. We show the efficiency of our algorithm using the system level simulation. The simulated network includes seven cells. Six cells surround the central cell. The frequency reuse factor is seven and the cell radius is 300 m. The carrier frequency is 2.4 GHz and the signal bandwidth is 10 MHz in each cell. Each sector has the maximum transmission power value of 20 W and the omni-directional antenna. Each user has the maximum transmission power value of 1 W and the omni-directional antenna. We have used the Vehicular B propagation channel model [13]. Figures 6.5, 6.6 and 6.7 illustrates the advantages of our algorithm compared to the traditional SINR-based algorithm in the Mobile WiMax network.

Fig. 6.5 Maximum sector load as a function of frame number for load-balancing and SINR-based horizontal handover algorithms in Mobile WiMax network

124

S. Filin et al.

Fig. 6.6 Load of the central sector as a function of traffic load for load-balancing and SINR-based horizontal handover algorithms in Mobile WiMax network

Fig. 6.7 Network throughput as a function of traffic load for load-balancing and SINR-based horizontal handover algorithms in Mobile WiMax network

Figure 6.5 shows the maximum sector load among seven sectors as a function of frame number. Figure 6.4a indicates that the traditional handover algorithm occasionally leads to overloading, that is, to the network condition when the QoS requirements are not satisfied for the users. Our load-balancing handover algorithm keeps the maximum sector load less than one, thus guarantees meeting the QoS requirements.

6


125

Figure 6.6 shows the load of the central sector as a function of the network traffic load. The SINR-based handover algorithm leads to overloading under the network traffic load value of about 22 Mb/s, whereas our load-balancing algorithm results in the overloading condition under the network traffic load value of about 70 Mb/s. Figure 6.7 shows the network throughput, that is, the throughput of seven sectors as a function of the network traffic load. The SINR-based handover algorithm reaches the maximum throughput value of about 22 Mb/s, while our load-balancing algorithm gains the maximum throughput value of about 70 Mb/s. Therefore, the load-balancing approach enables development of the efficient horizontal handover algorithm in the Mobile WiMax network. This algorithm guarantees QoS requirements and maximizes network capacity. Our load-balancing algorithm provides a considerable gain in the network capacity over the traditional SINR-based algorithm.

6.6 Dynamic Bandwidth Allocation In the Mobile WiMax network several sectors may form a group of sectors, where sectors within the group share the same signal bandwidth using different groups of subcarriers within this signal bandwidth. Time-frequency resources of these sectors do not overlap, but they may be adaptively distributed among the sectors on the frame-by-frame basis by changing groups of subcarriers used in each sector. This introduces the dynamic bandwidth allocation feature in the Mobile WiMax network. Serving sectors may be selected for the users in such a way that the total consumed system resources are minimized. If several sectors become overloaded, they may take system resources from the non-overloaded sectors. These are the key ideas of the joint handover and dynamic bandwidth allocation algorithm in the Mobile WiMax network. We proposed the load-balancing QoS-guaranteed joint dynamic bandwidth allocation and horizontal handover algorithm that maximizes the capacity of the Mobile WiMax network in. [17] This algorithm is similar to the horizontal handover algorithm described in previous section. It also intends to minimize the maximum sector load in all sets of sectors of the Mobile WiMax network. However, it has one more degree of freedom for optimization compared to the horizontal handover algorithm. In the horizontal handover algorithm, we can change consumed system resources of sectors only. In the joint dynamic bandwidth allocation and horizontal handover algorithm we can additionally adapt available system resources in each group of sectors sharing the same signal bandwidth. This is done by adaptive distributing of subcarriers used by each sector in a group. We evaluate the proposed joint handover and dynamic bandwidth allocation algorithm using system level simulation of the Mobile WiMax network. The simulated network includes seven cells, each having three sectors. Six cells surround the central cell. The frequency reuse factor is three and the cell radius is 300 m. The carrier frequency is 2.4 GHz and the signal bandwidth is 10 MHz in each cell. Three

126

S. Filin et al.

sectors of each cell forms a group of sectors. They share 10 MHz bandwidth. Each sector has the maximum transmission power value of 20 W and 120-degree sectored antenna. Each user has the maximum transmission power value of 1 W and the omni-directional antenna. We use the Vehicular B propagation channel model [13]. Within the coverage area of the Mobile WiMax network we distribute 3 high-rate Internet users and 21, 42, . . . low-rate Internet users. Each high-rate user has one downlink and one uplink service flows, each having traffic arrival rate 512 kb/s. Each low-rate user has one downlink and one uplink service flows, each having traffic arrival rate 128 kb/s. All these users pseudo-randomly move within the coverage area of the network during the simulation. For the described scenario, we simulated two cases. In the first case, we perform handover only using the algorithm described in the previous section. In the second case, we perform joint handover and dynamic bandwidth allocation. Figure 6.8 shows network throughput as a function of network traffic load for handover algorithm (“HO” curve) and for joint handover and dynamic bandwidth allocation algorithm (“HO & DBA” curve). When handover algorithm is used, the maximum network throughput is equal to 95 Mb/s. When joint handover and dynamic bandwidth allocation algorithm is used, the maximum network throughput is equal to 115 Mb/s. In other words, we have 20% network capacity gain. This network capacity gain may be interpreted as follows. When dynamic bandwidth allocation is enabled, a part of the users may switch to higher transmission rates due to better receiving conditions. Consequently, a part of the system resources becomes available for the additional users.

Fig. 6.8 Network throughput as a function of network traffic load for handover algorithm and for joint handover and dynamic bandwidth allocation algorithm

6


127

6.7 Conclusions We have described optimization of the Mobile WiMax OFDMA network using the load-balancing approach to RRM. To illustrate the advantages of the load-balancing approach, we have presented RRM algorithms, including call admission control, adaptive transmission, horizontal handover, and dynamic bandwidth allocation algorithms. These algorithms jointly maximize the network capacity and guarantee users QoS.

References 1. IEEE Standard 802.16–2004, IEEE Standard for Local and Metropolitan Area Networks – Part 16: Air Interface for Fixed Broadband Wireless Access Systems, (2004). 2. IEEE Standard 802.16e–2005, Amendment to IEEE Standard for Local and Metropolitan Area Networks – Part 16: Air Interface for Fixed Broadband Wireless Access Systems – Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands, (2005). 3. S. Frattasi, H. Fathi, F.H.P. Fitzek, R. Prasad, and M.D. Katz, Defining 4G Technology from the User’s Perspective, IEEE Network 20(1), 35–41 (2006). 4. D. Niyato and E. Hossain, Call Admission Control for QoS Provisioning in 4G Wireless Networks: Issues and Approaches, IEEE Network 19(5), 5–11 (2005). 5. V. Srivastava and M. Motani, Cross-Layer Design: A Survey and the Road Ahead, IEEE Communications Magazine 43(12), 112–119 (2005). 6. S.N. Moiseev et al, System Load Model for the OFDMA Network, IEEE Communications Letters 10(8), 620–622 (2006). 7. S.A. Filin et al, QoS-Guaranteed Cross-Layer Adaptive Transmission Algorithms for the IEEE 802.16 OFDMA System, IEEE Wireless Communications and Networking Conference (WCNC 2006) 2, 964–971 (2006). 8. S.A. Filin et al, QoS-Guaranteed Cross-Layer Adaptive Transmission Algorithms with Selective ARQ for the IEEE 802.16 OFDMA System, IEEE Vehicular Technology Conference (VTC 2006 Fall) (2006). 9. S.A. Filin et al, QoS-Guaranteed Cross-Layer Transmission Algorithms with Adaptive Frequency Subchannels Allocation in the IEEE 802.16 OFDMA System, IEEE International Conference on Communications (ICC 2006) 11, 5103–5110 (2006). 10. M.H. Ahmed, Call Admission Control in Wireless Networks: A Comprehensive Survey, IEEE Communications Surveys 7(1), 50–69 (2005). 11. S.N. Moiseev and M.S. Kondakov, Call Admission Control in Mobile WiMax Network, International Journal of Communication Systems, (unpublished). 12. S.A. Filin et al, Fast and Efficient QoS-Guaranteed Adaptive Transmission Algorithm in Mobile WiMax System, IEEE Transactions on Vehicular Technology, 2008 (unpublished). 13. Recommendation ITU-R M.1225, Guidelines for Evaluation of Radio Transmission Technologies for IMT–2000, (1997). 14. 3GPP2 Contribution C.P1002-C-0, cdma2000 Evaluation Methodology, (2004). 15. A.V. Garmonov et al, QoS-Oriented Intersystem Handover between IEEE 802.11b and Overlay Networks, IEEE Transactions on Vehicular Technology 57(2), 1142–1154 (2008). 16. S.N. Moiseev et al, Load-Balancing QoS-Guaranteed Handover in the IEEE 802.16e OFDMA Network, IEEE Global Communications Conference (GLOBECOM 2006), (2006). 17. S.N. Moiseev and M.S. Kondakov, Joint Handover and Dynamic Bandwidth Allocation in Mobile WiMax Network, IEEE Transactions on Mobile Computing, (unpublished).

Chapter 7

A Comparative Study on Random Access Technologies of 3G and B3G Mobile Communications Systems Jungchae Shin and Ho-Shin Cho

Abstract This chapter introduces and compares the various random access (RA) technologies designed for 3G and beyond 3G mobile communication systems including Mobile WiMax, IEEE 802.20, cdma2000, WCDMA, and 3G LTE. In terms of fundamental design issues of RA, such as multiplexing, RA procedures, backoff algorithm, power control and priority schemes, the competing systems are compared. Furthermore, the performances of ranging for Mobile WiMax are numerically evaluated by both link- and MAC (Media access control)-level simulations in terms of throughput, delay time, and ranging success probability. Keywords Random access · Ranging · Mobile WiMax · IEEE 802.20 · Cdma2000 · WCDMA · 3G LTE

7.1 Introduction Various kinds of 3rd generation (3G) and 3G plus mobile communication systems have been introduced such as cdma2000, WCDMA (Wideband code division multiple access), HSDPA (High speed downlink packet access), and Mobile WiMax (Worldwide interoperability for microwave access). Recently, the advanced versions of 3G and 3G plus systems called beyond 3rd generation (B3G) systems such as WiMax Evo, and 3G long-term evolution (LTE) are being intensively developed and are anticipated to start the services near 2010 [1–6]. These endeavors to evolve the systems are ultimately directed toward fourth generation (4G) mobile communication system which is characterized by global mobility, higher data rate up to 1 Gbps when stationary, and lower error probability. One of the crucial requirements to reach 4G era is to enhance random access (RA) performance. The RA is the first step for a mobile station (MS) to enter a mobile communication network. At the initial access stage, MS is out of synchronization to base station (BS) and thus, has no acquisition of system parameters and J. Shin (B) School of Electrical Engineering and Computer Science, Kyungpook National University, Daegu, Korea


129

130

J. Shin and H.-S. Cho

signaling/data channels. At this state, a preliminary procedure called RA is required to ready for communication with BS. In general, the RA follows the two steps [7]: 1. Uplink time/frequency/power synchronization with a BS. 2. Uplink resource acquisition for bandwidth requests. Being turned on, MS first searches the best BS to be connected and tries to find downlink information on time/frequency/power of the selected BS using a pilot signal. Then, MS tries to acquire uplink synchronization by sending an access probe. Receiving an acknowledgment (ACK) message which includes the modified uplink synchronization values, MS can adjust uplink synchronization. After that, the access probe may be retransmitted to obtain uplink channel on which bandwidth request message is sent. As explained above, RA plays fundamental roles at initial state of network entrance and affects the system performance, especially the ones related to users’ own experience such as the service starting time and the deliverable service quality. Thus, the mobile communication systems have made efforts competitively to enhance the RA performances. The main issues in RA are how long the access waiting time is, how fast and efficiently RA collision can be resolved, and how the quality of service can be differentiated according to the users’ priority, etc. This chapter introduces and compares the various RA technologies which have been developed for 3G and beyond 3G mobile communication systems including Mobile WiMax, IEEE 802.20, cdma2000, WCDMA, and 3G LTE. In terms of fundamental design issues of RA, such as multiplexing, RA procedures, backoff algorithm, power control and priority schemes, the competing systems are compared. Furthermore, the performances of ranging for Mobile WiMax are numerically evaluated by link- and MAC (Media access control)-level simulations in terms of throughput, delay time, and ranging success probability.

7.2 Initialization – Procedures Prior to Random Access A main function of RA is to acquire uplink synchronization and uplink resources. Before RA is executed, several steps called the initialization are required in order to synchronize timing and frequency on downlink with the serving BS. In Table 7.1, various initialization schemes of 3G and B3G systems are summarized.

7.2.1 Mobile WiMax A downlink preamble located at the first in a frame is used to search the most preferable radio access station (RAS) which is equivalent to BS and to acquire time and frequency synchronization. Then, a portable subscriber station (PSS) which is equivalent to MS decodes downlink channel descriptor (DCD) and downlinkmap (DL-MAP) in order to acquire the information on downlink channel configuration such as downlink subframe structure, burst profile, and physical layer

7

A Comparative Study on Random Access Technologies

131

Table 7.1 Initialization procedures and required information for various 3G and B3G systems Procedure Detect cell & sector Downlink synchronization Downlink & uplink channel parameters acquisition

Mobile WiMax Downlink preamble

DCD, DL-MAP, UCD, UL-MAP

IEEE 802.20

cdma2000

WCDMA

3G LTE

Superframe preamble pilots (TDM1, −2, −3) Superframe preamble (pBCH0, -1), Data channel (ExtendedChannelInfo message)

F-PICH F-SCH

SCH (Primary & secondary synchronization code), CPICH

SCH

F-BCCH F-CCCH

PCCPCH, BCH

BCH

characteristics. PSS subsequently tries to obtain uplink access information from uplink channel descriptor (UCD) and uplink-map (UL-MAP); inhere, uplink access information includes uplink subframe structure, ranging region, the number of accessible ranging codes, and backoff contention window size, etc [8–10]. Finally, PSS gets ready for RA which is especially called a ranging in IEEE 802.16 standards.

7.2.2 IEEE 802.20 IEEE 802.20 system has the similar initialization procedure as Mobile WiMax. The noticeable feature of IEEE 802.20 is only that access terminal (AT) which is equivalent to MS employs a hierarchical scan of three pilot signals named time division multiplexing 1 (TDM1), TDM2, and TDM3 in superframe preamble; superframe is comprised of multiple frames and superframe preamble is located at the first in superframe. This hierarchical scan is similar to the three-step cell search procedure in WCDMA and significantly reduces the searching complexity [11]. After searching a sector, AT achieves time and frequency synchronization with the serving access point (AP) and then decodes two primary broadcast channels (p-BCH0, p-BCH1) to obtain physical channel configuration information. After that, AT can finally obtain access information from ExtendedChannelInfo messages such as access cycle duration, maximum number of the access probes per access sequence, and ramping power step size [11, 12].

7.2.3 cdma2000 AT starts initialization procedures with a network determination through which the AT selects the best AP and the available frequency band based on predetermined access information [13]. Then, the AT needs to acquire the forward pilot channel (F-PICH) within a limited time duration (specified by 15 seconds). If it fails in the timely acquisition, the AT goes back to the network determination state. If it succeeds, the AT gets initial timing information such as PILOT PN (i.e. PN code offset) and 37-bit CDMA SYS TIME from synchronization message in forward

132


synchronization channel (F-SCH). After synchronization, the AT monitors two types of messages; page message and system parameter message transmitted on forward common control channel (F-CCCH) and forward broadcast control channel (F-BCCH), respectively. Then, the AT is now ready to access the system [3, 13].

7.2.4 WCDMA User equipment (UE) which is equivalent to MS obtains the downlink scrambling code and the common channel frame synchronization during a downlink synchronization procedure named a cell search. The cell search in WCDMA consists of three steps; slot synchronization, frame synchronization and code-group identification, and scrambling-code identification [14]. In the first step, UE exploits synchronization channel (SCH) and obtains primary synchronization code for the purpose of slot-synchronization. The primary synchronization code is same for all Node-Bs which are equivalent to BSs. In the second step, UE achieves the framesynchronization using secondary synchronization code in SCH and determines the code-group of the BS [15]. In the third step, UE finds the primary scrambling code through common pilot channel (CPICH). Finally, UE obtains access information such as available RA codes and RA slots from primary common control physical channel (PCCPCH) corresponding to the broadcast channel (BCH) in transport layer. And then UE can also determine the radio frame timing of all common physical channels [2, 15].

7.2.5 3G LTE Synchronization channel (SCH) and broadcast channel (BCH) are utilized in the initialization procedure of 3G LTE. UE acquires the primary information such as downlink symbol timing and frequency from SCH and the remaining cell/systemspecific information such as overall transmission bandwidth of the cell, cell ID, and radio frame timing information from BCH [7]. SCH and BCH are allocated one or multiple times in every 10 msec radio frame. For the purpose of the fast acquisition, the position of SCH and BCH in time and frequency domain is predetermined. Especially, the frequency band for SCH and BCH is fixed by 1.25 MHz and is located at the center of the total transmission bandwidth [7].

7.3 RA Procedures 7.3.1 Mobile WiMax In Mobile WiMax, RA is called ranging which is categorized into 4 types according to the purpose such as initial, handover, periodic, and bandwidth request. Initial and periodic ranging are designed to finely adjust the timing synchronization and

7


133

transmission power. Handover ranging is to continue current service in a new cell. Uplink channel request is performed by bandwidth request ranging [8–10]. PSS starts ranging process with transmitting a 144-bit length ranging code via ranging region which is located at first 3 OFDM symbols in uplink subframe. A 144-bit length ranging code is generated from pseudo random binary sequence (PRBS) generator using the cell ID as a seed value [8] and is transmitted by 144 subcarriers modulated by binary phase shift keying (BPSK). Since the total number of subcarriers is 864, the maximum 6 ranging codes can be transmitted in an OFDM symbol duration. However, the ranging regions are shared by non-ranging signaling such as channel quality indicator (CQI), and acknowledgement (ACK). Moreover, initial and handover ranging are repeated 2 times over 2 OFDM symbols for reliable transmission in unstable channel conditions. Thus, the available room for ranging [16] becomes much smaller than the maximum of 6. Upon receiving ranging codes from users, RAS operates parallel auto-correlations upon the received ranging codes with candidate PN codes, in order to recognize which ranging code is transmitted. The ranging code yielding the highest correlation value is selected. If the synchronization agreement and the power level of received ranging code are both acceptable, RAS notifies the ranging success by sending a control message named RNG-RSP carrying “success”. If some modifications are needed in synchronization and transmitting power level, RAS sends the adjustment values through RNG-RSP. If no ranging code yields auto-correlation output over a certain threshold level, then RAS cannot even recognize the ranging, and thus cannot give any ranging response. PSS waiting the ranging response retries a ranging when a timer named T3 is expired Figure 7.1. shows procedures of 4 different types of ranging. In cases of initial and handover ranging, afterward downlink synchronization and acquisition, PSS should take an initial backoff to transmit a ranging request (RNG-REQ) code. RAS has 3 types of ranging response; no response, “continue” message on RNG-RSP, and “success” message on RNG-RSP. If PSS does not receive any ranging response from RAS until T3 timer is expired, it makes decision on the access as failure, and then takes an additional random backoff with binary exponential random backoff algorithm [8]. Meanwhile, if PSS receives a “continue” message with adjustment values on time and power levels, PSS corrects the timing and power level and resends an RNG-REQ code. Otherwise, if PSS receives a “success” message on RNG-RSP and CDMA Allocation IE (i.e. information on uplink-channel allocation), PSS sends RNG-REQ message carrying its MAC address. Then, RAS binds the MAC address with a connection identifier (CID) and assigns the CID to PSS. The procedures of periodic and bandwidth request ranging are much shorter and simpler than those of initial and handover ranging because PSS already has his CID and good uplink synchronization. In order to maintain the synchronization between RAS and PSS, periodic ranging is required every expiration of T4 timer or irregularly when it is needed. In case of bandwidth request ranging, T16 timer runs in waiting for RNG-RSP message with CDMA Allocation IE. If PSS fails in receiving the RNG-RSP until T16 timer is expired, PSS has a random backoff from binary exponential random backoff algorithm. Even though several random backoffs may

134

J. Shin and H.-S. Cho PSS

PSS

RAS UCD/UL-MAP

Timeout T4

Initial Backoff Start T3

Initial Backoff

RNG-REQ Code

RNG-REQ Code

Start T3

Timeout T3 Random Backoff

RAS UCD/UL-MAP

Timeout T3 RNG-REQ Code

Random Backoff RNG-RSP Message (“continue”)

RNG-REQ Code

RNG-REQ Code

RNG-RSP Message (“continue”)

RNG-RSP Message (“success”) UL-MAP (CDMA_Allocation_IE)

RNG-REQ Code

RNG-REQ Message (MAC Address) RNG-RSP Message (Management CIDs)

RNG-RSP Message (“success”)

(b) Periodic ranging

(a) Initial and handover ranging

PSS

RAS UCD/UL-MAP

Start T16

RNG-REQ Code

Timeout T16 Random Backoff RNG-REQ Code UL-MAP (CDMA Allocation IE)

BW-REQ Header

UL-MAP (BW Allocation)

(c) Bandwidth request ranging

Fig. 7.1 Ranging procedures of 4 types of ranging in Mobile WiMax system

happen, in the end, PSS receives CDMA Allocation IE which indicates the uplink resources through which PSS sends the bandwidth request signaling.

7.3.2 IEEE 802.20 For RA, AT transmits an access probe named AccessSequenceID. 1024 orthogonal AccessSequenceIDs are grouped into 9 different sets according to AccessSequencePartion which is given by an access node (AN). Then, based on buffer state and received pilot strength, AT determines an AccessSequenceID set among which the AccessSequenceID is selected. The access probe selected by AT is transmitted through the reverse access channel (R-ACH). If AN successfully receives the access probe, the AT will receive the Access Grant message through forward shared signaling channel (F-SSCH). Access Grant message contains medium access control identification (MAC ID) and node ID which are scrambled with a hash of the AccessSequenceID [11, 12]. A simple example for access probe transmission is illustrated in Fig. 7.3. First, AT takes p-persistence test to determine the timing offset for access probe


Access probe power

p-persistence

Persistence interval

AccessCycleDuration

probe probe sequence

1

135

Access Probe sequence Grant Timer backoff time

2

3

Np

1

p-persistence

7

AccessCycleDuration

1

2

3

Np

2

Time Ns

Unit time: ControlSegmentPeriod[6 RL PHY frames] AccessCycleDuration(2bits)=[1,2,3,4] MaxProbePerSequence(4bits)=[1,2,…,16] MaxProbeSequences=3

ProbeRampUpStepSize(4bits)=0.5*(1+n)dB AccessRetryPersistence(3bits)=2–n AccessGrantTimer=5 PHY frames

Fig. 7.2 Access probe transmission algorithms in IEEE 802.20 system

transmission. The inter access probes duration is specified by AccessCycleDuration and the number of access probe transmissions is limited by Np . If the number of access probes transmitted reaches Np , further access probe transmission is prohibited and AccessGrantTimer starts to run. A new access attempts are allowed when the AccessGrantTimer is expired. Transmission power of the initial access probe is determined by an open loop power control, but the power of subsequent access probe is controlled by a power ramping scheme based on the parameter ProbeRampUpStepSize. Several important parameters of access probe transmission are described in Fig. 7.2. More detailed information on access parameter is explained in overhead message protocol (OMP) [12]. The algorithm for timing offset determination is explained by a flow chart in Fig. 7.3. When RA is triggered by AN with paging, if ProbeSequenceNumber is not equal to 1, the timing offset is generated from Geo(p) process truncated by a MaxProbeSequenceBackoff value, where Geo(p) is a geometric random variable with parameter p. If ProbeSequenceNumber is equal to 1 and QuickPage bit is not equal 1, timing offset is uniformly selected between 0 and 3 times PageResponseBackoff. Specially, if both ProbeSequenceNumber and QuickPage bit are equal to 1, AT immediately transmits access probe without delay. On the other hand, when RA is requested by PSS not by AN paging, if ProbeSequenceNumber is equal to 1, AT immediately sends an access probe And if not equal to 1, timing offset is determined by Geo(p) truncated by MaxProbeSequenceBackoff value.

7.3.3 cdma2000 Figure 7.4 shows an access procedure in cdma2000. An access probe consists of two parts; a preamble for pilot channel and a capsule for data and pilot channels as

136


Response to a paging?

No

Yes

No

ProbeSequence Number==1?

ProbeSequence Number==1?

No

Yes

Yes Yes QuickPage bit==1?

No min[Geo(p) RV, MaxProbeSequenceBackoff]

Uniform distribution [0,PageResponseBackoff*3]

Persistence interval=0

min[Geo(p) RV, MaxProbeSequenceBackoff]

Unit time: ControlSegmentPeriod[6 RL PHY frames] PageResponseBackoff=1

MaxProbeSequences=3

MaxProbeSequenceBackoff=8

Fig. 7.3 Timing offset determination of access probe transmission in IEEE 802.20 system

shown in Fig. 7.4. The parameters, PreambleLength and CapsuleLengthMax specify the lengths of preamble and capsule, respectively. And the unit length is 16 slots. A persistence test is performed at every transmission of access probe in order to effectively control a congestion on the Access channel. The maximum number of access probes allowed at once is specified by (1+NUM STEPs). The transmission power of every first access probe is determined by the initial power level specified by AN or the measured power level specified by AT with open loop power control. AT transmits repeatedly access probes with power ramped until an access acknowledgement message is successfully received from AN. In this case, the step of power increase is determined by AN [3, 13].

7.3.4 WCDMA After cell search and synchronization procedure, UE obtains some parameters for RA such as the number of signatures, preamble scrambling codes and subchannels from the BCH. The random access channel (RACH) and the physical random access channel (PRACH) which belong to the transport channel and the physical channel in a hierarchical WCDMA channel architecture, respectively, are used for RA. As a response to RA, the acquisition indicator (AI) is transmitted on the acquisition indicator channel (AICH). Figure 7.5 shows the RA procedure on AICH and PRACH where a 20 msec frame combined by two system frames is divided into 15 slots. Between AICH and PRACH, a timing offset by ␶p-a exists. That is, transmitting a preamble onto a randomly selected slots of PRACH, UE anticipates to receive the response (AI) from Node-B after the delay of ␶p-a . Unlike other systems where the RA response is

...

137

persistence

...

persistence

persistence

A Comparative Study on Random Access Technologies persistence

7

...

... Time

Access probe sequence Probe (1+NUM_STEPs) Probe 3 Probe 2 Probe 1

… Time

Data Channel Pilot Channel Pilot Channel

Time PreambleLength x 16 slots

CapsuleLength Max x 16slots

Fig. 7.4 Access probe transmission and its structure in cdma2000 system Radio Frame with SFN mod 2 = 0

AICH access slots Rx at UE

#0

#1

#2

#3

#4

Radio Frame with SFN mod 2 = 1

#5

#6

#7

#8

#9

#10

#11

#12

#13

τp-a τp-a 5120chips

PRACH access slots Tx at UE

#0

#1

#2

#3

#4

#5

#6

#7

#8

#9

#10

#11

#12

#13

4096chips

τp-p, min

τp-m Access slot set 1

Access slot set 2

10 msec Preamble

AI

10 msec Message part

Fig. 7.5 The timing relation between PRACH and the AICH in WCDMA system

#14

#14

138


valid at any time, thus not only MS needs to keep listening the response but also the response message should contain both the RA transmission time and the access code number, WCDMA requires UE to listen AICH at only predetermined time points, thus AI could be recognized by only the received signature information without the timing information. The transmission power of preamble is ramped up from Preamble Initial power by the Power Ramp Step. The successive preambles should be positioned apart at least ␶p-p,min and the last preamble acquiring the AI could be followed by a message with a length of 10 msec or 20 msec after ␶p-m . The preamble has 4096-chips length composed of 256 repetitions of a 16-chips signature which is randomly selected from a Hadamard code set. The AI has also 4096-chips length and the dispreading output at UE has the values of +1, −1, and 0; +1 for positive acknowledgement (ACK), −1 for negative acknowledgement (NACK), and 0 for no response. The resources for RA such as preamble signatures (i.e. access codes) and access slots (i.e. access time) are divided into 8 inequality groups according to Access Service Class (ASC) ; ASC 0 is the highest priority and ASC 7 is the lowest priority [17]. According to ASC, the probability for persistence test Pi also differs. The RA procedure is summarized as follows [14, 17]: 1. The UE randomly selects an RA resource according to ASC. 2. Taking a persistent test with Pi , the UE transmits a preamble through a selected uplink access slot with Preamble Initial Power. 3. The BS responds to the preamble with ACK or NACK. 4a. Receiving NACK or no AI, UE retransmits a preamble with the power ramped up after a random backoff. 4b. Receiving ACK, UE transmits a message after the delay ␶p-m from the last preamble.

7.3.5 3G LTE Both the non-synchronized RA and the synchronized RA are used. The nonsynchronized RA which is designed to acquire uplink synchronization and uplink resource has 2 different approaches of approach-1 and approach-2 as shown in Fig. 7.6. In approach-1, UE sends an RA burst and a bandwidth request message together, then, Node B responds with uplink timing information and up- and downlink channel assignments. On the other hand, in approach-2, UE sends RA burst first and then Node B responds to the burst with uplink timing information and uplink resource allocation for the bandwidth request. After that, UE requests a bandwidth through the assigned time-frequency resource. The synchronized scheme may be used when uplink synchronization has been already done. The procedure of synchronized RA is similar to that of non-synchronized RA except that it does not need the response of uplink timing information [7].

7


UE

Node-B

139

UE

Node-B

Access preamble (+Message)

Access preamble

Timing information

& Scheduling request resource allocation

Timing information

Scheduling request Uplink data resource allocation Uplink data resource allocation

UL data transmission

Preamble part

Payload part

UL data transmission

Preamble part

RA burst

RA burst

(a) Approach-1

(b) Approach-2

Fig. 7.6 Non-synchronized RA procedures in 3G LTE system

7.4 Technologies Comparison In this section, fundamental design issues in RA are addressed with comparison of various mobile communication systems. Table 7.2 summarizes the RA technology features of cdma2000, WCDMA, IEEE 802.20, 3G LTE, and Mobile WiMax. Note that standardization of 3G LTE is now under way, and thus, it is difficult to specify the technologies in detail at current state.

7.4.1 Multiplexing Scheme Access probe and user’s data are transmitted by time and frequency division multiplexed (TDM/FDM) manner in OFDMA based systems such as IEEE 802.20 and Mobile WiMax. On the other hand, code division multiplexed manner is used in CDMA based systems such as WCDMA and cdma2000. In 3G LTE, both FDM/TDM and CDM are considered.

7.4.2 RA Procedures In cdma2000 system, AT transmits preamble and data message together while in the other systems the access probe and data message are separately transmitted. The simultaneous transmission approach is being also considered in 3G LTE standard because it can reduce the latency of access time. However, if access probe is not correctly recognized at BS due to unreliable channel conditions such as collision, faulty synchronization and busy network, the user data message is discarded

140

J. Shin and H.-S. Cho Table 7.2 Main features of RA technologies according to mobile communication systems

Item

cdma2000

Multiplexing CDM Procedures

Inter-probe duration / backoff algorithm

Inter-probe power control

Priority

WCDMA

IEEE 802.20

CDM

TDM/FDM

3G LTE

Mobile WiMax

TDM/FDM, or TDM/FDM CDM Sequential Sequential The same time Sequential The same transmission transmission transmission transmission time trans(message (message (message of preamble mission of after after after and message preamble preamble) preamble) preamble) Or and Sequential message transmission (message after preamble) Fixed and Variable/ Fixed / random Under consid- Variable / variable/ random backoff eration truncated random backoff binary backoff exponential random backoff Power ramping Power ramping Power ramping Power ramping Power (zero step ramping size) or open and open loop power loop power control control Persistence Persistence Persistence Under consid- None test test, 8 access test, 9 access eration service probe set classes selection

to follow the failed access probe, which causes unnecessary system overhead and interference. Meanwhile, the separate transmission of access probe and user data can save unnecessary power consumption and interference; however, it may cause undesirable transmission delay.

7.4.3 Inter-Probe Duration and Backoff Algorithm Variable inter-probe duration is basically employed in most systems using persistence test and backoff algorithm. However, IEEE 802.20 uses a fixed inter-probe duration. In cdma2000, if the transmission time of access probe falls into the silence-interval, the access probe transmission is canceled. Thus the inter-probe duration does not maintain to be constant. In Mobile WiMax, the inter-probe duration is randomly generated by the truncated binary exponential random backoff algorithm [18].

7


141

7.4.4 Inter-Probe Power Control The initial power of access probe is predetermined and open-loop power control follows based on the received downlink pilot strength. If an access probe transmission fails, the next probe transmission power is ramped with fixed step size in cdma2000, WCDMA, Mobile WiMax, and IEEE 802.20 systems (Fig. 7.7 (a)). Especially, cdma2000 system adopts open-loop power control together. Meanwhile, 3G LTE system has not determined the power control scheme for access probe transmission but the per-burst open loop power control and zero-step power ramping schemes [7] are being studied as shown in Fig. 7.7. Initial power level

Initial power level

Per-burst open loop power control Power ramping Power

Step size

RA bursts

(a) Equal-step-size power ramping

Power ramping (zero step size)

RA bursts

(b) Zero-step-size power ramping

RA bursts

Time

(c) Per-burst open loop power control

Fig. 7.7 Inter-probe power control schemes addressed in 3G LTE system

7.4.5 Priority In cdma2000, the priority of RA is controlled by the persistent test where the probability value, p is given according to the priority level. If the persistent test fails successively 4/p times, then access probe is immediately transmitted without the persistent test. Similarly, WCDMA uses the persistent test where the probability, p has different 8 values according to access service class (ASC). And RA resource including preamble signatures (i.e. code) sub-channel (i.e. access slot) are allocated according to ASC by using the overlapping and the non-overlapping schemes [19]. In cdma2000 and WCDMA, the persistent test is required at every transmission of access probe. If the test fails, the next test is postponed by a backoff time in order to control the system traffic load and reduce the multiple access interference (MAI) caused by RA [20, 21]. In IEEE 802.20, the persistent test plays similar load control function where the retrial number of test is determined by geometric distributed random variable. Moreover, different 9 orthogonal-code sets are used for access probe according to buffer state of MS and received pilot strength. Meanwhile, Mobile WiMax manages 4 ranging code sets according to the ranging purpose, however, has not adopted any priority scheme. Standardization on RA priority of 3G LTE is in progress.

142


7.5 Performance Evaluation of Ranging in Mobile WiMax In this section, the performance of ranging in Mobile WiMax system is evaluated by link- and MAC-level simulations.

7.5.1 Link-Level Simulation Figure 7.8 shows the link-level model for Mobile WiMax system. Each MS selects and transmits an OFDM modulated access code. The BS receives an aggregated signal r (t) from all users and it is demodulated by FFT (Fast Fourier transform). Then, BS operates parallel auto-correlations upon the received access codes with G candidate access codes. The g-th auto-correlation value between the received signal r (t) and g-th access code cg is given by Ag =

L r (l)cg (l)

(7.1)

l=1

where L is an access code length. An access code is recognized at BS if the autocorrelation value exceeds a decision threshold. Thus, the detection successibility of access code is mainly dependent on threshold values and decision strategies [21]. In this work, we introduce a modified auto-correlation like G· Bg =

L r (l)cg (l)

l=1 L G

r (l)cg (l)

,

g = 1, 2, · · · , G.

(7.2)

g=1 l=1

If the modified auto-correlation is larger than a threshold denoted by Bth , the access code is successfully detected. Otherwise, it fails in detection. MS 1

Access code selection

AWGN OFDM (IFFT)

Fading

Code detection MS 2


OFDM (IFFT)

Fading

r(t) Base station

MS N


OFDM (IFFT)

Fading

Fig. 7.8 Link-level simulation structure

OFDM (FFT)

Autocorrelation process

7


143

7.5.2 MAC-Level Simulation Figure 7.9 shows the ranging procedure of Mobile WiMax. An MS transmits a randomly selected ranging code which was referred as access code in previous subsection. Then, the BS takes the detection process and broadcasts the result. If the detection fails due to deterioration of received signal or ranging collision, MS retries a ranging after a random backoff delay. Conversely, if the detection succeeds, MS transmits a response message to specify the detailed requests and the identification (i.e. MAC ID). Finally, BS allocates subchannels to MS based on MSs’ requests. In this procedure, three time entities are defined: (a) Response recognition time: the elapsed time from transmission of a ranging code to reception of detection result. (b) Success recognition time: the elapsed time from transmission of a ranging code to recognition of raging success by receiving a channel allocation message to specified MAC ID (c) Collision recognition time: the elapsed time from transmission of a ranging code to collision detection (MS is not able to recognize the ranging failure until it receives the erroneous channel allocation message with another MS’s MAC ID). As shown in Fig. 7.10, a retrial queuing model with a finite population is used for ranging traffic model with potential arrival rate λ p and population size M. The potential arrival rate λ p is defined by the ratio of new arrivals to population size in every frame. And the effective arrival rate λe is defined by the sum of new arrival rate λn and retrying rate λr . The ranging failure MSs should enter a buffer and retry ranging after a random backoff delay. Thus, only (M − m) λ p MSs newly try ranging where m represents the number of MSs waiting for retrying in buffer. Mobile WiMax system employs the truncated binary exponential random backoff time between 1 and backoff window Wn as shown in Fig. 7.11. The backoff window Wn doubly (binary) increases up to maximum backoff window Wmax (truncated). Detection success

Detection failure Ranging code transmission Backoff_time

Response_recognition_time

BS

Retrials time

MS Response_recognition_time

Success_recognition_time Collision_recognition_time

Ranging success Ranging failure: random backoff

Fig. 7.9 Simplified ranging procedure and three time entities in Mobile WiMax system

144


λn = (M − m ) ⋅ λ p Population: M

Success

New arrival rate Server

Potential arrival rate ( λ p ) Retrying rate

Failure

λr

Source

Effective arrival rate

λe = λn + λ r

Size: m Buffer

Fig. 7.10 Finite population arrival model

Arrival, Set n=0

Transmitting ranging codes

Success?

