MIMO-Aware Medium Access Control in IEEE 802.11 Networks

MIMO-aware Medium Access Control in IEEE 802.11 Networks

Abduladhim Mabruk Ashtaiwi

A thesis submitted to the Department of Electrical and Computer Engineering in conformity with the requirements for the degree of Doctor of Philosophy

Queen’s University Kingston, Ontario, Canada January 2009

ISBN:978-0-494-48492-0

Abstract

Wireless Mesh Networks (WMNs) are dynamically self-organized and self-configured, where the nodes in the network automatically establish an ad hoc network and maintain mesh connectivity. These properties make WMNs a key technology for next- generation wireless networking. However, supporting Quality of Service (QoS) to enable multimedia services is still one of the major issues in next-generation WMNs. In distributed systems like WMNs, the Medium Access Control (MAC) layer is considered very important in the IEEE 802.11-based wireless networks, as it supports many crucial operational functions. Hence, QoS support in WMNs can be enhanced through the efficient cross-layer design of MAC protocols that utilizes advanced physical layer technologies viz Multiple-Input Multiple-Output (MIMO) with its multiple spatial channels that are capable of simultaneous receive or transmit streams. MIMO has become a very attractive technology in providing support for different QoS requirements. In this thesis we propose a novel QoS MIMO-aware MAC Protocol (QMMP). QMMP is a MAC protocol framework that exploits the MIMO system gains to boost QoS support. The proposed MAC framework includes the following components. The first component enables concurrent sharing of the increased MIMO bandwidth, i.e., instead of allocating all the spatial channels to one connection, connections can i concurrently share the increase bandwidth via splitting the spatial channels. The second component reduces the medium access collisions problem. In distributed systems like WMNs, medium access collisions have a noticeably negative impact on resource (bandwidth) utilization as they leave the bandwidth unutilized for a long time. To address this problem, we propose a spatial channels sharing scheme during medium contention period. The third component boosts the bandwidth utilization during data transmission. We propose resource management schemes that adapt the physical data rate and the aggregation frame length according to the instantaneous channel quality. Then we propose a QoS-aware bandwidth provisioning mechanism that performs effective bandwidth distribution to further boost QoS support.

ii Dedication

I wish to dedicate this thesis to the memory of my mother, Fatma. She was a constant source of inspiration in my life. Although she is not here to give me strength and support, I always feel her presence, urging me to strive to achieve my goals in life. May Almighty Allah reward her good deeds. To my father, Mabruk, who taught me that even the largest task can be accomplished if it is done one step at a time. To my wife, who stood beside me and encouraged me constantly. To my children, Saja, Sara, and Nada for giving me happiness.

iii Acknowledgments

First of all I would like to express my countless Praise to ALLAH for His graciousness and guidance, without which I could never have made it to the end. I would like to express my sincere gratitude to my supervisor Dr. Hossam S. Hassanein, for his patience and guidance throughout the course of this work. Special thanks to the external examiner, Dr. Wessam Ajib for his constructive and useful comments that helped improve the quality of this thesis. I would also like to thank the members of the examination committee, Dr. M. Ibnkahla, Dr. M.H. Rahman, Dr. P.K. Jain, and Dr. G.K. Takahara, for their valuable remarks and recommendations. Thanks are due to my colleagues Bader Manthari, Ashraf Ali Bourawy, Waleed Alsalih, Afzal Mawji, Khaled Ali, and Hassan Ahmad for their valuable collaboration and assistance during the course of this work. My deepest gratitude goes to my family for their unﬂagging love and support throughout my life. Finally, my sincere thanks to my friends and colleagues at the Telecommunication Research Lab (TRL) at Queen’s University for their guidance and friendship. I am also grateful to the numerous individuals who have directly or indirectly contributed to the completion of this work. iv Contents

Abstract i

Dedication iii

Acknowledgments iv

Contents v

List of Tables viii

List of Figures ix

List of Acronyms xi

List of Symbols xvi

1 Introduction 1 1.1 Motivation and Objectives ...... 3 1.2 Thesis Contributions ...... 6 1.2.1 QoS MIMO-aware MAC Protocol ...... 7 1.2.2 MIMO-aware Collision Avoidance Enhancements ...... 8 1.2.3 MIMO-aware Bandwidth Utilization and QoS Support . . . . 8 1.3 Thesis outline ...... 9

2 Background and Framework Overview 10 2.1 IEEE 802.11 Standard–Wireless Local Area Networks ...... 11 2.2 IEEE 802.11 Amendment Activities ...... 13 2.3 IEEE 802.11e–MAC Quality of Service Enhancements ...... 15 2.4 Multiple Input Multiple Output (MIMO) ...... 17 2.5 IEEE 802.11n–Enhancements for Higher Throughput ...... 18 2.6 IEEE 802.11s–Wireless Mesh Networks ...... 20 2.7 Proposed Framework and Related Work ...... 22 v 2.7.1 MIMO-aware MAC Protocol ...... 23 2.7.2 MIMO-aware Collision Avoidance ...... 28 2.7.3 MIMO-aware Bandwidth Utilization ...... 33 2.8 Summary ...... 35

3 QoS MIMO-aware MAC Protocol 37 3.1 Introduction ...... 37 3.2 Motivation and Problem Formulation ...... 38 3.3 MIMO Channel Interference Model ...... 40 3.4 QoS MIMO-aware MAC Protocol (QMMP) ...... 44 3.4.1 TXOP Scheduling and Broadcasting ...... 51 3.5 QMMP Properties ...... 58 3.6 Discussion of other QMMP Advantages ...... 60 3.7 Performance Evaluation ...... 66 3.7.1 Simulation Model ...... 66 3.7.2 Traﬃc Model ...... 69 3.7.3 Simulation Parameters ...... 69 3.7.4 Performance Metrics ...... 70 3.7.5 Simulation Results ...... 70 3.8 Summary ...... 78

4 MIMO-aware Medium Access Collision Avoidance 80 4.1 Introduction ...... 80 4.2 Enhancements to the IEEE 802.11e EDCF Collision Avoidance Mech- anism ...... 82 4.3 Performance Evaluation ...... 91 4.3.1 Simulation Model ...... 91 4.3.2 Network Topology ...... 92 4.3.3 Simulation Parameters ...... 92 4.3.4 Performance Metrics ...... 93 4.4 Simulation Results–Single-hop Network ...... 93 4.5 Adaptive γ ...... 99 4.6 Simulation Results–Multi-hop network ...... 110 4.7 Summary ...... 112

5 MIMO-aware Bandwidth Utilization 114 5.1 Introduction ...... 114 5.2 IEEE 802.11n MAC-Layer Frames ...... 116 5.2.1 A-MSDU Aggregation Frame Structure ...... 116 5.2.2 A-MPDU Aggregation Frame Structure ...... 117 5.2.3 A-MSDU and A-MPDU Two-level Frame Aggregation Structure 118 vi 5.3 Exploiting 802.11n Capabilities to Support QoS in IEEE 802.11s . . . 119 5.3.1 Link Adaptation ...... 120 5.3.2 Aggregation Frame Length Adaptation ...... 122 5.3.3 Bandwidth Provisioning Scheme ...... 124 5.4 Performance Evaluation ...... 126 5.4.1 Simulation Model ...... 126 5.4.2 Network Topology ...... 127 5.4.3 Traﬃc Model ...... 128 5.4.4 Performance Metrics ...... 128 5.4.5 Simulation Results ...... 129 5.5 Summary ...... 136

6 Conclusions and Future Work 138 6.1 Summary of Contributions ...... 139 6.2 Future Research Directions ...... 143

Bibliography 146

vii List of Tables

2.1 VoIP Codecs ...... 23

3.1 The physical and MAC configuration attributes of the IEEE 802.11n and traffic specifications ...... 39 3.2 IEEE 802.11n configurations ...... 69

5.1 Channel modulation parameters ...... 121 5.2 Traﬃc speciﬁcations ...... 128

viii List of Figures

1.1 Wireless mesh network evolution ...... 2

2.1 Snapshot of 802.11 physical and MAC standardization activities . . . 14 2.2 The structure of the sounding frame ...... 20 2.3 Wireless mesh network’s components and hierarchy structure . . . . . 21 2.4 The basic structure and operation of the proposed MAC in [50] . . . 25

3.1 Average medium access delay versus transmission delay comparison of different access classes...... 40 3.2 Interference channel model ...... 41 3.3 Stage 1: depiction of the QMMP main phases, deferral, and concurrent transmissions ...... 46 3.4 Stage 2: depiction of the QMMP main phases, deferral, and concurrent transmissions ...... 47 3.5 The QMMP scheme under hidden node problem ...... 49 3.6 Finding a slot in case there exist previously reserved slots ...... 54 3.7 Some functional aspects of the QMMP MAC protocol ...... 63 3.8 Instantaneous signal decode-ability with ϕ=4 ...... 68 3.10 Achievable throughput as a function of communication and interference distance ...... 72 3.11 Medium access delay for different ranges ...... 73 3.12 Throughput for different ranges of requested rates ...... 74 3.13 Medium access delay for different values of D ...... 75 3.14 Throughput for different values of D ...... 75 3.15 Wireless mesh network model ...... 76 3.16 performance comparison between the IEEE 802.11 DCF and QMMP MAC protocol ...... 77 3.17 Comparison of requested and achieved rates ...... 78

4.1 The sounding frames transmission window ...... 83 4.2 The synchronization of the sounding frames transmission windows . . 85 4.3 Medium contention at transmitters ...... 87 ix 4.4 Medium contention at receivers ...... 90 4.5 Hidden node network topology ...... 92 4.6 Mean number of attempts per packet for different γ and network loads. 94 4.7 M-EDCF MAC delay for different γ and network loads...... 95 4.8 Channel utilization for different γ and network loads...... 96 4.9 Throughput for different γ and network loads...... 97 4.10 Actuating versus silencing the selection and termination properties . . 98 4.11 The effect of changing γ ...... 100 4.12 Virtual transmission period...... 103 4.13 γ variability ...... 104 4.14 The success probability as function of different values of γ and β . . . 107 4.15 Multi-hop network performance for different value of γ ...... 112

5.1 A-MSDU aggregation frame structure ...... 116 5.2 A-MPDU aggregation frame structure ...... 117 5.3 Two level frame aggregation ...... 119 5.4 Wireless mesh network model ...... 127 5.5 Medium access delay for different access classes using different aggregation techniques...... 130 5.6 Throughput comparison of different access classes using different aggregation techniques...... 131 5.7 Packet drop ratio of different access classes using different aggregation structures ...... 131 5.8 The allocated transmission opportunity difference between link adaptation (LA-T XOP j) and using fixed physical transmission rate (FR- T XOP j)...... 132 5.9 Medium access delay versus increasing network load using two-level aggregation structure ...... 133 5.10 The total throughput of all access classes with and without using aggregation technique ...... 134 5.11 The MAC delay versus increasing voice network traffic load with and without using aggregation technique ...... 135 5.12 The achieved throughput for increasingly voice network traffic load . 136

x List of Acronyms

A-MPDU Aggregated MAC Protocol Data Unit

A-MSDU Aggregated MAC Service Data Unit

ACK Acknowledgement

AGC Automatic Gain Control

AIFS Arbitration InterFrame Space

AP Access Point

APB Acknowledged Packet Bitmap

BF Backoﬀ

BG background

BHT Backhaul Tier

BPSK Binary Phase Shift Keying

BSA Basic Service Area

CFP Contention-Free Period

CP Contention Period

CSMA/CA Carrier Sense Multiple Access with Collision-Avoidance

CSP Channel Sounding Phase

CSQ Channels Sounding reQuest

CSR Channels Sounding Response

CT Coordination Time xi CTS Clear To Send

CW Contention Window

DCCA Dual Channel Collision Avoidance

DCF Distributed Coordination Function

DIDD Double Increment Double Decrement

DISF DCF Interframe Space

DSMA-S Double Sense Multiple Access-Single

DSSS Direct Sequence Spread Spectrum

EBA Early Backoﬀ Announcement

EDCA Enhanced Distributed Channel Access

EDCF Enhanced Distributed Coordination Function

EQM Equal Modulation

FCR Fast Collision Resolution

FHSS Frequency Hopping Spread Spectrum

FTP File Transfer Protocol

GPS Global Positioning System

HC Hybrid Coordinator

HCCA HCF Controlled Channel Access

HCF Hybrid Coordinator Function

LANs Local Area Networks

LTS Long Training Sequence

M-EDCF MIMO-aware EDCF

MA-MAC MIMO Aware-MAC

MAC Medium Access Control

xii MACA-BI Multiple Access Collision Avoidance By Invitation

MANs Metropolitan Area Network

MAP Mesh Access Point

MCP Medium Contention Phase

MCS Modulation and Code Scheme

MCT Mesh Clients Tier

MFS MIMO-based Frame Scheduling

MMSE Minimum Mean Squared Error

MP Mesh Point

MPDU MAC Protocol Data Unit

MPP Mesh Point Portal

NACAM Neighbor Aware Collision Avoidance MAC

NAV Network Allocation Vector

NiQ Node in Question

OFDM Orthogonal Frequency Division Multiplexing

OSI Open System Interconnection

OSUC Ordered Successive Cancellation

PAB Proposed Antenna Bitmap

PANs Personal Area Networks

PC Point Coordinator

PCF Point Coordination Function

PDU Protocol Data Unit

PIFS Point Interframe Space

QAP QoS-Access Point

xiii QBSS QoS Basic Service Set

QMMP QoS MIMO-aware MAC Protocol

QoS Quality of Service

QPSK Quadrature Phase Shift Keying

QSTA QoS Station

RA Reservation Attributes

RAF Rate Adaptive Framing

RF Radio Frequency

RIFS Reduced Interframe Space

RT Reservation Attributes

RTS Request to Send

SCCMA Spatial Channel Control Multiple Access

SCMA Stream Control Medium Access

SEEMesh Simple, Eﬃcient and Extensible Mesh

SF Sounding Frame

SI Service Interval

SIFS Short InterFrame Space

SIG Signal ﬁeld

SNR Signal to Noise Ratio

STBC Space Time Block Code

STS Short Training Sequence

SUC Successive Cancellation

TC Traﬃc Class

TID Traﬃc ID

xiv TP TXOP Phase

TSBP TXOP Scheduling and Broadcasting Phase

TSPEC Traﬃc Speciﬁcations

TXOP Transmission Opportunity

UEQM Unequal Modulation

UWB Ultra-Wideband

Vi Video

Vo Voice

VoIP Voice over Internet Protocol

WLANs Wireless Local Area Networks

WMS WLAN mesh services

ZF Zero Forcing

xv List of Symbols

nt The number of transmit antenna elements.

nr The number of receive antenna elements.

th βi The SFs transmission window used by the i node in the NiQ vicinity.

qi The Slot i at which the anith antenna transmits its sounding frame. oγ The optimal SFs transmission window length of the NiQ.

Ce The estimated number of active nodes in the NiQ vicinity.

tv The virtual transmission length.

pd The probability that a node transmits in a ξ slot.

th fi The i antenna element index.

wi The weight factor that is assigned to a ﬂow.

th mi The mean MSDU frame size of the i ﬂow.

ni The average number of arrival packets during a SI interval.

ra The available rate or bandwidth during a slot(Mbps).

th vi The i connection pair.

Hd The channel coeﬃcients matrix from/to the desired transmitter.

CT The maximum error-free rate that the channel can support.

Ci The partial used bandwidth by vith connection pair.

Ut The total currently used spatial channels.

th Ig The generated interference by the i connection pair. xvi Pe The probability of a bit error rate.

Pf The probability of having f error bits. vg The total active antenna elements in the NiQ vicinity. ˜ It The amount of tolerated interference.

˜e The end time of a TXOP slot.

˜s The start time of a TXOP slot. n˜t The number of active antenna elements per active node. ϕ The degree of freedom.

ξ The MAC layer slot length. n The total number of network nodes.

λ The packets mean arrival rate.

γ The SFs transmission window length of the NiQ.

I The number of consecutive idle slots during each tv interval. α The exponential moving average parameter. z The total number of antennas in the NiQ vicinity including the NiQ. v The total number of active antennas in the NiQ vicinity. m The number of antennas that may transmit in a generic slot qi. D The distance between the transmitter and its intended receiver.

R The distance between the interfering connection pair and the NiQ.

η The path loss exponent.

Ψ The number of connections that have already reserved a TXOP slot.

CWmin The contention window minimum.

CWmax The contention window maximum. rphy The physical layer data rate (Mbps).

xvii th txopi The allocated transmission opportunity of the i connection pair.

th ani The i antenna element of a node. ntg The number of active antenna elements of the NiQ.

AFl The optimal aggregation frame length.

i th Ua The current spatial streams used by the i connection pair.

i th HI The channel coeﬃcient matrix from/to the i interfering connection.

Ps(m) The probability that m antennas will transmit in a generic slot.

xviii Chapter 1

Introduction

IEEE 802.11-1997 standard [27] for Wireless Local Area Networks (WLANs), addresses local area networking where the connected devices communicate over the air to other devices that are within proximity to each other. The IEEE 802.11 WLAN standard covers both the Medium Access Control (MAC) and the physical layers. With the gaining popularity of wireless laptops and other devices, WLANs have become popular due to ease of installation and location freedom . Public businesses such as coﬀee shops or malls have begun to oﬀer wireless access to their customers, some are even provided as a free service. Large wireless network projects are being set up in many major cities. Google is even providing a free service to Mountain View, California, and has entered a bid to do the same for San Francisco [18]. WLANs rely on single-hop connectivity, whereby the wireless Access Points (APs) are back-hauled through wired connections. Each AP is connected to a number of Stations (STAs) to provide Internet access service as shown in Figure 1.1(a). Industrial and standardizations venues are currently growing into a new type of wireless network, namely Wireless Mesh Networks (WMNs) [29]. Then IEEE 802.11

1 CHAPTER 1. INTRODUCTION 2

Task Group “S” extends the IEEE 802.11 MAC standard by deﬁning architectures and protocols that support self-organized and self-conﬁgured networks. A characteristic of WMNs is that both back-haul and the access connectivity rely on wireless interfaces. The WMN’s back-haul structure is composed of Mesh Points (MPs), i.e., nodes can support mesh services such as mesh path selection and forwarding. A mesh point that is collocated with portal functionality (Mesh Point Portal (MPP)) also supports inter-network forwarding, i.e., mesh to portal, enabling frames to enter and exit the mesh. A mesh point that is collocated with an AP (Mesh Access Point (MAP)) supports mesh to/from AP frame forwarding as shown in Figure 1.1(b).

External network External network

Mesh Portal

Portal Mesh Point MP Access Point AP MP AP Mesh Links AP MP Mesh AP AP MP STA AP Station STA STA STA STA STA STA STA Non-mesh Stations STA

(a) IEEE 802.11 deployment model (b) Wireless mesh network deployment model

Figure 1.1: Wireless mesh network evolution

With the capability of self-organization and self-conﬁguration, WMNs can be deployed incrementally, one node at a time, as needed. As more nodes are installed, the relia- bility and connectivity available for users increases accordingly. WMNs have become a promising wireless technology for numerous applications, e.g., broadband home networking, community and neighborhood networks, enterprise networking, building CHAPTER 1. INTRODUCTION 3

automation, etc. The IEEE standard for WMNs is in its final phase and is expected to be released in November 2008 [29]. Commercial deployments of WMNs are currently made in many cities. For example, in 26 Aug 2008, the City of Moncton, New Brunswick, started providing free Wi-Fi access to its downtown area via implementing wireless mesh networking. The city of Thorold, Ontario, also covers 32 square miles with a WMN to provide inexpensive Internet access. However, to achieve the objectives of WMNs more challenges exist. One of which is the scarcity of link capacity. Overcoming such challenge requires a deviation from the traditional Single-Input Single-Output (SISO) transmission system that impede achieving higher data rates. Achieving higher link capacity requires considering more advanced physical layer techniques such as Multiple-Input Multiple-Output (MIMO) technologies. Another challenge is the lack of an efficient cross-layer design of a MAC protocol that takes into account the wireless link properties and coding and modulation techniques of physical layer technologies such as MIMO. Indeed, QoS support in WMNs can further be enhanced by designing efficient cross-layer MAC protocols that effectively utilize the enhanced communication systems of MIMO technology.

1.1 Motivation and Objectives

As the demand for WMNs with multimedia services support increases, utilizing more advanced physical layer techniques that oﬀer higher data rates, e.g. MIMO, becomes more important. MAC protocols, viz the IEEE 802.11 MAC, are essential as they provide a variety of functionalities such as channel allocation procedures, protocol data unit (PDU) addressing, frame formatting, error checking, and fragmentation and CHAPTER 1. INTRODUCTION 4

reassembly. MAC protocols implement such mechanisms using different interaction schemes such as inserting Interframe Spaces (IFS)–idle time intervals between frame transmissions, backoff, and exchanging control frames. According to MAC operations, we can divide the time into: 1) Coordination Time (CT) (i.e., the aggregated time the MAC layer spends in inserting IFSs, performing backoff, and exchanging control frames), and 2) Service-Time (ST), time during which the authorized connection pair start the data transmission phase. At the same time, the advancements in the physical layer techniques (for example MIMO systems) increasingly enable higher data rate links. Likewise, the advancements in codec technologies enable the support of applications with a lower required data rate. However, as the physical layer and codec technologies advance further, the difference between the achieved data rate by the physical layer and required data rate of application increase. On the one hand, a higher available data rate is good because it shortens the service-time and gets the system to quickly switch to serve the next pending application. On the other hand, in distributed systems, the medium access coordination consume a considerable amount of time required for medium coordination procedures. Thus, as the data rate (bandwidth) difference between the available and the required is increasing, the medium access coordination increasingly becomes the performance bottleneck. Hence, there is a need for bandwidth sharing MAC protocols where connection pairs can coordinate to concurrently use the higher achieved data rat of MIMO systems. Developing concurrent bandwidth sharing schemes, therefore, can boost the system performance in terms of the overall system usability and medium access delay. One of the most performance degrading factors of the medium access coordination CHAPTER 1. INTRODUCTION 5

is collisions, i.e., multiple stations accessing the medium at the same time. Unlike centralized systems, in distributed systems like WMNs, collisions have a prominent negative affect on the system performance as for each collided packet, the contending node doubles its backoff window size which leave the bandwidth unutilized for long intervals. The bandwidth wastage worsens when the collision occur during the data transmission period. In this case, the wastage includes (a) the IFS coordination idle intervals inserted between frame transmissions, (b) the channel contention phase wastage resulting from using backoff intervals, (c) the channel reservation phase wastage due to the time spent in exchanging control frames, and (d) the data transmission phase wastage resulted from sending at least one full frame before knowing that there was a collision. Hence, in order to enhance the multimedia services support, efficient cross-layer MAC protocols that exploit the characteristics of the physical layer to avoid medium access collisions is essential for the operation of distributed wireless systems such as WMNs. Increasing the system usability requires tight control of the available bandwidth. To address this challenge, enhanced bandwidth utilization schemes are required during the service-time specifically by reducing the following overheads: 1) the amount of time wasted in interframe spacing overheads used by the IEEE 802.11 MAC protocol to provide a required period of inactivity between frame transmissions. 2) The physical and MAC header overheads. Both overheads occupy the channel and reduce the capacity available for frame transmission. Generally, aggregating multiple packets in larger frames minimize the per packet physical and MAC header overhead and the medium access coordination overhead. Unfortunately, using large aggregation frame lengths is limited by the channel error rate. Consequently, adapting the aggregation CHAPTER 1. INTRODUCTION 6

frame length according to the accepted error rate threshold (i.e., the number of error bits per packet that can be corrected by the receiver using a desirable correction technique) can reduce the overhead per packet to its minimum possible value. Adapting the physical data rate (i.e., by selecting proper modulation and coding schemes) according to the instantaneous channel quality can provide higher available bandwidth and hance reduce the channel error rate.

1.2 Thesis Contributions

In this thesis, we propose a MIMO-aware MAC protocol with QoS diﬀerentiation for WMNs, which consists of three novel approaches to simultaneously achieve the following objectives:

1. Enhancing the system usability via sharing the high achieved data rate of MIMO technology.

2. Designing an eﬀective MAC protocol that utilize MIMO system gains.

3. Enhancing bandwidth utilization via reducing the physical and MAC layers overhead.

4. Devising QoS-diﬀerentiation schemes.

5. Alleviating medium access collisions and the hidden node problem.

The main contribution of this thesis is the design of a comprehensive framework for medium access with provisions for eﬀective bandwidth utilization. The framework includes the design of a QoS MIMO-aware MAC protocol that implements concurrent bandwidth sharing schemes, a MIMO-aware medium access collision avoidance CHAPTER 1. INTRODUCTION 7

scheme, and MIMO-aware bandwidth management schemes. Since IEEE 802.11 is the dominant MAC protocol in WMNs, our proposed techniques are based on the IEEE 802.11 standard. Indeed we devise enhancements to the IEEE 802.11e, IEEE 802.11n, and IEEE 802.11s protocols and demonstrate how they can be integrated to achieve improved performance. These proposals are further highlighted in the following sections.

