This action might not be possible to undo. Are you sure you want to continue?

by Shihyu Chang

A dissertation submitted in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy (Electrical Engineering: Systems) in The University of Michigan 2006

Doctoral Committee: Professor Wayne E. Stark, Co-Chair Associate Professor Achilleas Anastasopoulos, Co-Chair Professor Arthur G. Wasserman Assistant Professor Mingyan Liu

c

Shihyu Chang 2006 All Rights Reserved

ACKNOWLEDGEMENTS

I am most grateful to my Ph.D. advisors Professor Wayne E. Stark and Professor Achilleas Anastasopoulos, for their invaluable guidance and continued support during the course of this research work at the University of Michigan. I am also extremely grateful to Professor Stark for his patient rectiﬁcation of my English pronunciation and academic writing skills. I would like to thank the other members of my committee, Professor Mingyan Liu and Professor Arthur G. Wasserman, for their time and eﬀort in reading this work and providing worthwhile comments and suggestions. I would also like to present my gratitude to my colleagues, Kar-Peo Yar, ChihWei Wang, Jinho Kim and Changhun Bae, at the Wireless Communication Lab at the University of Michigan. I have enjoyed supportive and resourceful discussion with them during our group meetings. In addition, I would like to acknowledge my ﬁnancial support from the Oﬃce of Naval Research under Grant N00014-03-1-0232. Finally, my sincere thanks goes to my parents and siblings. Without their support and love, it is impossible for me to ﬁnish my Ph.D study at the University of Michigan. They have always been there to encourage me when I needed them.

ii

. . . . . . . 802. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Numerical Results . . . . . . . . . . . . .2. .1 PHY and MAC Cross-layer Analysis . . . . . . . 1 2 2 3 6 9 10 12 14 15 19 19 22 24 25 29 34 37 39 46 51 53 53 56 58 65 71 74 75 1. . . . . . .3 1.3. . . . . . . . . . . .TABLE OF CONTENTS ACKNOWLEDGEMENTS . . . . . . . . . . .1 Wireless Network Architectures and Protocol Layers . .3 Other Research . . . . . . . . . . . . 3. . . . . . . . . . . . . . . . .4 Average Energy with Delay Constraint . .2 Protocol Layers . . . ii v viii ix LIST OF TABLES . . . . . . . . . . . . Conclusion . . . . . Systems . . . . . . . . .6 III. . . .3. . . . . . . .1 Adaptive Energy Scheme for Wireless Network 3. . .3. . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . .2 Analysis . . . . . . . . . . . . . . . . . . .1 2. . . . 1. . . System Description . . . . . . .3 Mean System Energy Consumption and Delay . . . .4 II. . . . . . 2. . . . . . . . . . . . . . . . . . . 1. . . . . .1. . . . . . . . . . . 2. . . . . . . . . . . . . . . Literature Review for IEEE 802. . . . . . . . . . . . . . . .11 MAC Protocol . . . . . . . .2 1. . . 3. . . 1. . . . . . . . . . . . .3 Introduction .11 Protocol with ARQ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy-Delay Analysis of MAC Protocols in Wireless Networks . . . . . . . . . Numerical Results . .2 Joint Generating Function of Energy and Delay 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . Variations of 802. Energy-Delay Analysis . . . . . . . . . . . . . . .11 DCF Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . 3. . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LIST OF FIGURES . . . . . . . . .2 System Delay and Energy Analysis . . . . . .2. . . . . . . . . . . . .1. . . . . . . . . . . Motivation . . . . . . . . .1. . . . . 1. . . . . . . . . . . 2. . . 1. . . . . . . . . .3. 3. . . . .5 2. . 3. . . . . . . . . . . . . . . . . . . . . . . LIST OF APPENDICES . . . . . . . . . . .2 iii . . . . . . . . Energy-Delay Optimization . .4 2. . .2 2. Thesis Contributions and Outline . . . . . Introduction . . . . .1 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . .1 System Description . . . . . 2. .1 Three Nonlinear System Equations . . . . . . . .1 Wireless Network Architectures . . . CHAPTER I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Priority and Scheduling Analysis .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. . . . . . . . . . . .1. . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . Go-Back-N ARQ . . . . . . . . . . . .6 4. . . .2 Contributions . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . 4. . . . . Some Practical ARQ Systems . . . . . . . . . . . . . . . . . . . . . . . FSM Model. . . . . . . .6. . . . . . . .3 4. . . .4. . . . . . . . . . . . . . . . . . . .6. . . . . . . 4. . . .3 Existence of Codes for Memory Receiver . . . . . . . . . . . . .2 4. . iv . . . . . . . . . . . . . . 85 85 88 93 106 106 112 117 121 122 124 128 130 134 141 4. . . . . . . . . . . . . . . . . . . . .1 SW-ARQ over Time-Varying Channels: No Repetition . . . . . . . . . . . . Numerical Results . . . . . . . . . 80 82 IV. . .6.6. . . . .2. . . Analysis of Energy and Delay for ARQ Systems over Time Varying Channels . 4. . 142 5. . . . . . . .Part II : Imperfect Channel Knowledge . . . . . . . . . Generating Function Analysis for General ARQ Systems . . . . . . . . . . . . 4. . . . . .7 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 SW-ARQ over Time-Varying Channels: Repetition . . . . . .8 V. . . . . . . . . . . . . .2 Random Coding Bound for Memory Receiver . . . . Channel Model and Assumptions for ARQ Systems . . . . . . . . 4. . . . . . . . . . . . . . . . Cutoﬀ Rate for Memory Receiver over Memoryless Channels . . . . . . . . . . . . . . . . . . . 144 147 163 APPENDICES . . . . . . . . . . . . . . . . . . . . . .3 3. . . . . . . . . . . . . . . 142 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Cutoﬀ Rate Optimization . . . Conclusion and Future Research . . . . . . . . . . . .3 Numerical Results . . . . . . . . . . .1 5. . . . . . . . . . . .4. . BIBLIOGRAPHY . . . . .1 4. . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . .5 4. . . . . . . . . . . . . . 4. . . . . . . . . . . . . . Conclusion . . . . .4 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . .1 Random Coding Bound for Memory Receiver in AWGN channelPart I: Perfect Channel Knowledge . . . . . . . . .

2 1. and i = 1. . . . Energy-delay curves for various outage delay probabilities and diﬀerent values for the outage probability P r(Td > γd ) (m = 1. . . . . . . The hidden stations problem. Lines with a square symbol represent the case of KDT = 4400 and lines with a star symbol represent the case of KDT = 2400.4 30 47 48 2. . . . . . . . . .11 52 v . . average delay curves for diﬀerent n and KDT . . . . . . . . . . . . . . . . . . .7 49 2. W = 8. . . st . . . . KDT = 640). . . . . . . . . . .8 49 2. . . . . . Illustration of the random processes. . . . . . . . . Energy delay curves for diﬀerent number of users and packet sizes. . . . .2 26 28 2. . . . . .LIST OF FIGURES Figure 1. . . . .4 2. . . . . The numbers j0 and j1 represent random numbers at stage i = 0. . . . . . Transform variables X and Y are omitted for simplicity. . . . .3 1. . . . . .5 2. . Energy-delay curves for diﬀerent number of users n with KDT = 6400. The illustration of fundamental tradeoﬀ between energy and delay. . 3 4 7 8 An ad hoc network. . . . . . . . The average energy and delay curve (circle symbol) is also shown for comparison. . . . . Station A and B are hidden station of each other. . . . . Energy-delay curves for various outage energy probabilities and diﬀerent values for the outage probability P r(Et > γe ) (m = 1. . n = 10. . . . s(τ ) and their discrete-time counterparts bt . . . . . . . . . . . . The lines with squares represent numerical results and the lines with circles represent simulation results.9 50 2. n = 10. . . . . . Energy-delay curves for diﬀerent KDT with n = 10. . . . . . . .6 2. Timing diagram for the protocol under investigation. . . . . . b(τ ). . . . . . . . . . . . . . . . . . . . The average energy and delay curve (circle symbol) is also shown for comparison. . . . . . . . . Normalized standard deviation of delay vs.3 2. . . . . . . . . . . . . . . . . . . . . Each station communicates mutually without the help of AP. . . . . . respectively. Energy-delay tradeoﬀ curves evaluated from the approximation method (star symbol) and the exact energy-delay tradeoﬀ curves (square symbol) with n = 10 . . . . . . . . . .11 MAC protocol. 25 2. . . . . W = 8. . . . . . . . . State diagram representation of the 802. . . .1 1. . .10 51 2. . Markov chain for backoﬀ counter and contention window stage. KDT = 640). . . . . .1 An infrastructure network. . .

3.1

Timing diagram for protocol. The j0 and j1 are backoﬀ random numbers at CW stage i = 0, and i = 1, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . State diagram for the protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The ST d comparison of convolutional codes for diﬀerent KDT with code rate 1 and 2 n = 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The ST d comparison of Reed-Solomn codes for diﬀerent KDT with code rate 1 and 2 n = 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The ST d comparison of convolutional and Reed-Solomn codes for diﬀerent code rate with KDT = 128 and n = 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . The ST d comparison of Reed-Solomn codes for diﬀerent code rate with KDT = 128 and n = 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of energy and delay curves for diﬀerent ∆ with n = 10 and KDT = 64. Comparison of energy and delay curves for diﬀerent n and KDT with ∆ = 0.3(dB). State diagram for the protocol. Solid lines represent successful reservation or transmission, while dotted lines represent unsuccessful reservation or transmission. . . . Energy-delay curves for n = 10 and KDT = 1400. Solid lines represent the curves after packets optimization. Dashed lines represent the energy-delay curves for Ec /N0 of 0 and 3 dB using the optimal packet lengths. . . . . . . . . . . . . . . . . . . . . The dashed lines represent SWARQ after optimization. The number beside the curve is the redundant bits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Gilbert-Elliott channel model for time varying channel. . . . . . . . . . . . . . . .

57 62

3.2 3.3

67

3.4

68

3.5

69

3.6

70 71 72

3.7 3.8 3.9

77

3.10

83

3.11

84 91

4.1 4.2

An example for a general ARQ system with mR = 2. The ﬁrst slot represents the state of the channel and transmitter. The second and third slots represent the state of the receiver memory. The second (third) slot is used to represent the ﬁrst (second) position of the the receiver memory. . . . . . . . . . . . . . . . . . . . . . Stop-and-Wait ARQ protocol with increment redundancy.

95

4.3 4.4 4.5 4.6 4.7 4.8 4.9

. . . . . . . . . . . . . 107

FSM representation for SW-ARQ protocol with increment redundancy. . . . . . . . 108 Operation of sliding window. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 FSM representation for Stop-and-Wait ARQ protocol with m = 2 and mR = 1. . . 110 Three diﬀerent repetition ARQ protocols with m = 2 considered in this section. . 113

State diagram for R-SW-ARQ-ML. . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 FSM representation for repetition SW-ARQ systems with sliding window memory with m = 2 and mR = 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

vi

4.10

FSM representation for repetition SW-ARQ systems with non-overlapping window memory with m = 2 and mR = 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Cutoﬀ rate comparison for two diﬀerent receiver structures. . . . . . . . . . . . . . 129 The comparison of cutoﬀ rate for uniform and optimal priori probability for input signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 The comparison of cutoﬀ rate for diﬀerent average energy constraints for 8 QAM . 135 Energy-Delay curves for diﬀerent KDT . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Energy-Delay curves for diﬀerent λ with ﬁxed average SNR. . . . . . . . . . . . . . 138 Normalized average delay relations with λ. . . . . . . . . . . . . . . . . . . . . . . 139

4.11 4.12

4.13 4.14 4.15 4.16 4.17 4.18 4.19

Energy-Delay curves for diﬀerent average channel SNRs. . . . . . . . . . . . . . . . 140 Energy-Delay curves for repetition ARQ. . . . . . . . . . . . . . . . . . . . . . . . 140

Energy-Delay curves for GBNARQ with diﬀerent Nrt . . . . . . . . . . . . . . . . . 141

vii

LIST OF TABLES

Table 1.1 2.1 3.1 3.2 4.1 4.2 4.3 4.4 4.5 The OSI network’s seven-layer model. . . . . . . . . . . . . . . . . . . . . . . . . . System Parameters for Numerical Results . . . . . . . . . . . . . . . . . . . . . . . System Parameters for Numerical Results . . . . . . . . . . . . . . . . . . . . . . . System Parameters for Numerical Results . . . . . . . . . . . . . . . . . . . . . . . 4 47 66 81

Output functions and state transition probabilities for SWARQ . . . . . . . . . . . 108 Output functions and state transition probabilities for SWMARQ . . . . . . . . . . 111 Table for transition probabilities and output functions of R-SW-ARQ-ML . . . . . 115 Table for transition probabilities and output functions of R-SW-ARQ-SWM . . . . 116 Table for transition probabilities and output functions of R-SW-ARQ-NOM . . . . 119

viii

. 159 ix . . . . . . . . . . . . . . . . . . . . . . . . . .3 Memory Receiver : Non Overlap . . . . . . Generating Function Analysis for NR-ARQ Systems . . . . . . . . . . . . . . . . . . . . . Generating Function Analysis for R-ARQ Systems .2 Memory Receiver : Sliding Window . . . . . . . . . . 148 A. . . . . . . . . .1 Memoryless Receiver . . . . . . . . . . 153 B. 150 B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 B. . . . . . . . . . . .LIST OF APPENDICES Appendix A. . . . . . . . . . . .1 Memoryless Receiver . 153 B. . . . 148 A. . . . . . . . . . . . . . . . . . . . . . . .2 Memory Receiver . .

vertical handoﬀ algorithms need to be 1 .CHAPTER I Introduction Future wireless systems may change the way people communicate. • Because most devices are battery powered and the energy of a battery is a limiting resource. For example. globalization of business will be realized by trading among companies located at diﬀerent countries via internet. distance eduction enables students and teachers to communicate through information technology without the necessity for the students and teachers to be physically in same location and time. shop and work signiﬁcantly by establishing ubiquitous communication among people and devices. energy eﬃcient network techniques are crucial to design of a network system to meet some performance criteria. • In order to have seamless communication between all existing diﬀerent wired and wireless communication protocols. Some of them are: • Channel and network characteristics are random and time varying. mobile access to work environment will allow people to work anywhere in the world. These goals will be fulﬁlled by interconnecting any devices at anywhere and anytime through wired and wireless communication. A lot of challenges have not been solved in implementing the future wireless systems.

Therefore. For instance.1.2 developed. the requirements for optimizing performance are often contradictory and there is a fundamental tradeoﬀ between energy and delay. On the other hand. 1. we will concentrate on discussing the issue of energy eﬃciency. if less energy is used. To combat severe channel conditions of wireless networks compared to wired networks.1 Wireless Network Architectures and Protocol Layers We will describe brieﬂy network architectures and the wireless protocol stack. In an infrastructure network. • Advanced coding schemes and multiple antenna systems are indispensable and an intelligent controller for radio resources is still missing. In the rest of this thesis. 1. A wireless network in which stations communicate with each other by ﬁrst going through an access point (AP) is called an infrastructure network. the loss of data packets could increase the delay to transmit a packet successfully. In this thesis. we will determine the tradeoﬀ between energy and delay analytically. These two factors are crucial in designing an energy eﬃcient wireless network. decreasing the transmission delay by increasing the transmission rate often results in less energy eﬃciency. wireless stations can communicate with each other or can communicate through a wired network with other stations not in radio range.1 Wireless Network Architectures There are two diﬀerent network architectures: infrastructure and ad hoc networks. we need more energy in packet transmission to decrease packet error probability. A set of wireless stations which are connected to an AP is referred to as a basic service set (BSS). Most corporate wireless local area networks (LANs) operate in infrastructure mode .

3

because they require access to the wired LAN in order to use services such as ﬁle servers or printers. Fig. 1.1 shows an infrastructure network. The big circle around each AP represents the communication range of each AP.

AP Wired Backbone

: station AP

Figure 1.1: An infrastructure network.

A wireless network in which stations communicate directly with each other, without the use of an AP is called a network with ad hoc mode. An ad hoc network architecture is also referred to as peer-to-peer architecture or an independent basic service set (IBSS). Ad hoc networks are useful in cases that temporary network connectivity is required, and are often used for battleﬁelds or disaster scenes. Fig. 1.2 shows an ad hoc network.

1.1.2 Protocol Layers

The concepts of protocol layering provides a basis for knowing how a complicated set of protocols cooperate together with the hardware to provide a complete wireless network system. Although new protocol stacks such as the infrared data association (IRDa) protocol stack for point-to-point wireless infrared communication and

4

: station

Figure 1.2: An ad hoc network. Each station communicates mutually without the help of AP.

Table 1.1: The OSI network’s seven-layer model. Application and Services layer Presentation layer Session layer Transport layer Network layer Date link layer A. Logical control sublayer B. Media access control sublayer Physical layer

the wireless application protocol (WAP) Forum protocol stack for building more advanced services have been proposed for wireless networks [2, 1], we will concentrate our discuss on the traditional OSI protocol stack. There are seven layers regulated by OSI, shown in Table 1.1. Physical Layer: The physical layer (PHY) is made up by radio frequency (RF) circuits, modulation, and channel coding systems. The major functions and services performed by the physical layer are : • establishment and termination of a connection to a communications medium;

5

• conversion between the representation of digital data in user equipment and the corresponding signals transmitted over a communications channel. Data Link Layer: The function of data link layer is to establish a reliable logical link over the unreliable wireless channel. The data link layer is responsible for security, link error control, transferring network layer packets into frames and packet retransmission. The media access control (MAC), a sublayer of data link layer is responsible for the task of sharing the wireless channel used by stations in the network. Network Layer: The task of network layer is to rout packets, establish the network service type (connectionless vs. connection-oriented), and transfer packets between the transport and link layers. In the scenario of mobility of stations, this layer is also responsible for mobility management. Transport Layer: The transport layer provides transparent transfer of data between hosts. It is usually responsible for end-to-end error recovery and ﬂow control, and ensuring complete data transfer. In the Internet protocol suite this function is achieved by the connection oriented transmission control protocol (TCP). The purpose of the transport layer is to provide transparent transfer of data between end users, thus relieving the upper layers from any concern with providing reliable and cost-eﬀective data transfer. Session Layer: The session layer sets up, coordinates, and terminates conversations, exchanges, and dialogs between the applications at each end. It deals with session and connection coordination. Presentation Layer:

and context adaption are implemented in this layer. digital signal processing (DSP). a system with lower code rate needs longer delay to transmit the same information packet due to the ﬁxed bandwidth. we are going to determine the tradeoﬀ between energy and delay in wireless networks taking into account the data link and physical layers. the system requires higher energy per information bit to achieve the same reliability as a system with lower code rate. Another example is shown in Fig. multipath fading.. 1.2 Motivation Time varying channels (e. Services provided at this layer depend on the various users requirements. We plot a channel SNR realization with respect to the system .3.g. In the rest of section. an eﬃcient system requires a joint design across all these layers. shadowing) are often encountered in wireless and mobile systems. Given the energy used per information bit. Layered architectures have been used in most data networks such as Internet.6 The presentation layer converts incoming and outgoing data from one presentation format to another. The horizontal axis represents the system time and the vertical axis indicates the signal-to-noise ratio (SNR) of the channel. 1. However. a wireless communication system adopting forward-error-correction (FEC) can provide higher protection with lower code rate. In this thesis. 1.1. we focus on the lowest two layers of OSI seven layer architecture mentioned in Sec. If we want to decrease the delay to transmit the same information packet. Since all layers of the protocol stack aﬀect the energy consumption and delay for the end-to-end transmission of each bit. Application and Services Layer: Source coding.2.

it is likely that the station at the far edge of the circle can access the AP since the distance between the station and AP is less than the communication . Take a physical star topology with an AP with many stations surrounding it in a circular fashion. the transmitter waits until the channel SNR becomes high before transmissions. Hidden stations in a wireless network refer to stations which are out of the communication range of other stations. 1. In Fig. CH. For example. not each station can communicate with each other. the transmitter needs to spend more energy in transmission in order to keep the same reliability as in the previous case. SNR Large delay TX waits TX transmits (a) CH.7 time in Fig 1. 1.3(b). Therefore. the system will spend large delay to ﬁnish transmission. however. In Fig. the transmitter transmits immediately even when the channel SNR is low. However.3(a). each station is within communication range of the AP.3.3: The illustration of fundamental tradeoﬀ between energy and delay. SNR Large energy Time TX transmits (b) Time Figure 1. there is a fundamental tradeoﬀ between energy and delay of a wireless communication system. In this case.

Hidden stations problem leads to diﬃculties in media access control. The source station detects an unsuccessful channel reservation by the lack of a correct CTS packet. which occurs when some stations in the network are unable to detect each other. the channel reservation is successful and the source station will begin to send data and wait for the acknowledgement (ACK) packet.8 range. the 802. say r.11 protocol uses a mechanism known as request-to-send/clear-to-send (RTS/CTS). A AP B Figure 1. In order to solve the problem of hidden stations [33]. If the RTS packet is received correctly at the destination station (there is no collision with other RTS packets sent by other competing stations and the receiver correctly decodes the packet) the destination station broadcasts a clear-to-send (CTS) packet. the source station sends a request-to-send (RTS) packet. The MAC protocol investigated in . 1. Station A and B are hidden station of each other.4: The hidden stations problem. Before transmitting the data packet. These two stations are known as hidden stations with respect to each other. If the CTS packet is successfully received by the source station. Fig.4 illustrates the hidden stations problem of a wireless network. but it is unlikely that the same station can detect a station on the opposite end of the circle because the distance between these two stations exceeds the communication range r.

9

this thesis is RTS/CTS protocol adopted in IEEE 802.11 standard. It is a well known fact that a packet with more redundant bits have smaller error probability. This means that longer packets can be transmitted successfully with lower energy demand. However, longer packets require larger delay to transmit. In order to determine the tradeoﬀ between energy and delay of a wireless system, we need to have a relation between error probability, energy and delay for each type of packet in the system. In this thesis, we will use the reliability function bounds for a channel to determine the packet error probability. Let K be the number of information bits in a packet and N be the number of coded symbols for this packet. Then there exists an encoder and decoder for which the packet error probability Pe,K,N is bounded by Pe,K,N ≤ 2K−N R0 , (1.1)

where R0 is the cutoﬀ rate determined by channel characteristics. For an additive white Gaussian noise (AWGN) channel using binary input the cutoﬀ rate is R0 = 1 − log2 (1 + e−Ec /N0 ), where Ec /N0 is the received signal-to-noise ratio per coded symbol. 1.3 Literature Review for IEEE 802.11 DCF Analysis

We will begin with a brief introduction of IEEE 802.11 protocol followed by reviewing literatures about its analysis. The IEEE 802.11 protocol for wireless LANs is a multiple access technique based on carrier sense multiple access/collision avoidance (CSMA/CA). The basic operation of this protocol is described as follows. A station with a packet ready to transmit listens the activity of transmission channel. If the channel is sensed idle, the station captures the channel and transmits data packets. Otherwise, the station defers transmission and keeps in the backoﬀ state. There are

10

two basic techniques to access the medium. The ﬁrst one is called distributed coordination function (DCF). There is no centralized coordinator in the system to assign the medium to the users in the network. This is a random access scheme, based on the CSMA/CA protocol. Whenever the packets collide or have errors, the protocol adopts a random exponential backoﬀ scheme before retransmitting the packets. Our work is based on using DCF to access the medium. Another medium access method is called point coordinator function (PCF) to implement medium access control. There is a centralized coordinator to provide collision free and time bounded services. Because the analysis of network performance in this thesis is based on Bianchi’s distributed coordination function (DCF) analysis, we will give a literature review of papers which use Bianchi’s DCF analysis in this section for comparison. From these papers, we classify them into three categories. The ﬁrst category is to use Bianchi’s model to perform the cross-layer analysis (PHY and MAC). The second category is to use Bianchi’s model to perform the priority and scheduling analysis. We put other applications in the third category due to their variety. Some use Bianchi’s model to improve channel utilization and some modiﬁes Bianchi’s model to consider non-saturation traﬃc scenario. We will begin with the ﬁrst category.

1.3.1 PHY and MAC Cross-layer Analysis

In [24], a theoretical cross-layer saturation goodput model for the IEEE 802.11a PHY layer and DCF basic access scheme MAC protocol is developed. The proposed analytical approach relates the system performance to channel load, contention window (CW) resolution algorithms, distinct modulation schemes (BPSK and QPSK), FEC schemes (convolutional code), receiver structures (maximum ratio combining) and channel models (uncorrelated Nakagami-m fading channel). In [32], the authors study the impact of frequency-nonselective slowly time-variant Rician fading chan-

11

nels on the performance of single-hop ad hoc networks using the IEEE 802.11 DCF in saturation. They study the throughput performance of the four-way handshake mechanism under direct sequence spread spectrum (DSSS) with diﬀerential binary phase shift keying (DBPSK) modulation. In an ad hoc network, it is important that all stations are synchronized to a common clock. Synchronization is necessary for frequency hopping spread spectrum (FHSS) to ensure that all stations hop at the same time; it is also necessary for both FHSS and direct sequence spread spectrum (DSSS) to perform power management. In [25],the authors evaluates the synchronization mechanism, which is a distributed algorithm, speciﬁed in the IEEE 802.11 standards. By both analysis and simulation, it is shown that when the number of stations in an IBSS is not very small, there is a non-negligible probability that stations may get out of synchronization. The more stations, the higher probability of asynchronism. Thus, the current IEEE 802.11s synchronization mechanism does not scale; it cannot support a large-scale ad hoc network. To alleviate the asynchronism problem, this paper also proposes a simple modiﬁcation to the current synchronization algorithm. The modiﬁed algorithm is shown to work well for large ad hoc networks. In [46], a cross-layer analytical approach from both the PHY layer and the MAC layer to evaluate the performance of the IEEE 802.11 is developed. From the PHY layer, this analytical approach incorporates the eﬀects of both capture and directional antenna, while from the MAC layer, the model takes account of the CSMA/CA protocol. Through a cross-layer modelling technique, this analytical framework can provide valuable insights of the PHY layer impact on the throughput performance of the CSMA/CA MAC protocol. These insights can be helpful in developing a MAC protocol to fully take advantage of directional antennas for enhancing the

they study the impact of the probability of overhearing another packets priority index.11 priority back-oﬀ schemes to approximate the idealized schedule. 29]. By monitoring transmitted packets. They then incorporate this scheduling table into existing IEEE 802. RTS/DATA packets in IEEE 802. They solve the model numerically and show that the saturation throughput per station is strongly dependent not only on the stations position but also on the positions of the other stations. From these aspects. this model provides the total saturation throughput of the medium. they provide a scheduling scheme named multi-hop coordination so that downstream stations can increase a packets relative priority to compensate for excessive delays incurred upstream. and the random nature of medium access prohibit an exact realization of the ideal schedule. the proposed model computes its saturation throughput by relating to the positions of the other concurrent stations. Consequently. First.3.2 Priority and Scheduling Analysis In [28. the authors develop two mechanisms for QoS communication in multihop wireless networks. the proposed analytical model is provided for multi-hop coordination and it is used to compare the probability of meeting an end-to-end delay bound over a multi-hop path with and without coordination. Further. In the former mechanism. link errors. In [36]. Second. .12 performance of the WLAN.g. In the latter mechanism. 1. a technique that piggybacks the priority tag of a stations head-of-line packet onto handshake and data packets. they propose distributed priority scheduling. they observe that congestion. each station maintains a scheduling table which is used to assess the station’s priority level relative to other stations.11.. the authors develop an analytical model to quantitatively explore these two mechanisms according to the model of Bianchi. e.

This work extends previous work of Bianchi by adding a further state to model diﬀerent IFS for diﬀerent priority classes. based on the the estimated information. and. This paper proposes an analytical approach to evaluate throughput and delay performance of IFS based priority mechanisms for diﬀerent priority classes.11 enhanced DCF. In [7]. the model does not rely on traditional multi-dimensional Markov chains because the crucial assumption of a constant probability to access the channel in a given time slot is not always correct. the model fails when the diﬀerence between the IFS of two classes is greater than the minimum contention window. a number of service diﬀerentiation mechanisms have been proposed for. However. the authors develop a model-based frame scheduling scheme. in particular. computes the current network utilization. to enhance the capacity of IEEE 802. in general. they develop an analytical model that characterizes data transmission activities in IEEE 802. called MFS. CSMA/CA systems. In order to accurately calculate the current utilization in WLANs. In MFS each station estimates the current network status by keeping track of the number of collisions it encounters between its two consecutive successful frame transmissions. An eﬀective way to provide prioritized service support is to use diﬀerent inter frame spaces (IFS) for stations belonging to diﬀerent priority classes. and.11operated WLANs with/without the RTS/CTS mechanism. . and validate the model with ns-2 simulation.11-operated wireless LANs. The result is then used to determine a scheduling delay that is introduced (with the objective of avoiding collision) before a station attempts for transmission of its pending frame.13 In [31]. the 802. For an example.

The model explicitly takes into account both the carrier sensing mechanism and an additional backoﬀ interval after successful frame transmission. Furthermore. of the HOB (head of burst) immediate ACK. Another interesting conclusion is that the optimal selection between immediate and block ACK does not depend on the number of contending stations.11e.3. delay and frame dropping probabilities of the diﬀerent traﬃc classes in the range from a lightly loaded. Through the analytical modelling. given a data packet size and a retransmission limit. They quantify these comparisons by providing the eﬃciency thresholds needed to select the best possible mechanism.11 DCF. the authors analyze the performance of channel utilization. a model is proposed to predict the throughput. Finally. i.3 Other Research In [43]. while it is very attractive for high data rate transmissions. and some optimizations are possible. immediate/block ACK) on the overall system performance. they show that the block ACK is generally not useful for low data rates and low value for retransmission limit.14 1. which can be ignored under saturation conditions. non-saturated channel to a heavily congested. they show that. In [19]. the authors develop an analytic model for unsaturation throughput evaluation of 802.e. Most of analytical models proposed so far for EEE 802... the model describes diﬀerentiation based on diﬀerent AIFS-values (Arbitration . In [41. They develop an analytical framework to evaluate the impact of diﬀerent access modes (i.e.. they discuss the role of the block ACK protection mechanisms.11 DCF focus on saturation performance.e. supported by the emerging IEEE 802. Expressions are also derived by means of the equilibrium point analysis in [41]. 2-way/4way handshaking) and acknowledgment policies (i. For example. the access mode and the ACK policy have a great impact on the overall system throughput. based on Bianchi’s model. based on data burst transmissions. 30]. saturated medium.

4 Thesis Contributions and Outline The most important contribution of this thesis is to represent the operation of a system by a state diagram and use the generating function approach to derive its energy and delay consumption.15 Inter Frame Space). we will study a single link wireless ARQ communication system. [48] used computer simulations to evaluate the network throughput. . Previous research on performance evaluation of 802. 1. in addition to the other adjustable parameters (i. the authors propose a new contention algorithm called parallel contention algorithm that divides the subcarriers into multiple groups to reduce the contention time. In the ﬁrst part of this thesis. retransmission limits etc.11a standard. Bianchi [5] used a simple but accurate model that characterizes the random exponential backoﬀ protocol. and in the second part of this thesis. the system performance was evaluated by an analytical model. They analyze the proposed scheme by extending the Markov chain model and verify the accuracy of the analysis through the simulations. The goal of this thesis is to investigate the energy and delay tradeoﬀ of a wireless communication system. windowsizes. AIFS diﬀerentiation is described by a simple equation that enables access points to determine at which traﬃc loads starvation of a traﬃc class will occur. which is an important factor in wireless network. [6] and Weinmiller et al.) also encompassed by previous non-saturated models.e. The protocol performs well especially when the transmission speed and the number of users are getting higher. These papers did not incorporate channel noise in the analysis. In [52]. thereby achieving a better performance improvement ratio than the original IEEE 802. 11]. In [23.11 has been carried out by two methods. Crow [18]. we will concentrate on communication networks. 17.

Finally.11 protocol in which the transmitter will increase the energy level per coded symbol whenever it suﬀers an unsuccessful transmission. they did not relate the packet error probability to the energy used and the number of redundant bits used for error control coding. Next. The ﬁrst proposed protocol is an energy adaptation scheme with original IEEE 802. By changing the signal-to-noise ratio. We use the random coding bound to represent the packet error probability as a function of the delay and energy.16 Although Hadzi-Velkov and Spasenovski [22] considered the eﬀects of packet errors in the analysis. we propose an approximation method to express the energy-delay tradeoﬀ curves analytically and show the proposed approximation is extremely accurate especially when the number of information bits per packet is large. In their original work. they did not relate the packet error probability to the energy used and the number of redundant bits used for error control coding. We propose a state diagram representation the operation of the MAC layer and obtain the joint generating function of the energy and delay by incorporating the eﬀects of the PHY layer. Another contribution of this thesis is to apply our proposed generating function method to the analysis and design of other wireless network protocols. The numerical results show that the proposed protocol can improve system performance signiﬁcantly when the channel condition is bad. the energy-delay tradeoﬀ curves for minimum delay are obtained. By using Reed-Solomon codes we can optimize the system performance over . we optimize numerically over the code rate for each type of packet to minimize the average transmission delay. We extend the results from Bianchi [5] and Hadzi-Velkov [22] by considering the eﬀects packet errors in the analysis. The motivation for this thesis is to understand the role energy and codeword length (number of redundant bits) at the PHY layer have on the total energy and delay of the network.

The numerical results show that the IEEE 802. The eﬀect of transition probability which depends on the packet length is also investigated. From the system state diagram. we extend the traditional Markov chain model for the channel state as well as the transmitter state [13] by using a state diagram that takes into account the states of transmitter. The states of the transmitter can be used to model the diﬀerent packet lengths adopted by the transmitter and the states of the receiver can be utilized to model the receiver memory content [27].17 the code rate and energy per coded bit.11 original protocol and the proposed one have almost identical performances and are equally sensitive to the knowledge of the channel quality at the transmitter. The outline of the rest of the thesis is as follows. The motivation of this analysis is to demonstrate that the generating function approach can be applied to analyze more function layers jointly by including the analysis of logical link control (LLC) sublayer into the original protocol (MAC and PHY layers only). As the numerical results demonstrate. we are able to characterize the joint energy and delay distribution of the system incorporating physical layer characteristics (packet error probability as a function of energy and delay) through generating function approach. For wireless ARQ systems. the framework of our analysis can be used for other coding and modulation schemes over various wireless channels. the timevarying characteristics of the channel have a great inﬂuence on system performance especially at low channel SNR.11 protocol. we will use the . The second proposed protocol is an ARQ mechanism for data packets transmission with 802. Finally. receiver and channel jointly. In Chapter 2. Although the packet error probabilities are evaluated with Reed-Solomon codes over an additive white Gaussian noise (AWGN) channel. we also compare the system performance between Reed-Solomon codes and convolutional codes.