Yes

Departure

No

n=n+1

Backoff timer randomly generated between [1, min(Wmax, Wmin*2^(n))] No

Yes

Backoff timer expired?

Waiting

Fig. 7.11 Truncated binary exponential random backoff generation

7.5.3 Numerical Results Table 7.3 summarizes link-level simulation parameters referring to the Mobile WiMax standard [8]. It is assumed that frequency, time, and power are perfectly synchronized and the number of access codes is 60. And only one ranging region is used, thus, every MS transmits a ranging code to the common access region. Ranging codes are generated by the PN code generator 1 + x 1 + x 4 + x 7 + x 15 . Monte Carlo simulation method is used to obtain the detection success probability that an access code is correctly detected when the number of MSs is n, which is simply given by

7


145

Table 7.3 Link-level simulation parameters Parameter

Value

Number of access codes (G) Length of an access code (L) Size of FFT Sub-carrier spacing Duration of OFDM symbol Physical frame length Eb/No Center frequency Velocity Wireless channel model

60 144 bits 1024 9.765625 kHz 115.2 usec 5 msec 20 dB 2.3 GHz 3 km/h ITU-R pedestrian A [22]

PH I T (n) =

The number of MSs correctly detected The number of MSs transmitting an access code

(7.3)

Figure 7.12 shows the detection success probability where it is assumed that the threshold value Bth for successful detection is 2. It is easily shown that PH I T (n) seriously goes down if the number of MSs becomes higher. This is because the MAI severely deteriorates as the number of MSs increases. Especially, when 6 MSs simultaneously try a ranging, the detection success probability PH I T (6) becomes below 50%.

Fig. 7.12 Detection success probability

146

J. Shin and H.-S. Cho Table 7.4 MAC-level simulation parameters Parameter

Value

Response recognition time Collision recognition time Success recognition time Backoff algorithm

4 frames 8 frames 8 frames Truncated binary exponential random backoff 8 frames 64 frames

Minimun backoff window size Maximum backoff window size

Table 7.4 shows MAC-level simulation parameters. It is assumed that Response recognition time, Collision recognition time, and Success recognition time take 4, 8, and 8 frames, respectively. The retransmission timing is generated by truncated binary exponential random backoff algorithm [18, 23]. Minimum and maximum backoff window sizes are 8 and 64 frames, respectively. Figure 7.13 shows the throughput for the various population size and potential arrival rate λ p . To fix the number of potential arrivals under changing population, we define a potential arrival rate index as (λ p /0.01) × (M/200). That is, the average potential arrivals when the index is i is equal to i × 0.01 × 200. The throughput is defined by the average number of RA successes in a frame. As the λ p increases, the throughput increases. In general, a throughput in traffic theory increases up to a cer-

Fig. 7.13 Throughput

7


147

Fig. 7.14 Effective arrival rate λ p

tain level and then turns to decrease as arrival traffic increases due to abrupt increase of collision and interference [24]. In the observation range of our simulation where the arrival rate is sufficiently large, for example, arrival rate index 7 and population 600 corresponds to the number of MSs larger than 20 per 5 msec (Fig. 7.14), the throughput decrease does not appear. Figure 7.15 shows the average delay time elapsed from the first RA trial until RA success. The delay time is normalized by the frame size of 5 msec. As the λ p index and the population size M increases, the delay time asymptotically increases to a certain level. Figure 7.16 shows the ranging success probability at the first try. The larger λ p and M lead to the lower success probability because the higher effective arrival rate λe causes more collisions and severer MAI [20, 21]. In order to improve the performance, the collisions and MAI reduction methods such as enlargement of resource for ranging, priority scheme, and persistent test prior to ranging are needed.

7.6 Summary This chapter has introduced a comparative study on the initialization and RA schemes of 3G and B3G systems. The initialization procedure commonly includes, a cell search using pilot signal, downlink synchronization and system parameter acquisition in a sequence. Following the initialization, RA tries the uplink syn-

148

Fig. 7.15 Average delay time till RA success

Fig. 7.16 Success probability at the first try


7


149

chronization and the bandwidth request through access probe transmission and the response from BS. The fundamental design issues for RA such as inter-probe power control, backoff algorithm, and priority have been addressed to compare the various mobile communication systems. Especially, random access performance of Mobile WiMax has been evaluated with link- and MAC-level simulation in terms of throughput, delay time, and success probability. Acknowledgments This research was supported by Electronics and Telecommunications Research Institute (ETRI), Korea. We would like to thank Dr. Chulsik Yoon and Dr. Sungcheol Chang in WiBro Standardization Team, ETRI for their valuable comments and discussions.

References 1. Rahim Tafazolli, Technologies for the Wireless Future – Wireless World Research Forum (WWRF), Wiley, 2005. 2. Harri Holma and Antti Toskala, WCDMA for UMTS, Wiley, 2004. 3. Samuel C. Yang, 3G CDMA2000 Wireless System Engineering, Artech House, 2004. 4. WiMax Forum, White Paper v2.8, “Mobile WiMax – Part I: A Technical Overview and Performance Evaluation,” August 2006. 5. WiMax Forum, White Paper v3.3, “Mobile WiMax – Part II: A Comparative Analysis,” May 2006. 6. 3GPP, TR 25.913 V7.2.0, “Requirements for Evolved UTRA (E-UTRA) and Evolved UTRAN (E-UTRAN),” March 2006. 7. 3GPP, TR 25.814 V1.3.0, “Physical Layer Aspects for Evolved UTRA,” May 2006. 8. TTA, TTAS.KO-06.0082/R1, “Specifications for 2.3GHz band Portable Internet Service,” December 2005. 9. IEEE, Std 802.16-2004, “IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems,” October 2004. 10. IEEE, Std 802.16e-2005 and Std 802.16-2004/Cor 1-2005, “IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems-Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum 1,” February 2006. 11. IEEE, C802.20-05/68, “QFDD and QTDD: Technology Overviews,” October 2005. 12. IEEE, C802.20-06/04, “MBFDD and MBTDD: Proposed Draft Air Interface Specification,” January 2006. 13. 3GPP2, C.S0024-A. Version 2.0, “cdma2000 High Rate Packet Data Air Interface Specification,” July 2005. 14. 3GPP, TS 25.214 V7.0.0, “Physical layer procedures (FDD),” March 2006. 15. 3GPP, TS 25.211 V7.0.0, “Physical channels and mapping of transport channels onto physical channels (FDD),” March 2006. 16. Jisang You, Kanghee Kim, and Kiseon Kim, “Capacity Evaluation of the OFDMA-CDMA Ranging Subsystem in IEEE 802.16-2004,” in Proceedings Wireless and Mobile Computing, Networking and Communications 2005, vol. 1, pp. 100 – 106, August 2005. 17. 3GPP, TS 25.321 V7.0.0, “Medium Access Control (MAC) protocol specification,” March 2006. 18. Robert M. Metcalfe and David R. Boggs, “Ethernet: Distributed Packet Switching for Local Computer Networks,” Communications of the ACM, vol. 19, no. 7, pp. 395 – 404, July 1976. 19. 3GPP, TS 25.922 V6.3.0, “Radio Resource Management Strategies,” March 2006.

150


20. Jens Muchenheim and Urs Bernhard, “A Framework for Load Control 3rd Generation CDMA Networks,” in Proceedings Global Telecommunications Conference 2001, vol. 6, pp. 3738 – 3742, November 2001. 21. Ki-Nam Kim, et al., “The Scheme to Improve the Performance of Initial Ranging Symbol Detection with Common Ranging Code for OFDMA Systems,” in Proceedings ICACT2006, pp. 183–188, February 2006. 22. ITU-R, M.1225, Guidelines for Evaluation of Radio Transmission Technologies for IMT2000, 1997. 23. Byung-Jae Kwak, et al., “Performance Analysis of Exponential Backoff,” IEEE/ACM Transactions on Networking, vol. 13, no. 2, April 2005. 24. Alberto Leon-Carcia and Indra Widjaja, Communication Networks: Fundamental Concepts and Key Architectures, 2nd edition, McGraw-Hill, 2004.

Chapter 8

An Improved Fast Base Station Switching for IEEE 802.16e with Reuse Partitioning I-Kang Fu, Hsiang-Jung Chiu and Wern-Ho Sheen

Abstract FBSS (fast base station switching) is an important handover mechanism in IEEE 802.16e whose OFDMA (orthogonal frequency division multiple access) mode has been adopted as the mobile WiMax technology. By using one radio-link connection and multiple network connections for the handover user, FBSS strikes a good balance between complexity and handover performance, as compared to the hard handover and macro diversity handover in IEEE 802.16e. In this paper, an FBSS with reuse partitioning cell structure is proposed to improve the performance of the traditional FBSS. By reserving some of the radio resource for use of high reuse factor, and using that radio resource to accommodate those handover users with bad radio-link performance, packet loss rate of FBSS can be reduced by a ratio from 38.33 to 84.08%, at the slight expense of 1.67–4.21% cost on average cell throughput. Keywords Fast base station switching · Reuse partitioning · IEEE 802.16e · Mobile WiMax

8.1 Introduction Broadband mobile communication is targeted to support multimedia services over a variety of environments such as indoors, outdoors, low-mobility, high-mobility, etc. Data rate up to several tens Mbps is essential in order to support a multitude of services and QoS requirements [1]. OFDM (orthogonal frequency division multiplexing) is an effective modulation/multiplexing scheme for broadband communication for its ability to overcome severe inter-symbol interference (ISI) incurred by high-data-rate transmission [2]. By using parallel orthogonal sub-carriers along with cyclic-prefix, ISI can be removed completely as long as the cyclic-prefix is larger than the maximum delay I-K. Fu (B) Department of Communication Engineering, National Chiao Tung University, Hsinchu, 300 Taiwan, ROC e-mail: [email protected]


151

152

I.-K. Fu et al.

spread. OFDM can also be employed as an effective multiple access scheme in a multi-cell environment [2]. In particular, OFDMA (orthogonal frequency division multiple access), a form of OFDM in combination with FDMA, has been recognized as one of the most promising multiple access schemes for broadband mobile communication [3]. In fact, the OFDMA mode of IEEE 802.16e [4] has been adopted as the mobile WiMax technology [5]. There are three kinds of handover defined in IEEE 802.16e: hard handover, macro diversity handover (MDHO) and fast base station switching (FBSS). In hard handover, the system maintains only one connection in both the network and radio-link sections for the mobile station (MS), and therefore the network and radio-link connections to the target base station (BS) will be established only after breaking the existing ones. Hard handover is very simple, but services will be disrupted for a period of time needed for the establishment of the new connections. Contrary to hard handover, the system maintains multiple network and radio-link connections at the same time for the MS in MDHO. Therefore, both the network and radio-link connections to the target base stations can be established before breaking the existing ones. Since MDHO is essentially a soft handover, the service disruption can be eliminated by having multiple network and radio-link connections simultaneously. MDHO needs more than one receiver in the MS and hence increases its complexity. In addition, two copies of radio resource are needed for handover users and that leads to lower system spectrum efficiency and higher downlink interference. FBSS, on the other hand, takes advantage of the low service disruption time from MDHO and low MS complexity from hard handover [4, 6]. At network section, FBSS establishes connections with potential target BSs before breaking the existing one, so that the service disruption time can be reduced like in MDHO. Meanwhile, only one connection will be maintained at the radio-link section so as to keep low MS complexity as in hard handover. By switching between BSs fast enough, the MS can maintain its link performance and explore macro diversity gain. In FBSS, the radio-link will not be switched to the target BS before the establishment of the new network connection to the target BS; otherwise, packets might be lost and/or become obsolete. This is especially important for real-time services. Unfortunately, the time needed for the establishment of a network connection is uncertain (a random variable) and cannot be known in advance [7]; to initiate the establishment of network connection too early will waste the network resource, but too late the radio-link performance might be degraded to an unacceptable level and that incurs packet loss. In this paper, an improved FBSS with reuse petitioning (RP) cell structure is proposed for IEEE 802.16e. By reserving some of the radio resource for use of high reuse factor, and using that radio resource to accommodate those handover users with bad radio-link performance, the packet loss rate can be substantially reduced, at the slight expense on average cell throughput. The rest of this paper is organized as follows. The FBSS in IEEE 802.16e is presented in Section 8.2. The new FBSS with reuse partitioning is proposed in Section 8.3. Simulation results are given in Section 8.4, and the paper is concluded in Section 8.5.

8

An Improved Fast Base Station Switching

153

8.2 Fast Base Station Switching in IEEE 802.16e In the IEEE 802.16e system, an FBSS handover begins with a decision for an MS to transmit/receive data to/from the Anchor BS that may change within the diversity set [4]. Diversity set is a set containing a list of active BSs that are informed of the MS capabilities, security parameters, service flows and full MAC context information, and the Anchor BS is the BS in the diversity set that is designated to transmit/receive data to/from the MS at a given frame. Based on the received signal quality, the MS in FBSS handover can fast switch the Anchor BS so as to obtain the macro diversity gain. For good performance, the MS can scan the neighbor BSs and select those suitable ones to be included in the diversity set (diversity set selection/update), and the MS shall select the best BS from its current diversity set to be the Anchor BS (Anchor BS selection/update) [4]. Figures 8.1 and 8.2 are used to explain the FBSS operation in IEEE 802.16e in details. For simplicity only two BSs are considered. Assume the MS is moving from B S1 toward B S2 , and B S1 is the Anchor BS at the beginning. In IEEE 802.16e, the parameter H Add, a threshold used to trigger diversity set selection/update, is broadcasted through DCD (Downlink Channel Descriptor), and the characteristics of neighbor BSs (B S2 in this case) including BSID, PHY Profile ID, Preamble Index, etc. through the MOB NBR-ADV message. Base on the information, the MS may send MOB SCN-REQ to B S1 and get response from MOB SCN-RSP for requesting a period of time to facilitate scanning and/or association (an optional

Fig. 8.1 Cell structure, received radio-link signals, diversity set membership and Anchor BS selection of an MS involved in FBSS in IEEE 802.16e

154

I.-K. Fu et al.

Fig. 8.2 The message flow of FBSS in IEEE 802.16e

initial ranging) of B S2 , as shown in Fig. 8.2. When the CINR (Carrier to Interference plus Noise Ratio) difference of B S1 and B S2 is less than H Add, the MS sends MOB MSHO-REQ with a recommended BS list including B S2 . If B S2 is also in the recommended list in MOB BSHO-RSP, which is replied by B S1 , then the MS sends MOB HO-IND to request to add B S2 to the diversity set, i.e., to initiate the diversity selection/update procedure. In Fig. 8.1, this first happens at the time instant A. After that, B S1 sends HO-Request to B S2 through wire-line network to establish the network connection, and B S2 will reply HO-Response to B S1 if the establishment is complete. Accordingly, B S1 can update the MS new diversity set members through MOB BSHO-RSP. When the MS keeps moving and if CINR B S2 is higher than CINR B S1 , for example at the time instant B in Fig. 8.1, the MS may send MOB MSHO-REQ to request to

8


155

change Anchor BS from B S1 to B S2 (initiation of the Anchor BS selection/update procedure). The request, however, will not be granted in this example until the time instant C because the network connection to B S2 is not yet established (for some reason) before that instant, and B S2 will not be included in the diversity set. If the request is granted, through MOB BSHO-RSP, the MS will send MOB HO-IND to terminate the existing radio link connection and then perform fast ranging with B S2 , see Fig. 8.2. Note that B S2 needs to reserve an uplink contention-free ranging sub-channel for the MS and place Fast Ranging IE in the extended UIUC (Uplink Interval Usage Code) in a UL-MAP IE (Information Element) to inform the MS this ranging opportunity. The fast ranging process can be accomplished in two frames, where the uplink ranging opportunity is indicated by the downlink MAP in the first downlink sub-frame, and then the MS sends RNG-REQ in the successive uplink sub-frame based on the radio parameters recorded in the scanning interval. Then B S2 replies RNG-RSP along with the correction commends encoded in TLV formats in the second frame. After that, the MS begins to transmit/receive data to/from B S2 . As is mentioned, the MS is not able to change its Anchor BS to B S2 until the time instant C, although CINR B S2 is already higher than CINR B S1 at the time instant B. During the time period between B and C, the MS still talks toB S1 but with a degraded link performance, and that may result in packet loss. One simple remedy to this problem is to use a larger H Add; in other words, the request for diversity set update is initiated earlier. This, however, may waste network resource if B S2 is put into the diversity set too early. In the next section, an FBSS with reuse partitioning cell structure is proposed to improve the handover performance at the expense of slight loss in average cell throughput.

8.3 The Improved FBSS with Reuse Partitioning Reuse partitioning (RP) is a cell structure in which a regular cell is divided (ideally) into two or more concentric cell-regions, each with a different frequency reuse factor [8, 9]. This also implies that the radio resource of a cell has to be divided into the same number of resource-regions. A smaller reuse factor (small reuse distance) is allowed for the inner cell-regions because of the smaller transmit power, and a larger reuse factor is needed for the outer cell-regions so as to maintain the required signal quality. By allowing the inner regions to use a smaller reuse factor leads to a higher system capacity, as compared to the regular cell structure where the same reuse factor is used for the entire cell [9]. In this paper, the concept of reuse partitioning is used to increase the handover performance.

8.3.1 Reuse Partitioning in IEEE 802.16e Figure 8.3(a) shows a simplified TDD frame structure in IEEE 802.16e. Preamble, FCH (Frame Control Header), downlink MAP and uplink MAP are control signals

156

I.-K. Fu et al.

Fig. 8.3 A simplified IEEE 802.16e TDD frame structure with (a) regular K = 4 resource zone, and (b) regular K = 4 resource zone and K = 7 RP zone

of the frame. The cell-specific Preamble is mainly used for downlink synchronization. FCH contains DL Frame Prefix that indicates the length and coding scheme of the DL MAP message. DL MAP and UL MAP are MAC messages that define the starting point of the downlink and uplink bursts, respectively. In IEEE 802.16e, frequency reuse with factor K can be achieved by dividing the radio resource in a frame into K resource-regions and each one of them is allocated to different BS. In Fig. 8.3(a), K is equal to 4 so that the radio resource (both downlink and uplink) is divided into 4 resource-regions. Note that BSs need to be synchronized and follow the same sub-carrier permutation rule for this scheme to work [4]. In order to design a reuse partitioning scheme, we adopt the concept of resource zone in IEEE 802.16e [4]. As an example, in addition to the regular resource-zone for K = 4, a reuse partitioning (RP) zone is defined for K = 7, as shown in

8


157

Fig. 8.3(b). Again, within each respective zone, the radio resource is divided into K resource-regions and each one of them is allotted to different BS.

8.3.2 FBSS with Reuse Partitioning The concept of reuse partitioning is used here to increase the performance of FBSS handover. The basic idea is as follows. Under the reuse partitioning cell structure, the handover users who are in a bad channel condition are scheduled to a resourceregion with a large reuse factor so that a better CINR can be maintained, and therefore the packet loss rate is reduced. This method is very effective for the handover users who are waiting for the target BS to establish the network connection, as discussed in the previous section. Figure 8.4 shows the RP cell structure and the received radio-link signals, diversity set membership and Anchor BS of an MS involved in FBSS. K = 4 for the

Fig. 8.4 Reuse partitioning cell structure, received radio-link signals, diversity membership, Anchor BS selection and scheduled zones of an MS in FBSS with reuse partitioning

158

I.-K. Fu et al.

inner cell-region and K = 7 for the outer one with the resource-regions given in Fig. 8.3(b). In the K = 7 resource-region, since a large reuse factor is used, the received CINR of an MS is higher as compared to the K = 4 resource-region. Initially, the MS talks to B S1 by using resource in the K = 4 resource-region. As discussed in the previous section, during the time between B and C, although CINR B S2 is higher than CINR B S1 , the MS still talks to B S1 since the network connection to B S2 is not ready yet. In the case of using reuse partitioning, the difference from Fig. 8.1 is that now we have K = 7 resource-zone, and the handover users going through this period of time can be re-scheduled to that region to improve the radio-link performance if its CINR is less than the threshold ρr eq . Therefore, the packet loss during this period will be mitigated and the packet loss rate can be substantially reduced.

8.4 Simulation Results In this section, simulation results are given to illustrate the handover performance of FBSS with and without reuse partitioning.

8.4.1 Simulation Model The IEEE 802.16e downlink OFDMA system is simulated in a multi-cell urban environment, where cells are assumed to be synchronized to each other. Total of 19 cells is simulated with 1 km cell coverage under 15 Watts transmit power. K = 4 for the regular resource-zone and K = 7 for the RP resource-zone. Each cell has three sectors. The OFDM PHY parameters are given in Table 8.1. Table 8.1 Parameters for system-level simulation Parameter

Value

Channel bandwidth FFT size OFDM symbol duration (including cyclic-prefix) Cyclic-prefix Sub-carrier frequency spacing Frame duration Downlink sub-frame duration Sub-carrier permutation rule Number of bins Number of sub-carriers in a bin Number of sub-carriers for data transmission Number of sub-carriers for guard band Sub-channel definition Maximum diversity set size Threshold to schedule MS into the RP zone, ρr eq . Time to establish network connection

6 MHz 2048 336 ␮s 37.333 ␮s 3.348 kHz 20 ms 10 ms Adjacent sub-carrier permutation 192 9 1728 320 (2048–1728) 2 adjacent bins × 3 adjacent symbols 3 6 dB Uniform distributed between 200 ms ∼ 700 ms [7]

8


159

At the beginning of the simulation, MSs are generated by Poisson processes and located randomly in a cell. The path loss to every BS is calculated for each MS, and the log-normal shadow fading (with de-correlation distance of 50 m) and frequency-selective fading are generated according to the models given in [10]. An MS will send a service request to the BS with the highest effective CINR, and the request will be granted if there is radio resource available. Otherwise, the MS will be blocked and removed in the simulator. The effective CINR is evaluated on the received preamble signal part by the EESM (Exponential Effective SIR Mapping) criterion [11]. The location of an MS is periodically updated based on ITU vehicular mobility model [12] with mobility of 50km/hr. According to the mechanism proposed in Section 8.3, the diversity set of an MS may be updated according to the CINR variation. Moreover, the anchor BS is the BS in diversity set with the highest CINR, and the MS only transmits/receives the radio signal to/from the anchor BS. The G.729 VoIP traffic model is adopted in the simulation. The VoIP packet arrives every 20 ms with a packet size of 640 bits. The packet error rate is determined by the received SINR of the data traffic part and can be obtained by looking up table given in [4].

8.4.2 Simulation Results The performances of FBSS are given in Fig. 8.5, where PR P is the percentage of radio resource allotted to the K = 7 resource-region. Three sets of results are given including packet loss rate, average diversity set size and cell throughput. In all of the results PR P = 0 represents the case of traditional FBSS; that implies no RP resource-region is reserved. Figure 8.5(a) shows the result of packet loss rate. As can be seen, PR P should be large enough, says more than 5%, in order to obtain a sizable reduction on packet loss rate. More specifically, 38.33% to 84.08% reductions are achievable by increasing PR P from 7.29% to 21.88%, with respect to the case PR P = 0. Note that the packet loss rate reduction becomes saturated if PR P is larger than around 21.88%. In addition, for a PR P , a lower packet loss rate is obtained with a larger H Add. Figure 8.5(b) shows the average diversity set size, which is an indicator on the network resource usage. As expected, the diversity set size is unchanged with different PR P for a specific H Add. On the other hand, when increasing H Add, the neighbor BSs might be added to the diversity set too early before the MS really needs to change its Anchor BS. Therefore the average diversity set size is larger and more network resource is consumed. The cell throughput is shown in Fig. 8.5(c). Since frequency reuse factor is larger in RP zone, the larger PR P , the lower the cell throughput. The results show that less than 4.21% cell throughput will be lost even when PR P is increased to 21.88%. It is because the reduction on packet loss rate can increase the effective packets received by the MS, which can increase the cell throughput.

160 Fig. 8.5 The performances of FBSS with reuse partitioning

I.-K. Fu et al.

8


161

8.5 Conclusions In this paper, an FBSS with reuse partitioning cell structure is proposed to improve the performance of the traditional FBSS in IEEE 802.16e. By reserving some of the radio resource for use of high reuse factor, and using that radio resource to accommodate those handover users with bad radio-link performance, the packet loss rate of FBSS can be reduced by 38.33% to 84.08%, at the slight expense of 1.67% to 4.21% cost on cell throughput. Compared with the traditional method, using reuse partitioning cell structure can achieve more significant reduction on packet loss rate for FBSS in the IEEE 802.16e system.

References 1. Recommendation ITU-R, “Framework and overall objectives of the future development of IMT-2000 and systems beyond IMT-2000,” International Telecommunication Union, June 2003. 2. R. Van Nee and R. Prasad, “OFDM for Wireless Multimedia Communications,” Boston: Artech House, 2000. 3. H. Yang, “A road to future broadband wireless access: MIMO-OFDM-based air interface,” IEEE Communications Magazine, Vol. 43, Issue. 1, pp. 53–60, January 2005. 4. IEEE 802.16e-2005, “IEEE standard for Local and Metropolitan Area Networks, Part 16: Air interface for fixed and mobile broadband wireless access systems, amendment for physical and medium access control layers for combined fixed and mobile operation in licensed bands,” February 2006. 5. WiMax Forum, “Mobile WiMax Part I: A Technical Overview and Performance Evaluation,” August 2006. http://www.wimaxforum.org/news/downloads/ 6. S. Choi, G-H Hwang, T. Kwon, A-R Lim, and D-H Cho, “Fast Handover Scheme for RealTime Downlink Services in IEEE 802.16e BWA System,” IEEE Vehicular Technology Conference, Vol. 3, pp. 2028–2032, June 2005. 7. F. Feng and D. S. Reeves, “Explicit Proactive Handoff with Motion Prediction for Mobile IP,” IEEE Wireless Communications and Networking Conference, Vol. 2, pp. 855–860, March 2004. 8. 3GPP R1-05-0407, “Interference Coordination in new OFDM DL air interface,” 3GPP TSG RAN WG1 #41, May 2005. 9. T-P Chu and S. S. Rappaport, “Overlapping Coverage with Reuse Partitioning in Cellular Communication Systems,” IEEE Transactions on Vehicular Technology, Vol. 46, No. 1, pp. 41–54, February 1997. 10. I-K Fu, W. Wong, D. Chen, P. Wang, M. Hart and S. Vadgama, “Path-loss and Shadow Fading Models for IEEE 802.16j Relay Task Group,” IEEE C802.16j-06/045, July 2006. 11. WiMax Forum, “WiMax System Evaluation Methodology V1.0,” January 2007. 12. UMTS 30.03, “Universal Mobile Telecommunications System (UMTS); Selection Procedures for the Choice of Radio Transmission Technologies of the UMTS,” TR 101 112 V3.2.0, April 1998.

Chapter 9

Fast Handover Schemes in IEEE 802.16E Broadband Wireless Access System Qi Lu, Maode Ma and Hui Ming Liew

Abstract IEEE 802.16e is a promising system to provide broadband wireless access for wide area mobile communications. As the enhanced version of IEEE 802.16, with the mobility support, it becomes a potential candidate to satisfy the requirements of high data rate and wide coverage in the next generation wireless communication system. However, the handover procedures specified by the standard may cause large handover delay, which would not be suitable for the service of real-time applications. To improve the system performance and reduce the handover delay, several fast handover schemes have been proposed. In this chapter, an overview of the basic handover modes and the handover procedure defined in the standard IEEE 802.16e is presented. Some fast handover schemes, which have been proposed to reduce the handover latency, are reviewed. A more complex relay system and the corresponding new MAC frame structure to support the network relay are discussed. A handover scheme to reduce the handover latency in the WiMax relay system is summarized. The concept to construct a moving network and the fast handover scheme to support the network mobility is discussed. Keywords Handover · IEEE 802.16e · WiMax · Network mobility

9.1 Introduction IEEE 802.16e wireless metropolitan area networks support combined fixed and mobile broadband wireless access. As the enhanced version to previous IEEE 802.16 system, IEEE 802.16e can provide services for subscriber stations moving at vehicular speeds within the licensed bands below 6 GHz [1]. The standard specifies the Physical (PHY) and Medium Access Control (MAC) layers for system operation to support wireless services. Based on the orthogonal frequency division multiple access (OFDMA) technology, IEEE 802.16e system is designed to target

Q. Lu (B) School of Electrical and Electronic Engineering, Nanyang Technological Unversity, Singapore


163

164

Q. Lu et al.

non-line-of-sight (NLOS) communications with features such as high data transmission rate, wide coverage and mobility support. In such wireless systems, to support mobile users, the handover scenarios should be fully considered. Here, the handover is defined as a Mobile Station (MS) migrates from the air interface provided by one Base Station (BS) to the air interface provided by another BS [1]. Which means the MS is moving across the boundaries of air interfaces of two BSs, and the area of this air interface is usually called as a cell. As defined in IEEE 802.16e standard, three basic handover modes are supported to enable continuous data transmission and services when a MS moves across the cell boundaries of BSs, and they are: Hard Handover (HHO), Macro Diversity Handover (MDHO) and Fast Base Station Switching (FBSS). Among these handover modes, HHO is mandatory in IEEE 802.16e system, while MDHO and FBSS are optional. HHO adopts a break-before-make approach, which is less complex and easier to implement than MDHO and FBSS. However, it may introduce a long handover delay and affect the performance of delay sensitive services. To solve this problem, several fast handover schemes are proposed. In [2], a target BS selection strategy was suggested, which only selects one neighbor BS for scanning or association, and it can reduce the redundant scanning and association processes. In [3], a fast handover scheme for real-time downlink (DL) services was proposed, which defines a Fast DL MAP IE message to support DL traffic during handover process. In [4], a solution named as “Passport Handover” scheme was introduced, which enables data transmission during handover and shortens the service interruption with the proposed connection identity (CID) handling mechanism. As the enhancement to the basic system structure, the concept of Relay Stations (RSs) has been introduced into the WiMax network design to achieve the coverage extension or the capacity increment. To realize the idea to support RSs in the IEEE 802.16 networks, new MAC frame structures have been proposed in [5] for both distributed control of relays and central control of relays. Since a handover can also happen in the relay networks, one handover scheme to reduce the handover latency was proposed in [6], which reduces the occurrence of inter-cell handovers and tries to perform intra-cell handovers when two types of handovers are available. Different from the generic handover scenario with an individual MS moving which is regarded as node mobility, a more complex architecture with a group of MSs moving simultaneously is called network mobility, which has also received a lot of attention recently. For the handover processes in the WiMax relay networks, the concept of an Multi-hop Relay architecture has been introduced in [7], where a mobile RS is essentially to be connected by other MSs to construct a moving network and achieve the network mobility. In [7], a proposal of fast handover scheme to reduce the handover delay has also been detailed for the network mobility in the relay networks. The rest of this chapter is organized as follows. Section 9.2 gives an overview of the three basic handover modes and the standard handover procedures specified in the standard first. Then various proposals of fast handover schemes to enhance the standard handover operation have been described. In Section 9.3, the WiMax relay system and corresponding MAC frame structures are investigated. The possi-

9

Fast Handover Schemes

165

ble improvement on handover to reduce handover delay in relay system is discussed. Section 9.4 describes a system architecture of network mobility with relay structure, and the fast handover scheme for network mobility is reviewed. Finally, a summary and conclusion is given in Section 9.5.

9.2 Standard Handover and Improvements In this section, the three handover modes supported in IEEE 802.16e WiMax networks are introduced. Later, the features and the differences of each mode are briefly discussed. Then, an overview of the standard hard handover procedure is presented to show the detailed steps by using a flow chart. The options to the standard handover steps are also discussed. Finally, the state-of-the-art fast handover schemes proposed to speed up the handover procedure and improve the system performance are reviewed with comparison.

9.2.1 Handover Modes in IEEE 802.16e Networks To support a MS moving across cell boundaries of several BSs while maintaining data transmission, the handover operation is required. According to the IEEE 802.16e, the mandatory handover mode is Hard Handover. Besides this, Macro Diversity Handover and Fast Base Station Switching are also adopted by the standard, although they are optional. These three modes are illustrated in Fig. 9.1. In Hard Handover, the adjacent BSs use different frequencies for the information transmission, and a MS connects to only one BS at a time. In the case that a handover is required, the current connection with serving BS should be broken before new connection with target BS is established. This handover process is initialized when the signal level from neighbor BS exceeds the signal level from current serving BS for certain threshold. It is simple and easy to implement, but high latency may be caused and this latency may interrupt the service of delay sensitive applications. In Macro Diversity Handover, all the BSs use only one frequency for information delivery. A Diversity Set, which includes several BSs involved in the handover process, is maintained for the MS, which can communicate with all the BSs in this Diversity Set simultaneously. During the handover process, some diversity techniques could also be applied. The Diversity Set is maintained and updated based on the long-term statistical signal strength of BSs. When the long-term statistical signal strength of an active BS is less than H Delete Threshold, this BS will be dropped from the Diversity Set, while if long-term statistical signal strength of a neighbor BS is greater than H Add Threshold, this BS will be added to the Diversity Set. Among these BSs listed in the Diversity Set, one of them is selected as the Anchor BS. The MS is only synchronized and registered to this Anchor BS for the transmission of management messages.

166

Q. Lu et al.

Fig. 9.1 Handover modes

In Fast Base Station Switching, similar to Macro Diversity Handover, a Diversity Set for MSs is maintained. But the MSs can only connect to the Anchor BS for data transmission and management messages transmission. The Anchor BS can be changed during operation according to Anchor BS selection requirement. And the new Anchor BS and Diversity Set can be updated according. The switching of the connection from current Anchor BS to another BS in the Diversity Set can be intialized when a handover is required. Although the IEEE 802.16e supports these three handover modes, the MDHO and FBSS require more complex handover procedures and special configuration of the network architecture. In the standard system, the hard handover is mandatory and preferred.

9.2.2 Standard Hard Handover Procedures The standard handover process includes the following steps of cell reselection, HO decision & initiation, termination with the serving BS and network entry/re-entry. A brief overview of these steps is given in the following paragraphs. In the cell reselection phase, a MS may scan or associate with its neighbor BSs to determine the suitable one as the target BS for performing handover. This step is carried out before the handover request is made. In the system, the serving BS broad-

9


167

casts the topology information and the channel information of neighbor BSs with Mobile Neighbor Advertisement (MOB NBR-ADV) message periodically. The MS can acquire this information about neighbor BSs and use it for the consideration of cell reselection before scanning process is started. Later, the MS may use Mobile Scanning Request (MOB SCN-REQ) to request allocation of scanning intervals. Then, these scanning intervals are allocated and acknowledged by serving BS via Mobile Scanning Response (MOB SCN-RSP) to initiate the scanning process. During the scanning process, the MS measures the channel quality or signal strength of each neighbor BS. Some neighbor BSs may be selected as candidate BSs for the later actual handover. The cell reselection is performed prior to actual HO, and at this stage, the connection to the serving BS is still maintained. In HO decision & initiation phase, a handover process is initiated by Mobile MS Handover Request message (MOB MSHO-REQ) or Mobile BS Handover Request message (MOB BSHO-REQ) when the conditions to perform handover are satisfied. Both of the MS and the serving BS can request a handover activity. If the handover is requested by the MS, the MS may send MOB MSHO-REQ and indicates the possible target BSs based on the performance evaluation from previous scanning or association. The serving BS may negotiate with the recommended target BSs via backbone network, and it sends acknowledgement to the MS with Mobile BS Handover Response message (MOB-BSHO-RSP). If the handover is requested by the serving BS, the serving BS may send MOB BSHO-REQ to the MS, in which the suitable neighbor BSs are recommended. The MS can conduct handover to one of the recommended BSs, or reject this recommendation and attempt to perform handover to some other BSs. During the message exchange, dedicated ranging opportunity may be allocated to speed up ranging process for later network re-entry. Finally, Mobile Handover Indication message (MOB HO-IND) is issued to indicate the releasing of serving BS. After the exchange information of handover request and response, the MS terminates the connection with serving BS by sending MOB HO-IND, and it starts network entry/re-entry step. The whole network re-entry process includes ranging, re-authorization and re-registration. In this phase, the MS needs to synchronize with downlink transmission and obtain downlink and uplink transmission parameters with target BS. Then Ranging Request message (RNG-REQ) will be sent to start a ranging process, dedicated ranging opportunity may be available if it is allocated in the previous step, which can avoid contention-based ranging. Later, Ranging Response message (RNG-RSP) is transmitted, and in which the re-entry management messages that can be omitted are indicated. After the channel parameters are adjusted, the MS can communicate with target BS to negotiate channel capability, perform authorization and conduct registration. Some information about the MS may be transferred from the serving BS to target BS via backbone network to speed up this process. The handover procedure is completed thereafter, and the data transmission between the MS and the new serving BS could be started. An example flow chart of standard handover is shown in Fig. 9.2, and an example of handover procedure at MAC layer is given in Fig. 9.3 [1].

