Documente Academic
Documente Profesional
Documente Cultură
THSE
Prsente
par
Aymen Belghith
Professeur, ENSI Manouba Tunisie Professeur, LRI Universit Paris Sud Orsay Professeur, IFSIC Universit de Rennes 1 Matre de confrence, Tlcom Paris Tech Professeur, Tlcom Bretagne Matre de confrence, Tlcom Bretagne
Examinateurs
MM. : Bernard Marceaux
Cousin Coupechoux
Encadrants
MM. : Xavier Lout
Lagrange Nuaymi
Remerciements
Je remercie Abdelfettah Belghith, Professeur, ENSI Manouba Tunisie, et Khaldoun Al
Agha, Professeur, LRI Universit Paris Sud Orsay, d'avoir bien voulu accepter la charge de
rapporteur.
Je remercie Bernard Cousin, Professeur, IFSIC Universit de Rennes 1, et Marceaux Coupechoux, Matre de confrence, Tlcom Paris Tech, d'avoir bien voulu juger ce travail.
Je remercie Xavier Lagrange, Professeur, Tlcom Bretagne, et Lout Nuaymi, Matre de confrence, Tlcom Bretagne, qui ont dirig ma thse.
Je tiens aussi remercier tous les membres de ma famille pour leurs encouragements.
Table of Contents
1 Introduction 2 Overview of WiMAX
2.1 WiMAX Historical Background and Main Elements . . . . . . . . . . . . . . . . . 2.1.1 2.1.2 2.1.3 2.2 2.2.1 2.2.2 Evolution of the Cellular Systems . . . . . . . . . . . . . . . . . . . . . . . IEEE 802 Wireless Standards and WiMAX Creation . . . . . . . . . . . . WiMAX Main Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transmission Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . Frame Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2.1 2.2.2.2 2.2.3 2.3 2.3.1 Downlink Frame Structure . . . . . . . . . . . . . . . . . . . . . Uplink Frame Structure . . . . . . . . . . . . . . . . . . . . . . .
5 11
11 11 12 13 14 14 17 17 20 21 22 23 24 26 28 29 30 30 31 31 31 31 32 32 34 35
Physical layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Useful Throughput Computations . . . . . . . . . . . . . . . . . . . . . . . Service Flow Management . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1.1 2.3.1.2 2.3.1.3 Dynamic Service Addition . . . . . . . . . . . . . . . . . . . . . . Dynamic Service Change . . . . . . . . . . . . . . . . . . . . . . Dynamic Service Delete . . . . . . . . . . . . . . . . . . . . . . . UGS QoS class characteristics . . . . . . . . . . . . . . . . . . . ertPS QoS class characteristics . . . . . . . . . . . . . . . . . . . rtPS QoS class characteristics . . . . . . . . . . . . . . . . . . . . nrtPS QoS class characteristics . . . . . . . . . . . . . . . . . . . BE QoS class characteristics . . . . . . . . . . . . . . . . . . . .
2.3.2
2.4 2.5
Link Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scheduling in WiMAX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 2.5.2 Scheduling in WiMAX in PMP mode . . . . . . . . . . . . . . . . . . . . . Scheduling in WiMAX in mesh mode . . . . . . . . . . . . . . . . . . . . .
2.6
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Table of Contents
3 Design and Implementation of a WiMAX Module with QoS for NS-2 Simulator 37
3.1 3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . QoS-included WiMAX Module Description . . . . . . . . . . . . . . . . . . . . . . 3.2.1 3.2.2 3.2.3 Link Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Service Management . . . . . . . . . . . . . . . . . . . . . . . . . Request Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3.1 3.2.3.2 3.2.4 3.2.4.1 3.2.4.2 3.2.4.3 3.3 3.3.1 3.3.2 3.3.3 3.4 Unicast request opportunities . . . . . . . . . . . . . . . . . . . . Contention request opportunities . . . . . . . . . . . . . . . . . . Unsolicited Grant Service QoS class scheduling . . . . . . . . . . Real-time Polling QoS class scheduling . . . . . . . . . . . . . . . Best Eort QoS class scheduling . . . . . . . . . . . . . . . . . . 37 38 39 40 41 42 42 43 43 45 45 46 48 49 51 52
Scheduling Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Results of our QoS-included WiMAX Module . . . . . . . . . . . . . . . . . . . . Behavior of the UGS Scheduler . . . . . . . . . . . . . . . . . . . . . . . . Behavior of the rtPS Scheduler . . . . . . . . . . . . . . . . . . . . . . . . Behavior of the BE Scheduler . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Scheduling in WiMAX
4.1 4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Classical Schedulers Used in WiMAX . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Systematic Schedulers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1.1 4.2.1.2 4.2.1.3 4.2.1.4 4.2.1.5 4.2.2 4.2.2.1 4.2.2.2 4.2.2.3 4.3 4.3.1 Round Robin scheduler . . . . . . . . . . . . . . . . . . . . . . . Prorate scheduler . . . . . . . . . . . . . . . . . . . . . . . . . . Weighted Round Robin scheduler . . . . . . . . . . . . . . . . . . Decit Round Robin scheduler . . . . . . . . . . . . . . . . . . . Example of systematic schedulers applied in WiMAX context . . maximum Signal-to-Interference scheduler . . . . . . . . . . . . . Temporary Removal Scheduler . . . . . . . . . . . . . . . . . . . Opportunistic Decit Round Robin scheduler . . . . . . . . . . .
53
53 54 54 54 54 55 55 56 58 58 59 60 62 62 62 63 63 65
Schedulers Specically Proposed for WiMAX . . . . . . . . . . . . . . . . . . . . Schedulers Proposed for a Specic WiMAX QoS Class . . . . . . . . . . . 4.3.1.1 4.3.1.2 4.3.1.3 4.3.2 Adaptive rtPS scheduler . . . . . . . . . . . . . . . . . . . . . . . Adaptive Polling Service (aPS) . . . . . . . . . . . . . . . . . . . Adaptive Bandwidth Allocation Scheme (ABAS) . . . . . . . . .
Table of Contents 4.3.2.1 4.3.2.2 4.3.2.3 4.3.2.4 4.3.2.5 4.4 4.4.1 4.4.2 Uplink packet scheduler with a Token Bucket Call Admission Control (CAC) mechanism . . . . . . . . . . . . . . . . . . . . . Cross-layer Scheduling Algorithm with QoS Support . . . . . . . Hybrid Scheduling Algorithm . . . . . . . . . . . . . . . . . . . . Frame Registry Tree Scheduler (FRTS) . . . . . . . . . . . . . . Scheduler ensuring QoS requirements . . . . . . . . . . . . . . .
65 65 67 68 69 71 72 73 73 73 74 79 81 82 83
Proposal for the rtPS QoS Class . . . . . . . . . . . . . . . . . . . . . . . . . . . Modied maximum Signal-to-Interference Ratio (mmSIR) scheduler description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Performance evaluation of the mmSIR scheduler . . . . . . . . . . . . . . 4.4.2.1 4.4.2.2 Simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison between the mSIR and mmSIR Schedulers . . . . .
4.5 4.6
Synthesis of Dierent Schedulers Deployed in WiMAX . . . . . . . . . . . . . . . 4.6.1 4.6.2 4.6.3 Bad Radio Channel Scenario . . . . . . . . . . . . . . . . . . . . . . . . . Fluctuant Radio Channel Scenario . . . . . . . . . . . . . . . . . . . . . . Good Radio Channel Scenario . . . . . . . . . . . . . . . . . . . . . . . . .
4.7
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
85 86 86 86 87 88 88 88 88 89 89 90 90 90 90 91 91 93 94
Description of the Pricing Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . A Basic Pricing Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Proposed Pricing Scheme . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2.1 5.3.2.2 5.3.2.3 UGS users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . rtPS users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BE users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4
Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
User SNR and MCS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 User Arrivals and Departures . . . . . . . . . . . . . . . . . . . . . . . . . 100
Performance Evaluation of the Basic Pricing Scheme . . . . . . . . . . . . . . . . 101 Performance Evaluation of the Proposed Pricing Scheme . . . . . . . . . . . . . . 103 5.7.1 5.7.2 5.7.3 Eect of the Reference Price Pref . . . . . . . . . . . . . . . . . . . . . . . 104 Choice of the Parameter CrtPS 5.7.3.1 5.7.3.2 5.7.4 . . . . . . . . . . . . . . . . . . . . . . . . 106 Best Eort Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Fixed pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Auction-based Best Eort pricing . . . . . . . . . . . . . . . . . 111
5.8
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6 Conclusion
115
A Ordonnancement et tarication des ressources radio dans les rseaux WiMAX117 List of Acronyms Publications Bibliography List of Figures List of Tables 125 129 136 137 141
Chapter 1
Introduction
Thesis Contents
Over the past two decades, the wireless networks have known a tremendous expansion. In some places, the wireless networks have a wider market than the wired networks. The wireless access is now denitely a need for a large majority of people. In the nineties the voice service was the main need, quickly joined by the cunning and light Short Message Service (SMS) services. As for wired services, new services are expected or already included in wireless networks and are generating a new growth. Hence, wireless telecommunication networks aect dierent services like the email, le transfer, Internet-style data networks, voice-mail, phone, radio, television and many others some of them not being yet known. These applications do not have the same Quality of Service (QoS) requirements. For example, it is well-known that the phone service is a real-time application that has stringent delay and rate requirements. While the email allows us to reach people in a deferred manner ; this non-real-time application does not need QoS guarantees. In a multimedia system, it is possible that a mobile terminal simultaneously accesses services having dierent performance requirements. This diversity is already provided by some wired networks such as in Digital Subscriber Line (DSL) networks. This technology allows simultaneous voice, video, and data communication. The user can surf the web and talk on the phone in the same time and over the same access line without need to disconnect either its voice, internet or video connection. In a wireless multimedia network, the simultaneous transmissions of more than one service, each with its proper requirements lead the operator to have a system that can meet the dierent performance requirements of the dierent network services. In a wireless network, the channel radio characteristics are time-variant and the radio resource is scarce. In order to provide dierent services with high eciency of the frequency use, the radio resource should be allocated taking into account the QoS requirements of the users while trying to maximize the network throughput. Among the radio resource algorithms, scheduling is of particular importance in the system performance. In this thesis, we consider the Worldwide Interoperability for Microwave Access (WiMAX) 5
Introduction
system which is an important Broadband Wireless Access (BWA). At the beginning, WiMAX was the name of the certication organism (Forum) for the IEEE 802.16 standard. Now it is also a standardization body (mainly for the network denition) and as it was the case for WiFi/IEEE 802.11, WiMAX is the common name for the IEEE 802.16 standard. In this document we will sometimes refer to IEEE 802.16 standard as WiMAX, although WiMAX will be mainly used as the name of the technology. The WiMAX system provides both high throughput and large coverage. WiMAX support up to thousands of connections between subscriber stations (SSs) and the Base Station (BS) in Non Line of Sight (NLOS) conditions. These connections can have dierent QoS requirements simultaneously. The rst application of the WiMAX system is the deployment of wireless metropolitan area network (WMAN). WiMAX can provide a broadband Internet connection and support multiple communication services. Another possibility is the use of the WiMAX system as a wireless backhaul to connect client/subscriber access sections of networks with their core switching and management topologies. But the main expected use of WiMAX is as a Mobile Broadband Access System where WiMAX is a potential competing technology for UMTS or its evolutions such Long Term Evolution (LTE). The physical layer of WiMAX is based on the Orthogonal Frequency Division Multiplexing (OFDM) in its two main variants : OFDM and OFDMA. The OFDM is a very ecient transmission technique, before being a multiplexing technique. In addition, an OFDM system is a very open one which leaves doors open for many optimization possibilities. Much research have already been done for dierent WiMAX topics and we mention some of the published works in the related parts in this document. Yet, much have still to be done. More details of the WiMAX system are presented in Chapter 2. Like in any wireless network, the radio resource of WiMAX should be eciently allocated to the dierent users. In WiMAX networks, the resource allocation is dynamic and done at the beginning of each frame. The frame duration is constant but this frame is divided into variablesize parts, each of them allocated to one or several users. An important aspect of the resource allocation is that the number of bits transmitted in the frame depends on the radio channel characteristics of the dierent subscribers : the better the channel conditions are, the higher the number of bits is transmitted. This is the now well-known principle of link adaptation. The resource distribution also has to take into account the characteristics of each of the connections to be served, i. e., the system scheduler must take into account the QoS requirements of the connections. WiMAX radio resource allocation proposals are studied in simulation. We consider Network Simulator (NS-2), an open-source software. We use the network simulation in order to evaluate some scheduling algorithms and estimate the WiMAX capacity. So, we implement a QoS-including WiMAX module in NS-2. The starting point for this simulation module is the implementation done by National Institute of Standards and Technology (NIST). We add all WiMAX QoS classes as well as the WiMAX mechanisms used for the addition of connections as specied by the IEEE 802.16 standard. We also implement the mechanisms required to send
bandwidth requests in order to specify the amount of radio resource to reserve. As the IEEE 802.16 standard supports many Modulation and Coding Scheme (MCS) and the MCS to use depends on the channel quality of the subscriber, we add a link adaptation mechanism. We then implement some scheduling algorithms in order to compare between them and estimate the WiMAX capacity that depends on the scheduler used and the radio channel characteristics. As scheduling in WiMAX is a very active eld, we present a review of scheduling algorithms that can be applied in a WiMAX context. A scheduler can be classical or specically proposed for WiMAX context. We then propose a scheduler for WiMAX real-time applications and a synthesis of the use of dierent schedulers in WiMAX. We nally extend our WiMAX studies to WiMAX pricing policy analysis. An important topic related to radio resource allocation is the pricing policy. This issue has received relatively less attention in the dierent wireless networks and specically WiMAX. Yet, it has important eects of the system eciency. This is a particularly important issue in WiMAX, where ve QoS classes exist in addition to many standard-dened tools related to pricing. In addition, the WiMAX frame is very dynamic and its burst distribution can be changed in each frame. On the other hand, the pricing scheme proposed for this very open system has to consider one or many criteria. As any other wireless system, the criteria can be the radio resource use eciency, users data rate (or indicators based on this parameter), operator revenue, users' comfort and possibly others. A suitable pricing model allows the operator to dene its parameters according to these considerations. We propose a detailed pricing framework for WiMAX ; some elements in this model are rather simple the proposed framework can be extended to more complicated considerations. Hence, it is easy to reuse in future works by ourselves or colleagues. We propose an ecient pricing scheme which includes both for real-time and non-real-time applications. The pricing models proposed for real-time applications take into account the congestion periods. The models also dierentiate between the prices paid for the rates required to satisfy the real-time connections constraints and the prices paid for the extra rates. We propose two pricing schemes for non-real-time (Best Eort) applications. They are based on xed (dened by the operator) and variable (determined by auction) allocation unity price. In order to evaluate our proposed pricing schemes, we also consider a basic pricing model where all users are charged the same price independently of their QoS class and the network conditions. We compare between our proposed (sophisticated) pricing model and the basic pricing model. We rst determine the suitable parameters of the sophisticated pricing models in order to maximize the network revenue of the operator. We then show that the sophisticated pricing model outperforms the basic model and provides higher revenue for the operator. Furthermore, this is not at the expense of users'comfort.
Introduction
Document Organization
Chapter 2
presents an overview of WiMAX. We briey describe the transmission technologies supported by the WiMAX PHY layer. We consider only the Orthogonal Frequency Division Multiplexing (OFDM) layer. We describe the frame structure. A frame consists of a downlink subframe (for transmission from the Base station) and an uplink subframe (for transmission for the Subscriber Station). Since the OFDM symbol represents the unit of allocation when using the OFDM PHY layer, we present the computations for the OFDM symbol duration and then the number of useful symbols per frame. The useful throughput can be computed when all users have the same radio channel characteristics. Since the radio channel is variable and each subscriber has its own characteristics, the network simulation can be used in order to estimate the WiMAX capacity. In the second part of this Chapter, we present briey the dierent sublayers of the MAC layer. Then, we present the service ow management that consists of an addition of a new connection, a change of parameters of an existing connection, and a delete of an existing connection. This management is based on the exchange of Dynamic Service management messages between the Base Station and the Subscribers Stations. We then describe the dierent QoS classes dened by the IEEE 802.16 standard and its amendment 802.16e. These QoS classes are useful for the determination of the QoS requirements of the connections. We also describe the link adaption since WiMAX supports many Modulation and Coding Schemes (MCSs) and the MCS used depends on the Signal to Noise Ratio (SNR) of the subscriber. We nally introduce scheduling in WiMAX in both point-to-multipoint (PMP) and mesh modes. We consider only the PMP mode.
Chapter 3
presents the design and implementation of our WiMAX module with QoS for NS-2
Simulator. This WiMAX module is used for the performance evaluation of our proposed scheduling mechanism. We then estimate the WiMAX capacity using dierent schedulers in dierent SNR environments. The implemented WiMAX module supports all QoS classes, the service ow management mechanisms, the unicast and contention request opportunities (used to send the bandwidth requests as specied by the standard), the link adaptation, and schedulers for three QoS classes (UGS, rtPS, and BE). These three QoS classes provide an heterogenous WiMAX environment as they are designed for real-time (xed and variable bandwidth request sizes) and non-real-time applications, respectively. After describing our implementation, a validation of the QoS-included WiMAX module is presented by studying the behavior of the UGS, rtPS, and BE schedulers.
Chapter 4
classical or specically proposed for WiMAX. We classify the classical schedulers into systematic schedulers that do not take into account neither the radio channel conditions nor the WiMAX QoS requirements and radio channel conditions-aware schedulers. We classify the schedulers
specically proposed for WiMAX into schedulers proposed for a single WiMAX QoS class and schedulers that can be applied on many WiMAX QoS classes. In this Chapter, we also describe our proposals for the rtPS QoS class. Then we evaluate the proposed scheduler using our QoSincluded WiMAX module. Finally we compare between some schedulers and show the impact of the schedulers as well as the radio channel conditions on the estimation of the WiMAX capacity.
Chapter 5
presents our proposed pricing models for the UGS, rtPS, and BE QoS classes. The
pricing model for the UGS QoS class takes into account the congestion periods when dening the price actually paid. The pricing model for the rtPS QoS class also takes into account the congestion periods when dening the price actually paid and dierentiates between the price paid for reserving the mandatory minimum reserved rate and for reserving an extra rate. The pricing model for the BE QoS class is based on the denition of the symbol price. The symbol price is determined manually by the operator when the Fixed Symbol Price Model (FSPM) is applied. While the symbol price is determined automatically using auction when the Variable Symbol Price Model (VSPM) is applied. We then show that the scheduler used has to take into account the characteristics of the pricing model. Finally, we evaluate our sophisticated pricing model by comparing it with a basic pricing model. In the basic pricing model, all subscribers (whatever their QoS classes) are charged the same price.