18 proposed generating function method to analyze the energy and delay consumption of wireless networks and discuss the tradeoﬀ between energy and delay. Finally in Chapter 5. In Chapter 4. . we brieﬂy summarize the conclusions from the thesis and suggest possible future research directions. The application of generating function method in designing and analyzing other wireless network protocols are presented in Chapter 3. we give the analysis of energy and delay expense for ARQ systems over time varying channels and derive the cutoﬀ rate for diﬀerent memory receiver structure.

Some WLANs must operate solely on battery power. Thus there is some amount of delay in accessing the channel and there is energy used in reserving the channel. Typical examples include the time-division multiple access (TDMA). Generally users make reservations for transmissions in a decentralized way. which mitigates the eﬀect of channel noise at the receiver. The MAC protocol resolves conﬂicts between users attempting to access the channel.1 Introduction Recently there has been considerable interest in the design and performance evaluation of wireless local area networks (WLANs). In such cases it is important to consider energy consumption in the system design and analysis. error control coding reduces the energy needed for transmission at the expense of increased delay. Two critical components of a wireless network are the medium access control (MAC) protocol and the physical layer (PHY). code-division multiple access (CDMA). An important component of the PHY layer is forward error control coding. and contention-based 19 . By transmitting redundant bits in addition to information bits.CHAPTER II Energy-Delay Analysis of MAC Protocols in Wireless Networks 2. There are many MAC protocols that have been developed for wireless voice and data communication networks. It is possible to reduce energy consumption by increasing delay incurred.

20 protocols such as IEEE 802. the source station sends a request-to-send (RTS) packet.11 standard. The DCF is a random access scheme.11 has been based on an analytical model proposed by Bianchi [5]. In order to combat the problem of hidden terminals [33]. is employed when there is no centralized coordinator in the system to assign the medium to users in the network. the 802. There are two basic techniques to access the medium in the 802. In this paper we focus on the DCF protocol for accessing the medium. The ﬁrst one called the distributed coordination function (DCF). 32. based on carrier sense multiple access with collision avoidance (CSMA/CA). 24]. [4]. which occurs when some stations in the network are unable to detect each other. Another MAC method in the 802.11 protocol uses a mechanism known as request-to-send/clear-to-send (RTS/CTS). is used when there is a centralized coordinator to coordinate the access of the medium. If the CTS packet is successfully received by the source station.11 standard called point coordinator function (PCF). we adopt the MAC protocol used in the 802. Before transmitting the data packet. In this paper. Bianchi used a simple but accurate model that characterizes the random exponential backoﬀ protocol. When packets collide or have errors. The source station detects an unsuccessful channel reservation by the lack of a correct CTS packet. Many previous research on performance evaluation of 802. In [22. the transmitter performs a random backoﬀ before retransmitting the packets. the channel reservation is successful and the source station will begin to send data and wait for the acknowledgement (ACK) packet. the .11 standard.11 [3]. 46. If the RTS packet is received correctly at the destination station (there is no collision with other RTS packets sent by other competing stations and the receiver correctly decodes the packet) the destination station broadcasts a clear-to-send (CTS) packet.

they did not relate the packet error probability to the energy used and the number of redundant bits used for error control coding. The authors proposed an eﬀective way to provide prioritized service support by using diﬀerent inter frame spaces (IFS) for stations belonging to diﬀerent priority classes. 7. contention window (CW) resolution algorithms. the 802. receiver structures (maximum ratio combining) and channel models (uncorrelated Nakagami-m fading channel). the authors adopted Bianchi’s model to perform the priority and scheduling analysis. distinct modulation schemes (BPSK and QPSK). For example. The motivation for this paper is to understand the role of energy and codeword length (number of redundant bits) at the PHY layer have on the total energy and delay of the network. The proposed analytical approach relates the system performance to channel load. In [28. The results obtained . and. in [7]. in particular. For example. 29. a theoretical cross-layer saturation goodput model for the IEEE 802. By taking the partial derivative for the joint generating of the energy and delay. Although some of the above papers considered the eﬀects of packet errors in the analysis. This is a universal approach and could be applied to other MAC protocols. 31]. 2.11a PHY and MAC layers was developed in [24]. FEC schemes (convolutional code).11 enhanced DCF. a number of service diﬀerentiation mechanisms have been designed for CSMA/CA systems. we determine the average energy and delay of a successful packet transmission by taking the packet error probability into consideration. We propose a state diagram representing the operation of the MAC layer and obtain the joint generating function of the energy and delay by incorporating the eﬀects of the PHY layer. The contributions in this chapter are as follows: 1.21 authors used Bianchi’s model to perform the cross-layer analysis (PHY and MAC).

We propose an approximation method to express the energy-delay tradeoﬀ curves analytically. The proposed approximation method for energydelay tradeoﬀ curves is given in Section 2. each station with a ﬁxed position can hear (detect and decode) the .6 gives the conclusion and future research. The remainder of this chapter is organized as follows. we give a brief description for the protocol used in our analysis and introduce the system assumptions. Finally. We demonstrate that the energy and delay relationship with random coding under AWGN channel through numerical method in Section 2. Section 2. First. Then the average energy with outage delay constraint and average delay with outage energy constraint are analyzed with generating function.2.3. In Section 2. The comparison of the energy-delay tradeoﬀ curves evaluated from this approximation method with the exact energy-delay tradeoﬀ curves (from numerical optimization) indicates that this approximation method is extremely accurate especially when the number of information bits per packet is large. we discuss our system state diagram and utilize it to derive the joint generating function of energy and delay. In Section 2. By changing the signal-to-noise ratio. the energydelay tradeoﬀ curves for minimum delay are obtained.5. 4. 3. 2.22 from the generating function approach will be agree with the results derived from the renewal cycle method proposed in [22]. We optimize numerically over code rate to have minimum average transmission delay over diﬀerent packet by introducing the random coding bound to represent the packet error probability.4.2 System Description The wireless networks that we analyze here have the following network layer speciﬁcations.

. the transmitting station doubles the CW size (increases the contention stage i by one) and picks another random number j as before. a packet of K information bits is encoded into a packet of N coded symbols. Third.11 MAC protocol (more details can be found in [3] and [4]).11 MAC protocol. . i is the contention stage (initially i = 0). When a station is ready to transmit a packet. Second. . The packet error probability depends on the parameters K. If there is a collision of the RTS packet with other competing stations or a transmission error occurs in the RTS or CTS packet. N . In the following we give a brief description of the most salient features of the IEEE 802.. If the channel is sensed idle. If there are no collisions or errors in the RTS and CTS packets. uniformly distributed in {0. the station 1n − 1 is the number of stations in the communication range of the reference station. After transmitting the RTS packet.23 transmission of n − 11 other stations in the network. stops the count down when the channel is sensed busy. . . 1. and reactivates when the channel is sensed idle again. It is assumed that the receivers have no multiple-access capability (i. each station uses the 802. A backoﬀ time counter begins to count down with an initial value j: it decreases by one for every idle slot of duration σ seconds (also referred to as the standard slot) as long as the channel is sensed idle. they can only receive one packet at a time) and they cannot transmit and receive simultaneously. The station transmits an RTS packet when the counter counts down to zero. At the PHY layer. it senses the channel for DIFS seconds. and Ec /N0 . Wi − 1}. stations always have a packet ready to transmit. where Wi = 2i W is the contention window (CW) size. and W is the minimum CW size.e. the transmission station picks a random number j. the station will wait for a CTS packet from the receiving station. where Ec is the received coded symbol energy and N0 is the one-sided power spectral density level of the thermal noise at the receiver.

In this chapter. . but picks a random number in {0. 2. say CTS. the energy Et is deﬁned as the energy consumed by both transmitting and receiving stations in the duration of Td . we analyze the energy and delay characteristics of the wireless networks described above. However. . 2. . . 1. as shown in Fig. if the data or the ACK packet is not successfully received. It is also noted that there is a maximum CW size (or equivalently. when the transmitter is in this maximum stage and needs to join the contention period again. for notational simplicity we assume that the propagation loss between transmitter and receiver is one (0 dB). Without loss of the generality. say RTS and the beginning of a packet transmission.1. The delay Td of each data packet is deﬁned as the time duration from the moment the backoﬀ procedure is initiated until DIFS seconds after the ACK packet is received correctly by the transmitting station. We also assume that the propagation time is negligible. we only consider the energy consumption for packet transmission and omit the energy required for signal processing and channel sensing. 2. The contention stage is reset (i is set to zero) when the transmitting station receives an ACK correctly. The system parameter SIFS is deﬁned as the time between the end of a packet reception. a maximum contention stage.3 Energy-Delay Analysis In this section. it does not increase further the CW. Similarly.1. the CW size will also be doubled (the contention stage i will increase by one) and the transmitting station will join the contention period again. . A time diagram indicating the sequence of these events is depicted in Fig. Wm − 1}. m). This time includes the time required for decoding a packet and other processing functions at the receiver.24 begins to transmit the data packet and waits for an acknowledgment (ACK) packet.

respectively. . The numbers j0 and j1 represent random numbers at stage i = 0. In order to analyze the energy and delay characteristics it is suﬃcient to only consider the time instances the backoﬀ counter (and CW stage) changes value.1 Three Nonlinear System Equations In order to analyze energy and delay relationships.1: Timing diagram for the protocol under investigation. The second random process s(τ ) is used to represent the CW stage i ∈ {0. A realization of these random processes is shown in Fig. 1. . where τt is the time instance of the t-th change in value of b(τ ). In this realization. and the counter becomes active again. . .3.2.25 j0 ~ U(0. we need to deﬁne two random processes to characterize the backoﬀ counter state and the CW size. and i = 1. To this end we further deﬁne the discrete-time random processes bt = b(τt ) and st = s(τt ). 2. 2. . The ﬁrst process b(τ ) represents the backoﬀ time counter for the reference station (this is the station for which we evaluate energy and delay). m} of the station at time τ .W 0-1) DIFS frozen slot counter frozen slot counter T d begin Trc 0 DIFS j1 ~ U(0.W 1-1) reference station unsuccessful transmission Tcoe stage number is increased by 1 frozen slot counter 0 reference station successful transmission DIFS frozen slot counter T d end stage number is set back to 0 Figure 2. b3 = 4 is the value of the counter just before being frozen and b4 = 3 is the value of the counter after the channel has been sensed idle for DIFS seconds.

bt ) is a discretetime Markov chain. bt+1 = k | st = j.2: Illustration of the random processes. b(τ ). k + 1} = P {0. Wi pc .1a) (2. k | j.26 DIFS frozen slot counter reference station unsuccessful transmission 321 98 b(τ ) 765 4 τ τ s (τ ) = 1 s(τ ) τ 0τ 1τ 2τ 3 τ 4τ 5 τ 6τ 7 τ8τ9 s (τ ) = 0 τ Figure 2. st . Wm 0 ≤ k ≤ Wi − 2. k | i − 1. (2. With the assumption that errors in transmission can occur only due to collisions. W0 pc . k | i.1b) (2. 0 ≤ i ≤ m 0 ≤ k ≤ W0 − 1. The ﬁrst assumption made in this analysis is that the event of packet collision is independent of past collisions and thus independent of the contention stage.1d) . As a result of the above assumptions. 1−pc . bt = l} with P {i. k | i. 0} = P {m. The second assumption is that the packet collision is identical for all states (values of bt and st ) of a user. s(τ ) and their discrete-time counterparts bt . 1 ≤ i ≤ m 0 ≤ k ≤ Wm − 1. 0} = 1.1c) (2. l} = P {st+1 = i. 0 ≤ i ≤ m 0 ≤ k ≤ Wi − 1. the one-step transition probabilities developed in [5] are given by P {i. these two assumptions are extremely accurate when the number of stations in the network is large (say greater then 10). k | m. As veriﬁed in [5]. the two dimensional process (st . 0} = P {i.

3.27 where pc is the conditional collision probability. and since the events of packet collision and packet error are ..e. 2. This is also the transition probability from one contention stage to the next in the two-dimensional Markov chain. .5). collision in the RTS packet or error in any of the packets. Pe. We assume that the channel is memoryless between packets. Finally. modulation etc (a speciﬁc example will be given in Section 2.1) corresponds to the decrement of the backoﬀ counter at the beginning of each time slot. The second equation accounts for the fact that a new packet following a successful packet transmission starts at contention stage i = 0. .ACK . we can take into account packet errors as shown in Fig. . .DT and Pe.RT S . . 2. DT.e. the contention stage increases and the backoﬀ counter is initialized with a uniformly chosen value in the range {0. These probabilities depend on the particular channel. i. the last case models the fact that the contention stage is not increased in subsequent packet transmissions when the contention window size reaches the maximum. the probability of a collision given a packet transmitted on the channel. and ACK packets are received correctly.CT S . Wi }. We denote the error probability of the four kinds of packets in the system as Pe. W0 − 1}. Pe. Using similar assumptions as in [5] for the packet collision probability pc . 1. . 1. as shown in Fig. coding. CTS.. By modifying the Markov chain model described above. and thus the backoﬀ counter is initially uniformly chosen in the range of {0. As described in the third equation.3. . when an unsuccessful transmission occurs at contention stage i − 1. Let pce denote the probability of the complement of this event. The ﬁrst equation in (3. The other two cases describe the system evolution after an unsuccessful transmission. A successful packet transmission requires that the RTS. . i.

W m -1 pce/W m Figure 2.3: Markov chain for backoﬀ counter and contention window stage.1 0. 1 1 1 m.1 pce/W 1 1 1 1.W 1 -1 pce/W m 1 m.28 (1-p ce) /W 1 1 1 1 0 0.W m -2 1 m.0 0.W 1 2 1 1.0 1 1. 0 m.W 0 2 0. .W 0 1 1.

6) where ∆t . the probabilities pce . From the above three nonlinear equations.2 (n − 1)ptx (1 − ptx )n−2 . j) of each state (i. j) of the Markov chain. ptx and pc can be evaluated numerically. (2.DT )Pe. ∆e are parameters that determine the resolution of our analysis (we choose ∆t = Tb and ∆e = Ec in the remaining part of our analysis).RT S )(1 − Pe.2) Following the derivation in [5]. (2. It is denoted by ptx1 and can be expressed as ptx1 2.CT S )Pe.RT S + (1 − Pe.29 independent. (2. = pc (2. we can evaluate the stationary probability P (i.RT S )(1 − Pe.4) pc = 1 − (1 − ptx )n−1 .4). Y ) = i=0 j=0 P r(Td = i∆t . Et = j∆e )X i Y j . 0) = 2(1 − 2pce ) (1 − 2pce )(W + 1) + pce W (1 − (2pce )m ) (2.CT S + (1 − Pe. Another important probability that will be used in our analysis later is the probability of a transmission of an RTS packet from exactly one of the remaining n − 1 stations given that at least one of the remaining stations is transmitting.5) Joint Generating Function of Energy and Delay Our goal in this section is to obtain the joint generating function of energy and delay for a successful data packet transmission.3) (2. Let ptx be the probability of a transmitting station sending an RTS packet during each backoﬀ slots. The transmission probability ptx and the collision probability pc can be related as in [5] m ptx i=0 P (i. the probability of pce can be expressed as pce = pc + (1 − pc )[Pe.CT S )(1 − Pe. We further deﬁne the . which can be expressed as ∞ ∞ Gs (X.RT S )Pe.2)–(2.DT +(1 − Pe.3.ACK ].

NDIF S = TDIF S /∆t .2. channel reservation ? G rfG bs.m G rfG bs.4: State diagram representation of the 802.m successful. NDIF S and Nσ are integers. Based on the protocol description in Section 2. Wait DIFS G bs. For .2 G rs successful DT transmission ? G tf G bs.4 can be obtained.1 G rs successful DT transmission ? G tf G bs.m G rs successful DT transmission ? G tf G bs.0 G rfG bs. and Nσ = σ/∆t .2 G rfG bs.30 quantities NSIF S = TSIF S /∆t .m G ts G ts G ts Wait DIFS Figure 2. channel reservation ? successful.1 successful. and make the additional assumption that NSIF S . 2. Transform variables X and Y are omitted for simplicity. the state ﬂow diagram shown in Fig.11 MAC protocol. channel reservation ? G tf G bs. Each transition from state A to state B in this diagram is labelled with the conditional joint generating function of the additional energy and delay incurred when the protocol makes a transition from state A to state B.

or .9) the meaning of which is that a failure can be due to either a collision/RTS transmission error.ACK )X NDT +NACK +NSIF S +NDIF S Y NDT +NACK .RT S )(1 − Pe.CT S )X NRT S +NCT S +2NSIF S Y NRT S +NCT S .7) where (1 − pc )(1 − Pe. the generating function Grs corresponding to a channel reservation success is Grs (X. and [NRT S + NCT S ] × Ec is the corresponding energy consumed. the generating function Gts corresponding to a data packet transmission success once the channel is reserved can be expressed as Gts (X. (2. Y ) = (1 − Pe. Similarly.RT S )(1 − Pe. a data packet transmission failure is either due to a data packet transmission error. We ﬁrst describe the factors that are independent of the CW stage.RT S ]X NRT S +NDIF S Y NRT S + (1 − pc )(1 − Pe.DT )(1 − Pe. or a CTS transmission error. Similarly. Each product contains a factor that is independent of the particular CW stage. corresponding to a channel reservation failure is given by Grf (X.RT S )Pe. Y ) = [pc + (1 − pc )Pe.8) The generating functions associated with failure to either reserve a channel or to transmit a data packet are products of two generating functions. The generating function Grf . and a factor that depends on the contention stage i.CT S · X NRT S +NCT S +NSIF S +NDIF S Y NRT S +NCT S . (2. (2. Y ) = (1 − pc )(1 − Pe. [NRT S + NCT S + NSIF S + TDIF S ] × Tb is the delay incurred during this transition. after reserving the channel.CT S ) is the probability of not having a collision and having a correct reception of the RTS and CTS packets.31 instance.

11) where Goc (X) is the generating function of the delay due to an occupied slot. The probability of a busy slot due to the transmission of other stations is pc and this event is independent and identical for each slot from our previous assumptions. pc ).i (X) = j=1 1 iW 2 j k=0 j [(1 − pc )X Nσ ]j−k (pc Goc (X))k .DT )Pe. which is captured by the generating function Gtf as follows Gtf (X. there are four possible cases. This generating function characterizes the delay for the transmitting station from the instant of starting the backoﬀ procedure to the instant that the backoﬀ counter reaches to zero at stage i. is 2i W Gbs. Gbs. (2. Let j be the backoﬀ slot chosen uniformly from the above range.i of the state diagram.i is not a function of the variable Y .DT X NDT +NDIF S Y NDT + (1 − Pe. The ﬁrst case is when two or more RTS packets from the remaining n − 1 stations collide or have a packet error. At contention stage i.32 an acknowledgement packet transmission error. We deﬁne an occupied slot as a slot when the transmitting station senses the channel is busy due to the transmission of one of the n − 1 remaining stations. In an occupied slot. Hence. Thus. Y ) = Pe. The third case is when the RTS packet is correctly received without collision and there is no error in the CTS packet but there is an error . the range of possible backoﬀ slots is from 1 to 2i W . k (2. The second case is when the RTS packet is correctly received without collision but there is an error in the CTS packet transmission. Gbs.i . the generating function. We do not need to consider energy consumption here since the reference station is not transmitting during this period. The number of the occupied slots in these j slots is binomialy distributed with parameters (j.10) We now evaluate the generating functions denoted by Gbs.ACK · X NDT +NACK +NSIF S +NDIF S Y NDT +NACK .

Therefore. m−1 j Gs (X.CT S X NRT S +NCT S +NSIF S +NDIF S +ptx1 (1 − Pe. i=1 From the joint generating function. the corresponding generating function is Goc (X) = [(1 − ptx1 ) + ptx1 Pe.CT S )(1 − Pe. .DT ) · X NRT S +NCT S +NDT +NACK +NSIF S +NDIF S .RT S ]X NRT S +NDIF S +ptx1 (1 − Pe. ∂Y T d = ∆t (2.14b) Below.12) From the state diagram and Mason’s gain formula [20].RT S )Pe. Grs Gts .DT X NRT S +NCT S +NDT +NSIF S +NDIF S +ptx1 (1 − Pe. (2.33 in the DT packet transmission. Note that the time duration of successful or unsuccessful ACK transmission are the same. ∂X ∂Gs E t = ∆e |X=Y =1 . The last case is when there is a correct reception of RTS. we obtain the following backward recursive equations for the generating function Gs .i Gbs. we will utilize the generating function to derive the average energy with an outage delay constraint.RT S )(1 − Pe.m where i (1 ≤ i ≤ m) is the index of the CW stage.13) [Grf + Grs Gtf ]m 1 − [Grf + Grs Gtf ]Gbs.CT S )Pe. the average energy and delay can be easily evaluated as ∂Gs |X=Y =1 . CTS.RT S )(1 − Pe. Y ) = Gbs.14a) (2.0 1 + j=1 [Grf + Grs Gtf ]j i=1 m Gbs.i + (2. and DT and either correct or erroneous reception of the ACK packet.

The last case is to extend case three with a correct DT packet transmission so there is an ACK packet transmission. The ﬁrst case is that the RTS packet from the remaining n − 1 stations suﬀers collision or packet error. However.RT S )]Ts. 2. The average number of consecutive standard slots ns between two consecutive interrupted slots due to the transmission of the remaining n − 1 stations can be evaluated as ns = ∞ i(1 − pc )i pc = i=0 1 − 1. The average duration of a renewal cycle can be expressed as Trc = ns σ + [(1 − ptx1 ) + ptx1 Pe. the initial value of the backoﬀ counter has mean of (Wi + 1)/2. . 2. moreover there is no error in the CTS packet.DT ]Ts. The average delay analysis is presented ﬁrst.3 Mean System Energy Consumption and Delay In this section. .1. (2.RT S )(1 − Pe.3. . The second case is that the RTS packet from the remaining n − 1 stations is correctly received and collision free but there is an error in the CTS packet transmission.RT S )Pe.34 2. pc (2. The third case is that the RTS packet from the remaining n − 1 stations is correctly received and no collision occurs.ACK .RT S +[ptx1 (1 − Pe. there is an error happened in the DT packet transmission.CT S +[ptx1 (1 − Pe. 1. There are four possible cases in an interrupted slot.RT S )(1 − Pe. .DT )]Ts.16) .CT S )(1 − Pe. m}.CT S )Pe.CT S ]Ts.DT +[ptx1 (1 − Pe. so that the average number of deferred slots (include standard slots and interrupted slots) before a retransmission attempt is (Wi + 1)/2. In each CW stage i ∈ {0.15) We deﬁne a renewal cycle as the period between two consecutive transmission of the n − 1 remaining stations including multiple consecutive standard slots and an interrupted slot as shown in Fig. we analyze average energy and delay of our system.

RT S pce +(1 − pc )(1 − Pe.RT S )(1 − Pe.CT S )(1 − Pe.RT S )(1 − Pe. the average time duration between two consecutive retransmissions of the reference station is (Wi + 1)pc Trc /2.17c) Ts. and TDIF S .RT S )Pe.RT S )Ts.ACK ].DT )Pe. TSIF S are system delay parameters deﬁned by the standard.DT Ts.17b) (2. denoted as Tcoe . there are four possible cases of unsuccessful transmission. But the last case will be modiﬁed as that the reference station receives ACK packet incorrectly. The previous three cases are the same as the ﬁrst three cases of the above interrupted slot. i.35 and Ts.18) The average time before the reference station makes its (i + 1)-th retransmission ..RT S = NRT S Tb + TDIF S Ts.ACK = (NRT S + NCT S + NDT + NACK )Tb + 3TSIF S + TDIF S . (2.17a) (2.CT S Ts. Since the retransmission attempt of a reference station in contention stage i is on the average preceded by (Wi + 1)/[2(ns + 1)] = (Wi + 1)pc /2 renewal cycles of n − 1 remaining stations.ACK Ts. it is not a successful transmission. (2.CT S )Pe.e. If the reference station fails to receive the ACK packet correctly.DT +(1 − pc )(1 − Pe.CT S +(1 − pc )(1 − Pe.CT S = (NRT S + NCT S )Tb + TSIF S + TDIF S Ts. The average duration while the reference station itself occupies the channel during each unsuccessful retransmission attempt.DT = (NRT S + NCT S + NDT )Tb + 2TSIF S + TDIF S (2.17d) where Tb is the time duration to transmit each coded bit. is Tcoe = 1 [(pc + (1 − pc )Pe.

RT S )(1 − Pe.CT S Es. denoted as Tr.19b) for m ≤ i.i = k=0 = iTcoe + (i + 1) pc Trc + W (2i+1 − 1) pc Trc 2 2 iTcoe + (i + 1) pc Trc + W (2m+1 − 1 + 2m (i − m)) pc Trc 2 2 (Wk + 1) pc Trc + iTcoe 2 (2.DT +(1 − pc )(1 − Pe. Finally.20a) = Ts. It is expressed with Ts.21) .ACK + Tcoe pce 1 − pce pc W 1 − 2(2pce )m (1 − pce ) + pm (1 − 2pce ) ce +Trc + 2 1 − 2pce pc W m+1 2m pce m pc (2 −1+ )pce + .i ce (2.CT S )(1 − Pe.ACK + i=0 (1 − pce )pi Tr.CT S +(1 − pc )(1 − Pe.14) by taking the partial derivative of Gs (X.36 attempt. Y ) with respect to X. The four possibilities of consuming energy when the reference station fails to receive the ACK packet correctly are the same in the previous delay analysis.RT S )(1 − Pe. (2. Tcoe and Trc as ∞ T d = Ts.RT S )Es.ACK . 2 1 − pce 2(1 − pce ) (2. is i Tr.20b) The average transmission energy Et of each data packet is the total energy consumed in the transmitting and receiving stations from the moment the backoﬀ procedure is initiated until the moment the ACK packet is received by the transmitting station correctly. The average energy consumption can be expressed as Ecoe = 1 [(pc + (1 − pc )Pe.RT S )Pe.ACK Es.RT S pce +(1 − pc )(1 − Pe.DT Es.CT S )Pe.ACK ].i . We will consider only the energy spent in packet transmission and assume that the energy consumption for each packet transmission is N Ec for a packet of length N .DT )Pe. the average delay of a successful data packet transmission can be obtained from (2.19a) for 0 ≤ i ≤ m − 1 (2.

ACK and Ecoe as ∞ E t = Es. denoted X + . · · · ) by ∞ X (x) k=1 + nk xk .37 where Es.22b) (2.RT S .22c) (2.RT S = NRT S Ec Es. n2 . and Es.22d) In the above equations. Let nd = γd /∆t be the normalized delay constraint (which we assume is an integer).3.14) by taking the partial derivative of Gs (X. Es.4 Average Energy with Delay Constraint We deﬁne the transform.ACK + i=0 i(1 − pce )pi Ecoe ce pce . Hence. (2.24) The average energy with delay constraint γd is calculated as follows.ACK = (NRT S + NCT S + NDT + NACK )Ec .DT . the average transmitting energy for a successful data packet transmission can also be derived from (2.CT S . 1 − pce (2.DT = (NRT S + NCT S + NDT )Ec Es.23a) (2. Y ) with respect to Y . Then . for a sequence of numbers (n0 .23b) = Es. It can be written with Es. Es. (2. Es.ACK + Ecoe 2.22a) (2.ACK denote the energy consumed for the four above mentioned cases. n1 .CT S = (NRT S + NCT S )Ec Es.

Then we have l = 1. By setting ul = ∞ j=0 jP r(Td = l∆t . Et = j∆e ). (2. ∂Y (2.25) can also be expressed with system generating function Gs . . Et = j∆e ) nd i=0 P r(Td = i∆t ) nd i=0 In order to evaluate the denominator of (2. S(X) = Gs (X.25c) (2. The numerator of (2.25b) (2. We let sl = l i=0 P r(Td = i∆t ) and S(X) = ∞ ∞ l=0 sl X l . Y ) |Y =1 . Y ) |Y =1 . Multiplying both sides by X l . 1−X (2.26) sl − sl−1 = j=0 P r(Td = l∆t . the denominator of (2. Et = j∆e ).38 ∞ E[Et | Td ≤ γd ] = j=0 j∆e P r(Et = j∆e | Td ≤ γd ) ∞ j=0 nd i=0 nd i=0 (2.25) is the coeﬃcient of term in S(X) that has degree equal to nd and can be found by inverse transforming X + . Et = j∆e ) P r(Td ≤ γd ) ∞ j=0 jP r(Td = i∆t . Y ) |Y =1 .29) . and taking into account that s0 = ∞ j=0 P r(Td = 0∆t . Et = j∆e ) P r(Td ≤ γd ) ∞ j=0 jP r(Td = i∆t . 2.25a) (2.25d) = ∆e = ∆e = ∆e j P r(Td = i∆t . we have ∞ ∞ S(X)(1 − X) = l=0 j=0 P r(Td = l∆t . . Et = j∆e )X l (2. summing over l. .25) we proceed as follows. . Thus. we get U (X) = ∂Gs (X.28) Therefore. Et = j∆e ) and taking the transform of {ul } with dummy variable X.27) = Gs (X.

Y ) |Y =1 ∂Y (2. X nd ) 1−X coef( ∂Gs (X. We will use the reliability function bounds for a memoryless channel to determine the packet error probability. Let K be the number of information bits in a packet (speciﬁed by the 802. X nd ) 1−X ∆e . energy and delay for each type of packet in the system. E[Et | Td ≤ γd ] can be written as E[Et | Td ≤ γd ] = . (2.26) and (2. coef( Gs (X. The quantity E[Td | Et ≤ ne ∆e ] can be written as E[Td | Et < ne ∆e ] = 2. we need to have a relation between error probability. we can also determine the average delay with an energy constraint. coef( Gs (X.31) where coef(f (X).30) Finally. Y ne ) 1−Y coef( ∂Gs (X.K.4 Energy-Delay Optimization . X i ) is the coeﬃcient of the term X i in the polynomial expansion of f (X).11 standard) and N be the number of coded symbols for this packet.33) where R0 is the cutoﬀ rate.32) In order to evaluate the energy and delay characteristics.N is bounded by Pe.Y )|X=1 . Then there exist an encoder and decoder for which the packet error probability Pe. (2. An interesting observation is that there is an optimal N for minimum delay.K. By a similar approach as above. For small N the packet error probability of .N ≤ 2K−N R0 . the numerator becomes nd ui = ∆ e i=0 ∂Gs (X.39 Following a similar procedure as in (2.Y )|Y =1 .Y ) |Y =1 ∂Y 1−X . Y ne ) 1−Y ∆t .27). For an additive white Gaussian noise (AWGN) channel using binary input the cutoﬀ rate is R0 = 1 − log2 (1 + e−Ec /N0 ).Y ) |X=1 ∂X (2. where Ec /N0 is the received signal-to-noise ratio per coded symbol.

1 and 2.5. we propose an approximate method to determine the energy and delay tradeoﬀ curves mentioned above. In this section. For the case when the number of information bits is large. and optimize delay (average or energy constrained) over the corresponding packet lengths NRT S . Repeating the above procedure for diﬀerent Ec /N0 . After getting the optimal packet error probabilities and lengths. the packet error probability for a random code will be small but the transmission time will be large.3. Our goal is to ﬁnd the best N for each kind of packet to minimize system delay. Some approximations are made to obtain the estimates of the packet error probabilities that achieve minimum average delay. T d . Based on these optimal values.3. The ﬁrst step of this approximation is to derive the average delay T d when there is only one transmitting station in the network. NCT S . if N is large. we will concentrate on deriving the approximate equations for energy and delay tradeoﬀ curves. we get a curve quantifying the tradeoﬀ between energy and delay.3. NDT and NACK . we ﬁx Ec /N0 . we evaluate the energy consumption (average or delay constrained). On the other hand. To achieve this. Large packet error probability will increase the system delay and energy due to the high chance of packet retransmission. We then evaluate the corresponding packet lengths by solving (2. We begin by considering the average delay with only one transmitting station that uses the same protocol as described in Section 2. This means that there are .33) for N (used as an equality) with the optimal packet error probabilities. The numerical results of energy and delay curves will be given in Section 2.2.40 a random code will be large. The purpose to provide the approximation method to determine the energy and delay curve is that it can reduce the time dramatically in evaluating the energy and delay curve. the average energy E t and delay T d are evaluated according to Section 2.

CT S )Pe.DT Ts.DT )Pe. but. there is no error in the CTS packet. We need to modify the analysis for the single station case by excluding the eﬀects of packet collision.RT S + (1 − Pe.ACK Ts.CT S )(1 − Pe.CT S Ts.CT S pce +(1 − Pe. The last case is that the RTS. The ﬁrst case is packet error of the RTS packet.ACK ].CT S )Pe.34) There are four possible cases of an unsuccessful transmission.ACK .CT S + (1 − Pe. (2.RT S )(1 − Pe.RT S )Pe. The time duration of each above cases is Ts.RT S + (1 − Pe. The third case is that the RTS packet is correctly received. Now the probability pce can be rewritten as pce = Pe. We still use pce to represent the probability from one stage to next.41 no RTS packet collisions in the medium.RT S )(1 − Pe.35) The average elapsed time Tr. (2.RT S )(1 − Pe. CTS and DT packets are successful but there is an error at the ACK packet.RT S . denoted as Tcoe .RT S )(1 − Pe. the packet collision probability pc becomes zero.ACK . is Tcoe = 1 [Pe.CT S )(1 − Pe.CT S .RT S Ts.DT +(1 − Pe.i before the reference station makes its (i + 1)th retrans- . there is an error happened in the DT packet transmission.RT S )Pe. The average duration while the reference station itself occupies the channel during each unsuccessful retransmission attempt. The second case is that the RTS packet is correctly received but there is an error in the CTS packet transmission. An unsuccessful transmission can happen only due to packet errors.DT and Ts. Ts.DT )Pe. Ts. Therefore.DT +(1 − Pe.