168

Q. Lu et al.

Fig. 9.2 Handover flow chart

During the standard handover process, some assisting schemes are provided for optional adoption. All of these schemes may be used to provide alternative handover support or to speed up the handover process. In the association procedure, three levels of association can be performed according to the system configuration, and they are scan/association without coordination, association with coordination and network assisted association reporting. When the scan/association without coordination is used, the target BS has no knowledge about the MS, and only contention-

Fig. 9.3 MAC layer handover procedure

9


169

based ranging can be conducted. The MS randomly picks up a contention-based ranging code from the domain of target BS and transmits it during the contentionbased ranging interval. After the target BS receiving the ranging code, the uplink (UL) allocation would be provided to the MS to conduct association. If the association with coordination is chosen, the serving BS would coordinate the ranging process and provide the association parameters of target BS to the MS. By this method, dedicated ranging code and transmission opportunity would be assigned by the target BS, and it would be informed to the MS by serving BS. In the network assisted association reporting, the MS only transmits the ranging code to target BSs while the information about the channels from each neighbor BS would be sent to the serving BS via backbone network. The required information would be aggregated and sent to the MS later by the serving BS in the form of a report message Mobile Association Report (MOB ASC REPORT). Other options are also available as follows. Before the ranging process in the network re-entry step, Fast Ranging IE may be used to allocate a non-contention-based initial ranging opportunity in the case that target BS can receive handover notification from the serving BS via backbone network. And also in the network re-entry step, some information of MS can be transferred through the backbone network, and it allows skipping transmission of some re-entry management messages, such as capability negotiation, authentication and etc. All of these options may be used to improve the handover performance or to reduce the handover delay.

9.2.3 Proposed Fast Handover Schemes Although the standard hard handover process is able to maintain data transmission for a MS moving across cell boundaries, it still may cause serious interruption for delay sensitive services. To further reduce the handover latency, especially to lower the interruption for the real-time applications, several fast handover schemes are proposed in various research publications. In [2], a target BS selection algorithm is proposed to reduce redundant scanning or association during the scanning process and expedite the cell reselection procedure. When scanning process starts, the data transmission would be paused during the neighbor BS scanning or association. It may interrupt the services seriously, if the number of neighbor BSs needs to be scanned or associated is large, the time spent on scanning process will be long. In normal scanning process, MS needs to scan or associate with neighbor BSs one by one, and the scanning time increases as the increasing of the number of neighbor BSs. The proposed algorithm suggests to acquire physical information of neighbor BSs and to estimate the mean carrier to interference-plus-noise ratio (CINR) of each neighbor BS. And the BS with larger mean CINR is more likely to be the target BS. Later the MS needs to conduct the scanning process with only one selected BS. The whole scanning process would only contain one scan or association activity and the time spent on redundant association processes is reduced. An example of the improvement by the proposed

170

Q. Lu et al.

Fig. 9.4 Single BS scanning procedure

scheme is shown in Fig. 9.4 when Neighbor BS 1 is selected as the only target BS. The Neighbor BS 1 would be scanned or associated while Neighbor BS 2 would not be, and the total time for scanning is approximately reduced by half. In [3], the proposed scheme enables the DL data transmission immediately after DL synchronization with target BS. It can reduce the service disruption for DL applications. In normal case, the MS can receive data only in the normal operation mode after the whole handover process is finished, and the long handover process may cause data loss, which will impair real-time services much and degrade the system performance. A Fast DL MAP IE is defined in this suggested scheme, which contains the MAC address of the MS and resource allocation information for DL data transmission. During the handover procedure, the target BS can transmit data using Fast DL MAP IE to the MS with old CID, only after the downlink synchronization without the UL synchronization until new CID is updated. The proposed solution can shorten the delay in data transmission to MS. To maintain backward compatibility with normal handover procedure, one reserved bit in the generic MAC header is defined as the fast DL indication bit, which is used to indicate the existence of data that requires fast DL transmission. When a handover process occurs, the fast DL indication bit should be set to one for the DL data, which need to be fast transmitted. Then the data can be transmitted by the target BS using the Fast DL MAP IE during the handover. For other data with the zero value in the fast downlink indication bit, they would be transmitted after finishing the handover and returning to normal operation mode. The Fig. 9.5 shows an example that handover is performed from the serving BS to the target BS, and the target BS transmits fast DL data to the MS after DL synchronization. In [4], a so-called Passport Handover scheme is proposed to allow real-time applications to start downlink transmission and uplink transmission before the handover process is completed. Then the MS can communicate with target BS continuously during network re-entry process without stop of the services. In this scheme, the CID assigned by serving BS will be also accepted by target BS for

9


171

Fig. 9.5 Handover procedure for fast downlink transmission

data transmission. Similar as [3], the downlink transmission is resumed just after downlink synchronization using the old CID. But this scheme also allows the uplink transmission to be continued with old CID just after ranging response and before the authorization process. Until the handover process is over, the data transmission is conducted using updated new CID. To avoid collision of old CID with active CIDs used by target BS during the handover process, a CID assignment method is provided in [4]. Since by this scheme, a group of CIDs should be reserved by both serving BS and neighbor BS as the passport CID, the number of these available CIDs may not enough for handling all connections so that this scheme is supposed to be used only for delay sensitive applications. For other non-delay sensitive services, the regular handover would be performed. An example flow of using this fast handover scheme is shown in Fig. 9.6. The downlink data transmission is conducted just after downlink synchronization before ranging. And the uplink transmission is carried out only after ranging with uplink synchronization and before the network re-entry. The first BS selection scheme is mainly to reduce delay in the scanning of neighbor BSs during the cell reselection procedure before the actual handover request is sent. The basic idea is to reduce the redundant scanning or association processes and select only one target BS to be scanned or associated based on the estimated CINR from each neighbor BS. Although this strategy can reduce lots of time spent on multiple rounds of scanning or association, it has not provided a feasible way to estimate the CINR from neighbor BSs to the MS. For the second and third schemes, both of them are proposed to make data transmission possible during the handover process. So they may be used to improve the performance and reduce the service interruption for real-time applications. The second fast handover scheme could only realize downlink data transmission during the handover, therefore, it is only applicable to the real-time downlink services such video streaming, multimedia radio, and etc. However, the third fast handover scheme, which can achieve data transmission for either downlink or uplink, could provide much more benefits to help different types

172

Q. Lu et al.

Fig. 9.6 Passport handover procedure

of applications and improve performance of the networks in the handover process. But due to limited number of CIDs could be reserved, this scheme is recommended for mainly handling real-time applications. It is clear that the third scheme would be much better with wider usage than the second scheme because it is applicable to all types of real-time applications rather than only downlink services. And also the third scheme can be implemented with the first scheme together to achieve overall better performance in the handover process because they function at the different stages of the handover.

9.3 Relay Systems and A Handover Scheme In this section, the relay technique and the structure of relay system will be reviewed. The MAC frame structures used in the relay systems are discussed. And finally, one fast handover scheme proposed to shorten the handover process in the relay systems is studied.

9.3.1 Relay Systems The purposes to introduce relay topology into WiMax system are mainly to extend the network coverage and to increase the system capacity. In such WiMax relay systems, one or more RSs are required to serve as relay servers. Generally, a RS is a simplified version of a BS, and it can be fixed or mobile. A fix RS is installed at a certain position permanently while a mobile RS can move within certain region. All of them are connected to the BS via wireless radio interface. The scenarios that RSs are used to extend network coverage and to increase system capacity are shown in Figs. 9.7 and 9.8, respectively.

9


173

Fig. 9.7 RSs to extend network coverage

When RSs are used to extend the network coverage, a RS is usually located at the boundary of last RS or BS. Therefore, the BS can establish connection with the MS, which is far away and outside the BS’s own coverage. The communication between the BS and the MS is conducted via one or more RSs in the path. In this system relay structure, the RS can deliver data to the MSs within its coverage or forward data to next RS. When RSs are used to increase system capacity, a RS is placed within the coverage of the BS. The MS can connect to its local RS, and then communicates with the BS. So the total simultaneous connections are increased. Similar as previous scenario, the data transfer between the BS and the MS is forwarded by the RS. To implement the relay system in the WiMax system specified by IEEE 802.16e standard, some modifications are required. An important change on the MAC frame structure to support the relay networks was proposed in [5]. The solution enables multi-hop data transmission from the BS to the MS via multiple RSs. In the relay system, handovers also exist when the MS moves across the cell boundaries of RSs or BSs. A solution to reduce the handover delay in such relay systems was provided in [6] by reducing the inter-cell handover events. Moreover, if a mobile RS is installed, the solution can be further extended to construct a moving network with several mobile devices in motion simultaneously to support network mobility

Fig. 9.8 RSs to increase system capacity

174

Q. Lu et al.

as described in [7]. In this system, the MSs are connected to a Mobile Relay (MR), which is regarded as a mobile RS, and the handover could happen between the MR and BSs while the network moves across cell boundaries.

9.3.2 MAC Frame Structures for Relay Network By the IEEE 802.16e standard, the data transmission in WiMax system is base on a frame structure which is further divided into a DL sub-frame for the information delivery from the BS to MSs and an UL sub-frame for the information delivery from MSs to the BS. The DL sub-frame comprises of the bursts of control information broadcasted and several MAC data units in following DL bursts. The UL sub-frame comprises of contention intervals and UL bursts for transmission of data units. These two sub-frames are separated by gaps. The generic MAC frame structure is shown in Fig. 9.9 [1]. Based on the generic MAC frame structure, some modifications could be made to allow the data transmission in the relay systems. Two types of relays, which are de-centrally controlled relays and centrally controlled relays, are available, which can be selected to implement according to the system architecture. The modified MAC frame structure and connection management of these two relaying approaches are both discussed in [5]. In the de-centrally controlled relays, one RS has full control of MSs connected to it, and then it connects to the BS. The RS appears to be a BS from the view of MSs, and it seems to be a MS from the view of BS. To enable this relaying operation, some modifications on the MAC frame structure are needed. The new frame structure consists of a broadcast portion, 1st hop DL bursts, a contention interval, 1st hop UL bursts and multi-hop sub-frames. In the MAC frame, DL bursts and UL bursts of the first hop transmission between BS and RS are specified by the BS. Later, the DL bursts and UL bursts of second hop transmission between the RS and MS or the RS of the next hop are built by the RS at the current hop in the multi-hop sub-frame portion, when the frame is received by the RS. The overall frame structure needs to be compatible with the standard generic MAC frame. During the operation, the MSs can only observe the multi-hop sub-frame built by their RS. The MAC frame is shown in Fig. 9.10 [5].

Fig. 9.9 MAC frame structure

9


175

Fig. 9.10 Frame structure for De-centrally controlled relays

To maintain the overall connection in the de-centrally controlled relays, a RS appears to be a regular MS to the BS, and the BS only needs to set up and maintain the connection to the RS. At the same time, the RS appears to be a regular BS to MSs, and it needs to set up and maintain the connection to the MSs. In the centrally controlled relays, the BS has full control of RSs and MSs in the network. A RS woks simply like a router to forward frames from the BS to MSs. In the proposed frame structure, it consists of a broadcast portion, DL bursts for multiple hops, a contention interval and UL bursts for multiple hops. In the MAC frame, the BS constructs the DL bursts and UL bursts of first hop and the subsequent hops in a consecutive manner. Unlike the de-centrally controlled approach, there is no multi-hop sub-frame existed in the frame structure. The frame structure also needs to be compatible with the standard generic MAC frame. During the operation, the RS forwards frames between BS and MSs, and MSs can communicate with BS like through direct connection. The MAC frame is shown in Fig. 9.11 [5]. For the centrally controlled relays, the RSs just forward packets between the BS and the MS. The connection between the RS and the MS would correspond to the connection between the BS and the MS. To set up and maintain overall connection

Fig. 9.11 Frame structure for centrally controlled relays

176

Q. Lu et al.

from the BS to the MS, each time the MS establishes a connection to the RS, the RS needs to establish a corresponding connection to the BS. With the concepts of above two implementations of the relay structure and the MAC frame structures, the relay system could be functioned. And either one of these two types could be selected to function by the system engineers according to the system requirements. For the de-centrally controlled relays, the selection of the path for data transmission would be more flexible, but the BS has less control on the MSs. While in the centrally controlled relays, the BS has more control on the MSs, but the path for the data transmission should be predefined.

9.3.3 Fast Handover Scheme in Relay System In the WiMax relay systems, two types of handover may happen, and they are intracell handover and inter-cell handover. The intra-cell handover occurs when a MS moves across the boundaries of coverage of RSs connected to the same BS. While the inter-cell handover happens when the MS moves to another RS connected to a different BS. Since the BS manages all the RSs connected to it, and all the RSs can obtain the necessary information about the MS from the BS, so less information and procedures are required for the intra-cell handover. In contrast, more complex network re-entry process is needed for the inter-cell handover. Therefore, the intracell handover would be simpler and faster than inter-cell handover. To reduce the handover latency in the relay system, a fast handover scheme was proposed in [6] to abbreviate handover procedure shown in Fig. 9.12 [6]. It would reduce the occurrence of inter-cell handovers and preferably select intra-cell handover when it is available. To achieve the goal of reducing the inter-cell handovers, firstly, the MS should identify which BS the neighbor RS is belonging to. Two methods are proposed to use. The first one is called Hierarchical BS/RS ID, which changes the BSID format to include the IDs of RSs. The new structure of full BSID which can be acquired from the MOB NBR-ADV is shown in Fig. 9.13 [6]. The first 24 bits is used for the Operator ID as the standard BSID format, while the last 24 bits are used for both BS ID and RS ID. When the MS obtained this information, it can identify which BS the RS is connected to. Another method is to use an additional Type/Length/Value (TLV) encoding in the MOB NBR-ADV. It would include the BS ID of a RS, when the RS is found to be the neighbor of the MS. So during the broadcasting of MOB NBR ADV, it could notify the MS which BS the RS is connected to. The TLV encoding is shown in Fig. 9.14 [6]. By knowing the BS of the neighbor RS, the MS can preferably choose the RS within the same BS, which the serving RS is connected to, to perform handover. Therefore, when a handover is required, the MS would conduct intra-cell handover if the signal strength from the inter-cell RS is similar. Only if the signal strength

9


177

Fig. 9.12 Intra-cell handover procedure

Fig. 9.13 Hierarchical BS/RS ID format

from the inter-cell RS is much stronger than that from the intra-cell RS by certain threshold, the MS could conduct inter-cell handover. In the Fig. 9.15, it shows a scenario that an intra-cell handover is performed rather than an inter-cell handover when the MS moves to the boundary of the serving RS. While the relay network is available, with this handover algorithm, the occurrence of inter-cell handovers may be decreased and the intra-cell handover would

Fig. 9.14 TLV Encoding for BS ID

178

Q. Lu et al.

Fig. 9.15 Intra-cell handover is preferred

be preferably selected, when both inter-cell and intra-cell handovers are possible. Because the intra-cell RSs are connected to the same BS and all the RSs could have knowledge about the MS, the network re-entry procedures would be simplified. Therefore, the intra-cell handover would be relatively simpler and faster, and smaller handover latency would be introduced into the handover process.

9.4 Network Mobility and Fast Handover Scheme Some particular scenarios such as network mobility with a network moving can be supported with the relay structure in WiMax networks. To construct the moving network and realize network mobility, a MR could be used to connect multiple mobile stations and the BS. The MR in the system actually acts as a mobile RS, which is used to establish the mobile relay system. Therefore, each MS in the moving network can communicate with the BS via the MR. In the system, the handover may happen between the MR and BSs while the network moves across cell boundaries. The MSs would not perform handovers with the BSs directly but via the MR. The network structure for network mobility is shown in Fig. 9.16 [7]. To realize the network mobility, a network mobility handover algorithm is required. In [8], a primary handover scheme was proposed. Since it doesn’t consider link layer handover but only network layer handover, high handover latency may be produced by using this method. In conventional handover process, the link layer handover and the network layer handover should be conducted in series, which in shown in Fig. 9.17 [7], and it consumes quite long time to finish the overall handover process. To minimize the handover delay, a fast handover scheme was proposed in [7], which integrate link layer and network layer handover procedures to shorten this process.

9


179

Fig. 9.16 Network mobility system

In the suggested fast handover scheme in [7], two triggers: Link Going-Down (LGD) and Link Switch Complete (LSC) are introduced to achieve concurrent handover procedures at both network layer and link layer. During the neighbor discovery, the MOB NBR-ADV would carry both the channel information of neighbor BSs and the Access Router (AR) information. The MR would monitor the signal strength continuously, and may initiate a handover process depending on pre-defined threshold. When the threshold is satisfied, the MR starts the concurrent handover procedures at both layers using LGD trigger. After that, same procedure as specified in the standard will be performed at the link layer, but different process will be made to accelerate the process at the network layer. The MOB MSHO-REQ is sent to the serving BS and the Fast Binding Update (FBU) would be sent to the Home Agent (HA) at the same time. Then the HA would set up a tunnel and transmit the Handover Initiate message (HI) to the New Access Router (NAR). The NAR checks

Fig. 9.17 Conventional handover flow

180

Q. Lu et al.

Fig. 9.18 Fast network mobility handover procedure

the validity of New Care of Address (NCoA) with the Duplicate Address Detection (DAD) procedure and replies the Handover Acknowledge message (HACK) to the HA. Later, the HA would send the Fast Binding Acknowledge (FBACK) to the NAR. After receiving the MOB BSHO-RSP, the MR sends MOB HO-IND to start the execution of actual handover. By finishing the network re-entry, the LSC trigger is used to inform transmission of FBACK and finalize the concurrent handover process. Then the network can return to the normal operation. The detailed fast handover scheme is shown in Fig. 9.18 [7]. For the more complex architecture that a moving network is required, the relay concept could be used to construct such system with a mobile RS. By implementing this fast handover scheme, handover procedures at both link layer and network layer are allowed to be executed at the same time. Therefore, the overall handover process would be faster, and the handover latency could be reduced significantly. From this network mobility scenario, we can see that by using relay technique and proper fast handover schemes, various complicated scenarios could be established and fast handover for delay-sensitive applications also could be realized.

9.5 Summary and Conclusion In this chapter, an overview of the three fundamental handover modes and the standard hard handover procedure specified in the IEEE 802.16e has been presented first. Three fast handover schemes, which are used to expedite the handover process and to reduce the handover delay, are briefly reviewed. The first handover scheme is to reduce delay caused by redundant neighbor BSs scanning and association during the cell reselection procedure. Only one neighbor BS with highest estimated CINR would be selected for scanning or association. Another two schemes are both proposed to allow data transmission during the actual handover process. The third scheme is much better than the second one because it could support both downlink and uplink data transmission for real-time applications, while the second scheme

9


181

could only make data transmission possible for downlink services. Since the first scheme and the third scheme are proposed for different stages of handover process, they can be incorporated and implemented as one comprehensive scheme to deal with the entire handover process with much performance improvement. Secondly, the WiMax relay systems and the corresponding MAC frame structures are summarized. One handover scheme to reduce handover latency in relay network has been presented. Finally, the method to construct a moving network with the relay structure and one fast handover scheme to support network mobility are discussed. In the relay systems, both de-centrally controlled relay and centrally controlled relay structures could exist. The selection of the relay systems needs to satisfy the system requirement and system architecture. The handover scheme in the relay systems has been presented to reduce the occurrence of inter-cell handovers by preferring intracell handovers. Lastly, a fast handover scheme is suggested for network mobility to attempt concurrent handovers at both link layer and network layer. By employing the relay structure, various complex scenarios could be established. And proper use of fast handover schemes or future design and optimization of handover process could be carried out to achieve low latency handover and satisfy the system requirements.

References 1. IEEE Std 802.16e-2005, “IEEE Standard for Local and metropolitan area networks, Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems, Amendment 2 and Corrigendum 1”, February 2006. 2. D. H. Lee, K. Kyamakya and J. P. Umondi, “Fast Handover Algorithm for IEEE 802.16e Broadband Wireless Access System”, Proceedings of 1st International Symposium on Wireless Pervasive Computing, January 2006. 3. S. Choi, G.-H. Hwang, T. Kwon, A.-R. Lim and D. H. Cho, “Fast Handover Scheme for RealTime Downlink Services in IEEE 802.16e BWA System”, IEEE 61st Vehicular Technology Conference, May 2005, Vol 3, pp. 2028–2032. 4. W. Jiao, P. Jiang and Y. Ma, “Fast Handover Scheme for Real-Time Applications in Mobile WiMax”, Proceedings of IEEE International Conference on Communications, June 2007, pp. 6038–6042. 5. C. Hoymann, K. Klagges and M. Schinnenburg, “Multihop Communication in Relay Enhanced IEEE 802.16 Networks”, Proceedings of IEEE 17th International Symposium on Personal, Indoor and Mobile Radio Communications, September 2006, pp. 1–4. 6. J. H. Park, K.-Y. Han and D.-H. Cho, “Reducing Inter-Cell Handover Events based on Cell ID Information in Multi-hop Relay Systems”, Proceedings of IEEE 65th Vehicular Technology Conference, April 2007, pp. 743–747. 7. L. Zhong, F. Liu, X. Wang and Y. Ji, “Fast Handover Scheme for Supporting Network Mobility in IEEE 802.16e BWA System”, Proceedings of International Conference on Wireless Communications, Networking and Mobile Computing, September 2007, pp. 1757–1760. 8. V. Devarapalli, R. Wakikawa, A. Petrescu and P. Thubert, “Network Mobility (NEMO) Basic Support Protocol”, IETF RFC 3963, January 2005.

Chapter 10

Addressing Multiservice Classes and Hybrid Architecture in WiMax Networks Kamal Gakhar, Mounir Achir, Alain Leroy and Annie Gravey

Abstract This work presents two different propositions which mark new advances in WiMax. The first work addresses multiservice environment and service differentiation in WiMax networks. It argues that using only polling based priority scheduling at subscriber stations and demand based dynamic bandwidth allocation (DBA) at the base station it is possible to serve various traffic types in WiMax systems with only three service classes rather than four as proposed in the standard. It reduces the complexity of scheduling mechanisms to be implemented in WiMax interface cards thus bringing down overall capital expenditure (CAPEX) model for such system while providing QoS to applications. Both the transfer plane QoS, in terms of latency and jitter, and the command plane QoS, in terms of blocking probability are assessed. In particular, a simple, multiservice call admission control (CAC) mechanism is proposed that significantly improves on a previously proposed CAC mechanism by favouring real-time traffic over non-real-time traffic. The second work proposes an architecture for a hybrid system composed of WiMax (access network) and WiFi systems. A new “tightly coupled” approach considers matching parameters at MAC level which translates directly into the transfer of requirements from WiFi network to WiMax. A notion of jitter in WiFi systems is also introduced. Keywords WiMax · WiFi · Interoperability · Admission control · Scheduling techniques · Network simulation

10.1 Introduction This work presents the research work accomplished on the network architecture and the mechanisms to facilitate quality-of-service (QoS) for applications within project IROISE. It was conceived at Réseau National de Recherche en Télécommunications (RNRT), which is a national entity in France, including various industrial and academic partners. The project aims to demonstrate an uninterrupted wireless coverage permitting access to network (for example Internet) for a diverse population having K. Gakhar (B) 68 Rue Gallieni; 92100 Boulogne Billancourt, France e-mail: [email protected] M. Ma (ed.), Current Technology Developments of WiMax Systems, C Springer Science+Business Media B.V. 2009 DOI 10.1007/978-1-4020-9300-5 10,

183

184

K. Gakhar et al.

different application requirements. We discuss the transfer plane QoS, in terms of latency and jitter, and the command plane QoS, in terms of blocking probability are assessed. It also includes a simple, multiservice call admission control (CAC) mechanism that significantly improves on a previously proposed CAC mechanism by favouring real-time traffic over non-real-time traffic. The problematic was classified into various objectives which are discussed as follows.

10.1.1 IROISE Objectives The goal of project IROISE is to demonstrate a seamless wireless coverage in a diverse geographic zone for applications, being used by various kinds of users, having some minimum acceptable quality. Our research goal within the project is to propose an architecture and associated services which could support QoS for applications being served in a wireless hybrid network based on current standards. This work has aimed to accomplish the above-mentioned goal by:

r r

r

carrying out a study of the state-of-art of similar works and the existing techniques to support QoS in a wireless network specifically in the domain of computer networks in order to understand existing interworking techniques. demonstrating feasibility of uninterrupted network access aided by network architecture operating in wireless mode inspired from current broadband wireless access technologies, for example IEEE 802.16 [1], in addition to local wireless access facility offered by IEEE 802.11 [2] (also popularly known as WiFi). It was done via a small real-life demonstration of proposed techniques. proposing novel techniques to support QoS (especially from the point of view of different traffic types) for applications using existing and/or proffering new techniques. This could include studying classic problems like dimensioning of such a network to allocate bandwidth for traffic needing predefined QoS, buffer management and scheduling techniques to support QoS for applications.

The first work addresses multiservice environment and service differentiation in WiMax (Worldwide Interoperability for Microwave Access) networks. IEEE 802.16 defines 2–11 GHz version using OFDM, also called WirelessMAN-SC, which is adopted by WiMax FORUM and thus 802.16 is also referred to as WiMax in literature. Since few years now networks have evolved into multiservice environments. In any network where bandwidth is limited it is inevitable to treat various traffic types differently. This concern goes back to early 1990s when IntServ [3] was proposed. It was further readdressed by the proposition of DiffServ [4]. However, to minimize technical complexity and to propose economically viable solutions, the question of how to provide this differentiation has also been around from the same era. To quote a question from [5]: “Is service priority useful in networks?”. Indeed the resources needed to provide multiservice differentiation inflict hard to realize challenges like matching software performance to theoretical values. Thus for the sake of simplicity of the system and eventually to have a practical solution, it is desirable

10

Multiservice Classes and Hybrid Architecture

185

to minimize the number of differentiations that are conceived for a system. The multiservice environment includes the type of traffic which demands “privileged” treatment compared to the traffic which needs only best effort (BE) scheduling. However, a “preferential” treatment doesn’t come by default. The resources need to be requested in advance and if enough of them are not available in the system then specific policies need to be implemented which eventually help to achieve the preference over less privileged traffic. For a service provider it is important to ensure good QoS to applications in broadband wireless access (BWA) for business reasons. In BWA networks the challenge is even more severe because of classical limitations of radio environment which imposes restrictions on the number of users and the quality of network service that can be attributed to them. The question is, “How could we ensure that technically and economically the broadband wireless networks can be deployed and provide the required QoS to the clients?”. An effective use of available bandwidth spectrum via dynamic resource reservation handles one aspect the issue [6]. Another facet is how better can one handle the given services in a system so as to incur minimum costs of handling what has already been provisioned to the system. It is clear that more the number of services to be handled in the system the higher will be the signalling costs involved and complex it would become. It is known that 802.16 standard proposes two different types of uplink traffic scheduling to be implemented (periodic and polling based). The present work investigates whether a 802.16 network that only supports three polling based classes is still capable of providing the QoS levels expected for all types of applications. Our interest here lies in identifying policies ensuring that different traffic classes receive the required QoS. In our work we assume that resources are scarce, and that a real multiservice support is mandatory, which implies using the polling based classes. The second work proposes an architecture based on “tight coupling” for a hybrid system composed of WiMax (access network) and WiFi systems. Before moving onto details of our work it will be interesting to see some approaches which have been proposed earlier for the problems similar in nature. Section 10.2 discusses some works that are related to services differentiation in networks and multiservice differentiation in WiMax with a detailed study and some novel propositions. Section 10.3 discusses CAC in WiMax networks. Section 10.4 discusses some related works on the interworking between systems composing an hybrid network. In Section 10.5 we propose a novel architecture for WiMax-WiFi interworking. Section 10.6 concludes the work by summarizing the contributions and highlights future work to be accomplished.

10.1.2 Application Taxonomy Let’s first consider Table 10.1 which gives us taxonomy of applications in a possible multiservice environment to better visualize the problematic. We can classify such an environment as per nature and kind of traffic involved. The elastic and interactive traffic like on-line gaming can afford slight delays on underlying networks. Traffic

186

K. Gakhar et al. Table 10.1 Taxonomy of applications in a multiservice network

Nature and traffic

Elastic

Non-elastic

Interactive

Traffic: Mostly VBR, QoS: slightly time sensitive, Examples: e-commerce, on-line gaming

Traffic: CBR, QoS: highly time and loss sensitive, Examples: VoIP, Video-conference

Non-interactive

Traffic: Mostly VBR, QoS: not time sensitive, Examples: Emails, Web browsing, downloads

Traffic: CBR or VBR, QoS: loss sensitive, Examples: IPTV

which is non-elastic and interactive for applications such as video conferencing are highly sensitive to delays and jitter they could experience in the networks. Applications which are elastic and non-interactive in nature such as large file downloads, as normally seen in networks, are not sensitive to delays and their quality is not affected. For multimedia rich content applications such as IPTV, however, it is very important to have almost no losses as such an application though non-elastic and non-interactive in nature are very important for user experience. Lets see some related works to understand adjoining ideas.

10.2 Related Works This section discusses some of the recent works which presented some ideas concerning dynamic bandwidth allocation (DBA) and service differentiation in networks. With the development of high speed and high capacity optical networks, considered mainly for backbone traffic, the question of providing service differentiation for a hybrid scenario of wireless network with these networks has been around for some years [7, 8]. A recent work by Yuksel et al. [9] emphasized how service differentiation has become more and more important in recent years due to increasing diversity of applications. The work considers two traffic classes on a single CoS link: premium class (needs delay performance) and best-effort traffic. It proposes to quantify the value of having differentiated service class-of-service support in IP backbone by comparing the capacity of applications that require delay or loss assurances in comparison to a network that provides best effort service (and still has to meet the same performance assurances by provisioning extra capacity needed to this effect). In wireless environment the bandwidth resources are very much limited own to technology and regulations. The work by Zang et al. [7] proposes a service differentiation enabling network architecture for a hybrid system comprising 4G and wired networks. It considers factors such as error-prone behaviour, spectrum limitations, user mobility, and packet scheduling in wireless scenario. It shows via different packet scheduling policies that differentiation among traffic types is effectively attained. Service differentiation is more important than ever as we see many commercial deployments of wired-cum-wireless (and vice versa) products [10]. This work uses a generalized approach of a congestion control algorithm to differentiate between “re-

10


187

liable” and “unreliable” flows. The contending flows in the same bandwidth pipe are allocated the rates in proportion to weight associated with its service class. However, as mentioned in the work itself the work targets flows of 100kbps or less. Moreover modelling the system for only low rate flows won’t really go a long way to propose anything concrete for real-time traffic where the average rate is normally higher than the one targeted in their work. The QoS architecture considered by Cho et al. [11] proposed probability based analytical models for uplink bandwidth allocation scheduling and channel utilization in 802.16 systems. Even though theoretically it considered different priorities among various traffic classes the simulations assume only Poisson type traffic which once again does not represent all of the real network traffic characteristics. The work by Gusak et al. [12] on the performance of 802.16 MAC studies an algorithm for adaptive frame partitioning for data and voice flows though the packet size range considered for voice doesn’t seem realistic as found in commercial products. Moreover, the assumptions made for the charge on a subscriber station are over simplified (just one flow per SS). The work by Ogawa et al. [13] presents an approach that resembles somewhat our own policy of bandwidth allocation. It addresses service level QoS to different traffic types in CDMA. It proposes a MAC level protocol using QoS control scheme which is composed of two mechanisms: Dynamic Queuing Control and DBA Control. It proposes to treat traffic as per classes, however, these classes refer to the type of clients being served, i.e., Class C1 – industrial client and Class C2 – household clients. The classes are served as per allowable delay time (ADT) which is a client based choice. Though they do not make an apparent priority among classes the methodology proposes a delay function which is strict for C1 than C2. It introduces a notion of treating one client superior to another. DBA scheme is based on locating optimum radio resources to a mobile terminal dynamically. Even as wireless systems were coming into being few years ago there had already been suggestions on differentiating services in networks, for examples, in IEEE 802.11 which only supported best-effort service [14]. The work proposes two distributed estimation algorithms for distributed coordination function (DCF) environment. It supports service differentiation, radio monitoring, and admission control. It implements a virtual MAC (VMAC) algorithm that monitors the radio channel and estimates locally achievable service levels. It uses an algorithm at source level to estimate service quality that different flows get by VMAC. It goes on proposing that by applying distributed algorithms to admission control of a radio channel it can attain and maintain a globally stable state without the need of complex centralized radio resource management. Over the last few years, the need of providing service differentiation via scalable architecture has also been highlighted by Christin and Liebeherr in [15], citing the limitations of earlier works in IntServ and DiffServ. The work argues that classbased architectures for service differentiation have lower overheads than the ones based on flows. It also emphasizes an urgent need of service differentiation in access networks which mostly suffer from bottle-neck problems hindering QoS for applications.

188

K. Gakhar et al.

A recent work by Cicconetti et al. [16] assesses performance of 802.16 point to multipoint system. It considers that as the standard advocates fixed allocation for UGS service it is not interesting to study its performance at MAC level (which on the other hand is vital in our analysis further in the chapter). For scheduling policies, it uses Deficit Round Robin (DRR) on downlink from BS to SS and Weighted Round Robin (WRR) on uplink from SS to BS. An article by Mukul et al. [17] concerning WiMax suggested to use adaptivebandwidth scheduling algorithm at SS for rtPS traffic in which SS predicts the arrival of packets in advance and makes the bandwidth request. The method is based on differential time grant where it makes stochastic prediction for the additional time needed by the rtPS traffic to serve the total packets arrived in the queue between the moment it first makes the request and the grants received to be used. However, there is no comparison of performance with the “average” values, that is, without additional time added. Also, time durations considered for the simulations are so small that “real” scenarios cannot be analysed. After going through the above-cited approaches, over the next sections we discuss our approach towards service differentiation where we propose some simple methodology showing how our suggestions can lead to less complex yet efficient solutions to provide good QoS to real-time and non-real-time traffic in an IEEE 802.16 system.

10.3 System Methodology In this section we first describe the objectives that we are going to address over the next paragraphs:

r r r

The need for a simple bandwidth allocation policy at BS in a 802.16 system which is able to achieve service differentiation for various traffic types and justifies an appropriate mapping of various traffic types to service classes. A simple traffic management policy within SS keeping in mind that delay sensitive applications get served in time. Addressing performance for various traffic types seen in Table 10.1 in the context of an 802.16 system.

To address these objectives lets start by understanding the framework behind the work.

10.3.1 Framework Let’s revise the framework of our approach which makes the basis of analysis of our system. We have two “guiding actors” in the form of Tables 10.1 and 10.4. Table 10.1 presents taxonomy of applications in the networks. Table 10.4 highlights the mandatory parameters at flow level which help addressing QoS associated with each service class in the standard. When we analyse traffic engineering policies in

10


189

802.16 systems we first have to understand how to treat well various applications. Also as we analyse our policies, the focus needs to be on classifying the service types which could be used to serve the applications (this is inline with the guidelines of the standard). Though the above parameters are supposed to be associated with different service classes, we will identify how the classification of applications may be helpful to design policy decisions to serve various service classes. It means we will treat various applications traffic with our policies and depending on the results obtained on QoS we will propose an eventual mapping of applications traffic to the respective traffic classes in which they could be served. The analytical approach to model such a multiservice system was abandoned in favour of simulation studies of the system. It was considered just to use existing simulation utilities which one can use to model various traffic types more easily and accurately to analyse elementary system behaviour. Before going into detailed system study we describe application modelling, QoS metrics, and simulation model for our system and the changes we had to make to the existing network simulator (ns-2) to accomplish our algorithms [18, 19] (the reader is advised to see appendix for details).

10.3.2 Application Modelling Having a proper traffic profile is perhaps the most important factor to be considered before any traffic engineering policy can be deployed. It is in the best interest of the service provider to have this knowledge beforehand so that it knows what applications should it market. We can refer to Table 10.1 to describe application modelling. We can see that in general we have to model two types of applications: Elastic (on TCP) and NonElastic (on UDP). Different applications are modelled using different models of traffic, notably, Poisson, Exp-On/Off, and resource hungry applications. Poisson source is interesting for it models superposition of multiple CBR flows in the system which, in fact, could imitate a SS acting as an WiFi AP. The On/Off sources allow to check the robustness as the application rate varies (inducing busrtiness inthe system), and the resource hungry FTP corresponds to the cases when SSs are saturated (as in for BE traffic). Table 10.2 summarizes various applications types and the associated models we use further for our simulations. Let’s now understand what parameters are mainly considered to analyse QoS of applications.