1.2.1 QoS MIMO-aware MAC Protocol

In Chapter 3, we introduce the QoS MIMO-aware MAC Protocol (QMMP). QMMP is a distributed MAC protocol that enables nodes to locally cooperate with other nodes in their vicinities to share the high data rate of MIMO systems via sharing the spatial channels (streams). The core idea is based on first estimating the channel status and then translating the required data rate (required by upper layer applications) into spatial streams requirements. The spatial channels sharing proposed by QMMP allows multiple connection pairs to concurrently share the bandwidth. As the required rate and the available rate may differ, the QMMP enables immediate or delayed reservation of data transmission periods. Reserving the required bandwidth at different times (i.e., immediate or delayed) can also be exploited to support other services such as QoS enhancement, co-channel interference mitigation, and/or attaining a desirable data rate. The performance of QMMP under different scenarios for different values of requested rates, interference zones, and different communication environments shows very promising results, which can be exploited to achieve different system performance goals. CHAPTER 1. INTRODUCTION 8

1.2.2 MIMO-aware Collision Avoidance Enhancements

In Chapter 4, we introduce a novel MIMO-aware Enhanced Distributed Coordination Function (M-EDCF). M-EDCF is a MIMO-aware collision-avoidance scheme proposed to enhance the IEEE 802.11e EDCF collision-avoidance mechanism. The basic idea is the sharing of the multiple spatial channels of MIMO systems, during the medium contention period, to avoid medium access collisions. Spatial channels sharing mean that instead of accessing the medium using all the channels, a node uses only a subset of the available spatial channels. As the total concurrent spatial streams (channels) used are fewer than or equal to the spatial degree-of-freedom, receivers can still receive and decode the transmitted signals of multiple medium contenders. Overhearing multiple contenders can be exploited by the receivers to implement response coordination schemes to avoid other potential collisions.

1.2.3 MIMO-aware Bandwidth Utilization and QoS Support

To efficiently utilize the bandwidth and boost the QoS support in WMNs, in Chapter 5 we propose schemes that exploit the physical and MAC layer enhancements of the IEEE 802.11n amendment [30]. Particularly, during the service-time period, we propose that each connection pair incorporate the following enhancements: 1) based on the online link quality connection pairs must adapt a proper Modulation and Code Scheme (MCS) index (i.e., by utilizing the different data rates defined in the physical layer specifications), 2) based on the online link quality and the receiver error correction threshold, connection pairs must determine the aggregation frame length, 3) based on the determined aggregation frame length and the QoS requirements, connection pairs must adapt a proper aggregation frame type (i.e., either Aggregation-MAC CHAPTER 1. INTRODUCTION 9

Service Data Unit (A-MSDU), Aggregation-MAC Protocol Data Unit (A-MPDU), or A-MSDU and A-MPDU combined) that are deﬁned in the IEEE 802.11n MAC enhancements, and 4) during the packet packing process (i.e. inserting the packets in the aggregation frame) connection pairs must pack packets according to their QoS constraints using our proposed QoS bandwidth provisioning program.

1.3 Thesis outline

This thesis is organized as follows. Chapter 2 presents some background material and previous work that are necessary for understanding the discussions to follow. Chapter 3 introduces the QoS MIMO-aware MAC protocol that is designed to provide eﬃcient bandwidth management via sharing the spatial streams. Chap- ter 4 presents the MIMO-aware collision-avoidance enhancements to the IEEE 802.11e EDCF, which aims at mitigating common medium access collisions. Chapter 5 introduces the MIMO-aware bandwidth utilization scheme, which aims at reducing the physical and MAC layers overhead. Chapter 5 also presents the QoS optimization program that optimally schedules data for transmission according to their QoS requirements. Chapter 6 presents the conclusions drawn from the thesis and discusses possible future research directions. Chapter 2

Background and Framework Overview

In this chapter we present some background material and previous work that is necessary for understanding the proposed framework. First, we explain in detail the WLANs standard. Then we further explain the extension amendments of IEEE 802.11 WLANs that are related to our proposed MAC protocol. Following that with suf- ﬁcient impetus to our work, we provide an overview for our proposed MIMO-aware MAC framework. Related to our proposed MAC scheme, we outline the status quo in the literature and current standardization eﬀorts.

10 CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 11

2.1 IEEE 802.11 Standard–Wireless Local Area Net-

works

The IEEE 802.11-1997 is a set of standards for WLAN computer communication developed by the IEEE standards committee (IEEE 802) in the 5 GHz and 2.4 GHz public spectrum bands [27]. The IEEE 802.11 WLAN standard covers both the MAC and physical layers. The IEEE 802.11 standard specifies three kinds of physical layer techniques, which are an Infrared (IR) baseband, a Frequency Hopping Spread Spectrum (FHSS) radio and a Direct Sequence Spread Spectrum (DSSS) radio. The Basic Service Set (BSS) is the central building block of the IEEE 802.11 architecture. The BSS is defined as a group of stations that are under the direct control of a single coordination function (i.e., a Point Coordination Function (PCF) or Distrib- uted Coordination Function (DCF)). The PCF medium access-function relies on a Point Coordinator (PC) to perform polling, i.e., enabling polled stations to transmit without contending for the transmission medium. In this thesis we only concentrate on DCF medium access-function. The DCF functional operation is based on Carrier Sense Multiple Access with Collision-Avoidance (CSMA/CA). The carrier sensing is performed at both the air interface, referred to as physical carrier sensing, and at the MAC layer, referred to as virtual carrier sensing. Physical carrier sensing detects the presence of other IEEE 802.11 WLAN stations by detecting their relative signal strengths. A source station performs virtual carrier sensing by sending MAC Proto- col Data Unit (MPDU) duration information in the header of the control and data frames. Other stations in the BSS use the information in the duration field to adjust their Network Allocation Vector (NAV). The channel is considered busy if either the CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 12

physical or virtual carrier sensing mechanisms indicate the channel is busy. Priority access to the wireless medium is controlled through the use of Interframe Space (IFS) time intervals between the transmission of frames. The IFS intervals are mandatory periods of idle time on the transmission medium. Three IFS intervals are specified in the IEEE 802.11 standard: Short IFS (SIFS), PCF IFS (PIFS), and DCF IFS (DIFS). The SIFS interval is the smallest IFS, followed by PIFS and DIFS, respectively. To ensure that multiple stations do not transmit at the same time, after the DIFS period the DCF function performs a Backoff (BF). Let ξ represent the MAC protocol slot length, BF be defined as a multiple number of ξ that is randomly selected, using the uniform distribution function, from the Contention Window (CW) range. The backoff counter is decremented only when the medium continue to be idle for ξ time interval and is frozen when the channel is sensed busy. Each time the medium becomes idle, nodes first wait for the DIFS interval and then resume decrementing one slot from the BF pool for each passed ξ idle time interval. Once the backoff timer counter reaches zero, the node is authorized to access the transmission medium. Once a frame is transmitted, the receiving station determines whether the frame was received correctly. Upon receipt of a correct frame, the receiving station waits a SIFS interval and transmits a positive Acknowledgment (Ack) frame back to the source station, indicating that the transmission was successful. To minimize the amount of bandwidth wasted when collisions occur, the IEEE 802.11 standard introduces optional Request To Send (RTS) and Clear To Send (CTS) control frames to be exchanged prior to the transmission of an MPDU frame. All stations in the BSS, hearing the RTS frame, read the duration field and set their CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 13

NAVs accordingly. Note that stations are capable of updating their NAVs based on the RTS from the source station and CTS from the destination station, which helps combat the hidden terminal problem.

2.2 IEEE 802.11 Amendment Activities

There are many amendments that extend the IEEE 802.11 standard to provide further functionalities. In this section, we overview the IEEE 802.11 amendment activities that are related to our proposed MAC protocol framework. Figure 2.1(a) shows the physical layer amendment extensions to the IEEE 802.11 standard. In 1999, the IEEE defined two high rate extensions: 1) IEEE 802.11b based on DSSS technology, with data rates up to 11 Mbps in the 2.4GHz band, and (2) IEEE 802.11a, based on Orthogonal Frequency Division Multiplexing (OFDM) technology, with data rates up to 54 Mbps in the 5GHz band [23]. In 2003, the IEEE 802.11g amendment extended the IEEE 802.11b physical layer to support data rates up to 54 Mbps in the 2.4GHz band. In response to growing market demand for higher performance WLANs, the IEEE standard association approved the creation of the IEEE 802.11 Task Group “N” (IEEE 802.11 TGn) during the second half of 2003. The scope of TGn’s objective is to define modifications to the physical and MAC layers to deliver a minimum of 100 Mbps throughput at the MAC Service Access Point (SAP) (i.e., ready to use throughput at the network and above layers). Figure 2.1(b) shows the MAC extension amendments of the IEEE 802.11 standard. To start with, the IEEE 802.11e-2005 amendment defines a set of QoS modifications to the MAC protocol to provide QoS differentiation. IEEE 802.11s is another defined extension to the IEEE CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 14

802.11 MAC protocol to support both broadcast/multicast and unicast delivery using self-conﬁguring and multi-hop topologies which are required to realize the WMNs’ objectives. The TGn deﬁnes an extension amendment to the MAC layer to accommodate for the increased throughput of the IEEE 802.11n physical layer. To understand the proposed MAC protocol framework, in the next sections (i.e., 2.3, 2.4, 2.5, and 2.6) we respectively explain in detail the IEEE 802.11e amendment, MIMO technology which is required to understand the IEEE 802.11n amendment, IEEE 802.11n, and IEEE 802.11s amendments.

802.11 PHY Layer 802.11 MAC Layer

Infra-Red(IR), 1/2Mps 802.11e QoS Enhancements

2.4 GHz FHSS

802.11g rates are 802.11b rates are 5.5 2.4GHz DSSS 9,12,18,22,24,33,36, 802.11n or 11Mbps 48, or 54 Mbps MAC Enhancements

802.11a rates are 5GHz OFDM 6,9,12,36, or 54 Mbps

802.11s 802.11n MIMO Service Set Extension

(a) The physical layer (b) The MAC layer

Figure 2.1: Snapshot of 802.11 physical and MAC standardization activities CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 15

2.3 IEEE 802.11e–MAC Quality of Service Enhance-

ments

IEEE 802.11e [28] is an enhanced version of the IEEE 802.11 MAC protocol to support QoS differentiation by assigning data traffic with different priorities based on their QoS requirements. The IEEE 802.11 amendment defines four different Access Classes (ACs). Access to the transmission medium is then granted based on the priorities of data traffic such that each AC first waits the differentiated Arbitration IFS (AIFS[AC]) and then performs the differentiated BF. In IEEE 802.11e, the AP and STA that provide QoS services are referred to as QAP (QoS Access Point) and QSTA (QoS Station), respectively, and the BSS they are operating in is called the QBSS (QoS Basic Service Set). IEEE 802.11e defines a new coordination function called the Hybrid Coordination Function (HCF). HCF is a centralized coordination function that combines the aspects of DCF and PCF with enhanced QoS mechanisms to provide QoS differentiation. HCF provides both distributed and centrally controlled channel access mechanisms similar to DCF and PCF in the IEEE 802.11 standard. The distributed, contention-based channel access mechanism of HCF is called Enhanced Distributed Channel Access (EDCA), and the centrally controlled, contention-free channel access mechanism is called HCF Controlled Channel Access (HCCA). The IEEE 802.11e amendment introduces the concept of the Transmission Opportunity (TXOP), defined as the time period during which a QSTA has the right to use the transmission medium. TXOP is characterized by a starting time and a maximum duration, called TXOP limit, which must not be exceeded by any stations. CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 16

As presented in the IEEE 802.11e reference scheduler, the initiation of AC connection requires the QSTA to first send a QoS request frame containing a Traffic Specification (TSPEC) to the QAP. The TSPEC describes the QoS requirements of a AC, such as mean/peak data rate, mean/maximum frame size, delay bound, and Maximum Service Interval (MSI). A maximum MSI refers to the maximum duration between the start of successive TXOPs that can be tolerated by a requesting application. Upon receiving all TSPEC information, the QAP scheduler first computes the selected Service Interval (SI), which should be the highest submultiple value of the beacon interval (TB) as follows, T SI = B , i = 1, 2,... (2.1) d TB e min(MSIi) th where MSIi is the maximum service interval of the i connection. Then the 802.11e

TB is divided into an integer number of SIs, and QSTAs are polled sequentially during each selected SI. This way, all the admitted ACs should be polled once within the delay requirement of the most time-stringent AC. Lastly, the QAP scheduler computes

the corresponding HCCA TXOP (txopi) values for diﬀerent QSTAs by using the QoS

requirements speciﬁed in their TSPECs. To further illustrate this, let ri represent the mean data rate required by the ith connection. It simply follows that the number of

frame arrivals ni during the SI interval is approximately computed as

ri × SI ni = d e, (2.2) mi th where mi is the mean MSDU frame size of the i connection. Given the estimated

number of frames, the QAP can compute the txopi that is going to be allocated to the ith connection as follows

ni × mi mmax txopi = max( + O, + O), (2.3) rphy rphy CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 17

where rphy is the physical layer data rate and mmax is the maximum MSDU size, and O represents the physical and MAC layer overheads. During the allocated TXOP

(txopi), the QAP polls the associated connection to start communication.

2.4 Multiple Input Multiple Output (MIMO)

Under suitable channel fading conditions, having both multiple transmit and multiple receive antennas (i.e., a MIMO channel) provides an additional spatial dimension for communication and yields a spatial degree-of-freedom gain. These additional degrees- of-freedom can be exploited by spatially multiplexing several data streams onto the MIMO channel, and lead to an increase in the capacity [57], [40], [16]. The performance improvements resulting from the use of MIMO systems are due to array gain, diversity gain, spatial multiplexing gain, and interference reduction gain. Array gain can be made available through processing at the transmitter and the receiver which results in an increase in average receive Signal to Noise Ratio (SNR) due to a coher- ent combining eﬀect. Transmit/receive array gain requires channel knowledge in the transmitter and receiver, respectively, and depends on the number of transmit and receive antennas. Diversity techniques rely on transmitting the signal over multiple independently fading paths. If the channel fade independently and the transmitted signal is suitably constructed, the receiver can combine the arriving signals such that the resultant signal exhibits considerably reduced amplitude variability. Extracting spatial diversity gain in the absence of channel knowledge at the transmitter is also possible through using suitably designed transmit signals. The corresponding technique is known as space-time coding in which time is complemented with the spatial dimension [19], [2], [71], [70]. The spatial multiplexing gain is realized by transmitting CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 18

independent data signals from the individual antennas. Under conducive channel conditions, rich scattering and transmit and receiver antennas spaced sufficiently apart, the receiver can separate the different streams, yielding a increase in capacity [56], [14], [72], [6]. Schemes that extracting both types of gains are proposed [77] [63]. The differentiation between the spatial signature of the desired and co-channel signals can be exploited to reduce co-channel interference. The interference reduction gain requires knowledge of the desired signal’s channel although the exact knowledge of the interferer’s channel may not be necessary. Interference reduction (or avoidance) can also be implemented at the transmitter, where the goal is to minimize the interference energy sent towards the co-channel users while delivering the signal to the desired user.

2.5 IEEE 802.11n–Enhancements for Higher Through-

put

In January 2004 the IEEE committee announced that is had formed a new 802.11 Task Group “N” (TGn) to develop a new amendment to exploit the MIMO technique for the 802.11 standard [30]. TGn’s goal is to achieve 100 Mbps throughput, after subtracting all the overhead for protocol management features. The IEEE 802.11n include physical and MAC layer enhancements. The physical enhancements include deﬁning 76 Modulation and Code Scheme (MCS) indexes for each channel bandwidth, i.e., 20 MHz and 40 MHz. The MCS indices specify the physical layer parameters that consist of modulation type (BPSK, QPSK, 16-QAM, 64-QAM) and coding rate

(1/2, 2/3, 3/4, 5/6), number of coded bits per single carrier NBPSC = {1, 2, 4, 6}, CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 19

number of spacial streams Nss = {1, 2, 3, 4}, and whether Equal Modulation (EQM) or Unequal Modulation (UEQM) per spatial streams is used. The IEEE 802.11n high throughput physical specification also defines two guard interval values GI = 400ns and 800ns. By combining different physical layer parameter values, the IEEE 802.11n can be configured to support up to 304 different data rates that range between 6.5 and 600 Mbps. The MAC layer enhancements include defining the following aggregation frame formats: 1) aggregation of MAC Service Data Units (MSDUs) at the top of the MAC (referred to as MSDU aggregation or A-MSDU), 2) aggregation of MAC Protocol Data Units (MPDUs) at the bottom of the MAC (referred to as MPDU aggregation or A-MPDU), and 3) A-MSDU and A-MPDU two-level frame aggregation structure, utilizing A-MSDU and A-MPDU aggregation techniques over two combining aggregation stages. To help receivers estimate the channel status and therefore recover the transmitted signals, the IEEE 802.11n amendment recommends transmitting a known communication setup frame, called Sounding Frames (SFs), through all transmit antennas. As defined in the IEEE 802.11n amendment, the SFs can be either be attached to the MPDU frames if no channels feedback information is required at the transmitter node or can be through two way handshake exchanges. In the latter case, the transmitter sends a Channels Sounding Request (CSQ) and the receiver responds with a Channels Sounding Response (CSR). The CSQ and CSR exchange is performed if the channels feedback information is required at both sides, i.e., the transmitter and the receiver nodes. During the CSQ period, the transmitter node concurrently sends one SF through each transmit antenna element. Then the receiver, after receiving CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 20

the CSQ frames, responds by concurrently sending one SF through each transmit antenna element of the receiver node, where the latter is done during CSR period. Figure 2.2 shows the structure of the sounding frame format. It consists of Short Training Sequence (STS), Long Training Field (LTS), and Signal Field (SIG). STS is used for synchronization and Automatic Gain Control (AGC), LTS is used for channel estimation. The number of LTS slots correspond to the number of channels that need to be estimated. The signal ﬁeld is used to communicate physical layer control information to the receiver. The average SF length is equal to STS + LTS + (nt×

LTS) + STG, where nt is the number of transmit antenna elements.

STS LTS1 LTSnt SIG

Figure 2.2: The structure of the sounding frame

2.6 IEEE 802.11s–Wireless Mesh Networks

To provide high-bandwidth network over extended coverage area, the IEEE 802.11 Task Group “S” (IEEE 802.11 TGs) extends the IEEE 802.11 MAC standard by defining architectures and protocols that to support self-configured networks, broadcast/multicast, and unicast frame delivery capability. Among other technologies, legacy IEEE 802.11, [27], has been used in implementing WMNs [29]. The IEEE 802.11s defines WMNs as a network consisting of two or more wireless mesh points in- terconnected via IEEE 802.11 links and communicating via the WLAN Mesh Services (WMS). Figure 2.3 shows the main WMN components; namely, the Mesh Point (MP), Mesh Access Point (MAP), Mesh Point Portal (MPP), and Mesh Clients (MCs). The CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 21

MP is a node that can support mesh services, e.g., mesh path selection and forwarding. An MP may be collocated with one or more other entities (e.g., Access Point (AP) or Portal). An MP collocated with an AP is referred to as MAP. Such a conﬁg- uration allows this entity to provide both mesh functionalities and AP functionalities. Similarly, a MP collocated with a Portal entity is referred to as an MPP, where its functionalities are to interface the WMNs to other IEEE 802 LAN segments. The MC is mesh entity that uses the MAPs to gain access to the mesh network services. MC entities include stations that can support mesh services and those which can not.

External Network

Back-haul channel Portal MP

MP MP AP AP Mesh-client channel

Mesh client

BSS1 BSS2

Figure 2.3: Wireless mesh network’s components and hierarchy structure

As shown in Figure 2.3, each MAP is equipped with two independent MAC protocols, which utilize diﬀerent channel frequencies, Back-Haul Tier (BHT) and Mesh Clients Tier (MCT) MAC. The BHT MAC utilizes the IEEE 802.11e EDCF for back- haul medium access coordination. On the other hand, MCT MAC utilizes the IEEE CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 22

802.11e HCCA medium access-function to coordinate the medium access among the MAP and its connected MCs. As explained in Section 2.3, the QAP, which is similar

th to the MAP in the IEEE 802.11-based WMN structures, computes the txopi of the i connection using Equation 2.3. Next, the MAP computes the total required TXOPj for all connections connected to this MAP, C X TXOPj = txopi,j j = {1, 2,...,Z}, (2.4) i=1 where C is the number of connections that are associated with the jth MAP and

Z is the number of back-haul MAPs. Given the TXOPj, the MAPs commence the back-haul tier MAC procedure for delivery to the next MP. Since the joint proposal (i.e., Simple, Eﬃcient and Extensible Mesh (SEEMesh) and “Wi-Mesh”) of WMN was accepted in January 2006, WMNs have undergone rapid progress which inspired numerous deployments to support large area coverage with low cost [48], [60].

2.7 Proposed Framework and Related Work

WMNs are dynamically self-organized and self-conﬁgured, where the nodes in the network automatically establish an ad hoc network and maintain the mesh connectivity. These properties make WMNs a key technology for next-generation wireless networking. However, supporting QoS to enable multimedia services is still one of the major issues in next-generation WMNs [74], [37]. In 802.11-based WLANs the MAC layer is often viewed as the “brain” of the wireless networks, as it provides a variety of functions that support the operation of the network. Therefore to enhance the QoS support in WMNs, an eﬀective cross-layer design of a capable MAC protocol that CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 23

eﬃciently utilizes MIMO gains is essential. In the next three sections we introduce our proposed contribution and related work in the literature.

2.7.1 MIMO-aware MAC Protocol

The IEEE 802.11n amendment achieves a data rate that is many times greater than the IEEE 802.11a/g amendments. At the same time, the advancements in encoding techniques increasingly allow the development of applications with lower required data rates. Table 2.1 shows an example of the required data rate diﬀerence for diﬀerent codec standards of the Voice over Internet Protocol (VoIP) calls [78].

Table 2.1: VoIP Codecs

Standard data rate

G.711 128 (Kbps)

G.729A 16 (Kbps)

G.723.1 10 or 12.8 (Kbps)

Allocating the increased bandwidth of MIMO systems to applications is desirable because it enables faster service-time and hence faster switching to the next pending application. For example, given the 100 Mbps as the minimum data rate supported by the IEEE 802.11n amendment, the allocated data rate for the VoIP call is hundreds of times higher than the required data rate. This higher allocated bandwidth shortens the service-time of the VoIP call. However, in distributed systems like WMNs, switching to the next call requires a considerable amount of time that is wasted in medium access and coordination mechanisms. As the bandwidth diﬀerence between CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 24

the available and the required increases, the medium access and coordination increasingly becomes the performance bottleneck. Hence, we believe that there is a fundamental need for bandwidth sharing schemes where connection pairs can concurrently use the high supported data rate of MIMO systems. Furthermore, designing concurrent bandwidth sharing MAC protocols can boost the system performance in terms of overall system usability and medium access delay [31], and [49]. In Chap- ter 3, we introduce a novel QoS MIMO-aware MAC protocol (QMMP). QMMP is MIMO-aware MAC protocol that eﬃciently utilizes the MIMO higher achieved data rate and channel quality by virtue of measuring the links and then translating the required data rate into spatial stream requirements. QMMP is a distributed MAC protocol that does not require node synchronization. It enables nodes to locally cooperate with other nodes in their vicinities to share the medium.

In the related literature, there have been few attempts to design new MIMO-aware MAC protocols to exploit the beneﬁts of MIMO technology. The work in [67] performs rate scheduling using a Stream Control Medium Access (SCMA) protocol, where light- loaded connections concurrently share the available rate and overloaded ones (belong to multiple contention area) use the full rate. Nodes in the SCMA protocol locally identify the bottleneck links by sensing over two hops using RTS and CTS and then color them, red or white. The red color represents overloaded pairs, whereas the white color represents light-loaded pairs. Lastly, it adapts a persistence parameter that is tuned to their colored links. The persistence is then used to perform spatial streams scheduling. A major limitation of the work in [67] is that scheduling according to vicinity’s traﬃc load is very hard to achieve specially in distributed systems where CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 25

the hidden node problem and collisions are the norm. Furthermore, scheduling over two hops can significantly decrease the network performance. Likewise, according to the proposed protocol, it is not clear how white links can coordinate to share the spatial streams without violating the physical layer limitation as outlined in [67]. A different approach is introduced in [50], where the proposed MAC protocol employs a spatial multiplexing with antenna subset selection for data transmission. It assumes that the nodes in the network are synchronized by using the Global Posi- tioning System (GPS). As shown in Figure 2.4, the MAC frame is divided into a negotiation period and a contention-free period. The negotiation period consists of two contention slots, in which the transmitters contend by exchanging RTS and CTS control frames. Nodes that successfully finish the reservation procedure can then concurrently start the data transmission phase.