19) is derived under the assumption of large n while (2. as can be seen from the (2.i ce = Ts. (2..36) is not derived from (2.ACK and Tcoe . 1 − pce 2 (2.ACK + Tcoe pce W σ 1 − 2(2pce )m (1 − pce ) + pm (1 − 2pce ) ce + + 1 − pce 2 1 − 2pce 2m pce m (2m+1 − 1 + )p . (2. The reason is that (2. Finally.DT 1 and Pe.19) by substituting pc = 0.36) is derived separately for n = 1 since the large n approximation does not hold. TSIF S and TDIF S can be ignored in the evaluation of Ts.39) .34).ACK 1).38) Using the assumption that the number of information bits K is large for each kind of packet.37) can be approximated by keeping the ﬁrst order terms in Pce as T d ≈ Td = Ts. The validity of this assumption will be veriﬁed later.i = ( k=0 = iTcoe + iTcoe + Wk σ ) + iTcoe 2 W σ(2i+1 −1) 2 W σ(2m+1 −1+2m (i−m)) 2 for 0 ≤ i ≤ m − 1 for m ≤ i.ACK + Tcoe pce Wσ + (1 + 2pce ).36) Observed that (2.CT S 1. the average system delay for successful transmission is ∞ T d = Ts. (2.ACK + i=0 (1 − pce )pi Tr. Pe. then pce will also be very small compared to 1.ACK = (NRT S + NCT S + NDT + NACK )Tb + 3TSIF S + TDIF S ≈ (NRT S + NCT S + NDT + NACK )Tb .37) 1 − pce ce If we assume that the optimal error probability of each packet is quite small (i.42 mission attempt is i Tr. resulting in Ts. Pe.RT S 1. The average system delay in (2. Pe.e.

CT S (NRT S + NCT S )Tb +(1 − Pe.RT S )(1 − Pe. (2.DT )Pe.ACK (NRT S + NCT S + NDT + NACK )Tb + Wσ (1 + 2(Pe.DT (NRT S + NCT S + NDT )Tb +Pe.RT S + (1 − Pe. the approximate average delay of the system becomes Td = (NRT S + NCT S + NDT + NACK )Tb + 1 Pe.DT )Pe.RT S )Pe.40) By using (2.RT S + Pe.38).ACK · ((NRT S + NCT S + NDT + NACK )Tb + 3TSIF S + TDIF S )] ≈ 1 [Pe.RT S )(1 − Pe.ACK )).RT S (NRT S Tb + TDIF S ) + (1 − Pe.DT Ts.RT S )Pe.CT S )(1 − Pe.ACK Ts.CT S Ts.RT S )Pe.CT S )(1 − Pe. 2 (2.CT S )(1 − Pe.DT )Pe.RT S Ts.43 and Tcoe = 1 [Pe.RT S )(1 − Pe.DT ((NRT S + NCT S + NDT )Tb + 2TSIF S + TDIF S ) + (1 − Pe.DT + Pe.41a) ≈ (NRT S + NCT S + NDT + NACK )Tb + Pe.RT S )(1 − Pe.CT S )Pe.CT S (NRT S + NCT S )Tb + Pe.CT S + Pe.DT )Pe.RT S )(1 − Pe.CT S (NRT S + NCT S )Tb pce +(1 − Pe.RT S )Pe.RT S )(1 − Pe.DT +(1 − Pe.DT (NRT S + NCT S + NDT )Tb +(1 − Pe.39) and (2.41b) .RT S NRT S Tb +Pe.DT (NRT S + NCT S + NDT )Tb +(1 − Pe.CT S )Pe.CT S )Pe.ACK ] = 1 [Pe.RT S )(1 − Pe.ACK · (NRT S + NCT S + NDT + NACK )Tb ].RT S NRT S Tb + (1 − Pe.RT S NRT S Tb 1 − pce +(1 − Pe.CT S pce +(1 − Pe.ACK · (NRT S + NCT S + NDT + NACK )Tb + Wσ (1 + 2pce ) 2 (2.CT S ((NRT S + NCT S )Tb + pce TSIF S + TDIF S ) + (1 − Pe.CT S )Pe.CT S )(1 − Pe.RT S )(1 − Pe.

ACK ) − W σPe.DT 2−KDT ) + ln(Pe.RT S 2−KRT S ) + ln(Pe.ACK R0 ln(2) +Pe. (2. ∗ ∗ ∗ ∗ Our goal is to ﬁnd the packet lengths NRT S .RT S − KRT S ln(2) + ln Pe.42) By setting (2. we get four packet error probabilities that achieve minimum average delay. We will derive the optimal packet probability of ACK as an example. NCT S .ACK ) − W σPe.ACK [ln Pe.ACK R0 ln(2) +Pe.ACK R0 ln(2) +Pe. After taking the partial derivative with respect to the four packet lengths and setting them equal to zero. The other three optimal packet error probabilities can be obtained in a similar way.ACK R0 ln(2) −Pe.ACK ) − W σPe.42) equal to zero.ACK [ln(2−NRT S R0 ) + ln(2−NCT S R0 ) + ln(2−NDT R0 ) + ln(2−NACK R0 )] = Tb (1 + Pe.ACK R0 ln(2) − Pe.ACK − KACK ln(2)] ≈ Tb − W σPe.ACK (ln(Pe.ACK R0 ln(2)(NRT S + NCT S + NDT + NACK ) = Tb (1 + Pe.ACK ) − W σPe.CT S − KCT S ln(2) + ln Pe.ACK [ln(2−NRT S R0 2KRT S 2−KRT S ) + ln(2−NCT S R0 2KCT S 2−KCT S ) + ln(2−NDT R0 2KDT 2−KDT ) + ln(2−NACK R0 2KACK 2−KACK )] = Tb (1 + Pe.44 where we have used the assumption that the optimal packet error probabilities are small.ACK ) − W σPe.CT S 2−KRT S ) + ln(Pe. we obtain the optimal probability for the ACK packet .ACK ln(2)(KRT S + KCT S + KDT + KACK ).ACK R0 ln(2) + Pe. We have ∂ Td ∂NACK = Tb (1 + Pe.ACK 2−KACK )) = Tb (1 + Pe. NDT and NACK that minimize the average system delay Td .DT − KDT ln(2) + ln Pe.

43d) Note that the optimal packet error probabilities in (2.43b) (2.ACK + p∗ p∗ ce ce ∗ ∗ ∗ Tcoe + Bsingle .44b) R0 KDT + log2 (ln(2)((KRT S + KCT S + KDT ) + W σR0 /Tb )) ≈ (2.ACK . we can derive the approximate energy-delay values analytically from (2. Since p∗ is evaluated from (2.CT S ∗ Pe.ACK and Ecoe can be determined from (2.22).44d) For the single user case.45) 1 − p∗ 1 − p∗ coe ce ce . KCT S .45 as ∗ Pe.21). (2. Es.44).44).44c) R0 KACK + log2 (ln(2)(KRT S + KCT S + KDT + KACK ) + W σR0 /Tb )) ≈ . Es.43c) (2. R0 (2.RT S ≈ ∗ Pe. Tcoe .DT Tb Tb + ln(2)(Tb KRT S + W R0 ) Tb ≈ Tb + ln(2)(Tb (KRT S + KCT S ) + W R0 ) Tb ≈ .43a) Using a similar derivation as above.17). KDT and KACK are large.34) with (2.ACK ≈ ln(2)(KRT S + KCT S 1 . (2.43) and ce ∗ ∗ ∗ ∗ Ts.44a) R0 KCT S + log2 (ln(2)((KRT S + KCT S ) + W σR0 /Tb )) ≈ (2.33) as an equality. + KDT + KACK + W σR0 /Tb ) (2. we can also solve for the optimal packet lengths by using (2. Tb + ln(2)(Tb (KRT S + KCT S + KDT ) + W R0 ) (2.43) and (2. (2. and (2. thus obtaining ∗ NRT S ≈ ∗ NCT S ∗ NDT ∗ NACK KRT S + log2 (ln(2)(KRT S + W σR0 /Tb )) (2.35) with (2.43) are consistent with the assumption that they are small compared to 1 when KRT S .ACK + E ∗ ). the other optimal packet error probabilities are given as ∗ Pe. Et∗ ) = (Ts. Furthermore. the energy-delay values can be analytically obtained as: ∗ ∗ (Td .

we plot the average energy and delay tradeoﬀ curves for diﬀerent KDT . The explanation is that the delay or energy consumption of the overhead transmissions.46 ∗ where Bsingle is ∗ Bsingle W σ 1 − 2(2p∗ )m (1 − p∗ ) + p∗m (1 − 2p∗ ) 2m p∗ ce ce ce ce ce m+1 = [ + (2 −1+ )p∗m ].1 summarizes the system parameters used to obtain the numerical results. the ce c tx tx approximate energy-delay values for the multiuser case can be obtained in a similar manner to the single user case as ∗ ∗ (Td . Table 3. 2.3). p∗ and p∗ 1 and the optimal packet lengths obtained from (2.5 Numerical Results In this section we present and discuss our numerical results for various scenarios and system parameters.ACK + p∗ p∗ ce ce ∗ ∗ ∗ ∗ Tcoe + Bmulti Trc .5.4) and (2. we normalize the energy by N0 since all results depend on the ratio of energy to noise power spectral density.48) 2. such as RTS. 2. CTS and . In all plots.2). Es.ACK + E ∗ ). ∗ ∗ 2 1 − 2pce 1 − pce ce (2.43). p∗ . By using the optimal probabilities of p∗ . (2. In Fig. both energy and delay are normalized with respect to KDT to allow for a fair comparison. (2. Et∗ ) = (Ts. In addition. p∗ . p∗ and p∗ 1 will be evaluated ce c tx tx according to (2.44). (2. ∗ 2 1 − 2pce 1 − p∗ ce ce (2. the normalized energy and delay decrease.5) with (2. the probabilities of p∗ .46) For the multiuser case.47) ∗ 1 − pce 1 − p∗ coe ce ∗ where Bmulti is ∗ Bmulti = p∗ W σ 1 − 2(2p∗ )m (1 − p∗ ) + p∗m (1 − 2p∗ ) 2m p∗ c ce ce ce ce ce [ + (2m+1 − 1 + )p∗m ].5 is that while both energy and delay increase with KDT . The interesting observation in Fig.

we plot the average energy and delay tradeoﬀ curves evaluated via our . We also plot the single user case for the same protocol (802. 1/Tb 1 Mbit/s Slot Duration.1: System Parameters for Numerical Results KRT S 128 bits KCT S 128 bits KACK 128 bits Channel Bit Rate. 2. become less signiﬁcant for larger data packet lengths.5 2 10 15 20 25 30 Td/KDT (µ sec) 35 40 45 Figure 2.6 that both energy and delay increase with the number of users.11 MAC protocol.47 Table 2.11) and the simple ARQ. ACK.5 3 2. 2. To study the eﬀect of diﬀerent users. m 5 Minimum Contention Window Size. In Fig. We can see from Fig.5 KDT=6400 KDT=5400 KDT=4400 KDT=3400 KDT=2400 5 4. we used C++ programming language to write an event-driven custom simulation program for 802. To verify the energy and delay tradeoﬀ curves.5 Et/(KDTN0) (dB) 4 3.5: Energy-delay curves for diﬀerent KDT with n = 10.7. W 8 slots TSIF S 28µs TDIF S 128µs 5. we ﬁx the data packet length as KDT = 6400 and plot the average energy-delay curves for diﬀerent n. These two curves are lower bounds to the multiuser curves. σ 50µs Maximum Contention Stage.

We observe that the average energy and delay tradeoﬀ curves obtained from analytical model is very accurate. It is also clear from the ﬁgure that the normalized standard deviation is grown in proportion to the normalized average delay with a factor of 1.11 4.5 3 2. We observe that the normalized standard deviation of the delay decreases as KDT increases.5. NDT . numerical analysis and via simulation for diﬀerent KDT and n with parameters described in Table 3. In Fig.802. NCT S . 2 The delay variance can be calculated from the generating function as σd = ∆2 ∂ G2s |X=Y =1 +∆t T d − T d .1. 2. NACK be the packet lengths that minimize T d .9.5 4 Et/(KDTN0) (dB) 3. The average energy-delay tradeoﬀ curves with diﬀerent delay constraints are shown in Fig. The system 2 2 generating function with these packet lengths and Ec /N0 is denoted as G∗ . 2.5 2 1. the normalized average delay for diﬀerent n and KDT . ∗ ∗ ∗ ∗ Let NRT S . NCT S . NDT .6: Energy-delay curves for diﬀerent number of users n with KDT = 6400. NACK . We ﬁx Ec /N0 and optimize T d over NRT S . If outage s delay probability is given. we can evaluate both outage delay γd and E[Et | Td < γd ] .48 5 n=10 n=12 n=14 n=16 n=18 n=1.5 0 10 20 30 40 Td/KDT (µ sec) 50 60 70 Figure 2. we plot the normalized standard deviat ∂X tion of the delay vs. ARQ n=1.8.

8: Normalized standard deviation of delay vs.5 DT DT =3200 =6400 3 2. 160 140 120 standard deviation of delay/K DT 100 80 n=10 60 n=20 40 20 10 20 30 40 50 Td/KDT 60 70 80 90 100 Figure 2.49 5.7: Energy delay curves for diﬀerent number of users and packet sizes.5 n=15. average delay curves for diﬀerent n and KDT . K 5 n=15. The lines with squares represent numerical results and the lines with circles represent simulation results. .5 n=10. K 4. Lines with a square symbol represent the case of KDT = 4400 and lines with a star symbol represent the case of KDT = 2400. KDT=6400 3. KDT=3200 Et/(KDTN0) (dB) 4 n=10.5 2 0 10 20 30 Td/KDT (µ sec) 40 50 60 Figure 2.

2. The observations made for Fig. n = 10. we obtain the average s energy-delay tradeoﬀ curve with a delay constraint. we also plot the average energy vs. It is observed that the approximate method is very accurate over a . KDT = 640).01 20 E[Et|Td<γd]/KDT (dB) 15 0.11. we plot energy-delay curves of under diﬀerent KDT with n = 10 users in the network. the range of outage delay decreases and conditional average energy also decreases. average delay curve. Given an outage energy probability constraint.10. 2. In Fig. 2. For comparison.1 10 5 0.50 from G∗ .10 as well. 30 25 0. 2. We see that the average energy used can be quite diﬀerent depending on whether an average delay or a strict delay requirement is imposed. The average energy and delay curve (circle symbol) is also shown for comparison.9 shows that when the outage delay probability increases. Fig. W = 8.9 apply to Fig. The average energy-delay tradeoﬀ curves with diﬀerent energy constraints is given in Fig. Then using the same procedure as above we obtain the average energy-delay tradeoﬀ curve with an energy constraint. 2.5 0 0 50 100 150 200 γd/KDT (µ sec) 250 300 350 Figure 2.9: Energy-delay curves for various outage delay probabilities and diﬀerent values for the outage probability P r(Td > γd ) (m = 1. Repeating the above procedure for diﬀerent Ec /N0 . we can evaluate both the outage energy γe and E[Td | Et < γe ].

5 6 5 10 20 30 40 50 60 70 80 90 E[Td|Et<γe ]/KDT (µ sec) Figure 2. a very accurate analytical approximation for the minimum average delay and corresponding delay was derived that enabled us to get the optimal energy-delay tradeoﬀ curves analytically. etc. KDT = 640). Finally.1 7 0.10: Energy-delay curves for various outage energy probabilities and diﬀerent values for the outage probability P r(Et > γe ) (m = 1.51 13 12 11 10 N ) (dB) 0. W = 8. the average energy with a delay constraint.01 9 γ /(K e DT 0 8 0. This approach allows us to optimize the performance over the block lengths used for diﬀerent packets. n = 10. The generating function was used to obtain various statistics such as the average energy and delay. wide range of KDT . By representing the protocol using a state diagram we derived the joint generating function of energy and delay. 2. We believe that our methods can be extended to the analysis of other protocols. The average energy and delay curve (circle symbol) is also shown for comparison. including .6 Conclusion In this chapter we obtained the joint distribution of energy and delay for packet transmission using RTS/CTS-type MAC protocols taking into account the PHY layer.

52 segmentation of data packets or retransmission of data segments without channel reservation.11: Energy-delay tradeoﬀ curves evaluated from the approximation method (star symbol) and the exact energy-delay tradeoﬀ curves (square symbol) with n = 10 . 22 20 18 K DT =10 16 N ) (dB) E /(K t DT 0 14 12 KDT=50 10 KDT=100 8 6 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Td/KDT (µ sec) Figure 2.

1 Adaptive Energy Scheme for Wireless Network Systems Wireless networks consists of mobile and sensors devices. Decreasing the amount of energy used in transmission increases the packet error probability and can increase the delay for a successfully received packet. we apply our proposed generating function method to the analysis and design of two diﬀerent wireless network protocols modiﬁed from the original IEEE 802. Decreasing the transmission delay by increasing the transmission rate gen53 . The conclusion of this chapter is stated in Section 3. adapts the energy per coded symbol whenever it suﬀers a failed transmission. Higher energy eﬃciency protocols and signal design may extend the battery lifetime or result in using batteries with smaller required energy capacity.11 MAC Protocol In this chapter. The ﬁrst wireless network protocol. discussed in Section 3. However. Most such devices are battery powered. Since the energy of a battery is a limiting resource it is important to have an energy eﬃcient network design.11 protocol. The second wireless network protocol related to the application of ARQ to IEEE 802.2.CHAPTER III Variations of 802.1.11 is investigated in Section 3. 3. wireless channels have fading that typically requires more energy for transmission than an unfaded channel for the same packet error probability.3. The tradeoﬀ between energy and delay is studied for both proposed protocols.

Their analytical and simulation results demonstrated a signiﬁcant decrease in energy consumption.g. there is a fundamental tradeoﬀ between energy and delay in designing a wireless network. The duration of this time interval is called beacon interval. the authors presented a new MAC protocol for energy saving mode in IEEE 802. So high priority data ﬂows would get more chance to be served when energy in the battery becomes lower. In [8].11 by proposing a quorum-based sleep/wake-up mechanism. network topology and network traﬃc ﬂows. The energy eﬃciency of wireless networks has been a subject of current research. which conserves energy by allowing the host to sleep for more than one beacon interval if few transmissions are involved.11 standard.11 energy saving mode since the original scheme fails to adjust a station’s sleep duration according to its traﬃc and nearby network topology.. As actual energy of the battery becomes low. They studied and analyzed the performance of this scheme and compare its performance with the original IEEE 802. the proposed scheme gives more weight to high priority queue. the packet error probability is an important parame- . Hence. Their results showed that the proposed scheme improves overall end-to-end throughput as well as support service diﬀerentiation over multi-hop wireless networks. 42] proposed a novel MAC scheme that changes the duration of listen/sleep modes adaptively according to the information about the network. The authors in [51. e. In [16]. the authors proposed a new scheduling scheme called adaptive Weighted Round Robin (WRR) scheme that provides service diﬀerentiation in multiple data class queuing system by considering battery life and load of traﬃc.54 erally requires more energy. In order to synchronize diﬀerent stations in a wireless network. In a wireless environment. This new scheme is more energy eﬃciency compared to the original IEEE 802. the beacon needs to be sent out repeatedly for some time interval.

The motivation of this section is to understand the energy and delay relation of networks by simultaneously incorporating the packet error probabilities (physical layer) and basic access method (MAC layer) in the performance analysis. in all the above mentioned papers.3. we also compare the system performance between Reed-Solomon codes and convolutional codes. In Subsection 3. In Subsection 3. we present numerical results regarding the throughput. However.11 (RTS.1. In Subsection 3.55 ter in overall system performance. In this section. we propose and analyze an adaptive protocol in which the transmitter will increase the energy level per coded symbol whenever it suﬀers a failure transmission.1.1.1. By using Reed-Solomon codes we can optimize the system performance over the code rate and energy per coded bit. the framework of our analysis can be used for other coding and modulation schemes over various wireless channels. we introduce the generating function and determine the joint generating function for energy and delay. the eﬀects of energy consumption for each kind of packet error probability used in IEEE 802. CTS. . none of the above papers consider the energy and delay consumption jointly of transmitting a data packet in the system. The numerical results show that the proposed scheme can improve system performance signiﬁcantly compared to the original IEEE 802.2. The rest of this section is organized as follows. Moreover.11 protocol when the channel condition is bad. DATA and ACK packets) was not considered. Although we use Reed-Solomon codes over an additive white Gaussian noise (AWGN) channel to represent packet error probabilities. energy eﬃciency and energy/delay tradeoﬀ with Reed-Solomon codes over an AWGN channel. Finally. we introduce the system assumptions and give a brief description for the protocol used in our analysis.

. stops the count down when the channel is sensed busy. In the following we give a brief description of the most salient features of the IEEE 802.56 3. If there is a collision of the RTS packet with other competing stations or a transmission error occurs in the RTS or CTS packet. it picks randomly a number j. stations always have a packet ready to transmit. A backoﬀ time counter begins to count down with an initial value j: it decreases by one for every idle slot of duration σ seconds as long as the channel is sensed idle. If 1n is the density of stations. 1. . Third. Wi − 1}. the receivers have no multiple-access capability (i. . . These assumptions are valid for local area networks. and reactivates when the channel is sensed idle again. i is the contention stage (initially i = 0). each station uses 802. the transmitting station doubles the CW size (increases the contention stage i by one) and picks a random number j as before. we assume that the propagation loss between transmitter and receiver is one (0 dB) and also assume that the propagation time is negligible. After transmitting the RTS packet. the station will wait for a clear-to-send (CTS) packet from the receiving station.1. Fifth. Fourth. and W is the minimum CW size. all stations have ﬁxed position (no mobility). The station transmits a request-to-send (RTS) packet when the counter counts down to zero. where Wi = 2i W is the contention window (CW) size. Second. .11 protocol for medium access control (MAC). each station with a ﬁxed position can hear (detect and synchronize) the transmission of n−11 other stations in the network. uniformly distributed in {0. they can only receive one packet at a time). When a station is ready to transmit a packet.e. .1 System Description The wireless networks that we analyze here have the following speciﬁcations. Finally.11 MAC protocol (more details can be found in [3] and [4]). First.

1. The contention stage is reset (i is set to zero) when the transmitting station receives an ACK correctly.1 stage number is increased by 1 0 DIFS reference station successful transmission DIFS frozen slot counter Td end stage number is set back to 0 energy level is Ec.0 0 DIFS j1 ~ U(0. the station begins to transmit the data packet and waits for an acknowledgment (ACK) packet. if the data or the ACK packet is not successfully received. A time diagram indicating the sequence of these events is depicted in Fig. However.W 0-1) DIFS frozen slot counter frozen slot counter Td begin energy level is Ec.1.W 1-1) reference station unsuccessful transmission DIFS frozen slot counter energy level is Ec. a maximum contention stage. . It is also noted that there is a maximum CW size (or equivalently. it does not increase further the contention window. the transmitter increases the energy level per coded symbol whenever the CW stage increases and let Ec. the CW size will also be doubled (the contention stage i will increase by one) and the transmitting station will join the contention period again. m). The proposed adaptive scheme uses the same MAC protocol as above. . when the transmitter is in this maximum stage and needs to join the contention period again.0 Figure 3. however. .i represent the energy level per coded symbol at transmission with CW stage i − 1. respectively. 3. . but picks a random number in {0. . j0 ~ U(0. and i = 1. The j0 and j1 are backoﬀ random numbers at CW stage i = 0.57 there are no collisions or errors in the RTS and CTS packets. 2m W − 1}.1: Timing diagram for protocol.

58 3. the one- . we only consider the energy consumption for packet transmission and omit the energy required for signal processing and channel sensing since the energy consumption for packets transmission requires most energy compared to other energy consumption aspects. i.1.e. With the assumption that errors in transmission can occur only due to collision. . 3. The second random process st is used to represent the contention stage i ∈ {0.. . In this paper. . The delay Td of each data packet is deﬁned as the time duration from the moment the backoﬀ procedure is initiated until DIFS seconds after the ACK packet is received correctly by the transmitting station. It uses the same time scale as the backoﬀ counter process. The ﬁrst process bt represents the value of the backoﬀ counter while it is active. 1. we deﬁne two random processes to characterize the random backoﬀ scheme. the energy Et is deﬁned as the energy consumed by both transmitting and receiving stations in the duration of Td . the probability of a collision given a packet transmitted on the channel. Similarly. Nonlinear System Equations As in [5]. Let pcc be the conditional collision probability. bt ) by a discrete-time Markov chain. m} of the station. . The ﬁrst assumption made in this analysis is that the event of packet collision is independent of past collisions and thus independent of the contention stage. we will model the two dimensional process (st .1. these two assumptions are extremely accurate when the number of stations in the network is large (say greater then 10). As veriﬁed in [5]. as shown in Fig.2 System Delay and Energy Analysis In this section. we analyze the energy and delay characteristics of the wireless networks described above. The second assumption is that the packet collision is identical for all states (values of bt and st ) of a user. As a result of the above assumptions.

These probabilities depend on the particular channel. k | m. . 1. k | i − 1.1. Packets which are transmitted at stage i will be transmitted with energy level Ec. 0} = P {i. W0 −1}. Fig.1) The ﬁrst equation in (3. k + 1} = 1. . Wm 0 ≤ k ≤ Wi − 2.i . (3.i per coded symbol. coding. k | j.i . m}. 0 ≤ i ≤ m 0 ≤ k ≤ W0 − 1. Pe. bt = l} with P {i. 0 ≤ i ≤ m 0 ≤ k ≤ Wi − 1.3). 0} = 1−pc . bt+1 = k | st = j. l} = P {st+1 = i. k | i.i . .i represent the transition probability from CW stage i to . . modulation etc (a speciﬁc example will be given in Section 3. we can take into account packets errors with diﬀerent energy level per coded symbols. Wi }. . 1. . and thus the backoﬀ counter is initially uniformly chosen in the range of {0.1) corresponds to the decrement of the backoﬀ counter at the beginning of each time slot. We denote the error probability of the four kinds of packets at CW stage i in the system as Pe. As described in the third equation. The other two cases describe the system evolution after an unsuccessful transmission. . k | i. . P {0. the contention stage increases and the backoﬀ counter is initialized with a uniformly chosen value in the range {0.RT S. where i in {0. 4 in [5] depicts the aforementioned two-dimensional Markov chain. W0 pc . By modifying the Markov chain model described above. Finally. .59 step transition probabilities developed in [5] are given by P {i. Wi pc . We assume that the channel is memoryless between packets. 1. 1 ≤ i ≤ m 0 ≤ k ≤ Wm − 1.i and Pe. .CT S. the last case models the fact that the contention stage is not increased in subsequent packet transmissions when the contention window size reaches the maximum. when an unsuccessful transmission occurs at contention stage i − 1. .ACK. The second equation accounts for the fact that a new packet following a successful packet transmission starts at contention stage i = 0.DT. Let pce. 0} = P {m. Pe. .

pce.i ].DT. pce.i + (1 − Pe.m .60 the next.DT. k + 1} = 1.i )(1 − Pe.m + W (1 + ( j=1 i=0 2m m−1 pce.2) 0 ≤ k ≤ Wi − 2.CT S.i )Pe.RT S.i )Pe.i ) + D m−1 i=0 pce. 1 ≤ i ≤ m 0 ≤ k ≤ Wm − 1 (3.i /(1 − pce.i 1 − pce.i + (1 − Pe.i i=0 ).m (3. Wm (3.(i−1) . an unsuccessful transmission can happen not only due to packet collision. 0} = P {m. P {0. (3.i )(1 − Pe.i can be expressed as pce.i ) + 1 − pce. It can be shown as ptx = 2(1 + m−1 j=1 ( j−1 i=0 pce. in this paper. k | i.m )) . However.RT S.i + (1 − Pe. k | i − 1. but also due to packet errors.i )Pe.RT S.RT S.CT S. 0} = P {i. k | i.i . W0 pce. Wi pce. 4 at [5]. 0} = 1−pce.CT S.ACK. Using similar assumptions as in [5] for the packet collision probability pc .3) The packet transmission probability ptx which is the probability of a transmitting station sending an RTS packet during each backoﬀ slot is obtained by summing over the stationary probability of all the states at ﬁrst column of the two-dimensional Markov diagram as shown in Fig.4) where D is expressed as m−1 j−1 D =1+ ( j=1 i=0 m−1 j−1 pce.i = pc + (1 − pc )[Pe. k | m. 0 ≤ i ≤ m 0 ≤ k ≤ W0 − 1. and since the events of packet collision and the event of packet error are independent. 0 ≤ i ≤ m 0 ≤ k ≤ Wi − 1.i )(1 − Pe.i ) + 2 i+1 m−1 i=0 pce.5) . The one-step transition probabilities are modiﬁed as P {i. This is also the probability of an unsuccessful transmission attempt seen by the transmitting station when its packet is being transmitted on the channel.

Let TRT S .61 The collision probability pc can be expressed as pc = 1 − (1 − ptx )n−1 . . one can evaluate the probabilities pce. Another important probability that will be used in our analysis later is the probability of exactly one transmission of an RTS packet from one of the remaining n − 1 stations given that at least one of the remaining stations is transmitting. = pc (3.6).i . 3. Let Rt be the transmitting rate in bits per second. CTS. Similarly. From the above nonlinear equations.i /N0 . Solid lines represent successful reservation or transmission. N0 ) is N Ec. ECT S. The delay. the transmission delay of each packet is also proportional to the packet length.6) since the collision will happen if there are at least one remaining n − 1 stations transmitting packets.2 can be built. Similarly. (3.1. in seconds. ptx and pc . Based on the protocol description in Section 3. the state ﬂow diagram shown in Fig. TCT S . (3.i . let ERT S. TDT and TACK denote the time duration for the transmission of RTS. while dotted lines represent unsuccessful reservation or transmission. where i is the stage of CW. (3. respectively. for a packet transmission with packet length N is N/Rt . It is denoted by ptx1 and can be expressed as ptx1 (n − 1)ptx (1 − ptx )n−2 . Then the energy transmission for a packet of length N (normalized to the thermal noise energy level.1.4) and (3.2).i .7) Delay/Energy Joint Generating Function We assume that stations spend energy only in packet transmission and that the energy consumption for each packet transmission is proportional to the packet length. DT and ACK packets.

2 Gfr.0Gbs. 3. In all the derivations below. where the variables X. Y ) = (1 − pc )(1 − Pe. functions are of the form G(X. DT and ACK packets at CW stage i.1 successful DT transmission ? Gtf. associated with the transitions in the state diagram in Fig. channel reservation ? Gfr. respectively.i . (3.m Gbs.1 Gts.2: State diagram for the protocol.m-1 Gfr. let TDIF S and TSIF S be system delay parameters deﬁned by the standard. respectively.CT S.i +ECT S. The generating function Grs.8) Similarly.m-1 Gbs.i )(1 − Pe.m Gbs. Finally.1 successful.i and EACK.RT S.m successful.2. CTS. channel reservation ? Gtf. Y are the transform variables of the delay and energy. We will now derive the conditional joint generating function of the delay and energy random variables.0 Gts.0 successful DT transmission ? Gtf.2 Grs. Y ).i denote the energy for the transmission of RTS. channel reservation ? successful.1Gbs. EDT.m-1 Gtf.m Wait DIFS Figure 3. the generating function Gts corresponding to successful transmission of a .i corresponding to successful channel reservation at CW stage i is Grs.0 Gfr.62 Wait DIFS Gbs.1 Grs.m-1 Gbs.m successful DT transmission ? Gts.0Gbs.i ) X TRT S +TCT S +2TSIF S Y ERT S.i (X.1Gbs.m Grs.

or the acknowledgement packet transmission error.i ]X TRT S +TDIF S Y ERT S. which is captured by the generating function Gtf.i .i + (1 − pc )(1 − Pe. Y ) = [pc + (1 − pc )Pe.i (X.CT S.i +EACK.i .9) We now develop expressions for the generating functions associated with failure to either reserve a channel or to transmit data.i X TDT +TDIF S Y EDT.i +ECT S.ACK. We do not need to consider energy consumption here since the transmitting station stops any transmission at this period.i X TRT S +TCT S +TSIF S +TDIF S Y ERT S. a failure to transmit a data packet at CW stage i is either due to the data packet transmission error. corresponding to failure to reserve the channel at stage i is given by Gf r. (3. (3. The probability of .i (X.i ) X TDT +TACK +TSIF S +TDIF S Y EDT.i + (1 − Pe. We ﬁrst derive the generating function Gf r.i )(1 − Pe. Y ) = (1 − Pe.63 data packet at CW stage i can be expressed as Gts.i as follows Gtf.i of the state diagram. after reserving the channel. This generating function will characterize the delay for the transmitting station from the instant of starting the backoﬀ procedure to the instant that the backoﬀ counter reaches to zero at stage i.10) the meaning of which is that a failure can be due to either a collision or RTS transmission error.RT S. Y ) = Pe.ACK X TDT +TACK +TSIF S +TDIF S Y EDT.DT ) · · Pe.i .i +EACK. Similarly.RT S.11) We now evaluate the generating functions denoted by Gbs.i (X. or a CTS transmission error.DT. (3.i )Pe.i .DT.

The ﬁrst case is when two or more RTS packets from the remaining n − 1 stations collide or have a packet error. of backoﬀ procedure at stage i is 2i W Gbs. the correspond . At backoﬀ stage i.12) (pc Goc (X))k .0 /ptx . Note that the time duration of successful or unsuccessful ACK transmission are same.64 each slot being sensed busy due to the transmission of other stations is pc and this event is independent and identical to each slot from our previous assumptions. Let j be the backoﬀ slots randomly chosen uniformly from the above range.0 is the stationary probability of state that CW stage is i and the counter reaches to zero. the probability of the transmission come from CW stage i given only one remaining n − 1 station transmission is bi. Hence. Therefore. There are four possible cases of an occupied slot. and DT and either correct or erroneous reception of the ACK packet.i . then the number of the occupied slots in these j slots is binomialy distributed with parameters (j. Since there are m possible CW stages of packets transmission from the other n − 1 stations. where bi. We deﬁne an occupied slot as a slot when the transmitting station senses the channel is busy due to the transmission of one of the n − 1 remaining stations. The second case is when the RTS packet is correctly received without collision but there is an error in the CTS packet transmission. where Goc (X) is the generating function for an slot sensed busy by a station. the range of possible backoﬀ slots is from 1 to 2i W .i (X) = j=1 1 2i W j k=0 j [(1 − pcc )X σ ]j−k k (3. pc ). The last case is correct reception of RTS. the generating function. CTS. Gbs. The third case is when the RTS packet is correctly received without collision and there is no error in the CTS packet but there is an error in the DT packet transmission.

l ) X TRT S +TCT S +TDT +TACK +3TSIF S +TDIF S ].l X TRT S +TDIF S l=0 + (1 − Pe.RT S. Then the recursion is fm (X.i−1 + Grs.l X TRT S +TCT S +TDT +2TSIF S +TDIF S + (1 − Pe.m Gts.i fi Gs (X.RT S. Et.DT.1.65 generating function for this occupied slot is Goc (X) = (1 − ptx1 )X TRT S +TDIF S ptx + 1 ptx l=m bl.i−1 Gts. the mean values are given by Td.m fi−1 (X.m − Grs.l X TRT S +TCT S +TSIF S +TDIF S + (1 − Pe. ∂X ∂Y (3.15) In this section we present and discuss our numerical results for various scenarios and system parameters.m Gtf. we obtain the following backward recursive generating function to characterize the energy and delay of the system.l )(1 − Pe.avg = |X=Y =1 .RT S.CT S.CT S.i−1 Gtf. Table 3. Let Gs be the system generating function. Hence.l )Pe. Y ) = Grs.3 Numerical Results ∂Gs ∂Gs |X=Y =1 .14) where i is the index of the CW stage from 1 to m.13) From the state diagram and Mason’s gain formula.l )(1 − Pe.avg = 3. (3. (3.CT S.m Gbs.i−1 )Gbs. Y ) = Grs.1 summarizes the system parameters used to obtain .m Gbs.m 1 − Gf r.0 f0 .DT.0 [Pe.i−1 + (Gf r.l )(1 − Pe. In particular.l )Pe.RT S. Y ) = Gbs. we can evaluate any joint statistics of the energy and delay of a successful packet transmission.