10.3.3 QoS Metrics As it is clear from literature, QoS can mean different things for different users. Nevertheless, here we define what we are looking for while addressing QoS for various applications traffic in 802.16 system and why the parameters (policies) discussed

190

K. Gakhar et al. Table 10.2 Taxonomy of applications traffic and simulation parameters

Application type

Example

Simulation model

Parameters

Non-elastic interactive

VoIP

CBR on UDP, Poisson on UDP, On–Off Exponential on UDP

packet-240 bytes rate-80 kbps, packet-240 bytes rate-390 kbps

Non-elastic Non- IPTV streaming interactive

On–Off exponential on UDP

packet-240 bytes rate-390 kbps

Elastic interactive

On–Off exponential on TCP packet-512 bytes

Business applications

Elastic Web browsing, email non-interactive

On–Off exponential on TCP, FTP on TCP

packet-512 bytes

further are important from QoS perspective. A user in any system will use its services (and be willing to pay) only if the applications run to good satisfaction. In new BWN systems such as 802.16 it is, therefore, very important to understand what parameters provide the best indications of QoS for such a multiservice system. From Table 10.1 it can be understood that for applications having non-elastic traffic the QoS parameters are the loss, delay, and jitter they experience in the network. However, for the applications traffic elastic in nature, essentially throughput is a better indicator of application QoS. The delays of different applications traffic are calculated as the difference between the time a packet enqueues at the source and the instant it is received at the destination. Throughput of applications is calculated from the bytes received at the destination. For non-elastic traffic, it is usually necessary to implement a so-called “playout buffer” in order to deliver correctly the successive packets to a codec. The network induced delay is then be approximated by what is called “effective delay”, i.e., the sum of the latency and the jitter affecting the traffic. We modelled the latency as the mean delay, and the jitter as the difference between an upper and a lower quantile of the delay. The jitter in the system for applications is calculated as: J itter = (max − min)delay quantile. While real-time VBR traffic can safely suffer higher delays, the main QoS metrics for this class also remain the same as for real-time CBR traffic. Throughput is derived via the bytes received (and thus converted to the bandwidth used by individual flows) as they arrive at destination. After looking into Table 10.1 we can see that traffic which is Elastic, Interactive can experience some delays. The applications of such traffic type can afford to suffer higher delays compared to the two traffic types discussed above. However, throughput for such traffic types is important. The last category of traffic which is Non-interactive, Elastic in nature fits best to BE traffic. There are no strict QoS metrics for this class of traffic. Nevertheless, it is in the best interest of user that such applications are being entertained in a multiservice system as and whenever possible. From the above discussion we can summarize that delays and jitter (hence almost no losses) are the most important QoS metrics for real-time traffic types. The

10


191

eventual QoS policies have to ensure via simple or complex methods that given some practical bounds on traffic a system gets QoS it demands. We will later propose another classification of traffic types which can serve various applications of Table 10.1. Now we discuss the way we conceive and model our simulation scenarios. Later, we will discuss the scenarios accompanying with the results obtained and their implications.

10.3.4 Simulation Modelling Our system consists of a BS and several SSs. We assume that there are demands for various traffic types at SS. However, here we ignore wireless conditions for simplifying system behaviour. The physical layer conditions themselves make an interesting topic but we rely on simple network management policy approach here in order to understand basic system behaviour. Figure 10.1 represents the system we consider for our simulation studies in the section further. The sources, represented by S, generate a mix of traffic demands to be sent on the uplink bandwidth pipe. In fact, such a model for uplink traffic demand highlights the scenario that may arise in a zone where different types of users are trying to access the wireless network with their applications needing diverse requirements. Also, we don’t consider any delays associated with BS due to its own backbone traffic. The model was first tested with CBQ implementations of ns2 to have some benchmark based on which we made further simulations (we don’t, however, indulge into the details of the results obtained with CBQ). As presented in Fig. 10.2, real-time CBR applications generate traffic at regular intervals of time. The standard proposes that such traffic is served by the allocations negotiated at connection setup and are given regularly. As such, the demands for

CBR on UDP Exp-On/Off on UDP SS 1

Exp-On/Off on TCP

S

FTP on TCP CBR on UDP Exp-On/Off on UDP

BS

SS 2

n SS

Exp-On/Off on TCP

S

FTP on TCP

CBR on UDP Exp-On/Off on UDP Exp-On/Off on TCP FTP on TCP

Fig. 10.1 Simulation model representing the system under performance evaluation

S

192

K. Gakhar et al.

CoS1

CoS2

SS CoS3

CoS4

Fig. 10.2 Traffic queues of various classes within bandwidth pipe

such traffic can easily be predicted. The real-time VBR applications sources, however, generate packets randomly as can be modelled by some available analytical principles. It may happen, therefore, that their demands vary from time and again. For such a service class the standard says to adopt a polling mechanism. It will be an effective means to respond to bandwidth demands which would vary in the network over the course of time. Within traffic queues, we treat the packets using a FIFO approach. The non-real-time applications generate large amount of data and are modelled normally using large packets with random On/Off times. Last type of traffic is considered to be that of FTP application. This kind of traffic is carried mainly by transport protocols which influence the congestion control in the underlying network using their own policies. The standard nevertheless proposes to use polling methodology to serve these traffic classes. The next section presents the scenarios we looked at and the effects of the results, and how our methodology helps to define service differentiations. This is followed by the conclusion on what kind of traffic types could be associated with what kind of service class of 802.16 system.

10.4 Analysis of Traffic Engineering Approach The superframe in 802.16 is adaptive in nature and allows uplink and downlink demands to be fulfilled. The subscriber stations (SSs) in the system operate in TDMA mode thus accessing the channel only as per the grant allocated to them. On the downlink, the base station (BS) operates in broadcast mode and a SS responds if the data is aimed for it. Our interest is to demonstrate the handling of applications traffic in uplink mode where the system could inflict delays on different traffic classes if appropriate policies are not implemented. Based on the performance of various traffic types and guidelines in the stanadrd we would like to suggest a justified classifcations of various service classes and applications traffic. It includes mainly an allocation policy and Intra-SS scheduling policy responsible to handle various traffic types. This section presents the detailed analysis of

10


193

our system as we try and answer to adopt a simple yet effective traffic engineering approach. We would like to study a 802.16 system via simulations to understand how various traffic types behave when guided by a comprehensive policy of bandwidth allocation. It is reasonable to argue right from the beginning that we inevitably need bandwidth allocation policies both at the BS and the SS in such a system where two-hop scenarios are the natural implications. WiMax/802.16 systems are new in broadband wireless networks domain and given the scenario of limited bandwidth in BWNs, BS must make sure that demands from various SS are satisfied in a fair manner therefore some decision making has to be ensured too for allocations at BS. We, therefore, would like to check various basic questions associated with the QoS that various applications traffic experience on the uplink following our policy. We would like, further, to make a mapping between various traffic types and the service classes in 802.16 which could be used to formalize the system behaviour. We once again focus on IROISE architecture discussed in Section 10.7. In a 802.16 system we need to address uplink allocation policies at two levels. The bandwidth allocation algorithm (at BS) should be such that it does not starve a subscriber station (which might need uplink allocations). Also, within a given SS the scheduling policy should be such that the traffic needing real-time constraints is provided a better dealing. For now we consider only one BS and two SSs in the system. We can easily generalize the results obtained here for a system of more than two SSs as the uplink channel can be considered loaded, without a loss of generality, in similar ways for demands of more than two SSs. It is assumed that admission control and/or traffic control is performed by the operator in order to avoid congestion. However, it is important to note that these policies have to rely on results such as given in the remainder of this section in order to determine acceptance levels.

10.4.1 Validation for Dynamic Allocation We gradually discuss some novel propositions while answering, with the help of simulations, some of the basic questions linked to the objectives for our system (our propositions came out of the observations made here). The first concern for a new system is, “As 802.16 uses TDMA for various SSs to able to use uplink allocations, how should BS policy make allocations to SSs in the system? Should it make fixed allocations on the basis of number of SSs or take into considerations the demands from each SS?” To address this we perform the following simulation: We want to see given some real-time VBR traffic in the system how the delays are affected based on the allocations a SS gets. The configuration of the system is as; we use real-time VBR applications sources on different SSs. One SS sends double the amount of traffic than the other. In our system the available bandwidth is 40 Mbps. The superframe duration is considered to be 10 ms for our studies and should remain the same unless specified. It is the most common value advocated for NLOS of 802.16 standard in literature. We will also take an uplink value of about 50% of the superframe value as

194

K. Gakhar et al.

the content rich applications are supposed to make higher demands for bandwidth. We consider in this case thus total load is about 12 Mbps (4 Mbps from one SS and 8 Mbps from another). The basic parameters to create such traffic are the same as defined in Table 10.2 and should remain same throughout unless indicated. Thus, if a bandwidth allocation (BA) follows a fixed allocation policy it divides the uplink in proportion to the number of SSs considered (here 2). The SSs do not implement any particular policy on applications traffic; packets arriving at SS are forwarded to BS in FIFO mode. Observations: We see the result in Fig. 10.3. It shows a histogram of delay observed for the SS which sends more traffic (the SS which make sends smaller amount gets served with very small delay). We see that it suffers unacceptable delays because of the fixed allocations in the system. The packets it sends to BS don’t receive enough time allocations in which it could send data on the uplink. This simulation shows that, even if 10 Mbit/s is statically allocated by the system to a source sending 8 Mbit/s of on–off traffic, the delivered QoS is bad. It means that we need an allocation policy which takes care of such a problem by making allocations which takes traffic demands into consideration. Hence a dynamic (adjustable) policy of bandwidth allocation is justified. We want the uplink traffic to utilize full system bandwidth available therefore under such traffic conditions a dynamic allocation is desirable in order to ensure QoS for real-time applications. The grants given by DBA policy should depend on some policy of demand from various clients. We simulate another scenario to prove our point.

Histogram - Delays of a SS in Fixed DBA 6e-05 real-time VBR traffic

Frequency Distribution

5e-05

4e-05

3e-05

2e-05

1e-05

0 0

0.2

0.4

0.6

1.2 1 0.8 Delay (seconds)

1.4

1.6

1.8

2

Fig. 10.3 Fixed grant allocations affects delays for real-time VBR traffic classes – It emphasizes that a fixed allocation policy is not the best answer to serve applications in a 802.16 system

10


195

Histogram -- Delays for real-time VBR traffic using Polling 0.003

real-time VBR - SS1 real-time VBR- SS2

Frequency Distribution (Mbps)

0.0025

0.002

0.0015

0.001

0.0005

0

0

0.1

0.2

0.3

0.4

0.5

Delay (sec)

Fig. 10.4 Proportional allocations to SS results in better treatment of applications

In simulation, one SS sends the demand of about 6 Mbps while the other demands for 8 Mbps (higher demand in the system than what was considered above). BS makes a polling at the end of each superframe cycle in order to collect current demands of each SS. It then analyses how much demand is received from each SS and makes respective allocation in proportion to the sum of total demands received n(the proportion is decided from the available uplink frame as: Demand SSi ). Figure 10.4 shows the result of the simulation. It Demand SSi / i=1 shows the delays of each SS which sends demands for real-time VBR traffic. We can see that with proportional allocations the delays for SSs are less than what they get via fixed allocation policy. Thus a better allocation will be the one where available bandwidth is used in just proportion to the demand of each SS. Therefore, we can conclude that a dynamic policy of bandwidth allocation would favour better treatment of applications (especially the one with bursty nature).

10.4.2 Impact of BE Traffic in the System Next we consider that all traffic sources are sending traffic on the uplink and the demands of each SS is considered dynamically by the BS. Given this, we would like to understand the consequences of treating equally the BE traffic type with the rest of the applications types – the applications which need some minimum bandwidth guarantee.

196

K. Gakhar et al.

The simulation scenario is as: Two SSs send different applications on the uplink. One of them sends applications which need some minimum bandwidth (they make in total demand of about 13 Mbps) and the other sends the applications which need BE service (FTP over TCP). A SS i makes a demand for time to send (QueueSi ze(i) ∗ 8)bits. The BS then decides grants (in time units) for each SS i as following: Demand SS(i)/Link BW ; then the effective grant of a SS i is calculated from the total uplink time available as: Grant SSi

i

Grant SSi ∗ U plinkT ime.

The above calculations of grants mean that BS has a policy of allocation which is sensitive to the demands. However, once the grants are made to SS there is no further differentiation of these allocations within the SS. Observations: The simulation results seen in Fig. 10.5. It shows histogram of delay suffered by real-time applications. We see that most of the real-time applications packets are served with the delay up to 4 s. The mean delays were observed to be more than 2.5 s. It is because of BE traffic which is in fact greedy in its nature and fills uplink demands by aggressively increasing its rate over the period of time. BS therefore accords more allocations to SS via which BE traffic types sends

Delay for CB traffic induced due to BE traffic 0.00035 CB traffic - real-time CBR CB traffic - real-time VBR 0.0003


0.00025

0.0002

0.00015

0.0001

5e-05

0

0

0.5

1

1.5

2

2.5

3

3.5

4

Delay (seconds)

Fig. 10.5 Greedy BE applications traffic gets more grants as it increases its demand over time which affects the delays with which most of the real-time applications are served – It highlights the need of treating CB traffic and BE traffic types differently in 802.16 systems

10


197

its demands. Such a result tells us that there is a clear need to treat BE and CB bandwidth traffic separateley.

10.4.3 DBA – QoS Sensitivity We note that untill now we do not have a notion of traffic differentiation in simulations. All traffic is being treated as if a big service class is composed of various applications found in the system. The idea is to show how the service differentiation in the DBA impacts on the QoS of applications. This will help us understanding the basis of differentiation before we develop full DBA for our system. The simulation scenario is as: We put non-elastic interactive traffic (video conference for example) on–off exp traffics on one SS. On the other SS, we put non-elastic non-interactive (video streaming). Both of them are Exp-On/Off carried on UDP. The first situation is to explain what happens when the DBA does not differentiate between the classes that could carry these traffic types. Each SS makes a demand of 8 Mbps each. The uplink allocations are made as in the above scenario, i.e., in proportion to the demands. We take the above scenario so that the differentiation could be noticed for delay sensitive applications. Observations: We see that in Fig. 10.6 when no differentiation is made among classes. Non-elastic interactive, non-elastic non-interactive traffic get their demands satisfied without any differentiation. The average delay and jitter for non-elastic interactive traffic were observed to be 12.6 and 7.7 ms, where as, the average delay and jitter for non-elastic non-interactive traffic was observed to be 13.0 and 7.6 ms, Delay Distribution for Non elastic interactive and Non elastic non interactive 1.4 non elastic interactive- SS1 non elastic non interactive- SS1 1.2


1

0.8

0.6

0.4

0.2

0 0

0.005

0.01

0.015

0.02 Time (sec)

0.025

0.03

0.035

0.04

Fig. 10.6 DBA does not differentiate among service classes at SSs, even though the demands are satisfied here to present a clear scenario. However, traffic would suffer delays in the absence of differentiation when treated in another class

198

K. Gakhar et al. Delay Distribution for Non elastic interactive and Non elastic non interactive 1.4 non elastic interactive- SS1 non elastic non interactive- SS1 1.2


1

0.8

0.6

0.4

0.2

0 0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

Time (sec)

Fig. 10.7 DBA makes a differentiation among service classes at SSs, improving service to nonelastic interactive traffic

respectively. However, as more application traffic will be mixed up in the system these different applications will suffer in the absence of differentiation at SS level. In another scenario, DBA favours non-elastic interactive traffic (compartively more delay sensitive) over non-elastic non-interactive traffic. We see in Fig. 10.7 the delays experienced by these two classes. The average delay and jitter for nonelastic interactive class was observed to be 12.4 and 7.2 ms, respectively. However, those for non-elastic non-interactive class were 14.36 and 7.8 ms, respectively. This differentiation would be more significant when bursty conditions are introduced in the system. Comparing the above two scenarios we observe that effective delay for nonelastic interactive traffic improves whereas that of non-elastic non-interactive traffic suffers more. Thus we understand that DBA differentiation is going to play an important role in servicing various applications in the system and has an important role. We develop a full DBA algorithm to this effect in a later section.

10.4.4 Consideration for Standard Service Classes Before moving to further simulations, we would like to adopt an approach which will make our analysis more comprehensive. When we will be finished our results we can justify the choice we present here. The standard proposes that UGS service class consists of real-time CBR applications. Similarly, it proposes to use rtPS service class for real-time VBR applications; nrtPS service for serving non-real-time

10


199

large data files and BE service for applications which are not delay sensitive. We, thus, present further studies using these notions as suggested in the standard. We will later justify this supposition to serve different applications. From now on, unless specified, the system load is considered to be about 45% of the available bandwidth (it means when the sources are active the sum of their mean rates don’t go beyond 20 Mbps though we should consider BE service – FTP over TCP – excluded from this – the sum of the rest of the sources add up to 18 Mpbs) and the simulation parameters ensure that various applications types contribute different percentage of traffic on the uplink load. The simulations are run for about 120 s in which the sources start gradually in the beginning and stop gradually towards the end. We will nevertheless highlight whenever a different configuration is used for simulation purposes. Now the whole scenario is set for us to further our studies. We have a DBA which implements a policy of allocations to SSs in proportions of their demand of the total load send on the uplink. We choose to have a simple policy of setting priorities at SSs in the way various traffic types are dealt with in the system. We would like to understand the kind of QoS each of the applications receive once we put the DBA methodology we conceive and what implications such an approach might have to define a 802.16 system. As we made more and more observations regarding delays for different traffic during our simulations we had to put the basic question to ourselves and which was: “How many service classes should be sufficient to support various traffic types in an access network comprising IEEE 802.16 (or WiMax)? [20]”. The objective of service classes is to gather one or more traffic types present in the network so that they can be treated in the same manner. In this way the policies related to traffic handling can be influenced to improve over all network service. Instead of dealing with individual traffic types it was noticed that a collection of similar performance characteristics traffic can provide a positive influence in order to simplify traffic handling policies in the networks.

10.4.5 DBA – Polling Based Policy and QoS for Real-Time Traffic The next question we put to ourselves was whether we need two different service classes within real-time traffic types, i.e., one for CBR traffic and one for VBR traffic type? Along with this we also need to address the way in which the polling is done for SSs.

10.4.5.1 DBA Algorithm Our goal is to keep the system simple while proposing some efficient solution to a WiMax network. In our approach, the DBA algorithm and hence inter-SS allocations are based on demand proportions of each SS. The basic algorithm is shown as in Fig. 10.8. Our methodology is equally applicable to a hybrid mapping module (that

200 1. 2.

3. 4.

K. Gakhar et al. UGS grants are prenegotiated and are fixed within a superframe before the transmission starts. Then BS allocates grants to rtPS traffic in the following manner: if the sum of all demands for real-time VBR traffic is smaller than available allocation then all demands are satisfied otherwise the BS serves the demands as per their proportion in the total demand. If there are remaining grants, BS then handles the nrtPS traffic demands (needing some minimum bandwidth guarantee) in a similar manner. BE traffic gets grants only if there are remaining grants after serving the non-real-time traffic demands.

Fig. 10.8 DBA algorithm providing service differentiation for various traffic types in 802.16 system. It will be helpful to further differentiate traffic and map them onto corresponding service classes

we present in further sections) within a QoS architecture and provides necessary differentiation for traffic types. In this algorithm the demand from each traffic source makes up a part of the global demand of its corresponding service class at SS in the system. The SSs then sends demands to BS. The DBA policy first makes “fixed” allocations to UGS class which is determined as a function of uplink frame and data that can be sent in this time. The remaining uplink time for allcations is: T otalU plink = T otalU plink − i U G Sgranti . If T otalU plink > 0, then rtPS class is assigned allocations (in time units) as follows: rtPSgranti = rtPSdemandi /LinkBW The rest of the service classes (nrtPS and BE) are also served in the same manner depending on the availablity of uplink time (and if they were given allocations), i.e., if TotalUplink > 0. Once these allocations are decided by BS, each SS sends data on the uplink starting with UGS – SS1 is followed by SS2 on the uplink time frame. This is followed by time frames of rtPS class – again SS1 is followed by SS2. The polling is also accomplished during the superframe time (however, polling methodology can be vary and might have different effects on QoS of applications). The same process continues for other classes if only they have uplink allocations. Our mechanism ensures that – real-time CBR traffic always gets its grants, realtime VBR traffic is served before non-real-time traffic, and BE traffic is served only if the CB traffic does not need the grants.1

1 We need to reserve a bit more time for UGS than what is calculated in theory in order to accommodate inherent queueing delays in the system, for example, we utilize 1.2 ∗ U G StimeT heor y . This “comfortable” value was obtained by observing delays for UGS traffic during simulations by varying the values of fixed allocations given to traffic. We consider a distance of about 10 km thus the propagation delay is about 40 ␮s.

10


201

10.4.5.2 Results on UGS and rtPS Classes To answer the aforementioned concerns, first, we analyse whether it is possible to treat real-time CBR and VBR applications traffic together. We first show that treating together two real-time service classes is possible which means that eventually the scheduling within SSs could be much simpler. And, then we present different results of polling in SSs showing the effects it produces on delays. Here we consider that SSs are polled each at the end of their respective allocations, i.e., a SS is polled for its next allocation as soon as it stops sending packets on the uplink. We consider certain percentage of each traffic type in the network, as discussed in earlier paragraphs, which is more realistic. Figure 10.9 shows the histogram of the delays for UGS and rtPS traffic types in the system. We observe in Fig. 10.9 that there is no minimum delay for considered UGS packets. As the allocations are given with priority to UGS traffic its packets get served in the superframe with null minimum delay. We see that the majority of UGS traffic gets served in 1 superframe (10 ms) with the maximum delays goes upto about 2 superframes (20 ms). The rtPS packets wait at least half of a superframe size (this is due to polling), and the majority of them have to wait between 1 and 2 superframes. The jitter for rtPS traffic is however smaller than the one for UGS traffic although the rtPS traffic is more bursty than the UGS traffic because the polling mechanism ensures that resources are allocated when requested, and not Histogram - Delay 1.4 UGS-SS1 UGS-SS2 rtps-SS1 rtps-SS2

1.2 Mean-UGS1 = 6.12 ms, Jitter = 11.7 ms


1

Mean-UGS2 = 6.26 ms, Jitter = 12.1 ms Mean-rtPS1 = 13.39 ms, Jitter = 8.4 ms

0.8

Mean-rtPS1 = 13.25 ms, Jitter = 8.3 ms

0.6

0.4

0.2

0 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

Delay (sec)

Fig. 10.9 Histogram – Delay – UGS (real-time CBR receives fixed grants) and rtPS traffic (realtime VBR receives polling based grants) showing that the effective delays for real-time traffic are considerably similar

202

K. Gakhar et al. Histogram - Delay 1.4 UGS-SS1 UGS-SS2 rtps-SS1 rtps-SS2

1.2

Mean-UGS1 = 12.34 ms, Jitter = 9.6 ms Mean-UGS2 = 12.24 ms, Jitter = 9.6 ms


1

Mean-rtPS1 = 13.22 ms, Jitter = 8.2 ms 0.8

Mean-rtPS2 = 13.1 ms, Jitter = 8.1 ms

0.6

0.4

0.2

0 0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

Delay (sec)

Fig. 10.10 Delay distribution of UGS and rtPS traffic when served by the same policy as the one used only for rtPS traffic

periodically. Lastly, the effective delay for the two classes differs only by a few milliseconds. This fact shows that we could try to treat these two different real-time service classes as one single class. Next, we move on to present the results of treating UGS and rtPS traffic types together as single service type of rtPS traffic. In this case there is no prenegotiated allocation for UGS traffic in the system. The allocations for all service types are done as discussed in the algorithm (see Fig. 10.8) except Step 1. The SSs are polled each at the end of their respective allocations. Figure 10.10 shows the histogram of delay in the case when UGS and rtPS traffic is treated as single rtPS class in the system (other traffic types are not shown here for clarity). Delay distributions for both traffic types are very similar; moreover, the latency and jitter performance for real-time traffic are similar to the one for rtPS traffic when both classes are supported. This shows that the rtPS traffic is able to efficiently use the resources that were previously allocated to the UGS traffic and that UGS traffic can indeed be supported by the policy used for rtPS traffic. In the next section we present some results on polling and also address QoS for various application types.

10.4.6 Impact of Polling and QoS of Various Application Traffic In this section, we present three aspects of treating various traffic types together in 802.16 systems. We first shed more light, via simulations, on the ways polling could

10


203

be done in the system. This will help us understand which method of polling is more efficient in treating traffic demands. We address this objective using the following approach. The BS can choose to poll all SSs either at the end of each superframe or to poll each SS at the end of its uplink allocation time for its next cycle. We see in the following results the effects of these two different polling methods. In the second part, we will see more analysis of our proposed DBA methodology on QoS offered to non-elastic non-interactive, and elastic-interactive applications while they are served by nrtPS class. We also compare QoS offered by rtPS to nonelastic non-interactive applications. The third part presents the simulations of bursty demands in the system and its impact for the applications analysed in the second part above thus checking the robustness of DBA. 10.4.6.1 System Performance Under Different Polling Types We start by describing the scenario and its ingredients. We add two different types of nrtPS traffic from two different sources. The SS1 uses TCP (elastic) where the SS2 uses UDP (non-elastic) as transport protocol, both carrying exponential on/off traffic. This differentiation would allow us to see how the applications like on-line gaming (sent on UDP) get treated if they are carried as nrtPS class. For now traffic configurations remain the same in the system as used in Section 10.4.5 (UGS demands 2 Mbps, rtPS demands about 8 Mbps, nrtPS demands about 8 Mbps, and the rest is BE demand). Figure 10.11 shows the delays for various traffic types when the polling is done at the end of each superframe (i.e., each SS finishes its allocation limited in total by uplink allocations followed by downlink spell in the superframe and then the BS polls each SS). One SS uses Exp-On/Off traffic on TCP whereas another SS carries Exp-On/Off traffic on UDP, both representing nrtPS traffic. First thing which can be seen is that the delays for UGS and rtPS traffic are not symmetric (though still within the acceptable range where UGS is also served via polling). We can also see that nrtps traffic gets served quite early in the system. The majority of this traffic (both from SS1 and SS2) gets served within 2 ∗ superframe time. There is, however, a little difference in which TCP-nrtPS traffic is served compared to UDP-nrtPS traffic. This can be explained by the basic nature of TCP traffic. As UGS and rtPS traffic (prioritized) is served as per their demand (where rtPS demand varies more often), many times nrtPS traffic does not get served which eventually makes TCP to readjust its rate. Another interesting part that can be noticed is towards the tails of nrtps traffic. We see that both sorts of nrtPS traffic exhibit long tail behaviour though there are bumps in the one being served by TCP (which is again explained because of the readjustment of rate and thus more packets being served towards the final phases of traffic service as there are less number of UGS and rtPS packets, prolonging the service time and hence the delay for nrtPS packets). Figure 10.12 shows the simulation delay for various traffic types when the polling is done at the end of individual allocation of each SS. This way BS takes into

204

K. Gakhar et al. Histogram - Delay 2.5 UGS-SS1 UGS-SS2 rtps-SS1 rtps-SS2 Nrtps-SS1 Nrtps-SS2


2

1.5

1

0.5

0 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

Delay (sec)

Fig. 10.11 Histogram – Traffic Delay – Polling is done at the end of Superframe

Histogram - Delay 2.5 UGS-SS1 UGS-SS2 rtps-SS1 rtps-SS2 Nrtps-SS1 Nrtps-SS2

2


Mean-UGS1 = 12.42 ms, Jitter = 9.1 ms Mean-UGS2 = 12.39 ms, Jitter = 9.2 ms 1.5 Mean-rtPS1 = 13.1 ms, Jitter = 8.2 ms Mean-rtPS2 = 13.0 ms, Jitter = 8.1 ms 1

Mean-NrtPS1 = 16.27 ms, Jitter = 27.6 ms Mean-NrtPS2 = 17.6 ms, Jitter = 29.0 ms

0.5

0 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

Delay (sec)

Fig. 10.12 Histogram – Traffic Delay – Polling is done at the end of allocation for each SS

0.08

10


205

account, for the next demand, the packets that arrived after the first request and during the time when the concerned SSs were being served. This method, therefore, takes into account the packets for one SS, lined up to be served (for next cycle) while serving the packets of another SS arrived in its last polling. It can be seen that the delays for real-time traffic (UGS and rtPS) are symmetrical and their packets get delayed for the periods that are well within acceptable range. As the packets of UGS and rtPS traffic types wait for some instants of uplink, nrtPS traffic gets served better. A sharp rise of nrtPS traffic distribution towards the end of 1 ∗ superfame time can be attributed to the fact that some TCP packets get served very early in the allocations as the demands from UGS and rtPS traffic is relatively low and TCP increases its demand. But, we see that over the next superframe as more UGS and rtPS traffic demands arrive (and served with priority) TCP once again gets served later. There is, however, no such noticeable difference of service for nrtPS traffic being served on UDP. We would also like to see the impact of polling on throughput of nrtPS applications. To continue with the above scenarios of traffic demands and polling based DBA we can see nrtPS throughput in Fig. 10.13. The BS allocations were able to serve the demands (4 Mbps for each SS) – well within what was asked. SS1 gets its demands to send Exp-On/Off traffic on TCP. We see thus elastic interactive traffic (e.g., business applications) gets its throughput. Moreover, the effective delay suffered by this traffic, as shown in Fig. 10.12, is of the order of 53 ms. This order of interactivity should not be too cumbersome from a user point of view.

Throughput vs Time 10 nrtPS traffic from SS1

Bandwidth (Mbps)

8

6

4

2

0 0

20

40

60 Time (sec)

80

100

120

Fig. 10.13 Throughput demand of Elastic interactive traffic is well served in polling based DBA

206

K. Gakhar et al.

From these figures we can study the efficiency of polling schemes in providing symmetric delays (thus efficient allocations and usage in DBA policy) to real-time traffic and the throughput for non-real-time traffic being served as nrtPS traffic type. We have had some interesting observations regarding the way TCP traffic delay varied. By comparing these two we see that nrtPS traffic which can easily suffer delays of more than 10 ms is treated differently by different polling schemes. The more number of packets suffer longer delays when the polling is done at the end of superframe in DBA compared to the case where each SS is polled immediately after it finishes its previous allocations thus providing the evidence that the second option of polling is efficient than the first one. In totality of the observations made above, we can see that polling does not really affect delays for UGS and rtPS traffic types when they are served with in the same class. 10.4.6.2 QoS for Non-Elastic Non-Interactive Traffic Keeping the above traffic configuration (there are no bursts in the system and all rates are average with On and Off periods as 500 ms) in mind we first analyse the QoS offered to non-elastic non-interactive applications in the system (IPTV, streaming). This type of traffic is modelled using Exp-On/Off on UDP as seen in Table 10.2. Figure 10.14 shows the delays suffered by these applications in the system. Delay Distribution for Non elsatic Non interactive traffic 1.4e-06 non elastic non interactive- SS1 non elastic non interactive- SS1 non elastic non interactive- SS2 1.2e-06


1e-06

8e-07

6e-07

4e-07

2e-07

0

0

0.02

0.04

0.06 Time (sec)

0.08

0.1

0.12

Fig. 10.14 Delays suffered by non-elastic non-interactive applications in a system employing QoS sensitive DBA which makes service differentiation

10


207

The effective delay for traffic from SS1 (red curve) is observed to be about 22 ms (average delay = 13.06 ms and jitter = 8.4 ms). The effective delay for traffic from SS2 (green curve) is about 22 ms (average delay = 13.08 ms and jitter = 8.2 ms). However, when non-elastic non-interactive traffic is served by service class for nrtPS, the effective delay is observed to be about 53.5 ms (average delay = 16.56 ms, jitter= 27.0 ms). This means that applications like IPTV though can be served by service class for rtPS the effective delays are not so high under normal circumstances for it being served in nrtPS class. Also, we can say that serving elastic interactive traffic in the same class does not really poses any drastic delays for nonelastic non-interactive traffic. 10.4.6.3 QoS Under Bursty Conditions Another interesting aspect is to judge the robustness of our DBA methodology. We do this by analysing the system under bursty scenario. It is achieved by increasing the Off periods and decreasing the On periods and then adjusting the peak rates used for On/Off applications in the system. Here the On period is 200 ms where Off period is 800 ms. This means that peak rates of applications in the system are 5 times the mean rate in the system. Following the same DBA methodology, we obtain the following simulation observation in Fig. 10.15. The effective delay for traffic from SS1 (red curve) is observed to be 29.3 ms (average delay = 14.92 ms, jitter = 14.4 ms). The effective delay for SS2 non-elastic Delay Distribution for Non elsatic Non interactive traffic in Bursty conditions 1.2 non elastic non interactive- SS1 non elastic non interactive- SS1 non elastic non interactive- SS2


1

0.8

0.6

0.4

0.2

0 0

0.05

0.1

0.15 0.2 Time (sec)

0.25

0.3

0.35

Fig. 10.15 Delays in bursty conditions where peak rates for different on/off sources was 5 times the mean rate

208

K. Gakhar et al.

non-interactive traffic (green curve) is 32.5 ms (average delay = 14.52 ms, jitter = 18.0 ms). However, we can see that delays for applications served by nrtPS class are affected in this case as the effective delay is observed to be 387 ms (average delay = 89.0 ms, jitter = 298 ms). It means that under bursty scenarios, the applications like IPTV, if served under bursty conditions will suffer greater delays than the applications might be able to handle. A user experience will not be very satisfying under such scenario. It will be better to provide service via the class serving rtPS. We also take note of the effect that bursty conditions might have on throughput of elastic interactive traffic. Figure 10.16 shows the throughput of applications (like business transactions) when served via service class for nrtPS. Though the throughput does not suffer (demand of 4 Mb is served as clear from the figure) the application get more “zigzag” service. When we compare it with Fig. 10.13, the service such an application gets in normal conditions, it is clear that the application gets served in smaller bursts of times. We saw a detailed study of different traffic types in 802.16 networks. We analysed the necessity of having a simple yet efficient bandwidth allocation policy. With the help of this policy we made propositions of differentiating various service classes for different applications. Further, with the help of guidelines in the standard we map applications traffic to various service types available. Then we move a step further and propose an optimum number of service classes that might easily be used to serve various applications traffic. The results from the simulations show that our

Throughput of Elastic Interactive application in bursty environment Elastic Interactive in SS1 14

12

Bandwidth (Mbps)

10

8

6

4

2

0 0

20

40

60 Time (sec)

80

100

120

Fig. 10.16 Throughput of elastic interactive applications like online business applications under bursty conditions

10


209

propositions attain good values of QoS parameters for various applications traffic. We see that the delays are not affected in these cases when we use our DBA scheme associated with SS scheduling thus proving the efficiency and robustness of our method.

10.4.7 Inferred Results on Service Classes To refer back to Table 10.1 we can now conclude, with the help of above results the type of service classes that can be used to serve such taxonomy of applications (and also justify the notions of service classes used in 802.16). The scheduling is simplified and easy to implement for different traffic classes. The delay and jitter for UGS and rtPS traffic classes are observed to be well within the practical bounds. The following observations can be used as guidelines by a service provider:

r

r r

UGS and rtPS service classes should primarily be used to serve interactive and non-elastic (e.g., VoIP and Video Conferencing services), and non-interactive and non-elastic (e.g., IPTV) applications. We have seen that given some traffic demands in the network, the above-stated classes can be served well within given delay and loss constraints required by these applications. As we saw, UGS class doesn’t have to be bounded via fixed allocations and it can be served well in conjunction with rtPS in the same service class. Applications which are interactive and elastic in nature, like on-line gaming and e-commerce, could be served using nrtPS service class. However, the effective delays suffered due to the bursty nature of other applications in the networks. That would result into longer than expected delays for user applications. Lastly, applications of nature non-interactive and elastic can be served using either nrtPS or BE service classes. These applications being delay tolerant in nature won’t be suffered in their quality of presentation though a user might have to wait a bit longer than his/her nerves can bear for a promised fast network service.

Next, we discuss a novel and simple CAC which improves an earlier work. We present a detailed analysis of the system and advocates an acceptability curve for a service provider.

10.5 Call Admission Control Let us take a brief look on how admission control policy would work for a WiMax (or a general network to say!) system. Admission control (as the name indicates) implements a certain distribution of acceptance percentage among various traffic types in a given system. For example, for given bandwidth pipe in a WiMax system, as in Fig. 10.17, admission control has to provide a policy which, in some presumed conditions, allows different percentages of different traffic types on to available

210

K. Gakhar et al.

Fig. 10.17 Admission control for IEEE 802.16/WiMax system

x UGS

y rtPS Admission Control

z NrtPS m BE Bandwidth Pipe

bandwidth. However, final decision for supporting various traffic types remains to a provider to decide. The 802.16 networks have two characteristics that are seldom associated: the transfer mode is connection oriented but relies on polling to implement multiplexing in the uplink. This is why a straightforward application of existing multiclass CACs may not be appropriate. In this section, first, the state of the art on CAC mechanisms is presented followed by an evaluation of an existing work which forms the basis of our analysis and a new proposition.

10.5.1 Related Works In related works, Wang et al. [21] discussed a scheme which provides highest priority to UGS traffic, maximizing the bandwidth utilization by bandwidth borrowing and degradation modelling. Non-UGS requests are accepted only if the bandwidth is still available. The admission control plays an important role in networks where multiple services such as voice and data are treated. The work by Leong et al. [22] contributes a CAC policy for the cellular system integrating on/off voice and besteffort data services. It also develops a model to characterize the interaction of the voice and data traffic flows which leads to better resource utilization and QoS. It, nevertheless, proposes a CAC which is complex as it consists of two sub-parts, one handling admissions for voice calls and another for data calls. In our own approach, we put more emphasis on adopting a deterministic approach while trying to resolve the admission conditions based on QoS parameters considered in the system. A cross-layer approach for adaptive power allocation and admission control for WiMax networks was proposed by Qian and Chen [23]. The admission control treats all of the traffic from SS in aggregate and CAC manager treats uplink and downlink bandwidth requirements. It argues that in a last-mile scenario the upstream traffic could be a fraction of the downstream traffic which we consider is a simplified assumption. For uplink CAC it assumes complete sharing thus accepting a connection if and only if there are sufficient resources for it. This approach does not take into account any differentiation among various traffic types which can produce unwanted results for real-time traffic. Moreover, for downlink CAC it treats traffic as the one that needs QoS (assembling UGS, rtPS, and nrtPS) and the other one which needs BE.