Chapter 6
concludes the achieved work by presenting the main contributions and suggests
10
Introduction
Chapter 2
Overview of WiMAX
2.1 WiMAX Historical Background and Main Elements
2.1.1 Evolution of the Cellular Systems
The last decades witnessed a tremendous evolution in communications and, specically, in wireless communications. The rst generation (1G) of mobile networks was introduced in the late seventies. The 1G systems like Advanced Mobile Phone Service (AMPS), Total Access Communication System (TACS), Radiocom 2000 and others were analog and used to transfer a unique service which was voice. The second generation (2G) of mobile networks started to be deployed in the beginning of the 1990s. The main 2G mobile network, and the most successfully, by far, is the Global System for Mobile communications (GSM) [LGT00]. The rst and unique service was voice quickly joined by Short Message Service (SMS). The so-called second-and-half generation (2.5G or 2G+) such as General Packet Radio Service (GPRS) [SSP02] and Enhanced Data Rates for GSM Evolution (EDGE) [HRM03] added packet data services and higher data rate mainly for Internet-style access and e-mail. The theoretical maximum rate in the GPRS system is 115 Kbps. The EDGE system provides better theoretical maximum rate (up to 384 Kbps). Today, users need wireless high speed Internet access. Moreover, users wants to be able to access the Internet from a large area. In the late of 1990s, the third generation (3G) completed the 2G system in many geographical areas. The 3G system can support multimedia, data, video, and other services along with voice. The main 3G systems are the Universal Mobile Telecommunications System (UMTS) [HH02] and CDMA2000 [DVV04]. The rst deployment of CDMA2000 and UMTS took place in 2000-2001. Yet, the 3G systems are still in evolution. The rst data rates were in the magnitude of 1 Mbps. Nowadays much higher data rates are expected in both uplink and downlink with the High Speed Downlink Packet Access (HSDPA) and the High Speed Uplink Packet Access (HSUPA) evolutions [HT06] (see, for example, Release 7 of UMTS). Apart from the displayed physical data rates, application-level data rates are smaller. For example, in 2007, the HSUPA could reach 1 Mbps for a File Transfer Protocol (FTP) application [Kaf07]. 11
12
Overview of WiMAX
More recent versions of HSUPA have higher gures. The next step after 3G is (evidently) 4G or what is also known as B3G (Beyond 3G). The main 4G system is currently Long Term Evolution (LTE). The ocial name of this technology is Evolved Packet System (EPS). According to presently displayed gures, the LTE network provides a downlink throughput of 100 Mbps for Single In Single Out (SISO) antennas and 173 Mbps for 2x2 Multiple In Multiple Out (MIMO) antennas [Jac08]. The third Generation Partnership Project (3GPP) is developing the LTE Advanced standard in order to have a 4G access technology running from 2010.
13
April 2008, Nokia ocially unveiled the "N810 WiMAX Edition Internet Tablet" [Cha08]. In January 2009, Motorola unveiled the "Motorola wi4 WiMAX CPEi 775" device [Wri09]. It is expected that Apple, HTC and, Android show up their WiMAX devices in 2009 [Gab08]. The success of the WiMAX devices depends on the WiMAX network deployment. Until now, WiMAX is being deployed by major telecom operators around the world. Today, WiMAX is deployed in Japan [Noh08], Taiwan [Com06a], India [Mac08], Latin America (rst WiMAX deployment in Latin America was in Columbia by Siemens) [Com06b], France ("Free" is the single national licence holder of WiMAX). WiMAX is already deployed in some regions like Normandie and Alsace and it is expected de be deployed in all the regions before December 2010) [San08], Belgium [LVO+ 07], South Africa [url08b], Tunisia (rst deployment in Sfax by Divona Telecom) [url07a] and elsewhere. Many other countries are still interested in the deployment of WiMAX. For example, several East countries will deploy WiMAX in 2009 [url08a]. South Korea will deploy Wireless Broadband (WiBro) voice calls also in 2009 [Par08]. However, WiMAX is not deployed on huge scale world wide yet. Due to the slow WiMAX roll out, Nokia canceled the "N810 WiMAX Tablet" in January 2009 [Hic09].
14
Overview of WiMAX
are supported in WiMAX : Time Division Duplexing (TDD) and Frequency Division Duplexing (FDD). The TDD mode uses the same frequency in the downlink direction (from the BS to the SS) as in the uplink direction (from the SS to the BS). Two dierent frequencies are used in the FDD duplexing mode. In this Chapter, we present the dierent transmission technologies supported by the WiMAX physical layer, the uplink and downlink frame structures. Since we consider the OFDM PHY layer, we describe the OFDM symbol structure. The OFDM symbol represents the allocation unit of scheduling in this layer. We also present the computations of the OFDM symbol duration which are useful for the WiMAX capacity estimation when all SSs have the same channel radio conditions and all symbols are allocated. Then, we present the dierent sub-layers of the MAC layer, the management of the service ow that represents the MAC transport service, the dierent WiMAX QoS classes that the service ow can belong to, and the link adaptation in order to take into account the channel quality. Finally, we introduce scheduling in WiMAX.
Physical layer
Tab. 2.1 Number of Useful Bits per OFDM Symbol
15
MCS BPSK 1/2 QPSK 1/2 QPSK 3/4 16-QAM 1/2 16-QAM 3/4 64-QAM 2/3 64-QAM 3/4
Coding rate (c )
192 1 1/2 = 96 192 2 1/2 = 192 192 2 3/4 = 288 192 4 1/2 = 384 192 4 3/4 = 576 192 6 2/3 = 768 192 6 3/4 = 864
Wireless High-speed Unlicensed Metropolitan Area Networks (WirelessHUMAN) Air Interface : the standard does not specify the transmission technique to use (SC, OFDM, or OFDMA). Only the TDD duplexing mode is supported. This interface is used for NLOS transmission in the 2-11 GHz frequency band. WiMAX retains only the OFDM and OFDMA transmission techniques. In our work, we consider only the WirelessMAN-OFDM physical layer. In the OFDM physical layer, the unit of transmission is the OFDM symbol. OFDM uses an Inverse Fast Fourier Transform (IFFT) in order to generate a signal formed by orthogonal subcarriers. The number of subcarriers is generally large and is an integer power of 2. For example, there are 256 subcarriers in OFDM PHY layer. However, the generated subcarriers cannot totally be used for data transmission ; 192 subcarriers for data transmission, 8 pilot subcarriers for synchronization, and 56 null subcarriers for guard band. WiMAX supports dierent modulations and dierent coding rates. The supported modulations are Binary Phase Shift Keying (BPSK), Quadrature Phase Shift Keying (QPSK), 16Quadrature Amplitude Modulation (16-QAM), and 64-QAM. Let d be the number of bits transmitted per modulation symbol. Parameter d has one value among 1, 2, 4, 6. Let c be the coding rate of the Forward Error Correction (FEC). The supported coding rates are 1/2, 2/3, and 3/4. The IEEE 802.16 standard presents a combination between the modulation and the coding rate. The combination is called Modulation and Coding Scheme (MCS). The dierent MCS supported are shown in Table 2.1. In that Table, we present the number of useful bits per OFDM symbol depending on the MCS used. This number is simply :
k = Nused d c ,
where Nused is the number of data carriers (192 for OFDM PHY layer).
(2.1)
The basic radio resource allocated for transmission in the OFDM PHY layer is the OFDM symbol. Hence, it is useful to compute the number of OFDM symbols per frame. The number of OFDM symbols per frame is constant. First we compute the duration of an OFDM symbol. An
16
Overview of WiMAX
OFDM symbol is represented in Fig. 2.1. It includes the useful part of the symbol, which lasts
Tb , and a copy of the last period of the useful symbol, termed as Cyclic Prex (CP ). This prex
is used to provide robustness in multi-path. Let Tg be the duration of the cyclic prex. We have Ts = Tb + Tg . Generally the duration of the prex depends on the duration of the useful part. It is then . Four possible useful to introduce the ratio of CP time to useful time denoted by G : G = Tg b values are proposed in the standard : 1/4, 1/8, 1/16, and 1/32 (see Table 213 of [STS04]). Then :
T
Ts = Tb (1 + G ).
(2.2)
The useful time (Tb ) is equal to the inverse of the sampling frequency (fs ) multiplied by the total number of subcarriers (NFTT ). For the OFDM PHY layer, the total number of subcarriers is equal to 256 (NFTT = 256). As fs is equal to the channel bandwidth (BW) multiplied by the sampling factor (n ) and using (2.2), the time duration of an OFDM symbol (Ts ) can be computed as follows :
NFTT (1 + G ) (2.3) n BW Knowing the values of BW , n , and G , the OFDM symbol duration can be computed. In our Ts =
work, we consider BW = 5MHz and G = 1/4. The sampling factor (n ) is equal to 144/125 when the channel bandwidth is a multiple of 1.25 MHz and not a multiple of 1.75 MHz (n = 8/7) or 1.5 MHz (n = 86/75), see Table 213 of [STS04]. So, we consider a sampling factor of 144/125. Using (2.3), the OFDM symbol duration is equal to 55.5 s. After determining the OFDM symbol duration, the number of OFDM symbols per frame can be computed for a determined frame duration. The possible values of a frame duration are 2.0ms , 2.5ms , 4ms , 5ms , 8ms , 10ms ,
12.5ms , and 20ms (see Table 384a of [STS06]). We consider a frame duration of 20 ms and then
the total number of OFDM symbols per frame is equal to 360. Once the number of useful symbols is determined, the useful throughput can be computed and then we can estimate the WiMAX capacity. The frame duration remains constant and the Base Station (BS) has to distribute those OFDM symbols in order to allow the data transmission in the downlink and uplink directions as well as transmission of management messages. In the following Section, we present the structure of the downlink and uplink frames.
Physical layer
17
18
Overview of WiMAX
The FCH is followed by one or multiple downlink bursts, each one of them is transmitted with a burst prole and contains MAC Protocol Data Units (PDUs). These downlink bursts are transmitted in a decreasing robustness order. For example, with the use of a single Forward Error Correction (FEC) type, data begin with BPSK modulation, followed by QPSK, 16-QAM and nally 64-QAM. Each downlink burst consists of an integer number of OFDM symbols. Especially, the rst downlink burst contains the broadcast messages : downlink map (DLMAP) and uplink map (UL-MAP). These messages are possibly followed immediately by the Downlink and Uplink Channel Descriptors (DCD and UCD, respectively). DL-MAP and ULMAP indicate the location of the downlink and uplink bursts, respectively (see Fig. 2.5). DCD and UCD describe the physical characteristics of the downlink and uplink channels, respectively. The DL-MAP contains : four elds (Management Message Type, Base Station identier, ...) with a length equal to 64 bits, and as many elds as the number of SSs served in the downlink direction. These elds are called DL-MAP Information Element (DL-MAP IE). The DL-MAP IE contains elds with constant and variable length. The elds with constant length (like the Connection identier and the start time that indicates when the corresponding SS will receive its downlink burst) are equal to 32 bits. We do not use the elds with variable length in the rest of our work. Let NDL be the number of SSs served in the downlink direction. An example of the DL-MAP structure is represented in Fig. 2.6 for NDL = 3.
Physical layer
19
Fig. 2.5 Time Division Duplexing (TDD) frame structure in IEEE 802.16.
Fig. 2.6 Example of the DL-MAP structure. The number of served SSs in the downlink direction
is equal to 3. The UL-MAP contains : four elds (Management Message Type, uplink channel identier, ...) with a length equal to 56 bits, and as many elds as the number of SSs served in the uplink direction. These elds are called UL-MAP Information Element (UL-MAP IE). The UL-MAP IE contains elds with constant and variable length. The elds with constant length (like the Connection identier and the start time that indicates when the corresponding SS can send its uplink burst) are equal to 48 bits. We do not use the elds with variable length in the rest of our work. Let NUL be the number of SSs served in the uplink direction. An example of the UL-MAP structure is represented in Fig. 2.7 for NUL = 3. The length of the DL-MAP and UL-MAP messages are equal to 64 + 32 NDL , and 56 + 48
NUL , respectively. As the MCS used for the DL-MAP and UL-MAP management messages is
BPSK 1/2, the number of useful bits per OFDM symbol is equal to 96 (see Table 2.1). Therefore,
64+32NDL 96
56+48NUL 96
UL-MAP message.
20
Overview of WiMAX
Fig. 2.7 Example of the UL-MAP structure. The number of served SSs in the uplink direction
is equal to 3.
Physical layer
21
n BW Nused d c (2.4) NFTT (1 + G ) The theoretic throughput (Throughputtheoretic ) directly depends on the Modulation and CoThroughputtheoretic =
ding Scheme (MCS) used. The OFDM symbols are not only allocated for data transmission. Indeed, the preamble, FCH, and management messages are transmitted in the frame. So, the useful throughput (Throughputuseful ) can be computed as follows :
Throughputuseful = Throughputtheoretic
where : allocated only for data transmission.
Suseful , Stotal
(2.5)
Suseful : represents the number of useful OFDM symbols per frame. These symbols are Stotal : represents the total number of OFDM symbols per frame. This number is constant. Now we compute the number of useful symbols. This number depends on the type of the frame since each frame type has its management messages. First, we compute the number of useful symbols in the FDD downlink frame. We have to subtract from the total number of symbols per frame the number of symbols reserved for the preamble, FCH, and management messages (DL-MAP, UL-MAP, DCD, and UCD). The total number of symbols per frame is computed as shown in Section 2.2.1. Recall that two OFDM symbols are reserved for the preamble and one symbol for the FCH. When the management messages are sent using the most robust MCS (BPSK 1/2), not sent. We assume that BW = 5MHz , n = 144/125, and the frame duration is equal to 20 ms. Given
64+32NDL 96
56+48NUL 96
symbols are
used for the UL-MAP message. We assume that the DCD and UCD management messages are
G , the MCS used, and the number of SSs served in the downlink and uplink directions, we can
compute the useful throughput (see Table 2.2). We assume that the number of SSs served in the downlink direction (NDL ) is equal to the number of SSs served in the uplink direction (NUL ). This number is denoted NSS . Recall that we consider G = 1/4. The useful throughput in the uplink direction is computed as in the downlink direction when the FDD mode is selected. We have to subtract from the total number of symbols the number of symbols used for contentions (ranging and bandwidth contentions) and for the preambles of the uplink PHY PDUs.
22
Overview of WiMAX
Tab. 2.2 Total useful throughput (expressed in Mbps) for a downlink FDD frame of 20 ms.
NSS
1 10 1 10 1 10 1 10
1/4 1/4
In the TDD frame, the symbols are distributed between the downlink and uplink subframes. It is up to the operator to specify the proportion of the uplink and downlink subframes. Once this proportion is dened, the number of useful symbols can be determined following the same steps of the computations in the FDD frame. We note that the useful throughput can be computed when the MCS used is the same and all SSs always have data to send. However, each subscriber has its channel characteristics and then uses a suitable MCS. Subscribers may also have no data to send. Therefore, the useful throughput cannot be computed with simple equations. Moreover, when the subscribers use dierent MCSs, the useful throughput depends on the scheduling algorithm used. In order to have more realistic values of the useful throughput, we use network simulations for that. Our simulation module will be presented in Chapter 3 and some simulation results for the throughput and WiMAX system capacity will be presented in Chapter 4.
23
Common Part Sublayer (CPS) : is responsible for fragmentation and segmentation of MAC SDUs into PHY PDUs as well as QoS control and scheduling. Security sublayer : provides authentication, secure key exchange, and encryption.
24
Overview of WiMAX
Fig. 2.10 Addition of a new service ow. The SS initializes the dynamic service addition
request.
Fig. 2.11 Addition of a new service ow. The BS initializes the dynamic service addition
request. Tolerated jitter : represents the maximum delay variation (jitter) of the connection. Maximum latency : represents the maximum latency between the reception of a packet and the forwarding of that packet. The dynamic service management is used to add a new service ow. A subscriber can have many service ows simultaneously. Once a service ow is created, the subscriber can request to change its parameters or delete it.
25
Fig. 2.12 Exchanges between MAC and CS layers to add a new service ow.
To add a new service ow, the exchanges between MAC and CS are represented in Fig. 2.12.
26
Overview of WiMAX
27
Request (DSC-REQ), Dynamic Service Change Response (DSC-RSP), and Dynamic Service Change Acknowledgment (DSC-ACK).
28
Overview of WiMAX
a Service ow error set. The service ow error set shall be included for every failed service ow in the corresponding DSC-RSP message. This parameter is mandatory only if the transaction is unsuccessful. The detailed behaviors of an SS and a BS to change parameters of an existing service ow are indicated in Fig. 2.15 or Fig. 2.16 when the SS or the BS initializes the process, respectively.
29
30
Overview of WiMAX
Tab. 2.3 Essential Characteristics of the QoS Classes
QoS
Example
Classes ter(s) UGS Periodic, xed size data grants ertPS rtPS Periodic data grants Minimum data rate ; periodic unicast polling nrtPS BE Slightly better than BE Best Eort. No unicast request polling obligation for the BS
Allowed Allowed
VoIP MPEG
Allowed Allowed
Allowed Allowed
Link Adaptation
31
32
Overview of WiMAX
mandatory exit threshold" but yet lower than "16-QAM 3/4 minimum entry threshold". This avoids the ping-pong eect between dierent MCSs. The subscriber uses 16-QAM 3/4 (more ecient MCS) only when the SNR becomes greater than "16-QAM 3/4 minimum entry threshold" (see Fig. 2.19).
Scheduling in WiMAX
33
Fig. 2.20 Packets Scheduling in BS and SS. The uplink scheduler may have dierent scheduling
classes depending on the service type. and UL-MAP MAC management messages, respectively. These messages are transmitted at the beginning of the downlink subframe (see Fig. 2.5). When it receives an UL-MAP management message, the SS determines if it can access to the uplink channel during the current frame. Since the SS may have several simultaneous connections, an uplink scheduler is required in each SS (see Fig. 2.20). The scheduling algorithm determines the SSs that will receive data (in the downlink direction) as well as the SSs that will send data (in the uplink direction). Therefore, the length of the DLMAP and UL-MAP management messages and then the number of useful symbols depends on the scheduler used. Moreover, a preamble is added to each burst. So, the more SSs the BS serves per frame, the less number of useful OFDM symbols is. The length of the UL-MAP message also depends on the bandwidth requests sent by the dierent SSs. A bandwidth request can be sent using contention like for nrtPS and BE users. In this case, these users use the bandwidth contention. As a consequence, the size of the bandwidth contention aects the number of useful OFDM symbols and the collision probability between the bandwidth requests. The number of useful symbols decreases when the bandwidth contention size increases while the collision probability increases when the bandwidth contention size decreases. A bandwidth request can be sent using unicast request opportunities like for rtPS users. Theses unicast request opportunities are sent periodically by the BS and the period of transmission is determined by the BS using the maximum latency of the connection. Finally, it is useless to send bandwidth requests by
34
Overview of WiMAX
the UGS users since the grant requested remains constant during all the connection. Therefore, the IEEE 802.16 standard specied that a UGS user announces its grant size once when he creates a new connection using the Dynamic Service Addition (DSA) messages. Moreover, UGS users are forbidden to send bandwidth requests using unicast request opportunities or bandwidth contention.