We will use Reed-Solomon codes and convolutional codes to represent packet error probabilities.N = l=t+1 N Psl (1 − Ps )N −l .11 standard) and N be the RS number of coded symbols for this packet.66 Table 3. . W 8 slots TSIF S 28µs TDIF S 128µs the numerical results.K. For high SNR. 1/Tb 1 Mbit/s Slot Duration.3. 3.N over an AWGN channel with BPSK modulation can be expressed as N RS Pe. we observe that the throughput decreases with longer information packets since the packet error probability for convolutional codes increases when the length of information packets becomes larger.K. Note that Eb is the energy used per information bit.16) where t = N −K 2 and Ps = 1 − (1 − Q( is the symbol error probability. the relation between the average delay throughput ST d and the channel SNR with code rate 1 2 for diﬀerent KDT of convolutional codes is shown. if longer information packets are used. Let K be the number of information symbols (1 symbol = 8 bits) in a packet (speciﬁed by the 802. We expect that there is an optimal value of KDT which achieves the largest throughput for each given channel SNR. Moreover. the eﬀect of packet error probability to the ST d will be more signiﬁcant compared to the eﬀect of less overhead redundant for low channel SNR condition. For low SNR. m 3 Minimum Contention Window Size. σ 50µs Maximum Contention Stage. In Fig.1: System Parameters for Numerical Results KRT S 128 bits KCT S 128 bits KACK 128 bits Channel Bit Rate. l 2Eb K 8 )) N N0 (3. then the packet error probability Pe. we can ﬁnd that the throughput increases with longer information packets because the system with longer information packets has smaller overhead redundant in RTS/CTS MAC protocol.

For convolutional codes.3: The ST d comparison of convolutional codes for diﬀerent KDT with code rate n = 10.5 4 Figure 3.5 E /N (dB) ib 0 3 3.03 0. the relation between the average delay throughput ST d and the channel SNR with code rate 1 2 for diﬀerent KDT of Reed-Solomn codes is shown. we compare the ST d of convolutional and Reed-Solomn codes for diﬀerent rate with KDT = 128.4. we observe that the throughput increases with longer information packets because the packet error probabilities of Reed-Solomn codes decrease with the increase of the length of information packets.5. 1 2 and In Fig. We also found that the convolutional codes has better performance since it adopts soft decision instead of hard decision used in Reed Solomon codes. 3.02 STd 0.01 0. For low SNR. In Fig.005 2 2.025 0.015 KDT=128 K =256 DT KDT=384 K =512 DT K =640 DT K =768 DT K =896 DT K =1024 DT 0.67 0. the throughput decreases with longer information packets since the packet error probabilities of Reed-Solomn codes increases with the increase of the length of information packets. 3. we can see that the ST d increases with the decrease of code rate since the gain of ST d due to smaller packet error probabilities is more than the loss of ST d due to less information transmission . For high SNR.

5 9 Eib/N0 (dB) Figure 3. we observe that the ST d increases with the increase of code rate since the loss of ST d due to larger packet error probabilities is less than the gain of ST d due to higher information transmission eﬃciency per coded packet when the code rate becomes larger. packet error probabilities may not decrease since the energy used per coded bit also decreases with the lower code rate from (3. For lower code rate.6. Therefore.4: The ST d comparison of Reed-Solomn codes for diﬀerent KDT with code rate n = 10. we plot the ST d of Reed-Solomn codes with the channel SNR for diﬀerent code rate as KDT = 128. For higher SNR.5 5 5. since the packet error probabilities are almost zero. In Fig.68 10 −1 10 −2 10 −3 10 −4 STd 10 −5 10 −6 KDT=128 KDT=256 K =384 DT K =512 DT KDT=640 K =768 DT K =896 DT K =1024 DT 10 −7 4. For Reed-Solomn codes. The ST d is aﬀected by two factors.5 7 7. the value of ST d is dominated directly by the information transmission eﬃciency per coded packet (code . 1 2 and eﬃciency per coded packet when the code rate becomes smaller. the total system throughput ST d is the tradeoﬀ result of these two factors. we will have lower information transmission eﬃciency per coded packet.16). 3.5 6 6. packet error probabilities and the information transmission eﬃciency per coded packet (code rate). however.5 8 8.

r=16/16 Td 10 −12 10 −14 10 −16 10 −18 −2 −1. Therefore. r=7/8 RS. For lower SNR. r=2/3 RS. if N is large. r=1/2 RS.5: The ST d comparison of convolutional and Reed-Solomn codes for diﬀerent code rate with KDT = 128 and n = 10.7. We ﬁrst ﬁx Ec. Repeating the above argument for diﬀerent Ec.5 −1 −0. we get the energy-delay curve. we can evaluate both average delay and average energy consumption. r=1/4 CV. r=1/3 RS. NDT and NACK . rate).5 0 0. 3. r=2/3 CV. r=7/8 CV. NCT S . r=1/2 CV. An interesting observation is that there is an optimal N . we can ﬁnd that there is a cross of ST d values for curve with code rate 7 8 and curve with code rate 1 4 at 5.5 2 2. we compare energy and delay tradeoﬀ curves for diﬀerent values of . r=15/16 RS. r=3/4 CV. From this. Our goal is to ﬁnd the best N for each kind of packet to minimize system delay. since the gain of ST d due to smaller packet error probabilities is more than the loss of ST d due to less information transmission eﬃciency when the code rate becomes smaller. On the other hand. Large packet error probability will increase the system delay and energy due to the high chance of packet retransmission.69 10 −2 10 −4 10 −6 10 −8 S 10 −10 CV.5 1 1.5 Eib/N0 (dB) Figure 3. For small N the packet error probability of a random code will be large. the packet error probability for a random code will be small but the transmission time will be large.avg to get the corresponding packet lengths for NRT S . r=1/3 CV. In Fig.1 /N0 .6 dB.1 /N0 and optimize the Td. r=1/4 RS. r=3/4 RS.

The explanation of this is that the delay or energy consumption of overhead . 3.8 that both energy and delay increase with the number of users. the system incurs higher energy consumption (compared to the system with smaller value of ∆) to transmit a data packet successfully. The other fact shown in this ﬁgure is that both energy and delay increase with the increment of KDT but the normalized (with respect to KDT ) energy and delay decrease with increasing of KDT . For the eﬀect of diﬀerent users.70 10 −3 rate :1/2 rate :1/3 rate :1/4 rate :2/3 rate :3/4 rate :7/8 rate :15/16 10 −4 Td 10 −5 S 10 −6 10 −7 10 −8 2 3 4 5 6 7 8 Eib/N0 (dB) Figure 3. we ﬁx the data packet length as KDT = 64 symbols and plot energy-delay curve for diﬀerent n. This demonstrates that that the system performance was improved signiﬁcantly when the channel SNR is low and the average delay is reduced insigniﬁcantly even consuming more energy at high channel SNR case. although higher value of ∆ can reduce the average delay. At good channel condition. We observe that the adaptation scheme can improve the system performance by reducing both average energy and delay as the channel SNR is small.6: The ST d comparison of Reed-Solomn codes for diﬀerent code rate with KDT = 128 and n = 10. ∆. We can see from Fig.

the physical layer (lowest layer) handles the transmission of bits through a communication link and includes the forward error control (FEC) and modulator/demodulator.6 11.11 Protocol with ARQ Recently there has been considerable interest in the design and performance evaluation of wireless local area networks (WLANs). 3.7: Comparison of energy and delay curves for diﬀerent ∆ with n = 10 and KDT = 64. we can almost assume that delay or energy consumption comes solely from KDT . According to the OSI protocol layers speciﬁcation. The second layer is the data link layer which is responsible for establishing a reliable and secure logical link over the unreliable wireless link.6 dB t 40 50 60 d 70 DT 80 90 100 110 T /(8*K ) (µ sec) Figure 3. when the data packet length is large.2 11 10.3 dB ∆=0.6 10. .4 11. CTS and ACK will become less signiﬁcant for larger data packet lengths.71 11.2 10 9.8 10.6 30 ∆=0 dB E /(N 8*K ) (dB) 0 DT ∆=0. Hence. transmissions such as RTS.8 9. The error control coding technique reduces the required energy needed for transmission at the expense of increased delay and reduced data rate.4 10. The forward error control (FEC) coding technique is used to mitigate the eﬀect of channel noise at the receiver.2 802.

KDT =32 12 t 0 n=30. a wireless communication system adopting ARQ scheme can provide higher protection by allowing larger retransmission attempts. The lower sublayer of data link layer is the medium access control (MAC) protocol layer which resolves conﬂicts between two users attempting to access the channel. Thus there is some amount of energy and delay spent in reserving the channel. Several papers have investigated the eﬀects of using ARQ techniques in cross-layer .3(dB). If we want to decrease the delay to transmit the same data packet.8: Comparison of energy and delay curves for diﬀerent n and KDT with ∆ = 0. Given the energy used per data packet. The upper sublayer of the data link layer (DDL) is logical link control (LLC) sublayer that implements error control involving feedback from the receiver. the system requires higher energy per data packet to achieve the same reliability as a system with lower retransmission attempts. K 11 DT =64 10 n=20. The most common technique to combat errors used in this sublayer is automatic repeat request (ARQ). K 9 0 DT =64 100 50 Td/(8*KDT) (µ sec) 150 200 250 Figure 3. Therefore. However. a system with larger retransmission attempts needs longer delay to transmit the same data packet. there is an energy and delay tradeoﬀ problem happened in both layers (PHY and DLL).72 16 15 n=10. K 14 DT =16 ) (dB) DT 13 E /(N 8*K n=10. KDT =64 n=10.

a new cross-layer ARQ algorithm for video streaming over 802. Moreover. The algorithm combines application-level information about the perceptual and temporal importance of each packet into a single priority value. none of 2 A network system design methodology which considers at least two diﬀerent function layers jointly during the design task. In [45].73 design2 of wireless networks. which drives packet selection at each retransmission opportunity. the authors proposed a class-based adaptive ARQ scheme with QoS support for multimedia traﬃc to increase the throughput through the reduction of MAC overhead. The proposed design achieves the performance in the framework of a IEEE 802. DATA and ACK packets) is not discussed.5 dB PSNR gains using approximately half of the retransmission bandwidth. only the most most perceptually important packets are retransmitted. An OPNET simulation model is used to show that the proposed ARQ enhancement increases system performance. focusing on problems deriving from the usage of ARQ schemes in today’s wireless networks and proposed a novel and general cross-layer interaction scheme between the transport and the link layers.11 MAC-layer ARQ scheme. . CTS. the issue of energy consumption for each kind of packet error probability used in IEEE 802. Hence. Their results showed that the proposed method consistently outperforms the standard MAC-layer 802. in all above described papers. delivering higher perceptual quality and less bandwidth usage compared to the standard 802. delivering more than 1. In a wireless environment.11 wireless networks was presented. the value of packet error probability can aﬀect the overall system performance signiﬁcantly. However.11 retransmission scheme. the authors explored the possibilities of crosslayer design as a candidate tool for performance optimization. In [21].11 (RTS. In [10]. Diﬀerent QoS support in the MAC layer is implemented by using service diﬀerentiation and traﬃc class prioritization.11-based multi-hop wireless network which demonstrates the value of the proposed framework.

Suppose each transmitter has K (ﬁxed) information bits to be transmitted. The rest of this section is organized as follows. By using a random coding bound we can optimize the system performance over the code rate and energy per coded bit. In Subsection 3. In Subsection 3. 3.2.2. we introduce the generating function and determine the joint generating function for energy and delay. If the receiver decodes this packet correctly. We determine the generating function of the state diagram and use it to characterize the joint energy and delay distribution of the system incorporating physical layer and data link layer characteristics. we introduce the system assumptions and give a brief description for the protocol used in our analysis.2.11 standard except the transmission of data packets.1. This procedure will continue (using length Nl packets for the l-th transmission) until either the transmitter receives the ACK correctly or the . The contribution of this section is that we propose a state diagram to represent the operation of the system that allows us to analyze the energy and delay consumption of the system.2.2. In Subsection 3. we present numerical results to demonstrate the energy-delay tradeoﬀ with random coding.74 the above papers consider the energy and delay consumption jointly of transmitting a data packet by considering the eﬀects of physical layer and data link layer characteristics. the framework of our analysis can be used for other coding and modulation schemes and various wireless channel.1 System Description The operation of MAC protocol will be the same as 802. These K bits will be encoded to form a packet with length N1 and transmit. Otherwise. if decoding is not correct.3. after the expiration of a timeout the transmitter encodes the same number of information bits K to a packet of length N2 and transmits it again. it sends an ACK to the transmitter. Although we use random coding for an AWGN channel to represent packet error probabilities.

The state of the Markov chain represents the value of the backoﬀ counter and the contention window stage.3).2. The delay Td and Et of each data packet is deﬁned as before. We denote the error probability of an RTS packet as Pe.2. The ﬁrst case is that the RTS packet suﬀers a collision or packet error. we analyze the energy and delay characteristics of the wireless networks operated according to previous subsection. modulation etc. in this work.75 total number of attempts achieves a maximum allowed number. These probabilities depend on the particular channel. We begin by calculating some important event probabilities that are crucial in later analysis. 3. 22] in that packet error probabilities are included in the CW stage transition probability.ACK . Let pce represent the transition probability (as deﬁned in [5]) from one CW stage to the next. a CTS packet as Pe. We assume that the channel is memoryless between packets. 22]. (a speciﬁc example will be given in Section 3. The third case is . d. However. By modifying the Markov chain model described in [5.CT S and an ACK packet as Pe. where 1 ≤ i ≤ d.2 Analysis In this section. The probability Pe. we can incorporate the packet error probabilities into the analysis. The second case is that the RTS packet is correctly received and collision free but there is an error in the CTS packet transmission. an unsuccessful transmission can happen not only due to packet collision. This is also the probability of an unsuccessful transmission attempt seen by the transmitting station when its packet is being transmitted on the channel. There are four possible cases when the channel is sensed busy. The performance evaluation is based on the Markov chain model described in [5] and [22].RT S . but also due to packet errors.DTi represents the error probability of ith transmission of a data packet. Our model diﬀers from the model proposed in [5. channel coding.

ptx and pc . and the collision probability pc can be expressed as (see [5]) ptx = and pc = 1 − (1 − ptx )n−1 . Because the event of packet collision and the event of packet error are independent.17) (1 − Pe. there is no error in CTS packet.RT S )(1 − Pe.CT S ) i=1 d i−1 Pe. The last case is that the data packet is correctly received in one of these d transmissions but there is an error in the corresponding ACK packet. pc (3. The packet transmission probability ptx which is the probability of a transmitting station sending an RTS packet during each backoﬀ slot. pce can be expressed as pce = pc + (1 − pc )[Pe.19).DT0 = 1.DTi + (1 − Pe.CT S + d (1 − Pe.ACK ]. one can evaluate the probabilities pce .18) From the above three nonlinear equations. It is denoted by ptx1 and can be expressed as ptx1 = (n − 1)ptx (1 − ptx )n−2 .DTj (1 − Pe. .RT S )· (3.DTi ))Pe. Another important probability that will be used in our analysis later is the probability of a transmission of an RTS packet from exactly one of the remaining n − 1 stations given that at least one of the remaining stations is transmitting.20) We assume that stations spend energy only in packets transmission and that the energy consumption for each packet transmission is proportional to the packet length.CT S )( i=1 j=0 Pe. where pc is the packet collision probability and we set Pe.RT S + (1 − Pe. (3.76 that the RTS packet is correctly received and collision free.RT S )Pe. (3.17)–(3. but there is an error in all d transmissions of the DT packet.19) 2(1 − 2pce ) (1 − 2pce )(W + 1) + pce W (1 − (2pce )m ) (3.

9: State diagram for the protocol.2.0 G rfG bs. the state ﬂow diagram shown in Fig. let ERT S . in seconds. for a packet transmission with packet length N is N/Rt . We will now derive the joint generating . Finally.77 In particular. Similarly. Similarly. k-th DT and ACK packets. while dotted lines represent unsuccessful reservation or transmission. ECT S . Let TRT S . TCT S . CTS. let TDIF S and TSIF S be system delay parameters deﬁned by the standard.1.1 successful.2 G rfG bs.m G rs successful DT transmission ? G tf G bs. channel reservation ? G rfG bs. respectively. N0 ) is N Ec /N0 . channel reservation ? G tf G bs. let αEc be the energy required to transmit one coded bit. Let Rt be the transmitting rate in bits per second.m G ts G ts G ts Wait DIFS Figure 3. Then the energy received for a packet of length N (normalized to the thermal noise energy level. Based on the protocol description in Section 3. CTS.9 can be built. 3. where α > 1 is the inverse of the path loss between transmitter and receiver. TDTk and TACK denote the time duration for the transmission of RTS. The delay. channel reservation ? successful. EDTk and EACK denote the (received) energy for the transmission of RTS. Solid lines represent successful reservation or transmission. respectively. the transmission delay of each packet is also proportional to the packet length.m G rfG bs.2 G rs successful DT transmission ? G tf G bs.m successful.1 G rs successful DT transmission ? G tf G bs. k-th DT and ACK packets. Wait DIFS G bs.

22) We now develop expressions for the generating functions associated with failure to either reserve a channel or to transmit data.CT S ) X TRT S +TCT S +2TSIF S Y ERT S +ECT S .9. We ﬁrst describe the constant factors. . respectively.DTj )(1 − Pe. corresponding to failure to reserve the channel is given by Gf r (X. Y are the transform variables of the delay and energy.RT S )Pe. or a CTS transmission error. Y ) = (1 − pc )(1 − Pe. and a factor that depends on the contention stage i. the generating function Gts corresponding to successful transmission of a data packet can be expressed as d i−1 Gts (X. (3. (3.ACK ) Y Pi k=1 TDTk +TACK +iTSIF S +TDIF S EDTk +EACK . functions are of the form G(X.21) Similarly.78 function of the delay and energy random variables associated with the transitions in the state diagram in Fig. The generating function Grs corresponding to successful channel reservation is Grs (X. Y ) = [pc + (1 − pc )Pe.CT S X TRT S +TCT S +TSIF S +TDIF S Y ERT S +ECT S . In all the derivations below.RT S ]X TRT S +TDIF S Y ERT S + (1 − pc )(1 − Pe.23) the meaning of which is that a failure can be due to either a collision/RTS transmission error. (3.RT S )(1 − Pe. 3. Y ). these functions are products of two generating functions. Similarly. As can be seen from the state diagram. The generating function Gf r . where the variables X.DTi )(1 − Pe. Y ) = X Pi k=1 ( i=1 j=0 Pe. after reserving the channel. Each product contains a factor that is independent of the particular contention stage.

pc ). pc . At backoﬀ stage i. The event of each slot being sensed busy due to the transmission of other stations is independent for each slot and has the same probability.DTi )Pe. This generating function characterizes the delay for the transmitting station from the instant of starting the backoﬀ procedure to the instant that the backoﬀ counter reaches to zero at stage i.24) We now evaluate the generating functions denoted by Gbs.i (X) = j=1 1 iW 2 j k=0 j [(1 − pc )X σ ]j−k (pc Goc (X))k . We do not need to consider energy consumption here since the transmitting station stops any transmission during this period. (3. We deﬁne an occupied slot as a slot when the transmitting station senses the channel is busy due to the transmission of one of the remaining n − 1 stations.DTj (1 − Pe. then the number of the occupied slots in these j slots is binomially distributed with parameters (j.25) where Goc (X) is the generating function for an occupied slot. Hence.79 a failure to transmit a data packet is either due to the data packets transmission error. Using arguments similar to the .i .ACK · Y Pi k=1 Pi k=1 TDTk +TACK +iTSIF S +TDIF S EDTk +EACK ). k (3. the range of possible backoﬀ slots is from 1 to 2i W . Let j be the backoﬀ slots randomly chosen uniformly from the above range.i of the state diagram. of backoﬀ procedure at stage i is 2i W Gbs. or the acknowledgement packet transmission error.DTi )X d Pd k=1 TDTk +(d−1)TSIF S +TDIF S · Y X Pd i−1 k=1 EDTk + ( i=1 j=0 Pe. which is captured by the generating function Gtf as follows d Gtf (X. for each slot (from our previous assumptions). the generating function. Y ) = ( i=1 Pe. Gbs.

3 ∂Gs |X=Y =1 . Let Gs be the system generating function. the mean values are given by Td = 3.DTj · .CT S X TRT S +TCT S +TSIF S +TDIF S + d ptx1 (1 − Pe.11 standard) and N be the number of coded bits in a packet.28) Numerical Results We will use channel reliability based bounds for an AWGN channel to estimate the packet error probability. ∂X Et = ∂Gs |X=Y =1 .CT S )( i=1 Pe. Y ) = Grs Gts 1 − Gf r Gbs. (3.17).RT S )(1 − Pe. Y ) = Grs Gts + (Gf r + Grs Gtf )Gbs.2.m − Grs Gtf Gbs.RT S )Pe.80 ones used in the derivation of (3.RT S ]X TRT S +TDIF S + ptx1 (1 − Pe.RT S )(1 − Pe.27) where i is the index of the CW stage from 1 to m. we can express Goc (X) as follows Goc (X) = [(1 − ptx1 ) + ptx1 Pe.0 f0 . Y ) = Gbs. Let K be the number of information bits in a packet (speciﬁed by the 802. we can evaluate any joint statistics of the energy and delay of a successful packet transmission. we obtain the following backward recursive generating function to characterize the energy and delay of the system.DTi )X Pi k=1 TDTk +TACK +(i+2)TSIF S +TDIF S From the state diagram and Mason’s gain formula.i fi Gs (X. Hence. (3. Then the recursion is fm (X.m fi−1 (X. Then .26) (1 − Pe.CT S ) i=1 j=0 Pe.DTi )· + X TRT S +TCT S + Pd k=1 TDTk +(d+1)TSIF S +TDIF S d i−1 ptx1 (1 − Pe. ∂Y (3. In particular.

For both protocols there is an optimal N for each kind of packets. Using these values. 3. we get the energy-delay curves.28) to get the correspond∗ ∗ ∗ ing optimal packet lengths NRT S . NCT S . N ∗DT and NACK (note that in the case of (3. if N is large.2 summarizes the system parameters used in our numerical evaluations. We ﬁrst ﬁx Ec /N0 and minimize T d from (3.N ≤ 2K−N R0 . NDT 1 . which is essentially an SWARQ-FT protocol with d = 1. we also present similar curves for the original 802.2: System Parameters for Numerical Results KRT S 128 bits KCT S 128 bits KACK 128 bits Channel Bit Rate. and NDT 2 ). We observe that the energydelay curves for two protocols are almost identical. For comparison. On the other hand. Repeating the above procedure for diﬀerent Ec /N0 .N bounded as Pe.10 the energy-delay curves for the original protocol and the IEEE 802.K. where R0 = 1 − log2 (1 + e−Ec /N0 ) is the cutoﬀ rate determined by SNR.K.11 protocol.11 protocol with SWARQ-FT (d = 2) is evaluated for diﬀerent number of users and data packet lengths. we can evaluate both average delay and average energy consumption.11 MAC Protocol with SWARQ-FT (d = 2). σ 50µs TSIF S 28µs TDIF S 128µs there exist an encoder and decoder with packet error probability Pe. This shows that if we can select . Our goal is to ﬁnd the best N for each kind of packet to minimize system delay. which increases the system delay and energy due to the high chance of packet retransmission. For small N the packet error probability is large.81 Table 3. The energy-delay curves of the IEEE 802.29) d = 2 we need to optimize over two data packet lengths. Rt 1 Mbit/s Slot Time. We ﬁrst plot in Fig. the packet error probability is small but the transmission time is large. Table 3.

3.11 protocol with SWARQFT. is compensated by the increased delay needed to access the channel. This ﬁgure shows that the optimal packet lengths are on the order of 60 bits larger than the minimum length implied by the cutoﬀ rate. We observe that both systems are equally and very sensitive to SNR. 3.3 Conclusion In this chapter. For high SNR values as small as 10 will provide almost optimal performance.11 and IEEE 802.11. While ARQ provides some robustness to SNR uncertainty in conventional single user system. the number of redundant bits has to increase. Essentially. while for low SNR. the energy-delay tradeoﬀ is plotted for the SWARQ protocol for different values of quantity NDT − KDT /R0 (Ec /N0 ). the SWARQ-FT does not give signiﬁcant improvement. the reduction in delay the SWARQ achieves by immediately retransmitting an erroneously received packet. In Fig. we ﬁx the packet lengths.11 with SWARQ-FT) are sensitive to the knowledge at the transmitter of the . The second proposed protocol is to incorporate the IEEE 802. this is not the case in this multiuser scenario. Our results demonstrated that both protocols (original IEEE 802. To compare these two protocols with respect to their robustness to SNR.82 packet lengths properly.11 standard. we applied our generating function method to characterize the energy and delay of a wireless network for two proposed protocols modiﬁed from 802. The numerical results indicated that both throughput and energy eﬃciency are improved signiﬁcantly when the channel SNR is low. since other users use the same protocol. The ﬁrst proposed protocol is related to the adaptive energy scheme which is implemented by increasing the energy level per coded bit due to previous failure transmission attempts. vary Ec /N0 and plot energy and delay for both protocols.

Our methods can be extended to the analysis of other protocols or to consider a wireless network system where more function layers are involved by representing the operation of a protocol with a system state diagram. .5 Et/KDT (dB) 5 4. We expect similar results when LDPC or turbo codes are used. the generating function approach allows one to ﬁnd the entire system performance distribution based on the performance distribution of its elements by using algebraic procedures in a systematic way. Dashed lines represent the energy-delay curves for Ec /N0 of 0 and 3 dB using the optimal packet lengths. However. etc.5 4 10 3. one may select other variables in the generating function for their analysis purpose. such as event’s cost. In this thesis. Once the system state diagram is obtained. event’s reliability.5 3 15 20 25 30 35 Td/(KDTN0) (µsec) 40 45 50 Figure 3.5 6 5. Solid lines represent the curves after packets optimization. This is a consequence of the steep falloﬀ of packet error probability with respect to SNR for random coding. 7 6. we select energy and delay consumption as our interesting variables since we want to determine the energy and delay tradeoﬀ.83 received SNR.10: Energy-delay curves for n = 10 and KDT = 1400. Therefore. we shall be able to derive the joint generating function of interesting random variables for the state diagram.

The number beside the curve is the redundant bits. .84 9 210 8 110 60 7 N ) (dB) E /(K t DT 0 6 5 10 4 Optimal SWARQ 3 15 20 25 30 35 40 Td/KDT (µ sec) 45 50 55 60 Figure 3.11: The dashed lines represent SWARQ after optimization.

In SW-ARQ. Thus the code rate must be low enough for the worst-case channel. Therefore.CHAPTER IV Analysis of Energy and Delay for ARQ Systems over Time Varying Channels 4. ARQ protocols are classiﬁed into three basic schemes: stop-and-wait (SW). the disadvantage of a conventional FEC system without feedback is that the transmitter does not adjust the error correction capability according to the channel variations. This scheme preserves the order of packets but results in low channel utilization if the round85 . ARQ error control systems are more suitable than FEC systems for error control in data communication systems having time varying channels where a feedback channel is available [35. The advantages of both FEC and ARQ techniques can be obtained by combining the techniques within a single hybrid-ARQ protocol [34]. Although FEC can provide a communication system with a ﬁxed throughput which is equal to the code rate. go-back-N (GBN). the transmitter must receive the ACK of a packet before transmitting the next packet. An ARQ system changes the error correction capability to adapt to the channel variation by using a feedback channel. In general. 47]. and selective-repeat (SR).1 Introduction Error control techniques can generally be classiﬁed either as forward-error-correction (FEC) or as automatic-repeat-request (ARQ).

Similarly. Upto to N subsequent packets can be transmitted before receiving an ACK/NACK for a given packet. If we want to decrease the delay required to transmit the same information packet. a system with lower code rate needs longer delay to transmit the same information packet due to the ﬁxed bandwidth. packets are transmitted continuously without waiting for ACKs/NACKs. On the other hand. a system with high code rate requires higher energy per information bit to achieve the same reliability as a system with lower code rate. For a ﬁxed amount of energy used per information bit. the transmitter retransmits the negatively acknowledged packet and all subsequent packets regardless of their acknowledgments. diﬀerent block (packet) sizes and multicopy transmission schemes were used as adaptation mechanisms. 37. packets are transmitted continuously as in GBN. in [44. we may use more attempts (longer delay) to transmit the same information packet successfully and use smaller energy. In [9. 50]. 26. If a NACK is received. The round trip delay is assumed to be smaller than N times the packet duration and the transmitter is capable of storing N packets. a wireless communication system adopting FEC can provide higher protection by using a lower code rate. Many adaptive ARQ protocols have been suggested in the literature to improve the system delay (or equivalently throughput). the authors suggested to vary the . Of the three schemes. there is a fundamental tradeoﬀ between energy and delay in a wireless communication system. but only negatively acknowledged packets are retransmitted. On the other hand. Therefore. In SR-ARQ. 40]. In GBN-ARQ. respectively. However. SR-ARQ achieves the highest throughput but requires the most feedback.86 trip delay is large. an ARQ system requires higher energy to transmit an information packet successfully with fewer trials (shorter delay to transmit a packet).

The joint generating function enables us to characterize the joint energy and delay distribution of the system incorporating physical layer characteristics (packet error probability as a function of energy and delay). their work does not provide the exact method to evaluate the system delay. By extending the traditional Markov chain [13] that models the channel as well as the transmitter. the overall system operation can be represented by a state diagram that takes into account the joint operation of the transmitter. fast Rayleigh fading and log-normally shadowed Rayleigh fading. we propose an analytical method to determine the joint distribution of the energy and delay of automatic repeat request (ARQ) schemes over time varying channels.87 FEC code rate to compensate for the variations in the channel conditions. receiver and channel. the authors investigated the energy savings of an ARQ strategy within two diﬀerent time varying channels. the time-varying characteristics of the channel have a great inﬂuence on system performance especially at . The eﬀect of packet length on the channel transition probability is also investigated. In this chapter. Another important issue that previous research has not addressed is the relation of the channel transition probability with the transmission delay of a packet. However. We determine the joint generating function of the energy consumed and delay incurred of an ARQ system from its state diagram. As the numerical results demonstrate. they do not consider the energy consumption as part of the system performance. However. In [38]. Most previous papers just assume that the channel transition probability is the same for diﬀerent packet lengths. The model of the transmitter accounts for diﬀerent packet lengths used by the transmitter and the model of the receiver accounts for the receiver memory [27]. Although the above papers consider the analysis based on time varying channels. this assumption is not reasonable if the variation of packet lengths is large.

transitions and actions. The rest of the chapter is organized as follows.4.6.2. A transition indicates a state change and is described by a condition that would need to be fulﬁlled . In Section 4. we model an ARQ system with a ﬁnite state machine (FSM) by considering the transmitter. it reﬂects the input changes from the system start to the present moment.4. Channel Model and Assumptions for ARQ Systems A ﬁnite state machine (FSM) or ﬁnite automaton is a model of a system which is composed of states.2 FSM Model. study of computation and languages. Originally deﬁned in the automata theory. Several practical ARQ systems with speciﬁc packet lengths are considered in Section 4. Finite state machines are very widely used in modeling of application behavior. A state stores a cumulative information about the past. go-back-N ARQ (GBN-ARQ).5 and its joint generating function for energy and delay over time varying channels will be derived.2 will be presented in the appendices. Another kind of ARQ system. FSM’s are used in the theory of computation [12].1 and 4.4. receiver and channel states jointly. The detailed derivation of the generating functions for systems discussed in Subsection 4.e. In Section 4. A general ARQ system which allows diﬀerent packet lengths in each retransmission is discussed and analyzed in Section 4. 4. The content of a state depends on the characteristics and behaviors of the system that we are interested in. software engineering.88 low channel SNR. design of hardware digital systems. i. will be investigated in Section 4.8. the cutoﬀ rate for two kinds of receiver structures that contain memory is derived.3. The conclusions are given in Section 4.7. In Section 4. we present numerical results for the performance by considering the energy and delay relationship with random coding for Gilbert-Elliott channels.

The states of the transmitter can be used to model the diﬀerent coding or modulation schemes and the states of the receiver can be utilized to model the receiver memory content or diﬀerent decoding schemes. • S is a ﬁnite non empty set of states.. the next state. the total energy and delay generating function from the initial state to the ﬁnal state can be obtained from Mason’s gain formula. the overall ARQ system operation is represented by a modiﬁed FSM that also takes into account the states of the transmitter and the receiver. . S. By using the concept of signal ﬂow proposed by Mason [20]. An action is a description of an activity that is to be performed at a given state. By extending the traditional Markov chain that models the channel state.e. • Γ is the alphabet of the output variables. • ω is the output function ω : S × Σ → Γ. • s0 is an initial state. For our communication system model. i. the state for the transmitter. A ﬁnite state machine is a sextuple < Σ. Γ. channel and receiver we be consider jointly in the system state space S. Also for our purpose the output of modiﬁed transition functions δ. δ. To determine the energy and delay generating function of the system.89 to enable the transition. the output at each state is used to characterize the energy and delay consumed at the state. • δ is the state transition function: δ : S × Σ → S. is a random state with some probability distribution due to the uncertainty of the channel characteristic. ω >. where: • Σ is the alphabet of the input variables. s0 .