10


211

An approach to mix pricing as one of the parameters for designing an admission control policy was proposed by Hou et al. [24]. It uses dynamic network conditions to distribute incentives and thus shaping the aggregate traffic in the network. The performance of the system is analysed in terms of congestion prevention, achievable user utility, and earned revenues. In CDMA systems most of the CAC propositions are based on predefined levels of SIR (signal-to-interference ratio) at the receiver [25]. The intelligent CAC proposed by Shen et al. [26] contains a fuzzy call admission processor to make admission decision. These decisions are based on various stochastic parameters such as forced termination probability of handoff, the outage probability of all service types, the next step existing call interference, the link gain, and the estimated equivalent interference of the call request. As we know that we have four different service classes to handle in 802.16 networks so minimum the number of parameters we need to handle simpler our system will be. We, therefore, would rather keep our approach focus on basic traffic parameters needed to play a role on managing admission control. As wireless networks become prevalent the need for even complex admission control policies would arise as argued by Niyato and Hossain [25]. This work discusses general models and basic admission control approaches and then extends them into 4G wireless networks. It highlights the fact that in 4G networks a user will supposedly switch between different access technologies because of mobility hence requiring stringent controls in order to provide QoS. However, their approach made simplified assumptions like one packet in one frame in a TDMA environment and only voice and data packets only. The approach in [27] is multifold. CAC policy follows simple criterion of accepting a connection if there is bandwidth left. Though a differentiation is made for given number of connections of a given class yet only segregation is between BE and “the rest” of traffic type. It proposes Deficit Fair Priority Queue (DFPQ) in order to make sure that BE traffic gets served. Heterogeneous traffic in satellite networks has been considered in a work by Qian et al. [28] where the work proposes an integrated CAC and bandwidth on demand (BoD) MAC algorithm. It uses a multi-frequency time division multiple access (MFTDMA) scheme for uplink multiple access in the system. It proposes the design of CAC in conjunction with BoD. The CAC approach is based on the approach of static allocation and booked allocation.2 The approach of having dynamic CAC to address packet-level QoS while having constraints on connection-level QoS has been considered in [29] but it did not include traffic differentiation when applying CAC. In recent years, the approach of network calculus, initiated by Cruz [30, 31], advocates deterministic bounds on QoS criteria, and in particular the delay for an application. The approach emphasizes that each source, in practice, conforms to some traffic description. The worst case sources are studied in a system from which delay bounds are established. Inspired from this approach, Hébuterne et al. [32] pointed out that an admission control cannot be separated from the bandwidth mechanism used inside the network. Their work makes a comparison of complete

2

For more details please refer to the paper directly.

212

K. Gakhar et al.

partitioning (CP), complete sharing (CS), and generalized processor sharing (GPS) to its proposition of using a class-based partitioning (ClassP) for a CAC process. According to this proposition, the sources are grouped in classes according to their delay, and where classes are managed according to a CP scheme. This work, in fact, directly validates our approach of DBA and intra-SS scheduling where we indeed proposed class-based treatment of traffic providing delays (thus performance) which are well within practical constraints. As we will see in later sections our own CAC proposition (which improves an earlier proposal) advocates deterministic approach towards “delay-based” class-oriented CAC procedure rather than adopting a traffic rate based policy. Another work that was similar in spirit to our own methodology was done by Lenzini et al. [33]. The work investigates the problem of scalable admission control for real-time traffic in sink-tree networks employing per-aggregate resource management policies, like MPLS or DiffServ. They use a network calculus approach to define an admission control algorithm for real-time traffic in sink-tree networks. The algorithm is based on a worst-case delay which has been derived and proved to be tight by using network calculus approach. The algorithm proposed to admit a new flow, following three conditions being checked, if a guarantee can be given that the required delay bound, besides those of other already established flows are not exceeded. Having discussed some works of similar nature, we now present our approach which was inspired from an existing work, discussing pitfalls of their proposition and present a detailed analysis of our own system. Before presenting our own proposition and analysis for CAC in the following paragraphs, we first analyse the work which has been an inspiration for it.

10.5.2 Performance of a Multiclass CAC In the following paragraphs we discuss a mechanism for CAC in 802.16 networks. The proposed mechanism is based on the one presented by [34]. The admission or the rejection is conditioned by the bandwidth availability and the maximum delay requirement for rtPS connections. This mechanism takes into account the different traffic classes defined in this standard. We have different admission conditions for each traffic class. We first define the terminology used in [34]:

r r r r r r r r

f : superframe duration m i ∗ f : maximum delay requirement for connection i Cup : total capacity allocated for uplink transmission CU G S : current capacity allocated to UGS connection Cr t P S : current capacity allocated to rtPS connection Cnr t P S : current capacity allocated to nrtPS connection ri : average bitrate for connection i bi : leaky bucket size of connection i

10


213

The admission control policy for IEEE 802.16 traffic classes are given by the following conditions:

r r r

nrtPS ri < Cup − CU G S − Cr t P S − Cnr t P S UGS ri < Cup − CU G S − Cr t P S − Cnr t P S ∀ k, bk + f ∗ rk < m k ∗ f ∗ (rk /Cr t P S ) ∗ (Cup − CU G S − ri ) rtPS ri < Cup − CU G S − Cr t P S − Cnr t P S bi + f ∗ ri < m i ∗ f ∗ (ri /Cr t P S + ri ) ∗ (Cup − CU G S ) ∀ k, bk + f ∗ rk < m k ∗ f ∗ (rk /(Cr t P S + ri )) ∗ (Cup − CU G S )

For all traffic types, the first condition just checks bandwidth availability. Since it is assumed that nrtPS traffic has less priority than UGS and rtPS, this condition is sufficient for a new nrtPS connection. The second condition (the “delay” condition) for a UGS connection consists in checking that the maximum delay condition shall still be respected for each currently active rtPS connection if the new UGS connection is accepted. Indeed, the 802.16 standard states that the network should be able to guarantee a maximum delay for each rtPS connection. We have similar conditions for a new rtPS connection. For a given k, the delay condition ensures that every bit arriving in a given superframe can be transmitted in less than m i frames. Indeed, the maximum bits received in one superframe is limited by bi + ri ∗ f according to the leaky bucket specification. Moreover, the rate factor takes account of the fact than UGS has more priority than rtPS, and that rtPS has more priority than nrtPS, which is translated into the scheduling policy. However, we note that accepting a new UGS or rtPS connection involves as many tests as the number of rtPS connections already active. This is obviously less than practical since this number may be arbitrarily large. We address this particular point in the following paragraphs by proposing a global condition to check maximum delay performance. Let’s assess the performance of the CAC proposed by [34]. In this study, we compute the dropping probabilities versus offered load ρ versus traffic profile parameters. Table 10.3 resumes the simulation parameters. The tolerated delay of an Table 10.3 Simulation parameter values Parameters

Values

Cuplink ri f m min m max bi ρ ρU G S ρr t P S ρnr t P S

67 Mbps 1 Mbps 1 ms 10 ms 50 ms 30000 bits 30–70 erlang ρ/3 ρ/3 ρ/3

214

K. Gakhar et al.

rtPS connection is taken randomly between m min and m max . As in [34], we assume that the frame duration is 1 ms. However, the standard has evolved and a value of 10 ms is considered more appropriate for a superframe duration. We also present and discuss results with 10 ms frame as we move on.3 Figure 10.18 shows the blocking probabilities versus the leaky bucket size (b) for r = 1 Mbit/s. It shows clearly the influence of the burstiness (represented by b); when b increases, the blocking probabilities for UGS and rtPS traffic sharply increases while the blocking probability for nrtPS traffic decreases. This should obviously be avoided; although the conditions for high priority traffic may be more stringent, it is not desirable that low priority traffic starve high priority traffic. As a further example of the system we see in Fig. 10.19 the dropping probability of various traffic types as the system load increases. It is clear that with increasing traffic load in the system real-time systems suffer much more than the non-real-time systems. Therefore, while the conditions proposed in [34] achieve an admission control for various service classes in the standard their mechanism penalize the performance of higher priority traffic while favouring low priority traffic. 0.35 P(UGS) P(rtPS) P(nrtPS) 0.3

Dropping probability

0.25

Cuplink = 67 Mbps f = 1 ms ρ = 67 erlang r = 1 Mbps m : random [10,50] ms

0.2

0.15

0.1

0.05

0

0

0.5

1

1.5

2

2.5

3 4

Leaky bucket size, b (bit)

x 10

Fig. 10.18 Dropping probability versus leaky bucket size b 3 Considering the WirelessMAN-SC, NLOS, using OFDM has a frame of 10 ms, 25 MHz channel with bit rate 40 Mbps.

10


215

0.35 UGS rtPS nrtPS 0.3


0.25

0.2

0.15

Cuplink = 67 Mbps f = 1ms r = 1 Mbps b = 30 Kbit m : random [10,50] ms

0.1

0.05

0 30

35

40

45

50

55

60

65

70

Traffic load (ρ) Fig. 10.19 The dropping probability versus the traffic load ρ when b = 30 k

10.5.3 Novel Call Admission Control In this section we propose an improved approach of CAC. Our approach modifies the second condition of traffic acceptance compared to what has been discussed for UGS and rtPS traffic in [34]. Our first objective is to substitute a global condition to the set of individual conditions, one for each rtPS connection. Our second objective is to limit the unfairness of the CAC regarding UGS and rtPS traffic. The following equations illustrate the admission control conditions for each traffic class:

r r

nrtPS ri < Cup − CU G S − Cr t P S − Cnr t P S UGS ri < Cup − CU G S − Cr t P S − Cnr t P S k (bk + f.r k,r t P S ) + k ( f.r k,U G S ) + f.ri,U G S ∀k, < f.min(m k ) Cuplink

216

r

K. Gakhar et al.

rtPS ri < Cup − CU G S − Cr t P S − Cnr t P S bi + f.ri,r t P S + k (bk + f.rk,r t P S ) + k ( f.rk,U G S ) ∀k, < f.min(m k ) Cuplink

For UGS traffic, the admission condition of the connection concerns the bandwidth availability: ri < Cup − CU G S − Cr t P S − Cnr t P S . It must also guarantee a normal operation of all rtPS connections. To do this, the admission condition must consider the worst case. It means that a new UGS connection is accepted if the available bandwidth can be used to transmit the rtPS connection which has the smallest maximum tolerated delay. This results in the following delay condition:

k (bk

+ f.rk,r t P S ) + k ( f.rk,U G S ) + f.ri,U G S < f.min(m k ). Cuplink

Finally the condition to admit an rtPS connection can be achieved if the following condition is satisfied: bi + f.ri,r t P S +

+ f.rk,r t P S ) + Cuplink

k (bk

k(

f.rk,U G S )

< f.min(m k )

Figure 10.20 shows that the variation of the dropping probability versus the traffic load (ρ). The interest of this new mechanism is clearly emphasized since the dropping probability of the rtPS traffic decreases strongly. Indeed, the bandwidth used by the nrtPS traffic was so large with the mechanism of [34] that was having a smaller dropping probability. With the new mechanism the bandwidth is allocated in a better manner since the rtPS and the UGS traffic take a part of the nrtPS bandwidth, this decreases the dropping probabilities of these two traffics (UGS and rtPS) and increase a little the probability of the nrtPS traffic. In this new CAC, the delay condition for a UGS connection consists in checking that all the real-time traffic (UGS and rtPS) that arrives during a given frame should experience a delay which is smaller than the most stringent bound f.min(m k ) currently negotiated for rtPS connections. The delay for accepting a new rtPS connection is similar. This delay condition assumes that UGS and rtPS connections share the total available bandwidth Cup if requested, nrtPS traffic being served with less priority. This means that the CAC does not take into account that UGS traffic has higher priority than rtPS traffic, leaving the scheduling to the DBA and the intra-SS scheduling mechanism. In results, we see that the blocking probabilities for UGS and rtPS traffics are significantly better with new CAC, while the blocking probability for nrtPS traffic correspondingly increases. This is because we assume that UGS and rtPS traffic together have prior access to transmission opportunities; by relaxing the delay conditions, the blocking probabilities decrease for both UGS and rtPS traffic.

10


217

0.35

0.3

UGS new CAC rtPS new CAC nrtPS new CAC UGS rtPS nrtPS


0.25

0.2

CUplink = 67 Mbps 0.15

f = 1 ms r = 1 Mbps b = 30 Kbit m : random [10,50] ms

0.1

0.05

0 30

35

40

45

50

55

60

65

70

Traffic load (ρ) Fig. 10.20 The dropping probability versus the traffic load ρ (New CAC)

Moreover, since the delay conditions for UGS and rtPS traffic are now very similar, the blocking probability performance for the two traffic classes are now very close. This implies that supporting UGS traffic within the rtPS class should not be a problem and has not impact on the blocking performance for either rtPS or UGS traffic; a CBR traffic has a low bucket size, which means that its impact on the delay condition is going to be negligible.

10.5.4 System Analysis This section takes a critical view of the proposition presented in the previous section (Novel CAC) by evaluating 802.16 system under various traffic conditions. It is interesting to note that CBR traffic normally needs small bucket sizes in a deployment. It means that in case of collective demands the parameters b, r and f can play crucial role for realizing a deployment. In the following paragraphs we present some results showing various values of traffic parameters and study its performance for the blocking probabilities of different traffic types. The aim of this analysis is to present a balanced view of how traffic parameters could play a significant role in reality (for a service provider) when it comes to network management policies. We

218

K. Gakhar et al.

0.015 UGS rtPS NrtPS

Dropping Probability

0.01

Cuplink = 40Mbps f = 10 ms r = 0.1 Mbps MaxDelay = 20 ms b = 2700 bits rhoUGS = rho * 0.3 rhoRtPS = rho * 0.2 rhoNrtPS = rho * 0.5

0.005

0 10

15

20

25

30

35

Traffic Load (ρ)

Fig. 10.21 Dropping probability for small r and b values with maxdelay 20 ms

also used different traffic configurations which are more apt to the standard currently and are shown on each figure. Figure 10.21 shows the dropping probability versus traffic load for new CAC policy. The value of r is 0.1 Mbps and that of b is 2700 bits. Also the loads of various traffic types are different than the values used earlier; UGS = ρ∗0.3; rtPS = ρ ∗0.2; nrtPS = ρ ∗0.5. In this case we see that nrtPS traffic has almost no dropping probability whereas UGS and rtPS traffic dropping probability between 1% and 2%. The maximum delay considered for traffic types is 20 ms. Figure 10.22 shows the dropping probabilities for different traffic types when we increase the value of only maxdelay compared to the one in Fig. 10.21. We realize that as the delay tolerance increases the accepted load for UGS and rtPS increases a bit though there is also a corresponding increase in the dropping probability of UGS and rtPS traffic (Fig. 10.23). Next, we change the value of b and increases it to 20 kbits, that is, the bucket size is now 20 kb. We can see a sharp increase in the dropping probabilities of UGS and rtPS traffic which need to satisfy higher bandwidth values before they can be accepted successfully for new connections. The above scenario doesn’t change much even if we increase the maxdelay value to 30 ms as we can see in Fig. 10.24. It means that such traffic profiles won’t be very practical for a normal practical scenario as the corresponding loads of UGS and rtPS traffic won’t be significant enough. We can see that the blocking probabilities for UGS and rtPS traffic types have higher values than those of NrtPS traffic. The charge considered in the system consists of 50% of real-time traffic (UGS and

10


219

0.025 UGS rtPS NrtPS


0.02

Cuplink = 40 Mbps f = 10 ms r = 0.1 Mbps MaxDelay = 30 ms b = 2700 bits

0.015

rhoUGS = rho * 0.3 rhoRtPS = rho * 0.2

0.01

rhoNrtPS = rho * 0.5

0.005

0

0

5

10

15

20 Traffic Load (ρ)

25

30

35

40

Fig. 10.22 Dropping probability for r and b values with maxdelay 30 ms

0.4 UGS RtPS NrtPS


0.35

0.3

Cuplink = 40 Mbps f = 10 ms r = 0.1 Mbps b = 20000 bits MaxDelay = 20 ms

0.25


0.2


0.15

0.1

0.05

0 10

15

20

25 Traffic Load (ρ)

Fig. 10.23 High dropping probability for UGS and rtPS as b size increases

30

35

220

K. Gakhar et al.

0.35

UGS 0.3


0.25

RtPS

Cuplink = 40Mbps f = 10 ms r = 0.1 Mbps b = 20000bits MaxDelay = 30ms

NrtPS

rhoUGS = rho * 0.3 0.2

rhoRtPS = rho * 0.2 rhoNrtPS = rho * 0.3

0.15

0.1

0.05

0 0

5

10

15

20

25

30

35

40

Traffic Load (ρ)

Fig. 10.24 Higher delays have almost no effect on dropping probabilities for high b values

rtPS) and 50% of NrtPS (we assumed that BE traffic gets treated as per bandwidth availability). Now we see the case when the value of r is increased to 1 Mbps. Figures 10.25 and 10.26 show the case when b is 2000 bits and maxdelay is 20 and 30 ms, respectively. We observe that now we have some dropping probability for nrtPS traffic in the system. Even though some realistic value is about 2% dropping probability in implemented systems here we want to show how the system behaves for higher values of r . Also we want to make one comment on maxdelays considered here; in commercial products and in literature we find the values relatively higher than the values considered here, thus we present a strict system behaviour here. To see the worst case scenario, we check Fig. 10.27. Here the value of maxdelay is kept to 20 ms, r is 1 Mbps whereas b is increased to 20 kb; means that traffic need huge bursts with low values of delays to respect (that could be real-time critical situations). We see that the number of real-time connections that can be accepted with lower dropping probabilities are very small with the above system configurations. It is clear from these observations that we have certain limits on the number of connections that can be entertained by a multiservice environment. This limit on the number of connections comes into effect as soon as the required QoS for a traffic type starts to deteriorate beyond a certain value. And, this in fact, is decided by the admission and delay conditions used in a system as shown via [34] and our own proposition. This means that an admission control policy plays a role of deciding how different traffic types can be categorized in order to serve them in the system.

10


221

0.4

0.35

UGS RtPS NrtPS


0.3

0.25

Cuplink = 40 Mbps f = 10 ms r = 1 Mbps MaxDelay = 20 ms b = 2000

0.2


0.15


0.1

0.05

0 0

5

10

15

20

25

30

35

40

Traffic Load (ρ)

Fig. 10.25 Large dropping probabilities for UGS and rtPS when r is 1 Mbps with some blocking for nrtPS traffic 0.35

UGS RtPS NrtPS

0.3


0.25 Cuplink = 40Mbps r = 1Mbps f = 10 ms b = 2000 bits MaxDelay = 30 ms

0.2

rhoUGS = rho * 0.3 0.15

rhoRtPS = rho * 0.2 rhoNrtPS = rho * 0.5

0.1

0.05

0 0

5

10

15

20

25

30

Traffic Load (ρ)

Fig. 10.26 Large dropping probabilities for UGS and rtPS when maxdelay is 30 ms

35

40

222

K. Gakhar et al. 0.7 UGS RtPS

0.6

NrtPS Cuplink = 40Mbps r = 1Mbps f = 10ms b = 20000 bits MaxDelay = 20 ms


0.5

0.4 rhoUGS = rho * 0.3 rhoRtPS = rho * 0.2 0.3


0.2

0.1

0 0

5

10

15

20

25

30

35

40

Traffic Load (ρ)

Fig. 10.27 Worst case: with smaller delay and high value of b

We have seen in the earlier that novel CAC favours more UGS and rtPS traffic to NrtPS traffic thus giving better system behaviour than the one discussed in [34]. Nevertheless, this new policy depends on the system configuration. Note that the set of individual conditions in [34] (one for each active rtPS connection) are thus replaced in this new CAC by a single one. This is obviously more practical, while being realistic in an operational framework. Indeed, in such a framework, it is not up to the user to fix the maximum delay but up to the operator to state what type of traffic is supported. It is very likely that network operators shall propose limited sets of traffic classes, which fix delay conditions according to traffic engineering capabilities. If a single delay condition m is offered, the upper bounds in the delay conditions become f.m. Therefore, as an example of such a system we propose a limit on the number of connections that can be entertained. Assuming that we have rtPS and NrtPS connections in the system then for a given traffic configuration, where the blocking probability is 3.5%, the number of rtPS connections are limited to 22 and the number of NrtPS connections are limited to about 32 as shown in Fig. 10.28. These values are matched and found to be almost the same as obtained by Erlang’s formula. Thus it is shown that an 802.16 system can very well support real-time and non-real-time traffic given a required profile. As discussed in the above paragraphs, it is the service provider to decide on how best he wants to configure the system. Now we move on to the second contribution. We first discuss some related works found on interworking in related domains. Then we present our novel proposition of a “tight coupling” based interworking architecture.

10


223

Acceptability Curve for Blocking Proba of 3.5%

35 Acceptability Curve

30

Cuplink = 40 Mbps r = 1Mbps f = 10 ms MaxDelay = 20 ms b = 20000 bits

nrtPS Charge

25

20

15

Zone of Acceptability of new Connections

10

5

0

0

5

10

15

20

25

rtPS Charge

Fig. 10.28 Acceptability curve in 802.16 system

10.6 Related Works In this section we look at some of the approaches which treat similar problematics. The work in [35] discusses the interworking options among cellular networks and WLANs arising due to business opportunities. The authors present a generic management platform suitable for WLANs interworking with other wireless systems. It focuses on the implementation of a high-speed wireless environment and the ability to maintain multimedia services in hotspot locations. The system is based on a common all-IP platform facilitating mobility using mobile IP. The work is an approach that can be classified as “loose-coupling” approach wherein various management modules are used to ensure interworking among different networks. There have been some recent studies on interworking between UMTS and WiMax [36, 37]. The motivation in [36] is to address ubiquitous communication in various wireless technologies enabling easy handover among them. It focuses on an UMTS-WiMax mobile IP based interworking architecture hiding the heterogeneity of lower-layer technologies. It describes the handover from WiMax access networks to UTRAN and also from UTRAN access networks to WiMax with the help of standard modules already existing in them. The work in [37] suggests an interworking based on tight coupling with wireless access (TCWA). It improves on their earlier work by proposing that the data traffic in inter-network communications can be dynamically distributed as the added wireless links provide alternative routes.

224

K. Gakhar et al.

These works have been focusing more on consolidating handover aspects within a hybrid system so as to provide the quality of network access. One more work based on “loosely coupled” approach, integrating 3G and 802.16 networks was published in [38] and also contains many similar references. The authors present a novel network and mobile client architecture that provides seamless roaming and mobility while maintaining connectivity across the heterogeneous wireless networks, WiMax and 3G, and provides QoS support. By integrating the 3G and 802.16 wireless networks, users will experience seamless and transparent wireless connectivity in more places and with better service than what can be obtained by a single type of network. The proposed architecture incorporates the procedures for activating a QoS session, the translation between network-specific QoS classifications, and network and session layer QoS support. They also propose a SIP-based mobility scheme that is capable of providing QoS support across different networks. Another work [39] for a hybrid system of 802.16 and 802.11 networks proposes a scheme of vertical handoff between these networks while keeping the available bandwidth as the metrics. The proposed algorithm reduces the unnecessary handoff probability because of the signal strength temporarily dropping down. It also discusses a two hop architecture using dual-mode relay gateway (which could be mobile) within a hybrid network also consisting of AP for WLAN. It is, once again, can be classified as “loosely-coupled” architecture which does not address QoS at MAC level. Wireless mesh networks (WMNs) have also been in focus for some time now. IEEE 802.16a [40] amends 802.16 by adding the specification for an air interface for both point-to-multipoint and mesh systems operating in the 2–11 GHz frequency range. We refer to some existing work and try to find some commonalities among our objectives and the available works in mesh networks. WMNs are dynamically self-organized and self-configured with the nodes in the network automatically establishing an ad hoc network and maintaining the mesh connectivity. The architecture for mesh networks can be classified into three types: (a) Infrastructure (Backbone) WMNs, (b) Client WMNs, and (c) Hybrid WMNs. Hybrid WMNs, with their ability to combine both infrastructure and client sides, are supposed to play an important role in coming years to propagate broadband services and provide QoS for applications though some research challenges for capacity evaluation, security and MAC level protocols are to be evaluated. The work in [41] highlights the need of efficient routing combined with better scheduling to eventually achieve QoS for UDP connections (serving video and voice) and also for data connections in WMNs of 802.16 systems. It starts by proposing a routing algorithm which is fixed and does not vary with time as for wireless channels. It provides a good performance for UDP and TCP connections though it does not claim to be optimal for either of them. It considers detailed scheduling for CBR and VBR traffic taking into account the number of slots that have to be allocated to a connection flow given a certain dropping probability for that traffic type. Using the similar principle it develops models to provide QoS to TCP, first with a fixed allocation scheme and then with an adaptive fixed allocation scheme. It further develops a detailed study to assure QoS for different TCP flows

10


225

sharing the bandwidth and eventually showing that adaptive scheme provides more than minimum required throughput to most of the flows. An analytical distributed scheduler was proposed in [42] for 802.16 in mesh mode. The model assumes that the transmit time sequences of all the nodes in the control subframe form statistically independent renewal processes. It estimates distribution of the node transmission interval and connection setup delay which have effects on throughput and delay. The goals of IROISE have been to advocate policies to achieve and to maintain QoS that a user is going to experience in a two hop hybrid wireless network of 802.16 and 802.11. This puts focus more on intrinsic ability of the proposed architecture to achieve the aforementioned objectives. This lead us to think about an approach that would be more “tightly coupled” and hence won’t need much external modules to ensure it. The work in [43] has been partially an inspiration for our approach. It defined interworking mechanism between WMAN and WLAN, in particular HiperMAN (High Performance Radio Metropolitan Area Network) and HiperLAN (High Performance Radio Local Area Network), both the standards defined by the European Telecommunications Standards Institute (ETSI). It studied the interworking between HiperLAN/2 and HiperMAN at three levels4 :

r r

r

IP level: This option was studied using IPv6 header fields, giving even new interpretation for “Flow Label” field. However, the study itself pointed out that implementing such approach was not possible in 802.11 networks and the proposed interpretation could not be implemented. Ethernet level: This approach inspired from the bridging mechanism available in [44] and its subparts. With the limitations of the number of class of services in [44], the approach at Ethernet level has been considered practical. However, it wouldn’t be possible to actually translate traffic requirements on parametric level (that could be better addressed at MAC level). Data Link Control (DLC) level: The bridging approach at this level could have been the simplest solution but it has not been possible to map service flow parameters to QoS parameters between two standards.

It was recognized in our work that it was necessary to identify the potential parameters which could eventually be used for self-information transmission in an hybrid system. With the advent of 802.16 and 802.11e (addressing parameters that could support QoS) it was sensed that it could eventually be possible to establish interworking between these two different paradigms. The work was addressed eventually in [45]. It was considered to address dynamic resource reservation policies for 802.16 systems. As the market evolves for these systems, specially in licensed band, it will be useful for operators to have resource allocation policies which could bring profits and also satisfy users. The subject was studied with assumptions of hysteresis mechanism and the relevant results are published in [6]. This work also addresses

4

For details please refer to [43].

226

K. Gakhar et al.

traffic in a multiservice environment of WiMax. It is clear from current trends in applications that multiservice networks are going to be a essential for every new system. With that in mind, the results have been obtained which prove that instead of supporting number of service classes recommended in the standard (infact four), one can easily have all applications requirements fulfilled with only three service classes. We will see the subsequent details in the following chapters.

10.7 InterWorking Architecture This section proposes a tightly coupled architecture where MAC level mechanisms inherently help to achieve interworking among WMAN and WLAN systems. We address the “matching” between traffic parameters as found in IEEE 802.16 and in IEEE 802.11e systems. In IEEE 802.16, we deal with various application flows by handling them via various scheduling services as in Table 10.4. In IEEE 802.11e, the access mechanisms help in achieving QoS requirements for an application. However, this similitude does not mean a direct conversion of traffic category from one system into another and vice versa. Our proposition for a hybrid architecture can be seen in Fig. 10.29. The radio gateway (RG), as perceived, works as a Subscriber Station for the IEEE 802.16 network and also as a QAP for the IEEE 802.11e network. In order to address the goals set for the project work (providing real-time audio/video, audio/video on demand, precious data transfer) we have identified the following traffic classes which could be made up of various traffic parameters found in the drafts/standards. These classes are worked out depending upon a traffic type and its QoS requirements and should not be conflicted with categorized traffic services in Table 10.4.

r r

CBR with Real-Time Traffic (C1): Applications like real-time audio/video fall into this class. The desirable characteristics for this class are very limited packet losses, minimum latency delays and very little jitter. VBR with Real-Time Traffic (C2): Examples of traffic for this class include video on demand (streaming) and variable rate VoIP. Again, packet losses, minimum latency delays and jitter limits apply to such traffic though their values could be more tolerable compared to those of class C1.

Table 10.4 Mandatory QoS parameters for traffic categories UGS

Maximum sustained traffic rate, maximum latency, tolerated jitter, and request/transmission policy

rtPS

Minimum reserved traffic rate, maximum sustained traffic rate, maximum latency, and request/transmission policy

nrtPS

Minimum reserved traffic rate, maximum sustained traffic rate,, traffic priority, and request/transmission policy

BE

Maximum sustained traffic rate, traffic priority, and request/transmission policy

10


Fig. 10.29 IROISE: The proposed architecture

227 802.16 Radio Gateway SS MM 802.11e / WiMAX

BS

QAP non-AP QSTA

non-AP QSTA

non-AP QSTA

BS: Base Station SS: Subscriber Station QAP: QoS Access Point non-AP QSTA: QoS Stations MM: Mapping Module

r r

VBR with Precious Data (C3): This class addresses traffic type like large data files. However, in this case traffic characteristics are more delay tolerant with a need of minimum packet loss. Unspecified Type (C4): This class contains simple best effort type traffic such as web access, emails etc. So the traffic is purely BE.

Lets zoom-in on the radio gateway as in Fig. 10.30. The QAP module, after receiving a request from a non-AP QSTA, forwards the traffic identifier (TID) of an Radio Gateway (RG)

T R

S Subscriber Station (SS)

MM

Mapping Table QoS Access Point (QAP)

Fig. 10.30 Radio Gateway supporting QoS

T: Transmitter S: Scheduler R: Receiver

T R S

228

K. Gakhar et al.

application flow along with the priorities/parameters that convey QoS requirements of an application to the mapping module (MM). The MM then maps the incoming traffic parameters to the ones that are supported for an IEEE 802.16 application flow. Based upon the traffic priorities discussed in an IEEE 802.11e network (first 8-bits of TID) as well as the traffic classes (per-flow traffic characteristics), we propose two different kinds of mappings.5 The first kind of mapping is “prioritized mapping”. In this mapping, the traffic priorities for an application flow, as in 802.1D [44], coming from a WiFi network are mapped to the corresponding traffic class in an IEEE 802.16 network and vice versa. The second kind of mapping is per-flow “parameterized mapping” as illustrated in Fig. 10.31. It depends upon optional/mandatory traffic parameter requirements for an application flow though more optional parameters (found in the drafts/standards) could be used depending upon the technical and/or the financial requirements of a network. However, the handling for these two kinds of mappings remain MM implementation dependent. Following this mapping the whole process of connection setup in an IEEE 802.16 network (as discussed in [45]) requesting QoS for an application flow is executed by the SS module present on the RG. As discussed in [1], the QoS requirements for an application flow can be sent in MAC CREATE SERVICE FLOW.request along with the scheduling required. However, whether the request is served or not depends upon the resources available to the BS. Similarly, for the downlink, once the SS receives an application flow it is forwarded to the MM. The MM identifies the incoming flow with its SFID and associates it with the corresponding TID that it received with the request from a non-AP QSTA. This mapping between SFID and TID would then be used until the completion of data transmission for an application flow. Obviously, during this whole process we will need to buffer the incoming traffic at the RG being used. We now discuss the proposed mapping in detail. For this sort of mapping to work the traffic characteristics pertaining to a class, as seen in the mapping table, in one system (say IEEE 802.11e) should be interchangeable with the similar traffic characteristics in the other system (say IEEE 802.16).

r r r

5

Both Maximum Sustained Traffic Rate and Peak Data Rate specify the peak information rate of the service in bits per second. They do not include MAC overhead such as MAC and PHY headers. Maximum Latency and Delay Bound asserts the maximum latency periods within their respective networks, representing a service commitment, starting at the time of arrival of a packet at the local MAC sublayer till the time of successful transmission of the MSDU to its destination. The following terms are used in the following equations: D : Delay, max D : Delay Bound, Dq : “Queueing” Delay, Dt : Transmission Delay, J : Tolerated Jitter. We consider that Dq includes all types of delay (buffering, scheduling, retransmission) except transmission delay. Note that max D and data rate are not independent and also Dt is proportional to data rate. Indeed, we observe that:

For more details on QoS setup in IEEE 802.11e please refer to [45].

10


Fig. 10.31 Parameterized mapping

229

IEEE 802.11e

IEEE 802.16

Traffic Class C1

Traffic Class C1

Peak Data Rate

Maximum Sustained Traffic Rate

Delay Bound

Maximum Latency

(Minimum PHY Rate + Delay Bound)

Tolerated Jitter

Traffic Class C2

Traffic Class C2

Minimum Data Rate

Minimum Reserved Traffic Rate

Peak Data Rate


Delay Bound Burst Size

Maximum Latency Maximum Traffic Burst

Traffic Class C3

Traffic Class C3 Minimum Data Rate

Minimum Reserved Traffic Rate

Peak Data Rate


User Priority

Traffic Priority

Burst Size

Maximum Traffic Burst

Traffic Class C4

Traffic Class C4

Peak Data Rate


User Priority

Traffic Priority

D = Dq + Dt

(10.1)

min D ≥ min Dq + min Dt max D ≤ max Dq + max Dt .

(10.2) (10.3)

max Dq ≤ max D − min Dt

(10.4)

min Dq = 0.

(10.5)

Also, we can say that

230

K. Gakhar et al.

Jitter for an application can be defined as: J = max D − min D.

(10.6)

We introduce an upperbound for “jitter” experienced by an application in an IEEE 802.11e network (no notion exists in [46] ). Using the above equations we deduce an upper bound for jitter as: J ≤ min(max D, max D + max Dt − 2 ∗ min Dt ).

r r

r

(10.7)

Therefore from the values of Data Rate and Delay Bound of an application request from an IEEE 802.11e network, J in an IEEE 802.16 network could follow the upperbound deduced in Eq. (10.7). Both Minimum Data Rate and Minimum Reserved Traffic Rate are the minimum data rates specified in bits per second and map to similar requirements of an application flow. In this mapping MAC headers are not counted. Traffic Priority determines the priority among two service flows identical in all QoS parameters except priority. However, for class C3/C4 type traffic from an IEEE 802.11e network, traffic priority is mapped onto by user priority (UP) assigned to an application flow. So UP and Traffic Priority play a similar role when it comes to mapping. Burst Size specifies the maximum burst of the MSDUs belonging to a TS which arrives at the MAC SAP at peak rate. Maximum Traffic Burst describes the maximum continuous burst the system should accommodate for a service. It also assumes that the service is not currently using any of its available resources, i.e. the instant when an MSDU arrives at MAC.

10.8 Conclusion In the following paragraphs we summarize our work with small discussion on future work in sight. This chapter discussed the multiservice nature of a WiMax network. We have based our study on the four traffic classes as specified in the 802.16 standard. We have shown that neither transfer plane QoS delay parameters nor command plane blocking parameters are greatly impacted by supporting UGS traffic within the rtPS traffic class. This leads to our proposal of offering only three traffic classes instead of four. This may have a significant impact on the cost of Network Interface Cards for network elements in WiMax networks since dealing with two separate scheduling mechanisms (periodic for UGS and polling based for the other classes) greatly increases the complexity of the design. The paper also proposes very simple mechanisms, both in the command plane (CAC) and in the transfer plane (DBA and intra-SS scheduling) and presents preliminary results showing that that these mechanisms can indeed provide WiMax network with a robust multiclass support. This implies that packet scheduling can have a very small impact on the cost of

10


231

WiMax cards. Furthermore, we have shown that a simple multiservice CAC can support real time, CB data service and BE traffic with little complexity. It is up to the network operator to specify traffic profiles that ensure a good utilization of the link. In this work we also discuss various paradigms evolved these last years for interworking among different wireless technologies. To address the objectives within the project we decided to adopt a “tightly coupled” approach for an interworking architecture for 802.16 and 802.11. We proposed a matching among the parameters of these two different systems. The proposed mapping may evolve further in details but is not the only factor that will count when it comes to ensure QoS. The process of establishing data transmission including buffering, proposed mapping, setting up of a new connection etc. will take some initial “setup” time. A synchronization should be ensured between the arrival of data at the RG and transmission opportunity (TxOP) available to QAP module. That will largely depend upon the behaviour of the mapping module (MM) which ensures the mapping. The role of MM is multifold: It has to ensure the integrity of the incoming and the outgoing traffic (in either direction). The scheduling policy for the traffic inside the RG has to make sure that application flows are channelled to the corresponding connections (real/virtual). Besides the traffic handling inside the RG, scheduling policies within the individual networks should ensure that QoS sensitive applications get served in time along with the bandwidth constraints which in turn would also depends upon the dimensioning of such a system.