Conclusion
35
in the mesh mode, the mesh BS and SSs have the same importance in the scheduling decision. In our work, we consider only the PMP mode.
2.6 Conclusion
In this Chapter, we have introduced the PHY and MAC layers as well as scheduling in the PMP and mesh modes for the OFDM physical layer. In the rest of our work, we interest only on scheduling in the PMP mode. The scheduling algorithm used directly aects the performance of the system since the number of useful OFDM symbols depends on the scheduling decision. The system performance also depends on the channel characteristics of the dierent users since the Modulation and Coding Scheme (MCS) used (and then the number of useful bits) depends on the Signal-to-Noise Ratio (SNR) of the subscribers. We can estimate the WiMAX capacity using computations but we have to assume that all subscribers use the same MCS and always have data to send. Otherwise, the WiMAX capacity depends on the channel characteristics. It also depends on the scheduling algorithm used since the scheduler determines which subscribers will send and/or receive uplink and/or downlink bursts, respectively. Therefore, a more sophisticated alternative (other simple computations) is needed in order to study the performance of scheduling algorithms and estimate the WiMAX capacity. In our thesis, we use for that the network simulation. The next Chapter presents in details our NS-2 WiMAX module utilized to evaluate our proposed scheduling algorithm and to estimate the WiMAX capacity using dierent environments and dierent schedulers.
36
Overview of WiMAX
Chapter 3
Design and Implementation of a WiMAX Module with QoS for NS-2 Simulator
3.1 Introduction
The WiMAX technology promises a large coverage and high throughput. Theoretically, the coverage range can reach 50 km and the throughput can achieve 75 Mbps [tec05]. Yet, in practice the maximum observed coverage range is about 20 km. The practical throughput is also lower than the theoretical throughput. For example, the data throughput can reach 9 Mbps using User Datagram Protocol (UDP) and 5 Mbps using File Transfer Protocol (FTP) over Transmission Control Protocol (TCP) [tec06]. The theoretical values do not always t the reality. The network simulation presents a solution to test the performance of technologies. Network Simulator 2 (NS-2) [url07c] is a widely-used tool to simulate wireless networks. Until today, this simulator does not implement a denitive WiMAX module. Nevertheless, there are some WiMAX NS-2 modules implemented by National Institute of Standards and Technology (NIST) [url07d] and Network and Distributed Systems Laboratory (NDSL) [url07b] and [CWC+ 06]. These modules implement the physical (PHY) and Medium Access Control (MAC) layers of a WiMAX system. The NIST module implements the Orthogonal Frequency Division Multiplexing (OFDM) PHY layer while the NDSL module implements the Orthogonal Frequency Division Multiple Access (OFDMA) PHY layer. Both PHY modules use the same duplexing technique : Time Division Duplexing (TDD) mode. Both MAC modules contain the WiMAX management messages. The NIST WiMAX module also supports mobility, fragmentation, and reassembly of frames. The NDSL WiMAX module supports Call Admission Control (CAC). As we consider the Wireless-OFDM PHY layer, our NS-2 developments are based on the NIST implementation. Our contribution consists of the addition of the WiMAX QoS classes and their requirements, some mechanisms specied by the IEEE 802.16 standard, a link adaptation 37
38
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
mechanism, and some scheduling algorithms. These implemented schedulers depend on the corresponding QoS class and take into account its requirement. We use the Unied Modeling Language (UML) [mH06] to design our WiMAX module. UML class and sequence diagrams are introduced in the following sections in order to model our system in a visual manner. Then, a validation of the behavior of the WiMAX module is required. We have published the description and the validation of our WiMAX module in SimuTools 2008 [BN08b]. The source code has been requested by many persons and is available at : https ://perso.enst-bretagne.fr/aymenbelghith/tools/. Many works are using our QoS-included WiMAX module and we note that the module has been cited by [FT09] in addition to some papers currently submitted and under review.
39
peerNode : contains information about the SS or the BS such as their SNR. WimaxScheduler : is responsible for the ranging and registration and performs scheduling algorithms. It includes two schedulers : one for the BS (BSScheduler ) and one for the SS (SSscheduler ).
40
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
Tab. 3.1 Receiver SNR assumptions (values of the IEEE 802.16e standard)
Receiver SNR (dB) 3.0 6.0 8.5 11.5 15.0 19.0 21.0
41
message. The dierent steps of the creation of a new service ow, in our new module, are depicted in Fig. 3.2. All those modications are performed in the ServiceFlowHandler class. Recall that this class handles the list of the downlink and uplink connections and manages the creation of new connections.
42
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
43
44
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
determined transmission time and MCS used. The number of available symbols must be enough to serve all UGS connections. For that, we use a Call Admission Control (CAC) mechanism. Once the number of reserved symbols is determined, the BS updates the uplink map (UL-MAP) and allocates an uplink burst. The UGS scheduling algorithm is performed at the beginning of each frame.
45
46
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
47
the implementation of the link adaptation is required since the MCS used depends on the SNR of the subscriber. Extensive experiments are conducted to test and validate our QoS-included WiMAX module. In this Section, we study the behavior of the dierent kind of schedulers implemented : for UGS, rtPS, and BE QoS classes. The main parameters of the simulation model are represented in Table 3.2. We consider ve UGS, nine rtPS, and two BE subscribers. The subscribers can use the QPSK 1/2, QPSK 3/4, 16-QAM 1/2, 16-QAM 3/4, 64-QAM 2/3, and 64-QAM 3/4 MCSs.
48
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
Tab. 3.2 Main parameters of the simulation model
Parameter Frequency band Propagation model Antenna model Antenna height Transmit antenna gain Receive antenna gain System loss factor Transmit power Receive power threshold Carrier sense power threshold Link adaptation Frame duration
Value 5 MHz Two Ray Ground Omni antenna 1.5 m 1 1 1 0.025 W -86.93 dBm -78.23 dBm enabled 20 ms 1/4 1024 bytes 100 s
G
Packet length Simulation duration
49
trac load submitted in the network. Our UGS scheduler provides a very low mean sojourn time because it periodically allocates enough symbols to all UGS SSs. Moreover, a new UGS connection is added only when it can be satised.
50
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
Fig. 3.9 Mean sojourn time of UGS SSs versus oered UGS trac load.
51
Fig. 3.11 Mean sojourn time of rtPS SSs versus oered rtPS trac load.
for the real-time applications. We note that the Existing RR scheduler requires a large average delay to deliver a data packet. This is because this scheduler does not provide enough symbols to the SSs. It periodically allocates all the remaining symbols of an uplink subframe to one rtPS SS. This period is equal to the frame duration multiplied by the number of all SSs. We observe that our implemented scheduler exhibits a much better sojourn time. Indeed, this scheduler takes into account the bandwidth requests of the rtPS SSs. Moreover, it does not block any SS.
52
The Design and Implementation of WiMAX Module with QoS for NS-2 Simulator
scheduling implementation serves the UGS, then rtPS, and nally BE connections. So, the higher the number of the remaining symbols is, the higher is the BE throughput. This behavior ts with the standard specication as the BE connections has no QoS requirements.
3.4 Conclusion
In this chapter, we have described our new WiMAX module that takes into account the QoS classes and their requirements. Our module supports the dynamic service addition mechanisms, a link adaptation, and unicast and contention request opportunities. It also involves some UGS, rtPS, and BE schedulers. Extensive simulations have been conducted to show that the behavior of our UGS, rtPS, and BE schedulers ts with the QoS specications of the IEEE 802.16 standard. We have shown through simulations that our schedulers outperform the scheduler of the existing module of WiMAX. Moreover, throughput and mean sojourn time values are determined for some QoS classes. The next chapter presents, more in details, scheduling in WiMAX describing some algorithms as well as their performance evaluation.
Chapter 4
Scheduling in WiMAX
4.1 Introduction
In any wireless network, the radio resource has to be scheduled according to the QoS requirements. The WiMAX/IEEE 802.16 standard does not dene a mandatory scheduling algorithm. Only the framework is given in the standard. Therefore, the choice of the algorithm is left to the vendor or the operator. The choice of a scheduling algorithm for WiMAX/IEEE 802.16 is an open issue. Since scheduling of WiMAX is presently a very active eld, we describe some selected algorithms. The scheduler used in WiMAX system may be classical or specically proposed for WiMAX. The scheduling algorithm can distribute the radio resource between the dierent connections without taking into account their radio channel conditions or their QoS requirements as in the Round Robin (RR) and Weighted Round Robin (WRR) [Lie07] schedulers. The scheduling algorithm can take into account the radio channel conditions as in the maximum Signal-to-Noise Ratio (mSIR) scheduler and Temporary Removal Scheduler (TRS). The scheduler can take into account some QoS parameters like the latency in the Earliest Deadline First (EDF) [Els98] scheduler. A scheduler can be specically proposed for a single WiMAX QoS class like the Adaptive Bandwidth Allocation Scheme (ABAS) [CLL07] which is proposed for the BE class or the adaptive Polling Service (aPS) [NVY07] which is proposed for the rtPS class. A scheduler can also be proposed for many WiMAX QoS classes. For example, the Frame Registry Tree Scheduler (FRTS) [XPM05] is proposed for the UGS, rtPS, nrtPS, and BE QoS classes while the ensuring QoS requirement scheduler [SAKH06] is proposed for all the WiMAX QoS classes (ertPS is included). A scheduler can take into account all the QoS requirements of a dened WiMAX class like the ensuring QoS requirement scheduler or some of them like in the FRTS. Indeed, the FRTS takes into account the latency of the rtPS class but does not guarantee the minimum reserved rate. In this chapter, we describe some schedulers that can be used in WiMAX. A synthesis of 53
54
Scheduling in WiMAX
schedulers used in WiMAX is presented in order to summarize the advantages and disadvantages of the schedulers already described. Then we present our proposals for the rtPS QoS class. We nally point out the dependence of the estimation of the WiMAX capacity on the scheduling algorithm and the radio conditions environment. We have published a book chapter titled "Scheduling Techniques in WiMAX" in [Mao09]. The book chapter proposes WiMAX practical scheduling considerations, describes some scheduling algorithms and gives a synthesis of scheduling techniques. The title of the book is "Current Technology Developments of WiMax Systems". The book was published by Springer at the end of January 2009 [url09].
Nsymbols . Then, the higher the demand of a subscriber is, the more symbols are
55
Fig. 4.1 Packets queues for Weighted Round Robin (WRR) scheduling algorithm.
of the total resources if there are n subscribers and all subscribers have data to send.
An example of WRR algorithm execution is represented in Fig. 4.1. In this example, there
are three queues : A, B, and C. The weight of queues A, B, and C are equal to 2, 1, and 3, respectively. In WiMAX, the connections have dierent QoS parameters and the subscribers use dierent MCS. Moreover, the subscribers do not generally have the same trac. Therefore, the connections do not need the same resources. The use of the WRR scheduler can be suitable for WiMAX by assigning dierent values of weights to the queues.
i is eligible to be dequeued if DC (i ) is greater than the length of the packet waiting to be sent,
noted L(k ). In this case, DC (i ) is decremented by L(k ). At each round, one packet at most can be dequeued for each queue. An illustration of a DRR algorithm execution is represented in Fig. 4.2. There are three queues : A, B, and C. The decit counter (DC ) of queues A, B , C , are 350, 300, and 250, respectively. In Fig. 4.2, the reference of the packet number k of queue A, B , and C are ak , bk , and ck , respectively. The packet length is L(.).
56
Scheduling in WiMAX We note that packet a1 is immediately sent because its length (L(a1 )=200) is lower than
the decit counter of queue A (DC (A)=350). However, packet b1 waits two rounds before being sent. Indeed, its length (L(b1 )=500) shall be lower than the decit counter of queue B (2 *
DC (B )=600).
The DDR scheduler requires a minimum rate to be reserved for each packet ow before being scheduled. This characteristic can be useful in WiMAX because some WiMAx QoS classes require a minimum reserved rate.
4.2.1.5 Example of systematic schedulers applied in WiMAX context Choice of schedulers for the BS and SS Three schedulers must be chosen for the BS and
SS ; an uplink and downlink schedulers both located in the BS and an uplink scheduler in the SS (see Fig. 2.20 in Section 2.5). Authors in [CLME06] make the following choice. They apply the DRR scheduler to the BS scheduler in the downlink direction and to the SS scheduler in the uplink direction. Indeed, a station can use the DRR scheduler only if it has the knowledge of the packet size of each queue. What is left is to choose the uplink scheduler. The DRR scheduler cannot be applied because the BS does not have the knowledge of the uplink packet size of the SSs queues. However, the BS can estimate the amount of backlog of each connection when it receives the bandwidth requests from the dierent SSs. So, the WRR scheduler can be selected as the uplink scheduler in the BS (see Fig. 4.3).
Performance evaluation The performance of the DRR and WRR schedulers applied in WiMAX system is analyzed by simulation in [CLME06]. The main parameters of the simulation are the following : the frequency band is 2 - 11 GHz, the air interference is the WirelessMANOFDM (using orthogonal frequency-division multiplexing), the channel bandwidth is 7 MHz, the duplexing mode is Frequency Division Duplexing (FDD), and the frame duration is 10 ms. The metrics to evaluate are : Maximum achievable throughput : represents the maximum amount of data that a station can send successfully. Packet-transfer delay : represents the time dierence between the arrival time of the packet at the upper layer of the trac destination and its arrival time at the MAC layer of the trac source. Delay variation : represents the dierence between the maximum packet-transfer delay and the packet transmission delay. Two scenarios are considered : residential scenario and Small and Medium-sized Enterprises (SME) scenario. In the rst scenario, the BS provides Internet access to SS by means of Best Eort (BE) connections. The results of the average delay and maximum achievable throughput depending on the oered load and number of subscribers are represented in Table 4.1. In the second scenario, the BS provides three types of services :
57
Fig. 4.2 Packets queues for Decit Round Robin (DRR) scheduling algorithm.
58
Scheduling in WiMAX
Voice over Internet Protocol (VoIP) service : each station has four VoIP sources multiplexed into an rtPS connection. Video-conference service : each station has two video-conference sources multiplexed into an nrtPS connection. Data service : each station has a data source provided by a BE connection. The BS periodically grants a unicast poll to each VoIP and video-conference services in order to allow SS to send bandwidth request. The results of the average delay and delay variation depending on the oered load, number of subscribers, and service types are represented in Table 4.1.
59
Tab. 4.1 Performance evaluation of the residential and SME scenario. Downlink scheduler is
DRR and uplink scheduler is WRR. Metric/Results Results for downlink trac - Constant at low cell load. Residential scenario : Average delay Residential scenario : Maximum throughput achievable
- Sharply increase at high cell load. same as downlink trac. - Downlink delay < Uplink delay. Decrease when increase of number of SSs. Problem : DL-MAP and UL-MAP messages become too large when the number of SSs increases. - Constant for VoIP. Indeed, the delay of rtPS connections depends only on the unicast poll period. Decrease when increase of number of SSs. Problem : Each SS has its preamble. So the throughput decreases due to the increase of the number of these preambles.
Small
and
Medium-
- Smoothly increase at low cell load same as downlink trac. for video-conference and data. - Sharply increase at high cell load for video-conference and data. Delay variation of VoIP = delay variation of video-conference. Delay variation of VoIP < delay variation of video-conference.
sized Enterprises (SME) scenario : Average delay SME scenario : Delay variation
60
Scheduling in WiMAX
scheduler in order to determine the number of symbols that will be allocated to each SS belonging to the scheduling list. The TRS can be combined with a common scheduler such as the RR, and mSIR schedulers. Temporary Removal Scheduler + Round Robin : The TRS can be combined with the RR scheduler. The combined scheduler is called TRS+RR. For example, if there are k packet calls and only one of them is temporarily removed, each packet call has a portion, equal
1 to k 1 , of the whole channel resources. Temporary Removal Scheduler + maximum Signal to Interference Ratio : The TRS can
be combined with the mSIR scheduler. The combined scheduler is called TRS+mSIR. This scheduler assigns the whole channel resources to the packets having higher SNR and belonging to the scheduling list. The performance of the TRS+RR and TRS+mSIR schedulers, applied in WiMAX, is analyzed by simulation in [BTGK05]. The main parameters of the simulation are the following : the frequency band is 3.5 GHz, the air interference is the WirelessMAN-OFDM, the channel bandwidth is 3.5 MHz, and the frame duration is 2 ms. The schedulers to evaluate are the following : the RR, mSIR, TRS+RR, and TRS+mSIR schedulers. The results for File Transfer Protocol (FTP) 300 kByte download are described below : The mSIR and TRS+mSIR schedulers provide the highest throughput whereas the RR scheduler provides the worst results. The mSIR, TRS+mSIR, and TRS+RR schedulers provide the lowest download time whereas the RR scheduler provides worse results. The TRS+mSIR and TRS+RR schedulers provide the lowest channel utilization whereas the RR and mSIR schedulers provide better results. The TRS+mSIR and TRS+RR schedulers provide the lowest blocking rate whereas the RR and mSIR schedulers provide the worst results.
61
Fig. 4.4 Main steps of determination of scheduled SSs using the O-DRR scheduler.
it still has data to send. The scheduled set is changed dynamically. This change depends on the wireless link state of each eligible subscriber. At the beginning of a new scheduling epoch, the BS resets the eligible and scheduled set and performs the above process again. The main steps of the determination of the scheduled SSs are represented in Fig. 4.4. The assignment of the bandwidth using the O-DRR scheduler is performed as follows. Each SS has a quantum Qi . If the SINR of the wireless link of the subscriber i is lower than the threshold SINRth , then : The quantum Qi of the subscriber i is distributed among the scheduled subscribers. The lead/lag counter of the subscriber i is incremented by Qi . The lead/lag counter of the scheduled subscriber j is decremented by the amount that the subscriber j receives above its quantum Qj .
62
Scheduling in WiMAX
Determination of the polling interval k The BS polls all subscribers every k frames. A low
value of k causes a polling overhead. So, the eciency will decrease. Furthermore, a high value of
k causes an unfair trac and a non satisfaction of QoS requirements. Authors in [RBS06] propose
that the BS objective is the minimization of the worst-case relative fairness in bandwidth and the normalized delay.
Performance evaluation of the O-DRR scheduler The results of the number of slots
assigned depending on the number of subscribers and k value show that the number of slots assignment increases when k increases. Indeed, when k increases, the SINR becomes more variable for the dierent subscribers and the lead/lag counter has more inuence on the bandwidth assignment. The results of the fairness in bandwidth using Jain's Fairness Index [JCH84] depending on the number of subscribers and k value show that Jain's Fairness Index remains above 90%. Then the provider has more choices to choose an appropriate value of k at which the fairness and the bandwidth requirements are both satised.
Performance evaluation of the adaptive rtPS scheduler The adaptive rtPS scheduler
is compared with the weighted ow scheduler. The simulation results show that the adaptive rtPS scheduler provides better average delay in low and medium load. This is because this scheduler considers the data generated between the sending time of the bandwidth request and
63
the allocation time of the bandwidth. Moreover, the adaptive rtPs scheduler requires less buer size for the rtPS data queue. In high load, the adaptive rtPS and weighted ow scheduler have the same performance. This is due to the saturation of the network.
4.3.1.2 Adaptive Polling Service (aPS) Adaptive Polling Service (aPS) description A novel adaptive Polling Service (aPS) mechanism for the rtPS QoS class is dened in [NVY07]. The main idea of the proposed mechanism is to adjust the polling period. The polling period represents the period of the sending of two successive unicast request opportunities. The BS initializes the polling period with Tmin . Parameter
Tmin is determined using the average packet arrival rate. If the BS does not receive bandwidth
request after N polls, it exponentially increases the polling period until reaching Tmax . Parameter Tmax is determined using the tolerable delay of the connection since the aPS mechanism is dened for real-time applications. When the BS receives a bandwidth request, it resets the polling period to Tmin .
Performance evaluation of the aPS The aPS mechanism is compared with the rtPS mechanism dened in the IEEE 802.16 standard where the polling period is constant during all the connection. The simulation results of an ON/OFF TCP trac show that the aPS mechanism reduces the signalling overhead by 50% to 66%. This reduction comes from the increase of the polling period during the OFF periods. On the other hand, the aPS mechanism provides higher delay than the rtPS mechanism. However, the delay is still acceptable for almost all of the applications. The simulations results of a TCP-based application show that the aPS mechanism reduces the signalling overhead by 66%. Like for the ON/OFF trac, the increase of the delay is still acceptable. The simulation results of an online game application working over User Datagram Protocol (UDP) shows that the aPS mechanism reduces the signalling overhead by 50% to 75%. However, the delay increases and becomes not suitable with all the applications. The increase of the delay depends on the aPS mechanism parameters. In the dierent scenarios, the authors investigate only Tmax and show that when Tmax increases, the signalling overhead are reduced and the delay increases. Then, the suitable value of Tmax can be chosen depending on the tolerate delay of the application.
4.3.1.3 Adaptive Bandwidth Allocation Scheme (ABAS) Adaptive Bandwidth Allocation Scheme (ABAS) description An Adaptive Bandwidth
Allocation Scheme (ABAS) for 802.16 TDD systems is dened in [CLL07]. It aims to dynamically determine the suitable downlink-to-uplink bandwidth ratio. Indeed, the IEEE 802.16 standard
64
Scheduling in WiMAX
species the frame structure and it is up to the operator to choose the downlink-to-uplink bandwidth ratio. The proposed scheduler performs as follows. The BS determines the dierent information of the downlink and uplink connections such as the bandwidth requests and the number of downlink and uplink connections. Then, the BS determines the number of slots allocated to the downlink and uplink subframe and adjusts the split between the two parts of the frame. Finally, the BS informs the dierent SSs about its decision using DL-MAP and UL-MAP MAC management messages. The mechanism is repeated at the beginning of each frame. ABAS is proposed for the BE QoS class and all BE users have TCP connections. It is assumed that all connections have the same channel conditions.
s and The ratio of the numbers of mini-slots allocated to downlink and uplink directions (Nd s respectively) is determined as follows. Nu s Nd d ndSS Ldata + nuSS Lack 1 = , s Nu nbSS Nack + nuSS Ndata d B tms
(4.1)
where : d : represents the delayed ACK parameter. ndSS : represents the number of downloading TCP transfers. nuSS : represents the number of uploading TCP transfers. Ldata : represents the data packet size. Lack : represents the ACK packet size. Ndata : represents the number of mini-slots for transmitting one data packet. Nack : represents the number of mini-slots for transmitting one ACK packet. B : represents the total capacity allocated for BE trac. tms : represents the mini-slots duration (expressed in s).
10 : 1). This is due to the tacking into account the trac characteristics.
65
4.3.2.2 Cross-layer Scheduling Algorithm with QoS Support Cross-layer scheduler description In [LWG06], a novel cross-layer scheduling algorithm is
proposed for wireless networks. This scheduler can support multiple connections with various QoS requirements and in various kinds of wireless networks. The main idea of the cross-layer scheduler is the aectation of priorities to the dierent connections. These priorities depend on the QoS requirements of the connections. This scheduler is applied to the UGS, rtPS, nrtPS, and BE WiMAX QoS classes. For the UGS connections, the scheduler must guarantee a constant number of time slots allocated during the whole service time. For the rtPS and nrtPS connections, the scheduler must guarantee the latency and the minimum reserved rate respectively. For the BE connections, there
66
Scheduling in WiMAX
Fig. 4.5 Main steps of the uplink packet scheduler with CAC.
Schedulers Specically Proposed for WiMAX is no QoS guarantee but a Packet Error Rate (PER) should be maintained.
67
After serving all UGS connections, the scheduler allocates all the residual time slots to the rtPS, nrtPS, then BE connections that have the maximum value of a dened Priority Function (PFR). The PRFs, for the rtPS, nrtPS, and BE connections consequently depend on the delay satisfaction indicator, the rate of the average transmission rate over the minimum reserved rate, and the normalized channel quality, respectively. The PRFs details are presented in [LWG06].
Hybrid scheduler description The hybrid scheduling algorithm proposed in [VSJD06] for
WiMAX uses two dierent schedulers. The EDD scheduler is used for the real-time services while the WFQ scheduler is used for the non-real-time services. The EDD scheduler is based on dynamic priority. In an EDD queue, the packets are classied in order of their deadline values. The expected deadline time of a packet is equal to its arrival time plus its maximum service time. The WFQ scheduler allocates dierent shares of radio resources to the users. Each share depends on the weights of the dierent connections. Then, the WFQ scheduler can provide a required throughput rate for each connection by assigning suitable values to the dierent weights. The connection delay number i is computed as follows :
DWFQ (i ) =
where :
R wi
n k =1 wk ,
(4.2)
n : represents the number of the connections. wi : represents the weight given to the queue i .
Scheduling in WiMAX
In [TFB06], the same hybrid scheduler is proposed ; the EDD scheduler is used for the UGS and rtPS QoS classes and the WFQ scheduler is used for nrtPS and BE QoS classes. The only dierence is that the nrtPS QoS class has more priority than that of the BE QoS class.
Performance evaluation In [VSJD06], the hybrid scheduler is compared with the EDD scheduler employed for the real and non-real time services. The simulation results show that the hybrid scheduler provides, for real time services, less delay than the EDD scheduler. This is due to the competition of the non-real time packets for channel access when the EDD scheduler is used. However, the non-real packets wait for more time when the hybrid scheduler is used. Since only the real-time services have delay requirements, it is recommended to use the hybrid scheduler. In [TFB06], the number of contention slots is investigated. The contention slots are used by the SSs to send their bandwidth requests through contention. The simulation results show that the throughput increases when the number of contention slots increases without exceeding the half number of the BE connections. This is due to the decrease of the probability of bandwidth request collisions. The throughput decreases when the number of contention slots exceeds the half number of the BE connections. This is due to the decrease of the amount of radio resources reserved for data transmission.
4.3.2.4 Frame Registry Tree Scheduler (FRTS) Frame Registry Tree Scheduler (FRTS) description The Frame Registry Tree Scheduler
(FRTS) scheduler [XPM05] consists of three operations : packet/request arrival, frame creation, and subscriber's modulation type change or connection QoS service change. The basic idea of the packet/request operation is to distribute packet transmissions in time frames, based on their deadline. For UGS and rtPS services, the packet deadline is equal to its arrival time plus its maximal latency. The subtree of the last time frame where this packet can be transmitted is updated, if it exists. Otherwise, it is created. For nrtPS and BE services, the packet deadline does not need to be calculated. Then, the subtree of the last existing time frame is updated. The frame creation procedure decides on the frame contents. There are three cases : If the subtree of the rst time frame contains a number of packets equal to one time frame, all these packets ll up the frame content. If the subtree of the rst time frame contains a number of packets less than one time frame, the empty slots are occupied by packets from the next time frame subtrees and/or will be left for contention. If the subtree of the rst time frame contains a number of packets more than one time frame, packets for BE service are moved to the next time frame subtree. If there are still
69
excess packets to transmit, rst nrtPS packets, then rtPS packets and nally UGS packets are deleted until the number of packets t exactly into one time frame. A change in a subscriber's modulation type or connection QoS service causes a moving of the corresponding subtree to the right position.
Performance evaluation of the FRTS In [XPM05], the FRTS scheduler is compared with a
simple scheduler that serves the UGS, then rtPS, nrtPS, and nally BE connections. Simulation results show that the FRTS scheduler provides better throughput. This is due to the decrease of the number of lost packets. Indeed, the FRTS scheduler takes into account the deadline of the real time packets (UGS and rtPS). The FRTS scheduler provides better results when the BS serves more SSs. For example, the FRTS scheduler provides 20% and 25% of less lost packets than the simple scheduler when the number of served SSs is equal to 340 and 440, respectively. The simulations results also show that the FRTS scheduler can serve nrtPS and BE connections even if the load is high. This is because this scheduler prots from the latency tolerance of some real-time packets.
4.3.2.5 Scheduler ensuring QoS requirements Description of the scheduling proposal In [SAKH06], a scheduler that ensures the QoS
requirements for the dierent QoS classes is proposed. This scheduler consists of the allocation of the minimum bandwidth requirements to all connections and then the allocation of free slots to the rtPS, nrtPS, and BE connections without exceeding their maximum bandwidth requirements (see Fig. 4.6). The number of slots to allocate to the UGS and ertPS is constant since the minimum and maximum bandwidth requirements of these connections are the same. The minimum and maximum number of slots to allocate to the rtPS and nrtPS connections depend on the minimum and maximum bandwidth requirements, respectively. Since the BE connections have no QoS requirements, their minimum numbers of slots to allocate are equal to zero while their maximum numbers of slots depend on their bandwidth requests. After satisfying the minimum bandwidth requirements of all connections, the remaining free slots are allocated to the rtPS and nrtPS connections. Then, they are distributed between the BE connections.
Performance evaluation The simulation results show that the proposed scheduler satises
the UGS and ertPS connections. The scheduler also provides to the rtPS and nrtPS connections a bandwidth greater than their minimum bandwidth requirements. The remaining bandwidth is shared between the BE connections. The simulations results show that the proposed scheduler fairly distributes the bandwidth between the dierent BE connections. The simulation results also show that the throughput decreases when the number of the served connections increases. This is due to the increase of
70
Scheduling in WiMAX
71
the overhead size. Moreover, despite of the change of the MCS used and therefore the number of slots to reserve, the proposed scheduler provides the bandwidth requirements. We note that this scheduler requires a CAC since the BS shall serve all UGS and ertPS connections and provide the minimum reserved rate for all rtPS connections. Even if a CAC is added to this scheduler, it is impossible to satisfy all the requirements of the dierent connections when the radio channel conditions become very bad. This is because the subscribers have to use more robust MCS and then the number of useful bits becomes lower. We also note that this scheduler aims only to ensure the QoS requirements and may block BE connections.
72
Scheduling in WiMAX
73
Parameters
Values
G
Frame duration Mean packet length Simulation duration Propagation model Antenna model Antenna height Transmit antenna gain Receive antenna gain System loss factor Transmit power Receive power threshold Carrier sense power threshold Link adaptation
1.5m
1 1 1 0.025 W -86.93 dBm 0.9 * Receive power threshold Enabled
74
Scheduling in WiMAX
like the mSIR scheduler, favors those SSs having the highest SNR. If the BS does not serve an SS having unicast request opportunities, it gives priority to other SSs having lower SNR. Moreover, the mmSIR scheduler serves fewer SSs than the mSIR scheduler (see Fig. 4.11). As a preamble is added to each uplink burst (see section 8.3.5.1 of [STS04]), the BS schedules more useful symbols when it serves fewer SSs per frame. The mean number of served SSs per frame is represented in Fig. 4.11. The mSIR scheduler serves more SSs per frame since it always serves the SSs having the highest SNR. Then, these SSs usually do not have many data packets and share radio resources. On the other hand, the mmSIR scheduler does not serve an SS having unicast request opportunities. This SS generates more data packets and its bandwidth request can be taken into account when it will be served later.
75
Fig. 4.11 Mean number of served SSs per frame versus trac load
behaviors of the BS downlink scheduler and the SS uplink scheduler are the same. However, the uplink BS scheduler requires information about uplink SS queues. Therefore, the synthesis table can be restricted to the downlink and uplink schedulers in the BS (no need to represent the uplink scheduler in the SS).
76
Tab. 4.3 Synthesis of systematic schedulers
Scheduling in WiMAX
Disadvantages Not suitable. Subscribers do not have the same trac, radio characteristics, and QoS requirements. Subscribers, having less data, may be scheduled after an excessive delay. Dicult to choose the suitable weights. Can be used only for the downlink trafc ; BS does not know the size of packets at the SS queues.
Takes into account the requested bandwidth. Does not block any connection. Does not block any connection.
Advantages Provides high throughput. - Packet under low SNR are delayed, but can be
Disadvantages Subscribers having a poor SNR may be scheduled after an excessive delay.
Temporary
Re-
- Suitable for uctuant duler. Fair scheduler by using of the lead/lag counter. Can block SSs having low SNR.
After this synthesis, we can identify some schedulers that can be used for the downlink or/and the uplink trac. The RR, mSIR, and Prorate scheduler are not suitable to use in a WiMAX context for most of the cases. Indeed, the RR scheduler is only suitable for subscribers who have the same characteristics such as the SNR, QoS requirements, and the trac load. The mSIR and Prorate schedulers may block subscribers having a poor SNR value or less data quantity, respectively. Some schedulers can be used only in one direction. For example, the DRR scheduler can be used only in the downlink trac because the BS does not know the queue length of the subscribers. The uplink scheduler with CAC mechanism is only proposed for the uplink trac since the CAC is performed to control the connections that are transmitted by the subscribers. The WRR, TRS, O-DRR, cross-layer, hybrid, and FRTS schedulers can be used in the downlink and uplink trac. A use of the WRR or the hybrid (EDD+WFQ) schedulers requires the
77
Scheduler
Advantages
Disadvantages - Applied only for rtPS connections in the uplink direction. - The scheduler performance depends on the performance of the prediction algorithm of the data arrival.
Adaptive rtPS
- Reduces the signaling overhead. adaptive Polling Service - Does not compromise the delay when increasing the system capacity. - Does not block SSs having low SNR. - Reduces the signaling overhead. Modied mSIR - Provides high throughput. - Does not compromise the delay when increasing the system capacity. - Assumes that the channel conditions between SSs and BS are the same. Adaptive Bandwidth Allocation Scheme Determines dynamically the - Assumes that data are always available. - Applied only for TCP connections. downlink-to-uplink ratio. - Applied only for rtPS connections in the uplink direction. - Ecient when data are always available. Applied only for rtPS connections.
determination of the suitable weights. The removal time, penalty time, and maximum number of repetitions are the main parameters of the TRS scheduler. Dierent values of these parameters completely change the behavior and performance of these schedulers. For the O-DRR scheduler, the polling interval and minimum threshold of the SINR parameters have an immediate impact on the delay and throughput, respectively. We also note that some schedulers are proposed specically for a single WiMAX QoS class. For example, the adaptive rtPS and our mmSIR schedulers are proposed for the rtPS QoS class in the uplink direction. The parameters of the stochastic algorithm of the adaptive rtPS scheduler have to be dened. Our proposed scheduler requires only to specify the period of the unicast request opportunities for each SS. Proposed only to the BE QoS class, the ABAS scheduler requires that all SSs have TCP connections and the channel conditions are the same.
78
Scheduling in WiMAX
Tab. 4.6 Synthesis of schedulers proposed for many QoS classes
Scheduler
Disadvantages
- Does not block BE connections if Can block non-real-time applicathere is enough symbols to reserve tions.
Cross-layer scheduler
Hybrid (EDD+WFQ)
- Takes into account the deadline Does not take into account the miof the real-time connections. nimum reserved rate of the rtPS and nrtPS connections. - Need of a CAC. - BE connections can be blocked.
Takes into account the QoS requirements of all the kinds of connections.
There are schedulers proposed for many WiMAX QoS classes. Some of them take into account only the deadline of the real-time applications and do not provide their minimum reserved rates like the cross-layer and Hybrid (EDD+WFQ) schedulers. The minimum reserved rates of nrtPS connections are also not taken into account by these schedulers. Some of them may block the non-real-time applications like the uplink scheduler with CAC mechanism and scheduler ensuring QoS requirements.
79
4.6 Comparison between Schedulers and Impact of Scheduling on the System Capacity
It is required to compare between dierent schedulers using the same simulation module. For that, we use our NS-2 module for WiMAX [BN]. The full details of our WiMAX module are presented in Chapter 3. In this Section, we assess some scheduling methods by determining the spectrum eciency and mean sojourn time performance measures. The spectrum eciency is equal to the throughput divided by the bandwidth (expressed in bit/s/Hz). Recall that the mean sojourn time represents the average time a data packet spends from its generation until its delivery at the destination. Simulation results are given for three dierent scenarios : bad, uctuant, and good radio channel conditions. We can also compare theoretical values and simulation results of WiMAX spectrum eciency. The simulation model is the same of that of Section 4.4. There are nine UGS connections, nine rtPS connections, and two BE connections. The capacity estimation using dierent schedulers in dierent environments was published in a paper in the Vehicular Technology Conference (VTC) Spring 2008 [BN08c].
80
Scheduling in WiMAX
Fig. 4.12 Spectrum eciency versus rtPS trac load (bad radio channel scenario).
Fig. 4.13 Mean sojourn time versus rtPS trac load (bad radio channel scenario).
block any connection. Furthermore, the RR scheduler outperforms the WRR scheduler since the complete cycle of the WRR scheduler is longer. So, when the WRR scheduler is applied, the subscribers having small weights may receive longer delay.
81
Fig. 4.14 Spectrum eciency versus rtPS trac load (uctuant radio channel scenario).
82
Scheduling in WiMAX
Fig. 4.15 Mean sojourn time versus rtPS trac load (uctuant radio channel scenario).
Conclusion
83
Fig. 4.16 Spectrum eciency versus rtPS trac load (good radio channel scenario).
Fig. 4.17 Mean sojourn time versus rtPS trac load (good radio channel scenario).
4.7 Conclusion
In this Chapter, we propose a state-of-the-art of WiMAX Scheduling. The choice of the scheduling algorithm is highly dependent on the transmission service type and the trac shape in addition to other QoS requirements. We propose a synthesis table for some of the proposed sche-
84
Scheduling in WiMAX
duling algorithms. The eciency of a scheduling algorithm can be estimated through simulations. Evidently, WiMAX capacity is completely dependent on the scenario (the environment) and the scheduling algorithm used. In the next Chapter, we try to point out that the choice of the scheduling algorithm depends on other models of the system. We consider only the inuence of the pricing model. So, we implement a sophisticated pricing model for WiMAX real-time and non-real-time applications. This pricing model is compared with another basic one in order to show that a pricing scheme has to take into account the characteristics of the system and can prot from the willingness-to-pay of users using auction.
Chapter 5
86
pricing mechanisms are proposed in order to allocate the available resources in the Internet. The operator gives higher priority to the users that are more socially valuable. Indeed, the operator announces the service price. Users then can decide if they connect or not. So, only users having higher valuable packets accept the current price. It is up to the user to valuate its packets. Authors of [LCC06] propose a pricing model for a general wireless system (no specic technology is considered) where the utility function depends on the transmission rate and the quality of the channel. The main objective of the dened pricing model is the utility maximization. In [NH07], WiMAX is used as the backhaul for Wi-Fi Access Points (APs). The WiMAX BS uses a strategy game, called Stackelberg leadership model [Bag84], in order to share bandwidth between Wi-Fi users. The game strategy is suitable when a system contains many entities and each entity has its own objective(s). For example, the operator tries to maximize its revenue while the user tries to achieve its QoS requirements. The system is described as follows. The WiMAX BS charges the dierent Wi-Fi APs and then tries to maximizes its revenue while a Wi-Fi AP tries to get its required bandwidth. All the Wi-Fi APs are considered as rtPS users. So, the pricing strategy is proposed for only one QoS class. To the best of our knowledge and apart from the works mentioned above, there is not yet any published pricing proposal taking into account the characteristics of the WiMAX QoS classes in the research literature. Our starting point in the evident considerations that WiMAX operators will use dierent pricing strategies for the QoS classes. Our proposed sophisticated model allows the operator to distribute radio resources depending on its optimization constraints. A suitable choice of the pricing model parameters provides to the operator a various choice of constraints such as the maximization of the revenue, the maximization of the total throughput, and the minimization of the blocking rate for some QoS classes. We also implemented a CAC in order to control the admission of the real-time applications. In order to compare, we rst propose a basic pricing scheme that does not dierentiate between the dierent WiMAX QoS classes. Then we present our sophisticated pricing model proposed for the UGS, rtPS, and BE QoS classes. The scheduling algorithm has to take into account the pricing scheme of the WiMAX system. We point out the relation between the scheduler and the pricing model used. Finally, we compare the sophisticated pricing model with the basic pricing scheme.
System Model
87
utility, which is dened as the dierence between the valuation and the price. Let Pi be the price a user pays for a given service. Let Vi be the valuation the user assigns to the service. Let Ui be the utility. We have then :
Ui = Vi Pi
(5.1)
Intuitively, the utility is a monetary equivalent of the benet that the user gets from using the service. It is therefore natural to assume that users will act so as to maximize this benet. As a consequence, if any choice of the user brings him a negative utility then he will simply prefer not to connect to the service and then get utility 0. We then see that determining the price is an indirect way to control the admission of users to the service. In telecommunication services, the most visible indicator of the quality of service is the throughput. In WiMAX the instantaneous rates granted to users are decided at every frame by the scheduler. Note that the scheduler has a given number of data symbols to allocate. More precisely it does not allocate rates but data symbols. For one allocated data symbol, the rate depends on the MCS used. The closer the user is to the BS, the higher is the throughput. Let
ri be the rate requested by user i , si be the number of symbols allocated to user i , SNRi be the
SNR of user i , and k (SNRi ) be the number of bit/s corresponding to an OFDM symbol. So, we have :
ri = k (SNRi ) si .
(5.2)
SW :=
i =0..N
Ui .
i =1..N
(5.3)
Considering that the operator's utility is simply his total revenue ( (5.1), social welfare is equal to :
Pi ) and using
SW :=
i =1..N
Vi .
(5.4)
On the other hand, if the operator can choose the pricing mechanism to apply, it is reasonable to consider that, possibly in addition to some other criteria, he selects the one that maximizes
R :=
i =1..N
Pi .
(5.5)
VUGS (ri ) := i ri .
(5.6)
System Model
89
Each UGS user has its own parameter i . We introduce heterogeneity among the UGS users by assuming that i is randomly chosen, according to continuous uniform distribution on the interval [UGS ,UGS ] (in monetary units (MU) per Kbps). 1 2
VrtPS (ri ) := i
(5.7)
where denotes the min, i is a constant coecient expressed in monetary unit (MU) per Kbps, and Ri max is the maximum transmission rate of connection i . The valuation function is represented in Fig. 5.1. Notice that with this model, user i is not willing to pay to get a transmission rate higher than Ri max since his valuation function is constant above that value. The choice of Ri max for each user depends on its rtPS application. User heterogeneity comes from the random choice of the coecient i . Parameter i is randomly chosen according to continuous uniform distribution on the interval [rtPS ,rtPS ] (in MU by Kbps transmitted). 1 2
5.2.2.3 BE users
The BE connections have no QoS guarantees. Therefore, the BE users have exibility in their transmission rates. To model demand elasticity, we also consider an increasing and concave function of the transmission rate for those users :
90
VBE (ri ) := i
where :
(5.8)
Rmax : is a trac rate above which BE users are insensitive to a rate improvement (possibly because the transfer speed then becomes limited by computational capacities). We assume that parameter Rmax is the same for all the BE users. i : is a constant expressed in MU by rate unit. In order to introduce heterogeneity among BE users, we assume that i is randomly chosen BE for each user, according to a uniform distribution on the interval [BE 1 ,2 ].
P b (ri ) := Pref ri ,
(5.9)
where the superscript b is for the basic pricing scheme, Pref is the xed price charged per unit of transmission rate, and ri is the data rate given to user i . We can then verify that P b (ri ) is the price paid by user i for the rate he receives in a frame. If the data rate is zero (no allocated data), the formula is still valid.
91
s PUGS (r ) :=
PUGS PUGS
Low High
r,
r , otherwise
(5.10)
where : nbres : represents the number of reserved OFDM symbols. nbtot : represents the total number of OFDM symbols. thUGS : represents a threshold above which the network is considered congested, PUGS Low (resp. PUGS High ) : represents the price charged per unit of transmission rate during non-congested (resp. congested) periods.
(5.11)
PrtPS
where :
add
(5.12)
CrtPS : represents a pricing constant xed by the operator. nbrem min : represents the number of OFDM remaining symbols after all the UGS users are served and the rtPS users get their minimum transmission rates. nbrem min depends on congestion and thus the users are encouraged to use more bandwidth if the trac is low.
5.3.2.3 BE users
We propose to charge BE users a given price, denoted SymbolPriceBE , per reserved symbol. If BE user i obtains si symbols per second, he has to pay per time unit a price
s PBE (si ) := Symbol PriceBE si .
(5.13)
We consider two models. In the model, called Fixed Symbol Price Model (FSPM), the symbol price is constant and xed by the operation. While the symbol price is variable and is determined using auction between dierent users when the second model is applied. This model is called Variable Symbol Price Model (VSPM).
92
Fixed Symbol Price Model In FSPM, the operator charges a xed price SymbolPriceBE
per reserved symbol. After serving all the UGS and rtPS users, the remaining symbols are distributed between all the BE users, using the Round Robin (RR) scheduler. The number of reserved symbols to each connection is equal to the minimum of the following values : The share of symbols for this user. The number of symbols required to transmit the remaining part of the le. The number of symbols that the user can reserve. This number depends on its valuation function.
Variable Symbol Price Model In VSPM, the operator also charges based on (5.13), but with a variable price Symbol PriceBE per reserved symbol. The symbol price changes periodically.
We assume that the operator computes Symbol PriceBE as the market clearing price [BB06] for the symbols remaining after serving the UGS and rtPS users, i.e. the unit price for which total BE demand equals supply. The market clearing price Symbol PriceBE is then the solution p of the equation :
di (p ) = nbBE ,
i N BE
(5.14)
where N BE is the set of BE users, nbBE the number of remaining symbols to allocate to the BE users, and di () is the demand function of user i , i.e. the number of symbols that he would like to buy at unit price p to maximize his utility. As the valuation function depends on the rate, we have to express ri as a function of the number of symbols si reserved for user i (see (5.2)). Formally, following (5.8) and (5.2), the number di (p ) = arg(maxs [VBE (k (SNRi ) s ) p s ]) is equal to :
di (p ) =
(5.15)
Auction or ttonnement methods can be implemented to converge to this price [WS02]. However we do not consider the implementation of these auctions in our WiMAX framework and leave this issue for future research. We directly compute this price in our simulations. Notice that if nbBE >
i N BE
Synthesis of the sophisticated pricing model The valuation and pricing functions of our
sophisticated pricing model for the UGS, rtPS, and BE QoS classes are presented in Table 5.1. We proposed to dierentiate between the QoS classes since each class has its QoS requirements. We also dierentiated between the price paid per unit of transmission rate by an rtPS user for its minimum reserved rate Rmin and for its extra rate. Indeed, an rtPS user requires a rate greater than its Rmin and the BS is free to extend the rate up to Rmax . We also took into account the congestion of the network when serving the UGS users and reserving the extra rates
93
for the rtPS users. We nally proposed two pricing schemes for the BE QoS class. In the rst scheme, the operator has to manually choose the symbol price. This price will depend on the valuation functions of the dierent users. In the second scheme, the symbol price is automatically determined by auction.
Tab. 5.1 Synthesis of the sophisticated pricing model.
QoS class
Valuation function
Pricing function
Low
ri ,
VUGS (ri ) := i ri
res if nb nbtot
ri , otherwise
rtPS BE
:= :=
i i
min
+ 2ri Ri max
Rmax
+ 2ri Rmax
94
Fig. 5.2 Sequence diagram of a UGS connection blocked ("symbols < 0").
the user and the current price of the extra rate (extra price ).
0) then the SS
sends a DSA-ACK message that contains a positive response and the connection is established (see Fig. 5.3). Otherwise, the connection is rejected because the price is too high and the SS sends a DSA-ACK message containing a negative response (see Fig. 5.4).
95
management message, the BS computes the number of the OFDM symbols per frame to reserve for Rmin . If there are not enough symbols, the BS sends a DSA-RSP message that contains a negative response (see Fig. 5.5). Otherwise, the BS lls up the basic price eld by Pref ; the price per unit of transmission rate paid for the minimum transmission rate. When the basic scheme is selected, Then, extra price eld is also equal to Pref sinc there is not dierentiation between the basic and the extra rates of the rtPS user. If the sophisticated pricing scheme is selected (with FSPM or VSPM), the BS determines the price per unit of transmission of the extra rate using the pricing constant CrtPS and the number of the remaining OFDM symbols. This price is used to ll up the extra price eld. Then, the BS lls up the extra rate max that depends on the number of remaining OFDM symbols, the MCS used by the rtPS user, and (Rmax -Rmin ). Finally, it sends a DSA-RSP message that contains a positive response. If the valuation function of the SS is greater than the current price paid to reserve Rmin (utility (Rmin )
0) then the SS computes the extra rate that he can pay for and lls up the
extra rate requested eld. Then, the SS sends a DSA-ACK message that contains a positive
96
Fig. 5.4 Sequence diagram of a UGS connection blocked ("utility < 0").
Fig. 5.5 Sequence diagram of an rtPS connection blocked ("symbols < 0").
97
response and the maximum transmission rate that he will reserve (see Fig. 5.6). Otherwise (utility (Rmin ) < 0), the connection is rejected because the price paid for the basic rate (Rmin ) is too high and the DSA-ACK message contains a negative response (see Fig. 5.7).
98
Fig. 5.7 Sequence diagram of an rtPS connection blocked ("utility < 0").
serve ) and the maximum number of symbols that the SS accepts to pay (MaxSymbolsPay ). RemainingSymbolsReserve depends on the size of the le not transmitted yet and the MCS used. MaxSymbolsPay is constant when the basic or sophisticated with FSPM schemes are selected.
Indeed, the price of a symbol remains constant and is xed by the operator. However, the symbol price is determined by auction when the sophisticated with VSPM scheme is used and then it is variable. Evidently, The BS does not reserve more symbols than the SS needs (RemainingSymbols-
Reserve ) or the SS can pay for (MaxSymbolsPay ). The maximum number of symbols that will
be allocated, called MaxSymbols , is equal to the minimum of RemainingSymbolsReserve and
MaxSymbolsPay . If MaxSymbols is greater than ShareSymbols , the BS allocates the whole corresponding share to the SS and updates the number of remaining symbols to reserve (Remaining-
SymbolsReserve ). Otherwise, the BS allocates only MaxSymbols . The le is completely transmitted if MaxSymbols is equal to RemainingSymbolsReserve. The main steps of the BE scheduler are represented in Fig. 5.8.
Simulation Model
99
100
classes in order to have an heterogeneous WiMAX environment with real-time and non-real-time users : UGS, rtPS, and BE. We carried out Matlab simulations in order to analyze our pricing mechanisms. We use ow level simulation. We assume that once the price is determined for a UGS or rtPS connection, it remains constant during the whole connection. We consider that once a UGS or an rtPS user is admitted, the price he pays does not change during his service. When the FSPM is used, the symbol price paid by BE users remains constant during the simulation. However, the rate of a BE user changes when the number of remaining free symbols changes (at each arrival or departure of a user). When VSPM is used, recall that the symbol price is determined by auction. Since this price depends on the available free symbols, it changes with an arrival or departure of a user. The implementation of the pricing schemes in NS-2 will be done in future works. We always consider the OFDM Physical layer and we select the FDD mode. Recall that we have selected the OFDM PHY layer in the TDD mode when evaluating our proposed scheduler and estimating the WiMAX capacity. Our study could be extended to the OFDM layer in the TDD mode and/or to the OFDMA layer. Both modes should provide additional latitude. Indeed, in the TDD mode, the number of symbols allocated for the downlink subframe (and then for the uplink subframe) can be adjusted by modifying the downlink-to-uplink ratio of the frame. On the other hand, the allocation unit in the OFDMA layer is the sub-channel instead of the symbol in the OFDM layer. Since the allocation unit becomes smaller, there is more exibility in scheduling and pricing decisions. We select a bandwidth (BW ) equal to 5 MHz, sampling factor (n ) equal to 144/125, guard time ratio (G ) equal to 1/4, and frame duration of 20 ms. According to [Nua07], the OFDM
G symbol duration (symbol duration ) is equal to n 1+ BW . Then, we obtain 360 symbols per (uplink NFFT or downlink) frame.
The prices that the UGS, rtPS, and BE users are willing to pay (in MU by Kbps transBE mitted) are uniformly distributed on the intervals [UGS ,UGS ], [rtPS ,rtPS ], and [BE 1 2 1 2 1 ,2 ],
101
SNR range (dB) [6, 8.5[ [8.5, 11.5[ [11.5, 15[ [15, 19[ [19, 21[ [21,+[
Useful bits per symbol 192 288 384 576 768 864
Proportion of users 30 % 10 % 20 % 15 % 5% 20 %
Tab. 5.3 Repartition of UGS and rtPS users for each possible transmission rate
Transmission rate (Kbps) R = 300 R = 500 R = 800 [Rmin - Rmax ] = [100 - 150]
Proportion of users 25 % 25 % 50 % 25 % 25 % 25 % 25 %
rtPS
[Rmin - Rmax ] = [150 - 250] [Rmin - Rmax ] = [200 - 350] [Rmin - Rmax ] = [250 - 450]
We assume that the duration of UGS and rtPS sessions follows an exponential distribution with a mean of 10 minutes. BE sessions correspond to le transfer and their durations therefore depend on the transmission rates experienced as well as the initial le sizes. We consider three types of UGS users and four types of rtPS users. The proportion of UGS and rtPS users for each rate is given in Table 5.3. We consider a single type of BE users that have a le size following a normal law with mean 4500 Kbit and standard deviation 200 Kbit (non-positive values are removed).
102
of the le not transmitted divided by the total size of this le. The blocking rates "utility < 0" of the UGS and rtPS connections naturally increase with
Pref , as less and less users are willing to pay the asked price. The blocking rates "symbols < 0"
of the UGS and rtPS connections decrease when Pref increases, since more users refuse to pay the asked price and consequently the number of available symbols increases. When Pref increases without exceeding 0.5, all the BE connections are blocked. This is because the network is overloaded by the UGS and rtPS connections. For 0.5 < Pref < 1.1, the blocking rate of the BE connections decreases with Pref . This is because there are more symbols available. Indeed, UGS and rtPS have refused to pay the current price, and BE connections are still accepting the price asked. For Pref > 1.1, the blocking rate of the BE connections increases. This is because BE users refuse to pay the current price. Fig. 5.10 shows the mean revenue of the UGS, rtPS, and BE connections as a function of
Pref . When Pref increases, the revenue from the UGS, rtPS, and BE connections increases then
decreases. The behavior of the revenue depends on the increase of the price actually paid and the increase of the blocking rate. When the required price becomes too high, the operator looses too much customers and the resources are then under-utilized. If the objective of the pricing scheme is to maximize the revenue, then Pref should be chosen close to 0.7. In this case, the operator revenue is equal to 4000 MU/s, and the social welfare dened in (5.4) is around 5500 MU/s (see Fig. 5.11).