Whenever there is an information packet with length K generated. λ. λ+µ The Gilbert-Elliott channel model can be speciﬁed completely if the following four parameters are given : (r. In each channel state. we can determine the average channel SNR (SNRavg ). it will be encoded to a 1 The SNR is the power ratio between a signal (meaningful information) and the background noise. It uses two states to characterize the channel condition. SNRavg = µSNRG +λSNRB . If the time duration for a packet transmission is T∗ where the subscript is used to distinguish diﬀerent packet types. . From the above system parameters. i.1 General ARQ Systems A general stop and wait (SW) ARQ protocol is described as follows. the Gilbert-Elliott channel model is widely used [26].. the channel transition probability from bad state to bad (good) state is e−µT∗ (1 − e−µT∗ ). i. we model the channel as AWGN channel and assume that the channel signal-to-noise ratio (SNR)1 of good state is SNRG and the channel SNR of bad state is SNRB . We assume that the duration of the good and bad B states are exponentially distributed with mean 1/λ and 1/µ. Similarly. µ).. We deﬁne r as the ratio of the channel SNR between good SNR and bad states.e. The channel model is shown in the Fig.e. The crossover probabilities between channel states are determined by the time duration of occupying the current state before the event of next channel transition and the channel transition rates. SNRavg . the state for good channel condition is designated by G and the state for bad channel condition is designated by B. 4. The value λ is the transition rate from good to bad channel condition and µ is the transition rate from bad to good channel condition. r = SNRG . respectively.90 Gilbert-Elliott Channel Model To model time-varying channels. then the channel transition probability from good state to good (bad) state is e−λT∗ (1 − e−λT∗ ).

The demodulator of the receiver is implemented by a maximum a posterior probability (MAP) criterion in determining the transmitted signal.91 Good state (High SNR) Bad state (Low SNR) Figure 4. If it is de- . After receiving this packet with length N2 . the receiver will send an ACK packet back to the transmitter. We use the symbol Tb to represent the time duration needed to transmit a coded bit and the symbol Ec to indicate the energy consumption required to transmit a coded bit. We will assume binary phase-shift keying (BPSK) modulation scheme is used by the transmitter. it is propagated through the channel. The modulator maps the input sequence of bits to a signal waveform. After the modulated waveform is generated. The coded bits are passed to the modulator. If the data packet is decoded correctly at the receiver. data packet with length N1 . the data packet will be decoded. If the data packet is decoded correctly at the receiver. After demodulation. If it is decoded incorrectly at the receiver. a NACK packet will be sent back to the transmitter and the received signal samples corresponding to the erroneously decoded packet will be stored in the receiver memory. Then the transmitter will encode the same information packet again to a data packet with length N2 and transmit it.1: Gilbert-Elliott channel model for time varying channel. the receiver will send an ACK packet back to the transmitter. the receiver will decode this new packet jointly with those packets stored in the receiver memory.

· · · }. N3 . (3) delay is incurred only during data packet transmission. N2 . the receiver may update the receiver memory content by storing the newest erroneous packet and discarding the oldest erroneous packet. (5) the Gilbert-Elliott channel model will be adopted. We make following assumptions about our system: (1) The ACK and NACK packets are error free. If we use notation N to represent a set of natural numbers. If the number of consecutive packet errors is greater then mR . Let mR be the maximum number of packets that the receiver can store in its memory. For example. the sequence of packet lengths in transmission given an information packet has the following form: {N1 . the transmitter will encode the same information packet with length N3 and retransmit. it is straightforward to consider the cases with ACK/NACK packet error and the round-trip delay under our analytical framework. then the set of {N1 . therefore. (4) energy is consumed only during data packet transmission. the receiver memory will take some action to keep the number of stored packets smaller or equal than mR . · · · } can be written as {Ni |i ∈ N}. . the transmission energy of ACK and NACK packets are negligible. After receiving the NACK packet. Because we allow diﬀerent packet lengths used in each retransmission. N2 . a NACK packet will be sent back to the transmitter and the new erroneously decoded packet (N2 ) will be stored in the receiver memory. N3 . (6) the event of packet errors is independent of the event of channel transition.92 coded incorrectly at the receiver. The above procedure will be repeated until an ACK is received by the transmitter. When an ACK is received by the transmitter. a new coded packet is transmitted with length N1 . However. (2) the round-trip delay and the transmission time of the ACK and NACK packets are negligible.

g. The next pair contains the channel state and packet length of the previous erroneously received packet. B} since the Gilbert-Elliott model is adopted.. After receiving a packet with length N4 trans- . Σ = {ACK. B. Γ = {X Ni Tb Y Ni Ec |i ∈ N}.e.e.. The channel state space for one packet transmission is denoted as SC and SC = {G.. For example. If a memory location does not contain a packet. G. The state space for the receiver is denoted by SR which is a collection of vectors of length 2mR . i. then we represent that by a packet length N0 . i. i. For the notational clarity. • (State space) Because the transmitter can use diﬀerent packet lengths for each retransmission. the corresponding elements of the FSM for a generalized ARQ system are listed as follows : • (Input variables) Since the system will change its state when an ACK or NACK packet is received by the transmitter. NACK}. The ﬁrst pair contains the channel state and packet length of the most recently received erroneous packet. at startup. the state of the receiver for memory with mR = 3 might be [G.3 Generating Function Analysis for General ARQ Systems We will derive the generating function of energy and delay consumption for a general ARQ system in this section.e. N2 . The state space for the receiver represents the lengths of the packets stored in the memory and the corresponding channel condition. we use ST to represent the state space for the transmitter. we use Σ to represent the input space.. N1 ]. N3 . the symbol Γ is used to represent the output space.93 4. • (Output variables) For the requirements of the generating function approach for the energy and delay consumption between each state transition (as shown in Chapter 2). ST = {Ni |i ∈ N}. e.

then the the corresponding state space for the packet lengths is [N3 .94 mitted in the bad channel condition. the total system state can also be expressed as [Gi . the transmitter.4. The ﬁrst component of this vector represents the channel state and the second component of this vector represents the transmitter state. Therefore. A state of the system. N3 . we show an example about the transition of states for a system with mR = 2 with the sliding window operation in the receiver memory. B2 . N4 . The remaining 2mR components represent the receiver memory state. sR ]) for the case that the transmitter sends a packet with length Ni in good (bad) channel condition given that the receiver state is sR . In order to consider all possibil- . In Fig. N3 . For notational simplicity. 4. B. G1 ]. N2 .2. G. N2 . sT . G. The ﬁrst component of SR contains the packet length of the most recently received erroneous packet. B. if the state space for the receiver is [G. denoted as s. N3 . and we assume that there are a packet lengths in the vector sR . the total state space of the system which contains the information about the channel. The detail about the sliding window operation in the receiver memory will be presented in subsection 4. Given an element from SR . say sR . N1 ]. N1 ]. the next state of the receiver memory will be [B. We will use use a shorthand notation for the state of the receiver memory by representing Gi as a packet of length Ni received when this packet is transmitted in the good channel condition. sR ]. For later derivation convenience. So the receiver state [G. can be written as s = [sC .1. and the receiver is denoted as vectors of length 2(mR + 1). The next component of SR contains the packet lengths of the previous erroneously received packets. sR ] ([Bi . N2 ]. G. For example. N1 ] can be written compactly as [G3 . B. N2 . the state space of packet lengths for the receiver is denoted by SR which is a collection of vectors of length mR .

However. sR (2) = [G3 . the index l for 1 ≤ l ≤ 2a . with N2 in G channel GN2 GN1 NACK 4-th TX. For example. [B3 . The channel realization index l is assigned to the set of all channel realizations of these a packets according to the lexicographically order by assuming G > B and the position order of the receiver memory. By the above ordering regulation. we have sR (1) = [G3 . The symbol sR (l) represents a state of the receiver memory that the state of packet lengths for the receiver is sR and the channel realization index is l. N3 ]. The second (third) slot is used to represent the ﬁrst (second) position of the the receiver memory. G3 ] and sR (4) = [B3 . with N4 in B channel BN4 GN3 GN2 NACK 5-th TX.95 1-st TX. called channel realization index. G3 ]. B3 ]. all memory states associated with sR (l) have the same number of packets and packet length indices stored in the receiver memory for each channel realization index l.2: An example for a general ARQ system with mR = 2. [B3 . if mR = a = 2 and sR = [N3 . Each element of SR may diﬀer in the number of packets stored in the receiver memory. [G3 . with N3 in G channel GN3 GN2 GN1 NACK After receiving ACK. B3 ]. B3 ]}. sR (3) = [B3 . G3 ]. with N5 in B channel . The second and third slots represent the state of the receiver memory. is used to distinguish diﬀerent memory states introduced by the randomness of the channel. one is sR and the other is sR (l). the 1-st TX begin with N1 BN5 BN4 GN3 Figure 4. G3 ]. B3 ]. The ﬁrst slot represents the state of the channel and transmitter. with N1 in G channel 3-rd TX. ities for the receiver memory states that have these a packet lengths. Φ Φ Φ GN1 NACK 2-nd TX. We have two ways to represent the memory content. then all possible states of the memory are {[G3 .

Y ) = P r(event)X Tevent Y Eevent . sk ]. sR ]. For the energy and delay analysis of ARQ systems. then the state of the system sk will be mapped into the state sk+1 by the transition function δ as δ(sk . sk+1 ] and C T R C T R δ([sk . The k-th transmission is counted from the transmission of the original K information bits with packet length N1 (1-st transmission). If we are interested in evaluating the generating function for a set of disjoint events. then sini = G1 if the initial channel condition is good and sini = B1 if the initial channel condition is bad. sini . • (Output functions) The output function will map a system with state sk to X Nk Tb Y Nk Tc since the packet length used in the k-th transmission is Nk . sT . NACK) → [sk+1 . sk+1 . Recall that an event is a set of random outcomes to which a probability is assigned. The state transition function δ will be C T R C T R determined by speciﬁed protocol and this is often represented by a state diagram. we then deﬁne the joint generating function of an event as: Gevent (X. Since s = [sC . sini ]. This is a state used at the ﬁrst time transmission of an information packet. It will map an element of space S × Σ to an element of space S. sk . δ. • (State transition function) The state transition function.96 • (Initial state) The initial state is represented by the symbol sini . we also have δ([sk . ACK) → sini . sk . then the generating . will deﬁne the transition between states for diﬀerent input. ACK) → [sini .1) where Tevent (Eevent ) is the required delay (energy) for this event. (4. then the state of sk will be mapped into the state of sini by function δ as δ(sk . If the input is a NACK after the k-th transmission. If the memory content is empty at the ﬁrst transmission. NACK) → sk+1 . sk ]. If the input is an ACK.

Gi |si to represent the event that the transmitter receives a R NACK (failure) after the i-th transmission of a given packet in the good channel condition given that the receiver memory state is si . Y ) = R si−1 ∈SR R i−1 Gf.Bi |si ) is used to represent the packet error probability R R for the i-th transmission in the good (bad) channel condition given that the memory state is si . Similarly.Gi |si (X. sR ]. We use symbol Ef. Because the i-th (for i ≥ 2) unsuccessful transmission is preceded by (i − 1) unsuccessful transmissions. Y )e−λNi−1 Tb I(δ([Gi−1 .1). sR ]. si ] when this event happens because the packet R length used in the i-th transmission is Ni .Bi |si is used to represent the R R generating function for the event that the transmitter receives a NACK after the i-th transmission in bad channel condition given that the receiver memory state is si .Gi |si X Ni Tb Y Ni Ec R . (4. We will use the shorthand notation. α = 1 − α.Gi R (Pe. Y )e−µNi−1 Tb I(δ([Bi−1 . si ]) · R R i−1 Gf. NACK) = [Gi . NACK) = [Gi .Bi |si for ARQ R R systems can be expressed as Gf.Gi−1 |si−1 (X. The generating function of this event is represented by Gf. for the following derivation. Gf. R The probability Pe.97 function for all events in this group has the following form Ggroup (X.2) where i is the index for events in this group and Teventi (Eeventi ) is the required delay (energy) for this i-th event. If the memory content is empty at i-th transmission. the recursive formulas for Gf. Note that the state space of R the system can be expressed as [Gi . the probability Pe. Y ) = i∈group P r(eventi )X Teventi Y Eeventi .Gi |si and Gf.Bi−1 |si−1 (X.Bi ) is used to represent the packet error probability for the i-th transmission in the good (bad) channel condition. si ]) + R R Pe.Gi |si according to (4.Gi |si (Pe.

Bi |si (X.Bi−1 |si−1 (X.Gi |si e−λNi Tb Pe. NACK) = si ) is an indicator function. sR ]. si+1 ]) + R R R Gf. sR ]. si ].. NACK) = [Bi+1 .98 Gf. NACK) = si ) = 0 if the inside condition is false. Y ) = Pe. sR ]. For the ij -th transmission. si ]. Y )e−µNi−1 Tb I(δ([Bi−1 . the packet lengths used . NACK) = [Bi+1 .3) and Gf. NACK) = [Gi+1 . NACK) = [Bi . si+1 ]) + R R R R Gf. si ].Bi+1 |si+1 I(δ([Bi .Gi+1 |si+1 I(δ([Bi .Bi |si e−µNi Tb Pe. By assuming that the system begins with the state space [G1 ] or [B1 ] (the receiver memory content is empty) and the channel has reached a stationary condition.Gi |si e−λNi Tb Pe.Bi+1 |si+1 I(δ([Gi . λ+µ e. si ]) + R R i−1 Gf. Y ) = R si−1 ∈SR R i−1 Gf. NACK) = [Gi+1 .G1 X N1 Tb Y N1 Ec + Pe. si+1 ]) + R R R R Gf.4) where the index i represents the event that the system fails at the i-th transmission but succeeds in the (i + 1)-th transmission.Bi |si e−µNi Tb Pe.G1 (4. (4.G1 |s1 = R µ P X N1 Tb Y N1 Ec λ+µ e.Gi−1 |si−1 (X. Y )e−λNi−1 Tb I(δ([Gi−1 . NACK) = [Bi . si ]. NACK) = si ) is 1 if the inside condition is true and I(δ(si−1 . we set Gf.B1 X N1 Tb Y N1 Ec + λ+µ λ+µ R ∞ i=1 si ∈SR si+1 ∈SR R R Gf. s = [sT .Bi |si X Ni Tb Y Ni Ec R where I(δ(si−1 .B1 Since the (i + 1)-th successful transmission is preceded by i unsuccessful transmissions and a successful transmission at the (i + 1)-th trial. si ]) · R R Pe.Gi+1 |si+1 I(δ([Gi .B1 |s1 = R λ P X N1 Tb Y N1 Ec . si+1 ]) · R R R R X Ni+1 Tb Y Ni+1 Ec . the total generating function for a successful transmission can be written as µ λ Gs (X. i. The function I(δ(si−1 . The symbol s is used to the set of packet lengths transmitted and stored in the receiver.e.

1. −2. the number of mj must be ﬁnite. sT and sR . We have the following theorem about our general ARQ systems for the set of all transmission indices N. We will group i all transmission with transmission indices with form as (ij + kmj ) for k ∈ Z = {· · · . We cannot require that the system states are always diﬀerent after (2(d + 1))mR +1 failures by the Pigeonhole Principle. Therefore. By examining the states of packet lengths for all transmission indices from sini = s1 . The transmission indices in the same group have same packet length indices set s and each group has same period m. Because the state si1 +1+m1 will be equal to the state si1 +1 since they are the outputs of the . 2. if the state si1 +m1 is the ﬁrst repeating state of any one state for packet lengths state in previous (i1 + m1 − 1) transmissions. Proof : Let d be the number of diﬀerent packet lengths used in the system. Let mj be the mini- i i mum number (if it exists) such that sTj i +mj j = sTj and sR i i +mj j = sR . · · · } into a group and denote this group as gj . which is a ﬁnite integer. If a set C is the union of two sets A and B and A B = φ. The set H contains those indices which are not belonged to any group gj . then the set of T I.99 j for the transmitter state is sTj and the receiver state is sR . then sk = δ k (s0 . We claim that the state for packet lengths s is the same for those indices (i1 + km1 ) for k ∈ {0} N. Theorem: If the number of diﬀerent packet lengths used in the transmitter and the receiver memory size are ﬁnite. 0. denoted as T I. say si1 . Suppose the system begins with the initial state sini ≡ s0 . then the total possible system states for the system is less then (2(d + 1))mR +1 . NACK) represents the system state after k failures for k ∈ N. The number mj is called the transmission period of this group. −1. we then write C = A B. can be classiﬁed as T I = H m j=1 gj according to the state of packet lengths.

· · · .100 same inputs (si1 +m1 = si1 ). (i1 − 1)} from above construction since the indices {1. · · · }since . the state s(i1 +1) is diﬀerent from si1 since there are no repeating states in {s(i1 +1) . · · · . 2. · · · . then si1 +k m1 = si1 +m1 +k m1 . si1 +2 . Therefore. the set of T I will become H there is only one group. the state si1 +n will be equal to the state si1 +m1 +n for any positive integer n. we group the transmission indices ((i1 + j − 1) + km1 ) as group gj . If m1 = 1. for each j. g1 and g1 = {i1 . Then the system state s(i1 +1)+m1 will be equal to the system state s(i1 +1) since they are the outputs of the same inputs (si1 +m1 = si1 ) by state transition function δ. We can ﬁnd that the set of T I = {1. · · · . · · · } = H m j=1 gj . The elements in group H will be {1. if we set n = k m1 (k ∈ N). the state for packet lengths s is the same for those indices (i1 + km1 ) for k ∈ {0} N. m1 }. m1 }. we know that the states for indices ((i1 + j − 1) + km1 ) for k ∈ {0} N are same from mathematical induction argument shown in the previous paragraph. Again. · · · . 3. Therefore. If m1 = 1. hence. It is easy to see that gk1 gk2 = {((i1 + k1 − 1) + km1 )} {((i1 + k2 − 1) + km1 )} = φ if k1 = k1 . by mathematical induction. 2. si1 +m1 −1 } for si1 . the state with transmission indices i1 + km1 = i1 + k will be same. there are m1 groups since k1 ∈ {1. for each j ∈ {1. the state for packet lengths s is the same for those indices ((i1 + 1) + km1 ) for k ∈ {0} N. Therefore. From the mathematical induction. Therefore. 2. i1 + 1. 2. 2. the state s(i1 +1)+n will be equal to the state s(i1 +1)+m1 +n for any positive integer n. i1 + 2. The states for s(i1 +j−1) are diﬀerent from each other for 1 ≤ j ≤ m1 since m1 is the period of the group g1 and. In this case. And we set m1 = m. (i1 − 1)} do not belong to any groups in gk for 1 ≤ k ≤ m.

BT (ij ) |eR. Similarly. Then. The symbol sR.gj (lj ) is used to represent s s the group of the events that the transmitter receives a NACK after the (ij + km)th transmission in the bad channel state given that the receiver memory state is j sR. then the general formula of total generating function shown in (4.GT (ij ) |eR. Let Pf. which assigns the p-th transmission to q = T (p) where q is the index for packet length used in the p-th transmission.BT (i i +qm j) |eR. we have ∞ i +qm j) Gf.GT (i |eR.gj (lj ) .gj s (lj ) X Pij [( k=1 NT (k) )+q( ij +m P p=ij +1 NT (p) )]Tb × [( Pij k=1 Y NT (k) )+q( ij +m P p=ij +1 NT (p) )]Ec .gj (lj ) ) s be the probability of the event that the transmitter can not receive an ACK packet after the (ij + qm)-th transmission with good (bad) channel condition given that the receiver memory state is sR.gj (lj ) = s q=0 j Pf. We deﬁne a packet length index assignment function.GT (ij ) |eR.gj (lj ).gj j j (Pf.GT (i i +qm (l ) |e s j ) R. The generating function of this group is denoted as Gf. Because the generating function for those unsuccessful transmissions with transmission indices in group gj can be evaluated systematically with closed form. Then T (ij ) will indicate the packet length used for transmissions in group gj . T .101 If we consider a general ARQ system with ﬁnite number of diﬀerent packet lengths. The function T is determined by the state transition function δ.gj (lj ).gj (lj ).gj (lj ) (lj is the channel realization index) is used to represent a state of the receiver memory in group gj . Ef. We use symbol Ef.gj (lj ) to represent the group of events that the transmits ter receives a NACK after the (ij + km)-th transmission in the good channel state given that the receiver memory state is sR.GT (ij ) |eR.4) can be expressed in a more structured way due to the previous theorem.

102

ij +m P p=ij +1

∞

Gf,BT (ij ) |eR,gj (lj ) = s

q=0

j Pf,BT (i

i +qm

j)

[( |eR,gj s (lj ) X

Pij

k=1

NT (k) )+q(

NT (p) )]Tb

×

[(

Pij

k=1

Y

NT (k) )+q(

ij +m P

p=ij +1

NT (p) )]Ec

.

(4.5)

We should note that it is possible that the packet length used in the k1 -th transmission may equal to the packet used in the k2 -th transmission for k1 = k2 .

j We have the following recursion formulas for probabilities Pf,GT (i j Pf,BT (i

i +qm

j)

|eR,gj (lj ) s

and

i +qm

j)

|eR,gj (lj ) s

**from the state transition diagram speciﬁed by any protocol, ij +qm Pf,G s T (ij ) |eR,gj (2) . . . P ij +qm m s f,GT (ij ) |eR,gj (2 R,gj ) P ij +qm f,BT (ij ) |eR,gj (1) s ij +qm Pf,B s T (ij ) |eR,gj (2) . . . ij +qm P mR,g j
**

f,BT (ij ) |eR,gj (2 s ) ij +qm Pf,GT (i ) |eR,g (1) s j j

ij +(q+1)m Pf,G s T (ij ) |eR,gj (2) . . . ij +(q+1)m P f,GT (i ) |eR,gj (2mR,gj ) s j P ij +(q+1)m f,BT (ij ) |eR,gj (1) s ij +(q+1)m Pf,B s T (ij ) |eR,gj (2) . . . ij +(q+1)m P mR,g j

f,BT (ij ) |eR,gj (2 s ) ij +(q+1)m Pf,GT (i ) |eR,g (1) s j j

, (4.6)

gj cG(k)G(l) = gj cG(k)B(l)

j cB(k)G(l) j cB(k)B(l)

g

g

**where the symbol mR,gj represents the number of packets stored in the receiver
**

j memory for transmissions in group gj . The components of matrix cG(k)G(l) are the

g

transition probabilities for the system which suﬀers an unsuccessful transmission at the (ij + qm)-th transmission with good channel condition to the system which has an unsuccessful transmission at the (ij +(q +1)m)-th transmission with good channel

j condition. The component for the position (k, l) of matrix cG(k)G(l) is obtained by

g

considering the m consecutive packet errors and channel transitions from the system

103

**state [GT (ij ) , sR,gj (k)] to [GT (ij ) , sR,gj (l)]. Similar explanations could be applied for
**

j j j components in matrices cG(k)B(l) , cB(k)G(l) and cB(k)B(l) . j The value of cG(k)G(l) is derived as follows

g

g

g

g

(m−1)

j cG(k)G(l) =

fold ···

s(ij +1) ∈S

m−2

g

(

k=1

tsij +k )ts T (i [G s

ij +k+1

ij +1

[GT (ij ) ,eR,gj (l)] s ,eR,gj (k)] tsij +m−1 s j)

,

s(ij +m−1) ∈S s(ij +m−2) ∈S

**(4.7) where the summation symbol
**

sp ∈S

for p ∈ N denotes the sum over all possible sysp+1

tem states at the p-th transmission. The symbol tsp s

for p ∈ N denotes the state

transition probability from a state, sp = [sp , sp , sp ], at the p-th transmission to a C T R state, sp+1 = [sp+1 , sp+1 , sp+1 ], at the (p + 1)-th transmission. It is given by C T R −λNT (p) Tb (p+1) e = G, δ(sp , NACK) = s(p+1) Pe,GT (p) |sp , if sp = G, sC C R −λNT (p) Tb (p+1) e Pe,BT (p) |sp , if sp = G, sC = B, δ(sp , NACK) = s(p+1) C R (p+1) (p+1) −µNT (p) Tb tsp = Pe,GT (p) |sp , if sp = B, sC = G, δ(sp , NACK) = s(p+1) , s C R e −µN T (p+1) p T (p) b P e = B, δ(sp , NACK) = s(p+1) e,BT (p) |sp , if sC = B, sC R 0, otherwise (4.8) where sp and sC C

(p+1)

**are the channel condtions for the p and (p + 1)-th transmissions.
**

g g g

j j j The components in matrix cG(k)B(l) , cB(k)G(l) and cB(k)B(l) can be derived simi-

**larly as shown in (4.7) by just changing the channel condtions at the ij -th and (ij +m)th transmissions. If we multiply both sides of (4.6) by X
**

Pij [( k=1 NT (k) )+q(

ij +m P p=ij +1

NT (p) )]Tb

104

ij +m P

and Y

Pij [( k=1 NT (k) )+q(

**and sum over q, we obtain s s Gf,GT (ij ) |eR,gj (1) Af,GT (ij ) |eR,gj (1) Gf,GT (ij ) |eR,gj (2) Af,GT (ij ) |eR,gj (2) s s . . . . . . Gf,G Af,G mR,g mR,g j) j) s s T (ij ) |eR,gj (2 T (ij ) |eR,gj (2 = − Gf,BT (ij ) |eR,gj (1) Af,BT (ij ) |eR,gj (1) s s G A f,BT (ij ) |eR,gj (2) s f,BT (ij ) |eR,gj (2) s . . . . . . mR,g mR,g Af,B Gf,B j) j) s s T (ij ) |eR,gj (2 T (ij ) |eR,gj (2 s Gf,GT (ij ) |eR,gj (1) Gf,GT (ij ) |eR,gj (2) s . . . gj gj cG(k)G(l) cB(k)G(l) Gf,GT (i ) |eR,g (2mR,gj ) s j j , · gj gj Gf,BT (ij ) |eR,gj (1) cB(k)B(l) cG(k)B(l) s G f,BT (ij ) |eR,gj (2) s . . . mR,g Gf,B j) |eR,g (2 s
**

T (ij ) j

p=ij +1

NT (p) )]Ec

(4.9)

where Af,GT (ij ) |eR,gj (lj ) = G s Af,BT (ij ) |eR,gj (lj ) = G s

f,GT (ij ) |sR f,BT (ij ) |sR

T (ij )

, . (4.10)

T (ij )

The quanity in (4.10) can be evaluted from (4.3) by setting i in (4.3) as T (ij ) and si in (4.3) as sR,gj (lj ). R From (4.9), we can solve the generating functions of Gf,GT (ij ) |eR,gj (lj ) and Gf,BT (ij ) |eR,gj (lj ) s s for 1 ≤ lj ≤ 2mR,gj in closed form. The symbol sR,gj ,G represents the state of the

gj (lj ) e s where Gi (X.GT (i−1) |si−1 e−λNT (i−1) Tb Pe.G + Pe.gj . sR. (4.BT (ij +1) |sR.gj .gj (lj ) e s Pe. si−1 ]. for the system as i1 m 2 mR.gj . si ]) · R R R R X NT (i) Tb Y NT (i) Ec . Y ) = i=1 Gi (X.11) Gf. Y ) + j=1 lj =1 Gf. Therefore. the symbol sR. si−1 ].gj .B X NT (ij +1) Tb Gf.GT (ij +1) |sR.BT (i−1) |si−1 e−µNT (i−1) Tb Pe. NACK).B + Pe.GT (i) |si I(δ([GT (i−1) .gj .gj (lj ) e s −µNT (ij ) Tb −λNT (ij ) Tb −λNT (ij ) Tb Pe.GT (i−1) |si−1 e−λNT (i−1) Tb Pe. sR ]. This part of the generating function can be evaluated from (4. Similarly. si ]) + R R R Gf.105 memory for the system state generated by δ([GT (ij ) . sR. NACK). We use Gi (X.12) . NACK) = [BT (i) .BT (i) |si I(δ([GT (i−1) .BT (ij ) |eR. si ]) + R R R R i−1 Gf.3).gj (lj )].gj (lj )].g j Gs (X.BT (i) |si I(δ([BT (i−1) . The second part is the generating function for those successful transmissions with transmission indices form (ij + km + 1). The ﬁrst part is the generating function for those successful transmissions which do not belong to any indices of the form (ij + km + 1).GT (ij ) |eR.GT (ij +1) |sR. the total generating function is composed of two parts. si ]) + R R R i−1 Gf. Y ) to represent the generating function for the i-th successful transmission. sR ]. NACK) = [GT (i) . Y ) = si−1 ∈SR si ∈SR R R −µNT (ij ) Tb Y NT (ij +1) Ec (4.GT (ij ) |eR. we can express the total generating function.BT (i−1) |si−1 e−µNT (i−1) Tb Pe.B represents the state of the memory for the system state generated by δ([BT (ij ) . Hence. Gs .GT (i) |si I(δ([BT (i−1) .BT (ij ) |eR.G + Gf.gj (lj ) e s Gf. NACK) = [GT (i) .BT (ij +1) |sR. NACK) = [BT (i) .

4. · · · }. Nm . the receiver will send an ACK packet back to the transmitter. Then.4.106 4.4. · · · . · · · . we will ﬁrst consider a NR-ARQ system which the receiver does not contain any memory. · · · .1 SW-ARQ over Time-Varying Channels: No Repetition In this section. · · · }. If it is decoded incorrectly at the receiver. Whenever there is an information packet with length K generated. a NACK packet will be sent back to the transmitter. N2 . N2 . we will consider ARQ systems which have the sequence of packets with the following form : {N1 . If the data packet is decoded correctly at the receiver. This category will be discussed in Subsection 4. N2 . we will discuss a NR-ARQ system with memory in the receiver. N2 .4 Some Practical ARQ Systems In this section. Nm . This category will be discussed in Subsection 4. Memoryless Receiver The protocol of SW-ARQ with incremental redundancy is described as follows. We call this version of ARQ as nonrepeating ARQ (NR-ARQ). If the data packet is decoded correctly at the receiver. Then the transmitter will encode the same information packet again to a data packet with length N2 and retransmit it.1. it will be encoded to a data packet with length N1 . N1 . Nm . · · · . We call this repeating ARQ (R-ARQ). Nm .2. Let m be the number of diﬀerent packet lengths adopted by the transmitter. In the ﬁrst category. Nm . In the second category. The coded . we will consider ARQ systems which have the sequence of packets with the following form : {N1 . If it is decoded incorrectly at the receiver.4. N1 . the receiver will send an ACK packet back to the transmitter. we will specialize the general ARQ systems to two special categories. a NACK packet will be sent back to the transmitter.

The initial state is G1 or B1 depending on whether the channel condition is in a good or bad state. The state transition probabilities and output functions are given in Table 4. The symbol Gi (Bi ) for (1 ≤ i ≤ m) represents a packet of packet length index i ready to be transmitted and the channel condition is good (bad) in transmitting this packet. [SC . we can summarize the FSM states in Fig. · · · . X N2 Tb Y N2 Ec . The set Γ. e−µNi Tb and 1 − e−µNi Tb are state transition probabilities from G-G. which indicates the output functions of energy and delay for packets transmission. is composed of Γ = {X N1 Tb Y N1 Ec . consists of {G1 .3. B-B.1. The channel state transition probability from a good state to a bad state after transmitting a packet with length Ni is 1 − e−λNi Tb due to the exponential distribution of the channel state duration. 4. YES Start i=1 TX N i Suc .4. G2 . X Nm Tb Y Nm Tb } where Tb is the transmission time of a coded bit and the Ec is the transmission energy of a coded bit. The set of Σ is composed of {ACK. We do not assign states to the receiver because a receiver without memory only decodes the latest packet sent from the transmitter for the whole ARQ procedure. We can use a FSM to model this system. NACK}. respectively. ST ]. Note that P = 1 − P for a probability .107 packet lengths have the following relation : N1 < N2 < · · · < Nm .3: Stop-and-Wait ARQ protocol with increment redundancy. Gm . · · · . The transmitter/channel state space. According to our protocol shown in Fig. Similarly. and B-G.? End i=i+1 YES i<m ? NO i=m NO Figure 4. 4. B1 . B2 . Bm }. e−λNi Tb .

Bm X Nm T b Y Nm E c Bm G1 e−µNm Tb Pe.Gi X Ni Tb Y Ni Ec B1 e−λNi Tb Pe.Bi X Ni Tb Y Ni Ec Gm e−µNm Tb Pe. The MARQ protocol is the same as the SW-ARQ protocol described in subsec- . We begin with the description of our SW-MARQ protocol.Gm X Nm T b Y Nm E c Gm −λNm Tb G1 e Pe.Bi −µNi Tb B1 e Pe. output function Gi+1 e−λNi Tb Pe. Memory Receiver Memory ARQ (MARQ) is a method of improving the performance of ARQ over a non-stationary channel by storing and using received erroneous packets to aid decoding which are discarded by a memoryless ARQ protocol.Bi Bi Bi+1 e−µNi Tb Pe.108 Table 4.Gi X Ni Tb Y Ni Ec 1≤i≤m−1 −λNi Tb G1 e Pe.Gm X Nm T b Y Nm E c −µNi Tb P Gi+1 e X Ni Tb Y Ni Ec e.Bm X Nm T b Y Nm E c quantity P.Bm X Nm T b Y Nm E c B1 e−µNm Tb Pe. GN1 BN1 GN1 BN1 GN1 BN1 GN1 BN1 GN1 GN2 GN3 GNm BN1 BN2 BN3 BNm GN1 BN1 GN1 BN1 GN1 BN1 GN1 BN1 Figure 4.Gm X Nm T b Y Nm E c Bm e−λNm Tb Pe.Bi X Ni Tb Y Ni Ec 1≤i≤m−1 −µNi Tb P G1 e X Ni Tb Y Ni Ec e.1: Output functions and state transition probabilities for SWARQ previous state next state transition prob.Gm X Nm T b Y Nm E c B1 e−λNm Tb Pe.4: FSM representation for SW-ARQ protocol with increment redundancy.Bm X Nm T b Y Nm E c −µNm Tb Bm e Pe.Gi X Ni Tb Y Ni Ec −λNm Tb Gm e Pe.Gi X Ni Tb Y Ni Ec Gi Bi+1 e−λNi Tb Pe.

a state [G3 . In Fig. [G2 . The receiver will clear all stored packets in its memory when it receives a correct packet. the operation of the receiver memory is illustrated. the receiver will update the receiver memory content by storing the newest erroneous packet at the position 1 of the receiver memory and discard the oldest erroneous packet originally stored at the position mR . If the receiver receives an erroneous packet after the memory is full. 4. tion 4.5: Operation of sliding window. We need to add to the receiver state space SR to characterize the memory content. A state to describe SW-MARQ system consists of three parts. For example. The FSM model for a SW-MARQ system is the same as a SW-ARQ system except the receiver state.5.1 except that the receiver will store upto mR previously erroneous packets. 2 3-rd oldest 3 2-nd oldest deleted New Error packet 1 3-rd oldest 2 2-nd oldest 3 1-st oldest Er ro ra nd 1 New Error packet 2 3-rd oldest 3 2-nd oldest 4 1-st oldest Error but mR > 3 s es cc Su joint decode with new coming packet Ø Figure 4. B1 ]] represents that the transmitter transmits with packet length N3 with good (G) channel condition and the memory content of the receiver is two erroneous packets: the ﬁrst position of the receiver memory stores a packet with length N2 which received when the channel condition is good (G) and the second .109 1-st oldest 1 if m R= 3.4.