RTS RTS Data Node 0

Collision CTS Node 1

RTS Data

Node 2

CTS

Node 3 Negotiation Contention-free period period

Figure 2.4: The basic structure and operation of the proposed MAC in [50]

Our proposed protocol is diﬀerent from [50] in that it is fully distributed scheme that CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 26

does not require synchronization. In addition, the proposed work in [50] assumes that connections are evenly sharing the antennas without any prior channel estimation mechanism which requires assessing the supported data rate of each link. The authors of [51] propose a MIMO based MAC protocol that utilizes antenna selection. The proposed MAC protocol extends the control frames, RTS, CTS, and Ack, respectively to M-RTS, M-CTS, and M-Ack. The extended frames are then used to perform negotiation about the active antennas, channel estimation, and the selection of MIMO coding techniques, i.e., multiplexing versus diversity. The M-RTS frame includes a field called Proposed Antenna Bitmap (PAB) which is used to encode the chosen subset of the available antennas proposed for the pending data transmission. Based on the post-processing SNR values for each antenna, the receiver confirms which antennas should be active in the Confirmed Antenna Bitmap (CAB) field of M- CTS frame. The M-Ack frame contains a one byte long bitmap called Acknowledged Packet Bitmap (APB) field which is used to indicate positive or negative acknowledgement. The communication procedure proceeds as follows. First the transmitter sends an M-RTS frame, setting 1 in the PAB field for the available antennas for the next transmission. Then the receiver responds with an M-CTS frame, setting 1 in the CAB field for the antennas accepted for transmission. After the reception of the M-CTS frame, the transmitter transmits (one or more) packets based on the instruc- tions of the receiver about the antennas to be used. After the reception, the receiver checks the received packets, and creates an M-Ack frame, setting 1 in the APB for each correctly received packet. When the transmitter receives the M-Ack frame, it removes the packets from the queue and initiate another transmission. If the M-Ack frame is lost, the transmitter will initiate a retransmission after a timeout. CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 27

MIMO-based MAC schemes proposed in [62] and [38] take into consideration the encoding-decoding delay incurred by MIMO systems and the current supported channel rate. To reduce the coding delay effect on the system performance, the authors of [62] do not consider using the control packets: RTS, CTS, and Ack, only allow data transmission on the medium. Collision-avoidance is accomplished by enlarging the carrier sensing range. The key contribution of [38] is presenting a MIMO physical layer aware and rate adaptive MAC protocol. The proposed scheme includes transmit antenna and data rate selection based on the optimal tradeoff between spatial multiplexing and diversity coding technique. The selection of antennas and the data rate are based on the instantaneous channel conditions at the receiver. A feedback mechanism is then used to convey preferred data rate to the transmitter. The proposed schemes in [62] and [38] do not conflict with our proposed QMMP. Rather, they can be used as an extension to provide antenna selection to further improve the system performance. The work in [11], [52], and [35] introduce MAC protocols based on utilizing beamforming, which is a signal processing technique [41] that changes the directionality of the array to create a pattern of constructive and destructive interference. The work in [11] proposes a MIMO-aware MAC (MA-MAC) scheme for scheduling two simultaneous transmissions within a single collision domain based on adjusting antenna weights to selectively listen to or ignore a particular transmission using beamforming. Nodes in the proposed MA-MAC scheme have two main operations to perform before data transmission. The first operation is adjusting their antenna weights according to previously collected information so that its transmission does not interfere with the ongoing transmission. The second important operation is disseminating the antenna CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 28

weights to all neighboring stations during their scheduled packet transmission. Using the broadcasted weights, nodes can coordinate to perform simultaneous transmissions using MA-MAC protocol. The authors of [52] present a MAC protocol termed NULL- HOC. The proposed MAC protocol coordinates the control frames exchanged, which are used to adjust the antennas weights to nullify (cancel) the interference of other users involved in existing communication sessions. The proposed scheme is based on using the knowledge of MIMO channels between antenna arrays at diﬀerent nodes to design transmit and receive signal structure that achieve a speciﬁed gain to the desired node while nulling existing communications between other nodes. Similarly, the work in [35] proposes Space-MAC, a MAC protocol. Space-MAC prevents interference between connection pairs by selectively nulling the signals from potentially interfering transmissions. This is done in a totally distributed fashion utilizing Chan- nel State Information (CSI) at both the transmitter and receiver to adjust antennas weights for each packet transmission. The negative aspect of [11], [52], and [35] is that they are all proposed for single-hop network. Additionally, their performance depends on the weight exchanges. Hence, any failure by any station to convey its weight information to other nodes can result in a collision that lasts for a complete TXOP period.

2.7.2 MIMO-aware Collision Avoidance

As stated earlier, in distributed systems like WMNs, collisions have a noticeably negative impact on resource (bandwidth) utilization. To reduce the bandwidth wastage, designing MAC protocols with efficient medium access collision-avoidance schemes are essential to effectively support QoS in wireless LANs. The random backoff adapted CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 29

by the IEEE 802.11 MAC cannot completely eliminate the collisions, since two or more stations may finish their backoff procedures simultaneously. As the number of contending nodes increases, the number of collisions is also likely to increase. The DCF adopts a binary exponential backoff by increasing the CW size exponentially upon each transmission failure in order to reduce consecutive collisions for each traffic class. However, in certain situations, this exponential backoff results in inefficient channel utilization. In Chapter 4, we introduce a novel MIMO-aware EDCF (M-EDCF) protocol. M- EDCF exploits the spatial channels of MIMO systems to enhance the IEEE 802.11e EDCF collision-avoidance mechanism. The basic idea is sharing of the spatial channels during medium contention periods to avert medium access collisions. Spatial channels sharing means that instead of accessing the medium using all spatial channels (streams), a node uses only a set of the available spatial channels. As the concurrent spatial channels used are fewer or equal to the spatial degree-of-freedom, receivers can detect multiple medium contentions. Spatial channels sharing has other properties such as medium contention termination and medium contention selection. The former refers to the possibility of terminating the ongoing medium access attempt if the accessing node detects multiple concurrent medium accesses. Contention selection refers to the possibility of detecting multiple concurrent medium access attempts initiated by different transmitters and hence they can coordinate their responses to avoid collisions. An optimization program that estimates the optimal spatial channels sharing value based on the network load variations is also proposed. CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 30

In the related literature, there have been numerous studies to improve the performance of DCF. The analytical model proposed in [5] demonstrates that the performance of DCF strongly depends on the minimum CW and the number of stations. The work in [7], [39], [32], [65], [45], [53], [36], [24], [8], and [68] propose schemes to avoid collisions based on modifying the CW size. The authors of [7] propose a dynamic and distributed algorithm which allows each node to estimate the number of competing nodes and to tune its CW to the optimal value during runtime. The work in [39] proposes Fast Collision Resolution (FCR), which actively redistributes the backoff timer for all competing nodes, hence allowing more recent successful stations to use a smaller CW, and allowing other stations to reduce the backoff timer exponentially when they continuously meet some idle time slots. The key contribution of [32] is proposing a distributed reservation based MAC protocol, called Early Backoff Announcement (EBA), where stations announce their future backoff information in terms of the number of backoff slots via the MAC header of the frame being transmitted. All stations receiving the information avoid collisions by excluding the same backoff duration when selecting their future backoff value. The work in [65], [24], [45], [53], and [36] propose adapting the contention widow size according to the packet collision rate, and channel load. The author of [26] proposes a Model-based Frame Scheduling (MFS) scheme to enhance the capacity of the IEEE 802.11 MAC protocol. In the MFS scheme, each node estimates the current network status by keeping track of the number of collisions it encounters between its two consecutive successful frame transmissions. Then, based on the estimated information, the station computes the current network utilization. The result is then used to determine a scheduling delay. To mitigate collisions caused by hidden terminals the CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 31

work in [69] proposes the MACA-BI ((Multiple Access Collision Avoidance)-By Invi- tation) scheme in which the data transmission is initiated by the receiver. It assumes that the receiver can predict its future reception time in a network with periodic data traffic. Our proposed collision-avoidance scheme is based on spatial channels sharing during the channel contention period. Hence, the existing schemes can be viewed as complementary to our proposed collision-avoidance enhancement scheme. Another class of medium access collision-avoidance proposals are based on the tight selection of the CW range length. In order to avoid potential future packet collisions, the authors of [8] introduce the concept of the Double Increment Double Decrement (DIDD) algorithm, which is based on gently and gradually increasing the CW size after a unsuccessful packet transmissions. In case of a successful packet transmission, the DIDD halves the CW instead of going back to CWmin. In [68] the authors claim that they had found the minimum optimal CW size which is equal to 8.5 * N -5, where N is the number of active nodes. The authors of [46] propose a modified backoff mechanism that uses a logarithmic increment instead of an exponential extension of the CW size in order to eliminate the degrading effect of a random number distribution. Different suggestions of medium access collision-avoidance schemes that are es- tablished on exploiting the dual channels mechanisms are proposed in [4], [25], [36], [20], [21], [12], and [22]. In order to efficiently avoid collisions in all cases of hidden/exposed receivers and senders, the authors of [4] propose Dual Channel Collision Avoidance (DCCA) which employs two channels, one channel for signaling and one channel for data transmission. The work in [20], [21], [12], and [22] have a similar medium access collision avoidance-approach that is based on exploiting the dual busy CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 32

tone multiple access which uses two out-of-band tones to decouple communications in two directions. The medium is divided into a control channel and a data channel. Data packets are transmitted over the data channel, while control packets (e.g., RTS and CTS) are transmitted over the control channel. The authors of [25] propose a MAC protocol called Double Sense Multiple Access-Single (DSMA-S) channel. In the DSMA-S scheme collision-avoidance is achieved by adapting mandatory-waiting on a single channel, that is realized by two out-of-band busy tone signals: BTc and BTr signals. The signals are used on the receiver side only. BTc indicates a packet collision at the receiver side, while the BTr signal indicates a successful transmission. Then both stations (the transmitter and receiver) check the status of both BTc and BTr signals upon their medium access attempt. If either the BTc or BTr signal is sensed, i.e. BTc= 1 or BTr = 1, the station defers its medium access attempt. If both BTc and BTr signals are not sensed, i.e. BTc = 0 and BTr = 0, the station sends a RTS packet. The work in [36] proposes a MAC protocol capable of collision detection based on using pulses in an out-of-band control channel for exploring the channel condition. As the control channel still suﬀers from multiple medium access, multiple channels MAC protocols do not provide higher performance [49], [34]. A higher number of collided packet retransmissions causes transmitters to quickly drain their power. The proposals in [17] and [64] base their collision-avoidance schemes on the power drainage level by adjusting the CW size according to the power level. The work in [17] proposes an energy-aware MAC enhancement of the IEEE 802.11 DCF. The scheme changes the deferring time before transmitting fresh data packets and also after an unsuccessful transmission attempt based on a normal distribution with mean and variance that depend on the current node’s battery level. CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 33

The authors of [64] propose a Neighbor Aware Collision Avoidance MAC (NACAM) protocol. The NACAM protocol deﬁnes algorithms to estimate minimum and maximum CW sizes, i.e., CWmin, CWmax, taking into account the number of neighbors in one-hop neighborhood and the energy level of the battery. We remark that none of the existing medium access collision schemes utilize MIMO techniques as a collision mitigation scheme in CSMA/CA based WLANs.

2.7.3 MIMO-aware Bandwidth Utilization

Supporting multimedia services in distributed systems (e.g., WMNs) imposes many challenges which require eﬃcient management of system resources. This includes enhancing the bandwidth utilization via sharing the increased bandwidth (i.e., via MIMO systems) as proposed in Section 2.7.1, reducing the bandwidth wastage that results from the medium access collisions as proposed in Section 2.7.2, and enhancing the bandwidth utilization during the reserved TXOP. The latter is the goal of our third contribution for this thesis. TXOP is the scheduled period during which a connection pair starts and ends data and Ack frame exchanges. The data packet Ack exchange may last for period where many packets are delivered from the transmitter to the receiver. In order to avoid collisions, for each data packet Ack exchange, the MAC protocol inserts IFS intervals and control frames. These IFS intervals and control frames must be considered a part of the protocol overhead as it occupies the channel and reduces capacity available for data transmission. In Chapter 5 we propose MIMO-aware bandwidth utilization schemes that are based on exploiting new physical and MAC enhancements of the IEEE 802.11n CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 34

amendment. Specifically, we propose that during the TXOP period, the connection pairs adapt their modulation and code scheme indexes (by utilizing the different data rates defined in the physical layer specifications), their aggregation frame lengths which determined according to the online assessed link quality, and their frame structure type (i.e., either A-MSDU, A-MPDU, or A-MSDU an A-MPDU combined aggregation frame). Then perform QoS bandwidth provisioning by optimally aggregating packets according to their QoS requirements while maintaining fairness. Exploiting frame aggregation technique helps reduce the MAC delay and increases the throughput, and consequently reduce the packet dropping ratio. In the related literature, there exist several attempts aiming at increasing the utilization of network resources. The work in [58], [75], and Ultra-Wideband (UWB) networks [44] proposed the introduction of a packet aggregation scheme. The technique trades off service time for packet length where the increase of MAC service time is mitigated by assembling multiple upper layer packets into a single MAC burst. In [43] an analytical model is developed in order to study the impact of packet aggregation on delay. The authors of [42] propose frame aggregation and optimal frame size adaptation for IEEE 802.11n WLANs. The adaptive aggregation shows higher performance than fixed frames sizes. However, the paper does not include details on how the frames are delayed. Also the model was developed and verified for single hop WLANs only. The work in [59] proposes an IP-based adaptive packet concatenation multi-hop wireless networks algorithm which is an end-to-end aggregation scheme. The basic idea behind this method is to select the packet size based on the route quality. The algorithm uses routing metrics to determine the Maximum Concatenation Interval CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 35

(MCI) and the Maximum Concatenation Size (MCS). Packets at the ingress node are delayed by MCI. The aggregated packet size cannot exceed MCS. The work in [9] introduces Rate-Adaptive Framing (RAF) that jointly controls the channel rate and frame size according to the observed interference patterns and noise level at the receiver. RAF uses the interference pattern at the receiver to determine the optimal channel rate and the frame size that achieves the maximum throughput. The receiver then communicates the conﬁguration of the channel rate and frame size to the transmitter in a few bits in the per-frame acknowledgement. The transmitter then applies the channel rate and frame size in the next frame transmission. In [15] the authors propose several performance optimizations aimed at improving the VoIP support in WMNs where header compression is exploited to improve the network capacity in terms of number of voice calls supported. Other attempts, e.g., [13] and [61], aim at increasing the system capacity by using multiple radios per node, where the radios assignments are based on schemes such as centralized, static, and dynamic channel assignments. Nevertheless, there have been no proposals that enhance the bandwidth utilization in IEEE 802.11s based on exploiting the IEEE 802.11n physical and MAC enhancements.

2.8 Summary

In this chapter we reviewed the IEEE 802.11 standard. The three main components on which our proposed MAC protocol is based are then discussed. These are the IEEE 802.11e (the Quality of Service (QoS) enhancements), IEEE 802.11n (the higher throughput enhancements), and IEEE 802.11s (the Extended Service Set (ESS) mesh CHAPTER 2. BACKGROUND AND FRAMEWORK OVERVIEW 36

networking enhancements). In this chapter we also outlined the proposed QoS MIMO- aware Medium Access Control Protocol (QMMP) MAC protocol in Wireless Mesh Networks (WMNs). The QMMP is designed as a comprehensive MAC protocol that include bandwidth sharing, collision avoidance, bandwidth management, and QoS- diﬀerentiation schemes. Related works in the literature were discussed for all proposed schemes. Chapter 3

QoS MIMO-aware MAC Protocol

3.1 Introduction

The advancements of physical layer technologies such as MIMO systems enable the support of higher data rates, i.e., higher than 100 Mbps. Similarly, the advancements of codec make applications with lower required data rates possible. In distributed systems bandwidth allocation consumes a very considerable amount of the time required for channel coordination. To alleviate bandwidth wastage, in this chapter, we introduce the QoS MIMO-aware MAC Protocol (QMMP). QMMP enables nodes to locally cooperate with other nodes in their vicinities to share the MIMO higher achieved data rate. The core idea is based on sharing the bandwidth via concurrent sharing of the spatial streams. This new capability introduced by the QMMP enables multiple concurrent communications. The remainder of this chapter is organized as follows. Section 3.2 explains the motivation behind introducing the QMMP scheme. Section 3.3 explains the MIMO

37 CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 38

channel interference model. The QMMP is discussed in detail in Section 3.4 including the procedure of proper TXOP scheduling and broadcasting. Section 3.5 presents some proven properties of the QMMP scheme. In Section 3.6 we discuss the advantages of QMMP. Performance evaluation of our proposed MAC protocol is shown in Section 3.7. Finally, concluding remarks are presented in Section 3.8.

3.2 Motivation and Problem Formulation

As explained earlier that as the data rate (bandwidth) diﬀerence between the available and the required is increasing, the medium access coordination increasingly become the performance bottleneck. To illustrate this point let us compare the time spent in medium access, Ma, to that spent in data transmission or service-time, Td.

Ma = AIF S[AC] + BF × ξ, (3.1) where AIF S[AC] is approximated as P% AIF SN[i, j] AIF S[AC] = i=1 × ξ + SIF S, j = 1, 2, . . . , χ (3.2) % where % is the number of contending nodes and χ is the access classes (i.e., Voice (Vo), Video (Vi), or Background (BG)), and BF in Equation (3.1) is the average backoﬀ.

Rreq The data transmission delay, on the other hand, is modelled as Td = , where ra

Rreq is the required data rate and ra is the available rate or bandwidth. According to the IEEE 802.11n amendment, the ready to use throughput (i.e., on top of the MAC layer) is equal to 100 Mbps. Using the OPNET Modeler, we compare the medium access delay with the transmission delay of diﬀerent required data rates in single-hop network with four connection pairs. All nodes reside in each other’s transmission ranges. During each CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 39

simulation run, we configure the transmitters to generate the same access class type. Before transmission the transmitter randomly selects a destination node to which the traffic is forwarded. The transmitters first contend for the channel (using the IEEE 802.11e medium access procedure, explained in Section 2.3) and then send the generated data packet to the intended receiver. Table 3.1 lists the IEEE 802.11n physical layer and traffic specification parameters.

Table 3.1: The physical and MAC configuration attributes of the IEEE 802.11n and traffic specifications

parameter value parameter value parameter value

CWmin 15 CWmax 1023 SIFS 16 µs AIFSN(Vo) 2 AIFSN(Vi) 2 AIFSN(BG) 7

Rreq(BG) 400-600 kbps Rreq(V o) 26-64 kbps Rreq(V i) 1-1.5 Mbps

ξ 9 µs ra 100 Mbps

We average the medium access delay and the transmission time of each delivered data packet, i.e., the MAC coordination time and the service-time. Under unsatu- rated traﬃc conditions, we observe that for Video, the time wasted in the medium access is 30% of the data transmission time as shown in Figure 3.1(a). As the diﬀer- ence between the available and the required bandwidth is increasing, the operation of the MAC protocol is increasingly dominated by the medium access coordination procedure as shown in Figures 3.1(b) and 3.1(c). Accordingly, instead of allocating the maximum achievable data rate of MIMO systems which is many times greater than the required data rate of applications, in this CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 40

0.016 0.005 0.014 0.0045 0.004 0.012 0.0035 0.01 0.003 (Sec)

0.008 0.0025 Time(Sec) 0.006 Time 0.002 0.0015 0.004 0.001 0.002 0.0005 0 0 Average time spent in medium Average time spent in service‐time Average time spent in medium Average time spent in service‐ access delay access delay time

(a) Video traﬃc class (b) Background traﬃc class

0.005 0.0045 0.004 0.0035 0.003 0.0025

Time(Sec) 0.002 0.0015 0.001 0.0005 0 Average time spent in medium Average time spent in service‐ access delay time

Figure 3.1: Average medium access delay versus transmission delay comparison of diﬀerent access classes.

chapter we assert that sharing the bandwidth increases the system performance in terms of medium access delay and bandwidth utilization.

3.3 MIMO Channel Interference Model

We consider MIMO links in distributed networks with n stationary nodes. In this type of network, there is no central control for data transmissions and all nodes have similar characteristics. Hence we do not distinguish between uplink and downlink CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 41

channels. As shown in Figure 3.2, the network model has k connection pairs where

th vi, {i = 1, 2, . . . , k} denote the i connection pair. For simplicity each connection

tx rx pair has one node as a transmitter (vi ) and the other node as a receiver, (vi ) all

the time. All nodes are equipped with nt transmit and nr receive antennas, where

tx rx tx ani , {i = 1, 2, . . . , nt} and ani , {i = 1, 2, . . . , nr} represent the anith transmit and the rx rx anith receive antenna element, respectively. The vith receiver node receives a desired message from a desired transmitter belonging to the same vith connection pair (i.e., i = i) and Γ interfering messages from nearby active connections (i.e., i 6= i). Γ is a sub-set of k − 1 connections that concurrently communicate with their connection pairs during the same time in which this connection is active.

tx Hd v1 rx v1

i H i HI I

Hd rx tx vk vk

Figure 3.2: Interference channel model

rx The received signal Y at the vith receive node is modelled as:

X i i Yvrx = PdHd + HI PI + ns, (3.3) ith i∈Γ CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 42

where Yvrx is the (nr × 1) received signal vector, Pd is the total transmit power of ith i tx desired transmitter node, PI is the total transmit power of the vith interfering node rx [57]. Hd is the (nr × n˜t) channel fading coeﬃcients matrix between the vith receiver tx and its desired transmitter vith (i.e., when i = i).n ˜t denotes the number of active

antennas during the transmission. Hd is expressed as   hd{1,1} . . . hd{1,antx }  n˜t   . . .  Hd =  . .. .  (3.4)     hd{anrx ,1} . . . hd{anrx ,antx } nr nr n˜t tx where h rx tx is the channel fading coeﬃcient between the an th transmit antenna d{ani ,ani } i rx λ η to the an th receive antenna and is modelled as, h rx tx = + χ , where i I{ani ,ani } 4πD σ tx λ is the wavelength, D is the distance between the vith transmitter and its intended

receiver, η is the path loss exponent, and χσ is the shadow fading eﬀect and is modelled with a zero-mean, Gaussian random variable, with standard deviation, σ (in dB),

i added to the path loss. HI in Equation (3.3) is the channel fading coeﬃcient matrix

th rx i of the i interfering connection and the vith receiver. HI expressed as

  hI{1,1} . . . hI{1,antx }  n˜t  i  . . .  H =  . .. .  , i = 1, 2,..., Γ, (3.5) I     hI{anrx ,1} . . . hI{anrx ,antx } nr nr n˜t tx where h rx tx is the channel fading coeﬃcient between the an th transmit antenna I{ani ,ani } i rx element of the interfering connection to the anith receive antenna element of the

receiving node. h rx tx is modelled the same way as h rx tx except that I{ani ,ani } d{ani ,ani }

R, the distance between the vith interfering connection and the receiver node, is used

instead of D to compute the path loss. ns in Equation (3.3) is the noise. For simplicity

ns is modelled as complex Gaussian random variable with a zero mean and a variance CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 43

1. In this work assume the following:

• We consider a Rayleigh slow fading channel with a very rich scattering environment and that the transmit and receiver antennas are spaced suﬃciently

i apart such that the channel gain matrix Hd and HI are independent identi- cally distributed. We assume that the spatial degree-of-freedom (ϕ) is equal to

min(nr, nt) [33], [73].

• For simplicity we assume that the power allocated for all transmit antenna elements is the same and is equal to the total transmit power constraint divided

by the number of antennas nt.

• We assume that during the reception period, nodes always activate all the available receive antennas and that they only activate the required transmit antenna elements during the transmission period.