103
PUGS
Low
= PrtPS min = Pref , and UGS = 2, (i.e. PUGS High = 2 Pref ). So, a UGS user
104
is charged a price Pref per unit of transmission when no congestion and 2 Pref per unit of transmission otherwise. We assume that there is congestion when the number of reserved symbols is greater than the remaining free symbols (thUGS = 0.5). Then, we study three parameters of our sophisticated pricing scheme : Pref : represents the price paid, per unit of transmission, by the UGS users when no congestion and by the rtPS users for their minimum transmission rates. All these rates have to be reserved by the BS. CrtPS : represents a constant parameter that aects the price paid, per unit of transmission, by the rtPS users for their extra rates. Symbol PriceBE : represents the price paid, per unit of allocated symbol, by the BE users. Still assuming that the operator will x those values, we should select values that maximize the total revenue. We actually follow a suboptimal strategy, that consists in maximizing iteratively with respect to each of the parameters and x its value. In our work, we do not try to nd the exact optimal values ; we rather want to show that there exist values such that the sophisticated mechanism outperforms the basic one. We rst consider the parameter Pref , and x for the moment CrtPS := 100 and Symbol PriceBE :=
Pref > 0.25. This is due to the high price actually paid and so some UGS users cannot pay to
reserve for their data rates. When Pref is higher than UGS (UGS = 1), the throughput is equal 2 2 to 0 because the utility function is always negative. The throughput of the rtPS connections is also constant when Pref increases and without exceeding 0.25. The rtPS throughput then increases when Pref increases without exceeding 0.75. This is because there are more free symbols when some UGS connections are blocked. The rtPS throughput nally decreases when Pref increases and Pref > 0.75. Indeed, the rtPS users cannot extend their transmissions rate to Rmax and are served,when the utility function > 0, at a rate
r in [Rmin , Rmax ] (depending on the price actually paid). When the current price becomes too
high, the rtPS users cannot pay to reserve their Rmin and then their throughput become equal to zero. In order to show in details the cause of the throughput decrease, we represent the simulation results of the mean blocking rate of the UGS, rtPS, and BE connections as a function of Pref in Fig. 5.13. The blocking rate (because "utility < 0") of the UGS connections increases when
105
Pref increases and this is due to the increase of the price actually paid. When Pref is higher than
UGS , the blocking rate (because "utility < 0") becomes equal to 100% because all the UGS 2
connections are blocked (UGS throughput = 0, see Fig. 5.12). When Pref > 0.5, rtPS connections can be blocked because of the high price actually paid.
106
This happens when rtPS connections cannot pay to reserve Rmin . We observe that the blocking rates (because "symbols < 0") of UGS and rtPS connections decrease and converge to 0 when Pref increases and exceeds 0.25. Indeed, there are more blocked connections due to the high price to pay and then there are more available free symbols. Fig. 5.14 shows the mean revenue of the UGS, rtPS, and BE connections as a function of
Pref . The revenue from the UGS connections increases when Pref increases without exceeding
0.25. This is because no UGS connections is blocked (blocking rate = 0 when Pref < 0.25, see Fig. 5.13) and the price actually paid increases when Pref increases. When Pref increases and
Pref > 0.25, the revenue from the UGS connections decreases. This is due to the high increase
of the blocking rate. When Pref increases, the revenue from the rtPS connections increases when Pref < 0.8 and decreases when Pref > 0.8. The behavior of the revenue depends on the increase of the price actually paid and the increase of the blocking rate. We note that the total revenue increases when Pref < 0.7 and decreases when Pref > 0.7. To obtain a close-to-optimal revenue, we assign 0.7 to Pref , and keep this value for the rest of the performance evaluation.
107
CrtPS . The throughput of the rtPS connections decreases when CrtPS increases, since the price
to pay for an extra rate increases (see (5.11)). As an indirect eect, the throughput of the UGS connections increases, because less rtPS throughput means more symbols are available. Fig. 5.16 shows the blocking rate of the UGS, rtPS, and BE connections as a function of CrtPS . The blocking rate (because "utility < 0") of the UGS connections decreases when CrtPS increases. Indeed, there are more free available symbols when the throughput of the rtPS connections decreases (see Fig. 5.15). Therefore, more UGS connections are charged a price PUGS Low instead of PUGS High per unit of transmission when the proportion of the reserved symbols becomes lower then thUGS (see (5.10)). We also observe that the are less blocked BE connections when CrtPS increases. This is because the price of the extra rate for the rtPS connections decreases when
CrtPS increases and then there are more free available symbols that can be reserved for the BE
users. Fig. 5.17 shows the mean revenue from the UGS, rtPS, and BE connections as a function of CrtPS . When CrtPS increases without exceeding 200, the revenue from the rtPS connections increases. Despite of the decrease of the throughput of the rtPS connections, the price paid for the extra data increases and the rtPS users are still able to pay in order to have an additional rate. For CrtPS > 200, the revenue from the rtPS connections decreases when CrtPS increases. The price paid for extra data becomes too high. Therefore rtPS users decrease their rates until limiting them to their Rmin . The revenues from the UGS and BE users increase and converge when CrtPS increases. Indeed, the UGS and BE throughput increase and converge. Moreover, the price paid by the UGS and BE users does not depend on CrtPS . In order to maximize the
108
total revenue, we x CrtPS to 700 for the rest of the performance evaluation.
109
Fig. 5.18 Blocking rate versus symbol price (sophisticated pricing scheme + FSPM)
Symbol PriceBE . The mean throughput decreases when Symbol PriceBE increases. This is a
direct consequence of the increase of the blocking rate. Fig. 5.20 shows the mean valuation/revenue of the BE and all the connections as a function of
Symbol PriceBE . The revenue from the BE connections increases when Symbol PriceBE increases
without exceeding 0.6 MU/symbol. Indeed, despite the decrease of the satisfaction percentage, the price actually paid increases and most of the BE users still aord buying symbols.
110
Fig. 5.19 Mean throughput versus symbol price (sophisticated pricing scheme + FSPM)
Fig. 5.20 Mean valuation/revenu versus symbol price (sophisticated pricing scheme + FSPM)
When Symbol PriceBE > 0.6MU /symbol , the revenue from the BE connections decreases when Symbol PriceBE increases. This is because that the price actually paid is so high that the majority of the BE users refuse to reserve symbols. We note that there is no revenue from the BE connections when Symbol PriceBE exceeds 1.8 MU/symbol.
111
The value of Symbol PriceBE that maximizes the total revenue is around 0.6 MU/symbol. In this case, the operator collects about 5100 MU/s, and the social welfare (sum of the valuations of all users) is about 7800 MU/s.
Symbol PriceBE , the revenues from these users are the same when using FSPM or VSPM. The
revenues and valuations for each type of connection are given in Table 5.4. For our model and the parameters considered, we notice an increase of the total revenue from 5100 MU/s with xed pricing to 7400 MU/s with auction-based one (an increase of 45%). The drawback is that an auction-based pricing could be more dicult to implement. We do not get into implementation issues and leave this as a future topic of research.
Tab. 5.4 Valuation/revenue of the connections using VSPM
BE 550 450
Basic scheme Network revenue (MU/s) Social welfare (MU/s) 4000 5500
When using the sophisticated scheme with VSPM, the operator earns 77.5% more revenue than that of the basic scheme. When using the sophisticated scheme with FSPM, the operator earns 27.5% more revenue than that of the basic scheme. Moreover, this is not at the expense of social welfare (or users' comfort), since the sophisticated mechanism also outperforms the basic one with respect to that measure. Our proposed sophisticated pricing scheme exploits better the
112
user diversity in terms of service requirements and willingness to pay, and guarantees a better allocation of the available resource (OFDM symbols), by giving them to users who value it most. From the operator's point of view, the sophisticated scheme with VSPM is clearly the most protable (earns 45% more revenue than that when using FSPM). However, one might expect that auctioning for radio resource would be badly perceived by users as it is now well-known that users are risk-averse and prefer to know in advance the price they are going to pay. Another problem of VSPM is the practical realization of BE auctions. Then, the sophisticated scheme with FSPM could be the best option on the short term. We now verify that even if the parameters of our sophisticated scheme are not perfectly optimized, the network revenue largely exceeds the one yielded by the basic pricing mechanism. In Table 5.6, we show the results when the pricing parameters are dierent from the optimal ones in order to have indications about the sensitivity of our proposed pricing scheme. We assume that the operator selects the parameters values that maximize its revenue with a maximum digression of 20% (maximum dierence of +20% or -20% between the optimal value and the value chosen). In Table 5.6, spopt is the optimal value of parameter Symbol PriceBE when the FSPM is used, RFSPM (sp ) is the total revenue when Symbol PriceBE is equal to sp , spmin is the symbol price that minimizes the revenue when Symbol PriceBE belongs to [spopt 0.8, spopt 1.2], and RVSPM is the total revenue when VSPM is applied. The loss of revenue is at most around 10% when using VSPM or FSPM even if there is a 20% maximum digression of optimal values for three parameters : Pref , CrtPS , and Symbol PriceBE . Even if the operator does not choose optimal values, he earns 62.5% more revenue, when using the sophisticated scheme with VSPM, than that when using the basic scheme. When using the sophisticated scheme with FSPM, the operator earns 13% more revenue than that of the basic scheme. So our sophisticated pricing schemes (with VSPM or FSPM) provide better performance even if there is a dierence of 20% more or less between the optimal values and the values chosen. All these earnings are obtained when the parameter value of the basic pricing model is optimal. When the operator selects the parameter of the basic pricing model with a dierence of 20% more or less with the optimal value, he earns 3800 MU/s (-5.2%) and 3620 MU/s (-10.5%) instead of its maximum revenue (4000 MU/s, see Table 5.4), respectively. When using VSPM with optimal values, the operator earns 45% more revenue than that when using FSPM with optimal values. He earns 63.7% more revenue than that when using FSPM with a 20% maximum digression of optimal values. We note that when the operator well-denes the parameters of VSPM, earnings are very noteworthy. With a 20% maximum digression of the optimal values when using VSPM, the operator earns 27.4% more revenue than that when using FSPM with optimal values. He earns 43.8% more revenue than that when using FSPM with a 20% maximum digression of optimal values. We note that when the operator does not select the optimal values of VSPM parameters, earnings are still outstanding. When using VSPM with a 20% maximum digression of optimal values, the operator earns
Conclusion
113
Tab. 5.6 Total revenues when using VSPM and FSPM with a 20% maximum digression of
CrtPS CrtPS
= 700
(optimal value)
spopt
0.6 0.55 0.5 0.6 0.55 0.55 0.5 0.6 0.5
RFSPM (spmin )
5000 (-2%) 4970 (-2.6%) 5040 (-1.2%) 4540 (-11%) 4560 (-10.6%)
RFSPM (spopt )
RVSPM
5100
(opti-
7400 (optimal
revenue) 7200 (-2.7%) 7200 (-2.7%)
mal revenue) 5050 (-0.9%) 5090 (-0.2%) 4600 (-9.9%) 4620 (-9.5%)
CrtPS = 7001.2
(+20%)
CrtPS = 7000.8
(-20%)
760
6500 (-12.2%)
6520 (-11.9%)
(optimal value)
CrtPS = 7601.2
(+20%)
CrtPS = 7600.8
(-20%)
= 1000
(optimal value)
CrtPS = 1000
1.2 (+20%)
CrtPS = 1000
0.8 (-20%)
between 2.7% and 12.2% less revenue than its optimal one (7400 MU/s). When using FSPM with a 20% maximum digression of optimal values, the operator earns between 0.2% and 11.4% less revenue than its optimal one (5100 MU/s). In both schemes, the maximum loss in the earnings is when the value of parameter Pref is chosen with a dierence of +20% in comparison with the optimal value. Indeed, when Pref increases, the blocking rate of the UGS connections increases. Since Pref is greater than the optimal value, the UGS revenue decreases. However, when Pref decreases from the optimal value, the UGS revenue increases (while the rtPS revenue decreases) and then the loss of earnings is less signicant.
5.8 Conclusion
In this chapter, we have proposed pricing models for WiMAX systems. These models (basic and more sophisticated) are designed for the UGS, rtPS, and BE QoS classes. In the sophisticated pricing model, we have proposed two pricing schemes for BE : xed and variable symbol price (FSPM and VSPM, respectively). The latter is based on auction. Moreover, our proposed mechanism is easily understandable by users and reasonably fair, and could therefore be imple-
114
mented in some WiMAX systems. We have also described the scheduling algorithms used in a WiMAX context including pricing models. We show that the scheduler has to take into account the characteristics of the pricing model used. We have set a precise simulation framework to study the implications of our proposals, and exhibited some good properties of our pricing model in terms of revenue maximization, throughput and user satisfaction (via the blocking rate). Extensive simulations are conducted to compare between the pricing models and study the behavior of the FSPM and VSPM schemes against the dierent parameters of the pricing models. We showed that the sophisticated scheme outperforms the basic one with respect to two economic performance measures, namely network revenue and social welfare. We then verify that an ecient WiMAX pricing scheme should benet from the diversity of services and user willingness to pay. We also showed that when using the FSPM scheme in the sophisticated pricing model, the operator has to change the symbol price at each network trac change while the price symbol price is automatically determined by auction and can prot from the willingness of the users when using the VSPM scheme. However, the user is generally risk-averse and prefers that the price to pay is known in advance. We also studied the sensitivity of our proposed pricing model. We showed that the maximum loss of revenue is around 10% when using VSPM or FSPM even if there is a dierence of 20% more or less between the optimal values and the values chosen. Moreover, our proposed model still provides better revenue than the basic one.
Chapter 6
Conclusion
In this thesis, we study radio resource allocation in WiMAX. We rst remind some of the main elements of WiMAX. We propose a state-of-the-art of WiMAX Scheduling. We describe classical scheduling algorithms as well as schedulers specically proposed for WiMAX systems. The choice of the scheduling algorithm is highly dependent on the transmission service type and the trac shape in addition to other QoS requirements. We propose a synthesis table for some of the proposed scheduling algorithms. The eciency of a scheduling algorithm can be estimated through simulations. For that, we design and implement a new NS-2 WiMAX module that takes into account the QoS classes and their requirements. Our module supports unicast and contention request opportunities, the link adaptation, and the mechanisms required to add service ows as specied by the IEEE 802.16 standard. The WiMAX module also involves some UGS, rtPS, and BE schedulers. Extensive simulations are conducted to show that the behavior of our UGS, rtPS, and BE schedulers ts with the QoS specications of the IEEE 802.16 standard. We show through simulations that our schedulers outperform the scheduler of the existing module of WiMAX ; the RR scheduler. Moreover, throughput and mean sojourn time values are determined for some QoS classes and using dierent scheduling algorithms. We have also estimated the capacity of the WiMAX systems using dierent schedulers as well as in dierent SNR environments. We also highlight a problem that can appear with the rtPS class and propose a solution to prevent this problem. This allows us to propose a more spectrum-ecient maximum SIR scheduler. We show through simulations that interesting improvements are obtained with the proposed mmSIR scheduler. This scheduler provides a much better mean sojourn time and also delivers more data packets with regard to the mSIR. In this thesis, we have also proposed pricing schemes for WiMAX systems. These schemes are designed for the UGS and rtPS QoS classes that are designed for real-time applications in WiMAX and for the BE QoS class that is designed for non-real-time applications. We propose two dierent pricing schemes for BE : xed and variable symbol price (FSPM and VSPM, respectively). In the rst model, the symbol price is determined by the operator while it is 115
116
Conclusion
determined dynamically using auctions. We have developed a simulation framework to study the implications of our proposals, and exhibited some good properties of our pricing model in terms of revenue maximization, throughput and user satisfaction (via the blocking rate). These proposals include many parameters, which give way to interesting optimizations. The choice of the system parameters depends on the optimization criteria. We also showed that VSPM provides a higher revenue than FSPM. Indeed, VSPM is ecient by determining the suitable value of the price symbol that maximizes the revenue proting from the willingness of the users. Moreover, when FSPM is applied, the operator has to change the symbol price at each network change. However, the user is commonly risk-averse and prefers a constant price symbol. Therefore, it is up to the operator to choose the suitable pricing model. Once our sophisticated pricing model is dened, it is required to evaluate it by a comparison with a basic pricing model. We propose that all the users are charged the same price per unit of transmission when the basic pricing model is used independently of their QoS classes. So, extensive simulations were conducted in order to study the behavior of our pricing schemes. The sophisticated scheme outperforms the basic one with respect to two economic performance measures, namely network revenue and social welfare. We then verify that an ecient WiMAX pricing scheme should benet from the diversity of services and user willingness to pay. Furthermore, we have shown that our proposed pricing model is not very sensitive (maximum decrease of revenue is around 10%) and still provides better revenue even if there is a 20% maximum digression of all optimal values. Our work is done for the OFDM WiMAX PHY layer in the FDD mode. A direction for future work is the study of our proposed scheduler for the OFDMA WiMAX PHY Layer. This layer is used specically for mobile subscribers and the unit of allocation becomes the subchannel instead of the OFDM symbol. We believe our results could be extended with some eort. It is also interesting to extend our pricing scheme to the TDD mode since this duplexing mode provides more latitude. This latitude is provided by choosing the suitable downlink-to-uplink ratio. In our work, we do not consider Adaptive Antenna Systems (AAS). AAS use leads to average SNR increase. Multiple-Input Multiple-Output (MIMO) systems use allows the use of a better MCS for the same SNR values. Consequently, the use of intelligent antennas techniques would give a higher spectrum eciency gure for the realistic scenario. The study of scheduling algorithms in a WiMAX environment with intelligent antennas can be a future topic of research. It could be also interesting to study the WiMAX capacity for other scheduling algorithms and other environments. A WiMAX environment with intelligent antennas techniques can also be a future topic of research. In addition, we can mention the observation of other QoS indicators such as the delay (average or maximum) in our pricing models. New pricing optimization considerations should be addressed. Our remark about possible extension to OFDMA is particularly true as an OFDMA subchannel is a less rigid unit than an OFDM Symbol, especially if we take into account the dierent types of subchannels dened in IEEE 802.16 : full usage of subchannels (FUSC), partial usage of subchannels (PUSC) and Band Adaptive Modulation and Coding (AMC).