Appendix: BlocQ Implementation Next, we briefly discuss the simulation utility that has been used to realize our scenarios before we move to detailed analysis. The basic simulation utility used in our work was the discrete event based network simulator popularly called ns-2. At the time we started our work, to the best of our knowledge, there was no known implementation of MAC layer of IEEE 802.16 for ns-2. Thus we needed to make certain hypothesis for simulating TDMA mechanism of 802.16 MAC layer. We consider that there are no losses in the system due to wireless conditions. We implemented a buffer management technique, modifying DropTail implementation in ns-2, which we call BlocQ.6 To achieve “polling” mechanism BS regularly calls an “update function” within its allocation policy (which could be done at different instances). This function then sends the sum of present number of bytes in its various queues of traffic sources which, in fact, is translated as demand of an SS. The BS then makes uplink bandwidth allocation to a SS in proportion of the available link bandwidth as: Demand SSi /Link BW

6

The code is available with the author.

232

K. Gakhar et al.

These demands are then redistributed internally in a SS to individual queue demands of each source (again in proportion to the demand of a source to the sum of demands of all traffic sources). Following these allocations each source is either allowed or not to send traffic (zero allocation) on the uplink. On the downlink, BS braodcasts the allocations to all SSs. The links in this direction are FIFO in nature and carries the downlink traffic (acknowledgements etc) to SS. In simulations each flow is identified by its corresponding number (in addition to its source and destination) which facilitates the receptions of downlink traffic to its destination. It was necessary to define BlocQ because changing DropTail functions directly had an impact on the functionality of other modules in ns2 (as we observed for some simple CBQ examples). This extension to ns-2 enables it to simulate the time-division multiplexing (TDMA – static and dynamic) in the upstream direction. For the sake of simplicity, we have chosen not to explicitly simulate the control messages (though their time duration was kept during simulations which is of order of few ␮ sec). Instead, we have implemented two procedures, one for evaluating the requirements of SSs, the second for controlling transmission. The transmission windows in the next polling cycle for each SS are computed using the requirements evaluated, e.g. by measuring the buffer size of each traffic class. In order to implement TDM behaviour in ns2 we implemented a variant of existing DropTail scheme in the system. It is called “BlocQ” which a user can control so as to command a queue not to send packets when it is blocked. It provides the behaviour of “block” and “unblock” in order to introduce dynamic allocation behaviour without any impact on ns2 functioning in general. The proposed solution is to have “BlocQ” as per its functions described in the lines above. The general view is “Queue/BlocQ”. To use it in a program we just need to declare a link as “BlocQ”. In fact, the code had to be adapted to ns2 in order to have a clean functioning of allocation policy. Steps needed to implement “BlocQ” functions are as:

r r r r

We keep the “virtual” option that was introduced in “queue.h” file i.e. we convert “void resume()” to “virtual void resume()”. We change all “DropTail” by “BlocQ” (changing class names) in “droptail.cc” and “droptail.h”. We make the changes introduced in “droptail.h” but the new file is reffered to “blocq.h”. We need to add the following lines in “nsdefault.tcl” file found in “/HomeDir/ns2.28/tcl/lib/” directory so that default values for BlocQ are set and it is known to ns “environment”. Note that first line is a comment and all other values are exactly the same as used for “Queue/DropTail” option. # BlocQ Queue/BlocQ Queue/BlocQ Queue/BlocQ Queue/BlocQ

set set set set

drop_front_ false summarystats_ false queue_in_bytes_ false mean_pktsize_ 500

10

r r


233

We also need to add “blocq.o” option in “Makefile” found under “/HomeDir/ns2.28” directory. It is done by adding “queue/blocq.o” in a place where other entries like “queue/droptail.o” are found. Once these chages are made, first do “make clean” inside “/HomeDir/ns2.28/” directory followed by “make depend” and “make” in the end. If there is no compilation error that means you are ready to use it. In case of error please check if there are any class hierarchy changes in your version of ns compared to ns2.28 (actually used for this process) or simple compilation problems.

References 1. Part 16: Air Interface for Fixed Broadband Wireless Access Systems, IEEE Std. 802.16, 2004. 2. Part 11: Wireless LAN Medium Access Control (MAC) and Physical layer (PHY) specifications, IEEE Standard, 1999. 3. R. Branden, D. Clark, and S. Shenker, “Integrated Services in the Internet Architecture: an Overview,” RFC 1633, Jun 1994. 4. S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, and W. Weiss, “An Architecture for Differentiated Services,” RFC 2475, Dec 1998. 5. S. Bajaj, L. Breslau, and S. Shenker, “Is Service Priority Usefeul in Networks?” in ACM SIGMETRICS 98, 1998, pp. 66–77. 6. K. Gakhar, M. Achir, and A. Gravey, “Dynamic Resource Reservation in IEEE 802.16 Broadband Wireless Networks,” in Fourteeth IEEE International Workshop on Quality of Service (IWQoS 2006), Jun 2006, pp. 140–148. 7. Y. H. Zang, D. Makrakis, S. Primak, and Y. B. Huang, “Dynamic Support of Service Differentiation in Wireless Networks,” in Proceedings of the 2002 IEEE Canadian Conference on Electrical and Computer Engineering, 2002, pp. 1325–1330. 8. S. I. Maniatis, E. G. Nikolouzou, and I. S. Venieris, “QoS Issues in the Converged 3G Wireless and Wired Networks,” IEEE Communications Magazine, pp. 44–53, Aug 2002. 9. M. Yuksel, K. K. Ramakrishnan, S. Kalyanarama, J. D. Houle, and R. sadhvani, “Value of Supporting Class-of-Service in IP Backbones,” in Fifteenth IEEE International Workshop on Quality of Service (IWQoS 2007), Jun 2007. 10. T. Nandagopal, T. E. Kim, P. Sinha, and V. Bharghavan, “Service Differentiation Through End-to-End Rate Control in Low Bandwidth Wireless Packet Networks,” in IEEE International Workshop on Mobile Multimedia Communications (MoMuC), 1999, pp. 211–220. 11. D.-H. Cho, J.-H. Song, M.-S. Kim, and K.-J. Han, “Performance Analysis of the IEEE 802.16 Wireless Metropolitan Area Network,” in First International Conference on Distributed Frameworks for Multimedia Applications, 2005. 12. O. Gusak, N. Oliver, and K. Sohraby, “Performance Evaluation of the 802.16 Medium Access Control Layer,” Lecture Notes in Computer Science – Proceedings of ISCIS, vol. 3280, pp. 228–237, 2004. 13. M. Ogawa, T. Sueoka, and T. Hattori, “Dynamic Queuing and Bandwidth Allocation for Controlling Delay Time for QoS in CDMA Packet System,” in 12th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, vol. 2, Sep–Oct 2001, pp. 38–42. 14. A. Veres, A. T. Campbell, M. Barry, and L.-H. Sun, “Supporting Service Differentiation in Wireless Packet Networks Using Distributed Control,” IEEE Journal on Selected Areas in Communications, vol. 19, no. 10, pp. 2081–2093, Oct 2001. 15. N. Christin and J. Liebeherr, “A QoS Architecture for Quantitative Service Differentiation,” IEEE Communications Magazine, pp. 38–45, Jun 2003.

234

K. Gakhar et al.

16. C. Cicconetti, A. Erta, L. Lenzini, and E. Mingozzi, “Performance Evaluation of the IEEE 802.16 MAC for QoS Support,” IEEE Transactions on Mobile Computing, vol. 6, no. 1, pp. 26–38, 2007. 17. R. Mukul, P. Singh, D. Jayaram, D. Das, N. Sreenivasulu, K. Vinay, and A. Ramamoorthy, “An Adaptive Bandwidth Request Mechanism for QoS Enhancement in WiMax Real Time Communication,” in IFIP International Conference on Wireless and Optical Communications Networks, 2006. 18. E. Altman and T. Jiménez, NS simulator for beginners, Universidad de Los Andes, Mrida, 2003. 19. K. Fall and K. Varadhan, The ns Manual, Aug 2006. 20. K. Gakhar, M. Achir, and A. Gravey, “How many traffic classes do we need in WiMax?” in IEEE Wireless Communications and Networking Conference, Mar 2007. 21. H. Wang, W. Li, and D. P. Agrawal, “Dynamic Admission Control and QoS for 802.16 Wireless MAN,” in Wireless Telecommunications Symposium, 2005, pp. 60–66. 22. C. W. Leong, W. Zhuang, Y. Cheng, and L. Wang, “Call Admission control for Integrated On/Off Voice and Best Effort Data Services in Mobile Cellular Communications,” IEEE Transactions on Communications, vol. 52, pp. 778–790, May 2004. 23. B. Rong, Y. Qian, and H.-H. Chen, “Adaptive Power Allocation and Call Admission Control in Multiservice WiMax Access Networks,” IEEE Wireless Communication, pp. 14–19, Feb 2007. 24. J. Hou, J. Yang, and S. Papavassiliou, “Integration of Pricing with Call Admission Control to meet QoS requirements in Cellular Networks,” IEEE Transactions on Parallel and Distributed Systems, vol. 13, pp. 898–910, Sept 2002. 25. D. Niyato and E. Hossain, “Call Admission Control for QoS Provisioning in 4G Wireless Networks: Issues and Approaches,” IEEE Network, vol. 19, no. 5, pp. 5–11, 2005. 26. S. Shen, C.-J. Chang, C. Y. Huang, and Q. Bi, “Intelligent Call Admission Control for Wideband CDMA Cellular Systems,” IEEE Transactions on Wireless Communications, vol. 3, pp. 1810–1821, Sept 2004. 27. J. Chen, W. Jiao, and H. Wang, “A Service Flow Management Strategy for IEEE 802.16 Broadband Wireless Access Systems in TDD mode,” in IEEE International Conference on Communications, vol. 5, pp. 3422–3426, May 2005. 28. Y. Qian, R. Q. Hu, and C. Rosenberg, “Integrated Connection Admission Control and Bandwidth on Demand Algorithm for a Broadband Satellite Network with Heterogeneous Traffic,” IEICE Transactions on Communications, vol. E-89B, no. 3, March 2006. 29. L. Huang and C. C. J. Kuo, “Dynamic Call Admisson Control with Joint Connection-level and Packet-level QoS Support in Wireless Multimedia Networks,” in IEEE Symposium on Real-Time and Embedded Technologies and Applications, May 2004. 30. R. L. Cruz, “A Calculus for Network Delay, Part 1: Network Elements in Isolation,” IEEE Transactions on Information Theory, vol. 37, no. 1, pp. 114–131, Jan 1991. 31. ——, “A Calculus for Network Delay, Part 2: Network Analysis,” IEEE Transactions on Information Theory, vol. 37, no. 1, pp. 132–141, Jan 1991. 32. G. Urvoy, Y. Dallery, and G. Hébuterne, “CAC procedure for Leaky Bucket-Constrained Sources,” Performance Evaluation, vol. 41, no. 2–3, pp. 117–132, Jul 2000. 33. L. Lenzini, L. Martorini, E. Mingozzi, and G. Stea, “A Novel Approach to Scalable CAC for Real-time Traffic in Sink-Tree Networks with Aggregate Scheduling,” in Valuetools’06, Oct. 34. K. Wongthavarawat and A. Ganz, “Packet Scheduling for QoS support in IEEE 802.16 Broadband Wireless Access Systems,” International Journal of Communication Systems, vol. 16, no. 1, pp. 81–96, 2003. 35. D. Kouis, P. Demestichas, V. Stavroulaki, G. Koundourakis, N. Koutsouris, L. Papadopoulou, and N. Mitrou, “System for enhanced network management jointly exploiting WLANs and other wireless network infrastructures,” in IEE Proceedings Communications, vol. 151, no. 5, pp. 514–520, Oct 2004. 36. Q. N. Vuong, L. Fiat, and N. Agoulmine, “An Architecture for UMTS-WiMax Interworking,” in The 1st International Workshop on Broadband Convergence Networks (BcN 2006), 2006.

10


235

37. C. Liu and C. Zhou, “An Improved Interworking Architecture for UMTS - WLAN Tight Coupling,” in IEEE WCNC, pp. 1690–1695, Mar 2005. 38. D. Kim and A. Ganz, “Architecture for 3G and 802.16 Wireless Networks Integration with QoS Support,” in Proceeding of 2nd Internatioanl Conference on Quality of Service in Heterogenous Wired/Wireless Networks (QShine’05), Aug 2005. 39. J. Nie, X. He, Z. Zhou, and C. Zhao, “Communication with Bandwidth Optimization in IEEE 802.16 and IEEE 802.11 Hybrid Networks,” in Proceedings of ISCIT 2005, pp. 26–29. 40. IEEE Standard for Local and metropolitan area networks–Part 16: Air Interface for Fixed Broadband Wireless Access Systems–Amendment 2: Medium Access Control Modifications and Additional Physical Layer Specifications for 2-11 GHz, IEEE Std. 802.16a, 2003. 41. H. Shetiya and V. Sharma, “Algorithms for Routing and Centralized Scheduling to Provide QoS in IEEE 802.16 Mesh Networks,” in ACM Workhop on Wireless Multimedia Networking and Performance Modeling, pp. 140–149, Oct 2005. 42. M. Cao, W. Ma, Q. Zhang, X. Wang, and W. Zhu, “Modelling and Performance Analysis of the Distributed Scheduler in IEEE 802.16 mesh Mode,” in MobiHoc 05, pp. 78–89, May 2005. 43. “STRIKE : HIPERMAN/HIPERLAN/2 Interworking Methods,” Deliverable D3.1.1, Information Society Technologies, Jun 2003. 44. Part3: Media Access Control (MAC) Bridges, ANSI/IEEE Std. 802.1D-1998, 1998. 45. K. Gakhar, A. Gravey, and A. Leroy, “IROISE: A New QoS Architecture for IEEE 802.16 and IEEE 802.11e Interworking,” in 2nd IEEE/Create-Net International Workshop on Deployment Models and First/Last Mile Networking Technologies for Braodband Community Networks BroadNETS 05, Oct 2005. 46. Part 11: Wireless LAN Medium Access Control (MAC) and Physical layer (PHY) specifications: Amendement 7: Medium Access Control (MAC) Quality of Service (QoS) Enhancements, IEEE Standard, Rev. D13.0, 2005.

Chapter 11

Energy-Efficient Multimedia Delivery in WMAN Using User Cooperation Diversity Ki-Dong Lee, Byung K. Yi and Victor C.M. Leung

Abstract Most previous work on cooperative cellular networks has considered homogeneous relaying architectures where all nodes act as both sources and relays or considered heterogeneous relaying architecture where relays are fixed. In this paper, we examine the power consumption performance of heterogeneous cooperative cellular networks with two classes of nodes: source nodes that do not act as relays, and relay nodes that are dedicated to relaying functions with little concern about power consumption. In this architecture, source nodes are able to reap the benefits of cooperative communication, such as improvements in the achievable data rate and reductions in the transmit power, while reducing the overall power consumption since they do not act as a relay. We consider geometry of a cell. Then we consider random locations of the source node and relay node, respectively, to analyze the average power consumption over the region of interest under the assumption that the source node and relay node are distributed uniformly over the cell region. Keywords Cooperative diversity · Energy efficiency · Multihop cellular network · Wireless MAN

11.1 Introduction It is very common that the complexity in wireless networks is increasing as the source nodes (SN’s; or subscriber stations, SS’s) become increasingly smaller in size and numerous in number. However, the large number of nodes will enable such networks to benefit from space diversity, or multiantenna diversity. The advantages of multiple-input multipleoutput (MIMO) systems have been intensively studied recently. In wireless metropolitan area networks (WMANs), a typical example of user node is a laptop computer where MIMO can be implemented. However, in the case of palm computers which has better portability because of the small size, even

K.-D. Lee (B) LG Electronics Mobile Research, San Diego, CA 92131, USA e-mail: [email protected]


237

238

K.-D. Lee et al.

though space diversity is known advantageous, it may not be feasible [1, 2] to take good advantage of the known MIMO benefits. To overcome this problem and to reap some of the benefits introduced by MIMO systems, the concept of cooperative communications has been introduced and several cooperative diversity techniques have been studied [1–4]. The benefits of cooperative communications result from cooperative diversity [3], which can be achieved by relaying, i.e., the diversity at the receiver side: one from the source node (SS) and the other from the relay node (or relay station, RS). Although both cooperative communication networks and multihop networks employ data relaying by network nodes, the former are distinguished from the latter through the use of cooperative diversity realized from multiple virtual links established in the process of relaying. The Gaussian relay channel was introduced by van der Meulen [5] and intensively analyzed by Cover and El Gamal [6], where the achievable rates for three coding schemes were evaluated. This work have been extended to the case of multiple relays by Schein and Gallager [4] who considered two parallel relays between source and destination, and Gupta and Kumar [7] who considered multi-level relays. Cooperative diversity in cellular systems was introduced by Sendonaris et al. in [1] and [2]. In the papers, the authors implemented a two-user cooperative CDMA cellular system, where both users (or SS’s) are active at the same time and use orthogonal codes to avoid multiple access interference. Also, for ad hoc networks, Laneman et al. proposed various cooperative diversity protocols and analyzed their performance in terms of outage probability [3]. It is commonly considered that a SS needs to be small in size (e.g., it is preferred to have as small a handset as possible). However, it is not necessary that RS’s should be small in size. As mentioned above, this is because we may have different kinds of RS’s such as dedicated devices that have little concern about power consumption. Such RS’s can be installed in vehicles, as mobile relays, or on top of buildings, as fixed relays. Assuming that RS’s are available in a cellular network, the SS’s can potentially save a lot of power because they can take advantage of cooperative diversity offered by RS’s to reduce their own transmit power, and they need not act as relays. This heterogeneous cooperative cellular architecture is not only useful to improve the average achievable rate and reduce power consumption at SS’s, but the reduction of SS power emission may also offer benefits in allaying health concerns with regard to electromagnetic radiations, and reducing probability of interception of signals by unintended receivers. Figure 11.1 presents the proposed cooperative cellular network architecture with the two types of nodes mentioned above. Based on this network architecture, we consider a general case where the cooperation chance depends on the situation that there is a candidate RS around the SS. Thus, we take probabilistic approach to evaluate the power consumption in this situation. The objective of this paper is to evaluate the performance of a cooperative diversity technique in a Wireless MAN environment [8] with mobile cooperative relays, focusing on the average achievable rate and power consumption. When a SS is connected to the base station (BS) through one RS, the BS may receive the signals from both the cooperating RS and the SS, resulting in a performance gain via cooperative

11

Energy-Efficient Multimedia Delivery

239 RS

BS (D)

SS (S)

cell

Fig. 11.1 Illustration for the cooperative network architecture. Source node (SS) has a typical concern about power consumption whereas relay node (RS) has very little concerns. We consider a series of time-frames each of which consists of the 1st timeslot, where SS transmits and RS and BS receive separately, and the 2nd timeslot, where RS transmits (relays) and BS receives: then, BS can combine those two signals to get diversity

diversity. In most previous studies, the characteristics of nodes are assumed identical, or homogeneous, and it is considered that nodes act as both sources and relays; or when the relays and source nodes are heterogeneous, relays are usually assumed to be immobile. However, in practice, different kinds of nodes with diverse mobility characteristics, battery lifetime, and so on, may exist in the network. For example, a portable handset has a relatively short battery lifetime whereas a radio transceiver onboard a vehicle has a much longer battery lifetime and vehicles are moving around the service coverage area. We investigate how much power can be saved through the use of cooperative diversity when both SS’s and RS’s are mobile and, then, they are randomly located over the cell region. Also, we study how the population density of RS’s affect the power saving performance.

11.2 The Relaying Mechanism 11.2.1 The Model We consider a two-hop cellular network, such as point-to-multipoint (PMP) mode [8], employing orthogonal frequency division multiple access (OFDMA) over the air interface. OFDMA is one of the most promising transmission technologies that are gaining popularity in wireless networks [9]. In each cell, we consider uplink transmissions from the SS’s to the BS via RS’s acting as relays if available. In cooperative networks, it is usually assumed that SS’s and RS’s have full channel state information (CSI) so that SS’s and RS’s can be synchronized [10]. We consider the decode-and-forward scheme. Figure 11.2 depicts an example of frame structure through which the BS can reap the benefit of the cooperative diversity. As already used in IEEE 802.16j [13], a two zone frame structure is considered. During the first zone called access zone the SS transmits and the RS and BS receives. During the second zone called relay zone the RS relays what it has received during the first zone. Then the BS can combine both signals.

240

K.-D. Lee et al. R

S 1st

2nd Frame

Time

Fig. 11.2 An example of frame structure for a cooperative relaying mechanism. The SS transmits on the 1st timeslot (during access zone) whereas the RS relays the overheard signal from the SS, to the BS on the 2nd timeslot (during relay zone). The BS may combine the two signals

11.2.2 Description of the Relaying Mechanism Figures 11.1 and 11.4 illustrate the relaying mechanism for the aforementioned relaying architecture with two heterogeneous classes of nodes. Consider a user (SS) as shown in Fig. 11.4 where the transmission range of this SS is shown. If there exists at least one RS in the transmission range, the SS may ask the RS for relaying. We consider a series of time frames each of which consists of two timeslots: in the first timeslot, SS transmits its data whereas RS and BS receive it; in the second timeslot, RS relays SS’s data to BS. In this case, the BS needs to receive signals from both the SS and the RS to achieve cooperative diversity. The procedure of the relaying mechanism is detailed in Fig. 11.3. If a SS needs a new action, such as handover or new call setup requests, then the SS checks both the pilot signal(s) coming from the associated BS(s) and the pilot signals including channel gain information coming from RS’s. BS’s periodically broadcast the pilot signal so that SS’s and RS’s can estimate the channel gains. After estimating the channel gain, RS’s broadcast the pilot signal with the channel gain information so that SS’s can estimate the channel gain. Each SS takes advantage of the information to make decisions required in the relaying architecture. If the distance between a SS and its best BS is small enough, then the SS may be connected to the BS directly without any relaying. The cell region excluding such a region is called candidate region in this paper. The SS selects the best BS and best RS (selection is completed by handshaking). If RS’s are unavailable (i.e., relay = 0), the SS sends capacity request (CR) to the BS directly. Otherwise, the SS checks if one RS is available (i.e., relay = 1), the SS sends CR to the BS. Once the BS received the CR, the BS checks if the required capacity is available according to the admission control policy it uses. If there is no capacity available for the SS that has originated the CR, then the BS rejects the request. Otherwise, the SS is allocated a certain amount of capacity and starts transmitting information in accordance with the relay mode. Before transmitting information, the SS determines the transmission power level.

11


241

SS (user)

RS

BS pilot (periodic) broadcast

estimate channel gain b/w BS and user

estimate channel gain b/w BS and RS

pilot + gain info (periodic) broadcast

estimate channel gain b/w RS and user select best BS; select one RS in transmission region

transmission via this rs better than direct transmission

N

relay = 0 relay = 1 send “relay”

relay > 0?

N capacity request

Cooperation handshaking ACK capacity request (rejected) end

relay > 0?

(accepted)

N

capacity available?

capacity allocation

N

power mode=NORMAL traffic power mode=SAVING traffic broadcast

traffic multiple access

Fig. 11.3 The flow of relaying mechanism

242

K.-D. Lee et al.

11.2.3 Transmission Power Modes of SS The transmission power of an SS (or mobile SS, MSS) needs to be continuously adjusted according to channel conditions. However, the transmission power of a SS in the relaying mechanism is roughly divided into two modes: NORMAL mode and SAVING mode. In the event of relaying, the SS needs not transmit information at the normal power level (NORMAL mode). This is because cooperative diversity can be achieved. In this case, the SS may reduce its power level as far as its signal may reach both RS and BS (SAVING mode) given that the transmit power of RS, say p2 , is less than or equal to a given threshold.

11.3 Evaluation of the Average Achievable Rate and Average Power Consumption We have the following notation in the evaluation.

r r r r r r r r

R: radius of a circular cell L: radius of a transport range of an SS (relaying case) α: path-loss exponent p: transmit power of an SS (mW) σ 2 : thermal noise power (mW) BER: desired bit-error rate δ: population density of RS (or relaying SS) (RSs/m2 ) f p : fraction of the average power consumption through the use of relay relative to the average power consumption without using it

11.3.1 The Average Power Consumption for Transmission Consider a SS located at (a, θ ), 0 ≤ θ < 2π (see Fig. 11.4). Suppose that the relaying mechanism is not used (single relaying or no relaying), then the average power consumed to achieve the transmission rate of x per subcarrier is given by P0 (a) =

2x/W − 1 κ · G(a)

and, by unconditioning with respect to a, the average power per subcarrier of an arbitrary SS without using the relaying mechanism is given by

R

P0 = 0

P0 (a) · d

a2 R2

.

Following a similar procedure as in the previous section for rate analysis, we consider an RA whose relative location with respect to the location of the SS is

11


Fig. 11.4 The evaluation model. O(0, 0), U (a, 0), ∠O M1 U = ∠O M2 U = π/2, L ≡ M1 U . ∠M1 U O = cos−1 La . θ1 ≡ ∠ M 3 U M 1 , θ2 ≡ ∠M3 U M2

243 transmission range (related to relaying) transmission range (nominal)

R

M1

d

R0

O

r

a

L

U

(r, θ ). The required transmission power is given by P(r, θ ; a) = inf p1 : 0.5 · min{log2 (1 + κ p1 G(r )), log2 (1 + κ p1 G(r ) + κ p2 G(d))} = x, p2 ≤ p M √ where d ≡ d(r, θ ; a) = a 2 + r 2 + 2ar cos θ and p M is a given threshold. From this, the average transmission power of a SS in the case of using the relaying mechanism (parallel dual agent relaying) is given by

L

P1 (a) =

θ2

θ1

0

P(r, θ : a) ·

r dθ dr Ae (a)

When the relaying mechanism is used, the average transmission power of a SS located at (a, θ ), 0 ≤ θ < 2π is given by P(a) = P0 (a) · φ0 + P1 (a) · (1 − φ0 ). By unconditioning with respect to a over the candidate region, we can obtain the average power per subcarrier of a SS residing in the candidate region as

R

P=

P(a) · d

R0

a2 R 2 − R02

and that of a SS in the inner co-center circle is given by P1 = 0

R0

P0 (a) · d

a2 R02

244

K.-D. Lee et al.

Finally, the average power per subcarrier of an arbitrary SS is given by

R2 R2 P = P · 1 − 02 + P1 · 02 . R R The (relative) improvement in the average power consumption per subcarrier is obtained by fp =

P0 P

,

where f p is greater than unity if the relaying mechanism saves the power consumption.

11.3.2 Experimental Results We consider the OFDMA-based PMP mode system with infrastructure [8], where there are 128 subcarriers over the 3.2-MHz band. Each transmitter is synchronized with respect to the receiver’s clock reference to make the tones orthogonal. The transmission power of a SS per subcarrier is p = 50 mW, the thermal noise power is σ 2 = 10−11 W, and the desired BER is 10−2 . The path loss exponent α is three. Figure 11.5 presents the probability that a randomly chosen SS has no relays around itself. Here, we can observe how small the probability is in normal urban environments: 0.0015 RAs/m 2 in San Diego, 0.00037 RAs/m 2 in Montreal. Under the condition that the power consumption amounts are the same, we compare two different cases for achievable rate. For the purpose of relative comparison, we define f c as the fraction of the achievable rate by the relay mechanism, relative to the achievable rate without relaying. In Fig. 11.6, the average achievable rates 1

pr{no RS available}, φ0

R = 500m L = 100m

Fig. 11.5 The probability that a SS cannot find any relay around it vs. the density of relay population (RAs/m 2 )

0.1

0.01

1E–3

1E–4

1E–5 1E–6

1E–5

1E–4

1E–3

population density (RAs/m 2)

0.01


245 2.5

200 180

2.0

160 1.5

140 120 100

1.0

w/o relay w/ relay increase in rate (%)

increase in rate, fc

Fig. 11.6 The average achievable rate of a user per subcarrier vs. p (3.2-MHz band with 128 subcarriers, σ 2 = 10−11 W, BER = 10−2 , R = 500 m, L = 50 m, α = 3, δ = 0 : 01)

average rate (Kbps)

11

0.5

80 60 1E–6

1E–5

1E–4

1E–3

0.0 0.01

bit error rate, BER

per subcarrier versus transmission power are shown. It is observed that the average achievable rates of both mechanisms are increasing with a decreasing speed of increase, i.e., concave increasing, with respect to power increase. For small transmission power levels, even though the amount of rate increase is small, the relative increase f c is very large. As the transmission power increases, the relative increase f c decreases, slowly approaching 200%. Figure 11.7 shows the power consumptions versus the desired BER. A steady improvement in power consumption for the test values of BER is observed. A remarkable reduction in power is observed. The relative improvement in power f p is approximately 518 (51800%). Figure 11.8 shows the average achievable rates per subcarrier for both mechanisms versus the desired bit-error rate (BER). Theoretically, the average achievable rate increases if the desired BER increases and this phenomenon is observed in the figure. In addition, it is observed that the relative increase f c is very large when the desired BER is small. As the desired BER increases, the relative increase f c

500

average power (mW)

Fig. 11.7 The average power consumption of a user per subcarrier vs. BER (3.2-MHz band with 128 subcarriers, σ 2 = 10−11 W, BER = 10−2 , R = 500 m, L = 50 m, α = 3, δ = 0 : 01)

400

10

w/o relay w/ relay saving in power, fp

300

200 1 100 0.1 1E–6

1E–5

1E–4 bit-error rate, BER

1E–3

0 0.01

saving in power consumption, fp

100

246 300

2.5

average rate (Kbps)

250 2.0

w/o relay w/ relay increase in rate (%)

200

1.5

150 1.0 100


Fig. 11.8 The average achievable rate of a user per subcarrier vs. BER (3.2-MHz band with 128 subcarriers, σ 2 = 10−11 W, BER = 10−2 , R = 500 m, L = 50 m, α = 3, δ = 0 : 01)

K.-D. Lee et al.

0.5

50 0

0.0 1E–3

0.01

0.1

transmission power, p (mW)

decreases with a decreasing rate, i.e., convex decreasing, slowly approaching 217%. This demonstrates that the relaying mechanism is much more useful for such traffic that requires a low BER. Figure 11.9 shows the power consumptions versus required rates. The required power is increasing with respect to the required transmission rate. However, the relative improvement in power by using the relaying mechanism f p is steady around 518. This demonstrates that the relaying mechanism can reduce a lot of amount in power in the case of transmitting traffic at a high rate. In Fig. 11.10, the effect of the density of RS’s in the cell site on the average achievable rates per subcarrier is shown. We test the effect with RS densities ranging from 0.0001 to 0.01 (RAs/m 2 ). The density of 0.01 is chosen for representing the situation of population density in urban area (For example, the population density in New York City is approximately 0.01/m 2 ). Under the assumption that everyone has an auto that can act as a relaying agent, this value represents densely populated situations. On the other hand, the value 0.0001 (RAs/m 2 ) is used for representing the rural population situation. In the figure, it is observed that the average rate of

500 w/o relay w/ relay saving in power, fp

1E8

400

300 10000 200 1 100

1E–4 1

10

100

required rate (Kbps)

0 1000

saving in power consumption fp

Fig. 11.9 The average power consumption of a user per subcarrier vs. required rate (3.2-MHz band with 128 subcarriers, σ 2 = 10−11 W, BER = 10−2 , R = 500 m, L = 50 m, α = 3, δ = 0 : 01)

average power (mW)

1E12


247

2.0

200 w/o relay w/ relay increase in rate (%)

1.5

150 1.0

fraction = 1.0 (no increase)


Fig. 11.10 The average achievable rate of a user per subcarrier vs. δ (3.2-MHz band with 128 subcarriers, σ 2 = 10−11 W, BER = 10−2 , R = 500 m, L = 50 m, α = 3)

average rate (Kbps)

11

0.5

100

0.0 1E–4

1E–3

0.01

0.1

density of relaying agent, δ (RAs /m 2)

the relaying mechanism steeply increases as the density of RS’s increases and satiates around 0.002. Also, for population densities 0.00017/m 2 , 0.00032/m 2 , and 0.01000/m 2 , the relative increase f c is 115%, 135%, and 217%, respectively. This demonstrates that the relaying mechanism is much more useful for most cases of population density. Even though the subscription ratio is considered, we believe that the performance has practical meaning for financial and/or shopping districts, where the density is much higher than any other area. Figure 11.11 shows the power consumptions versus the density of RS’s in cell. It is observed that the relative improvement in power f p is rapidly increasing until δ approaches 0.004. According to the results, the relative improvement in power in a site of population density 0.00037/m 2 , such as in Montreal, is 140% whereas that of density 0.01/m 2 , such as in New York City, is 500. These results demonstrate that the relaying mechanism can provide a remarkable power saving effect in urban areas.

1000

100

average power (mW)

Fig. 11.11 The average power consumption of a user per subcarrier vs. δ (3.2-MHz band with 128 subcarriers, σ 2 = 10−11 W, BER = 10−2 , R = 500 m, L = 50 m, α = 3)

100 60 w/o relay w/ relay saving in power, fp

40

10

20

0 1E–4

1 1E–3

0.01

density of relaying agents, δ (RAs/m2)

0.1

saving in power consumption fp

80

248

K.-D. Lee et al.

11.4 Conclusions In this paper, we have examined the average achievable rate and the average power consumption in a cooperative diversity based wireless metropolitan area networks with two classes of nodes. Unlike previously proposed architecture employing homogeneous nodes that function as both sources and relays, we propose a functional differentiation between source and relay nodes so that source nodes may save power while relay nodes have little concern about power consumption. In addition to the benefits of cooperative communication, such as increasing the achievable rate and transmit power reductions, SS’s can achieve substantially greater power savings by not performing relay functions. The heterogeneous cooperative cellular network architecture proposed in this paper is both feasible and practical. According to analysis results, the proposed architecture achieves a good amount of increase in the average rate and a huge amount of reduction in the average power consumption, compared to the situation where no RS’s are available between each SS and the BS.

References 1. A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diversity Part I: System description,” IEEE Trans. Commun., vol. 51, no. 11, pp. 1927–1938, Nov. 2003. 2. A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diversity Part II: Implementation aspects and performance analysis,” IEEE Trans. Commun., vol. 51, no. 11, pp. 1939–1948, Nov. 2003. 3. J.N. Laneman, D. Tse, and G.W. Wornell, “Cooperative diversity in wireless networks: Efficient protocols and outage behaviour,” IEEE Trans. Info. Theory, vol. 50, no. 12, pp. 3062– 3080, Dec. 2004. 4. B. Schein, and R. Gallager, “The Gaussian parallel relay network,” in IEEE ISIT, June 2000, p. 22. 5. E.C. van der Meulen, “Three-terminal communication channels,” Adv. Appl. Prob., vol. 3, pp. 120–154, 1971. 6. T. Cover, and A. El Gamal, “Capacity theorem for relay channel,” IEEE Trans. Info. Theory, vol. 25, no. 5, pp. 572–584, Sep. 1979. 7. P. Gupta, and P.R. Kumar, “Towards an information theory of large networks: an achievable rate region,” IEEE Trans. Info. Theory, vol. 49, no. 8, pp. 1877–1894, Aug. 2003. 8. IEEE Standard for Local and Metropolitan Area Networks – Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems, oct, 2004. 9. Z. Han, Z. Ji, and K.J.R. Liu, “Fair multiuser channel allocation for OFDMA networks using Nash bargaining solutions and coalitions,” IEEE Trans. Commun., vol. 53, no. 8, pp. 1366– 1376, Aug. 2005. 10. A. Goldsmith, “Rate limits and cross-layer design in cooperative communications,” in WICAT Workshop on Cooperative Communications, Polytechnic University, Brooklyn, New York, Oct. 2005. 11. K.-D. Lee, Byung K. Yi, and V.C.M. Leung, “Power consumption evaluation in a wireless MAN using cooperative diversity when both sources and relays are randomly located,” in Proc. ACM QShine ’07, Vancouver, BC, Aug. 2007. 12. K.-D. Lee, and V.C.M. Leung, “Evaluations of achievable rate and power consumption in cooperative cellular networks with two classes of nodes,” IEEE Tran. Veh. Technol., to appear, 2008.

11


249

13. IEEE P802.16j/D1 “Multihop Relay System,” Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems, Aug. 2007. 14. IEEE C802.16m-07/080r3 Draft IEEE 802.16m Evaluation Methodology Document, Aug. 2007. 15. IEEE 802.16m-07 = 002r2, “Draft TGm Requirements Document,” IEEE 802.16m Task Group Draft Document, June 11, 2007.

Chapter 12

Game Theory Modeling of Social Psychology Principle for Reliable Multicast Services in WiMax Networks Markos P. Anastasopoulos, Athanasios V. Vasilakos and Panayotis G. Cottis

Abstract A major challenge in operation of WiMax network is to provide largescale reliable multicast and broadcast services. The main cause limiting the scalability of such networks is Feedback Implosion, a problem arising when a large number of users transmit their feedback messages through the network, occupying a significant portion of the system resources. Inspired from social psychology, and more specifically from the phenomenon of bystander effect, a novel scheme for the provisioning of large-scale reliable multicast services is proposed. The problem is modeled using game theory. Through simulations of the proposed scheme carried out to evaluate its performance, it is found that the novel approach suppresses feedback messages very effectively, while at the same time, it does not degrade the data transfer timeliness. Keywords Wi-Max Networks · Reliable multicast · Feedback suppression · Bystander effect · Game theory

12.1 Introduction As the demand for high-speed ubiquitous Internet access is increasing at a rapid pace, Broadband Wireless Access (BWA) networks are gaining increased popularity as an alternative to DSL last-mile technology with a constantly growing market potential [1]. WiMax [2], the industry consortium associated with the IEEE 802.16 family of standards, poses as an interesting addition to current broadband options, such as DSL, cable and Wi-Fi (IEEE 802.11), offering both prospects of rapid broadband access provisioning in areas with under-developed infrastructure and a technology capable of competing for urban market share. Already, deployment of large-scale WiMax networks capable of providing broadband connectivity

M.P. Anastasopoulos (B) Wireless & Satellite Communications Group, School of Electrical & Computer Engineering National Technical University of Athens, Greece


251

252

M.P. Anastasopoulos et al.

to millions of users in different parts of the world has initiated, making this technology a fast-rising wireless access solution with lucrative commercial prospects. BWA networks, as described in the IEEE 802.16 standard [3], were initially designed to operate on a Line of Sight (LOS) cellular basis providing backhaul and high speed communication services to fixed subscribers in the 11-66GHz spectral range. Driven by the need to accommodate increasing numbers of users, the non-LOS operation was mandated in the 802.16a/d which defined aspects of the utilization of the 2-11GHz spectrum [4]. The IEEE 802.16e standard for Mobile WiMax initiates a new era in BWA systems, as focus is shifted from fixed subscribers to subscribers moving in up to vehicular speeds [5]. Mobile WiMax, aiming at providing bit rates up to 15 Mbps at 5MHz utilization and adopting state-of-theart technologies such as HARQ and MIMO, emerges as an exciting new technology with commercial prospects as a competitor of CDMA mobile networks (3G, HSDPA). Although WiMax networks offer significant advantages, they are also hampered by restrictions imposed by the specific phenomena of the wireless channels. Radio channels introduce high error rates, correlated losses and dynamically varying propagation conditions, all of which affect the performance of the wireless applications. Radio resource management and system-specific protocol design are two critical challenges to be met in the attempt to both satisfy the Quality of Service requirements for network agnostic applications and efficiently utilize the available network bandwidth. In second generation BWA networks, it is assumed that propagation, in the non LOS cell areas is dominated by shadowing effects [6], whereas the LOS communication between Base Stations (BSs) and Subscribers (S) at frequencies above 10GHz is primarily degraded by rain fading [6]. Mobile WiMax is expected to be further affected by phenomena linked to user mobility such as multipath and Doppler shift. The current trend in modern wireless networks is the integration of communication applications over a common IP infrastructure. The investigation of performance issues concerning the TCP/IP protocol stack over wireless links that suffer from propagation losses is a topic of great significance, in order to provide seamless connectivity to the access network. In the case of large-scale multicast applications over WiMax networks the main challenge arising, is QoS provisioning, in terms of reliability, and efficient uplink spectrum utilization, taking into account accurately the impact of propagation phenomena.

12.2 Providing Reliable Multicast Services To provide reliable large-scale multicast services, the main problem to be solved is feedback implosion [7], which is arisen whenever a large number of subscriber terminals transmit feedback messages through the uplink channel. Those messages increase linearly with the number of users and may lead to congestion of the network.

12

Game Theory Modeling

253

Feedback suppression is a well studied problem and many approaches exist in the literature. In general, solutions to the feedback implosion problem are classified as structure-based or timer-based [7]. In the first classification, structure-based approaches rely on a designated site to process and filter feedback information. They organize multicast group members into some structure in order to filter the amount of feedbacks generated by the group. Timer-based solutions rely on probabilistic feedback suppression to avoid implosion at the source. Receivers delay their retransmission requests for a random interval, uniformly, exponentially or beta distributed between the current time and the one-way trip time to the source. It is evident that the problem of feedback implosion could be confronted, if a limited number of users, on behalf of all multicast receivers, sent negative acknowledgments (NACKs). Hence, the question that is arisen, is how the feedback suppression problem should be modeled in such a way so that, as the number of users increase, their incentive for sending feedback messages, is reduced? An answer to this question may be found in social psychology and more specifically in the phenomenons of bystander intervention and bystander effect [8]. In bystander intervention, psychologists found out that, solitary individuals will typically intervene if another person needs help. However, in case where more people are present, help is less likely to be given. The latter, is known as bystander effect or bystander apathy. In this phenomenon persons are less possible to intervene in an emergency situation when others are present than when they are alone. Both bystander intervention and bystander effect may be the solution for the problem of providing large scale reliable multicast services. In particular, in case where a limited number of users participate in the multicast services, there is no need to their suppress feedback messages. Their impact in the performance of the network is low, due to the fact that a small number of NACKs do not demand high network resources, bandwidth and computational cost. On the contrary, if the number of users is high, rather than being conscientious, it is preferable for a user to act with unscrupulousness, avoiding feedback sending. Due to the fact that, even one feedback message can help all multicast users to recover from losses, the contribution of apathetic users to the performance of network is much more important in comparison with scrupulous who send NACKs instantly. In this paper, game theory is employed to formulate the feedback suppression problem. All users would like that the lost or corrupted packets to be retransmitted, but neither of them is willing to send a feedback message because this action requires energy consumption. This factor becomes more important, especially in case of mobile terminals, where energy issues are critical for the survivability of the system. In the game under consideration, the optimal strategy is investigated. In the current approach, the problem will be modeled using classical game theory, which predicts that users will assign their strategy according to a Nash Equilibrium (NE), i.e., in a way such no player has an incentive to change their routing strategy unilaterally. Even though NE are interesting from a practical point of view as they represent stable and fair allocations, however, the lack of a dynamic theory, imposes classical game theory inappropriate for modeling real wireless networks that face time dependent fading conditions and users mobility. To overcome those difficulties

254


the application of Evolutionary Game Theory [9], in the same problem is a subject of future work. The rest of the chapter is organized as follows. In Section III a brief description of the preliminaries of game theory is presented, while in Section IV the problem of feedback suppression is being modelleed employing game theory. Then description of the channel model employed along with the numerical results is provided in Section 12.6.7. Finally, important conclusions are drawn in Section 12.7.

12.3 Preliminaries of Game Theory Game theory is a branch of applied mathematics, which deals with multiperson decision making situations. For example, when the only two electronic equipment sellers choose prices for their products, aware that their sales are determined jointly, they both participate in a game. Many applications of game theory are related to economics, but it has been applied to numerous fields ranging from law enforcement and voting decisions in European Union to biology, psychology and recently engineering. In this paper, the analysis is restricted in games where players act individually without exchanging information among them and contributes to the public good. These games are called non-cooperative.

12.3.1 Definitions The basic elements of a game are players, strategies and payoffs. Players are individuals who decide their movements. Based on the information that has arrived at each moment they pick up actions in order to maximize their profit. An action, or move by player i, denoted ai , is a choice he may make. The rule that tells him which action to choose at each instant of the game is called strategy si , while the set of all available strategies is called strategy profile Si = {si }. Hence, the Cartesian product, S = ×i Si , sometimes called strategy space of the game [9], is the set of strategy profiles of all players. For any strategy profile s ∈ S, πi (s) denotes the associated payoff to player i. The term payoff is referred to the expected utility that a player receives as a function of the strategies chosen by himself and the other players. In economics the payoffs are usually profits or consumers utility, while in biology payoffs usually represent the expected number of surviving offspring. Often, in the literature a game is written in normal form that may be summarized as a triplet G = (I, S, π ), where I is its players set, S its strategy space, and π its payoff function.

12.3.2 Equilibrium and Dominant Strategies A concept in game theory is that of equilibrium. Equilibrium s ∗ = ∗very important ∗ s1 , . . . , sn is a strategy combination consisting of a best (optimum) strategy for

12


255

each of the N players in the game [10]. The equilibrium strategies are the strategies players pick in trying to maximize their individual payoff. Discussing equilibrium concepts it is useful to define a metric that describes the strategies that all the other players use. If s = (s1 , . . . , s N ) is the strategy set of all players, then notation s−i = (s1 , . . . si−1 , si+1 , . . . , s N ) denotes the combination of strategies for every player except i. Vector s−i is of utmost importance because helps player i to choose the best available strategy. Player i ’s best response or best reply to the strategies s−i , chosen by the other players, is the strategy si∗ that yields him the greatest payoff. The strategy si∗ is a dominant strategy if it is a player’s strictly best response to any strategies the other players might pick, in the sense that whatever strategies they pick, his payoff is highest with si∗ . It deserves to be noted that, all his inferior strategies are called dominated strategies. Finally, dominant strategy equilibrium is a strategy combination consisting of each player’s dominant strategy. Unfortunately, few games have dominant strategy equilibrium.Nevertheless, dominance can be useful even when it does not resolve things quite so neatly. In those cases, the idea of iterated dominance equilibrium is used. Let us first define the concept of weakly dominance. Strategy si is weakly dominated if there exists some other strategy si for player i for which is possibly better and never worse, yielding a higher payoff in some strategy profile and never yielding a lower payoff. An iterated dominance equilibrium is a strategy combination found by deleting a weakly dominated strategy from the strategy set of one of the players, recalculating to find which remaining strategies are weakly dominated, deleting one of them, and continuing the process until one strategy remains for each player [10]. However, the majority of games lack even iterated dominance equilibrium. In those cases, modelers use Nash Equilibrium (NE). NE is one of the cornerstones of economic theory. In essence NE requires of a strategy profile s that not only should each component strategy si be optimal under some belief on behalf of the ith player about the others strategy, it should be optimal under the belief that s itself will be played. In terms of best replies, a strategy profile s is a Nash Equilibrium if it is best reply to itself [9]. When a game is in NE no player has incentive to change his strategy given that the other players do not deviate. Concluding, every dominant strategy is a NE, but not every NE is dominant strategy equilibrium. If a strategy is dominant, it is the best response to any strategies the other players pick, including their equilibrium strategies. If a strategy is part of a NE, it needs only to response to the other players equilibrium strategies.

12.3.3 Pure and Mixed Strategies So far, the analysis has been limited in the case where the action set of the players is finite. It is often useful and realistic to expand the strategy space to include random strategies. Those random strategies are called mixed, while finite are called pure strategies. In other words, a mixed strategy for player i is a probability distribution over his set of pure strategies.

256


12.3.4 Existence of Equilibrium The first basic feature of a game that favours existence of Nash Equilibrium is continuity of the payoffs in the strategies. A payoff is continuous if a small change in a player’s strategy causes a small or zero change in the payoffs. The second feature promoting existence is a closed and bounded strategy.

12.4 Feedback Suppression Game The bystander effect in this Section, is being modelled by applying game theory. As a typical example imagine an event where several people are eye-witness in a crime. Each one would like someone to call the police and stop the illegal action, because having it stopped add x units to his payoff. Unfortunately, none of them wants to make the call himself because the effort subtracts y units (x > y). Strongly inspired from this, a new approach for the solution of feedback implosion is presented.

12.4.1 Two Players Game The problem of feedback suppression is modeled using game theory. The problem belongs to the general category of contribution games [10]. This term is used in the literature to describe games in which each player has a choice of taking some action that contributes to the public good, but would prefer another player to take the rap. In feedback suppression game, each player that has lost a packet would like to send a feedback message asking for packet retransmissions, because replacing the lost or corrupted packets helps him to satisfy the QoS constraints for data transmission. However, nobody wants to send the acknowledgment because this is an energy consuming action. Table 12.1 shows the feedback suppression game between player 1 and player 2 that have lost the same packet. If player 1 can be assured that player 2 will send a feedback message, then there is no reason for him to send an acknowledgment. In this circumstance, the payoff for player 1 is a, while for player 2 is a −d E, where dE denotes the energy squandering due to feedback transmission.

Table 12.1 The feedback suppression game Player 1 Don’t send FBM(p) Player 1

Send FBM (1-p)

Don’t send FBM(p)

0, 0

a, a − d E

Send FBM (1-p)

a − d E, a

a − d E, a − d E

12


257

The above game has two asymmetric pure strategy, (a − d E, a) (a, a − d E), and a symmetric mixed-strategy equilibrium.

12.4.2 N-Players Game Then, we are interested in solving the N-player version of the above game. Let p, denote the probability for player i not to send a feedback message. In the problem under consideration, if nobody sends a feedback message the payoff for player i is 0. In case he himself sends a message the payoff is a − d E, and, if at least one of the other N-1 players send, the payoff is a. In accordance with the two players, this game also has an asymmetric pure-strategy and a symmetric mixed-strategy equilibrium. If all players use the same probability p, the probability that all players except player i not to send feedbacks is p N −1 . Therefore, the probability at least one to send a feedback message is 1 − p N −1 . Thus, equating players i pure strategy payoffs using the payoff-equating method of equilibrium calculation yields π playeri (send

f eedback)

= π playeri (not

send f eedback)

(12.1)

or equivalently (a − d E) p N −1 + (a − d E) 1 − p N −1 = p N −1 · 0 + 1 − p N −1 · a

(12.2)

Thus yields,

∗

p =

dE a

1/N −1 (12.3)

It is obvious that the probability, PF B , a user to send a feedback message is given by PF B = 1 −

dE a

N 1−1 (12.4)

From Eq. (12.4) it is obvious that if the number of users is low the probability to send a feedback is high. In this occasion, multicast receivers react with conscientiousness according to the phenomenon of bystander intervention. However, increasing N, PF B is reduced, and the users are gradually becoming unscrupulous, hence, feedback suppression efficiency is increased. The above concept is clearly illustrated in Figure 12.1.

258


Probability for feedback transmission

PFB 1–dE/a

0

2

N(Number of Users) Unscrupulousness (Symptom of bystander apathy)

Scrupulousness (Symptom of bystander intervention)

Fig. 12.1 Impact of users’ population in feedback message transmission probability

12.5 Syndrome of Genovese: The Need for Backup Mechanisms Even though bystander effect may be effectively used for feedback suppression, sometimes turns out to be insufficient. Its main weakness is that when the number of multicast receivers is very high, the nodes are becoming so unscrupulous where it is possible no-one to send a feedback. This event took place in 1964 where Kitty Genovese was stabbed to death by a mentally ill serial rapist and murderer. The murder took place over a period of about thirty minutes, during which dozens of alleged “witnesses” failed to help the victim. For this reason, the name Genovese syndrome or Genovese effect was used to describe the phenomenon at the time. The death of Deletha Word in 1995 after witnesses failed to thwart her attackers, as well as the James Bulger murder case, may have been other well-publicized cases of the effect. The probability the syndrome of Genovese, PGen may be easily calculated by the following equation PGen = (1 − PF B ) N =

dE a

NN−1 (12.5)

From the aforementioned it is deduced that bystander effect from its own accord cannot ensure the reliability in data transfer. For this reason, backup mechanisms are essential to be used. A simple implementation is that of a timer. Each user maintains a timer and as long as the timer expires, in case he hasn’t received the lost packets, sends a feedback message asking for retransmissions. In this simple case, the expected number of feedback messages is given by, E {Feedback Messages} = (PF B + PGen ) · N

12


259

or equivalently E {Feedback Messages} = 1 −

dE a

N 1−1

+

dE a

NN−1

·N

(12.6)

More sophisticated implementations for backup mechanisms may be found in [7]. In these approaches the backup timers that are used are separated into two parts; the first is a wait period and the second is a random interval. The waiting period allows nodes to take their decisions about whether to send feedbacks or not. Besides this, gives sufficient time to the retransmitted packets to reach all multicast receivers and to suppress their feedback. It is obvious that the waiting period must be slightly greater than the round trip time (RTT). The random interval accounts for feedback responses and reduces possible packet collisions.

12.6 Simulation Environment and Numerical Results 12.6.1 Channel Modeling The performance of the proposed algorithm was tested using a Matlab based simulation. The configuration of the cellular topology under consideration is depicted in Fig. 12.2. The network is a typical IEEE 802.16 consisting of base stations and

Fig. 12.2 Simulation topology – A typical 802.16 cellular configuration network with fixed subscriber terminals

260


static subscriber stations. In the proposed model, the layout of the network consists of the rectangular grid of cells, with 90◦ sectors. The operational frequency of the system is 40 GHz. At frequencies above 10 GHz, the dominant factor impairing the performance of LOS wireless links is rain attenuation. For this reason, a dynamical rain rate field has been implemented [11–14]. Protocols underlying channel exhibits both spatial and temporal characteristics. The spatial characteristics of rain were simulated using HYCELL, a model for rain fields and rain cells structure developed at ONERA [11,12]. HYCELL is used to produce typical two-dimensional rain rate fields, R (x, y), over an area corresponding to the size of a WiMax network, where R (x, y) denotes the rainfall rate at a point (x, y). These rain fields follow the properties of the local climatology. In addition, the temporal characteristics of the wireless channel, R (t), were simulated using the methodology described in [13], where R (t) denotes the rainfall rate at time t. A typical rain rate field generated using the above model for two different time instants (t, t + 15 min), concerning the province of Attica-Greece, is depicted in Fig. 12.3. Having implemented an accurate model for R (x, y; t), the next step is to determine A Ri j (t) that is the attenuation caused by rain at link (i, j) between base station i and multicast receiver j, at time t. This is achieved by integrating the specific rain attenuation A0 , (dB/km) over the path length L i j of the link (i, j) within the rain medium. L i j A Ri j (t) =

A0 dl

(12.7)

0i

with A0 = a R b

(12.8)

Where R is the rainfall rate (mm/h) and a, b are parameters depending on frequency, elevation angle, incident polarization, temperature and raindrop size distribution [15].

Fig. 12.3 A random generated dynamic rain rate field for a 100 × 100 surface in Attica-Greece

12


261

12.6.2 Numerical Results In Fig. 12.4 the average number of feedback messages versus (d E/a), for various numbers of users, is depicted. First of all it is obvious that the curves show dependence of the expected feedback messages on (d E/a) and are convex with minimums at (d E/a)∗ . For values of (d E/a) less than (d E/a)∗ it is observed that the number of feedback messages is increased due to the fact that the cost for a user to send an acknowledgement is negligible. In this case, users have more to earn, in terms of QoS, from sending an acknowledgment by their self, rather than waiting from the others to do it for them. The above is daily confirmed in the social life. For example, why should someone stay apathetic in an illegal event that is taking place in front of his him, when he has nothing to lose? Equivalently, why should a node act according to the social welfare, suppressing his feedback messages and probably risking his QoS, when he has nothing to earn? On the other hand, for values of (d E/a) greater than the (d E/a)∗ the cost for sending feedback messages is significant. In this circumstance, users have no incentive to send feedback messages and the syndrome of Genovese appears. Then, in Fig. 12.5 the proposed feedback suppression game is compared with the method of exponential timers, described analytically in 0. In this approach, when a subscriber is informed that a feedback message of another receiver will suppress its own feedback sending. Feedback messages are sent on a multicast feedback channel in order to be received also by the other receivers. If every receiver delays its multicast feedback sending by a random time, feedback implosion can be avoided. Furthermore, it is proved that optimum suppression is achieved if timers are chosen to follow exponential distribution [16].

Fig. 12.4 Impact of payoff function in feedback transmission

262


Fig. 12.5 A comparison between timer based and bystander effect feedback suppression algorithms

In case no feedback suppression algorithm is applied, every user sends a feedback message asking for packet retransmission. It is observed that the proposed algorithm achieves almost the same performance with the case where T = 6c and ␭ = 10. Note that, T is the time interval where exponential timer is applied, c is the RTT and λ is a parameter of exponential distribution. The parameters a and dE of the bystander effect algorithm are 1 and 10−2 , respectively.

12.7 Conclusions A game-theoretic based model for the solution of feedback suppression problem for reliable multicast protocols in WiMax networks has been presented. The formulation of the problem was inspired from social psychology, and more specifically from the phenomenon of bystander effect. When a high number of multicast receivers are present, users resemble the behaviour of apathetic human beings, so the feedback implosion problem is being solved satisfactorily. The Genovese syndrome is discussed and the need for back up mechanisms is investigated. The performance of the proposed algorithm is analytically quantified and by employing a physical channel model, extended simulations of a hypothetical WiMax network are presented. Acknowledgments Markos P. Anastasopoulos thanks Propondis Foundation for its financial support.

12


263

References 1. T. Kwok, “Residential broadband Internet services and application requirements,” IEEE Commun. Mag., p. 76, June 1997. 2. The WiMax Forum. http://www.wimaxforum.org 3. H. Sari, “Broadband radio access to homes and businesses: MMDS and LMDS”, Comput Networks, vol. 31, pp. 379–393, 1999. 4. A. Ghosh, D. R. Walter, J. G. Andrews, R. Chen, “Broadband wireless access with WiMax/8O2.16: Current performance benchmarks and future potential”, IEEE Comm. Mag., vol. 43, pp. 129–136 Feb. 2005. 5. J. Yun, M. Kavehrad, “PHY/MAC cross-layer issues in mobile WiMax”, Bechtel Telecomm. Tech. J., vol. 4, no. 1, pp. 45–56 Jan. 2006. 6. ITU-R, P 1410–2, “Propagation data and prediction methods for the design of terrestrial broadband millimetric radio access systems operating in a frequency range of about 20–50 GHz,” in Propagation in Non-Ionized Media, Geneva, 2003. 7. K. Obraczka, “Multicast transport protocols”, IEEE Commun. Mag., vol. 36, pp. 94–102, Jan. 1998. 8. B. Latane, J. Darley, Bystander “Apathy”, Am. Scientist, vol. 57, pp. 244–268, 1969. 9. J. W. Weibull, “Evolutionary game theory”, The MIT Press, Cambridge, Massachusetts, London, 1995. 10. E. Rasmusen, “Games and information”, Blackwell Publishing, Oxford, 2006. 11. L. Féral, H. Sauvageot, L. Castanet, J. Lemorton: “HYCELL: A new hybrid model of the rain horizontal distribution for propagation studies. Part 1 : Modelling of the rain cell”, Radio Sci., vol. 38, no. 3, p. 1056, 2003 12. L. Féral, H. Sauvageot, L. Castanet, J. Lemorton, “HYCELL: a new hybrid model of the rain horizontal distribution for propagation studies. Part 2: Statistical modelling of the rain rate field”, Radio Sci., vol. 38, no. 3, p. 1057, 2003. 13. L. Féral, H. Sauvageot, L. Castanet, J. Lemorton, F. Cornet, K. Leconte, “Large-scale modeling of rain fields from a rain cell deterministic model”, Radio Sci., vol. 41, no. 2, Art. No. RS2010, Apr. 29 2006. 14. A. D. Panagopoulos, J. D. Kanellopoulos, “On the rain attenuation dynamics: spatial–temporal analysis of rainfall rate and fade duration statistics” Int. J. Satell. Commun. Network, vol. 21, pp. 595–611, 2003, DOI: 10.1002/sat.763. 15. ITU-R P-838-2, Specific attenuation model for rain for use in prediction methods, Geneva, 2003. 16. J. Nonnenmacher, E. W. Biersack, “Scalable feedback for large groups”, IEEE/ACM Trans. Networking, vol. 7, no. 3, pp. 375–386, June 1999.

Chapter 13

IEEE 802.16: Enhanced Modes of Operation and Integration with Wired MANs Isabella Cerutti, Luca Valcarenghi, Piero Castoldi, Dania Marabissi, Filippo Meucci, Laura Pierucci, Enrico Del Re, Luca Simone Ronga, Ramzi Tka and Farouk Kamoun

Abstract The evolution of wireless technologies allows users to be always connected to IP-based services through IP-based devices. Moreover the bandwidth available to wireless connected users is becoming comparable to the one provided by copper-based access technologies (e.g., xDSL). Worldwide Interoperability for Microwave Access (WiMax) is one of the wireless technologies that potentially allows users to utilize an access capacity in the order of tens of Mb/s. So far, WiMax (i.e., IEEE 802.16) has been exploited and investigated mainly in the Point-to-MultiPoint (PMP) mode, while IEEE 802.16 enhanced-modes of operation are still at their early research stages. Furthermore, how to integrate Wireless Metropolitan Area Networks (WMANs) based on IEEE 802.16 and wired/optical MAN to guarantee seamless Quality of Service (QoS) across the two transport domains still remains an open issue. This chapter addresses the IEEE 802.16 enhanced-modes of operation and the wireless/wired Metropolitan Area Network (MAN) integration. The focus is on advanced physical layer technologies for wireless transmission such as Multiple Input Multiple Output (MIMO) antennas and Adaptive Modulation and Coding (AMC), the optional IEEE 802.16 Mesh mode of operation, and the integration of wireless and wired/optical MANs. Current status and issues are presented and solutions are proposed. Keywords IEEE 802.16 · WiMax operation modes · Wired MAN · Quality of service · AMC modes · MIMO system · GMPLS · Integrated PHY-MAC simulator

13.1 Introduction The explosive growth of communications is driven by two complementary technologies: optical transport and wireless communications. Optical fiber offers the massive

I. Cerutti and L. Valcarenghi (B) Scuola Superiore Sant’Anna, Pisa, Italy e-mail: [email protected], [email protected]


265

266

I. Cerutti et al.

bandwidth potential that has fueled the rise in Internet traffic, whilst wireless techniques confer mobility and ubiquitous access through bandwidth-constrained and impairment-prone wireless channels. To overcome the bandwidth limitations and impairment susceptibility of wireless communications, a wireless technology for broadband wireless access has been recently standardized: IEEE 802.16, also known as Worldwide Interoperability for Microwave Access (WiMax). IEEE 802.16-2004 standard [1] defines the air interface for fixed broadband wireless access systems supporting multimedia services. IEEE 802.16 offers a wireless alternative to wired Metropolitan Area Network (MAN) access protocols and technologies, with a potential capacity of up to 70 Mb/s. Wired MANs are used to connect the users to the public Internet in areas in which fixed interconnections between routers have been already deployed, such as in densely populated cities. On the other hand, Wireless MANs (WMANs) can be used to offer a broadband fixed wireless access to the public Internet in rural areas or in areas in which infrastructural costs for wired metropolitan networks are too high. So far, wireless and wired MANs have been evolving independently but an integration between these two technologies is necessary to provide high-speed and flexible multimedia services to wireless (possibly mobile) terminals. Thus, improvements are needed in the wireless segment, in the wired segment, and in the integrated wireless-wired network. In the wireless segment, the various operational modes at Physical (PHY) and Media Access Control (MAC) layers available in the IEEE 802.16 standard can be efficiently selected in order to provide high capacity. Multiple Input Multiple Output (MIMO) transmission schemes can exploit wireless channel spatial properties in a Point-to-MultiPoint (PMP) context, while Adaptive Modulation and Coding (AMC), used in the flexible Orthogonal Frequency Division Multiple Access (OFDMA) scheme, is able to mitigate the effect of the wireless channel degradation. The joint adoption of MIMO and AMC, as explained in the following sections, can be considered as an effective solution for future high-performance wireless networks. In addition, MIMO and AMC schemes can benefit from exploiting also channel status information through Network-MAC-PHY interaction (i.e., a crosslayer approach). Finally, multi-hop transmission, such as the optional IEEE 802.16 Mesh mode of operation, has the potential of improving WMAN utilization. On the other hand, the increase in capacity and bandwidth, offered to wireless users, requires an increasing data rate in the wired MAN and calls for the adoption of the optical fiber based transport. Traffic engineering and advanced routing strategies provided by the Generalized MultiProtocol Label Switching (GMPLS) protocol suite must be exploited in the wired network, to improve network utilization and assure the required Quality of Service (QoS). Finally, cross-domain, i.e., wireless/wired, traffic engineering schemes must be implemented to guarantee a seamless QoS to users connected to any network segment. This chapter aims at indicating the enhancements that IEEE 802.16 should accommodate for implementing high-capacity QoS-guaranteed WMANs and the

13

Enhanced Modes of Operation and Integration with Wired MANs

267

integration process toward a seamless convergence with wired networks. Open issues in the implementation of the enhanced-modes of operation and the integration process are outlined. Solutions are proposed for the considered issues and their advantages and drawbacks are evaluated.

13.2 IEEE 802.16 PHY Layer IEEE 802.16 standard defines four different PHY layers, each one addressing specific wireless channels. The two single carrier PHY layers (SC and SCa) are designed to operate typically in high capacity links, respectively over 11 GHz in a line-of-sight (LoS) environment and under 11 GHz frequencies in a non-line of sight (NLoS) environment. In the other PHY layers, Orthogonal Frequency Division Multiplexing (OFDM) and Orthogonal Frequency Division Multiple Access (OFDMA) techniques are designed to be resistant to heavy NLoS environment. OFDM and ODFMA are two variants of the same technology: both divide an extremely fast signal into many slow signals, each one spaced apart at precise frequencies, referred to as subcarriers. The advantage is that multipath fading can be mitigated: OFDM modulation offers robustness against intersymbolic interference and thanks to the multiple narrowband subcarriers, frequency-selective fading on a subset of subcarriers can be easily equalized. The difference between the two variants (i.e., OFDM and OFDMA) lays in the ability of OFDMA to dynamically assign a subset of all the subcarriers to individual users. This allows the allocation algorithms to shape the traffic load over the available resources, while keeping into account the channel state. Thanks to the greater flexibility, OFDMA has been selected by the WiMax Forum as the basic technology for mobile user services [2]. In addition, Scalable OFDMA (S-OFDMA) allows a variable channel bandwidth ranging from 1.25 to 20 MHz and the deployment of networks with a frequency reuse factor of 1, eliminating the need for frequency planning. According to the standard [2], before transmission, the subchannels can be permuted in order to guarantee some physical related characteristics. Permutation zones, both in the DownLink (DL) or UpLink (UL), define how a part of the frame has been permuted. The DL sub-frame or the UL sub-frame may contain more than one permutation zone. In Fully Used Subcarriers (FUSC) and Partially Used Subcarriers (PUSC) permutations, subchannels are pseudo-randomly spread over the bandwidth in order to average the channel state for all the users. In AMC permutation (adjacent permutation), each user is assigned to a set of bins, each one composed of 9 adjacent physical subcarriers. In this case, non homogeneous channel states can be assigned to various users, thus permitting advanced adaptive algorithms, also based on MAC-connection QoS requirements. The OFDM/OFDMA PHY supports frame-based transmissions. The frame duration is fixed for the whole network and may range from 2.5 to 20 ms, depending also on the available bandwidth. Each frame interval includes transmissions of BS and SSs, gaps and guard intervals. Each frame starts with a preamble, a repetitive

268

I. Cerutti et al.

pre-defined pattern, used for synchronization and channel estimation. Uplink and downlink transmissions can adopt frequency or time division duplex (FDD or TDD). To have a high transmission reliability, the following channel coding procedure must be applied to data bits before transmission: 1. Randomization. Each block of data to be transmitted in the uplink or downlink is randomized according to a pseudo-random binary sequence. 2. Forward error correction (FEC). Randomized data are encoded with a concatenation of a Reed-Solomon outer code and a rate-compatible convolutional inner code. 3. Interleaving. Each encoded data bit is interleaved with a block size corresponding to the number of coded bits per allocated subcarriers per OFDM symbol. The objective of the interleaver is to ensure that adjacent encoded bits are mapped onto nonadjacent subcarriers and alternately onto less and more significant constellations bits. After the interleaving, the data bits are mapped into the modulation constellation points by the constellation mapper. The modulation can be flexibly selected, among binary phase shift keying (BPSK), 16 and 64 quadrature amplitude modulation (QAM) or quadrature phase-shift keying (QPSK). The mandatory channel coding per modulation are: BPSK 1/2 (i.e., BPSK with code rate 1/2), QPSK 1/2 and 3/4, 16-QAM 1/2 and 3/4, 64-QAM1 2/3 and 3/4.

13.3 Advanced Techniques: MIMO and AMC This section highlights the advantages of applying advanced transmission and reception techniques to the IEEE 802.16 PHY layer. In wireless links, the overall system performance degrades markedly due to multipath fading, Doppler effect, and time dispersive effects introduced by the wireless propagation. In addition, the limitation of the wireless resources (i.e., bandwidth and power) requires that they have to be efficiently exploited. To enhance the spectral efficiency while adhering to QoS, multiple antenna systems (e.g., MIMO) and adaptive resource allocation can be applied and are considered next. With respect to a wired network, the nature of wireless channel does not allow the steady use of a bandwidth-efficient modulation, due to random fading fluctuations in the channel state. The idea of Adaptive Modulation and Coding (AMC) is to dynamically adapt the modulation and coding scheme to match the transmission parameters to propagation conditions and to achieve various trade-offs between data rate and robustness. The dynamic adaptation of the parameters must take into account also the information coming from the higher levels, in particular from the MAC layer. Therefore, it is important to optimize the different components jointly across layers. 1

The 64-QAM modulation is optional for license-exempt bands.

13


269

The same goal of increasing the spectral efficiency and the coverage can be reached by using a complementary approach based on the utilization of multiple antennas. Some multi-antenna transmission methods, such as beamforming and transmit diversity, have been standardized for large scale third-generation (3GPP) systems and their evolutions, and few of them are already available on the market. Enhanced solutions, such as Multiple Input Multiple Output (MIMO) systems, are included in IEEE 802.16e [2] (also referred to as “Mobile WiMax”), as well as in other broadband wireless standards, such as Wideband Code Division Multiple Access (WCDMA), IEEE 802.11 (WiFi), and IEEE 802.20. These advanced solutions are suitable for WiMax systems, in either PMP or Mesh networks, and can be adopted to increase network performance. In particular, they are expected to bring a significant improvement in the throughput of wireless networks and an adaptive robustness against wireless channel fluctuations, experienced especially by highly mobile stations.

13.4 IEEE 802.16 AAS and MIMO Support IEEE 802.16-2004 (also known as IEEE 802.16d) standard [1] introduces multi antenna support. The “Amendment 2 and Corrigendum 1 to IEEE std. 802.16-2004”, namely 802.16e version [2], adds several details about Adaptive Antenna Systems (AAS) and MIMO [3]. The support of mulitple antenna systems by IEEE802.16d/e is summarized in Table 13.1. Table 13.1 IEEE802.16d/e MIMO support; Alamouti Space Time Diversity (STTD), Spatial Multiplex (SM) PHY layer

SC

SCa

OFDM

OFDMA

802.16d

no

STTD, AAS

STTD, AAS

802.16e

no

STTD, AAS

STTD, AAS

MIMO 2-4 BS antennas TD and SM; AAS 2-3-4 TX antenna grouping and selection, STTD; Layered Alamouti; SM SM Uplink MIMO (collaborative SM); AAS with AMC permutation

13.4.1 Adaptive Antenna Systems (AAS) In Adaptive Antenna Systems (AAS), the signals from several antenna elements (not necessarily a linear array) are weighted (both in amplitude and in phase) and combined to maximize the performance of the output signal. The smart antenna’s beams are not fixed and can place nulls in the radiation pattern to cancel interference and mitigate fading, with the objective of increasing the spectral efficiency of the system.

270

I. Cerutti et al.

The IEEE 802.16 standard defines a signaling structure that enables the use of adaptive antenna system. A PMP frame structure is defined for the transmission on downlink and uplink using directional beams, each one covering one or more Subscriber Stations (SSs). The Base Station (BS) forms a beam based on channel quality reported by the SSs. The part of the frame with adaptive antenna transmission is included in an AAS zone, which spans over all the subchannels until the end of the frame or until the next permutation zone. AAS support for AMC permutation has been added in IEEE 802.16e version of the standard. When a Fast Fourier Transform (FFT) size greater than or equal to 512 is used, the BS can decide to allocate an AAS Diversity-Map Zone. The Diversity-Map Zone is placed accordingly to the permutation used. In PUSC, FUSC, and optional FUSC permutation, it is positioned on the two highest numbered subchannels of the DL frame, while in AMC it is positioned on the first and last subchannels of the AAS Zone. These subchannels are used to transmit the Downlink Frame Prefix (AAS-DLFP) whose purpose is to provide a robust transmission of the required BS parameters to enable SS initial ranging, as well as SS paging and access allocation. In order to enter the network using the DLFP, an AAS-SS follows a specific procedure. The Downlink Channel Descriptor (DCD) and Uplink Channel Descriptor (UCD) offer information useful for decoding and demodulation. A channel with broadcast Connection Identifier (CID) is readable by every SS. The channels are allocated accordingly to the DL/UL maps transmitted after the frame header. An SS entering the network in AAS mode follows these steps:

r r r r r r

the AAS-SS synchronizes time and frequency by using the DL preamble; AAS-SS receives the necessary messages to identify used modulation and coding (as the DCD and UCD pointed to by allocations made from the AAS-DLFP using the broadcast CID) the AAS-SS decodes the DCD and UCD and then it performs ranging; the AAS-SS receives a ranging response message through a DL-MAP allocation pointed to by an AAS-DLFP with the broadcast CID; the AAS-SS receives initial downlink allocations through a DL-MAP allocation pointed to by the AAS-DLFP with broadcast or specific CID; other allocations can be managed by private DL-MAP and UL-MAP allocations.

13.4.2 Multiple Input Multiple Output System (MIMO) Multiple Input Multiple Output (MIMO) support has been introduced since IEEE 802.16-2004. In IEEE 802.16e, several features have been added for multiple antennas at the transmitter and receiver for OFDM and OFDMA PHY layers. These systems can be used to achieve diversity gain in a fading environment or to increase capacity.