1 2 Γ 1 Let horzcat[HI ,HI ,...,HI ] represent the horizontal concatation operation of

multiple matrices. Let Hcat denote the concatenated matrix of all interfering matrices

that are currently interfering with this connection pair at the same time. Hcat is given by

1 2 Γ Hcat = horzcat[HI ,HI ,...,HI ]. (3.6)

We denote the maximum error-free rate that the channel can support by CT . If

CT is to be used by k connection pairs, the sum of the partially used capacities by

1 1 2 Γ Similar to the MATLAB notations, we use horzcat[HI ,HI ,...,HI ] to horizontally concatenate i i matrices. We also use Hcat(:,HI ) to denote the removing of the matrix HI from the matrix Hcat, [47] CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 44

all connections must satisfy the following, k X Ci ≤ CT (3.7) i=1 where Ci is the partial used bandwidth by the vith connection and is given by. n˜ t h i−1 X H ˜ ˜ H H Ci = log2 1 + Pihj Inr + horzcat PiHjHj ,PI HcatHcat hj . (3.8) j=1 H th ˜ hj is the transpose of the j column of the Hd matrix, Hj is the remaining matrix

th after removing the j column from Hd.

3.4 QoS MIMO-aware MAC Protocol (QMMP)

QMMP is a MAC protocol that allows multiple concurrent communications through enabling spatial channels sharing. The QMMP scheme implements this by ﬁrst estimating the channel status, translating the required data rate into spatial channels requirements, broadcasting the required spatial channels using RTS CTS exchanges, and exploiting the reserved spatial channels to start data transmission. The following phases cover the functional aspects of the QMMP MAC scheme.

1. Medium Contention Phase (MCP): The channel idle and busy status of the QMMP scheme is defined based on the remaining (unused) spatial channels. Upon observing the AIFS[AC] idle time interval the station performs backoff. Once the backoff time is elapsed, the station starts the medium access procedure.

2. Channel Sounding Phase (CSP): Small physical layer frames called sounding frames are transmitted, over all antennas to induce the CSI, from the transmitter to the receiver and vice versa.

3. TXOP Scheduling and Broadcasting Phase (TSBP): Using the CSI information, CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 45

the accessing connection pair ﬁnds the TXOP reservation attributes which include the start and end time, the number of used spatial channels, and the tolerated interference.

4. TXOP Phase (TP): The reserved TXOP period during which the connection pair starts a course of data and Ack packets exchange.

These phases are executed in order, i.e., MCP, CSP, TSBP, then TP. Figures 3.3, 3.4, and 3.5(a) depict the phases, frames exchange order (e.g., the transmission order of the sounding, RTS, CTS, Data, and Ack frames), contention areas (i.g., depiction of whether nodes at the same or different contention area), the used spatial streams by each connection (e.g., how many antennas a connection pair is currently using), the medium contention using AIFS[AC], and the countdown procedure (e.g., the numbers of backoff slots of each connection pair and how they are decremented and which node reaches zero first). The MCP starts when a node has data and is ready for transmission. The process then starts by monitoring the channel activity until an idle period, equal to AIFS[AC], has been observed. Unlike the IEEE 802.11 MAC protocol, the idle and busy channel status are defined based on the Spatial Channels Control Multiple Access (SCCMA) mechanism. Spatial channels control is physically observed through the air-interface

i th by detecting the current spatial streams (Ua) used by the i connection pair and virtually via storing the intended spatial channel(s) broadcasted by the ith connection pair in the NAV table whenever they overhear RTS and CTS. Hence, the NAV keeps track of the remaining and the reserved spatial streams. Based on the total currently

PΓ+1 i used spatial channels, Ut = i=1 Ua, the instantaneous channel status is represented CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 46

Ua=2 Same Contention CSP C A A A area T C C C 2 s K K K

R 1 T Data Data Data s

AIFS[AC] Ua=4 10 slots

C A A T C C 4 s K K AIFS[AC] R 3 T Data Data

AIFS[AC] s AIFS[AC]

15 slots Ua=2

C A T C 6 s K SIFS

R 5 T Data s AIFS[AC] AIFS[AC] AIFS[AC] 20 slots

Figure 3.3: Stage 1: depiction of the QMMP main phases, deferral, and concurrent transmissions

as,   Idle if Ut < ϕ;   Channel status = (3.9) Busy if Ut = ϕ;    collision if Ut > ϕ;

In the case where the medium is sensed busy, (Ut > ϕ), a BF is consequently selected from a deﬁned CW range, i.e., BF = uniform(0,CW ), where CW min < CW <

CW max. The backoﬀ counter is decremented by one, only when the Ut < ϕ for ξ

µs and is frozen when the Ut ≥ ϕ. Once the backoﬀ timer counter reaches zero, the node is authorized to start the sounding frames exchange. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 47

MCP CSP TSBP TP t The same contention Ua = 4 area

C A A Same contention T C C

SIFS 2 area SIFS S SIFS K K SIFS

1 R T Data Data S SIFS AIFS[AC] Ua = 2 10 slots

CSR C A A T C C s K K 4 SIFS SIFS SIFS SIFS Differing R

3 SIFS SIFS

SIFS T Data Data On node 1 side, it set NAV with s AIFS[AC] AIFS[AC] accordence RTS 15 slots

C A T C s K 6 SIFS SIFS SIFS SIFS SIFS

5 Differing R On node 2 side, it sets the NAV T Data s AIFS[AC] AIFS[AC] with accordance to CTS AIFS[AC] 20 slots Ua = 2

Figure 3.4: Stage 2: depiction of the QMMP main phases, deferral, and concurrent transmissions

The CSP is the phase during which the connection pair start their SF exchanges, i.e., the CSQ and CSR exchange. The CSP phase start when the connection pair transmitter sends a CSQ sequence, and ends when the connection pair receiver responds with the CRS sequence. In both cases, the nt SFs are transmitted from each node over all antennas. At the end of the CSP phase, both the transmitter and the receiver of the ith connection get the CSI of the channel between them. Nodes around the ith connection pair can use the received sounding frames to assess the channel status between them and the transmitter/receiver or both. This information is continuously stored per node in the NAV table and updated whenever a new transmission is heard CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 48

for the same ith connection pairs. Because nodes must be able to decode and store the CSI and the broadcasted TXOP Reservation Attributes (RA) in their NAV table, CSP and TSBP phases must use a known channel coding technique whereas the connection can exploit diﬀerent channel coding techniques during the TXOP period. The TXOP RA is described by a tuple ˜ of 4 values: the TXOP start time ( ˜s), end time ( ˜e), tolerated interference (It), and the intended number of antennas (n ˜t). CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 49

Ua = 2

Same CSP C A A A Contention area T C C C S K K K 2

1 R T Data Data Data S AIFS[AC] Ua = 4

C A A T C C s K K 4

3 R R R T T T Data SIFS Data

s s s SIFS AIFS[AC] AIFS[AC]

Ua = 2

C A A T C C s K K 6

5 R Same T Data Data Contention area s AIFS[AC]

(a) QMMP reaction for slot mismatch caused by hidden node problem

(b) The schematic of the hidden node topology

Figure 3.5: The QMMP scheme under hidden node problem

Contending nodes in the QMMP MAC protocol suspend medium access attempts when they observe the start of sounding frames exchange. This gives a chance for nodes to update their NAV table for all reserved TXOP periods and resources (antennas). For example, node3 in Figure 3.4 pauses medium access contention when it hears node1, which is in the same contention area, starting the sounding frames CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 50

transmissions. Because it shares the same contention area with the receiver, node5 in Figure 3.4 pauses when it hears the start of CSR transmission of the receiver. Figure 3.4 also shows that after recording the reservation information, the contending nodes can resume the medium access procedure if there are still remaining resources (i.e., unused spatial channels). Figure 3.3 shows how connection pair (3-4) reserve a future TXOP period, i.e., the TXOP period that will be used after ∆ time interval of the reservation time. This happens because they do not find a TXOP with desired spatial streams immediately. The QMMP MAC protocol protects the connection pair’s right to withhold the channel usage for the whole period starting from first accessing the channel followed by TXOP reservation, and last during the reserved TXOP period via the utilization of small interframe space (SIFS), as depicted by Figures 3.3, 3.4, and 3.5(a). To prevent the reserved TXOP from being violated by other absent connection pairs (connection pairs that did not hear the reservation because they were busy communicating), the reserving connection pairs are required to re-send a resource reclaiming frame (i.e., RTS) that has the TXOP RA. Figure 3.5(a) shows how connection pair (3-4) reserves a TXOP, while connection pair (1-2) is busy with communication during its TXOP period. Connection pair (3-4) sends another RTS frame after connection pair (1-2) ends transmission to let the latter pair know about the reserved TXOP. Figures 3.4 and 3.5(a) show that nodes are located in different contention areas which causes TXOP slots to mismatch, i.e., the TXOP slot on the transmitter side has a different RA from that of the receiver side. The scenario in Figure 3.5(a) depicts how the QMMP performs under a slot mismatch as illustrated by Figure 3.5(b). For instance, node3 sends one desirable TXOP slot coupled with all alternative TXOP CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 51

slots, so that the receiver chooses from if needed. If an alternative TXOP slot is selected, the transmitter broadcasts the newly selected TXOP slot in another RTS packet so that nodes on the transmitter side can update their reservation table.

After the CSP phase both the transmitter and the receiver use the stored CSI and NAV information to ﬁnd the proper TXOP slot with the desired bandwidth. Although the CSI may change, the degradation of the channel data rate can be accommodated by utilizing spatial diversity and coding gains. This procedure of ﬁnding the desired TXOP slot is explained in the following section.

3.4.1 TXOP Scheduling and Broadcasting

This section discusses the tasks of finding the appropriate TXOP period and broadcasting its RA. Both tasks start after finishing the channel sounding phase and col- lecting the CSI information. Using the latter plus any previously broadcasted TXOP slot reservation information (stored in the NAV table), the connection pair perform a local computation to find the appropriate TXOP period that has the required resources (i.e., required bandwidth and/or antennas) and then broadcast its RA using RTS and CTS frames exchange. To explain the procedure of finding an appropriate TXOP period and determining its reservation information, we first define the following variables:

• zi, i = {1, 2,..., Ψ} ≡ The zith connection out of the Ψ connection pairs that have already reserved a TXOP slot.

zi • s ≡ The TXOP start time of the zith connection. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 52

zi • e ≡ The TXOP end time of the zith connection.

zi • Ua ≡ The number of intended used antennas that is broadcasted by the zith connection.

zi • HI ≡ The channel state information of the zith interfering connection.

zi • It ≡ The tolerated interference for the zith connection.

Scheduling a TXOP procedure can be divided into two cases: scheduling a TXOP period while there are no other previously broadcasted reservations, i.e., Ψ = 0. In this case, the accessing pair can schedule their TXOPs at any time to satisfy their QoS requirements. The second case is scheduling the TXOP period while there exist other previously performed reservations, Ψ > 0. In this case, the accessing pair can schedule their TXOPs such that they satisfy their requirements and at the same time they do not violate the limitation broadcasted by other connections ( e.g., not exceeding the tolerated interference and that the total used spatial streams must be less or equal to ϕ ). In the following subsections these two cases are further explained.

Scheduling a TXOP period with Ψ = 0

Algorithm (1) shows the procedure followed by the accessing pair to determine the TXOP slot reservation information in the case of no other reservations in the connection pair vicinities, i.e., no previously performed TXOP reservation by any connection pairs around either the transmitter or the receiver. Reservation in such a case is a very simple and straight forward process. Based on the collected information, nodes ﬁrst determine the required transmit antennas, the desired start time, and the period during which the connection pair holds the required resources (i.e., spatial streams). CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 53

Algorithm 1 Determining the reservation information of the TXOP slot INPUT:

• Hd.

1: while Ci < Rreq andn ˜t ≤ ϕ do

2: increament(n ˜t)

3: Given (Hd, n˜t) compute Ci (from Equation (3.8))

4: end while

5: return Given (Hd, n˜t) compute I˜t (from Equation (3.10))

6: return ( ˜s, ˜e, n˜t, I˜t)

Where Ci in line (3) represents the achievable capacity while increasing the number of active used antennasn ˜t for each iteration. Ci is computed from Equation (3.8). The amount of interference that can be tolerated while maintaining the required data rate is computed in line (5) and is given by

nt nr X X rx tx I˜ = h{an , an }P tx (3.10) t j i ani i=n ˜t j=1 rx tx tx Where h{anj , ani } is the channel fading coeﬃcient from the anjth unused transmit- rx tx ter to the an th receive antenna element. P tx is the transmission power of the an th i ani i transmit antenna element.

Scheduling a TXOP period with Ψ > 0

Here we explain the procedure for ﬁnding a valid TXOP period in the case where there exist other previously performed reservations. To illustrate this procedure, let us consider that at the time when the accessing pair start the searching algorithm there exist two reservations: pair A and pair B reservations. the pair A reservation CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 54

A A B B starts from ξs to ξe and pair B starts from ξs to ξe . Pairs A and B use two antennas,

A B i.e., Ua = 2 and Ua = 2, respectively. Figure 3.6 sketches the TXOP reservation start end time and the used antennas of both pairs. Algorithm 2 details the procedure that the accessing connection follows in order to ﬁnd the appropriate TXOP period during which they exchange their data.

A A Єe Єs

A Atx B rx Єs B Єs

Start the searching B tx algorithm Brx

Spatial stream Accessing pair Channel accessing and sounding

Figure 3.6: Finding a slot in case there exist previously reserved slots CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 55

Algorithm 2 Finding the appropriate TXOP slot INPUT:

zi zi zi zi zi • (s , e ,Ua ,HI ,It ), zi, i = {1, 2,..., Ψ}

• Hd

zi zi 1: A ←− {. . . , ξj,... } ←− sort(s , e ), ∀i time 2: for all ξjth ∈ A do 3: if ξjth is a start of transmission then zj 4: H ←− horzcat[Hd, horzcat[HI ,HI ]] zj 5: Ua ←− Ua + Ua zj−1 zj 6: It ←− min(It ,It ) 7: else zj 8: H ←− horzcat(Hd,HI (:,HI )) zj 9: Ua ←− Ua − Ua zj 10: It ←− It + It 11: end if 12: Ur ←− (ϕ − Ua) 13: while Ci < Rreq andn ˜t ≤ Ur do 14: increment(n ˜t) 15: Ci ←− Equation(3.8) ←− (H, n˜t) 16: end while 17: if Ci ≥ Rreq then 18: Ig ←− Equation(3.11) ←− (Hd, n˜t) ˜ 19: SL ←− (ξj+1 − ξ) 20: if Ig ≤ It then req 21: if SL ≥ SL then 22: clm ←− size(H) 23: I˜t ←− Equation(3.12) ←− (H, clm, n˜t) 24: return (ξ,˜ ˜e, n˜t, I˜t) 25: end if 26: else 27: ξ˜ ←− ξj+1 28: end if 29: else 30: ξ˜ ←− ξj+1 31: end if 32: end for

H: Matrix. Ua: Used antennas. It: Tolerated interference. Ur: Remained or unused antennas. req Sl: Slot length. SL : Requested slot length. clm: Total number of columns of a matrix. ξ˜: Valid start time of a TXOP slot. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 56

Because the finish or start of a TXOP period results in either releasing or the start of using spatial channel(s), we check the channel condition at the start or the finish of any TXOP period. The procedure in line (1) of Algorithm 2 sorts these points according to time (in which the TXOP starts or finishes) in ascending order using sort() time function. The Algorithm in lines 3-11 differentiates between the start and the end of TXOP periods in terms of the change in actively used antennas, tolerated interference, and H which holds the final matrix after concatenating or de-concatenating the columns of starting or finishing connections, respectively. The Algorithm in lines 5, 6, 9, and 10 keep track of the changes in the used antennas and the tolerated interference as different TXOP slots start or finish. The code in line 12 computes the remaining or unused antennas, which are then incrementally used in the while

loop, lines 13-16, to estimate the attained capacity, Ci. Using H and Ur, the algorithm in lines 13-16 incrementally increase the number of active used antennas and compute the possible achievable rate. If the requested rate is reached by any number of the available resources the while loop returns the required antennas,n ˜t. If the requested rate is attained as indicated by the code in line 17, the algorithm checks whether the generated interference, by using the required antenna, can be tolerable by other connections. This is done by comparing the generated interference with the tolerated interference broadcasted by other connections. The generated interference

zi is computed using the interfering matrix of the zith connection, HI , and by assuming that the fading of the channel from transmitter to receiver is similar to that from the receiver to the transmitter. Given the required antennas,n ˜t, returned from the while

zi loop, and HI , the accessing node can compute Ig using the following equation,

n˜t nr X X rx tx I = h {an , an }P tx (3.11) g I j i ani i=1 j=1 CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 57

rx tx tx where hI {an , an } is the channel fading coeﬃcient from the an interfering an- j i vith th tenna of the zi connection to this connection. If a valid TXOP with the required supported data rate is found as stated in lines 20-26, the accessing pair compute the ˜ tolerated interference It, which should not be exceeded by other connections if concurrent TXOP periods are scheduled. The tolerated interference is computed by the following expression

clm nr ˜ X X It = h{j,i}P tx (3.12) an{j,i} i=n ˜t j=1 where clm is the total number of columns of the H matrix, h{j,i} is the channel fading coefficient at the jth row and the ith column of the H matrix. The code in lines 27 and 30 show the case when the current interval is not a valid TXOP slot. In this case the algorithm jumps to the next interval which is represented by the next channel characteristic change point. The execution of both algorithms result in defining multiple valid TXOP slots with ˜ ˜ their RAs, (ξ, ,˜ n˜t, It). These valid TXOP slots and their RAs are then broadcast using RTS and CTS exchanges and stored by nodes which overhear them. Multiple valid TXOP slots are required in the case of a TXOP slot mismatch which is caused by the hidden node problem. A TXOP slot mismatch is induced by the fact that ˜ ˜ the description of the founded TXOP slot (i.e., (ξ, ,˜ n˜t, It)) on the transmitter side is different from that on the receiver side. Nodes then use their selected TXOP slots to transmit their data. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 58

3.5 QMMP Properties

In this section we prove some properties of the QMMP MAC protocol that shows additional advantages.

• Proposition 1 : When scheduling multiple connection pairs with interfering distance R greater than their communication distance D, the system usability D increases as the ratio R decreases.

Proof : Referring to the system model explained in Section 3.3, we have k connection pairs. Receivers of these connections receive a desirable signal from their desirable transmitters and Γ interfering signals from interfering transmit-

ters. Let Ri, i−{1, 2,..., Γ} represent the distance between this connection pair

th and the Γ interfering connection pairs. Letn ˜t represent the active transmit

antennas at each connection pair. Let Ct denote the total achievable system capacity, given by   k n˜t H X X 1 + Pjhi hi CT = log2  h i . (3.13) I + P H˜ H˜ H P H HH j=ϑ=1 i=1 nr j ϑ ϑ j j j ∀ j:j6=ϑ tx where Pj is the transmission power of the anjth transmit antenna element at the tx th vjth transmitter node, hi is the i column of the channel coeﬃcients matrix of ˜ th the desired transmitter , Hϑ, Hϑ is the remaining matrix after removing the i

th column out of Hϑ, Hj is the channel fading coeﬃcients matrix of j interfering

node, Inr is the identity matrix.

For simplicity we assume that the transmission power of all transmitting anten-

rx nas are the same. Accordingly, the channel coeﬃcient matrix estimated by vith CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 59

receiver can be expressed as  k h{1,1,j} . . . h{1,n˜ ,j}  t   . . .  Hj =  . .. .  (3.14)     h{nr,1,j} . . . h{nr,n˜t,j} j=1

rx where h{nr,n˜t,j} is the channel fading coeﬃcient between the anith receive element tx tx of this node to the anjth transmit element of the vjth transmit node and is given by

 ( λ )η + χ if j = γ;  4πD σ h{nr,n˜t,j} = (3.15)  λ η ( 4πR ) + χσ if j 6= γ; Equation (3.13) can be rewritten as

k n˜t H X X 1 + Pjh hi C = log i . (3.16) T 2 I + P H HH j=1 i=1 nr j j j Hence, by ﬁxing D and assigning larger values for R the channel fading co-

eﬃcient factor, h{nr,n˜t,j}, of the interfering connections further decrease which increases the SNR per each connection pair. From equation (3.16) getting higher SNR per each connection can collectively produce higher system usability.

Advantages:

This property can be utilized to enhance the system usability by scheduling multiple concurrent connections with the least interference eﬀect on each other.

• Proposition 2 : The QMMP MAC protocol can converge to the IEEE 802.11 MAC protocol by switching oﬀ the channel sounding phase.

Proof : To translate the required rate into antennas requirements, the QMMP CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 60

MAC Protocol extends the IEEE 802.11 MAC protocol by adding the channel sounding phase, which is then used to compute the required antennas. By switching oﬀ the channel sounding phase and requesting the maximum bandwidth, the QMMP MAC protocol completely converges to the IEEE 802.11 MAC protocol.

Advantages: This property is very beneﬁcial in systems where stations are required to communicate with legacy IEEE 802.11n stations and at the same utilize the concurrent bandwidth partake mechanism with capable stations. For example, in wireless mesh networks, the mesh access point stations are required to communicate with peer mesh points and users ( for example as shown in Figure 2.3). Hence, the MAP can exploit the concurrent bandwidth sharing protocol with peer mesh points and switch to the IEEE 802.11 MAC to communicate with legacy users.

3.6 Discussion of other QMMP Advantages

In this section we discuss some advantages of the QMMP MAC protocol such as the TXOP scheduling manipulation which can be exploited to increase the overall system usability and to manipulate the delay to enhance the QoS differentiation. Another advantage is the independent code design of different links which can produce different link characteristics that suit different application requirements. Finally, at the end of each discussion we highlight the advantages.

• Property 1 : The QMMP MAC protocol permits an independent coding and decoding mechanism of individual pairs of nodes during the reserved TXOP CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 61

period.

Discussion: Because connection pairs and nodes around them do not exchange any coordination information during the data transmission phase, the QMMP MAC protocol permits connection pairs to implement their own desirable coding and decoding structures during their reserved TXOP period.

Advantages: Such property allows connection pairs during the TXOP phase to use any channel coding structure and decoding scheme that suits their instantaneous channel status. This results in improving the channel quality of diﬀerent connection pairs based on their instantaneous channel status. As well, connection pairs and nodes around them know the spatial signature of each other. Hence, they can consider the received signal from other pairs as an interference signal and they can apply any interference cancellation technique.

• Property 2 : With the QMMP, connection pairs can optionally schedule their transmission opportunity at any desired slot.

Discussion Given the possibility of searching for valid TXOP slots as explained by Algorithms 1 and 2, the QMMP has the following means to enforce the TXOP reservation time. 1) contending nodes pause when they hear the start of sounding frames exchange as depicted by Figures 3.3, 3.4, and 3.5(a). This allows nodes to update their NAV table for all reserved TXOP slots. 2) Successful four way handshakes (channel sounding and RTS CTS frames exchange) further guarantee the TXOP slots reservations. Nodes which miss receiving the sounding frames because of collision are highly likely to receive the reservation frame exchanges. In addition, transmitting the sounding frame and RTS CTS frames over multiple delayed slots can further assure the reception of CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 62

the CSP and/or TSBP frame exchanges. 3) Connection pairs are only allowed to reserve the TXOP slot if their transmission do not violate the constraints broadcasted by other connection pairs for that TXOP slot. These constraints include not exceeding the upper limit of the used spatial streams, the generated interference, and the start of the medium access based on the remaining spatial streams. 4) Reservation information of all reserved TXOP slots are rebroad- casted by any new following broadcasted reservation. 5) QMMP guarantees the reserved TXOP slot by allowing connections to use small interframe space (SIFS) prior to their reserved TXOP. 6) To prevent the new reserved TXOP from being violated by other absent connection pairs, the reserving connection pair re-send a resource reclaiming frame that have the TXOP slot reservation

information. Figure 3.7 shows how connection pair P2, which reserved a slot

while pair P1 is busy with TXOP transmission, sends a reclaiming frame after

P1’s end transmission to let the latter pair know about the reserved TXOP slot. 7) Connection pairs can timeout after AIFS[AC] if the allocated bandwidth is

not being used yet. Figure 3.7 shows that pair P3 timeout and start medium

access after AIFS[AC] on P2 as the latter did not start TXOP at the predeﬁned time. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 63

Same contention area Did not start DTP phase

P1 AIFS[AC] P2 AIFS[AC] P3

Start receiving data timeout

Used spatial Remained spatial Stations Postpone contention Resources reclaiming streams streams frame

Reservation broadcasting Start contending Sounding frame frames RTS and CTS.

Figure 3.7: Some functional aspects of the QMMP MAC protocol

Advantages:

– Providing delay differentiation: Flexibility of scheduling higher critical delay bound flows earlier than those with less delay bound constraints can help support more flows to meet their delay constraints.

– Enhancing system usability: Concurrently scheduling connection pairs which have less interference eﬀect on each other can increase the achievable capacity of each pair and hence increase the overall system usability.

– Enhancing the bandwidth utilization: Using all the remaining antennas in all intervals.

• Property 3 : By scheduling the appropriate TXOP to access classes based on CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 64

their delay bound and system resources, the QMMP MAC protocol can provide delay diﬀerentiation.

Discussion: Let FL = {fli : i = 1, 2, . . . , σ} represent a set of access classes for which TXOP slot is scheduled. Without loss of generality, assume the QoS

requirements of an access class, fli, are described by

i – rr: The requested bandwidth (bits/sec).

– db: The delay bound.

Let S = {si : i = 1, 2, . . . , δ} represent the set of TXOP slots. Each slot, si, is described by following tuples:

i – ra: The available rate.

i – s: The slot start time.

i – e: The slot end time.

Algorithm 3 allocates the appropriate TXOP slot to access classes according to the weight of their maximum tolerated delay bound and channel utilization. Scheduling access classes close to their delay bound can save slots for access classes with more critical delay bound. Coupling the previous allocation criteria with enhancing the bandwidth utilization can further be optimized to serve more access classes. The following algorithm describes the procedure of ﬁnding the most tolerable slot in term of delay constraint. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 65

Algorithm 3 Find most delay tolerable slot INPUT:

– FL: set of ﬂows

– S: set of slots

1: for all fi ∈ FL do

2: db ←− fi

3: rr ←− fi

4: for all si ∈ S do

i 5: e ←− si i 6: ra ←− si i e rr 7: Oi ←− + i db ra 8: end for

9: return max(Oi)

10: end for

For each access class the algorithm ﬁrst stores the delay bound and requested rate as shown with the code in lines 2 and 3. For each available slot we store the start time and the available rate as described by the algorithm in lines 4 and 6. Given the stored information, i.e., the slot start time, delay bound, requested rate, and the available rate, the algorithm in line 7 computes the tolerable ratio of the ith slot. Lastly, the algorithm in line 9 returns the maximum tolerable ratio of all slots. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 66

3.7 Performance Evaluation

In this section, we ﬁrst begin by describing the simulation model and the traﬃc model. Next we present the performance metrics, followed by detailed discussions and comparisons of the simulation results.

3.7.1 Simulation Model

Utilizing the OPNET Modules we have modiﬁed the built-in IEEE 802.11e physical and MAC layers to include all the details of QMMP MAC protocol. These modiﬁca- tions include the following:

• Adding multiple transmit and receive antenna elements. Each transmit and receive antenna element is modelled using the built-in wireless transmit and receive module OPNET Modeler, respectively. Accordingly, each node module is equipped with four separate transmit and receive antennas. We also modiﬁed the legacy model to performs spatial stream multiplexing and de-multiplexing.

• Modifying the interaction between the physical and MAC layers to accommodate for the increased number of transmit and receive antenna elements.

• Inserting the sounding frames transmission phase which include exchanging the sounding frames through all transmit and receive antenna elements of the transmitter and its intended receiver.

• Building the packet structure of the sounding frame using the Packet Module of OPNET. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 67

• Upon receiving the sounding frames transmitted, receive antenna models esti-

rx tx mate the channel fading coeﬃcient hd{ani , ani } of all the received sounding

frames. The results are then used to construct the channel sounding matrix Hd. Similarly, the overhearing receivers in the same coverage area use the received sounding frames to estimate the channel fading coeﬃcient and to construct the

i HI .

• Modifying the NAV table to maintain the updated number of remaining and used antennas (spatial streams).

• To estimate the achieved channel capacity under diﬀerent interference zones, we interfaced the OPNET Modeler to MATLAB simulation tools. Hence, the OPNET modeler is used to model the MAC layer whereas MATLAB is used

i to model the physical layer. The MAC layer sends Hd, HI , D, R, andn ˜t to the MATLAB. The latter then computes the achieved capacity and returns the result.

• The instantaneous channel status (i.e., busy, idle, or collision) was also modified following Equation 3.9. The modification include modifying the state transition diagram of the original model and the interaction between the physical and MAC layers. Figure 3.8 models the instantaneous signal decode-ability model that we used in our model. To explain the figure consider a receiver equipped with four receiving antennas. Also consider 5 transmitting nodes, each node equipped with one transmitting antenna. We consider that the receiver can receive and decode all the incoming spatial streams as long as the number of spatial streams is fewer or equal to the number of receiving antennas which is CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 68

equal to nr in this model. If these transmitters start transmitting their streams, the instantaneous signal decode-ability at the receiver is modelled as shown in Figure 3.8. Spatial streams that experience any collision period are discarded at the end. We also consider that if a collision occurs, the collision period lasts as long as the received signal has an overlapping portion with the collided signals.

Collision Period

Signal decodability decoded decoded decoded Undecoded period

Spatial stream

Finish receiving Discarded

Figure 3.8: Instantaneous signal decode-ability with ϕ=4

• Since IEEE 802.11n is not yet supported by OPNET Modeler, we have also modiﬁed the IEEE 802.11e module to include the deﬁned IEEE 802.11n physical and MAC enhancements. This includes adding multiple transmit and receive antennas that are attached to the multiplexing and de-multiplexing module.

• The values of the physical and MAC layers characteristics used for both models follow those deﬁned for the IEEE 802.11n amendment which are summarized in Table 3.2. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 69

Table 3.2: IEEE 802.11n conﬁgurations

parameter value

CWmin 15

CWmax 1023 SIFS 16 µs

PMDU 500-65535 byte

RIFS 2 µs

q 30 µs

3.7.2 Traﬃc Model

Nodes are configured to generate traffic that follows a Poisson distribution with mean arrival equal to λ. The latter is a simulation parameter that assumes different values to model the network loads. For each generated data packet, the nodes select a random neighbor as final destination for this packet. The transmission rate in Mbps, with no interference at D =250 meters is computed and used for MIMO-aware modified IEEE 802,11e and QMMP. The former schedules the entire supported data rate, whereas the QMMP MAC shares the supported data rate via utilizing the multiple spatial channels created by multiple antennas system.

3.7.3 Simulation Parameters

• Desired distance (D): the communication distance (in meters) between the transmitter and its intended receiver. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 70

• Interference distance (R): the distance (in meters) between the receiver and the transmitter of the received interfering signal.

• Requested rate (Rreq) window: window of request rates from which a desirable requested rate is selected using uniform distribution function.

• Similar to the IEEE 802.11n draft [30], we consider that nt = nr = 4.

3.7.4 Performance Metrics

• Average throughput: the total number of successfully delivered bits divided by the lifetime of the simulation.

• Average medium access delay: the average time from inserting the packet in the queue until it starts transmission.

• Achieved rate: the instantaneous supported physical data rate (Mbps) computed before transmission.

3.7.5 Simulation Results

In this section, we study the performance of the QMMP scheme under different interference zones, communication distances, and requested rates. In all the experiments 90% confidence levels are maintained with 10% confidence intervals based on 10 independent runs [3]. To examine the effects of the interference on the system throughput, we consider the scenario in Figure 3.9(a) where we vary two parameters D and the R for

two connection pairs P1 and P2. For this scenario no MAC protocol is used. Each CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 71

vtx transmitter generates data for its intended connection pair receiver. Before Pith transmission the vtx transmitter ﬁrst exchanges the sounding frames to get the CSI Pith

Pith information, Hd, which is then concatenated with the interference part HI of the other interfering connection. The result is then used to compute the total channel capacity using Equation (3.8) which is then used as the data transmission rate.

P1 tx rx

R D P2 P 4 R

P1 rx tx

P 3

(a) Interference model scenario (b) Modelling the distance aﬀect on the system performance

Figure 3.9

The results in Figure 3.10 show the system throughput at different values of D and R. No doubt larger values of R reduce the interference effect, and hence allow the other connection pair to achieve a higher data rate. Note that connections communicating over a smaller distance D are less affected by interference component. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 72

8 x 10 throughput (bps) vs Interference distance R (m) 7

D=50

5 D=100

4 D=150

3 T h ro u g h p u t (b p s ) D=200

2 D=250

0 0 50 100 150 200 250 Distance R (m)

Figure 3.10: Achievable throughput as a function of communication and interference distance

The performance of the QMMP is compared to the MIMO-aware modified IEEE 802,11e MAC protocol. To evaluate the performance of the QMMP under different values of D, R, and k, we place four connection pairs, i.e., P1, P2, P3, and P4, in a circle, as shown in Figure 3.9(b), with equal interference distance R from each other so that the effects of interference remain fixed during the simulation. For each

simulation run, we vary D and hence the requested rate Rreq, where the former is assigned to one of the following: 50, 100 and 150 meters. Whereas the latter is randomly selected from diﬀerent ranges: 50-85, 85-120, 120-155, 155-190, 190-260, and 260-300 Mbps. During each simulation run one range is considered and one value is selected as the requested rate using the uniform distribution function from the selected range. Figure 3.11 relates the requested rates to the medium access delay. Increasing the amount of the requested rate apparently increases the experienced CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 73

MAC delay. Larger communication distances D cause higher channel attenuation, which proportionally requires more antennas (spatial channels) to attain the same requested rate. On the other hand, communicating over smaller distances D leads to higher proportional supported data rate per spatial channel, hence requesting a higher data rate, as connections with shorter communication distance D tend to have better performance compared to satisfying the same requested rate at a larger communication distance. Higher sharing of the spatial channels of small requested data rates produce higher throughput, see Figure 3.12.

0.0009 0.00085 0.0008 0.00075 0.0007 D=50 0.00065 D=150 delay(Sec) 0.0006 D=250 0.00055 MAC 0.0005 0.00045 0.0004 55 100 137 172 207 242 280 Average requested rate per connection pair(Mbps)

Figure 3.11: Medium access delay for diﬀerent ranges CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 74

100

70 D50

60 D150

Throughput(Mbps) D250 50

40 0 50 100 150 200 250 300 Average requested rate per connection pair(Mbps)

Figure 3.12: Throughput for diﬀerent ranges of requested rates

To further examine the eﬀects of the communication distance D on the system performance, we ﬁx the interference distance R and the requested rate Rreq at each

simulation run. For each simulation run Rreq is assigned a value from 50, 200 or 300 Mbps. R is ﬁxed at approximately 180 meters. D, at each simulation run, is set to one of the following 50, 100, 150, 200, or 250 meter. Results in Figures 3.13 and 3.14 clearly show that a larger value of D results in increased delay and lower throughput. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 75

0.001 0.0009 0.0008 delay(Sec)

0.0007 Rreq=50 0.0006 access

Rreq=200 0.0005 Rreq=300 0.0004 Medium 0.0003 50 100 150 200 250 Communiction distance(Meters)

Figure 3.13: Medium access delay for diﬀerent values of D

100

Rreq=50 70 Rreq=200 60 Rreq=300 Throughput(Mbps)

40 0 50 100 150 200 250 300 Communication distance(Meters)

Figure 3.14: Throughput for diﬀerent values of D CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 76

To compare the performance of the QMMP scheme with the MIMO-aware mod- iﬁed IEEE 802.11e MAC protocol, a generalized WMN topology is considered as shown in Figure 3.15. Diﬀerent settings are obtained by varying D and R such that their average distance remains the same for all the 10 used WMN topology settings. The IEEE 802.11e MIMO-aware MAC protocol uses the entire achievable capacity,

whereas the QMMP based MAC protocol uses Rreq that is uniformly selected from the range 50-300 Mbps.

D R

Figure 3.15: Wireless mesh network model

Results in Figures 3.16(a) and 3.16(b) depict the average medium access delay and achievable throughput over 10 simulated cases, with average values of D and R at 150m and 100m, respectively. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 77

0.0012 7.00E+07

0.001 6.00E+07

5.00E+07

(Sec) 0.0008

4.00E+07 delay 0.0006

Access 3.00E+07

0.0004 Throughput(bps) 2.00E+07 Medium

0.0002 1.00E+07 Average 0 0.00E+00 MIMO‐based modified 802.11e QMMP based MAC protocol MIMO‐based modified 802.11e QMMP based MAC protocol

(a) Medium access delay diﬀerence (b) The throughput diﬀerence

Figure 3.16: performance comparison between the IEEE 802.11 DCF and QMMP MAC protocol

To further demonstrate the eﬀectiveness of the QMMP MAC protocol, we show in Figure 3.17 its transient performance over an interval of 0.02 seconds for a tagged node in the experiment above. The results show that the requested rate can be always attained throughout the simulation run. CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 78

4.00E+08 Achieved rate Rreq 3.50E+08

3.00E+08

2.50E+08

2.00E+08

1.50E+08

Throughput(bps) 1.00E+08

5.00E+07

0.00E+00 0 0.005 0.01 0.015 0.02 0.025

Figure 3.17: Comparison of requested and achieved rates

3.8 Summary

In this chapter, we introduced a novel MIMO-aware bandwidth sharing protocol. Instead of allocating the maximum bandwidth achieved by the MIMO system, as defined in the IEEE 802.11n, QMMP introduces an efficient concurrent bandwidth sharing scheme that is based on sharing the spatial channels. We detailed QMMP’s phases which include medium contention control, which is conditioned on the number of used spatial streams, the channel sounding phase, which exchanges small frames over all antennas between the transmitter and its intended receiver to assess the channel between them, the TXOP scheduling and broadcasting phase procedure, which uses the channel state information resulting from the channel estimation phase to find the appropriate TXOP period with the desired bandwidth (antennas) and CHAPTER 3. QOS MIMO-AWARE MAC PROTOCOL 79

transmits its reservation attributes using an RTS CTS exchange, and the TXOP phase during which the connection pair start and end their data and Ack packets exchange. The performance of QMMP was evaluated under different scenarios for different rate demands, interference affects, and communication environments. The results were compared to those of the MIMO-aware modified IEEE 802.11e MAC protocol. Results showed that our proposed QMMP scheme outperforms the IEEE 802.11n modified model in medium access delay and throughput. We also showed that nodes can always attain their requested rate. Chapter 4

MIMO-aware Medium Access Collision Avoidance

4.1 Introduction

One of the most performance degrading factors of the IEEE 802.11e EDCF protocol is medium access collisions, i.e., multiple stations accessing the medium at the same time. To alleviate the collision problem, in this chapter, we introduce a novel MIMO- aware EDCF (M-EDCF). M-EDCF is the ﬁrst such protocol that exploits the MIMO properties to enhance the IEEE 802.11e EDCF collision-avoidance mechanism. The basic idea is the sharing of the multiple spatial channels created by MIMO systems, during medium contention periods, to avoid collisions. As the concurrent spatial channels used are fewer than or equal the spatial degree-of-freedom, receivers can detect multiple medium contentions. Spatial channels sharing has other properties such as medium contention termination and medium contention selection. The former refers to the capability of terminating the ongoing medium access attempt if the accessing 80 CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 81

node detects multiple concurrent medium accesses. However, contention selection refers to the possibility of detecting multiple concurrent medium access attempts initiated by different transmitters and hence they can coordinate their responses. Both the medium contention termination and selection can further be exploited to reduce the medium access collisions and to provide QoS differentiation. In addition an optimal spatial channels sharing that is based on the network load is also proposed in this chapter. The optimal spatial channels sharing is based on estimating the number of active nodes (i.e., nodes have data and currently contending for the channel) and their probabilities of transmission to determine the spatial channels sharing value that avoids collision and does not degrade the system usability. The remainder of this chapter is organized as follows. In Section 4.2 we explain how the MIMO multiple spatial channels can be exploited to enhance the IEEE 802.11e EDCF collision-avoidance mechanism. In Section 4.3 we detail the simulation model including the performance metrics and network topology. In Section 4.4 we evaluate the M-EDCF scheme over single-hop network. Section 4.5, we propose an online scheme that finds the optimal spatial channels sharing value that further reduces medium access collisions and boosts the channel utilization over single-hop network. In Section 4.6 we also evaluate the M-EDCF scheme over multi-hop network. Concluding remarks are given in the Section 4.7. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 82

4.2 Enhancements to the IEEE 802.11e EDCF Col-

lision Avoidance Mechanism

We consider a distributed network model that consists of n stationary nodes, similar to wireless mesh networks deployed in cities. Each node is equipped with nt transmit and nr receive antennas and independently apply the proposed collision-avoidance scheme. At any node in the network, the system is viewed as this node (Node-in- Question (NiQ)) and its neighbors, i.e., nodes whose transmission ranges overlap with NiQ. Let z represent the total number of antennas in the NiQ’s vicinity, including

th the NiQ node, i.e., z = (ne + 1) × nt, regardless on which node the i antenna element is mounted, ne is the number of neighbors that have a transmission range

th overlapping with the NiQ node. Let fi, i = {1, 2, . . . , z} index the i antenna element.

Each fith antenna element when the backoﬀ countdown of its holding node reaches zero transmits one sounding frame in a slot qi, i = {1, 2, . . . , γ} that is randomly selected from the [0, γ] window using the uniform distribution function. qi represent the transmission time delay interval required to send one sounding frame, γ denotes the sounding frames transmission window as shown in Figure 4.1 which consists of multiple slots during which a node must transmit all its sounding frames over all transmit antenna elements. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 83

qi RIFS AIFS[AC]

Sounding frame Slot

Figure 4.1: The sounding frames transmission window

IEEE 802.11n uses the Reduced Interframe Space (RIFS) (as an idle time interval) to separate sequential sounding frame transmissions as shown in Figure 4.1. For example, if the NiQ wants to send two sounding frames over two consecutive slots, it first sends the first sounding frame at the first slot and then waits for RIFS time interval and then sends the second sounding frame. Antennas of the same node may send their sounding frames as follows: 1) concurrently at the same slot i by using a sounding frames transmission window equal to one (i.e., γ = 1). This method is implemented by the IEEE 802.11n MAC protocol. Hence, throughout this thesis when γ = 1 we refer to the IEEE 802.11n draft. 2) partially overlapped slots where only a subset of antennas are allowed to transmit in each slot. 3) distributed over multiple delay slots by using longer γ. In distributed systems wireless nodes synchronize the starting of the sounding frames transmission windows using common interframe space intervals to the channel CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 84

status changes which they all observe. To explain this, let us consider the example

illustrated in Figure 4.2 where we have three connection pairs: A(tx, rx), B(tx, rx),

and C(tx, rx) in the same contention area. Initially, connection pair B(tx, rx) is communicating while connections A(tx, rx) and C(tx, rx) are deferring to B(tx, rx). At the end of connection B(tx, rx) transmission, connections A(tx, rx) and C(tx, rx) observe that the channel status has changed and hence they start contending for the channel, the procedure starts by first waiting for the interframe space time to elapse and then performs the backoff countdown. In single-hop networks, there are two possible outcome cases. The first case, is where both connections, A(tx, rx) and C(tx, rx), reach their zero countdown at the same time. In this case, the start time of the sounding frames transmission window of both connections is at the same time. Accordingly, the receivers receive aligned frames. The other case happens when either connection,

A(tx, rx) or C(tx, rx), reaches its zero countdown earlier than the other connection. In this case, if the start time difference is less than the propagation delay pd, both connections start their sounding frames transmission window misaligned with pd time difference. On the other hand, if the start time difference is greater than the propagation delay pd, only one connection proceeds where the other defers because it notices the start transmission of the first connection. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 85

AIFS[AC]

Atx Arx

The start time difference of different sounding frames transmission windows

Btx Brx

AIFS[AC]

Ctx Crx

Figure 4.2: The synchronization of the sounding frames transmission windows

For a specific γ window length, the antennas use a uniform distribution function to select a candidate slot and then to send their sounding frames at the selected slots. Larger γ permit contending nodes to share the spatial channels during the channel sounding phase. Hence even with multiple concurrent transmissions (by different transmit nodes) receivers can still decode the received signals from different transmitters because their signals are distributed over multiple delayed slots. On the contrary, shorter γ can cause a higher collision probability because a higher number of signal overlap (i.e., greater that ϕ) can cause collisions. Received signals of different antennas mounted on different transmit nodes have independent channel fading characteristics, therefore the receivers can separate the received signals using the same spatial separation techniques used to separate the signals received from different antennas mounted on the same transmitting node. Hence the core of our proposed scheme in this chapter (to enhance the IEEE 802.11 EDCF collision-avoidance mechanism) is based on varying γ, i.e., avoid having more than ϕ signals transmitted at the same slot. The basic idea is as follows. When the backoff counter of a node CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 86

reaches zero, the node independently selects its γ length such that the selected value is likely to avoid collision with other contending nodes. Varying the γ length by the contending nodes can have the following:

• Early Medium Contention Termination (EMCT): if the contending nodes choose longer γ window lengths then they can switch to the listening mode during the idle slots, the contending nodes can observe the channel activity (i.e., how many of the neighboring nodes are currently contending too) while they are currently contending for the medium. This new property can trigger the EMCT mechanism, which refers to the possibility of stopping the sending of the CSQ or CSR sequential frames if the accessing node detects other ongoing CSQ or CSR frame transmissions. This property can introduce early collision resolution, which allows nodes to terminate medium access contention attempt earlier if it would have resulted in a collision. A node may also terminate its attempt in accordance with any system performance metric requirements such as supporting QoS (where only lower priority nodes are required to terminate), being collision- avoidance conservative (considering longer γ window lengths which make nodes more attentive to other nodes medium access contentions), and/or saving bandwidth (terminate medium access attempts that has a higher probability to end up with a collision). Node C in Figure 4.3(a) is performing early termination because it has detected another medium access attempt by node B. This is compared to Figure 4.3(b) where γ = 1. Node C does not detect the other medium access attempt and hence it has caused a collision to the data received at node B. Algorithm 4 lists the procedure followed by the transmitter during the contention period. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 87

Same contention area ABC D

CSQ Detect multiple medium access

CSR CSQ terminate

TSBP+TP

(a) Multiple medium contentions when γ > 1

Same contention area A B C D

CSQ

CSR CSQ

CSR

TSBP+TP Collision

TSBP+TP TSBP+TP

(b) Multiple medium contentions when γ = 1

Figure 4.3: Medium contention at transmitters CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 88

Algorithm 4 M-EDCF transmitter medium access procedure 1: if BF ← 0 then

2: determine(γ)

3: for ani, ∀i, i = 1, 2, . . . , nt do

4: A{... } ← qj ← uniform(0, γ)

5: end for

6: if EMCT ← TRUE then

7: sort(A)

8: if IDL > ∆ then

9: Switch to listening mode and observe the channel

10: if Node detects other medium access attempts then

11: Terminate.

12: end if

13: end if

14: else

15: Send the sounding frames at their speciﬁed slots without switching to listening

mode

16: end if

17: end if

γ: The SF transmission window length during which a node sends its sounding frames.

qj: The slot at which the anith antenna element transmits its SF. IDL: The idle time interval between sending multiple sounding frames.

∆: Suﬃcient time for switching and listening to the media.

• Medium Contention Selection (MCS): considering larger γ can help receivers CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 89

detect multiple medium access contentions of diﬀerent transmitters. This is possible because their sounding frames are distributed over multiple delayed slots. Detecting multiple medium access contentions can trigger the MCS property. MCS can help the receivers to coordinate their responses such as responding to a particular access class to enhance the QoS diﬀerentiation or keeping silent if the node’s response may cause a collision to other nodes. Figure 4.4(a) shows the scheme when the accessing nodes consider longer γ lengths. Node C receives CSQ from D and CSR from B, C does not respond to node D, as its response can cause collision to B. The same scenario is repeated for γ = 1, i.e., all accessing antennas send their sounding frames at the same slot, as shown in 4.4(b) the spatial streams from B and D exceed ϕ and therefore C cannot decode the received signal. Node C responds to the next CSQ request from D which causes a collision to the data received at B. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 90

Same contention area

A BC D

CSQ

Detect multiple medium access requests

CSR CSR CSQ

CSQ

(a) Multiple medium contentions when γ > 1

Same contention area A BC D

CSQ

Collision

CSR CSR CSQ

CSQ

Collision

TP CSR CSR

(b) Multiple medium contentions when γ = 1

Figure 4.4: Medium contention at receivers CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 91

Algorithm (5) lists the procedure followed the receiver during the contention period.

Algorithm 5 : M-EDCF receiver medium access procedure 1: if Single CSQ request is received then

2: Respond with CSR after SIFS time.

3: else if Multiple CSQs requests destined to this node then

4: Respond with CSR to the ﬁrst ﬁnished CSQ node.

5: else if Multiple CSQs and CSRs requests are received then

6: Do not respond.

7: end if

4.3 Performance Evaluation

In this section, we ﬁrst begin by describing the simulation model and network topology. The traﬃc model used is similar to Chapter 3. We then introduce the simulation parameters and performance metrics. Next we provide a detailed discussion and comparison of the simulation results.

4.3.1 Simulation Model

To study the performance of the proposed M-EDCF scheme, we further extend the simulation model discussed in Chapter 3 to include the details of M-EDCF scheme. The extended modiﬁcation includes adapting the γ length and selecting the slot at which each antenna element is going to transmit its sounding frame. For a result CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 92

comparison, we also use the MIMO-aware modiﬁed IEEE 802.11e MAC protocol used in Chapter 3.