Appendix A
118
nrtPS est conue pour les applications en temps non rel ayant des tailles de paquets variables mais exigeant quand mme une sollicitation rgulire du terminal, de sorte garantir un dbit minimal. Enn, la classe BE est conue pour les applications en temps non rels et o aucune qualit de service n'est garantie. Nous nous intressons dans notre thse aux trois classes de services UGS, rtPS, et BE. La couche physique (PHY) du standard IEEE 802.16 supporte deux technologies de transmissions : Single Carrier (SC) et Orthogonal Frequency Division Multiplexing (OFDM). A partir de ces technologies de transmission, le standard a dni cinq couches physiques : WirelessMANSC, WirelessMAN-SCa, WirelessMAN-OFDM, WirelessMAN-OFDMA, et WirelessHUMAN. Le forum WiMAX a retenu seulement deux couches physiques : OFDM et OFDMA. La couche OFDMA est conue pour les stations mobiles. Nous nous intressons la couche OFDM. L'unit d'allocation dans la couche OFDM est le symbole OFDM qui contient 256 porteuses ; 192 porteuses sont conues pour la transmission de donnes utiles, 8 porteuses pilotes pour l'estimation du canal, et 56 porteuses nulles pour lutter contre les interfrences co-canal. Le nombre de symbole OFDM par trame est xe. Ce nombre dpend, entre autres, de la bande de frquence. Cependant, le nombre de bits utiles par symbole (et par consquent par trame) est variable et dpend du schma de modulation et de codage utilis Modulation and Coding Scheme, MCS). Le MCS utilis dpend du rapport signal-bruit (Signal to Noise Ratio, SNR). Par exemple, si le SNR est entre 3 dB et 6 dB, le MCS utilis est BPSK et le nombre de bits utiles par trame est gal 96. Lorsque le SNR augmente, la station peut utiliser un MCS plus ecace. De plus, l'allocation de ressources dpend des demandes de bandes passantes qui sont faites d'une manire non sollicite comme dans le cas d'UGS, d'une manire sollicite en unicast comme dans le cas de la classe de service rtPS, ou d'une manire sollicite en contention comme pour la classe BE. Notons aussi que l'allocation de ressources dpend aussi des contraintes des direntes classes de service. En consquence, il est important de bien dnir un algorithme d'ordonnancement pour utilisation ecace des ressources radio surtout que ces ressources sont rares et que les conditions radio sont variables. Puisque nous nous intressons au mode de transmission point multipoint (Point to multipoint, PMP), la station de base va dcider quels sont les utilisateurs qui vont recevoir des donnes et par la suite elle a besoin d'un algorithme d'ordonnancement dans le sens descendant. C'est la station de base qui va aussi dcider quels sont les stations qui peuvent envoyer des donnes et en consquence elle a besoin d'un algorithme d'ordonnancement dans le sens ascendant. Etant donn qu'un utilisateur pourrait avoir plusieurs connexions, il a besoin d'un algorithme d'ordonnancement dans le sens ascendant pour dcider quelles sont les connexions qui vont tre servies (voir Figure A.1). Les dcisions prises par la station de base sont annonces dans les messages de gestion DL-MAP et UL-MAP qui dcrivent respectivement les allocations en sens descendant et ascendant. Ces messages sont transmis au dbut de chaque trame et leurs tailles dpendent du nombre de stations servis par trame. Nous avons propos une classication des algorithmes d'ordonnancement pour les rseaux
119
WiMAX. Un algorithme d'ordonnancement peut tre classique, c'est--dire indpendant de la technologie de transmission, ou spcialement propos pour WiMAX. Un algorithme classique peut tre systmatique, c'est--dire qui en tient en compte que des demandes de bandes passantes, ou tenant en compte d'autres caractristiques comme les conditions radio. Un algorithme spcialement propos pour WiMAX peut tre propos pour une seule classe de service ou pour plusieurs. Nous dcrivons brivement deux algorithmes d'ordonnancement pour chaque classe. Pour les algorithmes classiques systmatiques, le plus connu est le " tour de rle " Round Robin (RR) o la station de base distribue quitablement les ressources radio entre tous les utilisateurs. Le " tour de rle avec pondration " ou Weighted Round Robin (WRR) est bas sur des poids statiques (wi pour la station i ). La portion de donnes de la station i sera gale Pour les algorithmes d'ordonnancement classiques qui tiennent en compte des conditions radio, nous citons le maximum Signal to Inteference Ratio (mSIR) o la station de base sert les utilisateurs ayant les rapports signal bruit le plus levs. Il y a aussi le Temporary Removal Scheduler (TRS). Tout d'abord la station de base doit dterminer la listes des utilisateurs qui peuvent tre servis en comparant leurs rapports signal bruit avec un certain seuil. Si le rapport est infrieur au seuil, l'utilisateur sera retir temporairement de la liste avec une condition supplmentaire qu'un utilisateur ne peut en aucun cas tre retir plus que L fois successifs. Puisque TRS ne dtermine pas la quantit de ressources allouer chaque utilisateur, il doit tre combin avec un autre algorithme d'ordonnancement. Si TRS est combin par exemple avec le RR
w n i w . k =1 k
120
(TRS+RR), et il y a k utilisateurs dans la liste, la station de base alloue 1/k des ressources pour chaque utilisateur. Si TRS est combin avec le mSIR (TRS+mSIR), la station de base sert les utilisateurs ayant les rapports signal bruit les plus levs et appartenant la liste. Pour les algorithmes d'ordonnancement proposs spcialement pour une seule classe de service, rtPS, nous citons " adaptive rtPS ". Le standard IEEE 802.16 exige que l'utilisateur annonce l'tat de sa le d'attente en envoyant une demande de bande passante. Le problme que les paquets gnrs par la couche applicative entre le temps d'envoi de la demande et le temps de rception de ce paquet par la destination subissent des dlais supplmentaires, l'ide de cet algorithme est de prvoir l'arrive des paquets rtPS. Un autre algorithme d'ordonnancement propos spcialement pour une seule classe de service, BE, nous citons " Adpative Bandwidth Allocation Scheme " (ABAS). Cet algorithme d'ordonnancement est spcialement propos pour les connexions Tansfer Control Protocol (TCP). L'ide de cet algorithme est de dterminer la proportion entre la trame descendante et la trame ascendante. Cette proportion n'a pas t prcise par la norme. Pour les algorithmes d'ordonnancement spcialement proposs pour plusieurs classes de service, nous citons " uplink packet scheduler with Call Admission Control ". L'ide de cet algorithme est de servir les utilisateurs UGS et rtPS, Ensuite, la station de base sert les utilisateurs nrtPS sans dpasser un certain seuil et les utilisateurs BE sans dpasser un deuxime seuil. Enn, la station de base sert les utilisateurs nrtPS et BE s'il reste des symboles OFDM disponibles. Un autre algorithme d'ordonnancement spcialement propos pour plusieurs classes de service, nous citons " scheduler ensuring QoS requirements ". L'ide de cet algorithme est de servir tout d'abord les utilisateurs UGS et ertPS. Ensuite, la station de base fournit les besoins minimums de bande passante pour les utilisateurs rtPS et nrtPS. Puis, la station de base fournit les besoins maximums de bande passante pour les utilisateurs rtPS et nrtPS. Enn, les utilisateurs nrtPS et BE vont tre servis. Dans cette thse, nous proposons une modication aux algorithmes d'ordonnancement spcialement proposs pour la classe de service rtPS. Nous avons signal un problme de synchronisation entre la demande et l'allocation des ressources radio. En eet, lorsque la station de base alloue une opportunit de demande de bande passante et une allocation de donnes pour le mme utilisateur et dans la mme trame, la dernire demande de bande passante ne peut pas tre tenue en compte immdiatement par la station de base puisque la dcision de la taille d'allocation de donnes a t dj prise au dbut de la trame et annonce dans le message de gestion UL-MAP (voir Figure A.2). Nous avons propos alors une amlioration du droulement de ce mcanisme. Cette modication consiste servir seulement les utilisateurs qui n'ont pas des opportunits de demande de bande passante. Nous avons utilis la simulation pour valuer notre proposition qui est appliqu l'algorithme d'ordonnancement mSIR. Pour cela, nous avons implment un nouveau module WiMAX qui supporte les direntes classes de services dans le simulateur NS-2. Nous avons implment
121
Fig. A.2 Problme de synchronisation pour les algorithmes proposs pour la classe de service
rtPS. les dirents mcanismes pour l'identication des besoins de service en utilisant les messages de gestions : Dynamic Service Addition (DSA), Dynamic Service Change (DSC), et Dynamic Service Delete (DSD). Puisque les utilisateurs peuvent avoir des rapports signal bruit dirents, nous avons implment un mcanisme d'adaptation de lien pour slectionner le MCS utiliser partir du SNR de l'utilisateur. Nous avons aussi implment les mcanismes de demande de bande passante en unicast et en contention aussi que des algorithmes d'ordonnancements pour les classes de service UGS, rtPS, BE an d'valuer notre proposition, raliser une comparaison entre les dirents algorithmes et eectuer une estimation de la capacit des systmes WiMAX qui dpend du type d'algorithme d'ordonnancement utilis aussi bien de l'environnement du scnario. Les rsultats de simulations obtenus montrent que notre algorithme d'ordonnancement ore une amlioration du temps moyen de sjour. En eet, lorsque les utilisateurs, ayant des SNR levs, possdent des opportunits de demande de bande passante, la station de base peut servir les utilisateurs qui ont un SNR rduit et par consquent ces utilisateurs ne seront pas compltement bloqus. Nous avons bien vri qu'il n'y a pas de dgradation de dbit puisque notre proposition favorise aussi les stations utilisant les schmas de modulation et de codage les plus ecaces. En plus, notre proposition sert moins de stations par trame que l'algorithme d'ordonnancement mSIR. Dans le Tableau A.1, nous prsentons quelques rsultats d'estimation de la capacit des rseaux WiMAX en mesurant le paramtre " ecacit spectrale " qui est gal au dbit divis par la bande passante. Nous choisissons une bande passante gale 5 MHz. Les rsultats des simulations sont prsents pour trois scnarios : mauvaises, uctuantes, et bonnes conditions radio.
122
Tab. A.1 Ecacit spectrale pour dirents environnements et dirents algorithmes d'ordon-
spectrale
Tab. A.2 Revenu total pour les modles de tarication basique et sophistiqu.
Basique 4000
FSPM 5100
VSPM 7400
Notre deuxime axe de recherche porte sur les politiques de tarication de WiMAX. Parmi les nombreux travaux de recherches portant sur les aspects techniques de WiMAX, trs peu ont inclus les implications conomiques. La tarication est le fait d'aecter des prix aux ressources. Selon le prix propos, l'utilisateur va dcider s'il va se connecter ou non. Par consquence, la tarication a un eet important sur l'utilisation des ressources. Une optimisation dans l'utilisation des ressources partir d'un algorithme d'ordonnancement adquat et un modle de tarication adquat orent l'oprateur un revenu lev. Nous avons propos des modles de tarication pour les applications en temps rel (classes UGS et rtPS) ainsi que pour des applications en temps non-rels (classe BE). Les modles de tarication proposs pour les applications en temps rel dpendent de l'tat du rseau, notamment la congestion. Ces modles font aussi la direnciation entre les prix pays pour les dbits ncessaires pour rpondre tous les besoins en QoS des connexions temps rels et les prix pays pour les dbits supplmentaires. En ce qui concerne les modles de tarication proposs pour les applications en temps non-rels, ils se basent ou bien sur un prix x par l'oprateur (l'option est appele Fixed Symbol Price Model, FSPM) ou sur un prix variable qui se dtermine dynamiquement par des enchres (l'option est appele Variable Symbol Price Model, VSPM). Au passage, nous mettons en vidence que les algorithmes de scheduling doivent tenir en compte du modle de tarication utilis. Nos mcanismes sont valus et compars avec un modle de tarication trs basique qui charge tous les utilisateurs de la mme faon. Les rsultats de simulation montrent l'importance de dnir un modle de tarication adquat pour le systme WiMAX. Le Tableau A.2 prsente les revenus en utilisant le modle basique et le modle sophistiqu avec ses deux options (FSPM et VSPM). Du point de vue oprateur, l'utilisation du modle sophistiqu avec l'option VSPM ore le revenu le plus lev. Par contre, il faut noter que l'utilisateur prfre en gnral connatre le prix l'avance. C'est l'oprateur de slectionner l'option de notre modle propos selon le
123
Tab. A.3 Revenu total pour les modles de tarication basique et sophistiqu avec une dirence
de 20% , 30%, et 40% entre les valeurs choisies et valeurs optimales. Schma de ta- Dirence de 20% Dirence de 30% Dirence de 40% rication Basique FSPM VSPM avec les valeurs optimales 3620 (-10.5%) 4520 (-11.4%) 6500 (-12.2%) avec les valeurs optimales 3300 (-17.5%) 4320 (-15.3%) 6460 (-12.7%) avec les valeurs optimales 3270 (-18.3%) 3170 (-37.9%) 4560 (-38.4%)
type des utilisateurs. Nous avons aussi tudi la sensibilit du choix des paramtres du modle sophistiqu. Nous prsentons dans le Tableau A.3 les revenus obtenus en cas o l'oprateur n'arrive pas dnir exactement les valeurs optimales des paramtres qui maximisent le revenu (avec une dirence de 20%, 30%, et 40%). Les rsultats de simulation montrent que notre modle sophistiqu reste performant mme lorsqu'il y a une dirence de 30% entre les valeurs choisies et les valeurs optimales (revenu suprieur 4000 unit montaire/s). Nous remarquons que la perte de revenu devient assez importante (environ 38%) lorsque la dirence entre les valeurs choisies et les valeurs optimales gale 40%.
124
List of Acronyms
1G 2G 2.5G 3G 3GPP 4G AAS ABAS AMC AMPS AP aPS ATM BE BPSK BS BW BWA CAC CDMA CID COL CP CPS CS DCD DIUC DL DL-MAP
rst generation second generation second-and-half generation third generation third Generation Partnership Project forth generation Adaptive Antenna Systems Adaptive Bandwidth Allocation Scheme Adaptive Modulation and Coding Advanced Mobile Phone Service Access Point adaptive Polling Service Asynchronous Transfer Mode Best Eort Binary Phase Shift Keying Base Station Bandwidth Broadband Wireless Access Call Admission Control Code Division Multiple Access Connection Identier Competitive On-Line Cyclic Prex Common Part Sublayer Convergence Sublayer Downlink Channel Descriptor Downlink Interval Usage Code Downlink Downlink map 125
126
List of Acronyms
DRR DSA DSC DSCH DSD DSL EDF EDGE email EPS ertPS FCH FDD FEC FSPM FTP FUSC GPRS GSM HSDPA HSUPA HTTP IE IFFT IP LAN LOS LTE MAC MCS MIMO mmSIR MPEG MSH mSIR NDSL NIST NLOS
Decit Round Robin Dynamic Service Addition Dynamic Service Change Distributed scheduling Dynamic Service Delete Digital Subscriber Line Earliest Deadline First Enhanced Data Rates for GSM Evolution electronic mail Evolved Packet System extended real-time Polling Service Frame Control Header Frequency Division Duplex Forward Error Correction Fixed Symbol Price Model File Transfer Protocol full usage of subchannels General Packet Radio Service Global System for Mobile communications High Speed Downlink Packet Access High Speed Uplink Packet Access HyperText Transfer Protocol Information Element Inverse Fast Fourier Transform Internet Protocol Local Area Network Line of Sight Long Term Evolution Medium Access Control Modulation and Coding Scheme Multiple-Input Multiple-Output modied maximum Signal-to-Interference Ratio Moving Picture Experts Group Mesh maximum Signal-to-Interference Ratio Network and Distributed Systems Laboratory National Institute of Standards and Technology Non Line of Sight
List of Acronyms
127 non real-time Polling Service Opportunistic Decit Round Robin Orthogonal Frequency Division Multiplexing Orthogonal Frequency Division Multiple Access Protocol Data Unit physical point-to-multipoint partial usage of subchannels Quadrature Amplitude Modulation Quality of Service Quadrature Phase Shift Keying Request Round Robin Response Receive/transmit Transition Gap real-time Polling Service Service Access Point Single Carrier Service Data Unit Service Flow Identier Single In Single Out Small and Medium Enterprises Short Message Service Signal-to-Noise Ratio Subscriber Station Total Access Communication System Transmission Control Protocol Time Division Duplex Task Group m Temporary Removal Scheduler Transmit/receive Transition Gap Uplink Channel Descriptor User Datagram Protocol Unsolicited Grant Service Uplink Interval Usage Code Uplink Uplink map Universal Mobile Telecommunications System
nrtPS O-DRR OFDM OFDMA PDU PHY PMP PUSC QAM QoS QPSK REQ RR RSP RTG rtPS SAP SC SDU SFID SISO SME SMS SNR SS TACS TCP TDD TGm TRS TTG UCD UDP UGS UIUC UL UL-MAP UMTS
128
List of Acronyms
VOD VoIP VPN VSPM WiFi WirelessMAN WirelessHUMAN WLAN WiMAX WLL WRR
Video on demand Voice over Internet Protocol Virtual Private Network Variable Symbol Price Model Wireless Fidelity Wireless Metropolitan Area Networks Wireless High-speed Unlicensed Metropolitan Area Networks Wireless Local Area Network Worldwide Interoperability for Microwave Access Wireless Local Loop Weighted Round Robin
Publications
Book Chapter
1. Aymen Belghith and Lout Nuaymi (2009). Scheduling Techniques for WiMAX. In : Maode Ma. Current Technology Developments of WiMAX Systems. Springer. Pages 61-84.
International Conferences
1. Aymen Belghith, Lout Nuaymi, and Patrick Maill, "Pricing of dierentiated-QoS services WiMAX networks", IEEE Global Communication Conference, IEEE GLOBECOM 2008, pages 1-6, New Orleans, LA, USA, 30 November - 4 December 2008. 2. Aymen Belghith, Lout Nuaymi, and Patrick Maill, "Pricing of Real-Time Applications in WiMAX Systems", IEEE 68th Vehicular Technology Conference, VTC2008-Fall, pages 1-6, Calgary, Alberta, Canada, 21-24 September 2008. 3. Aymen Belghith and Lout Nuaymi, "Comparison of WiMAX scheduling algorithms and proposals for the rtPS QoS class", 14th European Wireless 2008, EW2008, pages 1-6, Prague, Czech Republic, 22-25 June 2008. 4. Aymen Belghith and Lout Nuaymi, "WiMAX capacity estimations and simulation results", IEEE 67th Vehicular Technology Conference, VTC2008-Spring, pages 1741-1745, Marina Bay, Singapore, 11-14 May 2008. 5. Aymen Belghith and Lout Nuaymi, "Design and Implementation of a QoS-included WiMAX Module for NS-2 Simulator," First International Conference on Simulation Tools and Techniques for Communications, Networks and Systems, SIMUTools 2008, Marseille, France, 3-7 March 2008.
129
130
Publications
Bibliography
[ARV99] Jrn Altmann, Bjrn Rupp, and Pravin Varaiya. Internet demand under dierent pricing schemes. ACM conference on Electronic Commerce, EC-99, pages 9 14, November 1999. [Bag84] [BB06] [BBM07] Arunabha Bagchi. Stackelberg Dierential Games in Economic Models. SpringerVerlag, June 1984. David Besanko and Ronald R. Braeutigam. Microeconomics. Wiley, 2 edition, September 2006. Abdelfettah Belghith, Aymen Belghith, and Miklos Molnar. Enhancing psm eciencies in infrastructure 802.11 networks. International Journal of Computing &
Tools and Techniques for Communications, Networks and Systems, SIMUTools 2008,
March 2008. [BN08c] Aymen Belghith and Lout Nuaymi. Wimax capacity estimations and simulation results. IEEE 67th Vehicular Technology Conference, VTC2008-Spring, May 2008. [BTGK05] Carstin Ball, Franz Treml, Xavier Gaube, and Anja Klein. Performance analysis of temporary removal scheduling applied to mobile wimax scenarios in tight frequency reuse. European Transactions on Telecommunications, 2(16) :888894, September 2005. 131
132 [CCC07]
Bibliography Ben-Jye Chang, Yan-Ling Chen, and Chien-Ming Chou. Adaptive hierarchical polling and cost-based call admission control in ieee 802.16 wimax. Wireless Communications
and Networking Conference, 2007, WCNC 2007, pages 1954 1958, March 2007.