For general values of m and mR .6: FSM representation for Stop-and-Wait ARQ protocol with m = 2 and mR = 1. 4. in the column of the ”next state”.110 GN1 BN1 GN1 BN1 GN1 BN1 GN2|GN1 GN1 BN1 GN2|GN2 GN1 BN1 GN1 BN2|GN1 GN1 BN1 BN2|GN2 GN1 BN1 BN1 GN2|BN1 GN1 BN1 GN2|BN2 GN1 BN1 GN1 BN1 BN2|BN1 BN2|BN2 Figure 4. φ. G1 will be equal to a system state mR terms [G1 . However. state transition probabilities. the subscript i is used to distinguish the diﬀerent position of the receiver memory. · · · .6. and output functions (denoted by energy and delay generating function) in table 4. The packet error probability corresponding to this state is Pe. e. we only write those components of the system state vector. the FSM diagram for SW-MARQ with m = 2 and mR = 1 is shown. In Fig. Ri represents the memory content of the (i + 1)-th position in the receiver memory. In the columns of the ”previous state” and ”probability”.G3 |G2 B1 .. we can summarize the FSM states. .2. position of the receiver memory stores a packet with length N1 which is received when the channel condition is bad (B).g. The symbol Ri could be Gj or Bj where j is the index of the packet length. Ri represents the memory content of the i-th position in the receiver memory. The symbol Ri represents the memory content of some position in the receiver memory. φ]. In the following tables of this section.

RmR ] B1 e−µNi Tb Pe. Rk ] e−µNi Tb Pe.B1 X N1 Tb Y N1 Ec [G2 . R1 . · · · . · · · . [Gi+1 . Gi .Gm i|R1 R2 ···Rk [Bm . R1 . Gm . R1 . R1 . B1 ] e−µN1 Tb Pe. Rk ] e−µNi Tb Pe. Bm . Rk ] e−λNi Tb Pe.Bi |R1 R2 ···RmR Ni Tb Ni Ec [Bi+1 . R2 .Gi |R1 R2 ···RmR Ni Tb Ni Ec X Y [Bi+1 .G1 G1 e−µN1 Tb Pe. R1 . · · · . Rk ] e−µNm Tb Pe.Bm |R1 R2 ···Rk Nm Tb Nm Ec [Bm .Bm |R1 R2 ···Rk k < mR . · · · .Gm |R1 R2 ···RmR [Bm . [Gi+1 .B1 −λNi Tb G1 e Pe. R2 .Gm |R1 R2 ···Rk X Nm Tb Y Nm Ec G1 e−µNm Tb Pe. R1 . Gm .111 Table 4. R1 . Gi . R1 . RmR −1 ] e−λNm Tb Pe. R1 . RmR −1 ] e−λNm Tb Pe. RmR −1 ] e−µNm Tb Pe. Rk ] e−µNm Tb Pe. k < mR . R1 .Gm |R1 R2 ···RmR [Gm . RmR −1 ] e−µNi Tb Pe. RmR −1 ] e−λNi Tb Pe. R2 . R2 . · · · . RmR −1 ] e−µNm Tb Pe. R1 . · · · .Bi |R1 R2 ···RmR X Y G1 e−λNm Tb Pe. G1 ] e−λN1 Tb Pe. · · · .Gm |R1 R2 ···Rk −λNm Tb P [Gm . · · · . · · · .Bi |R1 R2 ···Rk i < m. RmR −1 ] e−λNi Tb Pe.Bm |R1 R2 ···RmR 1 Nm Tb Nm Ec Y [Gm . Rk ] e−λNm Tb Pe. RmR −1 ] e−µNi Tb Pe. Rk ] e−λNm Tb Pe.Gi |R1 R2 ···Rk i < m. R1 . [Gi+1 . R2 .Bi |R1 R2 ···Rk [Bi . R2 . Bi .B1 B1 B1 e−µN1 Tb Pe. · · · .2: Output functions and state transition probabilities for SWMARQ previous state. G1 ] e−λN1 Tb Pe. · · · .Gi |R1 R2 ···Rk [Gi . Gi . R1 . Rk ] B1 e−µNi Tb Pe.Gi |R1 R2 ···Rk G1 e−µNi Tb Pe. · · · .Gm |R1 R2 ···Rk k < mR . R1 . [Gi+1 .Gi |R1 R2 ···RmR G1 e−µNi Tb Pe. R2 . · · · . Rk ] B1 e−µNm Tb Pe.Bi |R1 R2 ···RmR i < m. · · · . · · · . next state probability output G1 e−λN1 Tb Pe. R1 . RmR ] i < m. Gm . Bi . R1 .B1 [B2 . Rk ] e−λNi Tb Pe. · · · . Rk ] B1 e e.Gi |R1 R2 ···RmR [Gi . R1 . R2 . R1 . Bm . Gm . [Gm .G1 G1 B1 e−λN1 Tb Pe. R1 . · · · .Bm |R1 R2 ···RmR X [Bm .G1 [B2 . R1 . · · · .Bm |R1 R2 ···Rk X Y G1 e−λNi Tb Pe. k < mR . [Gm .Gm |R1 R2 ···RmR G1 e−µNm Tb Pe. B1 ] e−µN1 Tb Pe. · · · . Bm . Bm .Bm |R1 R2 ···RmR [Bm .Gi |R1 R2 ···RmR B1 e−λNi Tb Pe. Gi . Bi . RmR ] B e−µNm Tb Pe. R1 . · · · . RmR ] B e−λNm Tb Pe. R1 . R1 . Rk ] B1 e−λNi Tb Pe.Bi |R1 R2 ···Rk X Y −λNm Tb G1 e Pe. · · · .G1 X N1 Tb Y N1 Ec [G2 . Bi .Bi |R1 R2 ···RmR [Bi .Bm |R1 R2 ···RmR .Gi |R1 R2 ···Rk Ni Tb Ni Ec X Y [Bi+1 . · · · .Bm |R1 R2 ···Rk [Bm . · · · .Bi |R1 R2 ···Rk Ni Tb Ni Ec [Bi+1 . · · · .Gm |R1 R2 ···RmR 1 Nm Tb Nm Ec Y X [Gm . R1 .

4. the . the receiver will send an ACK packet back to the transmitter. named repetition SW-ARQ with memoryless receiver (R-SW-ARQML). Therefore. If a detectable but uncorrectable error occurs at the receiver. whenever there is an information packet with length K generated. If the data packet is decoded correctly at the receiver.4. we will consider an ARQ systems with repetition packet length. a NACK packet will be sent back to the transmitter. Let m be the number of diﬀerent packet lengths used by the transmitter. The protocol.2 SW-ARQ over Time-Varying Channels: Repetition In this section. a NACK packet will be sent back to the transmitter. If a detectable but uncorrectable error occurs at the receiver. the third ARQ system also equips the receiver with memory but the receiver memory contents will be updated to be empty after the receiver cannot decode correctly when the memory buﬀer is full. the transmitter will encode the information packet to a data packet with length N1 again. If the data packet is decoded correctly at the receiver. Then the transmitter will encode the same information packet to a data packet with length N2 and transmit the packet. There are three kinds of repetition ARQ systems analyzed in this section. the receiver will send an ACK packet back to the transmitter.1.112 4. operates as follows. Memoryless Receiver In this subsection. the ﬁrst ARQ system does not contain any memory in the receiver. it will be encoded to a data packet with length N1 . If it is decoded in error at the receiver. At the transmitter. we will consider ARQ systems with a memoryless receiver. the second ARQ system equips the receiver with memory and uses a sliding window scheme as mentioned in Section 4. if the transmitter can not receive an ACK by sending a data packet with length Nm at m-th transmission.

The repetition procedure of these m diﬀerent packet lengths will be terminated if an ACK packet is received. The operation of this protocol is shown in part (a) of Fig.7: Three diﬀerent repetition ARQ protocols with m = 2 considered in this section.7.2. By using the FSM model described in Section 4. Each piece of the bold line in Fig.7 indicates the packets in the receiver memory used to decode jointly when the transmitter has transmitted the last packet underscored by each piece of the bold line. transmitter will encode the same information packet to a data packet with length N2 again. 4.113 N1 N2 N1 N2 (a) No memory N1 N2 N1 N2 (b) Memory : Sliding Window N1 N2 N1 N2 (c) Memory : Renew Window Figure 4. 4. If the transmitter can not receive an ACK by sending a data packet with length Nm at 2m-th transmission. we can summarize the state . it will encode the information packet to a data packet with length N1 at 2m + 1-th transmission.

8: State diagram for R-SW-ARQ-ML. A state space to describe R-SW-ARQ-SWM system consists of three parts. The third part denotes the . The FSM model for repetition SW-ARQ systems with sliding window memory (R-SW-ARQ-SWM) is extended to a repetition SW-ARQ system by including the receiver state.114 GN1 BN1 GN1 BN1 GN1 BN1 GN1 BN1 GN1 GN2 GN3 GNm BN1 BN2 BN3 BNm GN1 BN1 GN1 BN1 GN1 BN1 GN1 BN1 Figure 4. Memory Receiver: Sliding Window The protocol discussed in this subsection is the same as the repetition SW-ARQ protocol described in 4. We need to add the receiver state SR to characterize the memory content. We set mR = m − 1 be the maximum number of packets that the receiver can store in its memory. If the receiver receives an erroneous packet after the memory is full. transition probabilities and output functions (denoted by energy and delay generating function) in Table 4.2 except that the receiver will store up to mR previously erroneous packets. the receiver will update its memory content by storing the newest erroneous packet and discarding the oldest erroneous packet. The ﬁrst part of the state represents the channel condition.4.3. The second part is used to represent the packet lengths for transmission. The receiver will clear all stored packets in its memory when it receives a correct packet.

Bm X Nm T b Y Nm E c B1 e−µNm Tb Pe. The packet error probability corresponding to this state is Pe. In Fig. state transition probabilities.Gi X Ni Tb Y Ni Ec B1 e−λNi Tb Pe.Bm X Nm T b Y Nm E c −µNm Tb B1 e Pe.Bm X Nm T b Y Nm E c contents of the receiver memory. Memory Receiver: Non Overlap In this section.Bm X Nm T b Y Nm E c Bm G1 e−µNm Tb Pe.Gm X Nm T b Y Nm E c −µNi Tb P Gi+1 e X Ni Tb Y Ni Ec e.4. we summarize the FSM states.Gi X Ni Tb Y Ni Ec Gi Bi+1 e−λNi Tb Pe. we consider the case that mR = m − 1 is the maximum number of packets that the receiver can store in its memory.G3 |G2 B1 .Gi X Ni Tb Y Ni Ec 1≤i≤m−1 −λNi Tb G1 e Pe.Bi Bi Bi+1 e−µNi Tb Pe.9.Gm X Nm T b Y Nm E c Gm −λNm Tb G1 e Pe. output function Gi+1 e−λNi Tb Pe. G2 .4. For example. The protocol is the same as the repetition SW-ARQ protocol with the receiver memory described in 4. B1 ] represents that the transmitter transmits with packet length N3 with good (G) channel condition and the memory contents of the receiver is two erroneous packets: the ﬁrst packet with length N1 is received when the channel condition is bad (B) and the second packet with length N2 is received when the channel condition is good (G). and output functions (denoted by energy and delay generating function) in Table 4.3: Table for transition probabilities and output functions of R-SW-ARQ-ML previous state next state transition prob.Gm X Nm T b Y Nm E c B1 e−λNm Tb Pe. state [G3 . 4.Gi X Ni Tb Y Ni Ec −λNm Tb G1 e Pe.Bi X Ni Tb Y Ni Ec G1 e−µNm Tb Pe. the FSM diagram for SW-MARQ with m = 2 and mR = 1 is shown.Bi −µNi Tb B1 e Pe. For general values of m and mR .2 except .Bi X Ni Tb Y Ni Ec 1≤i≤m−1 −µNi Tb P G1 e X Ni Tb Y Ni Ec e.115 Table 4.Gm X Nm T b Y Nm E c B1 e−λNm Tb Pe.

· · · . B1 ] e−µN1 Tb Pe. Bi . · · · . · · · .Gi |R1 R2 ···Ri−1 X Ni Tb Y Ni Ec G1 e−µNi Tb Pe. B1 ] e−µN1 Tb Pe. Ri−1 ] e−λNi Tb Pe. R1 . · · · . R1 .B1 B1 B1 e−µN1 Tb Pe. R1 . Bi . R1 . [Gi+1 . · · · . Bi . Ri−1 ] e−λNi Tb Pe.Gi |R1 R2 ···RmR G1 e−µNi Tb Pe.Gi |R1 R2 ···RmR e−λNi Tb Pe. · · · .Gi |R1 R2 ···RmR Ni Tb Ni Ec X Y [Bi+1 .Bi |R1 R2 ···RmR Ni Tb Ni Ec X Y [Bi+1 . R1 .116 Table 4. R1 . Gi . RmR −1 ] e−λNi Tb Pe. R1 .Bi |R1 R2 ···Ri−1 [Bi . R1 . next state probability output G1 e−λN1 Tb Pe. G1 ] e−λN1 Tb Pe. · · · .B1 −λNi Tb G1 e Pe. G1 ] e−λN1 Tb Pe.G1 G1 e−µN1 Tb Pe.Bi |R1 R2 ···RmR .Bi |R1 R2 ···RmR [Bi . · · · RmR −1 ] e−µNi Tb Pe. Ri−1 ] e−µNi Tb Pe. R2 .Gi |R1 R2 ···Ri−1 [Gi+1 . R2 . R1 .Bi |R1 R2 ···Ri−1 Ni Tb Ni Ec X Y [Bi+1 . R1 .B1 N1 Tb N1 Ec [G2 . Ri−1 ] e−µNi Tb Pe. · · · RmR −1 ] e−µNi Tb Pe.Gi |R1 R2 ···Ri−1 i<m [Bi+1 .G1 [B2 . RmR −1 ] e−λNi Tb Pe. Gi . RmR ] B1 1 ≤ i ≤ m.Bi |R1 R2 ···RmR 1 ≤ i ≤ m.G1 G1 B1 e−λN1 Tb Pe. Gi .Bi |R1 R2 ···Ri−1 G1 e−λNi Tb Pe. · · · .B1 X Y [B2 . Bi .Gi |R1 R2 ···Ri−1 −λNi Tb P [Gi . R1 . [Gi+1 .G1 N1 Tb N1 Ec X Y [G2 . R2 . R2 · · · RmR ] B1 e−µNi Tb Pe. R1 . Ri−1 ] B1 e−µNi Tb Pe.Gi |R1 R2 ···RmR [Gi . Gi .4: Table for transition probabilities and output functions of R-SW-ARQ-SWM previous state. Ri−1 ] B1 e e.Bi |R1 R2 ···Ri−1 i<m [Gi+1 . · · · .

In Fig. 4. the FSM diagram for SW-ARQ systems with non-overlapping window memory (R-SWARQ-NOM) with m = 2 and mR = 1 is shown. we can summarize the FSM states.117 GN1 BN1 GN1 BN1 GN1 BN1 GN2|GN1 GN1 BN1 GN2|GN2 GN1 BN1 GN1 BN2|GN1 GN1 BN1 BN2|GN2 GN1 BN1 GN2|BN1 BN1 GN1 GN1 BN1 BN2|BN1 BN1 GN2|BN2 GN1 BN1 BN2|BN2 Figure 4. Once the transmitter receives a negative acknowledge (NACK) packet after one roundtrip propagation time. state transition probabilities.10. that the receiver will also refresh its memory contents to be empty after the capacity of the receiver memory reaches mR as shown in part (c) of Fig. the transmitter retransmits the received erroneously decoded packet and succeeding packets. The transmitter sends packets with a sequence number contiguously.5.9: FSM representation for repetition SW-ARQ systems with sliding window memory with m = 2 and mR = 1.7. 4. For general values of m and mR . 4. and output functions in Table 4.5 Go-Back-N ARQ We begin this section by describing the operation of Go-Back-N ARQ (GBN-ARQ) protocol. If the receiver decodes a packet correctly. it will send .

10: FSM representation for repetition SW-ARQ systems with non-overlapping window memory with m = 2 and mR = 1.118 GN1 BN1 GN1 BN1 GN2|GN1 GN1 BN1 GN1 BN2|GN1 GN1 BN1 GN2|BN1 BN1 GN1 GN1 BN1 BN2|BN1 BN1 Figure 4. .

Bi |R1 R2 ···Ri−1 i<m [Gi+1 .119 Table 4.Gi |R1 R2 ···Ri−1 −λNi Tb P [Gi . G1 ] e−λN1 Tb Pe.G1 N1 Tb N1 Ec X Y [G2 .B1 B1 . R1 . Ri−1 ] B1 e e. B1 e−µN1 Tb Pe.Bi |R1 R2 ···Ri−1 Ni Tb Ni Ec X Y [Bi+1 . B1 ] e−µN1 Tb Pe. R1 . G1 ] e−λN1 Tb Pe. Gi . · · · .B1 N1 Tb N1 Ec [G2 .Gm |R1 R2 ···RmR X Nm Tb Y Nm Ec G1 e−λNm Tb Pe.B1 −λNi Tb G1 e Pe. · · · . Ri−1 ] e−µNi Tb Pe. R2 .G1 [B2 .5: Table for transition probabilities and output functions of R-SW-ARQ-NOM previous state. Bi . next state probability output G1 e−λN1 Tb Pe. · · · . R2 . Bi . B1 e−λN1 Tb Pe. Gi . R1 . Ri−1 ] e−µNi Tb Pe.Bm |R1 R2 ···RmR [Bm .Bi |R1 R2 ···Ri−1 −λNm Tb G1 e Pe.Bm |R1 R2 ···RmR B1 e−µNm Tb Pe. Ri−1 ] e−λNi Tb Pe. R1 . R1 . · · · .Bm |R1 R2 ···RmR .B1 X Y [B2 .Gi |R1 R2 ···Ri−1 X Ni Tb Y Ni Ec G1 e−µNi Tb Pe. Ri−1 ] B1 e−µNi Tb Pe.Gm |R1 R2 ···RmR G1 e−µNm Tb Pe. R1 .G1 G1 . RmR ] B1 e−µNm Tb Pe. RmR ] B1 e.Gm |R1 R2 ···RmR B1 e−λNm Tb Pe. · · · . · · · . R2 .Gi |R1 R2 ···Ri−1 i<m [Bi+1 . R2 .Gi |R1 R2 ···Ri−1 [Gi+1 .Gm |R1 R2 ···RmR −λNm Tb P e [Gm . B1 ] e−µN1 Tb Pe.Bm |R1 R2 ···RmR X Nm Tb Y Nm Ec G1 e−µNm Tb Pe. · · · . · · · . Ri−1 ] e−λNi Tb Pe. R1 .G1 G1 e−µN1 Tb Pe. R1 .Bi |R1 R2 ···Ri−1 [Bi .

Let T(Ni ) be the channel transition probability matrix after transmitting a packet with length Ni . The receiver will send a NACK packet back to the transmitter if a packet can not be decoded correctly. It can be written as: e−λNi Tb (1 − e−µNi Tb ) (1 − e−λNi Tb ) e−µNi Tb T(Ni ) = .13) Then the transition probability matrix after Nrt transmissions of a packet with length Ni is T(Nrt Ni ) = [T(Ni )]Nrt = 1 2 − e−λNi Tb − e−µNi Tb ( (1 − e−µNi Tb ) (1 − e−λNi Tb ) (1 − e−µNi Tb ) (1 − e−λNi Tb ) Nrt + (e −λNi Tb +e −µNi Tb − 1) (1 − e−λNi Tb ) −(1 − e−µNi Tb ) −(1 − e−λNi Tb ) (1 − e−µNi Tb ) (4. From this transition probability matrix.14) ). where the Nrt is the number of packets in a round trip delay. the transition probability from G to G after one round trip time G of transmitting packets with length Ni . (4. is G Trt i Gi = 1 − (1 − e−µNi Tb ) 1 − (e−λNi Tb + e−µNi Tb − 1)Nrt 2 − e−λNi Tb − e−µNi Tb Similarly. denoted as Trt i Gi . we can incorporate the previous SW-ARQ with incremental redundancy and the receiver memory scheme into GBN-ARQ protocol by using diﬀerent packet lengths and modifying the decoding structure (from a nonmemory receiver to a memory receiver). From the described protocol of GBN-ARQ. we can obtain the transition probability from G to B. B to G and B to B.120 an acknowledge (ACK) back to the transmitter. .

it is well known that .6 Cutoﬀ Rate for Memory Receiver over Memoryless Channels Let K be the number of information bits in a packet and N be the number of coded bits for this packet.G1 Trt i Bi + PB Pe.B2 Trt 2 G2 ]Z + Pe.121 They are G Trt i Bi = (1 − e−µNi Tb ) B Trt i Gi B Trt i Bi 1 − (e−λNi Tb + e−µNi Tb − 1)Nrt 2 − e−λNi Tb − e−µNi Tb 1 − (e−λNi Tb + e−µNi Tb − 1)Nrt = (1 − e−λNi Tb ) 2 − e−λNi Tb − e−µNi Tb 1 − (e−λNi Tb + e−µNi Tb − 1)Nrt = 1 − (1 − e−λNi Tb ) 2 − e−λNi Tb − e−µNi Tb (4.G2 (1 − Pe.4. 4.G2 )Trt 2 B2 ]Z)) G B G B 1 − [Pe.B2 Trt 2 G2 + Pe.B1 Trt 1 B1 )X Nrt N1 Tb Y Nrt N1 Ec GB .G2 − [(1 − Pe.B2 − [(1 − Pe. The generating function is Gs = PG (1 − Pe. We also can obtain the generating function of GBN-ARQ with the receiver memory scheme by using the same method as we derived before.B2 )Pe.B2 )Trt 2 G2 ]Z)) B G B G 1 − [Pe. The only change is to modify the channel transition probabilities and multiply the round trip factor Nrt in energy and delay components at each transmission.G2 (Trt 2 G2 + Trt 2 B2 − 1)Z 2 G B (X N2 Tb Y N2 Ec (1 − Pe.17) and Z = X Nrt N2 Tb Y Nrt N2 Ec .G2 Trt 2 B2 − Pe.B2 Trt 2 G2 − Pe.G1 )X Nrt N1 Tb Y Nrt N1 Ec + PB ∗ (1 − Pe.B1 Trt 1 G1 )X Nrt N1 Tb Y Nrt N1 Ec GG + G B (PG Pe. From the random coding argument.G1 Trt i Gi + PB Pe.G2 Pe.16) where GG = GB = B G (X N2 Tb Y N2 Ec (1 − Pe.B1 )X Nrt N1 Tb Y Nrt N1 Ec + G B (PG Pe.G2 )Pe.15) We will give an example about calculating the generating function for a GBNARQ system with m = 2 and mR = 0 by using the approach presented in subSection 4.15).B2 (Trt 2 B2 + Trt 2 G2 − 1)Z 2 (4.G2 Trt 2 B2 ]Z + Pe.B2 Pe.1 and the channel transition probabilities (4. (4.G2 Trt 2 B2 + Pe.B2 (1 − Pe.

we can relate the channel SNR and packet length N with system performance.N ≤ 2K−N R0 . · · · . Let si be the signal vector received at the receiver end. The cutoﬀ rate of two diﬀerent kinds of receiver structure will be derived in this section. It is si packet 1 packet 2 packet m = [si11 si12 · · · si1N1 si21 si22 · · · si2Nm · · · sim1 sim2 · · · simNm ]. By using the cutoﬀ rate to represent the packet error probability.k are selected from the signal set {a1 .K. The third cutoﬀ rate is derived by the same assumption as second cutoﬀ rate. Let Nj be the number of coded symbols used at j-th packet at the receiver memory.18) where R0 is the cutoﬀ rate determined by the channel SNR and other channel characteristics.122 there exist an encoder and decoder that has packet error probability Pe. The second cutoﬀ rate is calculated according to the imperfect channel state information by using estimated channel transition probability of each signal. (4. Because the channel SNR and packet length used for packets transmission are two crucial components in determining the energy and delay tradeoﬀ of an ARQ system. 4.Part I: Perfect Channel Knowledge Let m be the total number of packets combined at the receiver which are used to decode jointly.1 Random Coding Bound for Memory Receiver in AWGN channel. Here we consider a set of M -ary coded signals constructed from m j=1 Nj -dimensional q symbols.N bounded as Pe.K. · · · aq } of q elements. . 2. however. a2 . The ﬁrst cutoﬀ rate is evaluated by assuming that the receiver has all the channel state information. where j ∈ {1.19) where sij. the receiver with memory will utilize more information (the failure information) to decode. (4. m}.6.

M RX = We can interpret R0. there are B = q M of M signals since there are m j=1 Pm j=1 Nj possible sets Nj positions can be chosen from q symbols.i ≤ α. Let dl. a2 .i ≤ (M − 1) m ( k=1 lk =1 mk =1 plk pmk e − d2 m lk k 4N0 )Nk < M = 2 where R = log Pm 2 M k=1 Nk 2−Nk R0.123 According to the random coding argument.m = Ek al − am al − am 2 . We only .23) R0. if we set R0.k .i ≤ ( i=j k=1 lk =1 mk =1 q lk =1 plk pmk e − d2 m lk k 4N0 )Nk (4.2 m k=1 Nk + ··· + Nm R0.M RX is N1 R0. · · · pq }. (4.21) Therefore. The average error probability given the transmission of the i-th message is P e. Let Pe.20) If we can show that P e.i (k) be the error probability of signal set k given the i-th message was transmitted.i = B k=1 Pe.m .M RX as the cutoﬀ rate by combining m packets according to their packet length ratio to the total length of m packets at the receiver. · · · aq } is according to probabilities {p1 . (4. where Ek is the received energy level of packet k and is the distance between signal al and am .i (k).i ≤ i=j E[e − si −sj 2 4N0 ]. m k=1 Nk (4. p2 .22) k=1 P −( m Nk )(R0. By the union-Bhattacharyya bound for an AWGN channel.k = − log2 ( the right hand side of (4.i ≤ α.1 m k=1 Nk + N2 R0. Then we have m q q P e. If we assume that the selection of symbols {a1 . the average error probability of transmitting i-th message over the ensemble of codewords sets is P e.21) as q mk =1 plk pmk e − d2 m lk k 4N0 ) we can further express m q q P e. then there exists a signal set with Pe.M RX −R) k=1 is the code rate after packets combination and R0.

There are at least M/2 (otherwise the average error probability of this code will greater then /2). sR ) > p∗ (y|sj . Similarly.M RX ( Pm k=1 Nk ) codewords that the average error m k=1 probability of this code is less then /2 for large of the codewords with error probability less then Nk . We will show that we can ﬁnd a single codeword set such that the error probability of each signal of this codeword set approaches to zero as packets length approaches to inﬁnity.6. Choosing a code with M = 2 · 2R0. aq }. (4. a2 .k ∈ {a1 .124 show that there exist a signal. the receiver end has signal-to-noise ratio EG /N0 but the receiver uses ∗ the incorrect signal-to-noise ratio EG /N0 to decode. for which the average error probability can be made arbitrary small as packets length becomes large. sR )for all j = i}. Let si be the signal vector received at the receiver end. the receiver end has signal-to-noise ratio EB /N0 when the channel condition is bad but the receiver uses ∗ the incorrect signal-to-noise ratio EB /N0 to decode. When the channel condition is good. It is si packet 1 packet 2 packet m = [si11 si12 · · · si1N1 si21 si22 · · · si2Nm · · · sim1 sim2 · · · simNm ].M RX ( codewords with error probability small as 4. we consider that the receiver uses the channel transition probabilities p∗ (y|x) as if they were the true transition probabilities.2 m k=1 Pm k=1 Nk ) Nk −→ ∞. · · · . The decision region are ∗ Ri = {y : p∗ (y|si . we obtain a code with (at least) 2R0. Random Coding Bound for Memory Receiver .Part II : Imperfect Channel Knowledge In this section.25) . By deleting those bad codewords (their error probability is greater then ). say i-th signal.24) where sij. We will derive random coding bound of the receiver with memory with m packets. (4.

sR ) p∗ (y|sj . ∗ (y|s .j = {y : p∗ (y|si ) < p∗ (y|sj )}. Let Ri be the region where signal i is not decided. sR ) p∗ (y|si . sR ) p∗ (y|sj .30) where λ > 0.26) Pe. sR ). sR ) ∗ > 1.j ) < ∗ y∈Ri. s ) p i R we can further bound (4.j p(y|si .j . ∗ (4. (4. we have ∗ P (Ri. sR ) λ .31) . (4.125 where sR is the receiver memory state. Then we have ∗ ∗ ∗ Ri. y ∈ Ri.28) as follows (4.29) ∗ P (Ri.27) For each term of above summand. sR ) p∗ (y|si . we can upper bound (4.j p(y|si .i = P (Ri |Hi ) ≤ j=i ∗ P (Ri.30) as ∗ P (Ri. (4.28) Since p∗ (y|sj . sR ) λ .j ) < y∈R Pm N k=1 k p(y|si .j ).j .j ) = ∗ y∈Ri. By relaxing the integral domain. (4. Then Ri = j=i ∗ where Ri.

R1 ) yl1 ∈R p(yl2 |si2l2 .i is upper bounded by Pe.32) Nk lk =1 By taking the error probability over all possible codes and let p(si ) = p(siklk )). . R1 ) p∗ (yl1 |si1l1 . R1 ) p(yl1 |si1l1 . R1 ) ∗ p (yl1 |si1l1 . R2 ) p∗ (yl2 |sj2l2 . R2 ) =1 p(ylm |simlm . Rm ) . m k=1 ( (4. sR ) N1 λ = j=i N2 y∈R Pm N k=1 k p(yl1 |si1l1 . Rm ) = j=i N2 l1 =1 p∗ (yl1 |sj1l1 . R2 ) p(yl2 |si2l2 . R2 ) λ λ × ··· λ l2 =1 Nm yl2 ∈R lm =1 ylm ∈R p(ylm |simlm . R1 ) l1 =1 p∗ (yl1 |sj1l1 . the error probability Pe. sR ) p(y|si .126 Thus. Rm ) p∗ (ylm |simlm . Rm ) p∗ (ylm |sjmlm . R2 ) p∗ (yl2 |si2l2 . Rm ) p∗ (ylm |simlm . Rm ) ··· λ Nm lm =1 N1 p∗ (ylm |sjmlm . R1 ) λ λ × l2 p∗ (yl2 |sj2l2 . sR ) ∗ p (y|si . R2 ) ∗ p (yl2 |si2l2 .i < j=i Pm N y∈R k=1 k p∗ (y|sj .

1 (X1 .k (λ) as ∗ R0. Rm ) p∗ (ylm |simlm . (4. If let R0. X2 )]]N1 [E[Jλ. R1 ) p∗ (yl1 |sj1l1 .127 Then N1 P e. R2 ) ∗ p (yl2 |si2l2 . Then we can further upper bound of (4. Rk ) p∗ (yklk |sjklk .M RX = max λ N1 ∗ R0.M RX −R) (4. R2 ) yl2 ∈R p(simlm )p(sjmlm ) p(ylm |simlm . a2 .Rn ) n for the n-th packets in the receiver memory.33) n sinl n sjnl n p(sinln )p(sjnln ) ynln p(ynln |sinln .35) ∗ is the code rate after packets combination.k (λ) = p(sikl ) k max − log2 sikl sjkl k k p(siklk )p(sjklk ) kylk ∈R p(yklk |siklk . R1 ) λ λ × yl2 ∈R p(yl2 |si2l2 . R2 ) p∗ (yl2 |sj2l2 .n (X1 .m (λ) m k=1 Nk .1 (λ) m k=1 Nk + N2 ∗ R0. Rk ) p∗ (yklk |siklk .m (X1 .33) as P e. Rk ) λ . X2 )]]Nm ∗ where E[Jλ. (4.36) . Rm ) ··· λ Nm lm =1 simlm sjmlm ylm ∈R p∗ (ylm |sjmlm . R2 ) ··· λ lm p∗ (ylm |sjmlm .m (X1 .i < j=i N2 si sj l1 =1 p(si1l1 )p(sj1l1 ) p(si2l2 )p(sj2l2 ) l2 =1 Nm yl1 ∈R p(yl1 |si1l1 .1 (X1 .2 (X1 . Rn ) ∈R p∗ (ynln |sjnl .i < 2−( where R = log Pm 2 M k=1 Nk Pm k=1 ∗ Nk )(R0. X2 )]]N2 · · · [E[Jλ. and siklk ∈ {a1 .Rn ) λ p∗ (ynln |sinl .2 (λ) m k=1 Nk + ··· + Nm ∗ R0. R0. X2 )] is (4.2 (X1 . X2 )]]Nm ∗ ∗ ∗ = (M − 1)[E[Jλ. Rm ) ∗ p (ylm |simlm . R1 ) p∗ (yl1 |si1l1 . Rm ) ylm ∈R =1 N1 = j=i N2 l2 =1 si2l sj2l 2 p(si1l1 )p(sj1l1 ) l1 =1 si1l sj1l 1 1 yl1 ∈R p(yl1 |si1l1 . Rm ) p(simlm )p(sjmlm ) p(ylm |simlm . R2 ) p(si2l2 )p(sj2l2 ) p(yl2 |si2l2 . R1 ) p∗ (yl1 |si1l1 . X2 )]]N2 · · · [E[Jλ.34) . X2 )]]N1 [E[Jλ. R2 ) p∗ (yl2 |si2l2 .M RX is ∗ R0. Rm ) = j=i ∗ ∗ ∗ [E[Jλ. R1 ) λ λ × 2 p∗ (yl2 |sj2l2 . R1 ) p∗ (yl1 |sj1l1 . · · · aq }. The random variable X1 represents ∗ source signals silk and X2 represents source signals sjlk .