4.3.2 Network Topology

We evaluate the M-EDCF scheme in two network models:

• Single-hop network: in Section 4.4 we evaluate the proposed M-EDCF scheme in a single-hop network model. In this conﬁguration we have 16 nodes that are distributed in the transmission range of each other using the uniform distribution function and they hear the same channel status.

• Multi-hop network: in Section 4.6 the multi-hop network model is also configured with 16 stationary nodes. In this configuration nodes are distributed with distances ranging between (150-400) meters using the uniform distribution function in such a way that hidden nodes are formed as shown in Figure 4.5. Nodes (A, B), and (B, C) are in each other transmission range whereas nodes (A, C) are out of each other transmission range. The latter makes nodes all over the network observe different channel states.

A B C

Figure 4.5: Hidden node network topology

4.3.3 Simulation Parameters

• γ: the sounding frames transmission window length. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 93

• Net oﬀered load: the number of packets generated per node per second.

4.3.4 Performance Metrics

In addition to the average medium delay and throughput deﬁned in Section 3.7, we introduce the following performance metrics used in this chapter.

• Mean number of attempts per packet: average number of transmission attempts per successfully delivered packet.

• M-EDCF MAC delay: the time from inserting the packet in the queue until it starts transmission. The M-EDCF MAC delay is composed of two delay components, backoff delay and the CSP delay. The backoff delay refers to the backoff contention window lengths used before the medium access. This delay may include multiple attempts, which requires doubling the contention window upon each reception failure. The CSP delay refers to the used sounding frames transmission window lengths considered by each node to send its sounding frames before transmitting its data packet.

Td • Channel utilization: deﬁned as U = where Td is the time spent in trans- Tc+Td

mitting successful data and Tc is the total control time, which includes all the time spent in exchanging the coordination messages until a successful transmission.

4.4 Simulation Results–Single-hop Network

In this section, we study the performance of the M-EDCF scheme under diﬀerent network loads. The performance of the M-EDCF is also studied during activating CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 94

and silencing EMCT and MCS properties. In all the experiments 90% confidence levels are maintained with 10% confidence intervals based on 10 independent runs [3]. Figure 4.6 shows the mean number of attempts needed to transmit a packet for different γ lengths and network loads. As expected, the number of collisions for smaller γ tend to have rapid increase as the network traffic load increases. This is because smaller γ have a higher probability of signal overlap than those with longer transmission windows.

0.2

0.18 packet

0.16 per 0.14 γ=1 0.12 γ=2 attempts 0.1 γ=3 of

γ=4 0.08 γ=5 number 0.06 γ=6

Mean 0.04 γ=7 0.02 Oγ

0 0.66666 0.99 1.11 1.42 2 3.33 4 4.5 5 6

Net offered load

Figure 4.6: Mean number of attempts per packet for diﬀerent γ and network loads.

Figure 4.7 depicts the average M-EDCF MAC delay. Because collisions are rare in light load networks, the M-EDCF MAC delay is dominated by the CSP phase delay. As shown in Figure 4.7, the performance of the M-EDCF scheme using smaller γ tend to have lower M-EDCF MAC delay compared to those using larger γ. As the network load increases, the number of collisions also increases and, therefore, the M-EDCF CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 95

MAC delay is dominated by the backoff delay as collided packets have to wait for a longer backoff time and sometime the same packet experiences multiple collisions before being successfully transmitted which means longer and longer backoff delays.

0.6

0.5

γ=1 0.4 γ=2 delay(Sec)

γ=3 0.3 MAC

γ=4

γ=5 EDCF

‐ 0.2

M γ=6

γ=7 0.1 Oγ

0 0.66666 0.99 1.11 1.42 2 3.33 4 4.5 6

Net offered load

Figure 4.7: M-EDCF MAC delay for diﬀerent γ and network loads.

Next we study the impact of implementing the M-EDCF scheme on the channel utilization. Figure 4.8 shows the system utilization at diﬀerent values of γ lengths and network loads. The performance gap between longer and shorter γ increases as the network load increases. This is because collisions have higher negative impact on the channel utilization. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 96

0.99

0.98

0.97

0.96 γ=1 0.95 γ=2 utillization(%) γ=3 0.94 γ=4

Channel 0.93 γ=5 γ=6 0.92 γ=7 0.91 Oγ

0.9

Net offered load

Figure 4.8: Channel utilization for diﬀerent γ and network loads.

Figure 4.9 shows the effect of changing γ lengths and the network loads on the throughput. Under light network load all the arriving traffic is delivered before the end of the simulation time. That is because, even with collisions resulting from using shorter γ, there is still enough time to retransmit and deliver the collided packets. This explains the alignment of curves, at light network load, of all used γ. As the network load increases (after traffic rate of 5 per node per second packets), the performance under longer γ tend to perform better than those under shorter lengths. The higher performance gap in the throughput of longer γ result from the fact that longer γ avoid more collisions which saves some bandwidth to deliver more data. Hence, under high network load, longer γ offers higher average throughput than shorter γ. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 97

5.50E+07

5.00E+07

4.50E+07 γ=1

4.00E+07 γ=2

3.50E+07 γ=3

3.00E+07 γ=4 γ=5

Throughput(bps) 2.50E+07 γ=6 2.00E+07 γ=7 1.50E+07 Oγ 1.00E+07 1.4223.3344.55 6 Net offered load

Figure 4.9: Throughput for diﬀerent γ and network loads.

As stated earlier, the M-EDCF scheme possesses two properties: early medium contention termination (Tc) and medium contention selection (Sc). Figure 4.10(a) shows the number of attempts per packet difference between activating and silencing both the Tc and Sc modes. Activating both modes, ATcASc, can further avoid collisions compared to silencing both modes, STcSSc. Assuming longer γ make nodes detect more medium access attempts and hence they can perform a higher number of terminations which can result in fewer collisions. Figure 4.10(b) shows the difference in the M-EDCF MAC delay between the ATcASc and STcSSc modes. The combina- tion of having fewer collisions (i.e., via the termination mode) and selecting first CSQ requests ( i.e., via the selection mode) makes the ATcASc mode have lower M-EDCF

MAC delay compared to the STcSSc mode. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 98

0.45 0.4 Pkt

0.35 per

0.3 0.25 attempts 0.2 of

0.15 0.1 number

0.05

Mean 0 γ=1 γ=2 γ=3 γ=4 γ=5 γ=6 γ=7 γ=8 STcSSc 0.4006 0.2887 0.1289 0.0874 0.0454 0.0291 0.0193 0.0151 ATcASc 0.4006 0.2644 0.0656 0.0321 0.0141 0.0069 0.0041 0.003

(a) Mean number of attempts per packet diﬀerence between ATcASc and

STcSSc modes.

0.75

0.7

0.65

delay(S 0.6 ATcASc

MAC STcSSC 0.55

0.5

0.45 γ=1 γ=2 γ=3 γ=4 γ=5 γ=6 γ=7 γ=8

(b) M-EDCF MAC delay diﬀerence between ATcASc and STcSSc modes.

Figure 4.10: Actuating versus silencing the selection and termination properties CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE 99

4.5 Adaptive γ

Earlier results show the system performance for different γ lengths and network loads. It is clear that, even with a lightly loaded network, adapting longer γ can avoid a very considerable amount of collisions compared to those with smaller γ. We can also observe that longer γ at light network loads tends to have slightly higher M-EDCF MAC delay. Figure 4.11 shows the effects of changing the γ on the channel utilization and on the M-EDCF MAC delay at values of arrival rates, i.e., 1 per node per sec and 1.5 per node per sec packets. Figures 4.11(a) and 4.11(c) shows the medium access delay for different values of γ. Changing the γ produces M-EDCF MAC delay values with a concave up curve shape where the M-EDCF MAC delay first starts at higher values, then it drops, and lastly it increases again. Similarly, changing the γ has a concave down curve affect on the channel utilization as shown in Figures 4.11(b) and 4.11(d). By observing the overall system performance in Figure 4.11, we notice that the best system performance of different traffic loads (1 per node per sec and 1.5 per node per sec packets) happens at different γ values per each traffic load. For example, the maximum channel utilization and the minimum M-EDCF MAC delay for the traffic load of 1 per node per sec packets is achieved when γ = 3, whereas γ = 5 achieves the best system performance when the traffic load equal 1.5 per node per sec packets. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE100

0.059 0.977 0.0585 0.058 0.972 0.0575 0.057 0.967 (Sec) 0.0565 delay utilization(%)

0.056 0.962

MAC 0.0555 Channel 0.055 0.957 0.0545 0.054 0.952 γ=1 γ=2 γ=3 γ=4 γ=5 γ=6 γ=7 γ=8 γ=9 γ=10 γ=1 γ=2 γ=3 γ=4 γ=5 γ=6 γ=7 γ=8 γ=9 γ=10

(a) The effect of changing the γ on the M-EDCF (b) The effect of γ on the channel utilization MAC delay when the traffic load is equal to 1 when the traffic load equal to 1 per node per per node per sec packets sec packets

0.084 0.975 0.083 0.97 0.082 0.965 0.96 0.081 0.955 0.08 0.95 delay(Sec)

0.079 utilization(%)

0.945

MAC 0.078 0.94 Channel 0.077 0.935 0.076 0.93 0.075 0.925 γ=1 γ=2 γ=3 γ=4 γ=5 γ=6 γ=7 γ=8 γ=9 γ=10 γ=1 γ=2 γ=3 γ=4 γ=5 γ=6 γ=7 γ=8 γ=9 γ=10

(c) The effect of changing the γ on the M-EDCF (d) The effect of changing the γ on the channel MAC delay when the traffic load equal to 1.5 per utilization when the traffic load equal to 1.5 per node per sec packets node per sec packets

Figure 4.11: The eﬀect of changing γ

In this section we propose an online scheme by which each contending node, during the runtime and according to the variations of the traﬃc load, adopts its γ to an optimal oγ length so that the selected value avoids collisions and enhances the channel utilization. Our proposed γ adaptability scheme is considered in single-hop CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE101

networks, where all nodes hear the same channel. The proposed scheme is based on observing the channel activity to estimate the number of active antennas v, where

v = Ce × nt. Ce is the number of active nodes (those with data packet(s) ready for

transmission). Ce is estimated by monitoring the channel activity using the carrier

sensing during the virtual transmission (tv) interval. The latter, as shown in Figure 4.12, is the span of time between the observing of the AIFS[AC] period to the end

of a transmission attempt, i.e., successful or collided transmission attempt. The Ce estimation is based on counting the number of consecutive idle slots I in each virtual transmission period. This value is then substituted in the exponential moving average function to update the average E[I] of the current virtual transmission as follows

E[I]n+1 = α × E[I]n + (1 − α) × In+1, (4.1)

where α is smoothing factor. As discussed in [7], the updated average idle slots,

E[I]n+1, and the current transmission probability, P dn+1, can be used to estimate the number of active nodes, ln( E[I]n+1 ) E[I]n+1+Tslot Ccomp = . (4.2) ln(1 − P dn+1)

The computed active nodes, Ccomp, is then substituted in the exponential moving average function to smooth the variations that result from the network traﬃc load ﬂuctuations as follows

Cen+1 = α × Cen + (1 − α) × Ccomp, (4.3)

P dn+1, in Equation (4.2), expresses the probability that a node transmits in a randomly chosen slot [7]. From the updated average of idle slots we can estimate the probability to be 1 P dn+1 = . (4.4) 2 × E[I]n CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE102

In this section we diﬀerentiate between two types of sounding frame transmission windows: γ and E[β]. γ is the sounding frames transmission window allocated by the NiQ to transmit its sounding frames. E[β] is the average sounding frames transmission window lengths considered by the Ce nodes during the previous virtual transmissions. The E[β] computation is performed by observing the channel activity during each tv period and recording the lengths of the sounding frames transmission window,

βi, i = {1, 2,...,Ce}, considered by the contending nodes to transmit their sounding frames. During each virtual transmission and at the end of each SF transmissions of diﬀerent nodes, the NiQ detects the βi length and computes the average E[β]n+1 of the current virtual transmission as follows,

E[β]n+1 = α × E[β]n + (1 − α) × βin+1 . (4.5)

Thus (as shown in Figure 4.12) at the end of each tv interval nodes update the following parameters: P d, Ce, E[I],E[β], where P d is the probability that a node transmits in a randomly chosen slot time, Ce is the estimated number of active nodes, E[I] is the average number of consecutive idle slots, and E[β] is the carried moving average of the sounding frames transmission window lengths of nodes during the previous virtual transmissions. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE103

th th End of the n tv end of the (n+1) tv

Pdn Pdn+1 Cen Cen+1 E[I]n E[I]n+1 Backoff countdown CSP E[β]n E[β]n+1 Transmission attempt AIFS[AC]

β In+1 i

Figure 4.12: Virtual transmission period.

Since all slots of the sounding frames transmission window are equally selected

1 as candidate slots with uniform probability, i.e., ( γ ), we analyze the proposed collision avoidance mechanism in a generic slot (i.e. any randomly chosen). Let m =

1, 2,..., (v + nt) represent the number of contending antennas that may transmit in the generic slot. Of any node, m consists of two contending antennas groups: the antenna group of the NiQ (ntg) and the antenna group of the Ce contending nodes (vg). Let p(m) be the probability that m contending antennas transmit in a generic slot. p(m) is a function of the number of contending antennas of each contending group

1 1 (i.e., ntg and vg), and their transmission probability (i.e., γ and E[β] ), respectively. According to MIMO properties receivers can decode signals if the total number of spatial streams (transmitting antennas) are less than or equal to ϕ. Allowing higher than ϕ antennas to transmit in a slot causes collision and hence must be avoided.

Accordingly we denote ps(0 < m ≤ ϕ) as the probability of success which represents

1 receiving any decode-able signals. Hence, based on the knowledge of the vg, E[β] , and ϕ, the NiQ can adapt its transmission probability via varying its γ to maximize the deﬁned success probability, i.e., permitting any number between one and the spatial CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE104

degree-of-freedom of antennas to transmit. This is done by using the proper probability of transmission. Figure 4.13 shows how the NiQ can vary its sounding frames transmission window to increase the probability of having the desired number of allowed transmitting antennas. Figure 4.13 also shows that varying γ results in three cases i.e., γ < E[β], γ = E[β], and γ > E[β]. Let τ represent the number of slots. By letting τ = max(γ, β), see Figure 4.13, τ can be divided into two subsets of slots: slots set a and b. The ps(0 < m ≤ ϕ) of a slots are aﬀected by the transmission probability of both contending antennas groups, ntg and vg. b slots are only aﬀected by one or the other of the contending antennas group, i.e., either ntg or vg, depending on whichever is longer. Since the slots of set a are the slots impacted by both contending groups (i.e., ntg and vg), we can only study the collision probability in these slots.

E[β] E[β] E[β]

γ γ γ

ntg

a b a a b

Figure 4.13: γ variability

Let Υ represent the sample space of all combinations of transmission number of ´ ntg and vg at a generic slot of slots set a. Let Υ be the outcomes that represent the success events, i.e., having a number of transmit antenna combinations within this CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE105

range 0 < m ≤ ϕ. Therefore, the success outcome events can be formally deﬁned as, ϕ 1 X v 1 1 p (0 < m ≤ ϕ) = (1 − )nt ( )r(1 − )v−r+ s w × γ r E[β] E[β] i r=1 n n nt " nt−j # X nt 1 1 X v 1 1 ( )j(1 − )nt−j ( )r(1 − )v−r (4.6) j w × γ w × γ r E[β] E[β] j=1 i i r=0 n n The ﬁrst term in Equation 4.6 computes the probability that none of the NiQ antennas will transmit multiplied by the probability that from 1 to ϕ antennas of the v antennas will transmit. Similarly, the second term computes the probability that

j, j = 1, 2, . . . , nt antennas of the NiQ will transmit multiplied by the probability

that nt −j antennas of the v antennas will transmit. wi is the weight of a class i which is introduced to modify the the transmission probability considered by different contending nodes to enable medium access differentiation among different access classes. The higher the access class is, the lower the assigned weight factor. Assigning smaller weight factors to wi, permits the contending nodes of higher priority traffic to finish the transmission of their sounding frames earlier and hence they can be selected by the respective receivers before those nodes with higher weight factor. On the other hand, assigning larger weight factors to wi makes nodes with lower priority traffic adapt longer sounding frames transmission window lengths which has the following advantages: 1) nodes with longer γ have more idle slots where they can listen to the channel activity which makes them more attentive to other transmissions (i.g., transmission of higher priority traffics). 2) Longer γ permit other nodes, which adapt to shorter γ, to get selected first. Generally, modifying the transmission probability of nodes via changing their assigned weight factors can be adapted to achieve different desirable performance metrics such as QoS prioritization, early collision resolution, and being more conservative, as explained in Section 4.2. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE106

Using Equation (4.6), Figure 4.14 shows the success probability for diﬀerent values of γ and β for vg = 18 and ntg = 4. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE107

0.8

0.7

0.6 ) s 0.5 β=3

β=4 0.4 β=5

β=6 0.3

Probability of success (P β=7

β=8 0.2 β=9

β=10 0.1

0 0 5 10 15 20 25 30 The node-of-question transmission probability (1/γ)

(a) The case when either vg or their transmission probability is high

0.9

0.85 ) s

0.8

0.75 β=17

Probability of success (P β=19

β=18

β=21 0.7 β=23

β=25

β=27 0.65 0 5 10 15 20 25 30 The node-of-question transmission probability (1/γ)

(b) The case when either vg or their transmission probability is low

Figure 4.14: The success probability as function of diﬀerent values of γ and β CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE108

From the curves in Figure 4.14 the probability of success as a function of γ can be divided into two cases. The first case occurs when the transmission probability of the vg antennas is high (i.e., when β is small). In this case maximizing the success probability requires the NiQ to consider very low transmission probability, i.e., using larger γ. The second case occurs when the transmission probability of the vg antennas is low (i.e., when β is high), in this case the maximum success probability is achieved by considering high transmission probability by the NiQ, shorter SF transmission window lengths. Based on the behavior of Equation (4.6) under different values of γ and β, we formulate an optimization program which can independently used by the NiQ to find the optimal sounding frames transmission window length, oγ. The optimization problem is formulated as follows,

max (ps(0 < m ≤ ϕ)) (4.7)

Subject to:

γmin ≤ γ ≤ γmax (4.8)

where γmax is computed using the following   ps(0 < m ≤ ϕ) γ=i+1 γmax = i, if  ≥ κ , ∀i, i = 1, 2,... (4.9) ps(0 < m ≤ ϕ)  γ=i where the second term in Equation (4.9) represents the eﬀect of changing γ. κ is a constant used as a lower bound to prevent overextending the sounding frames transmission window length. Overextending the window length case happens when either the number of contending antennas vg and/or their transmission probability is very high. In this case the NiQ keeps increasing its sounding frame transmission CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE109

window length as each increased slot achieves a higher probability of a success value. Overextending the SF transmission window length can result in bandwidth wastage because the contending nodes persist for a long time to ﬁnish their sounding frames transmission. Based on the behavior of the probability of success resulting from the formulated function, the optimal value can be found using the nonlinear numerical function,

fminbnd(fun, γmin, γmax), which locally ﬁnds the maximizer of a constrained function with one variable γ. fminbnd() is a nonlinear optimization program similar to ﬁnding the root program with error tolerance threshold [47]. The fminbnd() function is provided with the optimization tool box of MATLAB-7. Accordingly the formulated optimization problem becomes

fminbnd(−(ps(0 < m ≤ ϕ)), γmin, γmax). (4.10)

Algorithm 6 lists the procedure of ﬁnding the optimal SF transmission window length,

Algorithm 6 Finding the optimal SF transmission window

1: Use Equation (4.9) to ﬁnd the γmax. 2: Set γmin = 1 3: Use Equation (4.10) to ﬁnd the optimal value oγ. 4: for ani, ∀i, i = 1, 2, . . . , nt do

5: qani ← uniform(1, oγ) 6: end for

To study the impact of adapting the SF transmission length on the system performance, we use the same simulation model built before. In this model, instead of assigning a ﬁxed γ, nodes exploit the proposed optimization program to ﬁnd the optimal oγ length, the procedure proceeds as follows. First each active node estimates the number active nodes, Ce, and the average SF transmission windows, E[β], used CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE110

during the previous virtual transmissions. Then it ﬁnds its optimal SF transmission window length oγ as explained in Algorithm 6, and lastly it starts transmitting its sounding frames at the respective slots. Figure 4.6 shows that using the optimal SF transmission window length achieves lower number of attempts per transmitted packet for all values of used network loads. Figure 4.7 depicts that using the proper SF transmission window lengths also achieves lower M-EDCF MAC delay compared to using ﬁxed γ. The optimal SF transmission window also achieves higher channel utilization and throughput as shown in Figures 4.8 and 4.9, respectively.

4.6 Simulation Results–Multi-hop network

The inherent complexity of analysis of a multi-hop network together with fact that the behavior of a node is dependent not only on its neighbors’ behavior, but also on the behavior of other unseen nodes makes estimating the active nodes Ce in multi-hop network extremely difficult. This complexity grows exponentially with the number of nodes in the network. We are unaware of any multi-hop analysis that estimates the number of active nodes. However, if any such model becomes available adapting M-EDCF scheme for multi-hop networks is a straightforward plug-in. Hence, in this work we study the impact of exploiting the M-EDCF scheme over multi-hop network for different network traffic loads using fixed γ assignments. Because the medium access is dependent not only on its neighbors’ behavior but also on the behavior of unseen nodes, collisions in multi-hop networks tend to have higher rate compared with single-hop networks. The collision rate increases as the vulnerable time increases. See Figure 4.5, the vulnerable time is the time interval CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE111

between the medium access time of node A to the observation of node C to A’s transmission. Figure 4.15 shows the effect of changing the sounding frames transmission widow on multiple contention area networks. Figure 4.15(a) shows the mean number of attempts per packet for different γ. Evidently, longer γ have higher distinctive performance difference compared with simulation runs that use shorter lengths. The difference in the number of collision for different γ result in a M-EDCF MAC delay, channel utilization, and aggregate throughput difference as shown in Figures 4.15(b), 4.15(c), and 4.15(d), respectively. Note that longer γ achieve lower M-EDCF delay, higher channel utilization, and higher throughput compared with those using shorter γ. CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE112

0.35 0.7

0.3 0.6 pk 0.25 0.5 per

0.2 0.4 γ=1 γ=1 attempts delay(Sec)

0.15 0.3 of γ=2 γ=2 Mac 0.2 0.1 γ=3 γ=3 nuber

0.05 0.1 Mean 0 0

Net offered load Net offered load

(a) Mean number of attempts per packet for dif- (b) M-EDCF MAC delay for diﬀerent γ and net- ferent γ and network loads. work loads.

1 6.00E+07

5.00E+07 0.95 4.00E+07 0.9 γ=1 3.00E+07 γ=1 utilization(%) 0.85 γ=2 2.00E+07 γ=2 Throughput(bps) γ=3 γ=3 channel 0.8 1.00E+07

0.75 0.00E+00

Net offered load Net offered load

Figure 4.15: Multi-hop network performance for diﬀerent value of γ

4.7 Summary

In this chapter we proposed an M-EDCF scheme which exploits the MIMO properties to enhance the IEEE 802.11e EDCF collision-avoidance mechanism. The basic idea of the M-EDCF scheme is sharing the MIMO multiple spatial channels during the medium access period to avoid medium collisions. Spatial channels sharing can help CHAPTER 4. MIMO-AWARE MEDIUM ACCESS COLLISION AVOIDANCE113

transmitters to detect ongoing medium access contentions (i.g., by other transmitters in their vicinity) and hence they can perform early medium contention termination if their transmissions cause the collision. Spatial channels sharing can help intended receivers to perform response coordinations to avoid collisions. We evaluated the M- EDCF scheme over single-hop and multi-hop networks. The M-EDCF scheme results showed considerably fewer collisions for both networks. Activating the contention termination and contention selection modes further reduced the number of collisions compared to those simulation runs with silence mode. To further improve the performance of the M-EDCF scheme, an adaptive spatial channels sharing scheme during the medium access period was proposed. The proposed scheme is based on the online estimation of the number of active nodes. The NiQ uses the estimated active node to ﬁnd the optimal sounding frames transmission window length via an optimization program that boosts the system performance in terms of the number of medium access collisions and channel utilization. Chapter 5

MIMO-aware Bandwidth Utilization

5.1 Introduction

This chapter aims at enhancing bandwidth utilization during TXOP periods. The computation of the TXOP length depends on the following: 1) number of queued packets, 2) the per packet MAC and physical header, medium access coordination overhead, 3) the physical data rate, and the length limit of the MAC frame. Aggre- gating multiple packets in larger frames, generally, minimizes the per packet physical and MAC header overhead and the medium access coordination overheads. The larger the aggregation frames are, the higher the saved bandwidth. Unfortunately, the aggregation frame length is limited by the channel error rate. Consequently, given the channel quality, adapting the aggregation frame length according to the accepted error rate threshold (i.e., the number of error bits per packet that can be corrected by the receiver using a desirable correction techniques) can maximize the gained beneﬁts 114 CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 115

of using larger aggregation frames. As detailed in the previous chapters, the IEEE 802.11n physical enhancements include defining different data rates that can be achieved by a different selection of a number of transmit antennas. Additionally, the IEEE 802.11n MAC enhancements include defining new MAC frame types that can be used to aggregate multiple subframes. Hence, based on the aforementioned physical and MAC enhancements, we propose incorporating the following schemes during the reserved TXOP period to boost bandwidth utilization by reducing the per packet overhead:

• Link adaption: based on online link assessment nodes adapt the appropriate data rate that best suits the instantaneous channel quality.