[CH04] Ben-Jye Chang and Ren-Hung Hwang. Distributed cost-based update policies for qos routing on hierarchical networks. Information Sciences, 159 :87 108, January 2004. [Cha08] [CJG05] Bonnie Cha. Nokia n810 internet tablet wimax edition makes ocial debut. http ://computershopper.com.com/8301-12261 7-9907141-51.html, April 2008. Jianfeng Chen, Wenhua Jiao, and Qian Guo. Providing integrated qos control for ieee 802.16 broadband wireless access systems. IEEE 62th Vehicular Technology
Bibliography
133
[GBK+ 07] Emiliano Guainella, Eugen Borcoci, Marcos Katz, Pedro Neves, Marilia Curado, Fausto Andreotti, and Enrico Angori. Wimax technology support for applications in environmental monitoring, re prevention and telemedicine. Mobile WiMAX Sym-
bretagne.fr/ xlagrang/Publis/R2000D01-XL.pdf/, 2000. Jang-Won Lee, Mung Chiang, and A. Robert Calderbank. Price-based distributed algorithms for rate-reliability tradeo in network utility maximization. IEEE Journal
134 [LWG06]
Bibliography Qingwen Liu, Xin Wang, and Georgios B. Giannakis. A cross-layer scheduling algorithm with qos support in wireless networks. IEEE Transactions on Vehicular
2009. http ://www.intomobile.com/2008/12/28/wibro-voice-calls-coming-to-southkorea-wimax-networks-in-2009.html, December 2008. [RBS06] Hemant Kr Rath, Abhijeet Bhorkar, and Vishal Sharma. An opportunistic drr (odrr) uplink scheduling scheme for ieee 802.16-based broadband wireless networks.
Modeling analysis and simulation of wireless and mobile systems, MSWiM '06, pages
108 117, October 2006. [San08] Nil Sanyas. lent. [SSP02] [STS04] [STS06] [Tan02] [tec05] [tec06] [TFB06] Le wimax en france accuse un dveloppement bien trop http ://www.pcinpact.com/actu/news/46016-WiMAX-boucle-locale-radio-
arcep.htm, September 2008. Emmanuel Seurre, Patrick Savelli, and Pierre-Jean Pietri. Gprs for Mobile Internet. December 2002. IEEE Computer Society, IEEE Microwave Theory, and Techniques Society. Part 16 : Air interface for xed broadband wireless access systems. October 2004. IEEE Computer Society, IEEE Microwave Theory, and Techniques Society. Part 16 : Air interface for xed and mobile broadband wireless access systems. Febrary 2006. Andrew S. Tanenbaum. Computer Networks. Prentice Hall PTR, 4 edition, August 2002. Deploying license-exempt wimax solutions. Technical report, Intel, 2005. Considerations for deploying mobile wimax at various frequencies. Technical report, Nortel, 2006. Daniele Tarchi, Romano Fantacci, and Marco Bardazzi. Quality of service management in ieee 802.16 wireless metropolitan area network. International Conference
Bibliography Seamless and secure mobility. http ://www.antd.nist.gov/seamlessandsecure.shtml, September 2007. Middle Wimax east operators in committed south to wimax for deployment 2008. over http 2009. ://rfdeOchttp ://www.ameinfo.com/175230.html, November 2008. deployment africa sign.com/next generation wireless/news/africa wimax deployment 1028/, tober 2008.
[url09] [VSJD06]
Signals & communications. http ://www.springer.com/engineering/signals/book/9781-4020-9299-2 ?detailsPage=toc, January 2009. Kalyan Vinay, N. Sreenivasulu, D. Jayaram, and Debabrata Das. Performance evaluation of end-to-end delay by hybrid scheduling algorithm for qos in ieee 802.16 network. Wireless and Optical Communications Networks, 2006 IFIP International
List of Figures
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 Orthogonal Frequency Division Multiplexing (OFDM) symbol structure. . . . . . Time Division Duplexing frame format. . . . . . . . . . . . . . . . . . . . . . . . Frequency Division Duplexing frame format. . . . . . . . . . . . . . . . . . . . . . Downlink subframe format. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Time Division Duplexing (TDD) frame structure in IEEE 802.16. . . . . . . . . . Example of the DL-MAP structure. The number of served SSs in the downlink direction is equal to 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Example of the UL-MAP structure. The number of served SSs in the uplink direction is equal to 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Uplink subframe format. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Medium Access Control (MAC) sublayers. . . . . . . . . . . . . . . . . . . . . . . request. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.11 Addition of a new service ow. The BS initializes the dynamic service addition request. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.12 Exchanges between MAC and CS layers to add a new service ow. . . . . . . . . 2.13 Detailed behavior of an SS-initiated DSA-REQ. . . . . . . . . . . . . . . . . . . . 2.14 Detailed behavior of a BS-initiated DSA-REQ. . . . . . . . . . . . . . . . . . . . 2.15 Detailed behavior of an SS-initiated DSC-REQ. . . . . . . . . . . . . . . . . . . . 2.16 Detailed behavior of a BS-initiated DSC-REQ. . . . . . . . . . . . . . . . . . . . 2.17 Detailed behavior of an SS-initiated DSD-REQ. . . . . . . . . . . . . . . . . . . . 2.18 Detailed behavior of a BS-initiated DSD-REQ. . . . . . . . . . . . . . . . . . . . 2.19 Example of a burst prole threshold usage. . . . . . . . . . . . . . . . . . . . . . 2.20 Packets Scheduling in BS and SS. The uplink scheduler may have dierent scheduling classes depending on the service type. . . . . . . . . . . . . . . . . . . . . . 2.21 Resources allocation in mesh distributed scheduler. . . . . . . . . . . . . . . . . . 3.1 3.2 3.3 class diagram of NIST module. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Steps of the creation of a new service ow. . . . . . . . . . . . . . . . . . . . . . . Unicast request opportunities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 33 34 39 41 42 24 25 26 27 28 28 29 29 32 20 20 23 24 19 16 17 17 18 19
2.10 Addition of a new service ow. The SS initializes the dynamic service addition
List of Figures Contention request opportunities. . . . . . . . . . . . . . . . . . . . . . . . . . . . Unsolicited Grant Service (UGS) scheduling. . . . . . . . . . . . . . . . . . . . . . real-time Polling Service (rtPS) scheduling. . . . . . . . . . . . . . . . . . . . . . Best Eort (BE) scheduling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . UGS throughput versus oered UGS trac load. . . . . . . . . . . . . . . . . . . Mean sojourn time of UGS SSs versus oered UGS trac load. . . . . . . . . . . 43 44 46 47 49 50 50 51 52 55 57 58 61 66 70 71 72 74 75 75 80 80 81 82 83 83 89 94 95 96 96 97 98 99
3.10 rtPS throughput versus oered rtPS trac load. . . . . . . . . . . . . . . . . . . 3.11 Mean sojourn time of rtPS SSs versus oered rtPS trac load. . . . . . . . . . . 3.12 BE throughput versus oered UGS trac load. . . . . . . . . . . . . . . . . . . . 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 Packets queues for Weighted Round Robin (WRR) scheduling algorithm. . . . . . Packets queues for Decit Round Robin (DRR) scheduling algorithm. . . . . . . . Use of DRR and WRR schedulers. . . . . . . . . . . . . . . . . . . . . . . . . . . Main steps of determination of scheduled SSs using the O-DRR scheduler. . . . . Main steps of the uplink packet scheduler with CAC. . . . . . . . . . . . . . . . . Main steps of a scheduler ensuring QoS requirements. . . . . . . . . . . . . . . . . Allocation of symbols for rtPS connection. . . . . . . . . . . . . . . . . . . . . . . Main steps of the proposed mmSIR scheduler. . . . . . . . . . . . . . . . . . . . . Mean sojourn time versus trac load. . . . . . . . . . . . . . . . . . . . . . . . .
4.10 Number of delivered data packets versus trac load. . . . . . . . . . . . . . . . . 4.11 Mean number of served SSs per frame versus trac load . . . . . . . . . . . . . . 4.12 Spectrum eciency versus rtPS trac load (bad radio channel scenario). . . . . . 4.13 Mean sojourn time versus rtPS trac load (bad radio channel scenario). . . . . . 4.14 Spectrum eciency versus rtPS trac load (uctuant radio channel scenario). . . 4.15 Mean sojourn time versus rtPS trac load (uctuant radio channel scenario). . . 4.16 Spectrum eciency versus rtPS trac load (good radio channel scenario). . . . . 4.17 Mean sojourn time versus rtPS trac load (good radio channel scenario). . . . . 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 Valuation function of the rtPS QoS class. . . . . . . . . . . . . . . . . . . . . . . Sequence diagram of a UGS connection blocked ("symbols < 0"). . . . . . . . . . Sequence diagram of the establishment of a UGS connection. . . . . . . . . . . . Sequence diagram of a UGS connection blocked ("utility < 0"). . . . . . . . . . . Sequence diagram of an rtPS connection blocked ("symbols < 0"). . . . . . . . . . Sequence diagram of the establishment of an rtPS connection. . . . . . . . . . . . Sequence diagram of an rtPS connection blocked ("utility < 0"). . . . . . . . . . Main steps of the BE scheduler. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mean blocking rates (%) in the system versus Pref (basic pricing scheme). . . . . 102
5.10 Mean revenue versus Pref (basic pricing scheme). . . . . . . . . . . . . . . . . . . 103 5.11 Mean valuation versus Pref (basic pricing scheme). . . . . . . . . . . . . . . . . . 103
List of Figures
139
5.12 Mean throughput versus Pref (sophisticated pricing scheme). . . . . . . . . . . . . 105 5.13 Mean blocking rate versus Pref (sophisticated pricing scheme). . . . . . . . . . . . 105 5.14 Mean revenue versus Pref (sophisticated pricing scheme). . . . . . . . . . . . . . . 106 5.15 Mean throughput versus CrtPS (sophisticated pricing scheme). . . . . . . . . . . . 107 5.16 Mean blocking rate versus CrtPS (sophisticated pricing scheme). . . . . . . . . . . 108 5.17 Mean revenue versus CrtPS (sophisticated pricing scheme). . . . . . . . . . . . . . 108 5.18 Blocking rate versus symbol price (sophisticated pricing scheme + FSPM) . . . . 109 5.19 Mean throughput versus symbol price (sophisticated pricing scheme + FSPM) . . 110 5.20 Mean valuation/revenu versus symbol price (sophisticated pricing scheme + FSPM)110 A.1 Scheduling dans la station de base et l'utilisateur. . . . . . . . . . . . . . . . . . . 119 A.2 Problme de synchronisation pour les algorithmes proposs pour la classe de service rtPS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
140
List of Figures
List of Tables
2.1 2.2 2.3 3.1 3.2 4.1 4.2 4.3 4.4 4.5 4.6 5.1 5.2 5.3 5.4 5.5 5.6 Number of Useful Bits per OFDM Symbol . . . . . . . . . . . . . . . . . . . . . . Total useful throughput (expressed in Mbps) for a downlink FDD frame of 20 ms. Essential Characteristics of the QoS Classes . . . . . . . . . . . . . . . . . . . . . Receiver SNR assumptions (values of the IEEE 802.16e standard) . . . . . . . . . Main parameters of the simulation model . . . . . . . . . . . . . . . . . . . . . . Performance evaluation of the residential and SME scenario. Downlink scheduler is DRR and uplink scheduler is WRR. . . . . . . . . . . . . . . . . . . . . . . . . Main Parameters of the Simulation Model . . . . . . . . . . . . . . . . . . . . . . Synthesis of systematic schedulers . . . . . . . . . . . . . . . . . . . . . . . . . . . Synthesis of radio channel conditions-aware schedulers . . . . . . . . . . . . . . . Synthesis of schedulers proposed for one QoS class . . . . . . . . . . . . . . . . . Synthesis of schedulers proposed for many QoS classes . . . . . . . . . . . . . . . Synthesis of the sophisticated pricing model. . . . . . . . . . . . . . . . . . . . . . 59 73 76 76 77 78 93 15 22 30 40 48
Distribution of the MCSs for Arriving Users . . . . . . . . . . . . . . . . . . . . . 101 Repartition of UGS and rtPS users for each possible transmission rate . . . . . . 101 Valuation/revenue of the connections using VSPM . . . . . . . . . . . . . . . . . 111 Comparison of economic performance of pricing schemes . . . . . . . . . . . . . . 111 Total revenues when using VSPM and FSPM with a 20% maximum digression of optimal values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
A.1 Ecacit spectrale pour dirents environnements et dirents algorithmes d'ordonnancement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 A.2 Revenu total pour les modles de tarication basique et sophistiqu. . . . . . . . 122 A.3 Revenu total pour les modles de tarication basique et sophistiqu avec une dirence de 20% , 30%, et 40% entre les valeurs choisies et valeurs optimales. . . 123
141
Rsum
La technologie de transmission large bande sans l (BWA, Broadband Wireless Access) IEEE 802.16, certie par le Forum WiMAX (Worldwide Interoperability for Microwave Access Forum), connue donc sous le nom WiMAX est un des premiers systmes conu comme multiservice ds le dpart et avec une transmission oriente paquet. Par consquent, l'ordonnancement (scheduling) joue un rle particulier dans l'utilisation ecace des ressources radio. L'ordonnancement reprsente la manire d'allouer les ressources aux dirents utilisateurs en fonction des direntes demandes et, d'autre part, de la charge du rseau. Le systme WiMAX/IEEE 802.16 a dni cinq classes de services ainsi que leurs exigences. Cependant, la norme n'a pas spci le(s) dirents algorithme(s) radio utiliser. Dans cette thse, nous commenons par tudier le scheduling de WiMAX. Nous prsentons un tat de l'art de ces algorithmes, o nous direncions entre les algorithmes classiques et ceux proposs spcialement pour les rseaux WiMAX et indiquons les avantages et inconvnients de chacun de ces algorithmes. Nous nous intressons la classe de service rtPS en signalant un problme de synchronisation entre la demande et l'allocation des ressources radio. Nous proposons alors une amlioration du droulement de ces mcanismes en l'appliquant sur plusieurs exemples. Nous utilisons la simulation pour valuer nos propositions. Pour cela, nous implmentons un nouveau module WiMAX qui supporte les direntes classes de services ainsi que certains mcanismes dnis par la norme. Nous implmentons aussi notre proposition ainsi que d'autres autres algorithmes d'ordonnancements pour les classes de service UGS, rtPS, BE an d'valuer notre proposition, raliser une comparaison entre les dirents algorithmes et eectuer une estimation de la capacit des systmes WiMAX qui dpend du scheduler et de l'environnement. Notre deuxime axe de recherche porte sur les politiques de tarication de WiMAX. Parmi les nombreux travaux de recherches portant sur les aspects techniques de WiMAX, trs peu ont inclus implications conomiques. Nous proposons des modles de tarication pour les applications en temps rel (classes UGS et rtPS) ainsi que pour des applications en temps non-rels (classe BE). Les modles de tarication proposs pour les applications en temps rel dpendent de l'tat du rseau, notamment la congestion. Ces modles font aussi la direnciation entre les prix pays pour les dbits ncessaires pour rpondre tous les besoins en QoS des connexions temps rels et les prix pays pour les dbits supplmentaires. En ce qui concerne les modles de tarication proposs pour les applications en temps non-rels, ils se basent ou bien sur un prix x par l'oprateur ou sur un prix variable qui se dtermine dynamiquement par des enchres. Au passage, nous mettons en vidence que les algorithmes de scheduling doivent tenir en compte du modle de tarication utilis. Nos mcanismes sont valus et compars avec un modle de tarication trs basique qui charge tous les utilisateurs de la mme faon. Les rsultats de simulation montrent l'importance de dnir un modle de tarication adquat pour le systme WiMAX. Les rsultats de simulation montrent aussi que notre modle sophistiqu reste performant mme lorsque l'oprateur n'arrive pas dnir exactement les valeurs optimales des paramtres.
Abstract
WiMAX is a Broadband Wireless Access (BWA) technology. It is based on the IEEE 802.16 standard and its amendment IEEE 802.16e. Scheduling algorithms are very important in WiMAX for ecient use of radio resources. The IEEE 802.16 standard denes ve QoS classes and their requirements as well as the management messages. A scheduling algorithm should take into account the WiMAX QoS classes and service requirements while applying the WiMAX/IEEE 802.16 scheduling framework. The standard does not specify which scheduling algorithm(s) should be used. In this thesis, we present a state-of-the-art of scheduling techniques for WiMAX. We analyze the proposed use of some generic algorithms for WiMAX and then some scheduling algorithms specically proposed for WiMAX. We also draw a comparison between the dierent possible scheduling methods and highlight the main points of each of them. We then focus on the real-time Polling Service (rtPS) QoS class. We highlight a problem that may exist with the WiMAX rtPS QoS class and we provide solutions for it. As an application of these solutions, we propose an enhancement of the maximum Signal-to-Interference Ratio (mSIR) scheduler, called modied maximum Signal-to-Interference Ratio (mmSIR). We use the network simulations in order to test the performance of the proposed scheduler. We propose and implement a new WiMAX module for Network Simulator (NS-2) simulator. This module is based on the National Institute of Standards and Technology (NIST) implementation of WiMAX. Our contribution consists of the addition of the QoS classes as well as the management of the QoS requirements, unicast and contention request opportunities mechanisms, and scheduling algorithms for several QoS classes. We show through extensive simulations that this enhancement provides better mean sojourn time in addition to an improvement in throughput. We determine the spectrum eciency of the WiMAX system determined through simulations for dierent scheduling algorithms. This allows us to estimate the WiMAX capacity for dierent schedulers. We also present some capacity estimation results for dierent radio channel conditions. While dierent technical aspects of this technology have been addressed, the pricing of WiMAX technology has received little attention until now. In this thesis, we propose a WiMAX pricing model suitable for both real-time applications as well as non-real-time applications. The pricing proposal for real-time applications aims at coping with congestion while maintaining a sucient QoS and/or yielding sucient revenue to the operator. The pricing proposal for nonreal-time applications has two variants : an OFDM symbol price and an auction-based variable OFDM symbol price. Our mechanisms are evaluated under a specic model of user preferences and WiMAX situations and compared with a basic pricing model. The comparison of the different results shows the importance of a well-planned pricing framework for WiMAX. Moreover, our pricing scheme outperforms the basic scheme over a large range of input parameters. It then increases the revenue even if the operator cannot nely tune the algorithm.