4.1. However. it does not guarantee that the same code ensemble can satisfy the random coding bound for any usage of previous k codes for k < m.3 Existence of Codes for Memory Receiver In the previous derivation of cutoﬀ rate for the receiver with memory.1 (indicated by symbol I) and the memory receiver with imperfect channel knowledge described in Subsection 4.2 (represented by symbol II) in Fig. it is stored in the receiver memory. i.6. Cm ) satisfying the random coding bound if there m−1 packets stored in the receiver memory.11.. We observe that the cutoﬀ rate of the memory receiver structure I is higher than the cutoﬀ rate of the memory receiver structure II because the receiver of structure I has much more information about the channel condition. 4.6. Therefore. we know that there is an ensemble random codes of m blocks. there are only k − 1 packets . We assume that the ﬁrst packet with length N1 is decoded erroneously after receiving over the good channel condition. The cutoﬀ rate with symbol + is higher then the cutoﬀ rate with symbol o because the lines with symbol + have higher bad channel SNR compared to the lines with symbol o. the cutoﬀ rate represented by the solid lines is lower then the cutoﬀ rate represented by the dashed lines. The cutoﬀ rate is plotted with respect to good channel SNR for the memory receiver with perfect channel knowledge described in Subsection 4. · · · .e.6. The cutoﬀ rate is evaluated by considering the ﬁrst erroneous packet stored in the receiver memory and the second incoming packet with length N2 from bad channel condition jointly.128 The channel model used in the following cutoﬀ rate evaluation is Gilbert-Elliott channel model described at the beginning of subsection 4. C2 . Since the cutoﬀ rate with parameters N1 = 2N2 has higher chance for the joint packet (the packet combining the ﬁrst erroneous one and the second incoming one) to stay at good channel condition compared to the cutoﬀ rate with parameters N2 = 2N1 .4. (C1 .

11: Cutoﬀ rate comparison for two diﬀerent receiver structures. II. I.6 2 0. II. N =2N 2 1 r=8.1 0 −3 −2 −1 0 1 EG/N0 (dB) 2 3 4 5 Figure 4.8 0. N2=2N1 r=2.k > β2−( Pk k l=1 )(R0. Note that Pe.k is a random number since we use random coding selection for codes construction.M RX −R) ) ≥ 1− (4.k > β2 P k −( k )(R0.M RX −R) l=1 m }) = 1 − P r( m {Pe.37) where Pe. · · · . This implies that there exist at least an .5 R0 0.38) by choosing a suitable β. N1=2N2 r=8. Hence. I. say β = 2m. Theorem : There exists an ensemble of codes (C1 .2 0. Proof : By applying Markov inequality. II. I. m P r( {Pe. (4. Ck ) ≤ 2−( Pk k l=1 )(R0.M RX −R) }) k=1 k=1 ≥ 1− k=1 P r(Pe. we have P r(Pe.k is the probability of error for previous k codes. II. N2=2N1 r=8. I. The following theorem indicates such ensemble codes exist [49]. N =2N 1 0.4 0.M RX −R) )≤ 1 . · · · .3 0.7 0.k > β2−( Pk k l=1 )(R0. N1=2N2 r=2. β for 1 ≤ k ≤ m. N2=2N1 r=2. C2 . stored in the receiver memory.129 0.M RX −R) for 1 ≤ k ≤ m.9 r=2. N1=2N2 r=8. C2 . Cm ) satisfying Pe (C1 .k > β2−( m >0 β Pk k l=1 )(R0.

z denotes a received signal..k by the quantity z max − log Pj j k Pj Pk z f (z|k)[f ∗ (z|j)f ∗ (z|k)]λ dz . C2 . (4.e. for 1 ≤ k ≤ m.k) where dB (j. we need to optimize the following cutoﬀ rate per waveform given λ R0 (λ) where 1. f (z|k) = z f ∗ (z|k) and the system is operated in an AWGN channel. j. (4.6.39) Here we can see that the factor β has no eﬀect on the cutoﬀ rate. assuming the j-th symbol was transmitted. k) is Bhattachryya distance.M RX −R) ). Denote Dj. 4. 3. 2. 4.4 Cutoﬀ Rate Optimization In the previous derivation of cutoﬀ rate. We want to ﬁnd the priori probability of the input symbol Pj in two cases.40) f (z|k)[f ∗ (z|j)f ∗ (z|k)]λ dz. Pj is the a priori probability of the j-th symbol selected by the transmitter. i. k are the index for symbols. (C1 . The superscript symbol ∗ indicates the incorrect channel transition probability used by the receiver to decode. f (z|j) is the true channel transition probability.130 ∗ ∗ ∗ ensemble of m blocks. Cm ) with Pk ∗ Pe. the value f (z|k)[f ∗ (z|j)f ∗ (z|k)]λ dz is equal to e−dB (j. If λ = 1/2.k ≤ β2−( k l=1 )(R0. · · · . the probability density function of the receiver output. . (1) there is no constraint of Pj except that its components are sum up to one and (2) there is an energy constraint of input symbols.

45) where [D.2 · · · 1 (4.2 · · · D1. (4. From Kuhn-Tucker theorem.42) From the signal constellation.i Ps ≥ −2detD M s=1 det[D. we can write the term PT DP as M PT DP = i=1 Pi2 + 2 1≤i<j≤M Di. .=i Ds.k ]} (4.M D2. The equality will hold for all i such that Pi > 0. .131 We will consider the case (1) ﬁrst. the matrix D has the following form D1. under constraint Pj = 1. From this. .1 DM. . .i due to the symmetry of the Bhattachryya distance function. it will also maximize the cutoﬀ rate R0 .45) minimizes PT DP. . The optimization problem of this case can be formulated as follows: R0 = max{− log[ Pj j T k Pj Pj Pk Dj. DM. the necessary and suﬃcient conditions on the probability vector P that minimizes PT DP are M Pi + s=1. . . 1s ] . 1s ] is matrix obtained by replacing the s-th column of matrix D by one vector.j Pi Pj .43) and Di.41) = max{− log[P DP]}. for all i (4. Because the probability vector obtained in (4. . ··· . it is a convex tion.M 1 1 · · · D2.1 . therefore. j (4.44) func- The quantity PT DP is a quadratic function of P.j = Dj.

M=4 Opt. The amplitude levels are assigned equally spaced over the interval √ √ [− Ec .M=3 Unif. . we can calculate the cutoﬀ rate for optimal a priori probability input signals. However. We observe that there is about 40% increase of cutoﬀ rate at -10 dB for M = 3 and about 60% increase of cutoﬀ rate at -10 dB for M = 4. We also note that when the channel SNR is low. we plot the cutoﬀ rates for uniform and optimal priori probability input signals for PAM modulation scheme with M = 3 and M = 4.12: The comparison of cutoﬀ rate for uniform and optimal priori probability for input signals. This indicates that the suitable selection of a priori probability for the input signals is crucial especially for the bad channel condition. the cutoﬀ rate of smaller input signals is higher then the larger input signals due to larger distance between signals. when the channel SNR is high.132 According to the previous derivation.12. In Fig. the cutoﬀ rate of smaller input signals is lower then the larger input signals due to high bandwidth eﬃciency. 10 1 10 0 R0 10 −1 10 −10 −2 Unif.4. + Ec ].M=4 −5 0 5 10 15 E /N (dB) c 0 20 25 30 Figure 4. M=3 Opt.

(4. we will consider the case (2). 1i ] − λ2 det[D. we obtain a system of linear equations of Pi M M (4.=i Setting the derivative to 0.=i Ds. . the Pi can be expressed as −λ1 − λ2 E1 −λ1 − λ2 E2 det[D. The optimization problem of this case can be formulated as follows: R0 = max{− log[ Pj j k Pj Pj Pk Dj. In order to minimize PT DP with these two constraints. we write the constrained minimization using Lagrange multipliers as the minimization of M M M J(p) = i=1 Pi2 +2 1≤i<j≤M Di. we get ∂J Ds. under constraint Pj = 1. (4.47) where Ej is the energy used in j-th signal and Eav is the average energy constraint.46) = max{− log[PT DP]}. (4.51) .j Pi Pj + λ1 ( i=1 Pi − 1) + λ2 ( i=1 Pi Ei − Eav ). = 2Pi + ∂Pi s=1. j and j Pj Ej = PT E = Eav . . (4.k ]} (4. Ei ] detD .i Ps + λ1 + λ2 Ei .i Ps = −λ1 − λ2 Ei . . for 1 ≤ i ≤ M .50) Solving this. −λ1 − λ2 EM ] i Pi = = detD −λ1 det[D.49) 2Pi + s=1.133 Now.48) Diﬀerentiating with respect to Pi .

2 for all i (4.7 Numerical Results In this section.51) into the two constraints. The channel model used for the numerical evaluation is the . b= e −det[D. we can solve for λ1 and λ2 . .134 where Di.=i Ds. In Fig. · · · . Ei ] is obtained by replacing its i-th column by the energy vector E = [E1 .12 ] e detD T b 1 bT E . . EM ]T .52) .54) According to the previous derivation. bT 1 bT E T a 1 1 det aT E Eav λ2 = T T a 1 b 1 det aT E bT E (4.i Ps ≥ −λ1 − λ2 Ei .i = 2 for 1 ≤ i ≤ M .SW-MARQ and repetition ARQ over time varying channels. The solutions are 1 det Eav λ1 = T a 1 det aT E where a= e −det[D. we can calculate the cutoﬀ rate for a optimal priori probability with average energy constraints. . the necessary and suﬃcient conditions on the probability vector P that minimizes PT DP are M Pi + s=1.4.1M ] e detD e −det[D. . Substituting (4. (4.E1 ] e detD e −det[D.13.E2 ] e detD . E2 .53) e −det[D. 4. .EM ] e detD From the Kuhn-Tucker theorem.11 ] e detD e −det[D. we will compare the system performance of SW-ARQ . The matrix [D. The modulation adopted in the transmitter is BPSK.j = Di.j for i = j and Di. we plot the cutoﬀ rate for diﬀerent average energy constraints for 8 QAM.

135 10 1 10 0 R 0 10 −1 10 −10 −2 0.B} = 1 − log2 (1 + e−Ec. the packet error probability can be written as: Pe.N.6 Ec/N0 0.{G.SR ≤ 2K−N R0.{GN.B}.BN }.4 Ec/N0 0. Let K be the number of information bits in a packet and N be the number of coded bits for this packet.{GN.N.3 Ec/N0 0. (4.6 by assuming that the receiver has perfect knowledge of the channel state information.56) where R0.5 Ec/N0 0. For the receiver with memory. then there exist an encoder and decoder that has packet error probability Pe.SR .N.BN }.7 Ec/N0 −5 0 5 10 Ec/N0 15 20 25 30 Figure 4.{G.B} ≤ 2K−N R0.N bounded by Pe.K.K. All packet error probabilities are evaluated with the random coding bound derived in Section 4.55) where R0.K.{G.{G.{G.B} /N0 ) is the cutoﬀ rate determined by the channel SNR which will depend on the packet is transmitted in the good or bad channel state.SR is the cutoﬀ rate determined by the channel SNR of transmitting the describe packet with packet length N and the receiver memory state .B} .13: The comparison of cutoﬀ rate for diﬀerent average energy constraints for 8 QAM two state Gilbert-Elliott model. (4.

G . We ﬁrst ﬁx Ec /N0 and optimize the average delay to get the corresponding packet lengths for Ni . For example.4.55) and (4. then the packet error probability becomes large.0001 and r = 2.GN2 . the two systems have almost the same performance because the transmitter has a higher chance to get an ACK on the ﬁrst transmission with higher channel SNR. m = 2. The eﬀect of the receiver memory improvement to the system performance goes away due to the lack of the receiver memory.14. is Pe. All the following ﬁgures show the energy-delay curves when the bad channel SNR varies from −3 to 3 dB.B +N2 R0.BN1 where R0. mR = 1. Therefore. the packet error probability of transmitting a packet with length N2 through the good channel when the memory content is BN1 (the ﬁrst packet with length N1 is transmitted through the bad channel and decoded incorrectly). we get the energy-delay curve. The system uses the following parameters. We also note that the normalized average delay and . N1 +N2 From the packet error probability formulas (4.6. λ = µ = 0. This increases the system delay and energy due to the high possibility of packet retransmission.136 SR as derived in Section 4. Our goal is to ﬁnd the best Ni to minimize the average delay. We observe that the average delay of MARQ at low channel SNR is smaller by 20% compared to ARQ. although the packet error probability is small. The reason is that if N is small. MARQ outperforms ARQ since MARQ can decode more eﬃciently by utilizing the previous erroneous packets.56). On the other hand. the transmission time per packet is large. From this. Repeating the above argument for diﬀerent Ec /N0 .GN2 . we can evaluate both average delay and average energy consumption.BN1 = N1 R0. if N is large.BN1 = 2K−(N1 +N2 )R0.GN2 . At high channel SNR. We plot SW-ARQ and SW-MARQ for diﬀerent K in Fig.K. we observe that there is an optimal N for each information packet that minimizes the average system delay.

(4.K=75 ARQ. In [15]. We plot energy-delay curves of SW-ARQ and SW-MARQ for diﬀerent λ in Fig.55).137 energy of systems decreases as K increases.K=100 7 6.5 MARQ.avg /K = 1+ log2 (ln K) K R0 . Because the average performance (delay or energy) of ARQ systems over time varying channels could be treated as a mix of performance of ARQ systems over a good channel and a bad channel with some proportion to each. r=2 6 t. m = 2. we showed the optimal error probability that achieves the minimum average delay is approximated as 1 1+K ln 2 when the packet error probability represented by (4.K=50 MARQ.001.K=75 MARQ.5 /K (µ secs) DT 3 3.5 5 4. K = 100 and r = 2(3dB). 7.K=100 ARQ.5 2 T d.avg /K E DT 5. The systems use following parameters.K=50 ARQ.4. hence. we can express the normalized average delay with respect to the number of information bits K as Td.15 with a ﬁxed average channel SNR.57).5 (dB) λ=µ=0. From (4. we know that the normalized delay decreases as K increases for any channel condition (cutoﬀ rate). mR = 1.57) for a ﬁxed channel condition with cutoﬀ rate R0 .5 4 Figure 4. this explains why the normalized average values decrease as K increases as indicated in the ﬁgure.avg 2.5 4 1 1. From this. We observe that the average delay and energy .14: Energy-Delay curves for diﬀerent KDT .

K = 100 and λ = 0.005 ARQ.15: Energy-Delay curves for diﬀerent λ with ﬁxed average SNR.5 4 Figure 4.0001 ARQ. the systems have higher chance to transmit a packet successfully on the ﬁrst attempt. λ=0.138 decreases with the increase of λ. The reason is that the average delay and energy values are the function of λ and they decrease as λ increases as shown in Fig.5 2 2. Fig. mR = 1. When the channel SNR becomes large.4. λ=0.0001 MARQ.005 MARQ.5 5 4. The systems use following parameters.001.16. 4. λ=0.5 MARQ. We also observe that MARQ systems are more sensitive to changes of parameters (λ and µ) of time varying channels because the MARQ systems can improve the system performance signiﬁcantly compared to ARQ systems due to .01 7 6.5 6 Et. 7.5 1 1. λ=0. m = 2. λ=0.avg/KDT (µ secs) 3 3. λ=0. We ﬁnd that the energy-delay curves have largest values for λ = 4µ because the channel has a probability of 1 3 4 5 to stay in the bad channel state compared to 1 2 for λ = µ and for µ = 2λ. The eﬀects of λ and µ to the average energy and delay become less signiﬁcant as the channel SNR increases. therefore the average values aﬀected by λ and µ through channel transition probabilities become less signiﬁcant.5 4 3.17 shows the energy-delay curves for diﬀerent average channel SNRs.5 Td.avg/KDT (dB) λ=µ. K=100 5.01 ARQ.

mR = 1. Accordingly. It shows that the average delay and energy consumed by the systems are proportional to the quantity Nrt since this is a go-back-Nrt protocol.18. we need to retransmit all the . Therefore. we see that the system has worst performance when it does not utilize the previous error packet information (memoryless at the receiver). The eﬀect of diﬀerent Nrt for GBN-ARQ is shown in Fig. 4. 4. For the receiver with memory. The comparison of three diﬀerent repetition ARQ schemes is given in Fig. The systems uses the following parameters. mR = 1. the R-ARQ-SWM has better performance compared to R-ARQ-NOW because R-ARQ-NOW refreshes the memory buﬀer with the repetition period of packet lengths even the previous packet is erroneous received. it cannot utilize all the information of previous erroneous packets.18. m = 2. better error correction capability.16: Normalized average delay relations with λ. 4.0001. MARQ has better performance but is more sensitive than ARQ.139 10 5 µ=4λ µ=λ µ=1/4λ 10 4 10 d.avg 3 T /K DT 10 2 10 1 10 −7 10 0 10 −6 10 −5 10 −4 10 λ −3 10 −2 10 −1 10 0 Figure 4. K = 100 and λ = 0.001. K = 100 and λ = 0. The systems use following parameters. m = 2.18. From Fig. Therefore.

λ=4µ MARQ. µ=2λ Et. λ=4µ ARQ. λ=µ MARQ.avg/KDT (dB) 6 5.λ=0.5 Figure 4.5 2 2.5 Figure 4.avg E /K 8 7 6 5 1 1.5 Td. µ=2λ ARQ.0001 9 (dB) DT t.5 5 4.5 2 T d.5 1 1.18: Energy-Delay curves for repetition ARQ. λ=µ ARQ.17: Energy-Delay curves for diﬀerent average channel SNRs.avg /K DT 2.5 4 3.5 7 6.avg/KDT (µ secs) 3 3. .5 MARQ. 11 RARQ RARQSWM RARQNOM 10 K=100.140 7.5 (µ secs) 3 3.

avg/KDT (µ secs) 12 14 7 2 Figure 4. In other words.Nrt=2 MARQ. receiver and channel. the numerical results demonstrated the time-varying characteristics of the channel have a great inﬂuence on the system performance. According to the state diagram constructed from the FSM. 4. we derived the joint generating function of energy and delay. 14 13 12 Et. the system needs to spend Nrt times in delay and energy for each packet transmission.Nrt=4 4 6 8 10 Td. This comprehensive approach allows us to analyze the overall system by taking into account the states of the transmitter.N =2 rt ARQ. Finally.N =3 rt MARQ. The generating function was used to obtain the energy and delay tradeoﬀ curves of an ARQ system. .19: Energy-Delay curves for GBNARQ with diﬀerent Nrt .avg/KDT (dB) 11 10 9 8 MARQ.141 following Nrt packets when the transmitter receives a NACK packet.N =4 rt ARQ.8 Conclusion In this chapter we modeled an ARQ system with a FSM to obtain the joint distribution of energy and delay for packet transmission over time varying channels.N =3 rt ARQ.

Furthermore. In Chapter 2. In the ﬁrst part (Chapter 2 and 3) of this thesis. In the second part (Chapter 4) of this thesis. we studied an ARQ communication system. we analyzed the energy and delay for a communication network. we proposed a state diagram representation of the operation of the MAC layer to obtain the joint generating function of the energy and delay by considering the eﬀects of the PHY and MAC layer jointly. Using these analysis tools. we began by extending the approach proposed in [22] to determine average energy and delay of a successful packet transmission by taking into consideration the dependence of packet error probability on energy and delay. channel coding was incorporated in the analysis by estimating the packet error probability using error-exponent-based bounds under an AWGN channel. We investigated the energy and delay tradeoﬀ of a wireless communication system.CHAPTER V Conclusion and Future Research In this chapter. Thereupon. the code rate and the 142 . 5. we conclude this dissertation by summarizing the content of the thesis and proposing possible future research directions.1 Contributions The most important contribution of this thesis is to represent the operation of a system by a state diagram and use the generating function approach to derive the statistics of the energy and delay consumption.

say channel SNR. By using Reed-Solomon codes we optimized the system performance over the code rate and energy per coded bit.11 protocol in which the transmitter increases the energy level per coded symbol whenever it suﬀers an unsuccessful transmission. By changing the signal-to-noise ratio.11 original protocol and the proposed one have almost identical performances and are equally sensitive to the knowledge of the channel quality. the framework of our analysis can be used for other coding and modulation schemes over various wireless channels. we applied our proposed generating analysis method to analyze and design other wireless network protocols. We ﬁrst designed an energy adaptation scheme with original IEEE 802. The numerical results showed that the proposed protocol can improve the energy and delay throughput of the system signiﬁcantly when the channel condition is bad. The comparison of the energy-delay tradeoﬀ curves evaluated from this approximation method with the exact energy-delay tradeoﬀ curves (from numerical optimization) demonstrated that this approximation method is extremely accurate especially when the number of information bits per packet is large. an approximation method to express the energy-delay tradeoﬀ curves analytically was provided. at the transmit- .143 signal-to-noise ratio per dimension were optimized to achieve minimum average delay per information bit.11 protocol. The numerical results showed that the IEEE 802. The purpose of this analysis is to demonstrate that the generating function approach can be applied to analyze more protocol layers jointly by including the analysis of the logical link control (LLC) sublayer into the original protocol (MAC and PHY layers only). the energy-delay tradeoﬀ curves for minimum delay were obtained. Finally. In Chapter 3. Although we used Reed-Solomon codes over AWGN channel to represent packet error probabilities. The second proposed protocol is an ARQ scheme for data packets transmission with 802.

we modeled an ARQ system with a FSM to obtain the joint distribution of energy and delay for successful packet transmission over a time varying channels. we derived the joint generating function of energy and delay. the numerical results demonstrated the time-varying characteristics of the channel have a great inﬂuence on the system performance.2 Future Research In this thesis. we also can investigate the best energy allocation strategy for transmitting stations. packet error probabilities are represented by random coding bounds. such as LDPC or turbo codes. For a diﬀerent traﬃc model. By regulating the desired quality of services (QoS). we can further consider the delay spent in a buﬀer. The cutoﬀ rate for two diﬀerent memory receiver structures was derived and the optimization of the cutoﬀ rate under average energy constraint was discussed. . In Chapter 4.144 ter. While ARQ provides some robustness to SNR uncertainty in a conventional single user system. this is not the case in this multiuser scenario. One extension is to consider the packet error probability for speciﬁc coding schemes. We also observed that the reduction in delay of a SWARQ-FT system achieved by immediately retransmitting an erroneously received packet. The other research direction that can be explored is to relax the assumption of the saturation traﬃc by analyzing the non-saturation traﬃc. receiver and channel. since other users use the same protocol. Finally. The generating function was used to obtain the energy and delay tradeoﬀ curves of an ARQ system. This comprehensive approach allows us to analyze the overall system taking into account the states of transmitter. is compensated by the increased delay needed to access the channel. According to the state diagram constructed from FSM. 5.

In particular. The spatial dimension provided by multiple antennas is known to dramatically increase system capacity. 14. We believe that the proposed generating function method described in this thesis could be applied to other MAC protocols with more complicated channel conditions. As a result. Future research is needed to quantify how the energy and delay consumption changes when stations are mobile. we do not consider mobility issue in this thesis. the combination of MIMO spatial diversity with temporal diversity has led to widespread development of space−time coding (STC) schemes. For ARQ systems. combining multiple-input multiple-out (MIMO) with ARQ systems has become an important and popular research subject. Therefore. 15] considered the eﬀects of packet errors in their analysis. this assumption is not realistic in practice. In commercial wireless networks. We also can consider other energy consumption. we only considered error free case for the ACK and NACK packets in this thesis. they all assumed that all stations in a wireless network share an identical channel condition. the wireless communication-links between any pair of source and destination stations may diﬀer from each other due to the diﬀerent physical objects in between. However. The energy considered in the thesis is just transmission energy. one possible future research is to relax the restriction of the homogeneous channel condition for all the contending stations by considering the channel inhomogeneity incurred among diﬀerent users. such as hardware processing energy.145 Although some research in [22. One future research direction is to analyze an ARQ system by taking into consideration the nonzero ACK and NACK packet error probabilities and the energy and delay consumption for ACK and NACK packets. however. antenna switching energy between transmitting and receiving modes. Many approaches have been . mobility is an important issue. 39.

Another future research direction is to develop some new ARQ protocols for MIMO systems which adapt the signal constellation of each ARQ retransmission.146 proposed for STC. and space−time block coding. space−time bit-interleaved coded modulation (BICM). space−time trellis coding. . for example.

APPENDICES 147 .

we consider a system with some speciﬁc parameters.G1 ) + PB (1 − Pe.3. in this system and T (i1 ) = T (2) = 2. From the state transition diagram shown in Fig. · · · }. 4. In order to illustrate the meaning for those formulas. Y ) and G2 (X. We ﬁrst derive the ﬁrst part of the generating function for those successful transmissions which are not belonged to (2 + k · 1 + 1). Y ) = (PG (1 − Pe.1) . The generating function for the event that the packet is transmitted successfully in the ﬁrst transmission is G1 (X. · · · } into a group since these transmissions use same packet length N2 to transmission. (A.148 APPENDIX A Generating Function Analysis for NR-ARQ Systems A. g1 .B1 ))X N1 Tb Y N1 Ec . we use P = 1 − P for an event with probability value P .1 Memoryless Receiver In this section. 2. For all appendices. We assume that the sequence of packet lengths used for each transmission is {N1 . we can group all transmissions with transmission indices from as (2 + k · 1) for k ∈ {0. 1. We can see that there is only one group.3. we need to evaluate the generating functions for G1 (X. we will determine the joint generating function of the energy consumed and delay incurred for a NR-ARQ system without the receiver memory according to the general formulas provided in Chapter 4. N2 . Therefore. Y ). N2 .

B2 e ) and the initial conditions are Af.(A.G2 and Gf.B1 (1 − e−µN1 Tb )]Pe. · · · }.B2 X (N1 +N2 )Tb Y (N1 +N2 )Ec . 1.G2 and Gf.B2 by solving a set of linear equations. µN2 Tb µN2 Tb Pe. 2.G1 e−λN1 Tb + PB Pe. For transition matrix.9).G1 (1 − e−λN1 Tb ) + PB Pe. we need to evaluate the generating function of G2 by using the recursion formula shown in (4.B2 for further notational simplicity since this is a memoryless system.3) Pe.G1 e−λN1 Tb + PB Pe.B2 = [PG Pe. Because min i1 = 2.149 where PG and PB are the probabilities that the initial transmission was in good and bad states. Y ) = [PG Pe. (A.12) and obtain G2 (X. it is Pe.G2 and Gf. From the systems of equations shown in (4.G2 = [PG Pe.B2 will be determined if the transition probability matrix for the system states and the initial conditions are given.B1 e−µN1 Tb ]Pe.G2 X (N1 +N2 )Tb Y (N1 +N2 )Ec + [PG Pe.G1 e−λN1 Tb + PB Pe.B2 X (N1 +N2 )Tb Y (N1 +N2 )Ec .4) . respectively.B2 (1 − e ) Pe. we need to obtain the generating functions of Gf. We omit the memory content parts in Gf.G2 eλN2 Tb (A. We assume that PG is equal to stationary probability of µ λ good channel ( λ+µ ) and PB is equal to stationary probability of bad channel ( λ+µ ) if the system has operated for a long time period already.G2 (1 − eλN2 Tb ) .B1 e−µN1 Tb )]Pe.G2 X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.B1 e−µN1 Tb ]Pe. the generating functions of Gf.2) For the second part of the generating function for those successful transmissions with transmission indices form as (2+k ·1+1) for k ∈ {0.

B2 (1 − e−µN2 Tb )(1 − Pe. They are G1 (X. G2 . We ﬁrst derive the ﬁrst part of the generating functions for those successful transmissions which are not belonged to indices (3 + k · 1).5) A. we can group all transmissions with transmission indices from as (3 + k · 1) for k ∈ {0. Y ) = [PG Pe.B2 |B1 e−µN2 Tb Pe.B2 e−µN2Tb (1 − Pe. N2 .G1 (e−λN1 Tb Pe.G2 e−λN2 Tb (1 − Pe. N2 . · · · } into a group since these transmissions use same packet length N2 to transmission. From the state transition diagram shown in Fig. 1. 2. · · · }.G2 |G1 e−λN2 Tb Pe.B2 |G2 + e−µN1 Tb Pe. Gf. Since i1 = 3. in this system and T (i1 ) = T (3) = 2.G2 |G2 + e−µN1 Tb Pe.150 Therefore.6.B2 |B2 )] · X (N1 +N2 +N3 )Tb Y (N1 +N2 +N3 )Ec .B2 )]X N2 Tb Y N2 Ec(A. g1 .B2 ) + Gf. Y ) = (PG (1 − Pe.B1 ((1 − e−µN1 Tb )(1 − Pe.G2 |G1 ) + (1 − e−λN1 Tb )(1 − Pe.B2 |G1 e−µN2 Tb Pe.G2 |B1 e−λN2 Tb Pe.B2 |G1 )) + PB Pe. (A.G2 ) + Gf. we need to evaluate G1 .G2 |G1 e−λN2 Tb Pe.G2 (1 − e−λN2 Tb )(1 − Pe.G1 (e−λN1 Tb (1 − Pe.B2 |G1 e−µN2 Tb Pe.B2 |B2 ) + PB Pe.B2 |G2 + e−λN1 Tb Pe. We assume that the sequence of packet lengths used for each transmission is {N1 .12). We can see that there is only one group. we will determine the joint generating function of the energy and delay required for a NR-ARQ system with the receiver memory.G2 |B2 + e−µN1 Tb Pe. the total generating function for this system is Gs = G1 + G2 + [Gf.G2 |B1 e−λN2 Tb Pe.G1 ) + PB (1 − Pe.B2 |B1 e−λN2 Tb Pe. Y ) = [PG Pe.G2 ) + .B2 |B1 ))] · X (N1 +N2 )Tb Y (N1 +N2 )Ec G3 (X.G2 |B2 + e−λN1 Tb Pe.B1 (e−µN1 Tb Pe.B1 ))X N1 Tb Y N1 Ec G2 (X.G2 |G2 + e−λN1 Tb Pe. 4.6) . and G3 recursively from (4.G2 |B1 ) + e−µN1 Tb (1 − Pe.2 Memory Receiver In this section.

G2 |G2 · X (N1 +2N2 )Tb Y (N1 +2N2 )Ec Af.8) .G2 |G2 .B2 |B2 by solving a set of linear equations.G2 |G1 + PB Pe. 1.B2 |B2 will be determined if the transition probability matrix for the system states and the initial conditions are given.G2 |B2 = (PG Pe.B1 e−µN1 Tb Pe.G2 |B2 and Gf.G2 |B1 )e−λN2 Tb Pe.G2 |G1 + PB Pe.B2 |G2 .B2 |B2 · X (N1 +2N2 )Tb Y (N1 +2N2 )Ec (A.B2 |B2 eλN2 Tb and the initial conditions are Af.9).B2 |G2 eµN2 Tb 0 Pe.G1 e−λN1 Tb Pe.B1 e−µN1 Tb Pe.G2 |G2 = (PG Pe.G2 |B2 eλN2 Tb 0 Pe.B2 |G2 · X (N1 +2N2 )Tb Y (N1 +2N2 )Ec Af.B2 |B2 eλN2 Tb 0 Pe.B2 |G1 + PB Pe.G2 |B2 · X (N1 +2N2 )Tb Y (N1 +2N2 )Ec Af.G2 |G2 eλN2 Tb 0 Pe.G2 |B2 eλN2 Tb 0 Pe.B2 |G1 + PB Pe.G1 e−λN1 Tb Pe.G1 e−λN1 Tb Pe.G2 |G2 .G2 |B2 and Gf. it is λN2 Tb 0 Pe.G2 |G2 e 0 Pe. 2. Gf. For transition matrix.B2 |B1 )e−µN2 Tb Pe.B2 |G2 . Gf. · · · }.G2 |B1 )e−λN2 Tb Pe. From the systems of equations shown in (4. Gf. we need to obtain the generating functions of Gf.B1 e−µN1 Tb Pe.7) .B2 |G2 = (PG Pe.B1 e−µN1 Tb Pe.G1 e−λN1 Tb Pe.B2 |B1 )e−µN2 Tb Pe.B2 |G2 eµN2 Tb 0 Pe.151 For the second part of the generating function for those successful transmissions with transmission indices form as (3 + k · 1 + 1) for k ∈ {0. (A. Gf.B2 |B2 = (PG Pe. the generating functions of Gf.

G2 |G2 ) + Gf.G2 |B2 )]X N2 Tb Y N2 Ec .B2 |B2 e−µN2 Tb (1 − Pe.G2 |G2 e−λN2 Tb (1 − Pe.G2 |B2 e−λN2 Tb (1 − Pe.9) .B2 |B2 ) + Gf.B2 |B2 ) + Gf. the total generating function for this system is Gs = G1 + G2 + G3 + [Gf. (A.B2 |B2 e−µN2 Tb (1 − Pe.G2 |B2 ) + Gf.G2 |G2 e−λN2 Tb (1 − Pe.B2 |G2 ) + Gf.B2 |G2 e−µN2 Tb (1 − Pe.G2 |G2 ) + Gf.B2 |G2 ) + Gf.152 Therefore.B2 |B2 e−µN2 Tb (1 − Pe.G2 |B2 e−λN2 Tb (1 − Pe.

N2 . g1 . we will determine the joint generating function of the energy consumed and delay incurred for a R-ARQ system without the receiver memory according to the general formulas provided in Chapter 4. We consider a system with some speciﬁc parameters by assuming that the sequence of packet lengths used for each transmission is {N1 .G1 ) + PB (1 − Pe.1) where PG and PB are the probabilities that the initial transmission was in good and bad states. Y ). 2. From the state transition diagram shown in Fig. we can group all transmissions with transmission indices from as (1 + k · 2) for k ∈ {0. 1.3. in this system and T (i1 ) = T (1) = 1.8.1 Memoryless Receiver In this section. We assume that PG is equal to stationary probability of . · · · } into a group g1 and (2 + k · 1) for k ∈ {0. (B. 1. There are two groups. g2 . the only term in the ﬁrst part of the generating function is G1 (X. T (i2 ) = T (2) = 2. We ﬁrst derive the ﬁrst part of the generating function. N2 .153 APPENDIX B Generating Function Analysis for R-ARQ Systems B. N1 . Y ) = (PG (1 − Pe.B1 ))X N1 Tb Y N1 Ec . 4. 2. i2 = 2. N1 . · · · } into another group g2 . N2 . · · · }. respectively. Since i1 = 1. The generating function for the event that the packet is transmitted successfully in the ﬁrst transmission is G1 (X.