• Adaptive aggregation frame length: based on the link quality and the used modulation, we compute the probability of error bits, which can be used with the receiver correction threshold to determine a suitable aggregation frame length.

• QoS bandwidth provisioning algorithm: given the allocated bandwidth and the aggregation frame length, we provide an optimization program that distributes the bandwidth to guarantee the required QoS while maintaining fairness among connections.

The remainder of this chapter is organized as follows. Section 5.2 further details the MAC enhancements defined in the IEEE 802.11n amendment. Section 5.3 introduces our proposed mechanisms for link adaptation, adaptive frame length aggregation, and QoS bandwidth provisioning. In Section 5.4, the OPNET Modeler interfaced with MATLAB is used to evaluate the incorporated enhancements under different flow and packet size scenarios. Concluding remarks are given in Section 5.5. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 116

5.2 IEEE 802.11n MAC-Layer Frames

As discussed in Section 2.5, the IEEE 802.11n amendment has enhanced the physical and the MAC layers performance. In this section we further elaborate on the MAC enhancements, i.e., the frame aggregation structure where multiple small subframes are carried in the same frame. In the following sections three diﬀerent aggregation techniques and their constraints (deﬁned in the IEEE 802.11n draft) are explained.

5.2.1 A-MSDU Aggregation Frame Structure

The basic principle of an A-MSDU aggregation is to send multiple MSDUs frames within a single MPDU frame. Figure 5.1 depicts the A-MSDU aggregation structure.

Bytes: 6 62 0-2304 0-3

DA SA Lnth MSDU Padding

Subframe header

Subframe 1 Subframe 2 Subframe 3 Subframe 4 Subframe N

Phy HDR MAC HDR A-MSDU FCS

PSDU DS: Destination address SA: Source address

Figure 5.1: A-MSDU aggregation frame structure

An A-MSDU aggregation is constrained by the following: 1) all MSDUs subframes must have the same traﬃc ID (TID), 2) the A-MSDU expiration occurs only when all A-MSDU’s constituents expire, 3) all the Destination Addresses (DAs) and Sender CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 117

Addresses (SAs) of all aggregated subframes must be the same, i.e., broadcasting or multi-casting is not allowed, 4) the maximum MPDU length that can be transported using A-MSDU aggregation technique is 4095 octets, and 5) The MPDU constructed from multiple aggregated MSDUs subframes cannot be fragmented.

5.2.2 A-MPDU Aggregation Frame Structure

In general, the concept of the A-MPDU aggregation is to join multiple MPDU subframes in order to diminish the physical layer header overhead. Figure 5.2 depicts the A-MPDU aggregation structure.

MPDU delimiter Variable

reserved MPDU lnth CRC delimiter MPDU Padding

Subframe 1 Subframe 2 Subframe 3 Subframe 4 Subframe N

Phy HDR A-MPDU

PSDU

MSDU = MAC service data unit PSDU = Physical service data unit CRC = Cyclic redundancy code

Figure 5.2: A-MPDU aggregation frame structure

However, restricting the aggregating of frames with matching TIDs is not an essential factor with A-MPDUs, but there are other constrains such as: 1) the maximum length that an A-MPDU can obtain is 65535 octets, 2) all the MPDUs within an CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 118

A-MPDU are addressed to the same receiver address, and 3) the maximum number of aggregated subframes is 64.

5.2.3 A-MSDU and A-MPDU Two-level Frame Aggregation

Structure

The two-level frame aggregation comprises a blend of A-MSDU and A-MPDU over two stages as depicted by Figure 5.3. Over the first stage, multiple MSDUs form an A-MSDU with packets that have the same TID. Packets that have different TIDs move over the second aggregation stage where they are packed together with the A-MSDUs from the first stage by using A-MPDU aggregation CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 119

DA SA Lnth MSDU Padding Subframe header

Subframe 1 Subframe 2 Subframe 3 Subframe 4 Subframe N

Phy HDR MAC HDR A-MSDU FCS

MPDU delimiter Variable

reserved MPDU lnth CRC delimiter MPDU Padding

Subframe 1 Subframe 2 Subframe 3 Subframe 4 Subframe N

Phy HDR A-MPDU

PSDU

Figure 5.3: Two level frame aggregation

5.3 Exploiting 802.11n Capabilities to Support QoS

in IEEE 802.11s

Based on the IEEE 802.11n physical and MAC enhancements, we propose incorporating the following mechanisms to enhance the WMN usability and further QoS CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 120

support. The ﬁrst two proposed enhancements are based on assessing the WMN’s link quality to approximate the physical data rate, rphy, and the aggregation frame length,

AFl. The third proposed enhancement is a QoS bandwidth provisioning mechanism implemented in the MAP to perform intelligent bandwidth distribution to further boost QoS support. For simplicity, we assume that channel status to stay constant over the TXOP period. The three proposed enhancements are respectively detailed in the following sections.

5.3.1 Link Adaptation

As stated earlier, diﬀerent combinations of the IEEE 802.11n physical layer parameters values can produce 304 diﬀerent data rates. Based on the modulation and code scheme index parameters, the following formula can be used to compute the actual transmission data rate in Mbps, rphy,

rphy = C × BW × Nss × NBPSC × Cr × GI, (5.1) where C is a constant equal to 0.65, BW is the channel bandwidth factor that is equal to 20 when the 20 MHz channel bandwidth is used and is equal to 40 when the 40

MHz channel bandwidth is used, Nss is the number of spatial streams equal to any value from 1 to 4, GI is the guard interval that is equal to 1 when GI=800 ns and is equal to 1.10769 when GI=400 ns is used, NBPSC is the number of coded bits per subcarrier, and Cr is the code rate [30]. According to the modulation type, NBPSC and Cr take diﬀerent values as depicted in Table 5.1. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 121

Table 5.1: Channel modulation parameters

Modulation Type NBPSC Cr BPSK 1 1/2

QPSK 2 1/2

QPSK 2 3/4

16-QAM 4 1/2

16-QAM 4 3/4

64-QAM 6 2/3

64-QAM 6 3/4

64-QAM 6 5/6

As previously stated, we consider WMNs that utilize the IEEE 802.11n and the STBC is the used coding structure. Accordingly, the link’s SNR is estimated as follows

ρ 2 SNRSTBC = kHdkF (5.2) nt where ρ is the total transmitting power from all antennas per node [57].

In essence, the objective is to adapt the link data rate according to the instantaneous link quality. Algorithm 7 explains the steps for ﬁnding the instantaneous physical data rate. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 122

Algorithm 7 Link data rate adaption INPUT:

• Hd, stored after exchanging SFs.

1: Compute SNR from Equation 5.2.

2: Compute Ci from Equation (3.8) of Chapter 3.

3: Compute max(rphy) , st: rphy + ∆r ≤ Ci, from Equation (5.1) ∀(ModulationT ypes)

5.3.2 Aggregation Frame Length Adaptation

As explained earlier, IEEE 802.11n introduced three different aggregation techniques: A-MSDU, A-MPDU and two-level aggregations. Incorporating frame aggregation enhances the MAC performance by minimizing the MAC headers and medium access overheads are needed to transmit the same amount of data. The disadvantage of considering large aggregation frames is the failure of frame decoding upon experiencing a bad link quality. Retransmitting large frames can dramatically decrease the system utilization. In this section we introduce a new algorithm that estimates the aggregated frame length according to the link quality, SNR, and to the receiver error correction threshold, f. The latter is the upper bound of the number of error bits that can be corrected by the receiver. Let Pf be the probability of f error bits occurring in a packet of N bits in length . The basic idea is to choose an aggregation frame length such that Pf does not exceed a predefined threshold. Therefore, the Pf parameter can be optimized by the MAP to estimate the aggregation frame length, AFl, based on the current link status. The following steps list the procedure to find the aggregation CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 123

frame length:

• Assess the link’s SNR using equation (5.2).

• Using the modulation type determined in Section 5.3.1, ﬁnd the bit error rate

(Pe), from the Pe versus SNR estimated curve plotted using the Modulation Curve Module of OPNET Modeler [55], [76]. Given the aggregation frame

length (N), Pe can be used to ﬁnd the probability of having f error bits using the binomial distribution, N P = P f (1 − P )N−f . (5.3) f f e e

• To ﬁnd the aggregation frame length, the MAP, in each iteration, adds a MSDU

frame, increments the total accumulated number of aggregated bytes, AFl, and

computes the probability of having Pf . If the Pf value exceeds its predeﬁned

value, the MAP returns the previous AFl as the appropriate frame length in bytes.

Given the aggregation frame length, AFl, and TXOPj, computed in Equation (2.4) of Chapter 2, the MAP computes the useful time, Udata, of the TXOPj that is used to send useful data. Udata is the remaining time after subtracting all the physical and

H medium coordination overheads. To compute Udata, ﬁrst let Ophy denote the physical

AF H header overhead, R = AFl − Ophy is the remaining aggregation frame length after subtracting the physical header. Udata can be then computed as follows, U = TXOP − dP AF e × OAF − OTXOPj , (5.4) data j TXOPj T MC

P AF is the possible number of aggregation frames transmitted in the TXOP TXOPj j CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 124

period TXOP − OTXOPj P AF = j MC , (5.5) TXOPj OAF + RAF T rphy AF where OT in Equations (5.4) and (5.5) is the total overhead required to transmit one aggregation frame H AF Ack Ophy OT = 2 ∗ SIFS + + . (5.6) rphy rphy

TXOPj OMC in Equations (5.4) and (5.5) is the MAC coordination overhead that is re-

TXOPj quired by the MAC protocol to perform AIFS[AC], BF, RTS, CTS, and SIFS. OMC is given by

TXOPj OMC = AIF S[AC] + bk + RTS + CTS + 2 ∗ SIFS (5.7)

bk is the moving average and computed as follows

bk = α × bk + (1 − α) × bk (5.8)

where α is a smoothing factor parameter.

Udata is then passed to the optimization program, described in the next section, to ﬁnd the optimal bandwidth distribution among diﬀerent classes.

5.3.3 Bandwidth Provisioning Scheme

We propose an adaptive bandwidth provisioning scheme (modiﬁed from High Speed Downlink Packet Access (HSDPA) [1]), to the MAP in WMNs, which enables QoS

guarantees and fairness by optimally distributing the Udata, among the contending connections.

Given rphy computed in Section 5.3.1, Udata computed in Section 5.3.2, in addition to other locally known information such as the queue size, qk, and the mean MSDU CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 125

size, mk, of the k class, we devise an optimization program that can optimally distrib-

k ute the bandwidth among access classes. Let breq(A) represent the time in seconds required to empty the kth queue class using the Ath aggregation technique and is computed as follows:  k k H H q (m +sm )+OMAC  r if A = AMSDU; k  phy breq(A) = (5.9) qk(mk+DL+OH )  MAC if A = AMP DU. rphy H H where sm is the subframe header (DS, SA, frame length, and padding), OMAC is the MAC header, and DL is the MPDU delimiter. Also, let ak represent the actual allocated time in seconds to the kth traﬃc class. The following optimization program can optimally distribute the bandwidth, k X j j j max w × a × breq(A) (5.10) 1≤j≤k j=1 subject to k X j a = Udata (5.11) j=1

k k k lb ≤ a ≤ breq(A) (5.12) where wi is a class prioritization weight assigned to diﬀerent traﬃc, and lbk is the

th k lower bound time assigned to k traﬃc class and computed as follows: let Ftx denote

th the average number of transmitted frames of the k class during each TXOPj period.

k k k Ftx = α × Ftx + (1 − α)Ftx. (5.13)

Also let Ftx represent the total number of frames transmitted from all access classes each TXOPj period.

Ftx = α × Ftx + (1 − α)Ftx. (5.14) CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 126

The lower bound bandwidth of each class is computed as  k  1 − Ftx lbk = β × Ftx × U (5.15)  k data Pk (1 − Ftx ) i=1 Ftx where β is the percentage of bandwidth that the kth traffic class is voluntarily re- linquishing to other flows. Although the optimization program Equation (5.10) may favor flows with higher queue sizes, fairness is guaranteed by Equation (5.12). Upon computation of ak by the optimization program for the k classes, the MAP starts transmitting data from k classes based on their bandwidth shares, ak.

5.4 Performance Evaluation

In this section, we ﬁrst begin by describing the simulation model, network topology, and the traﬃc model. We then introduce the other simulation parameters and performance metrics. Next we provide detailed discussions and comparisons of the simulation results.

5.4.1 Simulation Model

We have extended the MIMO-aware IEEE 802.11e MAC protocol modiﬁed in Chapter 3 to include the following.

• Adapting the data transmission rate of each transmit antenna element module according to the computed physical data rate.

• Building the aggregation frame format structure (i.e., the A-MSDU and A- MPDU) using Packet Module of OPNET. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 127

• Modifying the sending frame function of the original model to perform frame aggregates according to the desired aggregation frame format and length.

5.4.2 Network Topology

A WMN topology similar to the one in Figure 5.4 that incorporates the proposed enhancements is used. The back-haul topology consists of MAPs uniformly distributed around the MPP with distance d ranging between 100 to 300 meters. In this topology, each MAP utilizes the IEEE 802.11e HCCA MAC protocol reference scheduler to coordinate medium access of ui users, where i = {1, 2,...,Z} and Z is the number

of mesh back-haul access points. In this section we let Z = 5 and ui is varied in each scenario to examine diﬀerent system performance aspects.

External Network

MP Portal MP AP MP AP

MP AP MP AP MP AP

Figure 5.4: Wireless mesh network model CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 128

5.4.3 Traﬃc Model

Using built-in OPNET traffic modules, each mesh client is configured to generate one of the following traffic types, i.e., either VoIP, Vi, File Transfer Protocol (FTP), or Hypertext Transfer Protocol (HTTP). The traffic specifications of each traffic type are detailed in Table 5.2. A traffic class of the same traffic type may generate traffic with different data rates that is confined in the specified range per each traffic type. The traffic received, from associated users, by each MAP, is queued in four queue classes. After that the traffic is forwarded to the mesh portal.

Table 5.2: Traﬃc speciﬁcations

TSPEC VoIP (G.729A) Video (MPEG-4) FTP HTTP

Mean data rate 26-64 kb/s 1-1.5 Mb/s 400-600 kb/s 350-600 kb/s

delay bound 100 ms 150 ms - -

Nominal MSDU 160 octets 1024 octets 1200 octets 500 octets

Max service interval (MSI) 25 ms 30 ms - -

5.4.4 Performance Metrics

We use the following performance metrics to evaluate our schemes.

• Dropped packet ratio: percentage of the total dropped to the total received.

• Combined TXOPj length: The instantaneous total TXOP length requested by all connections of a MAP. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 129

5.4.5 Simulation Results

In this section, we study the performance of the proposed schemes for different traffic flow combinations and network loads. In all the experiments 90% confidence levels are maintained with 10% confidence intervals based on 10 independent runs [3]. In the first scenario, we associate each MAP with four different traffic types. Each MAP is associated with four users. Figure 5.5 shows the effect of the incorporated enhancements on the medium access delay. Due to shortening the wasted bandwidth consumed by the MAC headers and MAC medium accesses, sending data using A- MSDU or two-level aggregations tend to have lower MAC delay compared with the case without data aggregation. Larger aggregated frames promote higher MAC service savings, compared with A-MSDU, the two-level aggregation allows packing of mixed traffic types. These aspects make the two-level aggregation technique result in lower MAC delay compared to those techniques which send data with A-MSDU aggregation structure. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 130

0.2 0.18 0.16 0.14 0.12 (Sec)

y 0.1 WO‐agg

dela 0.08 A‐MSDU agg 0.06 Two‐level agg 0.04 0.02 0 FTP HTTP VI VO

Traffic flow type

Figure 5.5: Medium access delay for diﬀerent access classes using diﬀerent aggregation techniques.

In addition to achieving lower MAC delay, the saved bandwidth from aggregation can also be used to support more access classes. Figure 5.6 depicts the throughput performance diﬀerence among A-MSDU, two-level, and without aggregation techniques. Results show that schemes incorporating data aggregation outperform those which do not. As they achieve higher throughput, schemes implementing the aggregation technique also tend to have lower packet dropping ratios as shown in Figure 5.7. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 131

1.60E+06

1.40E+06

1.20E+06 (bps) 1.00E+06

8.00E+05 WO‐agg

6.00E+05 A‐MSDU agg

Throughput Two‐level agg 4.00E+05

2.00E+05

0.00E+00 FTP HTTP VI VO

Traffic flow type

Figure 5.6: Throughput comparison of diﬀerent access classes using diﬀerent aggregation techniques.

70 60 50 ratio(%) 40 cket

a 30 p

20 10 Dropped 0 FTP HTTP VI VO WO‐agg 23.47507565 58.84854432 12.10179421 24.40300949 A‐MSDU agg 0.015926103 0.081779522 0.027462468 4.530585541 Two‐level agg 0.063704412 0.098135427 0.032039546 2.502453386

Figure 5.7: Packet drop ratio of diﬀerent access classes using diﬀerent aggregation structures

Good link quality can be utilized to send at higher data rates, which can shorten CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 132

the service time required to transmit the same amount of data. This means nodes can finish earlier before their allocated transmission opportunity. The saved bandwidth can either be dedicated for further data transmission by the transmission opportunity’s owner or can be added to the contention period time. Figure 5.8 shows a snapshot of 7 seconds of the adaptive link’s transmission opportunity compared to that of fixed data rate for video traffic class. The horizontal line in Figure 5.8 represents the transmission opportunity value of video traffic class using a fixed physical rate (100 Mbps), similar to IEEE 802.11n physical data rate, while the dashed line represent the transmission opportunity values using link adaptation. Points below the fixed data rate line represent the saved bandwidth that has been added to the system resources.

0.0009

0.0008 (sec)

0.0007 0.0006 length 0.0005

OPj LA‐TXOPj X 0.0004 T

0.0003 FR‐TXOPj 0.0002

Combined 0.0001 0 02468 Time (Sec)

Figure 5.8: The allocated transmission opportunity diﬀerence between link adaptation (LA-T XOP j) and using ﬁxed physical transmission rate (FR- T XOP j)

In the second scenario we examine our scheme under diﬀerent network loads by CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 133

increasing the number of network access classes. Figure 5.9 relates the MAC delay to the increasing number of access classes, where the ratio between different traffic, i.e. voice, video, FTP, HTTP, is set to 1:1:1:1. As the combined transmission opportunities by MAP grow with an increasing the number of flows, the possibility of more bandwidth savings, contributed by shortening the MAC service time, also increases. According to the exchanged traffic specifications, each admitted traffic class is granted with an adequate transmission opportunity, txopi,j. Due to the bursty nature of traf-

fic, the allocated txopi,j might not be sufficient for some access classes with higher QoS constraints or bad link qualities. Combining the transmission opportunities of a larger number of access classes permits flexible data packing management which saves bandwidth for backlogged QoS flows. Consequently, the saved bandwidth is dynamically reallocated to backlogged flows based on QoS requirements.

0.04

0.035

0.03 FTP 0.025

ay(Sec) HTTP l de 0.02 VI 0.015 VO Average 0.01

0.005

0 4 8 12 16 24 Number of traffic flows

Figure 5.9: Medium access delay versus increasing network load using two-level aggregation structure

Larger combined transmission opportunities also allows accommodating more data CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 134

per each transmission opportunity period. Figure 5.10 shows the throughput for all access classes with and without data aggregation. The performance gap between the results with and without data aggregation expands as the number of access classes increases due to the advantage of using more combined transmission opportunities.

3.50E+07

3.00E+07

2.50E+07

2.00E+07 W‐agg 1.50E+07 throughput(bps) WO‐agg

Total 1.00E+07

5.00E+06

0.00E+00 4 8 12 16 24

Number of traffic flows

Figure 5.10: The total throughput of all access classes with and without using aggregation technique

In addition to the effect of combining transmission opportunities, the data aggregation technique is also affected by the packed packet sizes. In the third scenario, we evaluate the performance of our schemes under different packet sizes and the number of access classes. In this scenario, we increase the voice access classes number, i.e., in each simulation run the access classes load is changed to: 1, 3, 6, 12, 18 voice access classes. As each frame causes MAC service overhead, transmitting small packet sizes severely wastes the scarce bandwidth. On the other hand, aggregating small frames allows a larger number of frames accommodation. These properties cause a large performance difference gap between the case with and without data aggregation in CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 135

terms of MAC delay as shown in Figure 5.11.

WO‐agg W‐agg 0.1 0.09 0.08 0.07

) 0.06 0.05

Delay(Sec 0.04 0.03 0.02 0.01 0 1 3 6 12 18 Number of Voice flows (N)

Figure 5.11: The MAC delay versus increasing voice network traﬃc load with and without using aggregation technique

Figure 5.12 shows the total achieved throughput under a small packet size scenario. As clearly shown in the ﬁgure, increasing the transmission opportunity via increasing the number of access classes and using small packet sizes can achieve higher throughput. CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 136

1.60E+06

1.40E+06

1.20E+06

1.00E+06 WO‐agg 8.00E+05

throughput(bps) W‐agg 6.00E+05

Total 4.00E+05

2.00E+05

0.00E+00 1 3 6 12 18 Number of Voice flows(N)

Figure 5.12: The achieved throughput for increasingly voice network traﬃc load

5.5 Summary

In this chapter, we proposed enhancement schemes which exploit the IEEE 802.11n physical and MAC enhancements to improve the IEEE 802.11s QoS support constrained by link bandwidth deﬁcit and QoS scheduling. The proposed schemes include: 1) adapting the appropriate data rate based on the instantaneous channel quality, which is done by selecting the proper modulation and code scheme index, 2) an adaptive aggregation frame length based on the instantaneous link quality to maximize the advantage of using large frame aggregation and to prevent large frame retransmissions, 3) a balanced optimization program that distributes the bandwidth to guarantee the QoS requirements, while maintaining fairness among connections. Online interaction between OPNET Modeler and MATLAB was used to evaluate the incorporated schemes. Performance results demonstrated that incorporating link CHAPTER 5. MIMO-AWARE BANDWIDTH UTILIZATION 137

adaptation increases the achieved throughput. Exploiting frame aggregation techniques reduced the MAC delay, and thus, decreased the packet dropping ratio. As shown by the performance results, packet aggregation techniques have a crucial impact on system performance when aggregating small packet sizes. Chapter 6

Conclusions and Future Work

Wireless mesh networks (WMNs) have increasingly become a popular option for providing high speed network access to users in the context of home, enterprise, and community networks [10], [54], and [66]. Infrastructure-based WMNs consist of stati- cally positioned mesh points. Such back-haul network architecture is reliable, scalable, cost-effective, and easy to deploy. Among other technologies, legacy IEEE 802.11 has been used in implementing WMNs. Supporting Quality of Service (QoS) to enable multimedia services is foreseen to be vital for the success of next generation WMNs. The IEEE 802.11 Medium Access Control (MAC) layer is often viewed as the “brain” of the wireless networks, as it provides a variety of functions that support the operation of 802.11-based wireless LANs. Hence to enhance QoS support in WMNs an efficient cross-layer design of a capable MAC protocol is the key to success. Because the IEEE 802.11n amendment combines both the latest advancements in the physical and MAC layer techniques, it delivers higher data rate and better link quality than previously realized versions. We believe that designing an efficient cross-layer MAC protocol that utilizes the IEEE 802.11n new defined physical and MAC enhancements

138 CHAPTER 6. CONCLUSIONS AND FUTURE WORK 139

is the key to success for multimedia services support in WMNs. Accordingly, in this thesis we studied the problem of designing and developing an eﬃcient QoS MIMO- aware MAC Protocol (QMMP). QMMP is a distributed MAC protocol that fully exploits the IEEE 802.11n physical and MAC enhancements to boost QoS support in WMNs. In the remainder of this chapter, we summarize and discuss the conclusions from this thesis and provide directions for future research work.

6.1 Summary of Contributions

We classiﬁed the QoS MIMO-aware MAC protocol framework into three related levels, namely MIMO-aware bandwidth sharing MAC protocol, MIMO-aware collision- avoidance mechanism, and MIMO-aware bandwidth utilization schemes. The framework was designed to simultaneously achieve and balance the following objectives:

• Eﬃciently utilizing the higher achieved data rate of MIMO systems.

• Reducing the encountered medium access delay via allowing multiple concurrent communications.

• Reducing medium access collision.

• Enhancing the system usability of the WMNs.

• Supporting QoS diﬀerentiation among admitted traﬃc while maintaining fairness.

• Exploiting the physical and MAC enhancements deﬁned in the IEEE 802.11n. CHAPTER 6. CONCLUSIONS AND FUTURE WORK 140

In Chapter 3, we introduced a novel QoS MIMO-aware MAC Protocol (QMMP) which is MIMO aware medium access control protocol which efficiently utilizes the increased data rate and higher channel quality offered by MIMO systems by virtue of assessing the links and then translating the required data rate into antenna requirements. The functional operation of QMMP MAC protocol is divided into four main phases, and during each phase an intelligent operational aspect is executed. Namely; 1) medium contention phase (i.e., medium access coordination procedure adapted by the contending nodes) , 2) the channel sounding phase (i.e., exchanging small frames over all antennas between the transmitter and its intended receiver to assess the channel between them; nodes around them also can use the overhead sounding frames to assess the channel between themselves to the transmitting nodes), 3) TXOP scheduling and reservation phase (i.e., a procedure that uses the channel state information produced during the channel sounding phase to schedule a TXOP period with the desired bandwidth (antennas) and to broadcast its reservation attributes using RTS CTS exchange), and 4) TXOP phase (i.e., the period during which the connection pairs start and end their data Ack packet exchanges). Given data awaiting transmission, the start of the QMMP medium access attempt is based on observing the spatial streams remaining. Nodes have to defer transmission if no spatial channels are available. The operation of the QMMP scheme also exploits the IEEE 802.11 NAV table to decide when to defer transmission and when the available sources are at different interval. Before a node can attempt to send a frame, the total number of used spatial streams in its NAV table must be less than the spatial degree-of-freedom. CHAPTER 6. CONCLUSIONS AND FUTURE WORK 141

The operations of QMMP MAC protocol were expanded to accommodate for the hidden node problem. Such a problem causes the number of reserved antennas to be different at the transmitter and receiver sites. To address this problem, QMMP searches for all possible TXOP slots on each site and during the reservation phase (i.e., exchanging RTS CTS frames) both sites decide on a desirable matching TXOP slot. The performance of the QMMP scheme was tested under different scenarios for different required rates, interference impacts, and communication environments. The results were compared to those of the MIMO-aware modified IEEE 802.11e MAC protocol. Results showed that our proposed QMMP scheme outperforms the IEEE 802.11n modified model in medium access delay and throughput. Results also showed that nodes can always attain their requested rate. In Chapter 4, we introduced a novel MIMO-aware EDCF (M-EDCF). M-EDCF exploits the MIMO multiple spatial channels to enhance the IEEE 802.11e EDCF collision-avoidance mechanism. The basic idea of M-EDCF is sharing the multiple spatial channels during the medium contention period to avert medium access collisions. Spatial channels sharing means instead of accessing the medium using all spatial channels (i.e., all antennas), a node uses only a set of the available spatial channels (i.e., at each time interval only subset of antennas are active). As the concurrent spatial channels used are fewer than or equal to the spatial degree-of- freedom, receivers can decode the transmitted signals. Managing the sharing of the spatial channels of each connection pair can be steered to produce other advantages such as medium contention termination and medium contention selection. Medium contention termination refers to the capability of pausing the ongoing medium access CHAPTER 6. CONCLUSIONS AND FUTURE WORK 142

attempt if the accessing node concurrently observes other medium access attempts by other nodes. However, contention selection refers to the capability of observing multiple concurrent medium access attempts of diﬀerent transmitters. This capability allows receivers to coordinate their responses which may include responding to any particular node to satisfy a performance metric such as minimizing the medium access delay. Responses can also be coordinated to avoid potential collisions.

In addition, in Chapter 4, we also proposed optimal spatial channels sharing that based on the network load. The optimal spatial channels sharing is based on estimating the number of active nodes to determine the spatial channels sharing value that avoids potential collisions and at the same time and it is does not waste bandwidth by using longer sounding frames transmission windows (i.e., the number of slots during which a node starts and ﬁnishes its sounding frames transmissions). The performance of the M-EDCF scheme was tested over single-hop and multi-hop networks. The results obtained showed considerably fewer collisions for both networks. Activating the contention termination and contention selection modes further reduces the number of collisions compared to those simulation runs with silence mode. The results of the adaptive spatial channels sharing demonstrated consistent improved performance under diﬀerent network load conditions. In Chapter 5, further bandwidth management during the TXOP period was considered. To fully utilize the allocated bandwidth during the TXOP period, we proposed enhancement schemes based on exploiting the IEEE 802.11n physical and MAC layers. The schemes entail the following: 1) based on the high achieved data rate of the IEEE 802.11n amendment, we proposed adapting an appropriate data rate based CHAPTER 6. CONCLUSIONS AND FUTURE WORK 143

on the instantaneous channel quality. Such a scheme can accommodate for channel rate variability at different time intervals. The basic idea was to first assess the channel quality and then select a proper modulation and code scheme index from defined list. 2) Based on the different aggregation frame types defined by the IEEE 802.11n amendment, we propose an adaptive aggregation frame length based on the instantaneous link quality. Such a scheme aims at maximizing the advantage of using large aggregation frames and to prevent large frame retransmissions. 3) Based on the determined aggregation frame length, we designed a balanced optimization program that packs packets according to their QoS requirements while maintaining fairness among overall connections. Performance results demonstrated that incorporating link adaptation increases the achieved throughput. Exploiting the frame aggregation technique, on the other hand, reduced the MAC delay, and thus, decreased the packet drop ratio. As shown by the performance results, packet aggregation techniques had a crucial impact on system performance when aggregating small packet size.

6.2 Future Research Directions

There are several directions by which the work in this thesis can be extended. In this section, we highlight some of these directions.

We have shown in Chapter 3 the operational design of our MIMO-aware MAC protocol. Based on sharing the spatial channels, QMMP enables multiple concurrent CHAPTER 6. CONCLUSIONS AND FUTURE WORK 144

communication of multiple connection pairs. As discussed in Chapter 3, the connection pairs can implement independent coding and decoding during their reserved TXOP period. Further extensions during this period include independent coding and decoding per link. To illustrate this, designing a good code that improves the performance in terms of the channel quality and data rate is limited by the decoding complexity at the receiver. There are many receivers that utilize different decoding techniques such as Zero Forcing (ZF), Minimum Mean Squared Error (MMSE), Successive Cancellation (SUC), Ordered Successive Cancellation (OSUC), etc. These receivers are different in their goals and in the techniques they use. A solid extension to our MAC protocol would be designing and incorporating different decoding techniques that are based on the per link requirements. The vision is having multiple concurrent communication with each link having its own coding and decoding technique. The QMMP scheme only activates antennas that achieve the required data rate. Hence, incorporating antenna selection techniques and beamforming signal processing during the TXOP period of independent connection pair would be another potential extension to the QMMP to achieve higher performance. In Chapter 4 a MIMO-aware collision-avoidance scheme is proposed. The core idea for avoiding collisions is based on varying the sounding frames transmission window length based on the online active nodes. Estimating the active node is only available for single-hop network. To the best of our knowledge there is no work that analytically estimates the number of active nodes in a multi-hop networks. However, proposing a multi-hop active nodes estimation scheme and incorporating it in QMMP can further reduce the effect of the hidden node problem. As explained in Section 4.2 varying the sounding frames transmission window CHAPTER 6. CONCLUSIONS AND FUTURE WORK 145

length has other properties such as early medium contention termination and medium contention selection. The former property allows transmitters to listen to the channel activity while they are contending and hence to decide whether to continue or terminate the medium access attempt. This property can be exploited to provide QoS differentiation where different nodes adapt their sounding frames transmission widow lengths based on their priority level. Adaptive schemes that couple collision avoidance with QoS-differentiation support during selecting the sounding frames transmission window is a possible extension. In Chapter 5 MIMO-aware bandwidth utilization schemes were proposed. Namely, physical data rate adaptation, aggregation frame length adaptation, and QoS based packet packing algorithm. According to our system, each node performs two operations: 1) packing packets inside one of the IEEE 802.11n defined aggregation frames, i.e., A-MSDU aggregation frame, A-MPDU aggregation frame, or A-MSDU A-MPDU combined aggregation frame. 2) Sending the aggregated frame during the allocated TXOP. In Chapter 5, we only considered packing packets according to QoS requirements and fairness among connections. Further extension would be optimizing the aggregation frame lengths not only based on the channel error rate but also on the TXOP length. The designed scheme would include the difference of the defined aggregation frames, their constraints as discussed in Section 5.2, QoS differentiation, and maximizing the TXOP utilization. Bibliography

[1] Al-Manthari B., Abu-Ali N., Nasser N., and Hassanein, H. “Dynamic Bandwidth Provisioning with Fairness and Revenue Considerations for Broadband Wireless Communication”. IEEE International Conference on Communications, pages 217–225, May 2008.

[2] Alamouti S. M. “A Simple Transmit Diversity Technique for Wireless Commu- nications”. IEEE JOURNAL on Select Areas in Communications, pages 1451– 1458, Oct 1998.

[3] Altman D., Machin D., Bryan T., and Gardner M. “Statistics With Conﬁdence: Conﬁdence Intervals and Statistical Guidelines”. British Medical Journal, page 240, May 2000.

[4] Bharghavan V. “Performance evaluation of algorithms for wireless medium access”. IEEE Computer Performance and Dependability Symposium, pages 86–95, Sept 1998.

[5] Bianchi G. “Performance analysis of the IEEE 802.11 distributed coordination function”. IEEE Journal on Selected Areas in Communications, pages 535–547, March 2000.

146 BIBLIOGRAPHY 147

[6] Bolcskei H., Gesbert D., and Paulraj A. J. “On the capacity of OFDM-based spatial multiplexing systems”. IEEE Transaction on Communications, pages 225–234, Dec 2002.

[7] Cali F., Conti M., and Gregori E. “IEEE 802.11 protocol: design and performance evaluation of an adaptive backoﬀ mechanism”. IEEE Journal on Selected Areas in Communications, pages 1774–1786, Sept 2000.

[8] Chatzimisios P., Boucouvalas A. C., Vitsas V., Vafiadis A., Economidis A., and Huang P. “A simple and effective backoff scheme for the IEEE 802.11 MAC protocol”. International Conference on Cybernetics and Information Technologies, Systems and Applications, pages 546–555, July 2005.

[9] Chen C., Luo H., Seo E., Vaidya N.H., and Xudong W. “Rate-Adaptive Framing for Interfered Wireless Networks”. IEEE International Conference on Computer Communications, pages 1325–1333, May 2007.

[10] Cisco Wireless Mesh Networks solutions. www.cisco.com.

[11] Dan J., Khalim A. Shami A., and Primak S. “A Novel MIMO-Aware Distributed Media Access Control Scheme for IEEE 802.11 Wireless Local Area Networks”. IEEE Conference on Local Computer Networks, pages 125–132, 2007.

[12] Deng J., and Haas J. “Dual busy tone multiple access (DBTMA): a new medium access control for packet radio networks”. IEEE International Conference on University Personal Communications, pages 973–977, Oct 1998. BIBLIOGRAPHY 148

[13] Draves R., Padhye J., and Zill, B. “Routing in multi-radio, multi-hop wireless mesh networks”. IEEE international conference on Mobile computing and networking, pages 114–128, May 2004.

[14] Foschini G. J. “Layered space-time architecture for wireless communication in a fading environment when using multielement antennas”. Bell Labs Technology, pages 41–59, July 1996.

[15] Ganguly S., Kim N. V. K., Kashyap A., Niculescu D., Hong R. I. S., and Das, S. “Performance optimization for deploying VoIP services in mesh networks”. IEEE Journal on Selected Areas in Communications, pages 2147–2158, Sept 2006.

[16] Gesbert D., Shaﬁ M., Da-shan S., Smith P. J., and Naguib A. “From theory to practice: an overview of MIMO space-time coded wireless systems”. IEEE Journal on Selected Areas in Communications, pages 281–302, April 2003.

[17] Gobriel S., Melhem R., and Mosse D. “BLAM: An Energy -Aware MAC Layer Enhancement for Wireless Ad Hoc Networks”. IEEE Wireless Communication and Networking Conference, pages 340–348, June 2004.

[18] googleblog.blogspot.com/2006/08/free-citywide-wiﬁ-in-mountain-view.html. 8/16/2006.

[19] Guey J., Fitz M. P., Bell M. R., and Kuo W. “Signal design for transmitter diversity wireless communication systems over Rayleigh fading channels”. IEEE Transaction on Communications, pages 136–140, Aril 1999.

[20] Haas J, and Deng J. “Dual busy tone multiple access (DBTMA)-performance evaluation”. IEEE Vehicular Technology Conference, pages 314–319, May 1999. BIBLIOGRAPHY 149

[21] Haas J, Deng J., and Tabrizi S. “Collision free medium access control scheme for ad-hoc networks”. IEEE Military Communication Conference, pages 276–280, Nov 1999.

[22] Haas Z.J., and Jing D. “Dual busy tone multiple access (DBTMA)-a multiple access control scheme for ad hoc networks”. IEEE Transactions on Communi- cations, pages 975–985, Jun 2002.

[23] Heiskala J., and Terry J. “OFDM Wireless LANs: A Theoretical and Practical Guide”. Sams, Indianapolis, IN, USA, 2001.

[24] Hu J., and Raymond C.D. “A statistics based design of MAC protocols with distributed collision resolution for Ad Hoc networks”. ACM International Workshop on Mobility Management and Wireless Access, pages 917–922, June 2005.

[25] Huang F., and Yang Y. “Energy eﬃcient collision avoidance MAC protocol in wireless mesh access networks”. IEEE international conference on Wireless communications and mobile computing, pages 272–277, Oct 2007.

[26] Hwangnam K., and Jennifer C. “Improving protocol capacity with model-based frame scheduling in IEEE 802.11-operated WLANs”. international conference on Mobile computing and networking, pages 190–204, July 2003.

[27] IEEE 802.11-2007. “Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Speciﬁcations”. IEEE Standard, Last updated 2007.

[28] IEEE Amendment 8: 802.11e. “Medium Access Control (MAC) Enhancements for Quality of Service (QoS)”. IEEE Draft, June 2005. BIBLIOGRAPHY 150

[29] IEEE Draft Amendment, P802.11s/D2.0,. “ ESS Mesh Networking”. IEEE Draft, March 2008.

[30] IEEE P802.11n/D4.00. “Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) speciﬁcations: Enhancements for Higher Throughput”. IEEE Draft, March 2008.

[31] Iyengar R., Sharma V., Kar K., and Sikdar B. “Analysis of Contention-Based Multi-Channel Wireless MAC for Point-to-Multipoint Networks”. IEEE Inter- national Symposium on on World of Wireless, Mobile and Multimedia Networks, pages 453–455, May 2006.

[32] Jaehyuk C., and Joon Y. “EBA: An Enhancement of the IEEE 802.11 DCF via Distributed Reservation”. IEEE Transactions on Mobile Computing, pages 378–390, June 2005.

[33] Jafar S. A. and Fakhereddin M. J. “Degrees of Freedom for the MIMO Interfer- ence Channel”. IEEE Transaction on Information Theory, pages 962–968, July 2007.

[34] Jeonghoon M., Wilson H. S., and Walrand J. “Comparison of multi-channel MAC protocols”. ACM international symposium on Modelling, analysis and simulation of wireless and mobile systems, pages 209–218, May 2005.

[35] Joon-Sang P., Nandan A., Gerla M., and Heechoon Lee. “SPACE-MAC: enabling spatial reuse using MIMO channel-aware MAC”. IEEE International Conference on Communications, pages 3642–3646, May 2005. BIBLIOGRAPHY 151

[36] Jun P., Liang C., and Biplab S. “A Wireless MAC Protocol with Collision Detection”. IEEE Transactions on Mobile Computing, pages 1357–1369, Oct 2007.

[37] Kone V., Das S., Zhao B. Y., and Zheng H. “QUORUM: quality of service in wireless mesh networks”. The Journal of SPECIAL ISSUES on Mobility of Systems, Users, Data and Computing, pages 358–369, Oct 2007.

[38] Kulkarni G., Nandan A., Gerla M. , and Srivastava M. “MIMAC: A Rate Adap- tive MAC Protocol for MIMO-based Wireless Networks”. UCLA Computer Sci- ence Department Technical Report number 040035, 2004.

[39] Kwon Y., Fang Y., and Latchman H. “Novel MAC Protocol with Fast Collision Resolution for Wireless LANs”. IEEE INFOCOM, April 2001.

[40] Larsson E. G., Stoica P., and Ganesan G. “Space-Time Block Coding for Wireless Communications”. Cambridge University Press, New York, NY, USA, 2003.

[41] Li X., Jin S., Gao X., and McKay, R. “Transmitter Optimization and Optimality of Beamforming for MIMO Systems in the Presence of Double Scattering”. IEEE Global Telecommunications Conference, pages 3639–03644, Nov 2007.

[42] Lin Y., and Wong V. “Frame Aggregation and Optimal Frame Size Adaptation for IEEE 802.11n WLANs”. IEEE Global Telecommunications Conference, pages 1–6, Nov 2006.

[43] Lu K., Wang J., Wu D., and Fang Y. “Performance of a burst-framebased CSMA/CA protocol for high data rate ultra-wideband networks: analysis and BIBLIOGRAPHY 152

enhancement”. International conference on Quality of service in heterogeneous wired/wireless networks, page 11, June 2006.

[44] Lu K., Wu D., and Fang Y. “A novel framework for medium access control in ultra-wideband ad hoc networks, dynamics of continuous, Discrete and Impulsive Systems”. IEEE Transaction on Vehicular Technology, pages 427–441, Dec 2005.

[45] Malli M. “Adaptive Fair Channel Allocation for QoS Enhancement in IEEE 802.11 Wireless LAN”. IEEE International Conference on Communications, pages 2081–2093, July 2004.

[46] Manaseer S., and Ould-Khaoua M. “Logarithmic based backoﬀ algorithm for MAC protocol in MANETs”. Technical Report, University of Glasgow, 2006.

[47] MathWorks. www.mathworks.com.

[48] Mehmet S., and Tugcu T. “A survey on emerging broadband wireless access technologies”. IEEE Computer Networking., pages 3013–3046, June 2007.

[49] Michelle X.G., Scott F. M., and Shiwen M. “A cross-layer approach to channel assignment in wireless ad hoc networks”. IEEE Mobile Network and Application, pages 43–56, Nov 2007.

[50] Minyoung P., Heath R. W., and Nettles, S.M. “Improving throughput and fairness for MIMO ad hoc networks using antenna selection diversity”. Global Telecommunications Conference, pages 3363–3367, April 2004.

[51] Mirkovic J., Orfanos G., Reumerman, H.-J., and Denteneer D. “A MAC Protocol for MIMO Based IEEE 802.11 Wireless Local Area Networks”. IEEE Wireless Communications and Networking Conference, pages 2131–2136, March 2007. BIBLIOGRAPHY 153

[52] Mundarath J. C., Ramanathan P., and Van Veen B. D. “NULLHOC : a MAC protocol for adaptive antenna array based wireless ad hoc networks in multipath environments”. IEEE Global Telecommunications Conference, pages 2765–2769, Dec 2004.

[53] Naffa A., Ksentini A., and Mehaoua A. “Sliding Contention Window (SCW):Towards Backoff Range-Based Service Differentiation over IEEE 802.11 Wireless LAN Networks”. IEEE Network, pages 45–51, July 2005.

[54] Nortel Wireless Mesh Networks solutions. www.nortel.com.

[55] OPNET Modeler 14.5. www.opnet.com.

[56] Paulraj A. J. and Kailath T. “Increasing capacity in wireless broadcast systems using distributed transmission/directional reception”. Patent, (5,345,599), 1994.

[57] Paulraj A., Nabar R., and Gore D. “Introduction to Space-Time Wireless Com- munication”. Cambridge University Press, 2003.

[58] Pellegrini F. D., Maguolo F., Zanella A., and Zorzi, M. “A cross-layer solution for VoIP over IEEE 802.11”. Wireless Personal Multimedia Communications Symposium, 2005.

[59] Raghavendra R. “IPAC An IP-based Adaptive Packet Concatenation for Mul- tihop Wireless Networks”. Asilomar Conference on Systems, Signals and Com- puting, pages 2147–2153, Nov 2006.

[60] Raheleh B., Dilmaghani, and Ramesh R. “Hybrid wireless mesh network deployment: a communication testbed for disaster scenarios”. International Conference on Mobile Computing and Networking, pages 90–98, 2006. BIBLIOGRAPHY 154

[61] Raniwala A., Gopalan K., and Chiueh T. “Centralized channel assignment and routing algorithms for multi-channel wireless mesh networks”. ACM SIGMO- BILE Mobile Computing and Communications, pages 50–65, April 2004.

[62] Redi J., Watson B., Ramanathan R., and Basu P. “Design and implementation of a MIMO MAC protocol for ad hoc networking”. Wireless Sensing and Processing, pages 231–238, May 2006.

[63] Rezki Z., Haccoun D., Gagnon F., and Ajib W. “Impact of Spatial Correlation on the Finite-SNR Diversity-Multiplexing Tradeoﬀ”. IEEE Transaction on Wireless Communications, pages 1184–1189, April 2008.

[64] Romaszko S., and Blondia C. “Neighbour -Aware, Collision Avoidance MAC Protocol (NCMac) for Mobile Ad Hoc Networks”. IEEE International Sympo- sium on Wireless Communication Systems, pages 322–326, Sept 2006.

[65] Romdhani L., Ni Q., and Turletti T. “Adaptive EDCF: Enhanced Service Dif- ferentiation for IEEE 802.11 Wireless Ad-Hoc Networks”. IEEE Wireless Com- munication and Networking Conference, pages 1373–1378, May 2003.

[66] Strix Wireless Mesh Networks solutions. www.strixsystems.com.

[67] Sundaresan K., Sivakumar R., Ingram M. A., Chang T. “Medium access control in ad hoc networks with MIMO links: optimization considerations and algorithms”. IEEE Transactions Mobile Computing, pages 350–365, Dec 2004.

[68] Taifour M., Nait-Abdesselam F., and Simplot-Ryl D. “Neighborhood backoﬀ algorithm for optimizing bandwidth in single hop wireless ad-hoc networks”. BIBLIOGRAPHY 155

IEEE International Workshop on Mobility Management and Wireless Access, June 2005.

[69] Talucci F., Gerla M., and Fratta, L. “MACA-BI (MACA by invitation)-a receiver oriented access protocol for wireless multihop networks”. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, pages 435– 439, Sept 1997.

[70] Tarokh V., Jafarkhani H., and Calderbank A. “Space-time block codes from orthogonal designs”. IEEE Transactions on Information Theory, pages 1456– 1467, July 1999.

[71] Tarokh V., Seshadri N., and Calderbank A. R. “Signal design for transmitter diversity wireless communication systems over Rayleigh fading channels”. IEEE Transactions on Information Theory, pages 744–765, Mar 1998.

[72] Telatar I. E. “Capacity of multi-antenna Gaussian channels”. European Trans, pages 585–595, May 1999.

[73] Tse D. and Viswanath P. “Fundamentals of Wireless Communication”. Cam- bridge University Press, 2005.

[74] Wang B., and Mutka M. “QoS-aware fair rate allocation in wireless mesh networks”. IEEE Computer Communincaton, pages 1276–1289, July 2008.

[75] Wang W., Liew S.C., and Li V. O. K. “Solutions to performance problems in VoIP over a 802.11 wireless LAN”. IEEE Transaction on Vehicular Technology, pages 366–384, Nov 2005. BIBLIOGRAPHY 156

[76] Zhao Y., Adve R., and Lim T. J. “Optimal STBC precoding with channel covariance feedback for minimum error probability”. Journal on Advances in Signal Processing, pages 1257–1265, June 2004.

[77] Zheng L. and Tse D. N. C. “Diversity and multiplexing: A fundamental tradeoﬀ in multiple-antenna channels”. IEEE Transaction on Information Theory, page 10731096, May 2003.

[78] ZyTrax Communication. www.zytrax.com. February 2007.