13


271

When transmit diversity is desired, multiple copies of the same data stream are transmitted over independent spatial channels which are created by employing multiple antennas. Transmission is more robust to wireless channel fluctuations, since it is unlikely that all the channels fade simultaneously. When higher capacity is needed, spatial multiplexing is utilized to transmit various streams of data simultaneously over different antennas in the same time slot, over the same frequencies. If statistical decorrelation among antenna elements is available, multiple transmit and receive antennas can create independent parallel channels and the transmitted symbol can be correctly reconstructed at the receiver. Decorrelation condition can be satisfied by using antennas well separated (by more than λ/2) and/or with different polarizations. IEEE 802.16e standard defines three MIMO operation modes: Alamouti Space Time Transmit Diversity (STTD) [4], Layered Alamouti Space Time Coding (LSTC) [5] and Spatially Multiplexed Vertical Bell Laboratories Layered Space-Time (SM VBLAST) [6, 7]. Switching among these modes permits to follow the channel state [8]. In addition, adaptive spatial modulation, jointly with adaptive modulation and coding techniques, can offer high flexibility at the physical layer and allows to maximize data throughput and coverage. STTD is standardized for SCa, OFDM, and OFDMA PHY Layers. The Alamouti scheme needs two transmitting antennas at the BS and provides a transmit diversity of two. Alamouti relies on a constant channel response in two adjacent symbols: channel variations that occur during two-symbol time interval are the main source of performance degradation. The receiver is a linear combiner [4] where symbols can be reconstructed by using orthogonal properties of space-time coding matrix A. For SCa (Single Carrier for NLoS operation in frequency bands below 11 GHz), STTD is used at a burst level. A burst is composed by a number of QAM symbols equal to F. The bursts are arranged on a time basis as shown in Fig. 13.1. The STTD requires the processing of a pair of time bursts. The burst are coded accordingly to the well-known Alamouti matrix, where rows represent the transmitting antennas and the columns represent the burst time slots:

U Tx Ant 1

CP

N F

U

Payload 0: s0 [n]

U CP

Cyclic Prefix, last U symbols

U Tx Ant 2

CP

N F Payload 1: s1 [(F-n)mod(F)] Cyclic Prefix, last U symbols

Fig. 13.1 Single carrier Alamouti transmission

N F

U

Payload 1: -s1 [n] Cyclic Prefix, last U symbols

U

U CP

N F Payload 0: s0 [(F-n)mod(F)] Cyclic Prefix, last U symbols

U

272

I. Cerutti et al.

A=

s0 −s1∗

s1 s0∗

(13.1)

where ()∗ is the conjugate operator and sk is the burst at time k. First antenna should transmit two sequences of F symbols while the second antenna shall not only reverse the order in which burst are transmitted but also conjugate the transmitted complex symbols and time-reverse the sequence of data within each burst. The index n is the running position inside the burst. The time reversing operation is realised by the (F − n)mod(F) operation, as in Eq. (13.1), where the transmission for a pair of bursts is reported. A portion of the bursts (U symbols) is copied to form the Cyclic Prefix (CP). The receiver using the STTD scheme in SCa PHY layer can be found in Section 8.2.1.4.3.1 of [1]. In OFDM and OFDMA PHY layers STTD transmission operates differently. STTD encodes information at the symbol level and not as bursts. OFDM and OFDMA are differentiated by the minimum data unit that can be manipulated. In OFDM PHY layer, STTD operates on two subsequent on two subsequent OFDM symbols, while, in OFDMA, it can operate on a single group of subcarriers in the time-frequency allocation grid. In both cases, the symbols are coded accordingly to the matrix in Eq. (13.1), where rows represent the transmitting antennas and columns represent OFDM symbols or subcarriers for OFDM/OFDMA PHY layer respectively. Compared to the SCa PHY layer case, no time-reverse operation is needed, since transmitter and receiver are performing IFFT and FFT processing. When the BS has three or four antennas, it is not possible to achieve a fulldiversity approach, since it has been demonstrated that a full-rate, fully orthogonal Space Time Code only exists for two antennas [9]. When a full rate transmission is desired, LSTC schemes have to be used. Data rate is increased at the expense of diversity gain, linearity or orthogonality. In IEEE 802.16e, the proposed scheme for four antennas is expressed by the coding matrix B. Orthogonality is lost but the full diversity gain and the receiver linearity are preserved. This scheme is a tradeoff between a full-diversity STTD and SM VBLAST approach, which offers maximum capacity gain but no diversity gain. Layered schemes are subject to interference among symbols transmitted from different antennas. This results in Bit Error Rate (BER) degradation with respect to the orthogonal case. The receiver needs a number of receiving antennas Mr > Nblocks , where Nblocks are the orthogonal blocks which have been spatially multiplexed (or layered). In the case of a 4-antenna BS, the LSTC matrix is: ⎛

s1 ⎜−s ∗ 1 B=⎜ ⎝ s3 −s4∗

s2 s2∗ s4 s3∗

s5 −s6∗ s7 −s8∗

⎞ s6 s5∗ ⎟ ⎟. s8 ⎠ s7∗

(13.2)

Two Alamouti blocks are transmitted at the same time from a subset of the available BS antennas. This solution achieves the same transmit diversity as in the classic

13


273

2 × 1 Alamouti scheme and a spatial gain which is upper limited to two. The spatial gain depends on the channel response: the optimum is achieved when the channel matrix is orthogonal. In this case, contrary to the Alamouti case, no detection loss is experienced since cross-stream interference is null. The receiver operates as follows. Let us consider a system with a number of transmitting antennas Mt = 4 and a number of receiving antennas Mr = 2. The channel coefficient between the i-th transmitting element and j-th receiving element at the n-th signaling time is h i j(n) and s = [s1 , s2 , s3 , s4 ]T . At the receiver, the following composition of the two Mr × 1 received vectors y at time n and n + 1 is considered:

y(n) y∗(n+1)

⎡

=

h 11(n) ⎢ h 12(n) ⎢ ∗ ⎣h 21(n+1) h ∗22(n+1)

h 21(n) h 22(n) −h ∗11(n+1) −h ∗12(n+1)

⎤ h 41(n) h 42(n) ⎥ n(n) ⎥ s + −h ∗31(n+1) ⎦ n(n+1) ∗ −h 32(n+1)

h 31(n) h 32(n) h ∗41(n+1) h ∗42(n+1)

where n(n) is the Mr × 1 noise vector at time n. Channel matrix H has peculiar orthogonality properties that can be exploited in order to separate the spatially multiplexed Alamouti blocks. Let us define H = [H1 |H2 ]. The Least-Square (LS) receiver is obtained from the Moore-Penrose pseudoinverse H+ , as follows: ⎡ H H1 H1 sˆ = H+ y = UV y = ⎣ H2H H1

H1H H2

⎤−1 ⎡ ⎦

⎣

H2H H2

H1H

⎤ ⎦ y.

(13.3)

H2H

The matrix U has the auto-products of Alamouti blocks H1 and H2 on the principal diagonal and it has two cross-products representing the interferences due to spatial multiplexing. Let ( ) H and ( )T be the hermitian and real transpose operators, respectively. Due to Alamouti orthogonality, the auto-products of Alamouti blocks are: H1H H1 = k1 I

H2H H2 = k2 I

where k1 = h1 2 + h2 2

k2 = h3 2 + h4 2 .

The vector hi contains M R ×1 channel coefficients from the i th transmitting antenna. The anti-diagonal elements of matrix U satisfy: H1H H2 H2H H1 = ( + χ )I 2 2 where = h1H h4 − h2T h∗3 and χ = h1H h3 + h2T h∗4 .

274

I. Cerutti et al.

It can be shown [5] that a Moore-Penrose channel inversion can be attained using linear processing exploiting orthogonal properties of the STC. The estimated transmit vector sˆ can be written as: F2

$ ⎡

sˆ =

k2 I 1 ⎣ k1 k2 − ( + χ ) −H H H

1

2

F1 %& ' $ %& ⎡ ⎤' ⎤ −1 H1H −H1H H2 ⎦ ⎣ ⎦y k1 I H2H

(13.4)

The vector hi contains M R ×1 channel coefficients from the i th transmitting antenna. First, the receiver vertically processes two spatially multiplexed streams with matrix F1 . Then, with matrix F2 , it cancels cross-products (interferences). According to Eq. (13.4), Signal-to-Noise Ratio (SNR) undergoes a reduction that is proportional to ( + χ ). The SNR values for the estimation of the received symbols are: S N R12 =

k1 − ( + χ ) k2

and

S N R34 =

k2 − ( + χ ) . k1

(13.5)

−1

10

−2

BER

10

−3

10

−4

10

Alamouti 4X2 ant, 2 Layers 4 QAM (4 bps/Hz) Alamouti 2X2 ant, Single Layer 16 QAM (4 bps/Hz) VBLAST 2X2 ant, 2 Layers 16 QAM (8bps/Hz)

−5

10

0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

SNR

Fig. 13.2 STTD, LSTC, and VBLAST

The main idea of this family of STC comes from multiuser detection strategies. As a matter of fact, IEEE 802.16e standard offers the support for spatially multiplexed streams in multiuser scenarios. In downlink, each SS decodes the streams directed to itself on the basis of the map broadcast at the start of each frame. As a third MIMO option, the system can switch to a VBLAST transmission [6]. Independent data streams are spatially multiplexed, i.e., they are transmitted from different antennas in the same OFDM symbol time; each data stream is referred to as a layer. Since the code is operating over a single OFDM symbol, VBLAST is

13


275

a space-only coding. The coding matrix (in the case of two transmit antennas) is reduced to a vector: s C= 1 . s2

(13.6)

Layers can be separated at the receiver if the number of antennas is Mr > Nstr eams . While STTD of LSTC do not exploit all the freedom degrees of the MIMO channel, VBLAST can extract the complete spatial gain. On the other hand, VBLAST reaches smaller diversity gain with respect to the Alamouti 2 × 2 case. VBLAST receiver decodes the streams in successive steps. Receiving algorithm starts from a first signal detection. The detected symbol is then cancelled from each one of the remaining received signals, and so on. It is evident the analogy with the multiuser interference cancelled sequentially. The drawback of this method can be envisaged in the sequential order of the estimation and cancellation: a wrong decision propagates the error from one step of detection to the followings. Optimum decoding order has been demonstrated in [10] and is achieved by detecting the highest SNR layer at each successive cancellation step. Then, linear combinatorial nulling (Zero Forcing) is applied to remove interference caused by undetected layers. VBLAST receiver based on Zero Forcing (ZF) linear detection technique can be briefly summarised with the following steps: 1. compute the weight vector wki selecting entries from inverted channel matrix Gi wki = (Gi )ki 2. find the highest SNR received signal among the Mr available signals 3. perform signal detection (weighting and slicing) yki = wkTi ri aˆ ki = Q(yki ) 4. operate interference cancellation ri+1 = ri − aˆ ki (H)ki 5. update channel matrix before next iteration (+ Gi+1 = H k i

(ki is the matrix obtained by zeroing columns where (Gi ) j is the j-th row of Gi , H k1 , k2 , . . . , ki , and ( )+ denotes the Moore-Penrose Pseudoinverse. The process is repeated until the last interference-free stream has been detected. Figure 13.2 shows a performance comparison among STTD, LSTC, and VBLAST schemes. In uplink transmission, an additional feature is allowed: the Collaborative SM scheme. In a multiuser Collaborative SM context, two users may share the same subchannel in the uplink. In this case, an SS can request an allocation in the uplink to be used in coordination with a second SS in order to perform a Collaborative Spatial Multiplex. SS participating at the Collaborative SM can be equipped with a single antenna.

276

I. Cerutti et al.

13.5 Adaptive Modulation and Coding In wireless systems, the signal transmitted to and by a station can be modified to counter-react to the signal quality variations through a process commonly referred to as link adaptation. This allows to improve system capacity, peak data rate, and coverage reliability. Traditionally, wireless systems use fast power control as the preferred method for link adaptation. In a system with power control, the power of the transmitted signal is adjusted in order to meet a target carrier-to-interferenceplus-noise ratio at the receiver. Therefore, typically, the transmit power is low when a user is close to the BS and it increases when the user moves away from the BS. Adaptive Modulation and Coding (AMC) is offering an alternative link adaptation method that promises the increase of the overall system capacity. In a system with AMC, the power of the transmitted signal is held constant, but the modulation and coding formats are changed to match the current received signal quality. AMC provides the flexibility to match the modulation-coding scheme to the average channel conditions of each station. Users close to the BS are typically assigned higher-order modulations and high code rates. The modulation-order and/or the code rate may decrease as the distance from the BS increases. In particular, in an OFDM/OFDMA wireless system, the inherent multi-carrier nature of OFDM allows the use of AMC according to the behaviour of the narrowband channels (subcarriers) and different subcarriers can be allocated to different users to provide a flexible multiuser access scheme, that exploits multiuser diversity. Adaptive Modulation and Coding is twofold supported by IEEE 802.16 standard [2]. First, a large selection of modulation and channel coding is available at BS and SS. All the possible combinations between the modulations QPSK, 16-QAM, and 64-QAM and the coding rates 1/2 and 3/4 are allowed. In addition, the 64QAM modulation can be also combined with a coding rate 2/3. Then, a special AMC permutation scheme is defined, where subchannels are composed by groups of contiguous subcarriers. The contiguous permutations include downlink AMC and uplink AMC and have the same structure. A bin consists of 9 contiguous subcarriers (eight data subcarriers and one pilot subcarrier) in the same symbol. Let N be the number of contiguous bins and M be the number of contiguous symbols. A slot in AMC is defined as a collection of bins of type (N x M), with (N x M) = 6. Thus, the allowed combinations are: (6 bins, 1 symbol), (3 bins, 2 symbols), (2 bins, 3 symbols), and (1 bin, 6 symbols). In addition to power control and AMC, Dynamic Subcarrier Assignment (DSA) schemes can be used in OFDM/OFDMA systems. In particular, due to the contiguous subcarrier allocation, each user can experience highly-variable channel conditions and may benefit from multiuser diversity by choosing the subchannel with the best frequency response. High spectrum efficiency can be achieved with adaptive subchannel allocation strategies. In conclusion, there is a wide degree of flexibility for radio resource management in the context of OFDM. Since channel frequency responses are different at different frequencies or for different users, adaptive power allocation, AMC, and DSA, can significantly improve the performance of OFDM systems. The link adaptation algorithms

13


277

Base Station

Burst Composer

Carriers Permutation

IFFT and CP insertion

AMC Decision Process

to RF stages

CSI from SS

Subscriber Station

from RF stages

CP removal and FFT

CSI to BS

Carriers De-permutation

RX Burst Detection

Channel Estimation

Fig. 13.3 AMC usage graph

can be designed to maximize the overall network throughput or to achieve target error performance. The former objective may be appropriate for best-effort services but does not meet QoS requirements in terms of error performance. The latter one may ensure the QoS requirements, at the expense of network throughput. Dynamic link adaptation in the downlink is obtained through SS feedback, by providing the transmitting BS with channel state information (CSI) estimates, as illustrated in Fig. 13.3. In the uplink the SS can change its transmission parameters based on its own channel estimates. The accuracy of the channel estimates and the latency of the feedback affect the AMC algorithms performance. Another practical consideration in AMC is how quickly the transmitter must change its constellation size. Since the constellation size is adapted to an estimate of the channel fade level, several symbols may be required to obtain a good estimate. These practical considerations are relevant in a mobile context with fast varying channels and require accurate solutions such as prediction models for the channels.

13.6 IEEE 802.16 Mesh Mode Overview IEEE 802.16 offers a wireless alternative to “last-mile” access protocols and technologies. Using IEEE 802.16 protocol, stations connected to the wired network act as BS and offer broadband services to the SSs, as illustrated in Fig. 13.4. The main

278 Fig. 13.4 IEEE 802.16 deployment scenario

I. Cerutti et al.

IEEE 802.16 Mesh Mode

IEEE 802.16 PMP Mode BS

Mesh BS

Wired Network

SS SS

l ca od Lo rho o hb d ig de d ne ten oo Ex borh h ig ne

Mesh SS

Mesh SS

operation mode of IEEE 802.16 is based on PMP communications from/to the BS and thus is referred to as PMP mode. In addition, IEEE 802.16 defines an optional operation mode, called Mesh mode. The two different operation modes of IEEE 802.16 are sketched in the Fig. 13.4. In the PMP mode, the BS communicates directly with each SS (i.e., downlink transmission). SSs communicate with the BS (i.e., uplink transmission) on-demand. MAC is connection-oriented and offers the flexibility to choose among different types of uplink scheduling depending on the required QoS (i.e., Unsolicited Grant Service (UGS), extended real-time Polling Services (ertPS), Real-time Polling Service (rtPS), Non-real-time Polling Service (nrtPS), and Best Effort (BE)). In the optional Mesh mode, the Mesh BS is the system directly connected to backhaul services, such as those offered by wired networks. The Mesh SSs may communicate to the Mesh BS either directly or through multi-hop. Unlike PMP SSs, Mesh SSs can also communicate among themselves directly, by exchanging traffic or forwarding traffic on behalf of other Mesh SSs. The transmission can be classified into uplink and downlink transmissions, depending on the direction of the traffic, i.e., toward to and away from the Mesh BS, respectively. Mesh mode permits to efficiently schedule uplink and downlink transmissions, in a centralized or distributed manner, as it will be explained in the following sections. Furthermore, an extension of MAC and PHY layer of PMP mode is currently being investigated by the IEEE 802.16’s relay task group [11]. The objective is the definition of mobile multi-hop relay (MMR) functionalities between BS and SSs, and also between BSs and a central controller connected to the wired infrastructure. Proposals for more complex network architectures, such as a tree of PMP networks or a hybrid PMP/Mesh network, are also being discussed. Mesh mode networks are limited to operate on licensed bands below 11 GHz (typically in the 2–11 GHz range) using OFDM. Mesh mode may operate also on license-exempt band below 11 GHz (typically in the 5–6 GHz range) using OFDM.

13


279

The next sections will further describe the PHY and MAC layers of IEEE 802.16 for the optional Mesh mode. In the rest of the paper, Mesh BS and Mesh SS will be addressed as BS and SS, in order to simplify the notation.

13.7 Mesh Mode Physical Layer In this section the main features of the Mesh mode physiscal layer are described. Other features are described in Section 13.2. In each frame, uplink and downlink transmissions are time division duplexed (TDD). As depicted in Fig. 13.5, each frame is divided into:

r

a control sub-frame. The control sub-frame has a length that can be flexibly selected as a multiple of 7 OFDM symbols2 and it is encoded using the robust QPSK-1/2 encoding with the mandatory coding schemes. The control sub-frame contains opportunities for the transmission of MAC Management messages. As time Frame n

Frame n+1

Network configur.

MAC PDU with Long MSH−NENT preamble

Network configur.

PHY tr. burst from SS #j

Guard symb. Guard symb. Guard symb.

Network entry

Data subframe

Long MAC PDU with preamble MSH−NCFG

Schedule Control subframe PHY tr. burst from SS # j

Long MAC PDU with preamble MSH−CSCH


Distributed scheduling

Data subframe

Long MAC PDU with preamble MSH−CSCF


Centralized Centralized scheduling configur.

PHY tr. burst from SS #k


Network Control subframe

Frame n+2

PHY tr. burst from SS # k

Long MAC PDU with preamble MSH−DSCH


Frame n−1

Fig. 13.5 Mesh frame structure

2

The length in multiple of 7 OFDM symbols is defined by the MSH-CTRL-LEN field in the mesh configuration message (MSH-NCFG).

280

r

I. Cerutti et al.

shown in Fig. 13.5, the control sub-frame can be further classified into two different categories: network control sub-frame and schedule control sub-frame, depending on the type of carried information. The network control sub-frame carries MAC Management messages for the set-up and maintenance of the mesh topology. The schedule control sub-frame carries MAC Management messages for the coordinated scheduling of data transmissions. a data sub-frame. The data sub-frame is divided into minislots that are available for either uplink or downlink data transmissions.

13.8 Mesh Mode Medium Access Control The medium access control layer of IEEE 802.16 Mesh mode is subdivided into a security sublayer, MAC common part sublayer (CPS) and the service specific convergence sublayer (CS). The security sublayer provides access control and confidentiality to MAC data transmissions. Security associations are over a single-hop link, thus the traffic needs to be encrypted and de-crypted at each SS. Traffic encryption keys are exchanged by using Rivest-Shamir-Adleman (RSA) algorithm with 1024-bit keys. The service specific convergence sublayer (CS) interfaces the MAC common part sublayer (CPS) with various higher layer protocols. CS functions include the mapping of external network data into service data unit (SDU) to be delivered to MAC CPS and the reporting of process results to CS. Currently, the CS specifies a packet convergence sublayer for the support of IPv4 and IEEE 802.3/Ethernet packets and an ATM convergence sublayer. Only the packet convergence sublayer is required to be supported in Mesh networks. Next subsections describe in more details some of the MAC CPS specifications.

13.8.1 Mesh Node and Link ID At the MAC layer, the wireless network operating in Mesh mode can be represented as mesh topology. The nodes of this mesh topology are BSs and SSs. Nodes of the wireless network are uniquely identified by a 48-bit universal MAC address. However, for IEEE 802.16 operations, a 16-bit node address (Node ID) is used. The Node ID is assigned to a candidate node by the BS upon authorization to join the network. A link connects two nodes when a neighbor relationship exists between the corresponding stations. Two stations (BS or SS) are defined as neighbors, when they directly communicate through a radio channel. The local neighborhood of a node includes all the neighboring nodes, one-hop apart from the considered node. The extended neighborhood of a node includes the local neighborhood and the neighbors of the local neighborhood, i.e., all the nodes that are one or two hops apart from the considered node. The left hand side of Fig. 13.4 depicts the local and extended

13


281

neighborhood for the indicated Mesh BS. An 8-bit link identifier (Link ID) is used for addressing the links in the local neighborhood of a node.

13.8.2 MAC PDU

LSB

MSB

The generic format of MAC PDU is illustrated in Fig. 13.6a. It contains a fixed size (48 bits) generic MAC header, an optional extended subheader, a 16-bit Mesh subheader, a variable size optional payload, that can carry either CS data or a MAC Management messages, and a CRC. Thanks to the variable size, higher layer traffic can be accommodated in the payload and tunnel through the MAC. The generic MAC header (see Fig. 13.6b) contains information on the header type (HT), encryption control (EC), type encodings (Type), extended subheader field (ESF) to indicate the presence of the extended subheader, CRC indicator presence (CI), encryption key sequence (EKS), MAC PDU length (LEN), connection identifier (CID), header check sequence (HCS) for error detection in the header. As shown in Fig. 13.6c, depending on the unicast/broadcast nature, the CID field in Mesh networks may include service parameters such as Type (e.g., MAC Management, IP),

Extended subheader

Generic MAC header (48)

(optional)

Mesh subheader

Payload

CRC

(optional)

(16)

LEN LSB (8) CID LSB (8)

EKS (2)

Rsv (1)

CI=1 (1)

Type (6)

ESF (1)

EC (1)

HT=0 (1)

MSB

a) MAC PDU format

LEN MSB (3)

CID MSB (8) HCS (8)

b) Generic MAC header format Type (2)

Rel (1)

Unicast

Broadcast Prio

DP

Link ID

NetID

0xFF

(3)

(2)

(8)

(8)

(8)

c) CID format Management Management Message Payload Message Type d) Payload format for MAC Management messages

Fig. 13.6 MAC message format

282

I. Cerutti et al.

reliability (Rel), priority/class (Pri) and drop precedence (DP), link identification (Link ID) and network identification (NetID). The Mesh subheader contains the 16-bit Node ID of the transmitting node. When operating in Mesh mode, the Mesh subheader is mandatory and precedes the other subheaders, except for the optional extended subheader (Fig. 13.6a). To indicate the presence of the Mesh subheader, the most significant bit of the Type field in the MAC header is set to 1. The MAC Management messages are carried in the payload of the MAC PDU and consist of a Type field and a payload, as illustrated in Fig. 13.6d. The Type field identifies the type of MAC Management message. The use of the different MAC Management messages in the mesh-specific procedures of Synchronization, Network entry, Scheduling and ordinary operations is described next.

13.8.3 Synchronization The MAC layer incorporates mechanisms to achieve synchronization within the network and with nearby networks. Synchronization within the Mesh network is achieved through a number of externally synchronized nodes (e.g., GPS-connected) that act as master time keepers. The other nodes must synchronize to the neighbors, that are closer to the externally synchronized nodes (i.e., nodes with lower synchronization hop count). For this purpose, Mesh network configuration messages (MSH-NCFG) contain a timestamp and synchronization hop count information, that can be used to achieve synchronization. MSH-NCFG messages are sent periodically by the Mesh nodes (i.e., both BSs and SSs), during the network control sub-frame transmit opportunities, following an election-based scheduling. In addition, Mesh network entry messages (MSH-NENT) permit new nodes entering the network to achieve synchronization. MSH-NENT messages are sent only in the first transmit opportunity of a network control sub-frame, on a contention-based access. Finally, for synchronization to the start of the frame, MSH-NCFG, MSH-NENT and distributed scheduling messages (MSH-DSCH) are used. Between nearby Mesh networks, synchronization and coordination of the channel usage can be achieved by using the information carried by MSH-NCFG and MSH-NENT messages.

13.8.4 Network Entry A node entering into a Mesh network, referred to as candidate node, should perform the following steps:3

3

In the description, the notation MESSAGE:Type is utilized.

13

r

r r

r

r

r r


283

Scan the possible channels in the frequency bands of operations for active networks and establish coarse synchronization with the network by acquiring the timestamp from MSH-NCFG:Network Descriptor messages. MSH-NCFG: Network Descriptor message advertises the network channels and the PHY PDU profiles (i.e., the FEC code type and the mandatory exit and entry carrier-to-interference-and-noise ratio threshold) supported in the network. Obtain network parameters from MSH-NCFG:Network Descriptor messages and select a candidate Sponsor Node. Request a sponsor channel, i.e., a temporary schedule for the communications to the candidate Sponsor Node, by sending a Mesh network entry message, MSH-NENT:NetEntryRequest. The sponsor channel is opened when the candidate Sponsor Node replies positively by advertising the candidate node MAC address in the MSH-NCFG:NetEntryOpen message. The sponsor channel is available when the candidate node – now new node – acknowledges the response with a MSH-NENT:NetEntryAck message. Negotiate the basic capabilities of the links. A link is established to neighboring nodes by advertising an SS basic capability request message (SBC-REQ). The node at the other end of the link may accept the negotiation by replying with an SS basic capability response message (SBC-RSP), containing the intersection of both nodes’ capabilities. Request authorization to the Authorization Node4 by tunneling a privacy key management request message (PKM-REQ) through the Sponsor Node. Upon verification of the certification, the Authorization Node authorizes the new node to join the network by tunneling a privacy key management response message (PKM-RSP) through the Sponsor Node. Request registration by sending a registration request message (REG-REQ) to the Registration Node by tunneling through the Sponsor Node. Upon reception of a registration response message (REG-RSP), the new node is assigned a Node ID. Establish IP connectivity by acquiring IP address using DHCP, establish time of the day using IETF RFC 868-defined protocol and, if necessary, transfer the operational parameters by downloading the Configuration File via TFTP. Once the TFTP download is completed, the new node acknowledges the transfer with a TFTP completed message (TFTP-CPLT) and the BS establishes the provisioned connection by sending a TFTP response messages (TFTP-RSP). All these operations are taking place over the sponsor channel.

In the case of abnormal behavior of an SS, the BS may force the SS MAC re-initialization and the SS network entry process, by sending a reset command message (RES-CMD).

4 The Authorization Node and the Registration Node are defined in the sponsor node configuration file that each SSs can download via trivial file transfer protocol (TFTP).

284

I. Cerutti et al.

To establish links to neighbors, other than to the Sponsor Node, a mesh node utilizes MSH-NCFG:Neighbor Link Establishment IE message whose exchange is based on the following procedure:

r r r

A node sends a MSH-NCFG:Neighbor Link Establishment IE challenge to the challenged node, i.e., the terminal node of the candidate link. The challenge contains private keys. Upon reception, the challenged node replies with a positive or negative acknowledgment, MSH-NCFG:Neighbor Link Establishment IE challenge response, depending on the matching of the received and stored information, including the private keys. In the case of positive reply, upon a match of the information at the challenging node, the link is established when the MSH-NCFG:Neighbor Link Establishment IE accept message is received at the challenged node.

13.8.5 Scheduling Two scheduling mechanisms are defined for Mesh mode: distributed and centralized. Both mechanisms can be concurrently supported, by flexibly reserving the number of data minislots for centralized scheduling and the number of distributed scheduling Management messages, in the MSH-NCFG message.

13.8.6 Distributed Scheduling Distributed scheduling can be performed in coordinated or uncoordinated manner. In coordinated distributed scheduling, all the nodes in the extended neighborhood coordinate their scheduling. To achieve coordination, each node informs periodically the neighbors about its schedule, by transmitting an MSH-DSCH message in the schedule control sub-frame. In uncoordinated scheduling, the two nodes of a link coordinate their scheduling, ensuring that it does not interfere with centralized or coordinated distributed scheduling. Contention-based transmissions of MSH-DSCH messages occur in the data sub-frame without interfering with the existing schedule (but MSH-DSCH message collisions may occur). In both cases, the distributed scheduling is agreed upon a three way exchange of MSH-DSCH messages. When making a request through a MSH-DSCH:Availabilities message, the node includes information of potentially available slots and actual schedule. Upon request, an indication of the availabilities fitting the request is issued in a MSH-DSCH:Grant message. The schedule is acknowledged with a MSH-DSCH:Grant message sent by the requester to confirm the schedule. During the exchange of MSH-DSCH:Grant messages, the neighbors of the requesting and granting nodes are automatically informed of such schedule.

13


285

13.8.7 Centralized Scheduling The centralized scheduling permits coordinated and collision-free transmissions on the links of a routing tree covering the mesh network. The root of the routing tree is the BS and the links are those established from any SS to its corresponding Sponsor Node. The routing tree is advertised in the centralized scheduling configuration message (MSH-CSCF). The BS broadcasts the MSH-CSCF message, that contains the updated routing tree and the PHY PDU profile of each link, to all the neighbors that in turn rebroadcast it to their neighbors. Using the centralized scheduling message (MSH-CSCH), each SS may request bandwidth to the BS for each node in its own subtree. Uplink requests start in the last frame in which a node received the previous schedule. The BS collects the resource requests from the SSs, decides the resource assignment and broadcasts it through a MSH-CSCH message to all the neighbors that in turn rebroadcast it to their neighbors with higher tree depth. Each SS determines its own data transmission schedule in a distributed fashion by dividing the data sub-frame reserved for centralized scheduling proportionally to the assignments. Nodes, that are distant a number of hops equal or greater than the channel reuse field in MSH-NCFG message, may concurrently transmit in the same opportunity [12–14]. The schedule is valid for the time required to collect the request and distribute the next schedule, as shown in Fig. 13.7 for a tree topology composed of a BS (node 0), two BS’s child nodes (node 1 and 2), and two child nodes of node 1 (node 3 and 4). The scalability of the centralized scheduling and the delay between the arrival of an higher layer PDU and its transmission have been evaluated in [15]. validity of previous schedule MSH−CSCH requests

MSH−CSCH grants

tree depth

0 1 2

2

1

3 4 control data subframe subframe

2 3 4

control data control data subframe subframe subframe subframe

control data subframe subframe

Fig. 13.7 Schedule validity

13.8.8 Ordinary Operations Once the SS is registered into the network, it is responsible for performing the following operations:

r

keeping updated information on the local or extended neighborhood. Each SS maintains a neighborhood list with neighbor-specific information on MAC address, Node ID, hops to the BSs, synchronization hop count, and next expected MSH-NCFG message from the neighbor;

286

r

r

r

I. Cerutti et al.

periodic network configuration broadcasting. Periodically, the SS sends a MSHNCFG message with the updated neighborhood list, network information (such as available burst profiles and operator identifier), channel characterization, and time frame scheduling information (such as frame duration, control sub-frame length, and centralized scheduling length); data transmission. The SS can transit its own data or forward the data to its Sponsor Node on behalf of its child nodes (and viceversa), during the minislots allocated by the centralized scheduling (MSH-CSCH message). In addition, the SS may transmit data to the neighbors during the minislots allocated by the distributed scheduling message (MSH-DSCH); Management message tunneling and broadcasting. The SS is responsible for UDP/IP tunneling the Management messages (such as PKM-REQ, PKM-RSP, REG-REQ, REG-RSP) and forwarding data and Management messages (such as MSH-CSCH, MSH-CSCF) for/to its child nodes in the tree.

13.9 Open Issues and Solutions As described in Section 13.2, IEEE 802.16 allows different MIMO and link adaptation strategies to improve the link performance, increase the system capacity, and support mobility (i.e., IEEE 802.16e). However, the standards lack to indicate the receiver structure and the adaptation algorithms to be used for MIMO and AMC strategies. The study of these solutions is still an open issue. In addition, different and enhanced versions of MIMO and AMC should be investigated as they may help to further increase the system capacity and flexibility. For instance, the choice of implementing MIMO strategies based on spatial multiplexing or based on space time coding (STC) depends on the specific service requirements: high capacity or high reliability. The possibility to exploit both techniques, alternatively, may help to achieve both objectives. New diversity transmission schemes can be proposed by relaxing the orthogonal properties in order to achieve power balancing at the transmitter, as in the case of Alamouti transmission with more than two antennas. Moreover, new STC schemes with higher coding rate are suitable for increasing capacity or serving an higher number of users in a multiuser downlink system. In general, Alamouti has been demonstrated to be the unique full-rate full-orthogonal complex modulation STC. A maximum rate of 3/4 can be achieved in the case of 3 or 4 antennas. An orthogonal code with five or more antennas has a maximum rate of 1/2. Quasi-orthogonal codes achieve full rate transmissions and hybrid schemes achieve rates greater than 1 by adding spatial gain. Thus, BS with more than four antennas can select transmission modes from a wide range of possibilities. From this point of view, algorithms that are able to switch between the different diversity schemes, as the ones shown in Fig. 13.8, are necessary. These algorithms can operate by using either PHY information, such as channel state information, or MAC layer information, such as service type and queue status.

13


Burst Composer

Diversity scheme

Diversity decision algorithm

287



to RF stages



to RF stages

CSI from SS Qos and Mac param

from RF stages

from RF stages

CP removal and FFT

CP removal and FFT

Carriers De-permutation Diversity receiver

RX Burst Detection

Channel Estimation

TX Burst Composer

Carriers De-permutation

Fig. 13.8 Adaptive diversity scheme

As explained in Section 13.2 concerning the link adaptation strategies, power control, modulation and coding selection, and subchannel allocation are three key elements in the design of an efficient and flexible multiuser OFDM system. The adaptation algorithms that efficiently assign the resources (power, subcarrier and AMC scheme) must be defined and represents an important open issue.

13.10 Advanced Modulation and Coding Different schemes and techniques can be implemented to match transmission parameters to time-varying channel conditions and to meet QoS requirements. The aim is to maximize the system performance in terms of some QoS metrics, with a particular attention to error probability and throughput. In [16], two adaptive modulation techniques for the WiMax system are proposed. The first technique, called Maximum Throughput (MT), aims at maximizing the system throughput without any explicit constraint on target SER (Symbol Error Rate). The second technique, called Target SER (TSER), aims at guaranteeing a given maximum target Symbol Error Rate (SER) that is imposed on the basis of a target QoS level. The target value of the SER can be fixed for every SNR or can vary with it. In the second case, the modulation can be adapted in order to meet the

288

I. Cerutti et al. α

Current Technology Developments of WiMax Systems

Current Technology Development Of Wimax Systems

WIMAX New Developments

Current developments in bioerosion

WiMAX: A Wireless Technology Revolution

Current Developments in Mathematics 1998

WiMAX: Technology for Broadband Wireless Access

WiMAX-MobileFi: advanced research and technology

Magnetospheric Current Systems

New developments in parsing technology

Fundamentals of WiMAX

The Business of WiMAX

The Business of WiMAX

New Developments in Parsing Technology

Joint replacement technology, New Developments

Current Developments in Solid-state Fermentation

WiMAX Handbook

Mobile WiMAX

Developments in the Design of Thermal Systems

Autonomous Systems: Developments and Trends

Current developments of green IPPs: experiences, challenges, and strategies

Flexible Manufacturing Systems: Recent Developments (Manufacturing Research and Technology) (Manufacturing Research and Technology)

Lightwave Technology: Telecommunication Systems

Intelligent Support Systems Technology

Mobile WiMAX: A Systems Approach to Understanding IEEE 802.16m Radio Access Technology

Lightwave Technology Telecommunication Systems

Technology Systems and Management

Mobile WiMAX: A Systems Approach to Understanding IEEE 802.16m Radio Access Technology

New Nanotechnology Developments (Nanotechnology Science and Technology)

New Developments and Applications in Sensing Technology

Current Technology Developments of WiMax Systems

Current Technology Development Of Wimax Systems

WIMAX New Developments

Current developments in bioerosion

WiMAX: A Wireless Technology Revolution

Current Developments in Mathematics 1998

WiMAX: Technology for Broadband Wireless Access

WiMAX-MobileFi: advanced research and technology

Magnetospheric Current Systems

New developments in parsing technology

Fundamentals of WiMAX

The Business of WiMAX

The Business of WiMAX

New Developments in Parsing Technology

Joint replacement technology, New Developments

Current Developments in Solid-state Fermentation

WiMAX Handbook

Mobile WiMAX

Developments in the Design of Thermal Systems

Autonomous Systems: Developments and Trends

Current developments of green IPPs: experiences, challenges, and strategies

Flexible Manufacturing Systems: Recent Developments (Manufacturing Research and Technology) (Manufacturing Research and Technology)

Lightwave Technology: Telecommunication Systems

Intelligent Support Systems Technology

Mobile WiMAX: A Systems Approach to Understanding IEEE 802.16m Radio Access Technology

Lightwave Technology Telecommunication Systems

Technology Systems and Management

Mobile WiMAX: A Systems Approach to Understanding IEEE 802.16m Radio Access Technology

New Nanotechnology Developments (Nanotechnology Science and Technology)

New Developments and Applications in Sensing Technology

Recommend Documents