Mg1 .B1 Pe. we need to obtain the generating functions of Gf.G1 and Gf. 1.B2 by solving a set of linear equations. for the system states and the initial conditions are given.2) Similarly.B2 e−λN1 Tb e−µN2 Tb 12 cg1 = Pe.B1 Pe. the generating functions of Gf.G2 e−λN1 Tb e−λN2 Tb + Pe. 2.G1 Pe. for those successful transmissions with transmission indices form as (2 + k · 2 + 1) for k ∈ {0.G1 Pe. for the system states and the initial conditions .B1 X N1 Tb Y N1 Ec .3) (B.B1 Pe. Similarly. Mg2 . For the second part of the generating function for those successful transmissions with transmission indices form as (1 + k · 2 + 1) for k ∈ {0.G2 e−λN1 Tb e−λN2 Tb + Pe.9). 2. it is g1 g1 c11 c12 Mg1 = . 22 And the initial conditions are Af.B2 e−µN1 Tb e−µN2 Tb .G1 X N1 Tb Y N1 Ec Af.B1 Pe.B2 e−µN1 Tb e−µN2 Tb 21 cg1 = Pe.154 µ λ good channel ( λ+µ ) and PB is equal to stationary probability of bad channel ( λ+µ ) if the system has operated for a long time period already.G2 e−µN1 Tb e−λN2 Tb + Pe. g1 g1 c21 c22 where cg1 = Pe. For transition matrix.G1 Pe. 1. · · · }. we need to obtain the generating functions of Gf.B2 will be determined if the transition probability matrix. · · · }.G1 and Gf.G2 and Gf.G2 e−µN1 Tb e−λN2 Tb + Pe.4) (B.B1 by solving a set of linear equations. the generating functions of Gf.G1 Pe. From the systems of equations shown in (4.B1 will be determined if the transition probability matrix.G1 = PG Pe. (B.B2 e−λN1 Tb e−µN2 Tb 11 cg1 = Pe.B1 = PB Pe.G2 and Gf.

N2 .B1 (e−µN1 Tb Pe.B1 e−λN2 Tb e−µN1 Tb 11 cg2 = Pe.G2 (e−λN2 Tb Pe.B1 e−µN2 Tb e−µN1 Tb .B2 = [PG Pe.G2 Pe.B1 e−µN2 Tb e−µN1 Tb 21 cg2 = Pe.G2 + e−λN1 Tb Pe.G2 Pe.B2 )]X N2 Tb Y N2 Ec + [Gf.6) cg2 11 cg2 21 cg2 12 cg2 22 .G1 (1 − e−λN1 Tb ) + PB Pe.2 Memory Receiver : Sliding Window (B.B1 e−µN1 Tb ]Pe.G1 (e−λN1 Tb Pe.G2 Pe.5) In this section. For transition matrix.G1 e−µN2 Tb e−λN1 Tb + Pe.G1 e−λN2 Tb e−λN1 Tb + Pe.8) (B.G1 e−λN1 Tb + PB Pe.B1 ) Gf.B2 (e−µN2 Tb Pe.B2 X (N1 +N2 )Tb Y (N1 +N2 )Ec . N2 .G2 = [PG Pe.G1 e−µN2 Tb e−λN1 Tb + Pe.B2 Pe. (B. 22 And the initial conditions are Af. we will determine the generating function for a R-ARQ system with the receiver memory which updates its memory contents by the sliding window operation.7) Therefore.B2 Pe.B1 )]X N1 Tb Y N1 Ec B.G1 + e−µN2 Tb Pe.G2 Pe.B1 (1 − e−µN1 Tb )]Pe.G2 + e−µN1 Tb Pe.B2 ) Gf. N1 . We assume that the sequence of packet lengths used for each transmission is {N1 . . it is Mg2 = where cg2 = Pe.B2 Pe.B2 Pe. N2 . N1 .155 are given. the total generating function for this system is Gs = G1 + [Gf.G1 e−λN2 Tb e−λN1 Tb + Pe.G1 + e−λN2 Tb Pe. · · · }. (B.B1 e−λN2 Tb e−µN1 Tb 12 cg2 = Pe.G2 X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.

2. Y ) = [PG Pe. 4. 2. and Gf.G2 |G1 . T (i2 ) = T (3) = 1. Gf. Gf.156 From the state transition diagram shown in Fig. 1. and Gf.G2 |G1 . Gf. the terms in the ﬁrst part of the generating function are G1 .G2 |B1 . Gf.G2 |B1 )]X (N1 +N2 )Tb Y (N1 +N2 )Ec . 1. There are two groups. respectively. For the second part of the generating function for those successful transmissions with transmission indices form as (2 + k · 2 + 1) for k ∈ {0.9).G2 |B1 . We ﬁrst derive the ﬁrst part of the generating function.B1 |G2 . We assume that PG is equal to stationary probability of µ λ good channel ( λ+µ ) and PB is equal to stationary probability of bad channel ( λ+µ ) if the system has operated for a long time period already. Since i1 = 2. we need to obtain the generating functions of Gf.G1 (e−λN1 Tb Pe. Mg1 . · · · } into another group g2 .9) where PG and PB are the probabilities that the initial transmission was in good and bad states. · · · } into a group g1 and (3 + k · 2) for k ∈ {0.B2 |B1 will be determined if the transition probability matrix. 2.B2 |G1 . Gf.G2 |G1 + e−λN1 Tb Pe.G1 |G2 . From the systems of equations shown in (4. for those successful transmissions with transmission indices form as (3 + k · 2 + 1) for k ∈ {0. we can group all transmissions with transmission indices from as (2 + k · 2) for k ∈ {0. For transition matrix.B1 ))X N1 Tb Y N1 Ec G2 (X. They are G1 (X. G2 .G1 ) + PB (1 − Pe.B2 |B1 + e−µN1 Tb Pe. in this system and T (i1 ) = T (2) = 2.B2 |G1 ) + PB Pe.B2 |G1 . . and Gf. · · · }. Similarly. i2 = 3. Y ) = (PG (1 − Pe. · · · }. g1 . we need to obtain the generating functions of Gf.B2 |B1 by solving a set of linear equations. 1. the generating functions of Gf. for the system states and the initial conditions are given. Gf. (B.B1 |B2 by solving a set of linear equations. 1.B1 (e−µN1 Tb Pe.9.G1 |B2 . g2 . 2.

G1 |G2 e−λN2 Tb e−λN1 Tb .G2 |B1 Pe. 41 cg1 = Pe.G2 |G1 = PG Pe.G1 |G2 .G1 Pe.B2 |B1 = PB Pe.B2 |G1 Pe. and Gf.B2 |G1 Pe.B1 Pe.G2 |B1 Pe. 2. (B.B2 |B1 e−µN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec . Mg1 (B. .B1 |G2 e−λN2 Tb e−µN1 Tb 24 cg1 = Pe. 23 cg1 = Pe. cg1 = Pe.G1 |B2 e−µN2 Tb e−λN1 Tb .12) cg1 = Pe.B1 Pe. 11 cg1 = Pe.G1 |B2 .B1 |B2 by solving a set of linear equations.B2 |B1 Pe. cg1 = Pe.11) Similarly. Gf. And the initial conditions are Af.G2 |G1 e−λN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.B1 |B2 e−µN2 Tb e−µN1 Tb 42 cg1 = Pe.B2 |G1 e−λN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.B2 |B1 Pe. for those successful transmissions with transmission indices form as (3 + k · 2 + 1) for k ∈ {0. 43 .G2 |G1 Pe.B1 |B2 e−µN2 Tb e−µN1 Tb 44 (B. · · · }.157 it is g1 g1 g1 g1 c21 c22 c23 c24 = g c 1 cg1 cg1 cg1 31 32 33 34 cg1 cg1 cg1 cg1 41 42 43 44 cg1 11 cg1 12 cg1 13 cg1 14 .G2 |G1 Pe.G2 |B1 Pe.10) where cg1 = Pe.G1 |B2 e−µN2 Tb e−λN1 Tb .B2 |G1 Pe.B1 |G2 e−λN2 Tb e−µN1 Tb 22 cg1 = Pe.B2 |G1 = PG Pe.B2 |B1 Pe.G1 |G2 e−λN2 Tb e−λN1 Tb .G1 |B2 e−µN2 Tb e−λN1 Tb .B1 |G2 .G1 |G2 e−λN2 Tb e−λN1 Tb .B2 |B1 Pe.G1 |G2 e−λN2 Tb e−λN1 Tb .G1 |B2 e−µN2 Tb e−λN1 Tb .B1 |G2 e−λN2 Tb e−µN1 Tb 13 14 cg1 = Pe.G1 Pe.G2 |G1 Pe.G2 |G1 Pe.G2 |B1 Pe.G2 |B1 = PB Pe. we need to obtain the generating functions of Gf. 21 cg1 = Pe.B1 |B2 e−µN2 Tb e−µN1 Tb 32 31 cg1 = Pe. cg1 = Pe.B1 |G2 e−λN2 Tb e−µN1 Tb 12 cg1 = Pe. Gf.G2 |B1 e−µN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.B1 |B2 e−µN2 Tb e−µN1 Tb 33 34 cg1 = Pe. 1.B2 |G1 Pe.

G1 |G2 Pe. cg2 = Pe.B1 |B2 will be determined if the transition probability matrix. Mg2 .B2 |B1 e−µN1 Tb e−µN2 Tb 32 cg2 = Pe.B1 |G2 . cg2 = Pe.B1 |B2 Pe.G1 |B2 .G2 |G1 e−λN1 Tb e−λN2 Tb .G1 |G2 Pe. for the system states and the initial conditions are given.B2 |G1 e−λN1 Tb e−µN2 Tb 21 22 cg2 = Pe.G2 |G1 e−λN1 Tb e−λN2 Tb .B1 |G2 Pe.G1 |B2 Pe. 31 cg2 = Pe.G1 |B2 Pe.G1 |G2 .G2 |G1 e−λN1 Tb e−λN2 Tb . Gf. the generating functions of Gf. it is g2 g2 g2 g2 c21 c22 c23 c24 = g c 2 cg2 cg2 cg2 31 32 33 34 cg2 cg2 cg2 cg2 41 42 43 44 cg2 11 cg2 12 cg2 13 cg2 14 .B2 |B1 e−µN1 Tb e−µN2 Tb .B1 |G2 Pe.158 From the systems of equations shown in (4. cg2 = Pe.13) where cg2 = Pe.B2 |G1 e−λN1 Tb e−µN2 Tb 24 23 cg2 = Pe. 33 cg2 = Pe.G1 |G2 Pe.G1 |B2 Pe.B1 |B2 Pe. cg2 = Pe.B2 |B1 e−µN1 Tb e−µN2 Tb 42 41 cg2 = Pe. cg2 = Pe.G2 |B1 e−µN1 Tb e−λN2 Tb . 43 44 (B.G1 |B2 Pe.B2 |B1 e−µN1 Tb e−µN2 Tb 34 cg2 = Pe. and Gf.G1 |G2 Pe.G2 |B1 e−µN1 Tb e−λN2 Tb .14) .G2 |B1 e−µN1 Tb e−λN2 Tb . cg2 = Pe.G2 |G1 e−λN1 Tb e−λN2 Tb .B2 |G1 e−λN1 Tb e−µN2 Tb 12 11 cg2 = Pe.G2 |B1 e−µN1 Tb e−λN2 Tb . Mg2 (B.B1 |B2 Pe.B1 |G2 Pe. For transition matrix. Gf.B2 |G1 e−λN1 Tb e−µN2 Tb 13 14 cg2 = Pe.B1 |G2 Pe.9).B1 |B2 Pe.

B2 |B1 )(e−µN2 Tb Pe.G2 |B1 )(e−λN2 Tb Pe. we can group all transmissions with transmission indices from as (1 + k · 2) for k ∈ {0. g2 .3 Memory Receiver : Non Overlap (B.B1 Pe.B1 Pe.B1 |G2 ) + (Gf. we will determine the generating function of energy and delay to transmit a data packet successfully for a R-ARQ system with the receiver memory.G2 |G1 + e−λN1 Tb Pe. 1.B2 |G1 e−λN1 Tb e−µN2 Tb + PB Pe.15) In this section.G1 |B2 · X (2N1 +N2 )Tb Y (2N1 +N2 )Ec Af. the total generating function for this system is Gs = G1 (X. .G2 |G1 e−λN1 Tb e−λN2 Tb + PB Pe.B1 Pe. N1 .G1 Pe.16) (B.G1 |B2 = (PG Pe. From the state transition diagram shown in Fig.G1 Pe.10.B2 |B1 )]X N2 Tb Y N2 Ec . T (i2 ) = T (2) = 2. B. 2. 1.B1 |G2 + Gf. Y ) + [(Gf.B2 |G1 + Gf.G2 |B1 e−µN1 Tb e−λN2 Tb )Pe.B1 |G2 · X (2N1 +N2 )Tb Y (2N1 +N2 )Ec Af.G1 |G2 + e−λN2 Tb Pe. There are two groups.B1 |B2 = (PG Pe.G1 |G2 · X (2N1 +N2 )Tb Y (2N1 +N2 )Ec Af. N2 . Y ) + G2 (X. 4.G2 |G1 e−λN1 Tb e−µN2 Tb + PB Pe.B1 |B2 )]X N1 Tb Y N1 Ec + [(Gf.B1 Pe. · · · } into a group g1 and (2 + k · 2) for k ∈ {0.G1 |G2 = (PG Pe. g1 .B2 |B1 e−µN1 Tb e−µN2 Tb )Pe.G2 |G1 + Gf. We assume that the sequence of packet lengths used for each transmission is {N1 .G2 |B1 + e−µN1 Tb Pe.B2 |G1 ) + (Gf.G1 Pe. N2 .159 And the initial conditions are Af.G1 Pe. 2.B2 |B1 e−µN1 Tb e−µN2 Tb )Pe. N2 . N1 .B1 |G2 = (PG Pe.G1 |B2 )(e−λN1 Tb Pe.B1 |B2 )(e−µN1 Tb Pe. · · · } into another group g2 . in this system and T (i1 ) = T (1) = 1.G1 |B2 + e−µN2 Tb Pe.G2 |B1 e−µN1 Tb e−λN2 Tb )Pe.G1 |G2 + Gf.B1 |B2 · X (2N1 +N2 )Tb Y (2N1 +N2 )Ec .B2 |G1 e−λN1 Tb e−µN2 Tb + PB Pe. · · · }. Therefore.

Mg1 . · · · }. Y ) = (PG (1 − Pe. and Gf.G1 and Gf. it is g1 g1 c11 c12 Mg1 = .G1 ) + PB (1 − Pe. for those successful transmissions with transmission indices form as (2 + k · 2 + 1) for k ∈ {0. For the second part of the generating function for those successful transmissions with transmission indices form as (1 + k · 2 + 1) for k ∈ {0.160 We ﬁrst derive the ﬁrst part of the generating function. 1. i2 = 2. (B. Y ). Gf. Similarly.18) .G1 Pe. · · · }.G2 |B1 e−λN1 Tb e−λN2 Tb + Pe.B1 Pe.B1 will be determined if the transition probability matrix. We assume that PG is equal to stationary probability of µ λ good channel ( λ+µ ) and PB is equal to stationary probability of bad channel ( λ+µ ) if the system has operated for a long time period already.B1 Pe.B2 |B1 e−µN1 Tb e−µN2 Tb 22 (B. respectively. It is G1 (X.B1 Pe.G1 and Gf. Since i1 = 1. Gf. From the systems of equations shown in (4. cg1 cg1 21 22 where cg1 = Pe.G1 Pe.G2 |B1 e−µN1 Tb e−λN2 Tb + Pe.B2 |G1 e−λN1 Tb e−µN2 Tb 12 cg1 = Pe. for the system states and the initial conditions are given.B1 ))X N1 Tb Y N1 Ec .B2 |B1 by solving a set of linear equations.B1 Pe. 1.G1 Pe. the only term in the ﬁrst part of the generating function is G1 (X.B2 |G1 .G2 |G1 e−λN1 Tb e−λN2 Tb + Pe.G2 |G1 .17) where PG and PB are the probabilities that the initial transmission was in good and bad states.G2 |B1 . we need to obtain the generating functions of Gf. the generating functions of Gf.G1 Pe.B2 |G1 e−λN1 Tb e−µN2 Tb 11 cg1 = Pe. 2. 2.B2 |B1 e−µN1 Tb e−µN2 Tb 21 cg1 = Pe. For transition matrix.B1 by solving a set of linear equations.9). we need to obtain the generating functions of Gf.G2 |G1 e−λN1 Tb e−λN2 Tb + Pe.19) (B.

31 cg2 = Pe. 33 cg2 = Pe.G1 = PG Pe.G2 |B1 Pe.B1 e−µN2 Tb e−µN2 Tb 42 cg2 = Pe. Mg2 .G2 |G1 Pe.G2 |B1 .B1 e−µN2 Tb e−µN1 Tb 34 cg2 = Pe.G2 |G1 Pe.G2 |G1 Pe.B1 e−λN2 Tb e−µN2 Tb 12 cg2 = Pe.B2 |G1 Pe. 21 cg2 = Pe.G2 |B1 .B1 e−µN2 Tb e−µN1 Tb .9). for those successful transmissions with transmission indices form as (2 + k · 2 + 1) for k ∈ {0.G1 X N1 Tb Y N1 Ec Af.B2 |G1 Pe. and Gf.B1 e−λN2 Tb e−µN1 Tb 13 14 cg2 = Pe.22) 44 .B2 |G1 Pe.G1 e−µN2 Tb e−λN1 Tb .B2 |B1 Pe.G2 |B1 Pe.B2 |G1 .21) where cg2 = Pe.G1 e−λN2 Tb e−λN1 Tb . it is g2 g2 g2 g2 c21 c22 c23 c24 = g c 2 cg2 cg2 cg2 31 32 33 34 cg2 cg2 cg2 cg2 41 42 43 44 cg2 11 cg2 12 cg2 13 cg2 14 . From the systems of equations shown in (4.B1 e−λN2 Tb e−µN1 Tb 24 cg2 = Pe.B1 e−λN2 Tb e−µN1 Tb 22 cg2 = Pe.B1 X N1 Tb Y N1 Ec . for the system states and the initial conditions are given.B2 |B1 will be determined if the transition probability matrix.G2 |G1 . Gf. Gf. 41 cg2 = Pe.B1 = PB Pe.B2 |G1 . the generating functions of Gf. (B. 23 cg2 = Pe.B2 |B1 Pe. we need to obtain the generating functions of Gf.20) Similarly.G1 e−λN2 Tb e−λN1 Tb .G1 e−λN2 Tb e−λN1 Tb . · · · }.G2 |G1 Pe. 43 cg2 = Pe. and Gf.B1 e−µN2 Tb e−µN2 Tb 32 cg2 = Pe.G1 e−λN2 Tb e−λN1 Tb .B2 |G1 Pe.G2 |G1 . Mg2 (B.161 And the initial conditions are Af.G2 |B1 Pe.B2 |B1 by solving a set of linear equations. For transition matrix.G1 e−µN2 Tb e−λN1 Tb . 2. cg2 = Pe.G1 e−µN2 Tb e−λN1 Tb .B2 |G1 Pe.G2 |B1 Pe.(B.G1 e−λN2 Tb e−λN1 Tb . 11 cg2 = Pe.B2 |B1 Pe. 1. Gf. Gf.

G1 + e−λN2 Tb Pe.G2 |G1 + Gf.G2 |B1 + e−µN1 Tb Pe.B2 |G1 = PG Pe. Therefore.B2 |B1 = PB Pe.(B.B2 |B1 )]X N2 Tb Y N2 Ec + [(Gf.B2 |B1 e−µN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec .G1 + e−µN2 Tb Pe.B1 (e−µN1 Tb Pe.23) .G1 Pe.162 And the initial conditions are Af.B1 Pe.B1 Pe. the total generating function for this system is Gs = G1 + [Gf.G1 Pe.B2 |G1 ) + Gf.B2 |B1 )(e−µN2 Tb Pe.B2 |G1 e−λN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.G2 |B1 )(e−λN2 Tb Pe.24) (B.B1 ) + (Gf.G2 |B1 e−µN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.B1 )]X N1 Tb Y N1 Ec .G2 |B1 = PB Pe.B2 |G1 + Gf.G1 (e−λN1 Tb Pe.G2 |G1 = PG Pe.G2 |G1 e−λN1 Tb X (N1 +N2 )Tb Y (N1 +N2 )Ec Af.G2 |G1 + e−λN1 Tb Pe.

BIBLIOGRAPHY 163 .

pp. Oct. Conf. Globecom Conf. Energy-delay analysis of MAC protocols in wireless networks. INFOCOM. Dec.asp. (San Francisco. Bianchi. A buﬀer and energy based scheduling in mobile ad hoc networks over link layer. Oliveri.html. Proc. Stark. [14] S. Fratta. 3027–3031.irda. Performance analysis of IEEE 802. and M. Kim. 407–411. E. 2003. Performance analysis of the IEEE 802. on Telecom.11 wireless LAN: capacity analysis and protocol enhancement. Analysis of priority mechanisms based on diﬀerentiated inter frame spacing in csma-ca. Dec. and A.264 video streaming over 802. [4] IEEE standard for 802.org/what/technical. Y. Communications 34 (1986). [10] P. architecture speciﬁcation (2000). Proc. Mag. G. and P. Bianchi and I. [8] M.11 wireless networks. Bruneel and M. Intn’l Communications Conf. Proc. 2004. Indoor. Performance analysis of an adaptive SR ARQ scheme for time-varying rayleigh fading channels. pp. Chang. R.org/standards/speciﬁcations. Davini.11 CSMA/CA medium access control protocol. Select. Proc. Abouaissa. 3. Advanced Intn’l.. 244–249. Tinnirello. [5] G. [9] H. Brahma. Chakraborty and M. E. Vehicular Tech. Cao and Y.11 wireless lan medium access control (MAC) and physical layer (phy) speciﬁcations. Mar. [6] G. (Dallas. 1999. Liinaharja. Lorenz. Nov. Ho. 10 (1990). CA). IEEE Control Systems. 1997. IEEE J. On throughput performance of some continuous ARQ strategies with repeated transmissions. and W. 2478–2482. [12] X. http://www.11 wireless lan medium access control (MAC) and physical layer (phy) speciﬁcations. [11] F. .. IEEE Trans. IEEE Trans. no. Cross-layer perceptual ARQ for h. pp. M. Finland). Oct. L. June 2001. Cali. Filippi. 1998. Areas Commun. K. [7] G. Communications (2005). IEEE 802. Moeneclaey.11 distributed coordination function. Energy-delay analysis of wireless systems with random coding. E. 535–547. 142–149. [2] Irda sir data speciﬁcation (2000). pp. Proc. Stark. Y. 3270 – 3274. 2006. W.164 BIBLIOGRAPHY wireless applicaion protocol [1] Forum. 1401–1405. pp. and Web Applications and Services.wapforum. C. http://www. 1996. Masala. Gregori. E. Globecom Conf. and E. [13] S. Hachimi. Conti. Bianchi. 69–76. and J. USA). De Martin. M. IEEE Intn’l Symposium on Personal. Taiwan). Conf. Proc. A. S. C. W. Proc. Mar. [15] S. Models of discrete event dynamic systems. 2004. Anastasopolous. Anastaspoulos. [3] IEEE standard for 802. pp. Bucciol. 18 (2000). (Helsinki. Chang. (Submitted). 19–25. Nov. pp. A. E. and Mobile Radio Communications (Taipei.

Dec. 173–182. pp. Italy). 560–570. Chou. Non-saturation and saturation analysis of IEEE 802. S. vol. and E. Modeling the 802. Feedback control of dynamic systems. J. Chen. pp. (Stockholm. Li.11 distributed coordination function in non-saturated conditions. B. [21] F. Quebec. 3. [27] S. no.11 ad hoc networks. Performance modeling of asynchronous data transfer methods of IEEE 802. part II-the hidden terminal problem in carrier sense multiple access and the busy tone solution. Kallel. pp. 2003. Addision Wesely. Taiwan). P. Bakhtiyari. [24] R. Proc. 1–6. P. University of Arizona. . Proc.11 wireless LANs. Indoor. C. [33] L. 219–224. P. Wireless Networks 3 (1997). M Chao. Proc. and Mobile Radio Communications (Taipei. 1417–1433. Costello (eds. 2051–2055. E. 48 (1999). pp. and I. Vehicular Tech. [22] Z. Mobile Computing 5 (2006).11 DCF in fading channel.. 1996. C. SIGMOBILE (Rome. [18] B. 455–466. Cross-layering for performance improvement in multi-hop wireless networks. IEEE Trans. Franklin. 715–717. 121–126. Gupta. Tucson. California. MSWiM (Montreal. Hadzi-Velkov and B. A cross-layer saturation goodput analysis for IEEE 802. pp. MOBIHOC (Lausanne. Sabharwal. Sept. Spasenovski. [20] G. Conf. June 2002. Kanodia. Error control coding: Fundamentals and applications. Hoefel. [23] T. Tobagi. no. Kleinrock and F. pp. Hou. 224–233. A. 2004. USA). Performance evaluation and enhancement of the CAMA/CA MAC protocol for 802.).11a networks. Performance evaluaion of the IEEE 802. [31] H. Sadeghi. 12. 1996. Powell. Wireless Networks 8 (2002). Analysis of memory and incremental redundancy ARQ schemes over a nonstationary channel. Kallel. IEEE Trans. and A.11 wireless local area network protocol. Crow. IEEE Intn’l Symposium on Personal. H. C. Distributed priority scheduling and medium access in ad hoc networks. Saturation throughput-delay analysis of IEEE 802. IEEE Trans. On the scalability of IEEE 802. N. Improving protocol capacity with model-based frame scheduling in IEEE 802. Engelstad and O. Intn’l Communications Conf. 2005. Ho and K. An adaptive quorum-based energy conserving protocol for IEEE 802. Proc. 2005. Vehicular Tech. pp. [32] . Knightly. D. 2004. Sheu. Communications 40 (1992). J. [34] S. Oct. Osterbo. 9 (2005). 200–209. F.165 [16] C. Proc. Kim and J. Canada). Huang and T. [29] [30] .11e edca with starvation prediction. [26] S. [28] V.11 MAC protocol. Granelli and D. Distributed multi-hop scheduling and medium access with delay and throughput constraints. Proc. Packet switching in radio channels. Switzerland). S. Kliazovich. Oct. Link. 392–396.). 891–899. R. Intn’l Symposium on Parallel Architectures. . Sept. Lai. Conf. 1474–1480. 1994. MobiCom (San Diego. Lin and D. Communications Lett. 217–234. Throughput performance of memory ARQ schemes. Proc. pp. Communications 33 (1975). Proc. pp. Mar. [19] P. 1983. Master’s thesis. May 2005. and S.11-operated wlans. Sweden). Chhaya and S. Modeling single-hop wireless networks under rician fading channels.11 ad hoc networks. Proc. AZ. Emami-Naeini (eds. C. [25] L. J. 1. [17] H. 14–19. July 2001. Algorithms and Networks. F.

Huang. Choi. [46] L. J. Y. 20–23. Parallel contention algorithm with csmaka for ofdm based high speed wireless lans. 1879–1884. An eﬀective go-back-N ARQ scheme for variable-errorrate channels. Weinmiller. IEEE Trans.. Wall and J. June 2005. A traﬃc aware. Indoor. 2581–2585. Bahk. C. (Seoul. 737–739. [37] A. B. [49] Lie Hua Wong. IEEE Intn’l Symposium on Personal. on Personal Wireless Communications. Metzner and J. Schlager. 2004. H. R. Performance study of access control in wireless LANs IEEE 802. pp. pp. D. Department of Elec. Intn’l Symposium on World of Wireless Mobile and Multimedia Networks. Chen. . 3455– 3460. Tinnirello and S. Manshaei. J. University of Michigan.11e wlans. S. An adaptive ARQ enhancement to support multimedia traﬃc using 802. Communications 31 (1988). pp. 21–27. pp. A. IEEE Trans. Communications 42 (1994).11 wireless lans. ARQ protocols with adaptive block size perform better over a wide range of bit error rates. May 2005. Rice and S. M. Chung. Communications 43 (1995). Turletti. [52] J. S. Engin. IEEE Trans. A modiﬁed selective-repeat type-II hybrid ARQ system and its perofrmance analysis. May 2005. and M. Chikada. [36] M. Automatic-repeat-request rrror control scheme. and Comp.D. Heidemann. [43] I. Yao. Eﬃcient energy utilization with time constraints in mobile time varying channels. Mobile Networks and Applications 2 (1997). [40] M. 55–67. M. and Mobile Radio Communications. and A. no. Medium access control with coordinated adaptive sleeping for wireless sensor networks. 2005. Proc. [38] J. 593–608.11 MAC and physical layer protocol. Martins and J. Miller. Y.11 DFWMAC and ETSI RES 10 HIPERLAN. 5–17. Tech. Ko. Wolisz. 22 (1984). 2975–2978. J. Proc. [42] C. C. Vucetic. Proc. Mag. energy eﬃcient MAC protocol for wireless sensor networks. B. 1533–1543. Error exponent regions for multi-user channels. Khan. IEEE Network 19 (2005). Analysis of unsaturation throughput of IEEE 802. Intn’l Conf. Estrin.11e wireless networks. Globecom Conf. 49 (2000). [51] W. 12. Sept. Dec. 2003. Veh. pp. Wicker. Communications 38 (1990). 1. D. Lin. Wang and S.. Korea). Costello. Suh and Y. Intn’l Symposium on Circuits and Systems. Intn’l Communications Conf. IEEE Commun. pp. 88–97. [47] Y. IEEE Trans. Yun and S. [39] Qiang Ni. Ye. 2004. On the throughput performance of csma-based wireless local area network with directional antennas and capture eﬀect: A cross-layer analytical approach. IEEE Trans. [41] K. and T. 653–663. Wang. Cantieni. [48] J. Barakat. Communications 39 (1991). Eﬃciency analysis of burst transmissions with block ack in contentionbased 802.166 [35] S. Festag. Mar. 2005. Proc. no. 493–506. G. M. Lin. IEEE Trans. and J. and D. Ph. 1169–1177. Performance analysis of the IEEE 802. pp.11 dcf. An adaptive coding scheme for time-varying channels. 3037–3041. C. J. Networking 12 (2004). Proc. A sequential scheme for adaptive error control over slowly varying channels. C. Proc. [44] B. IEEE/ACM Trans. Alves. Ann Arbor. Sakakibara. 134–138. and A. [50] Y. Jan. thesis. Yamakita. [45] J. Performance analysis and enhancements for IEEE 802.

This framework allows us to optimize over the system parameters for various objective . there are two main themes. Further. such as the average energy with an outage delay constraint. we present a generic framework that allows us to obtain the joint statistics of energy and delay through their joint generating function. Stark Co-Chair: Associate Professor Achilleas Anastasopoulos In this thesis. Several important design tradeoﬀs are studied from the joint generating function. We ﬁrst determine the average delay incurred and the average energy consumed when the eﬀects of an imperfect channel are incorporated in the model. and delay. The ﬁrst part of the thesis is to study the tradeoﬀ between energy and delay for wireless networks.ABSTRACT Generating Function Analysis of Wireless Networks and ARQ Systems by Shihyu Chang Co-Chair: Professor Wayne E. we use the reliability function bounds for diﬀerent channel models on the error probability of a coded system. A network using a request-to-send (RTS) and clear-to-send (CTS) type medium access control (MAC) protocol is considered. In order to incorporate the relation between packet error probability. energy.

functions. From the state transition diagram of the FSM. An approximation method is also proposed to calculate the average energy and average delay analytically. As the numerical results demonstrate. . A ﬁnite state machine (FSM) is used to model the transmitter. the time-varying characteristic of the channel have a great inﬂuence on the system performance. The inhomogeneous and time-varying eﬀects for the tradeoﬀ between energy and delay are also studied. such as average delay. This approximation is found to be quite accurate for a wide range of lengths. The second part of the thesis is to propose an analytical method to determine the joint distribution of the energy and delay of automatic repeat request (ARQ) protocols over time varying channels. We also consider a receiver containing memory of previously received samples and derive the cutoﬀ rate for three diﬀerent receiver structures. receiver and channel. the generating function of energy and delay consumption can be evaluated according to the Manson’s gain formula while incorporating physical layer characteristics (packet error probability as a function of energy and delay).

- Mi Morad Arthesis
- ldpcphdhteiss
- Ben Salem Aymen 2014 Thesis
- Classical Feedback Control With MATLAB [Lurie B., Enright P]
- schedulinginumts-120923103839-phpapp01
- PhD Thesis Eric
- E Book DSP Theory Ios Java Android Sim Final-REV-SPANIAS
- Array Theory
- GIRD Systems CORDIC Tutorial
- Digital Notes
- Learning Material - ITC
- Teaching With Scicoslab
- shihyu_thesis.pdf
- SWITCHMODE Power Supply Reference Manual
- viterbi_1971
- Thesis Matthias
- Emerging Wireless Standards2
- ztran
- harq-2011
- 22.11.23293844 OFDM Based Cognitive Radio for DSA Networks
- BluetoothMobileIP
- X-tel Hsdpa Brochure
- IQ22_pgs52-56
- 24672554-David-K-Cheng-Field-and-Wave-Electromagnetics

Sign up to vote on this title

UsefulNot useful- WirelessHart - Simulating of Process Control w Packet Losses
- Elprince_wirelessprojectcontentsfinal0
- TransWirelessFebruary2010
- Bi an Chi 1996
- A New Dual-channel MAC Protocol for Multihop Adhoc Networks
- Optimizing the Redundancy Using Chorus in Wireless Networks
- C.S0038-B_v1.0_HRPD_SigConf_Spec-090402
- Key Recovery Based Preventing Method for Blocking Attacks
- Addressing the Multichannel Selection, Scheduling and Coordination Cost
- Final Report
- The Impact of Transmission Overheads on IEEE 802.11 Throughput
- Mq 3624532158
- Collision Avoidance and Minimization of Delay using CSMA，CA with Jam Signalling in MANET
- Network Components2
- SSD8 All Muti Choice
- netwarep
- newtworking
- A Documented Network is Easier to Troubleshoot
- EN 2005 to 2007
- quiz1.doc
- desd_questions
- Whitepaper Migration to IP - The Basics
- Lecture 1
- Fundamentals
- Developer Tutorial Aqua-Sim
- Distributed Databases
- Computer Communications Access Mmethods
- 10 Cau Hoi Can Biet
- MiWi
- Computer Networks Networks
- shihyu_thesis.pdf

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd