UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Performance analysis of destination multiplexing for wireless LANs Zhuang, Youli 2003

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2004-0131.pdf [ 2.73MB ]
Metadata
JSON: 831-1.0091330.json
JSON-LD: 831-1.0091330-ld.json
RDF/XML (Pretty): 831-1.0091330-rdf.xml
RDF/JSON: 831-1.0091330-rdf.json
Turtle: 831-1.0091330-turtle.txt
N-Triples: 831-1.0091330-rdf-ntriples.txt
Original Record: 831-1.0091330-source.json
Full Text
831-1.0091330-fulltext.txt
Citation
831-1.0091330.ris

Full Text

PERFORMANCE ANALYSIS OF DESTINATION MULTIPLEXING FOR WIRELESS LANS by YOULI ZHUANG MS in Biomedical Engineering, Shanghai Jiaotong University, Shanghai, PR China, 1988 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE STUDIES DEPARTMENT OF ELECTRICAL ENGINEERING We accept this thesis as conforming to the required standard T H E UNIVERSITY OF BRITISH COLUMBIA November 2003 © Youli Zhuang, 2003 Library Authorization In presenting this thesis in partial fulfillment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. y0(1 Li ZffiiMfr Name of Author (please print) Date (dd/mm/yyyy) Title of Thesis: Performance /hwk& &f- P^n^fli?L Degree: The University of British Columbia Vancouver, BC Canada Year: JL<?z>^ Department of EJe,dricaL GlAine£rtAf Abstract This thesis describes the performance of the IEEE 802.11 Medium Access Control (MAC) protocol with and without destination multiplexing. The IEEE 802.11 MAC protocol, the most widely used standard for Wireless Local Area Networks (WLANs), allows the wireless channel to be effectively shared by portable computers or wireless stations. In this thesis we consider the most common WLAN structure, one where wireless stations connect to a "backbone" wired LAN through a fixed base station or access point (AP). We consider the traffic from APs to wireless stations since typically most of the data flows in this direction. We also take into account fading since it is unavoidable in real wireless channels. An AP using First In First Out (FIFO) packet scheduling transmits or retransmits a data frame until it is successfully received. Fading that lasts for several retransmissions will degrade the performance of the system. Destination multiplexing selects a different destination after a failed transmission. Because of the statistical independence of the fading between the AP and different wireless stations, a transmission to another destination is more likely to be successful. We performed a computer simulation study of throughput and average delay for the overall system. The following factors were studied in our simulations: different data rates, differ-ent number of nodes, collision effect, different data frame length, effect of RTS/CTS control frames, different algorithms in selection of next frame, and the effect of fading channel parame-ters. Our results show that under some conditions destination multiplexing can improve the throughput more than 20 to 30 percent and decrease the average delay significantly. ii Table of Contents Abstract ii List of Tables v List of Figures vi Acknowledgment viii Chapter 1 Introduction 1 1.1 Problem for the traditional frame scheduling 1 1.2 Review of previous work 2 1.3 Objectives 4 Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 6 2.1 Ad-hoc network 6 2.2 Infrastructure Network 7 2.3 MAC Layer 8 2.4 Backoff Timer 12 2.5 IEEE 802.11 simulation model and state machine design 13 2.6 OMNeT++ Introduction. 17 2.7 Validation of IEEE 802.11 WLAN simulation model 18 Chapter 3 Wireless Channel and Destination Multiplexing 20 3.1 Multipath Fading and Shadowing Channel 20 3.2 Gilbert Channel Model 21 3.3 Destination Multiplexing 22 3.4 Retry Algorithm for Destination Multiplexing 23 Chapter 4 Simulation Results and Discussions 29 iii iv 4.1 One AP and four nodes with different data rates 33 4.1.1 IEEE 802.11 FHSS PHY 1 Mbit/sec : 33 4.1.2 IEEE 802.11 DSSS PHY 2 Mbit/sec 35 4.1.3 IEEE 802.11b DSSS PHY 11 Mbit/sec 37 4.1.4 IEEE 802.11a OFDM PHY 54 Mbit/sec 39 4.1.5 Effect of limited queue size on delay 41 4.2 One AP and multiple nodes greater than four 44 4.3 Two APs and four nodes 45 4.4 One AP and four nodes with different data frame length ....47 4.5 One AP and four nodes with no-RTS/CTS data frame transferring 53 4.6 Different algorithm to select the next frame for transmission 55 4.7 Channel fading effect on the data transmission 58 Chapter 5 Conclusions 61 5.1 Conclusions drawn 62 5.2 Suggestions for future research 64 Glossary 65 Bibliography 67 List of Tables Table 2.1 EEEE 802.11 MAC layer state transition description 14 Table 4.1 Simulation parameters 30 Table 4.2 Simulation parameter values for different PHYs 31 Table 4.3 The distribution of data frame length 47 Table 4.4 The duration of data frame 48 Table 4.5 Fading parameters used in the simulations 58 V List of Figures Figure 2.1 A simple ad-hoc network [1] 6 Figure 2.2 A simple infrastructure network [1] 7 Figure 2.3 Transmission of MPDU without RTS/CTS [1] 11 Figure 2.4 Transmission of MPDU with RTS/CTS [1] 12 Figure 2.5 IEEE 802.11 MAC layer state machine design 13 Figure 2.6 Throughput for 9 stations in an ad-hoc network 19 Figure 2.7 Throughput for 9 stations in an ad-hoc network showing the confidence interval 19 Figure 3.1 Gilbert Channel Model 22 Figure 3.2 WLAN of one AP and four nodes 23 Figure 3.3 The retry algorithm used in almost all simulations 26 Figure 3.4 The "production" algorithm 27 Figure 3.5 The channel state dependent packet - round robin algorithm 28 Figure 4.1 Throughput vs. offered load for FHSS PHY 1 MbitVsec 34 Figure 4.2 Average delay vs. offered load for FHSS PHY 1 Mbit/sec 35 Figure 4.3 Throughput vs. offered load for DSSS PHY 2 Mbit/sec 36 Figure 4.4 Average delay vs. offered load for DSSS PHY 2 Mbit/sec 37 Figure 4.5 Throughput vs. offered load for DSSS PHY 11 Mbit/sec 38 Figure 4.6 Average delay vs. offered load for DSSS PHY 11 Mbit7sec 38 Figure 4.7 Throughput vs. offered load for OFDM PHY 54 Mbit/sec 40 Figure 4.8 Average delay vs. offered load for OFDM PHY 54 Mbit/sec 40 Figure 4.9 Average delay vs. offered load for DSSS PHY 11 Mbit/sec (queue size = 40) 41 Figure 4.10 Average delay vs. offered load for DSSS PHY 11 Mbit/sec (queue size = 200) ...43 vi Figure 4.11 Average delay vs. offered load for DSSS PHY 11 Mbit/sec (queue size = 3000) .43 Figure 4.12 Throughput vs. number of nodes 44 Figure 4.13 Average delay vs. number of nodes 45 Figure 4.14 Throughput vs. offered load for two APs and four nodes 46 Figure 4.15 Average delay vs. offered load for two APs and four nodes 46 Figure 4.16 Throughput vs. offered load for 64 byte data frame 50 Figure 4.17 Average delay vs. offered load for 64 byte data frame 50 Figure 4.18 Throughput vs. offered load for 1520 byte data frame 51 Figure 4.19 Average delay vs. offered load for 1520 byte data frame 51 Figure 4.20 Throughput vs. offered load for variable data frame length 52 Figure 4.21 Average delay vs. offered load for variable data frame length 52 Figure 4.22 Throughput vs. offered load for no-RTS/CTS case 53 Figure 4.23 Average delay vs. offered load for no-RTS/CTS case 54 Figure 4.24 Throughput vs. offered load for different retry algorithms 56 Figure 4.25 Average delay vs. offered load for different retry algorithms 56 Figure 4.26 Variance of average delay vs. offered load 57 Figure 4.27 Throughput vs. fraction of time faded 59 Figure 4.28 Average delay vs. fraction of time faded 59 v i i Acknowledgment I would like to express my heartfelt thanks to my graduate advisor, Dr. Ed Casas, for many useful suggestions, constant encouragement, invaluable guidance, detailed proofreading during my research and thesis writing. Without his supervision, it would be totally impossible to finish my research and thesis. I would also like to thank the many people who gave me encouragement during my studies. viii Chapter 1 Introduction Computer communications are used in many aspects of our daily life and work. It would be difficult to imagine what our life would be like if there were no computer communications. There are two kinds of computer communications, one is established by a wire or a cable, another is accomplished through a wireless link or channel. A Local Area Network (LAN) is connected by a wire or a cable and allows computers to communicate in a relatively small area, such as within an office, office building or campus environment. LANs connect the computers and peripherals with wired networks so that the users can access host computers, databases, applications, files, printers, various peripherals, and Wide Area Network (WAN) connections. A Wireless Local Area Network (WLAN) is the LAN without a wire or a cable connection. Antennas are used to transmit and receive the signals. Since there are no wires, the WLAN can reduce costs. The WLAN is most useful in areas where the wires and cables are difficult to deploy or will be abandoned soon due to various reasons. The IEEE 802.11 committee has established a standard for WLAN [1] [2]. It is called the IEEE 802.11 Medium Access Control (MAC) protocol, by which the wireless channel can be effectively shared among a set of portable computers or wireless terminals. Later, the IEEE 802.11a [3] and IEEE 802.11b [4] were published for higher data rates. In this thesis, we will use the IEEE 802.11 MAC protocol. 1.1 Problem for the traditional frame scheduling There are currently two kinds of WLAN which will be described in Chapter 2. We mainly consider the WLAN which are used to gain access to a wired backbone LAN. The Access Point 1 Chapter I Introduction 2 (AP) works as the connection between the wired LAN and the WLAN. In this thesis, we will study the downlink data frame transmission because most data is carried in this direction. The downlink transmission refers to the frame sent from the Access Point (AP) to other wireless terminals. Due to multi-path transmission and possible obstacles between the AP and wireless terminals, wireless links are fading channels, the characteristics of which are bursty and time varying. The transmission of data frames can fail because of fading-induced errors in the received frame. The traditional frame dispatch scheduling is based on the First In First Out (FIFO) method. When the AP fails to send a frame to some wireless terminal, the FIFO method keeps the AP continuously sending the same frame to the same destination. Actually, it is very likely that the AP could successfully transmit another frame to another destination. The performance of overall system is therefore degraded because of the FIFO, method. 1.2 Review of previous work Channel state dependent scheduling was proposed in an earlier work by Bhagwat et. al [5] [6]. Their work is based on the observation that the FIFO may cause Head Of Line (HOL) blocking when the wireless link to the destination of the frame at the head of outgoing queue is in a bad state. The HOL means the first frame in the queue which will be transmitted. Their purpose was to improve TCP performance. P. Bhagwat et al studied the channel state dependent packet (CSDP) scheduling methods. They used as many queues as the number of wireless terminals in the AP. The packets in each Chapter 1 Introduction 3 queue were transmitted on the FIFO principle. The HOL packet of each queue is transmitted depending on the current estimated channel state between the AP and its terminal. They concluded that this kind of scheduling could increase the overall throughput and reduce the unfairness problem. However, their researches were restricted on the followings: • One AP and ten wireless terminals (without collision) • FTP Traffic mode • TCP performance analysis S. Desilva et al [7] did an experimental evaluation of channel state dependent scheduling using Lucent Technology's wireless LAN, Pentium based laptops and PCs. They concluded that experimental performance evaluation with UDP streams and FTP sessions demonstrate signifi-cant performance benefits of CSDP. M. Inoue et al [8] [9] [10] [11] employed the slotted ALOHA wireless message transport protocol to carry out the computer simulations for their channel state dependent resource schedul-ing. Results showed that the CSD mechanism has a good effect on the scheduling algorithms. Furthermore, they demonstrated that CSD - Equal Sharing algorithm could obtain significant performance improvement. C Fragouli et al [12] pointed out that having a high channel utilization (throughput) and ensuring distribution of bandwidth to different connections according to their allocations (fairness) are major requirements in transporting real time multimedia traffic over wireless links. They described that Class Based Queuing (CBQ) and CSDP scheduling can be combined to enable controlled wireless link sharing. However, modification to the simple combination of CBQ Chapter I Introduction and CSDP needed to be made to work well with wireless links. 4 Yu-Kwong Kwok et al [13] [14] did the research on the uplink access control problem in a cellular phone network where every user could use the mobile phone to transmit voice or file data. They proposed a novel TDMA-based uplink access protocol, which used a channel state dependent allocation mechanism. Their simulation results showed that the proposed protocol considerably outperforms five other state-of-the-art protocols in terms of packet loss, delay, and throughput. 1.3 Objectives The objective of this thesis is to do more detailed research on this topic. The research will be focused at the MAC layer. Various conditions are taken into account to analyze the perfor-mance. The IEEE 802.11 MAC layer protocol is used to simulate the traditional FIFO scheduling and the CSDP scheduling. The traditional FIFO scheduling and the CSDP scheduling are named as non-destination multiplexing and destination multiplexing respectively in this thesis. The IEEE 802.11a and IEEE 802.11b physical layer (PHY) are also considered for higher data rate simula-tions. The results from the two schedulings will be compared and analyzed. The effect of the following will be studied in our simulations: • different data dates • more than four nodes • collisions caused by multiple APs • different data frame lengths • use of RTS/CTS control frames Chapter 1 Introduction 5 • different algorithms to select the next transmitted data frame • effect of channel fading We perform computer simulations to study the performance (throughput and average delay in this thesis) for the overall system. The following chapters are outlined here. Chapter 2 briefly talks about the IEEE 802.11 WLAN architecture and MAC layer protocol. The IEEE 802.11 MAC layer state machine design will be given in this chapter. Then, the simulation tool, OMNeT++, will be introduced and our simulation software will be validated. Chapter 3 presents the characteristics of wireless channels and our solution to enhance the overall system performance. Also in this chapter, two algorithms for next frame selection will be discussed. The results and discussions are shown in Chapter 4. In the final chapter, Chapter 5, the conclusions are presented. Chapter! IEEE802.il WLAN Architecture and MAC Layer 6 Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer In this chapter, we will briefly describe the IEEE 802.11 WLAN architecture and its MAC layer. Then, the IEEE 802.11 MAC layer state machine design will be presented. Finally, we will introduce the simulation tool, OMNeT++, and validate our simulation software developed from it. The IEEE 802.11 architecture includes one or more Basic Service Sets (BSS). The geographical area covered by the BSS is called the Basic Service Area (BSA). In the BSS, there are a few stations which are controlled by the coordination functions. All stations in the BSS are conceptually considered to be able to directly communicate with all other stations in the same BSS. There are two main architectures of WLANs, ad-hoc and infrastructure networks. 2.1 Ad-hoc network Figure 2.1 shows an ad-hoc network. Figure 2.1 A simple ad-hoc network [1] An ad-hoc network is an infrastructureless network with no fixed APs, or wireless base stations. A single BSS can be used to form an ad-hoc network. In such a network, all wireless Chapter 2 IEEE802.il WLAN Architecture and MAC Layer 7 stations can move and can be connected dynamically and arbitrarily. Any station can establish a direct communication with any other station in the BSS. A minimum of two stations can make an ad-hoc network. In the IEEE 802.11 standard, Independent Basic Service Set (IBSS) is the formal name of the ad-hoc network. 2.2 Infrastructure Network Figure 2.2 illustrates an infrastructure network. Figure 2.2 A simple infrastructure network [1] Infrastructure networks are established to give wireless stations specific services and range extension. In the IEEE 802.11, the fixed APs together with some wireless stations are used to form the infrastructure networks. The AP is also a wireless station and offers range extension by connecting multiple BSSs. An Extended Service Set (ESS) is composed of multiple BSSs and Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 8 a common Distribution System (DS). An ESS can also provide gateway access for wireless stations into a wired network such as the Internet, which is done via a device known as a portal. The portal is a logical entity on the DS where the IEEE 802.11 network integrates with a wired IEEE 802 LAN. 2.3 M A C Layer The IEEE 802.11 WLAN MAC layer is responsible for the channel allocation, protocol data unit (PDU) transmitting, frame formatting, error checking, and fragmentation and reassem-bly. In IEEE 802.11 there are three different types of frames: management, control, and data. The management frames are used for station association and disassociation with the AP, timing and synchronization, and authentication and deauthentication. Control frames are used for obtain-ing the status of the wireless media or channel and providing the positive acknowledgment during the contention period. RTS (Request-To-Send), CTS (Clear-To-Send), and ACK (Acknowledg-ment) are control frames. Data frames are used to transmit data. Inter-Frame Space (IFS) is time interval between frames and is used for priority access to the wireless media. Four IFS intervals are specified in the standard: Short-IFS (SIFS), Point Coordination Function-IFS (PIFS), Distributed Coordination Function-IFS (DIFS), and Extended-IFS (EIFS). They are listed in order of duration and priority level. The SIFS interval is the smallest and provides the highest priority access. In this thesis, only SIFS and DIFS are used for analysis and simulation. So, we don't study the PIFS and EIFS. The SIFS shown in Figure 2.3 and Figure 2.4 is used between RTS and CTS, between CTS and Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 9 data frame, between data frame and ACK. Simply speaking, the SIFS is used after stations have gained the channel and want to keep it for the completion of the frame transmission. The DIFS interval is for the stations operating under the DCF. A station is allowed to access to the channel if the channel is free for at least DIPS period and it is not in the backoff status. The MAC layer coordinates the use of a shared medium. The MAC layer protocol specified in IEEE 802.11 is the distributed coordination function (DCF) known as carrier sense multiple access with collision avoidance (CSMA/CA). Another access capability is Point Coordi-nation Function (PCF), which is connection-oriented, and enables the polled stations to transmit data without contention. Since the PCF has nothing to do with destination multiplexing which we will discuss in this thesis, we won't touch on it. CSMA/CA, which is implemented in all wireless stations and APs, is designed to reduce the collision probability when multiple stations access a medium. In the IEEE 802.11, carrier sensing (CS) can be achieved both through physical carrier sensing at the air interface and virtual carrier sensing at the MAC layer. Physical carrier sensing decides if there are other wireless stations transmitting frames in the BSS by measuring signal strength to detect if the channel is active. Virtual carrier sensing mechanism is realized through the use of duration information embedded in the control and data frames. In the next paragraph, we will give a detailed description of virtual carrier sensing because it will be used in our simulation for destination multiplexing Virtual carrier sensing is used by a source station to inform all other stations in the BSS of how long the channel will be utilized to successfully transmit a MAC protocol data unit (MPDU). Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 10 An MPDU is a complete data unit that is passed from the MAC layer to the physical layer. In the MPDU, there are header, payload, and trailer information. The source stations set the time duration in the MAC data frames, or RTS and CTS control frames. The time duration, which is called Network Allocation Vector (NAV) in the IEEE 802.11, indicates the amount of time from the end of the present frame to the time that the source station successfully finishes the transmis-sion of the frame. Other stations except destination station adjust their NAV after they receive the transmitted MPDU containing the time duration field. Those stations stop access to the channel waiting for the end of NAV duration and then begin sampling the channel for idle status. The channel is thought of as being busy if either the physical or virtual carrier sensing indicates the channel is being used. When using the CSMS/CA mechanism, if a station has a frame to be transmitted, it may transmit if the medium is free for at least DCF inter frame space (DIFS) time. If the medium is busy, it follows the backoff procedure to set a random backoff timer. The timer will decrease only when the medium is clear for a slot time period and will be frozen during the busy period. When the backoff timer reaches zero, the station transmits the frame. Since the probability that two stations choose the same backoff timer is small, frame collisions are minimized. Figure 2.3 illustrates the transmission of a data frame where there are no RTS and CTS involved. If a collision occurs, the source station won't know that the collision happens until it completes the transmission of the data frame. So, the channel bandwidth is wasted. To minimize the wasted bandwidth, the RTS and CTS are used. The length of RTS/CTS is usually shorter than the data frame. Normally the RTS and CTS are used when the data frame exceeds the RTS_Threshold (manageable parameter). Using RTS/CTS, the data transmission procedure is as Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 11 Source DIFS D A T A Destination SIFS ^—fe. ^ — A C K Others DIFS ^ fe Contention Window ~ w /// Defer Access Backoff After Defer Figure 2.3 Transmission of MPDU without RTS/CTS [1] follows: the source station sends RTS after DIFS or backoff procedure. After short inter frame space (SIFS), the destination station sends CTS back if it received the RTS. Both RTS and CTS contain the NAV explained before, which indicates the time duration that is reserved for transmit-ting the actual data frame. This information is transmitted to all other stations, which will stop transmission during this period to avoid collision and solve the hidden station problem [1] [2]. Then, after SEFS, the source station sends the data frame if it received the CTS. When the frame is received successfully, as determined by the cyclic redundancy check (CRC), the destination station transmits an acknowledgment (ACK) frame to the source station after SIFS. The whole transmission mechanism is shown in Figure 2.4. Retransmissions happen when the source station doesn't receive CTS within CTS_Timeout period or ACK within ACK_Timeout period. The short retry limit or long retry limit is the maximum number of retransmissions of a data frame due to a collision or interference. For the data frame smaller than RTS_Threshold, the number of retransmissions is set to Short_Retry_Limit. For the data frame larger than RTS_Threshold, the number of retransmissions is limited to Long_Retry_Limit. Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 12 Source DIFS < • RTS SIFS D A T A Destination SIFS CTS SIFS A C K Others DIFS < • Contention Window N A V (RTS) ///// N A V (CTS) Defer Access Backoff After Defer w Figure 2.4 Transmission of MPDU with RTS/CTS [1] 2.4 Backoff Timer When a station receives a frame to be transmitted, it listens to the medium to ensure that there is no other station transmitting. If the medium is free, then it starts to exchange RTS/CTS and transmit the data frame. Otherwise, the backoff procedure will be followed. In the IEEE 802.11 protocol, the random backoff timer is set as follows: Backoff Timer = Random() * Slot time (2.1) RandomO is the pseudo random integer drawn from a uniform distribution between 0 and CW. CW, which stands for Contention Window, is an integer between C W ^ and C W m a x . For the current data frame's first transmission CW is set to be CW^j,. After each collision (indicated by not receiving CTS or ACK), CW for the current data frame is doubled until it reaches C W m a x according to CA mechanism. Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 13 2.5 IEEE 802.11 simulation model and state machine design In this section, we will talk about how to design the IEEE 802.11 simulation model for the computer to do simulations. Figure 2.5 IEEE 802.11 M A C layer state machine design Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 14 Our main purpose is to study if the destination multiplexing mechanism can improve the overall performance of the whole system. So, when we design the simulation model, we omit some irrelevant factors and don't make the simulation model too complicated. The PIFS and EIFS are not taken into account in the simulation model since they have little to do with the perfor-mance of destination multiplexing. The data frame fragmentation is also not considered due to its complexity. The first step of designing the model is the state machine design. After reading the IEEE 802.11 MAC layer standard, we have designed fifteen states to simulate it. The state machine and the transitions of the states are shown in Figure 2.5 and Table 2.1. Table 2.1 IEEE 802.11 MAC layer state transition description State Next State Transition Conditions New frame (this is a transition state, no frame queued, new frame arrival) Defer The channel idle time is less than the defer timer DIES when a new frame arrives. RTS Tx The channel idle time is greater than or equal to the defer timer DIES when a new frame arrives and the frame length is greater than RTS_threshold. DATA Tx The channel idle time is greater than or equal to the defer timer DIFS when a new frame arrives and the frame length is less than or equal to RTS_threshold. Frame queued (waiting for DIFS) Defer All queued frames must first go to the Defer state to wait for the DIES period. Defer (waiting for Defer timer to expire) Backoff The Defer timer expires and the channel is still free, then set the backoff timer and go to the Back-off state. Busy When in the Defer state, if a transmission has been sensed, go to the Busy state to receive the frame. Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer Table 2.1 IEEE 802.11 MAC layer state transition description State Next State Transition Conditions Backoff (waiting for Backoff timer to expire) Busy When in the Backoff state, if a transmission has been detected, go to the Busy state to receive the frame. RTS Tx The Backoff timer expires and the channel is still free, go to the RTS Tx state to send the RTS frame if the data frame is greater than RTS_threshold. DATA Tx The Backoff timer expires and the channel is still free, go to the DATA Tx state to send the data frame if the data frame is less than or equal to RTS_threshold. RTS Tx (transmitting RTS con-trol frame) Wait for CTS After finishing transmission of RTS frame, set the timer to CTS_Timeout and go to the Wait for CTS state. Wait for CTS (waiting for the CTS control frame) SIFS after CTS If the CTS frame is received before the timer for CTS_timeout expires, go to this state and delay SIFS time. Idle If the following conditions have been satisfied. • The timer for CTS_timeout expires or other frames which are not CTS or CTS with the des-tination not to this STA are received. • The data frame waiting for transmission has reached its lifetime or its retry count is greater than the retry limit, which means that the data frame will be discarded. • There is no data frames in the queue. Frame queued If the following conditions have been satisfied. • The timer for CTS_timeout expires or other frames which are not CTS or CTS with the des-tination not to this STA are received. • There will be data frame in the queue even if the data frame waiting for transmission may be dis-carded. SIFS after CTS (waiting for SIFS to expire after CTS con-trol frame is received) DATA Tx After the SIFS delay expires, go to DATA Tx state to transmit the data frame. DATA Tx (transmitting data frame) Wait for ACK After finishing transmission of data frame, set the timer to ACK_timeout and go to the Wait for ACK state. Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer Table 2.1 IEEE 802.11 MAC layer state transition description State Next State Transition Conditions If two following conditions have been satisfied. • The timer for ACK_timeout expires or other frames which are not ACK or ACK with the destination not to itself are received. Idle • The data frame waiting for transmission has reached its lifetime or its retry count is greater than the retry limit, which means that the data frame will be discarded. Or if the following condition has been satisfied. Wait for ACK (waiting for ACK frame) • The ACK frame is received before the timer for ACK_timeout expires, which means that the data frame is sent successfully. And there is no data frame in the queue. The transition occurs • Although the ACK frame is not received within the ACK_timeout period, the data frame won't be discarded. or • The data frame will be discarded because of the reasons described above, but there are data frames in the queue. or • The data frame is transmitted successfully and there are data frames in the queue. Frame queued Idle (the channel is free, no frame queued) Busy The transmission via the channel is detected. New frame A new frame arrives. Idle The channel is free and there is no data frame in the queue. Busy (the channel is busy) Busy The frame whose destination is not this STA. So set the timer to the NAV field of the frame and go to the Busy state again. SIFS after RTS The RTS frame is received. SIFS after DATA The data frame is received. SIFS after RTS (waiting for SIFS to expire after RTS con-trol frame is received) CTS Tx After delaying SIFS period, go to the CTS Tx state. Chapter 2 IEEE 802.11 WLAN A rchitecture and MA C Layer 17 Table 2.1 IEEE 802.11 MAC layer state transition description State Next State Transition Conditions SIFS after DATA (waiting for SIFS to expire after the data frame is received) ACK Tx After delaying SIFS period, go to the ACK Tx state. CTS Tx (transmitting CTS . control frame) Idle Transmitting the CTS frame. If the channel is free and there is no data frame in the queue, then go to the Idle state. Frame queued Transmitting the CTS frame. If the channel is free and there is/are data frame/frames in the queue, go to the Frame queued state Busy Transmitting the CTS frame. If the channel busy is detected, go to the Busy state. ACK Tx (transmitting ACK frame) Idle Transmitting the ACK frame. If the channel is free and there is no data frame in the queue, then go to the Idle state. Frame queued Transmitting the ACK frame. If the channel is free and there is/are data frame/frames in the queue, go to the Frame queued state. Busy Transmitting the ACK frame. If the channel busy is detected, go to the Busy state. 2.6 OMNeT++Introduction OMNeT++ [15] is developed by Dr. Andras Varga in Department of Telecommunication at the Technical University of Budapest and is an object-oriented simulator tool for discrete event simulation. OMNeT++ is the acronym of Object Modular Network Testbed in C++. It is very useful for the communication protocols modeling, networks and traffic modeling. Our reasons for selection of OMNeT++ as the simulation tool are described as follows. Chapter 2 IEEE 802.11 WLAN Architecture and MAC Layer 18 • OPNET is a commercial simulation tool. It can only be used in our lab. OMNeT++ is a non-commercial simulation tool. We can download it for free and use it freely in the office or home environment. • The previous graduate students have already chosen Ptolemy and OPNET as their sim-ulation tool. We are thinking that we need to use new simulation tool for research work. • The NED language in OMNeT++ is easy to learn, read, and write and can be edited by any text editor. It can be used to describe arbitrarily complex network topology in a much easier way. • OMNeT++ is an open-source simulation tool and well documented. It will be conve-nient for us to find the bugs in the tool if we encounter some bugs related to the tool when we do simulations. 2.7 Validation of IEEE 802.11 WLAN simulation model We developed the IEEE 802.11 WLAN model using the OMNeT+-i- tool based on the MAC layer protocol state machine designed in this chapter. To verify this simulation model, we performed simulations for 9 wireless stations in an ad-hoc network and no fading. We plot the throughput vs. the offered load curve in Figure 2.6. Compared to the result curve plotted in [16], we can see that the curve shape and sample values are very close. We can deduce that our simula-tion model is good for IEEE 802.11 WLAN simulation. For each sample value in the curve, we calculate its standard deviation and find out that a 95% confidence interval within + 5% of the average values showing in Figure 2.7. So, in the next Chapter 2 IEEE802.il WLAN Architecture and MAC Layer chapter, we assume that the confidence interval is the same and don't specify it again. 250 200 - 150 Xi 3 OX) S 2 H 100 50 1 1 1 1 1 -— Simulation results "x" William Cheung's results 0 100 200 300 400 Offered load (kbyte/sec) Figure 2.6 Throughput for 9 stations in an ad-hoc network 500 600 250 100 200 300 400 Offered load (kbyte/sec) 500 600 Figure 2.7 Throughput for 9 stations in an ad-hoc network showing the confidence interval Chapter 3 Wireless Channel and Destination Multiplexing 20 Chapter 3 Wireless Channel and Destination Multiplexing In this chapter, the real wireless channel, the fading channel, is introduced. Due to the fading phenomena, wireless stations may be unable to successfully receive a data frame from the source station. Fading degrades the performance of the whole system and causes low throughput and high average delay. Destination multiplexing or channel dependent packet scheduling was recently proposed to improve the system performance when the fading occurs. We use the IEEE 802.11 standard as the MAC layer protocol. We compare two data frame dispatch schemes, First In First Out (FIFO) and destination multiplexing to see which is better. 3.1 Multipath Fading and Shadowing Channel In the real world, the wireless station sometimes is completely out of sight of the AP (i.e. there is no signal path traveling to the receiver via line of sight). In this case, the received signals are made up of a group of reflections from objects, and none of the reflected paths is any more dominant than the other ones. The different reflected signal paths arrive at slightly different times, with different amplitudes, and with different phases. Because there are many different signal paths, constructive and destructive interference can result. Consequently, the "fades" occur. The signal power in the direct path decreases relatively slowly with the increasing distance of the wireless station. However, the obstacles that partially block the signal path can cause drops in the received power. This decrease in power occurs over many wavelengths of the carrier and is thus called slow fading or shadowing. Shadowing is usually modeled by a log-normal distribution with mean power and standard deviation. The reason for the log-normal distributed shadowing is that the received signal is the result of the transmitted signal passing through or reflecting off many different objects. Each object attenuates the signal to some extent, Chapter 3 Wireless Channel and Destination Multiplexing 21 and the final received signal power is the sum of transmission factors of all these objects. As a consequence, the logarithm of the received signal equates to the sum of a large number of transmission factors. As the number of factors becomes large, the central limit theorem dictates that the distribution of the sum approaches a Gaussian, even if the individual terms are not Gaussian. Since there exists multi-path fading and shadowing between the wireless stations and the A P , wireless channels are bursty and time varying. Based on the paper by Gilbert [ 1 7 ] , we can model the burst error wireless channel as a two-state Markov process, or Gilbert model, where one state is described as the good state "G" and another is described as the bad state "B". Figure 3.1 shows the Gilbert channel model. The good state may transit to the bad state and vice versa. If we assume that P G B is the probability from good state to bad state and P B G 1 S the probability from bad state to good state, then, in the steady state, we will have the following equations for PQ , the probability that the channel is good, and P B , the probability that the channel is bad. If we assume that P G G is the probability from good state to good state and P B B is the probability from bad state to bad state. We will have 3.2 Gilbert Channel Model P G ~ P B G / ( P G B + P B G ) P B - P G B / ( P G B + P B G ) (3 .1 ) P G G - 1 " P G B P B B - 1 " P B G (3.2) Chapter 3 Wireless Channel and Destination Multiplexing 22 We assume that when the channel is in the good state the data frame can be received successfully and when the channel is in the bad state the receiver station can't receive the data frame. P E G P G B Figure 3.1 Gilbert Channel Model 3.3 Destination Multiplexing We take a simple WLAN into consideration. Figure 3.2 illustrates a WLAN of one AP and four wireless stations where all stations operates on the IEEE 802.11 WLAN MAC layer protocol. The wireless link is usually shared when the wireless stations communicate with the AP. Since the wireless channel provides only a fraction of the available bandwidth compared with wired line, data frames arriving from the wired line network form a queue at the AP. As shown in Figure 3.2, We can imagine that four queues are formed according to the destination of frames instead of one queue. The general data frame dispatch scheme would be the FIFO order in which these data frames are transmitted over the wireless channel to the wireless stations in the order they were received. However, the FIFO data frame scheduling in the AP degrades throughput and causes unfair allocation of wireless bandwidth. This is because the AP must repeat the transmission of Chapter 3 Wireless Channel and Destination Multiplexing 23 the Head Of Line (HOL) data frame due to the channel burst errors and the data frames destined for the other wireless stations are blocked. These data frames may have been successfully transmitted during the repeated transmission period because of the statistical independence of wireless links to other wireless stations. This situation necessitates consideration of channel state dependent scheduling or destination multiplexing. Simply speaking, destination multiplexing selects the "best" destination when the AP has data frames queued for more than one destination. According to this scheduling, in the AP there are many queues, each of which corresponds to a wireless station. In each queue, the data frames are transmitted based on the FIFO principle. The AP continues sending from a queue only while RTS frame and data frame transmissions are acknowledged. This type of selection of the next data Figure 3.2 WLAN of one AP and four nodes frame will be used in almost all the destination multiplexing simulations in this thesis. 3.4 Retry Algorithm for Destination Multiplexing The retry algorithm is the one used to select the next data frame to be transmitted after the current data frame has been sent out successfully or unsuccessfully. Chapter 3 Wireless Channel and Destination Multiplexing 24 In previous section, we actually talk about one retry algorithm for destination multiplexing (called normal destination multiplexing algorithm in this thesis), which will be implemented in almost all the following simulations. The flowchart of this retry algorithm is presented below in Figure 3.3 (dest is the abbreviation of destination). Suppose that we have one AP and n nodes and all data frames are sent by the AP and are received by the nodes. In the AP, there are n queues for the data frames whose destinations correspond to the n nodes. The timestamp of the data frames from the head to tail of the queues is in the oldest to the latest order. In one sentence, this retry algorithm is to select the same destination when the last transmission is successful or select the oldest data frame whose destination is different from that of last data frame if possible when the last transmission is unsuccessful. The engineers from a wireless company use one kind of retry algorithm which we called the "production" algorithm [19] to implement their WLAN cards. From personal discussion, the "production" algorithm is being used in many WLAN cards in the market. The "production" algorithm only keeps one queue in the AP. All the data frames are inserted into the queue. The timestamp of the data frame from the head to tail of the queue is in the oldest to the latest order. Assume that in the BSS there are one AP and n nodes. The flowchart of the "production" algorithm is shown in Figure 3.4. In one sentence, the "production" algorithm is to always select the oldest data frame when the last transmission is successful or select the oldest data frame whose destination is different from that of last data frame if possible when the last transmission is unsuccessful. The main difference between two retry algorithms is whether we choose a data frame with the same destination or the oldest data frame if the last frame is successfully transmitted. Chapter 3 Wireless Channel and Destination Multiplexing 25 Another retry algorithm we will try in this thesis is called the channel state dependent packet - round robin (CSDP-RR) which was described in [5]. The principle for this algorithm is shown in the flowchart of Figure 3.5. In the AP, there are not only n queues but also n link statuses corresponding to the n nodes. The queues with good status are in one circular list and the queues with bad status are in another circular list. The A P looks at the queues in the good-status circular list in a round robin mode to select the next frame for transmission. If the transmission fails, the link status for this node wil l be marked as bad and the queue wil l be put into the bad-status circular list. A timer with this link wil l be started and its duration is set to the average fading duration of the system. If the timer expires, the corresponding link status will be marked as good and the queue will return back to the good-status circular list. If there is no queue in the good-status circular list, the AP chooses the queues in the bad-status circular list also in a round robin mode. If the transmission still fails, the link status and the queue are kept the same as before and no timer will be started. If the transmission is successful, the link status will be set to be good and the queue will go to the good-status circular list. As for the implementation complexity, the "production" algorithm is simplest and has only one queue. The CSDP-RR algorithm is most complicated in all three algorithms. It requires not only the link status but also several timers in the AP. Chapter 3 Wireless Channel and Destination Multiplexing 26 Set n = number of nodes in BSS. t Setup n queues for the data frames corresponding to the n nodes. t_ Select the first data frame arrived at the AP. Assume it is in queue i. Set dest = i Figure 3.3 The retry algorithm used in almost all simulations Chapter 3 Wireless Channel and Destination Multiplexing 27 Setup one queue for data frames t Pick up one data frame from the head of queue. Assume its destination is node i. Set dest = i Send this data frame Yes Select next data frame from the head of the queue. Assume its destination is node k Search the queue from head to tail for a data frame whose destination is different from dest Set dest = k Yes Select the same data frame Select this data frame. Assume its destination is node j t Set dest = j Figure 3.4 The "production" algorithm Chapter 3 Wireless Channel and Destination Multiplexing 28 Set n = number of nodes in BSS. a Setup n queues for the data frames corresponding to the n nodes. J Create an array of size n to represent the status of n wireless links. Each link status is initialized to be good. Select the first data frame arrived at the AP. Assume it is in queue i. Yes Set this link status good. Cancel the timer if there is one associated with this link. Set this link status bad. Start a timer if there is no existing one with this link. The timer duration is the average fading duration. Start from dest to search in cardinal order the first queue whose status is good for a data frame. Set dest to this queue Yes Start from dest to search in cardinal order the first queue whose status is bad for a data frame. Set dest to this queue Figure 3.5 The channel state dependent packet - round robin algorithm Chapter 4 Simulation Results and Discussions 29 Chapter 4 Simulation Results and Discussions The simulations are performed under both destination and non-destination multiplexing conditions. In the preliminary IEEE 802.11 Standard, only two kinds of Physical Layer, FHSS and DSSS, are introduced and the data rates are not high, only 1 Mbit/sec and 2 Mbit/sec. Later, in IEEE 802.11a and IEEE 802.11b, OFDM and DSSS are used and the data rates can reach as high as 54 Mbit/sec and 11 Mbit/sec. Nowadays, people are more and more interested in high data rate transmission. In the simulations, different data rates will be studied. In the next paragraph, we will introduce seven scenarios which are studied in this chapter. We will first study the scenario of (1) one AP and four nodes with different data rates. Usually the BSS is composed of one AP and many associated wireless stations or nodes. Although the number of associated nodes may exceed four, generally only a few are active. The effect of the different data rates is also of interest. We will also consider the situation of more nodes. The second scenario which will be studied is (2) one AP and multiple nodes greater than four. When the APs are transmitting a frame, collisions may happen. The third scenario is (3) two APs and four nodes to study the collision effect. For the above scenarios, we assume that all the data frames have the same length. In the practical situations, transmitting different data frame lengths is very common. The fourth scenario we will study is (4) one AP and four nodes with different data frame lengths. In all the simulations above, we send RTS/CTS frames before sending out the data frame. The RTS/CTS mechanism is very useful for detecting the fading channel or collisions, but it will add the overhead and cause the throughput to be lower. So, we are interested in the fifth scenario (5) one AP and four nodes with no-RTS/CTS data frame transfer-ring. If we use the destination multiplexing mechanism, we need to decide which data frame to be Chapter 4 Simulation Results and Discussions 30 sent next after the previous data frame was unsuccessfully transmitted. The different selection criteria make us study the sixth scenario (6) Different algorithm to select the next frame for transmission. In the final section, we will study the seventh scenario (7) Channel fading effect on the data transmission. Table 4.1 describes all the simulation parameters we will use when we do the computer simulations. Table 4.1 Simulation parameters Parameter Description SIFS Short inter frame space DIFS Distributed Coordination Function inter frame space Slot Time RTS Request to send frame length CTS Clear to send frame length ACK Acknowledgment frame length DATA Data frame length Bit Rate Data rate CTS_Timeout The maximum waiting time for the CTS frame ACK_Timeout The maximum waiting time for the ACK frame Contention Window Maximum and minimum size of Contention Window (CW^,,, cwm a x) Short Retry Limit Maximum number of retransmission of a frame because of no receipt of the CTS Long Retry Limit Maximum number of retransmission of a frame because of no receipt of the ACK Probability of good state to bad state This value must be defined when Gilbert model is used Probability of bad state to good state This value must be defined when Gilbert model is used Update rate for Gilbert channel Channel state change rate for the Gilbert channel Queue Size There are different queue sizes for destination multiplexing and non-destination multiplexing. These two queue sizes follow certain rule in this thesis Chapter 4 Simulation Results and Discussions 31 Table 4.2 shows the values of all the simulation parameters for IEEE 802.11 FHSS PHY 1 Mbit/sec, IEEE 802.11 DSSS PHY 2 Mbit/sec, IEEE 802.11b DSSS PHY 11 Mbit/sec, IEEE 802.11a OFDM PHY 54 Mbit/sec. Table 4.2 Simulation parameter values for different PHYs Parameter FHSS 1 Mbit/sec DSSS 2 Mbit/sec DSSS 11 Mbit/sec OFDM 54 Mbit/sec SIFS 28 us 10 us 10 us 16 n s DIFS 128 us 50 us 50 us 34 us Slot Time 50 us 20 us 20 us 9 ns RTS 293 n s 272 us 207 us 25 ns CTS 244 jxs 248 us 202 us 24 ns ACK 244 us . 248 us 202 ]xs 24 ns DATA 8378 u. s 4192 us 919 n s 171 ns Bit Rate 1 Mbit/sec 2 Mbit/sec 11 Mbit/sec 54 Mbit/sec CTS_Timeout 300 us 300 us 300 us 300 ns ACK_Timeout 300 us 300 us 300 us 300 ns Contention Window (15, 1023) (31, 1023) (31, 1023) (15, 1023) Short Retry Limit 7 7 7 7 Long Retry Limit 4 4 4 4 Probability of good state to bad state 0.01 0.01 0.01 0.01 Probability of bad state to good state 0.09 0.09 0.09 0.09 Update rate for Gil-bert channel 10 Hz 10 Hz 10 Hz 10 Hz Queue Size 10 10 10 10 In the IEEE 802.11 standard, we can find the values for SIFS, DIFS, Slot Time, Bit Rate, Contention Window, Short Retry Limit, and Long Retry Limit. The values of CTS_Timeout and ACK_Timeout can be deduced from the Standard. The values are 300 ns. The same fading Chapter 4 Simulation Results and Discussions 32 conditions are used except the seventh scenario. So the values for "Probability of good state to bad state", "Probability of bad state to good state", and "Update rate for Gilbert channel" are chosen as in Table 4.2. How to calculate the RTS, CTS/ACK, and DATA duration will be discussed in the following sections. In all simulations, a limited queue size is employed so that we can simulate the behaviors of an actual IEEE 802.11 system. A frame will be discarded if it arrives at the AP and finds the queue is full. There is only one queue for non-destination multiplexing. But, there are several queues for destination multiplexing. The number of queues in the AP for destination multiplexing is equal to the number of its nodes. In this thesis, the queue size for destination multiplexing is specified as the queue size for non-destination multiplexing divided by the node number. The default queue size for destination multiplexing is chosen to be 10 shown in Table 4.2. Hence, the default queue size for non-destination multiplexing is 40. All simulations are run using these two default values unless specified. In the following seven sections, the simulation conditions and parameters will be described first, and these are followed by the simulation results and discussions. We study the throughput and average delay based on the offered load. The throughput is calculated as the number of frames successfully transmitted within a certain time period divided by that time period. Throughput = No. of frames successfully sent / time period (4.1) The average delay is obtained after dividing the sum of the time required for each data frame to be successfully sent to the destination node after it arrives at the AP by the total success-Chapter 4 Simulation Results and Discussions , 33 fully sent data frames. Average delay = Sum of times taken for each data frame / no. of frames successfully sent (4.2) According to the definition, the offered load is how many kbytes are received within one second. If we assume that the data frame length is 8000 bits, which is 1 kbyte (1000 byte). The offered load can be interpreted as the number of data frames received per second. 4.1 One AP and four nodes with different data rates In this section, the one AP and four nodes scenario is studied. It is assumed that the fading channel is simulated as the Gilbert channel model. The Poisson frame arrival traffic model is employed. Below we will show the simulation parameters and simulation results for IEEE 802.11 FHSS PHY 1 Mbit/sec, IEEE 802.11 DSSS PHY 2 Mbit/sec, IEEE 802.11b DSSS PHY 11 Mbit/ sec, and IEEE 802.11a OFDM PHY 54 Mbit/sec. 4.1.1 IEEE 802.11 FHSS PHY 1 Mbit/sec The simulation parameter values are shown in the second column of Table 4.2. Based on the IEEE 802.11 Standard, we can calculate the RTS, CTS/ACK, and DATA duration if we know how many bits there are. For FHSS PHY, we know that PPDU duration is defined as follows. PPDU = aPreambleLength + aPLCPHdrLength + aMPDUDurationFactor * 8 * PSDULength (octets) / data rate (4.3) where aPreambleLength is the PHY's Preamble Length, aPLCPHdrLength is the PHY's PLCP Header Length, aMPDUDurationFactor is the overhead added by the PHY to the MPDU as it is Chapter 4 Simulation Results and Discussions 34 transmitted through the wireless medium expressed as a scaling factor applied to the number of bits in the MPDU. According to the IEEE 802.11 Standard, RTS is a 160 bit frame, CTS/ACK is a 112 bit frame. The data frame can be variable length. We assume that there are always 1 kbyte (8000 bits) data frames to be sent. From the FHSS PHY attributes table in the Standard, the default values for aPreambleLength, aPLCPHdrLength, and aMPDUDurationFactor are 96 us, 32 us, and 1.03125 respectively. So, the RTS, CTS/ACK, and DATA durations can be derived to be 293 u s, 243.5 n s, and 8378 ns. The curves of the throughput and average delay vs. the offered load are shown in Figure 4.1 and Figure 4.2. Single AP and four nodes 120 T 0 1 , , , r , , , 1 0 20 40 60 80 100 120 140 160 Offered load (kbyte/sec) Figure 4.1 Throughput vs. offered load for FHSS PHY 1 Mbit/sec Chapter 4 Simulation Results and Discussions 35 Single AP and four nodes 1 0.9 8. 0.7 £ 0.6 £ 0.5 a) 0.4 2 0.3 d) > 0.2 0.1 •Non-destination Multiplexing • Destination Multiplexing 20 40 60 140 160 Offered load (kbyte/sec) Figure 4.2 Average delay vs. offered load for FHSS PHY 1 Mbit/sec From the diagrams, we can see much improvement from using destination multiplexing. The throughput is increased by more than 20%. The average delay is decreased significantly. 4.1.2 IEEE 802.11 DSSS PHY 2 Mbit/sec In the IEEE 802.11 Standard, DSSS is also recommended as the physical layer. Here, we study the 2 Mbit/sec data rate DSSS. The simulation parameters will be changed to the third column in Table 4.2. In Table 4.2, RTS, CTS/ACK, and DATA are calculated using the PLCP frame format defined in the IEEE 802.11 Standard. PPDU = PLCP Preamble + PLCP Header + MPDU / data rate (4.4) where PLCP Preamble is 144 bits and PLCP Header is 48 bits. From the Standard, we know that the entire PLCP Preamble and Header shall be transmitted using the 1 Mbit/sec data rate. So, Chapter 4 Simulation Results and Discussions 36 PLCP Preamble + PLCP Header will be 192 us. The RTS is defined to be 160 bits frame length, CTS/ACK is defined to be 112 bits frame length, DATA are assumed to have 1000 kbyte (i.e. 8000 bits), thus these frame lengths are 272 u s, 248 n s, 4192 u s. The simulation results are shown in Figure 4.3 and Figure 4.4. Single A P and four nodes 250 0-1 , , , , , 1 0 50 100 150 200 250 300 Offered load (kbyte/sec) Figure 4.3 Throughput vs. offered load for DSSS P H Y 2 Mbit/sec We can also see much improvement is achieved in the throughput and average delay. Roughly saying, the throughput is increased by approximately 30%, the average delay is half of the non-destination multiplexing value. Chapter 4 Simulation Results and Discussions 37 0.5 ~ 0.4 o <D CO >> ra 0) •o a at a 0) > < 0.3 0.2 0.1 0 Single AP and four nodes • Non-destination Multiplexing • Destination Multiplexing 100 150 200 Offered load (kbyte/sec) 300 Figure 4.4 Average delay vs. offered load for DSSS PHY 2 Mbit/sec 4.1.3 IEEE 802.11b DSSS PHY 11 Mbit/sec In IEEE 802.11b, a high-speed physical layer in the 2.4 GHz band is introduced. The data rate has four choices 1 Mbit/sec, 2 Mbit/sec, 5.5 Mbit/sec, 11 Mbit/sec. We choose the highest data rate in this set, 11 Mbit/sec, to perform the simulation. Since the different data rate of 11 Mbit/sec is used, the simulation parameter table is changed to the fourth column in Table 4.2. To calculate RTS, CTS/ACK, and DATA durations, equation (4.4) shown above is used. So, 160 bit RTS is equivalent to 207 us, 112 bit CTS/ACK is equivalent to 202 us, 8000 bit DATA is equivalent to 919 u s. Under the above conditions, we perform the computer simulations. The results are shown Chapter 4 Simulation Results and Discussions in Figure 4.5 and Figure 4.6. 38 Single AP and four nodes 600 0-1 , , , , , , 1 0 100 200 300 400 500 600 700 Offered load (kbyte/sec) Figure 4.5 Throughput vs. offered load for DSSS PHY 11 Mbit/sec 0.25 _ 0.2 in >; o.i5 a o> •o <u 0.1 o> n L. > 0.05 0-I Single AP and four nodes • Non-destination Multiplexing - Destination Multiplexing 100 200 300 400 500 600 700 Offered load (kbyte/sec) Figure 4.6 Average delay vs. offered load for DSSS PHY 11 Mbit/sec From the result diagrams, we can see the throughput is increased by more than 25% and the average delays of non-destination multiplexing are at least double those of destination Chapter 4 Simulation Results and Discussions 39 multiplexing. 4.1.4 I E E E 802.11a O F D M P H Y 54 Mbit/sec The I E E E 802.11a is the standard for the high-speed physical layer in the 5GHz band. The O F D M system is used in the PITY. Right now, people are more and more interested in it because its data rate can reach 54 Mbit/sec. Whether or not destination multiplexing can improve the performance of the system is our concern. So, we do the simulations to see the results. The simulation parameters are changed according to the 54 Mbit/sec data rate. The fifth column in Table 4.2 shows all the parameter values used to simulate destination multiplexing and non-destination multiplexing at 54 Mbit/sec data rate. Calculating RTS , C T S / A C K , and D A T A duration for O F D M P H Y is different from that for F H S S P H Y or DSSS P H Y . A simplified equation may be used to calculate O F D M T X T I M E . T X T I M E = T P R E A M B L E + T S I G N A L + (16 + 8 x L E N G T H + 6 ) / D A T A R A T E + T S Y M (4.5) where T P R E A M B L E : P L C P preamble duration whose value is 16 us. T S I G N A L : Duration of the S I G N A L B P S K - O F D M symbol whose value is 4.0 u s. T S Y M : Symbol interval whose value is 4 US. Then, we can derive the following values. RTS duration =164-4 + (16 + 160 + 6)/54 + 2 =25 u s. C T S / A C K duration = 16 + 4 + (16 + 112 + 6)/54 + 2 = 24 n s. D A T A (1 kbyte ) duration = 16 + 4 + (16 + 8000 + 6)/54 + 2 = 171 u s. Figure 4.7 and Figure 4.8 are the simulation results. Chapter 4 Simulation Results and Discussions 40 Single AP and four nodes 3000 o-l , , , , , , , 1 0 500 1000 1500 2000 2500 3000 3500 4000 Offered load (kbyte/sec) Figure 4.7 Throughput vs. offered load for O F D M P H Y 54 Mbit/sec u 0> >. ro v TJ <u ro > < 0.05 0.04 0.03 0.02 0.01 Single AP and four nodes • Non-destination Multiplexing • Destination Multiplexing 500 1000 1500 2000 2500 3000 3500 4000 Offered load (kbyte/sec) Figure 4.8 Average delay vs. offered load for O F D M P H Y 54 Mbit/sec From the simulation results, we can see that the above conclusions are still correct. The throughput has been improved by more than 25% and the average delay of destination multiplex-ing is less than half of that of non-destination multiplexing. Chapter 4 Simulation Results and Discussions 41 4.1.5 Effect of limited queue size on delay In some figures of this chapter, the average delay curves for non-destination multiplexing show unusual fluctuations which are difficult to explain. For example, Figure 4.6 shows unexpected drops in delay as the offered load increases. Hence, further work is done in order to get a reasonable answer. Changing the simulation duration to be four times longer, we obtain the results shown in Single AP and four nodes — • — Non-destinatbn Multiplexing — • — Destination Multiplexing / — — * — • • • • • • —• m-—9r~^mr 0 100 200 300 400 500 600 700 Offered load (kbyte/sec) Figure 4.9 Average delay vs. offered load for DSSS PHY 11 Mbit/sec (queue size = 40) Figure 4.9. From Figure 4.9, we can see that the curve is smoother. So, the simulation length does have some effects on the curve fluctuations. This is due to very slow convergence of the average delay of non-destination multiplexing, The curve in Figure 4.6 shows some unusual "bumps". However, we still find that when the offered load varies from 100 to 450 kbyte/sec the average delay looks almost flat and slightly decreased. Chapter 4 Simulation Results and Discussions 42 After investigation, we guess that it is probably caused by the limited queue size. For the current implementation, if a frame arrives and finds that the queue is full, it will be dropped. The possible explanation would be that although the offered load increases, lots of frames will be discarded if the AP keeps sending the HOL frame due to the bad link and the queue fills up. Hence the average delay seems not to increase with the increasing of the offered load. To verify this idea, we increased the queue size from the current value 40 to 200. Figure 4.10 shows the curve after the simulation is done. We find that the flat area is shortened. It is only from 150 to 350 kbyte/sec. When we run the simulation, we also notice that the queue size is still not large enough. So, we enlarge it to be 3000 and run the simulation again. With this queue size we obtained the curve shown in Figure 4.11. So, the unexpected behaviors are caused by the combined effects of the limited queue size and the simulation length. We also see that actual average delay in Figure 4.11 is more than that in Figure 4.9. It is because that we consider more frames in the queue and those frames are signifi-cantly delayed due to the blocked HOL frame when the link is bad. Chapter 4 Simulation Results and Discussions 43 >. ra O •o <v at ra k_ cu > < 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Single AP and four nodes •Non-destination Multiplexing • Destination Multiplexing 100 200 300 400 500 Offered load (kbyte/sec) 600 700 Figure 4.10 Average delay vs. offered load for DSSS PHY 11 Mbit/sec (queue size = 200) Single AP and four nodes 10 7 700 Offered load (kbyte/sec) Figure 4.11 Average delay vs. offered load for DSSS PHY 11 Mbit/sec (queue size = 3000) Chapter 4 Simulation Results and Discussions 44 4.2 One AP and multiple nodes greater than four Usually, in an infrastructure BSS, there is one AP and many nodes or stations. Those nodes may be active or inactive. It is likely that the number of active nodes can be more than four. Thus, it is of much interest to look at the behavior of more nodes existing in the BSS. We simulate the scenario of one AP and multiple nodes greater than four. We still use the Gilbert channel as the fading channel and Poisson frame arrival traffic model. We assume that the average offered load for each node at the AP is 20 kbyte/sec. The simulation parameter values are the same as those in the second column of Table 4.2. We plot the simulation results in Figure 4.12 and Figure 4.13. Figure 4.12 shows the relation between throughput and number of nodes. The throughputs Single A P and multiple nodes 120 8.100 I 80 Si r 60 a. sz at 3 O 40 20 -• •-• Non-destination Multiplexing - Destination Multiplexing 6 8 10 Number of nodes 12 14 16 Figure 4.12 Throughput vs. number of nodes go up steadily with the increase of number of nodes from 4 to 6. The ideal throughputs for 4, 5, 6 nodes would be 4 x 20 kbyte/sec, 5 x 20 kbyte/sec, 6 x 20 kbyte/sec. For destination multiplexing, the values are close to the ideal ones but less than those because of unsuccessful transmission. Chapter 4 Simulation Results and Discussions 45 Single AP and multiple nodes 5 o 0) in 4 - • — Non-destination Multiplexing <•— Destination Multiplexing ra cu T3 0) O) cs 3 < cu > 2 0 0 2 4 6 8 10 12 14 16 Number of nodes Figure 4.13 Average delay vs. number of nodes From 6 to 14, the throughputs for destination multiplexing are saturated and become flat, and the throughputs for non-destination multiplexing also trend to be unchanged and go down a little bit because the AP keeps sending the unsuccessful data frame to the same node. Figure 4.13 shows that the average delay of non-destination multiplexing is always worse than that of destination multiplexing due to the FIFO frame dispatch scheme. 4.3 Two APs and four nodes For the normal BSS, we have only one AP and dozens of nodes. We now assume there are two BSSs overlapped. Two APs are simultaneously interacting with four nodes. Here, we would like to investigate the effect of collision on destination multiplexing and non-destination multiplexing. The fading channel is still the Gilbert channel and the frame data arrival still obeys a Poisson distribution. The simulation parameter values are as shown in the second column of Table 4.2. The simulation results are shown in Figure 4.14 and Figure 4.15. Chapter 4 Simulation Results and Discussions 46 120 100 80 o 4) * 60 3 a. 2 20 Two APs and four nodes • Non-destination Multiplexing • Destination Multiplexing 10 20 30 40 50 Offered load per AP (kbyte/sec) 60 70 Figure 4.14 Throughput vs. offered load for two APs and four nodes Two A P s and four nodes ra a> TJ o> O) ro > < 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 • Non-destination Multiplexing • Destinatbn Multiplexing Offered load per AP (kbyte/sec) 70 Figure 4.15 Average delay vs. offered load for two APs and four nodes Since two APs are sending data frames to four nodes, collisions are unavoidable. From Figure 4.14, we can see that the throughput of destination multiplexing for two APs is lower than that for one AP. It is the collision that causes such change. However, the throughput of destination Chapter 4 Simulation Results and Discussions ' 47 multiplexing is still at least 15% more than that of non-destination multiplexing. Figure 4.15 shows the average delay of destination multiplexing is definitely better than that of non-destina-tion multiplexing almost being a half of that for the worst case. 4.4 One AP and four nodes with different data frame length For Internet traffic, it is not practical if we only consider the average data frame length. Normally, there are three peaks in the distribution of data frame length. The distribution in Table 4.3 has been proposed [18] for the data frame lengths. Table 4.3 The distribution of data frame length Data frame length Distribution 64 bytes long 60% 596 bytes long 17% 1520 bytes long 23% Table 4.3 shows: • The size of most frames is small. • The distribution of medium size frames is close to that of large size frames. • Number of small size frames almost triples that of medium size frames or large size frames. Since the IEEE 802.11b is widely used and there is not much difference in the perfor-mance advantage of destination multiplexing for the different data rates used in section 4.1, we think that the IEEE 802.11b 11 Mbit/sec data rate can be used to obtain results that represent all other data rates. The simulation parameter values are all from the fourth column of Table 4.2 except DATA (data frame length). From equation (4.4), we can obtain Table 4.4. Chapter 4 Simulation Results and Discussions 48 Table 4.4 The duration of data frame Data frame length DATA (us) 64 bytes long 239 596 bytes long 625 1520 bytes long 1297 We can also deduce that the average DATA duration would be: 239 x 60% + 625 x 17% + 1297 x 23% = 548 us Based on Table 4.4 and the average DATA duration, we perform the simulations as follows. • Do the simulation using the data frame length which is equal to 64 bytes. • Do the simulation using the data frame length which is equal to 1520 bytes. • Do the simulation using variable data frame length which has Table 4.3 distribution. The purpose of doing the simulation for data frame length 64 bytes and 1520 bytes is that we can compare the results with the result for variable data frame length. The simulation results for 64 bytes data frame length are shown in Figure 4.16 and Figure 4.17. Figure 4.18 and Figure 4.19 show the simulation results for 1520 byte data frame length. The simulation results for variable data frame length can be seen from Figure 4.20 and Figure 4.21. From Figure 4.16 to Figure 4.21, we have the following conclusions. Chapter 4 Simulation Results and Discussions 49 • If we compare Figure 4.18 and Figure 4.20, we can see the throughput of variable data frame length is less than that of 1520 byte data frame length. No matter whether the destination multiplexing is used or not, the throughputs of 1520 byte data frame are all greater (the throughputs for 1520 byte data frame are approximately 1520 * 400 for destination multiplexing and 1520 * 300 for non-destination multiplexing, which are greater than 489 * 700 and 489 * 500 of the throughputs for variable data frame). Since there is much overhead involved with sending each frame, the time it takes to send frames of different lengths is similar although the data frame lengths are differ-ent. So, longer frames have more bytes to be transmitted and higher throughput. • From Figure 4.19 and Figure 4.21, the average delays of 1520 byte data frame with or without destination multiplexing are smaller than those of variable length data frame (according to equation (4.2), the actual average delay needs to be divided by the frame length). The reason is the same as above. Longer frames have sent out more bytes and have got less actual average delay for one byte. • Figure 4.16 and Figure 4.20 show that the throughput of 64 byte data frames is less than that of variable length data frames for destination multiplexing and non-destina-tion multiplexing (64 * 700 throughput is less than 489 * 700 for destination multi-plexing, 64 * 500 throughput is less than 489 * 500 for non-destination multiplexing). • The average delay of 64 byte data frames is greater than that of variable length data frames in both destination multiplexing and non-destination multiplexing cases, which can be seen in Figure 4.17 and Figure 4.21. Chapter 4 Simulation Results and Discussions 50 Figure 4.16 Throughput vs. offered load for 64 byte data frame Single AP and four nodes 0.3 n 0.25 0.2 a o> 0.15 •a 0) S1 o.i 0) < 0.05 0 - • — Non-destination Multiplexing - • — Destination Multiplexing 200 400 600 800 Offered load (64 byte/sec) 1000 1200 Figure 4.17 Average delay vs. offered load for 64 byte data frame Chapter 4 Simulation Results and Discussions Figure 4.18 Throughput vs. offered load for 1520 byte data frame Single AP and four nodes 0.3 0.25 0.2 >. to ai 0.15 0 100 200 300 400 500 600 700 Offered load (1520 byte/sec) — • — Non-destination Multiplexing — • — Destination Multiplexing Figure 4.19 Average delay vs. offered load for 1520 byte data frame Chapter 4 Simulation Results and Discussions 52 u V * o> 00 •*-• 3 Q . 3 O 800 700 600 500 400 300 200 100 0 Single A P and four nodes • Non-destination Multiplexing • Destination Multiplexing 200 400 600 800 Offered load (489 byte/sec) 1000 1200 Figure 4.20 Throughput vs. offered load for variable data frame length Single A P and four nodes 0.3 T  _ 0.25 — * — Non-destination Multiplexing — • — Destination Multiplexing , » • • • • • 0 200 400 600 800 1000 1200 Offered load (489 byte/sec) Figure 4.21 Average delay vs. offered load for variable data frame length Chapter 4 Simulation Results and Discussions 53 4.5 One AP and four nodes with no-RTS/CTS data frame transferring Up to now, we all perform the simulations and discuss the results under the general condition that RTS/CTS is transmitted first before the actual data frame is sent. If RTS/CTS is used, we know that it actually adds a lot of overhead and causes the throughput to be lower. So, it is practical to consider the special condition that there is no RTS/CTS and only the data frame and ACK are transmitted. We also do the simulations in this case. We use the same simulation parameter values as shown in the fourth column of Table 4.2 without considering RTS/CTS in the simulations. Figure 4.22 and Figure 4.23 show the results. Single A P and four nodes 700 •, 600 cu « 500 n 400 • 300 a Ol 200 3 o £ 100 -1-0 -• Non-destination Multiplexing • Destination Multiplexing 0 100 200 300 400 500 600 700 800 900 Offered load (kbyte/sec) Figure 4.22 Throughput vs. offered load for no-RTS/CTS case From Figure 4.22 and Figure 4.23, the following conclusions can be made. The throughputs of destination multiplexing and non-destination multiplexing without RTS/CTS are all approximately 20% greater than those with RTS/CTS in Figure 4.5 because of fewer overheads. Chapter 4 Simulation Results and Discussions 54 Single AP and four nodes 0.2 0.18 -5- 0.16 1 0.14 > 0.12 n o 0.1 TJ 0) 0.08 ui 2 0.06 | 0 0 4 0.02 0 - Non-destination Multiplexing - Destination Multiplexing 100 200 300 400 500 600 700 800 900 Offered load (kbyte/sec) Figure 4.23 Average delay vs. offered load for no-RTS/CTS case • The throughput of destination multiplexing is increased by more than 30% compared to that of non-destination multiplexing. More advantage can be obtained if we com-pare the destination multiplexing scheme with the non-destination multiplexing scheme when RTS/CTS is not used. The reason is that the overheads are reduced in the no-RTS/CTS case. • The average delays of destination multiplexing and non-destination multiplexing with-out RTS/CTS are all about 0.01 sec and 0.02 sec less than those with RTS/CTS in Fig-ure 4.6 respectively when the curves trend to be stable. • The average delay of destination multiplexing is still approximately half of that of non-destination multiplexing. Chapter 4 Simulation Results and Discussions 55 4.6 Different algorithm to select the next frame for transmission From the descriptions in Chapter 3, we can see that the only difference between the normal destination multiplexing algorithm and the "production" algorithm is whether we choose a data frame with the same destination or the oldest data frame if the last frame is sent successfully. The CSDP-RR algorithm is to choose the data frame from the next good-status queue in the circular list. We run the simulations for all three algorithms with the simulation parameter values shown in the fourth column of Table 4.2. Figure 4.24 and Figure 4.25 show the simulation results. The "PA" and "RR" in Figure 4.24 and Figure 4.25 stands for "Production Algorithm" and "Round Robin" respectively. Figure 4.24 and Figure 4.25 show the following: • Three algorithms have almost the same throughput and average delay except at high load. • The throughput of the "production" algorithm is a little bit less than those of the nor-mal destination multiplexing algorithm and the CSDP-RR algorithm when the offered load gets high. • The throughput of the normal destination multiplexing algorithm and the CSDP-RR algorithm are almost the same. • The average delay of the "production" algorithm is worst when the offered load be-comes high. • The average delay of the CSDP-RR algorithm is longer than that of the normal desti-nation multiplexing algorithm at high load. In the real implementation, the "production" algorithm is used because it has only one Chapter 4 Simulation Results and Discussions 56 Figure 4.24 Throughput vs. offered load for different retry algorithms Single A P and four nodes 0.08 _ 0.07 o 0.06 m * 0.05 — • — Destination Multiplexing (PA) ^ - T . — • — Destinatbn Multiplexing —A— Destinatbn Multiplexing (RR) /y — • = » i • • • * 0 100 200 300 400 500 600 700 Offered load (kbyte/sec) Figure 4.25 Average delay vs. offered load for different retry algorithms queue and is simplest to implement. From the results, it appears that the average delay is decreased most if the data frame with the same destination is chosen when the last transmission was successful. That the data frames from the same queue are chosen causes the overall average delay to be lowest. This conclusion made us question the fairness of the algorithms. We thus Chapter 4 Simulation Results and Discussions 57 looked at the variance of the delay for the three algorithms. We computed the variances for three algorithms and plot them in Figure 4.26. CD O c cs J o JZ fl) CD O) CB > < 2 .50&03 2.00E-03 1.50E-03 1.00E-03 5.00E-04 0.00 E+00 Single AP and four nodes — • — Destination Multiplexing (PA) — • — Destination Multiplexing — * — Destination Multiplexing (RR) Jr/-+ A 100 200 300 400 500 Offered load (kbyte/sec) 600 700 Figure 4.26 Variance of average delay vs. offered load Figure 4.26 shows that the delay of the normal destination multiplexing algorithm has the biggest variance among all three retry algorithms when the A P has high load of the data frames, which is what we expect because if the AP continues sending the data frames to the same destina-tion some data frames' delay will be very small and others' delay will be very large. The CSDP-RR algorithm is changing the queues or the data frame destinations constantly so that it shows the medium variance. Chapter 4 Simulation Results and Discussions 58 4.7 Channel fading effect on the data transmission As described in Chapter 3, the Gilbert channel model is always used to simulate the fading channel. Equations (3.1) and (3.2) are normally used to calculate the fading statistics. In all simulations done so far, P G B is set to 0.01 and P B G is set to 0.09, which means that P G = 0.9 and P B = 0.1. P B , the probability that channel is bad, can be thought as the fraction of time the channel is faded. In theory, P B is a function of signal strength at the receiving nodes and depends on the distance from the client node to the AP. In practical use, a BSS system would normally be deployed to keep P B very small. Generally, P B will be 10"1 or less. So, in our simulation of this section, we choose P B to be 0.1, 0.01, 0.001, 0.0001. The Table 4.5 shows the parameters used in the simulations. Table 4.5 Fading parameters used in the simulations P B P G B P B G 0.1 0.01 0.09 0.01 0.001 0.099 0.001 0.0001 0.0999 0.0001 0.00001 0.09999 A l l other simulation parameter values can be found in the fourth column of Table 4.2. the offered load is set to be 600 kbyte/sec. The following Figure 4.27 and Figure 4.28 show the simulation results. From the simulation results, we can draw the following conclusions. Chapter 4 Simulation Results and Discussions 59 600 500 400 o o> •8 * 300 3 a. o o 200 100 Single AP and four nodes • Non-destination Multiplexing • Destinatbn Multiplexing 1 2 3 4 Fraction of time faded (-log,0(P(B))) Figure 4.27 Throughput vs. fraction of time faded Single A P and four nodes 0.12 0.1 0.08 >. IS 0) TJ 0.06 d) O) RJ 0.04 > < 0.02 0 • Non-destination Multiplexing - Destinatbn Multiplexing 2 3 4 Fraction of time faded (-log10(P(B))) Figure 4.28 Average delay vs. fraction of time faded • From P B = 0.001 to 0.0001, the throughputs of destination multiplexing and non-desti-nation multiplexing remain the same. It means that when the channel is in a good state, destination multiplexing and non-destination multiplexing have the same throughput. Chapter 4 Simulation Results and Discussions 60 • From P B = 0.001 to 0.0001, the average delays for destination multiplexing and non-destination multiplexing are kept constant although their average delays are different. This is again because the P B is so small that the channel almost always is in the good state. • There are some differences of average delay for destination multiplexing and non-des-tination multiplexing when the channel is almost not faded. The reason is that the se-lection of the next frame is different for destination multiplexing and non-destination multiplexing. For destination multiplexing, the data frame in the same queue is chosen when the transmission is successful. However, for non-destination multiplexing, the oldest data frame is always chosen. Chapter 5 Conclusions 61 Chapter 5 Conclusions This thesis describes the performance of destination multiplexing for the IEEE 802.11 MAC layer protocol. Since fading is unavoidable in a real wireless channel, frames will sometimes fail in transmission. How to select the next frame to send is the main topic studied in this thesis. The normal FIFO scheduling would cause the AP to keep sending the same frame to the same destination. It is obvious that the performance of such system can be low. If we use the throughput and average delay as the criteria of performance, then the throughput would be low and the average delay would be high. This enables us to consider a better frame scheduling scheme to improve the overall system performance. Destination multiplexing is one of the solutions to this problem. The idea of destination multiplexing is to pick a frame which has the biggest probability to be successfully sent to the destination. In this thesis, we mainly study the infrastructure WLAN. To simplify the problem and not lose the generality, we consider the basic structure of one AP and four nodes in most cases. It actually represents a practical situation since many nodes may be associated to an AP while the number of active nodes is much smaller. The IEEE 802.11a and IEEE 802.11b standards were published when this thesis research was being done. They use OFDM and DSSS as their PHY and have much higher data rates. We first studied the basic performance of destination multiplexing and non-destination multiplexing under the different data rates. We also studied the multiple APs and many-node scenarios. Then the variable data frame length was investigated to see how destination multiplexing affects the performance. We also studied the no-RTS/CTS data frame transmission scenario. The algorithm of selecting the next frame is also of interest. We look at two different algorithms' performance. Finally, the fading parameters have been changed to see how Chapter 5 Conclusions ' 62 the performance of destination multiplexing and non-destination multiplexing are affected. 5.1 Conclusions drawn From the results at different data rates, we have shown that destination multiplexing increases throughput by 20% to 30% and the average delays of destination multiplexing are decreased to about half of the non-destination multiplexing ones. So, we are sure that destination multiplexing does improve the overall system performance no matter what data rates are employed in WLAN. The results for more than four active nodes illustrates the performance improvement when the number of nodes is increased. When the number of nodes is close to four, the throughput is linearly going up with the increase of the node number. When the node number is more than seven, the throughput is saturated depending on the offered load. The throughput of the destina-tion multiplexing scheme is 30% to 40% more than that of the non-destination multiplexing scheme when saturation happens. For the average delay, we get better performance for destination multiplexing with less than half the average delay of non-destination multiplexing. If collisions occur in the system, performance can also be improved when destination multiplexing is used. However, the performance improvement is not as large as that for no-collision cases. This is because the collisions cause the throughput to be lower. Destination multiplexing improves the throughput by approximately 15%. For different data frame lengths, the results are what we would predict. The medium size frame's throughput and average delay are between that of the small size frame and that of the large size frame. The main reason is that overheads involved in each frame transmission makes Chapter 5 Conclusions 63 the time required for sending data frames of different sizes very close. Thus, longer frames achieve higher throughput and less average delay per byte of data. When RTS/CTS is not used before the actual data frame, we expect that the throughput increases and the average delay decreases because of the reduced overheads. This can be seen from our simulation results. The performance of destination multiplexing is still better than that of non-destination multiplexing. The improvement rate is almost the same as that with RTS/CTS because RTS/CTS is not used in both destination multiplexing and non-destination multiplexing. From comparing three algorithms for selecting the next transmitted frame, we conclude that there is not much difference among the throughputs of three algorithms but the average delay and the variance of the delay show some changes at high loads. The normal destination multiplex-ing algorithm that keeps sending to the same destination after successful transmission of the last data frame has the best performance improvement in all three algorithms based on its throughput and average delay. However, as expected, its standard deviation is greatest. The "production" algorithm is easiest to implement and the CSDP-RR algorithm has the most implementation complexity. We varied the fraction of time the channel is faded to see the effect of fading. We conclude that the throughput and average delay for both destination multiplexing and non-destination multiplexing are not changed when the fraction of time faded is less than 10" 2. When the fraction of time faded is small, the channel is nearly ideal with no fading. Thus the throughput and average delay keep almost constant. The smaller average delay for destination multiplexing is because the selection of the next transmitted data frame for destination multiplexing is different from that for non-destination multiplexing. The same effect can also be seen in two algorithm simulations. Chapter 5 Conclusions 64 The overall conclusion is the performance of the system using destination multiplexing is significantly better. The throughput increases up to more than 20% and there are even more significant reductions in average delay. 5.2 Suggestions for future research Some interesting areas for future research are described below. The first research project is to derive the mathematical model for destination multiplexing and non-destination multiplexing for performance study. If we have mathematical formulas to describe how the throughput or the average delay depends on frame arrival, channel model, and destination multiplexing or non-destination multiplexing, we can have better understanding of the overall system performance. The second research project is to study the performance when the fragmented data frames are used. Some data frames are fragmented before sending to reduce the impact of frame loss. It would be useful to consider such a scenario. The third research project is to employ other fading channel models to do simulations. Although, the Gilbert channel model is a good model to simulate the fading wireless channel with bursty error, we would like to verify that destination multiplexing also has better performance if other channel modes, such as Rayleigh fading channel, are used. The fourth research project is to study how destination multiplexing would affect the IEEE 802.lie quality of service (QOS) extensions. It is a new topic that has not been touched on yet. Glossary AP Access Point ACK Acknowledgment BSA Basic Service Area BSS Basic Service Set CBQ Class Based Queuing CRC Cyclic Redundancy Check CS Carrier Sense CSDP Channel State Dependent Packet CSMA/CA Carrier Sense Multiple Access with Collision Avoidance CW Contension Window DSSS Direct Sequence Spread Spectrum DCF Distributed Coordination Function DIFS Distributed Coordination Function Interframe Space DS Distribution System EIFS Extended Interframe Space ESS Extended Service Set FHSS Frequency Hopping Spread Spectrum FIFO First In First Out FTP File Transmission Protocol HOL Head Of Line IBSS Independent Basic Service Set IFS Interframe Space LANs Local Area Networks MAC Medium Access Control MPDU MAC Protocol Data Unit MSDU MAC Service Data Unit NAV Network Allocation Vector 65 OFDM Othogonal Frequency Division Multiplexing OMNeT++ Object Modular Network Testbed in C++ OPNET OPtimized Network Engineering Tools PHY Physical Layer PCF Point Coordination Function PTES Point Coordination Function Interframe Space PLCP Physical Layer Convergence Protocol PPDU PLCP Protocol Data Unit PSDU PLCP Service Data Unit QOS Quality Of Service RTS/CTS Request To Send / Clear To Send SIFS Short Interframe Space TCP Transmission Control Protocol TDMA Time Division Multiple Access UDP User Datagram Protocol WAN Wide Area Network WLANs Wireless Local Area Networks Bibliography [I] IEEE 802.11 - Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, May 9, 1997. [2] ISO/IEC 8802-11 - Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, August 20, 1999. [3] IEEE Std 802.11a-1999 - Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: High-speed Physical Layer in the 5 GHZ Band, September 16, 1999. [4] IEEE Std 802.11b-1999 - Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: Higher-Speed Physical Layer Extension in the 2.4 GHz Band, September 16, 1999. [5] P. Bhagwat, et al, "Using channel state dependent packet scheduling to improve TCP throughput over wireless LANs" Wireless Networks, pp 91-102, 1997. [6] P. Bhagwat, et al, "Enhancing throughput over wireless LANs using channel state dependent packet scheduling", Proceedings of IEEE INFOCOM 96, pp. 1133-1140. [7] S. Desilva, et al, "Experimental evaluation of channel state dependent scheduling in an in-building wireless LAN", Proceedings 7th International Conference on Computer Communications and Networks, pp 414-421, 1998. [8] M. Inoue, "Channel state dependent resource scheduling for wireless message transport with framed ALOHA-reservation access protocol", IEICE Transaction on Fundamentals of Electronics, Communications and Computer Sciences, pp 1338-1346, July 2000. [9] M. Inoue, et al, "Channel state dependent resource scheduling for wireless message transport", IEEE Vehicular Technology Conference, pp 1264-1268, 1998. [10] M. Inoue, et al, "Link-adaptive resource scheduling for wireless message transport", IEEE Global Telecommunications Conference, pp 2223-2228, 1998. [II] M. Inoue, et al, "Resource scheduling with channel state information for wireless message transport", ICUPC '98, Proceedings of IEEE 1998 International Conference on Universal Personal Communications, pp 249-253, 1998 67 Bibliography 68 [12] C. Fragouli, et al, "Controlled multimedia wireless link sharing via enhanced class-based queuing with channel state dependent packet scheduling", Proceedings of IEEE INFOCOM 98, pp 572-580, 1998. [13] Y.-K. Kwok, et al, "A novel channel-adaptive uplink access control protocol for normadic computing", IEEE Transactions on Parrallel and Distributed Systems, pp 1150-1165, 2002. [14] V.K.N. Lau, Y.-K. Kwok, "A channel state dependent bandwidth allocation scheme for integrated isochronous and bursty media data in a cellular mobile information system", Proceedings - IEEE Military Communications Conference MILCOM, pp 524-528, 2000. [15] A. Varga, OMNeT++ discrete event simulation system, http://www.hit.bme.hu/phd/vargaa/ omnetpp.htm. [16] W. K. W. Cheung, "Performance Study of the IEEE 802.11 Wireless Media Access Control Protocols", M. A. Sc. thesis, Dept of Electrical and Computer Eng, Univ of British Columbia, Vancouver, Canada, 1995. [17] E. N. Gilbert, "Capacity of a Burst-Noise Channel", The Bell System Technical Journal, September 1960, pp 1253-1265. [18] D. Kramer, "The Network Processor Revolution Fast Pattern Matching and Routing at OC-48, http://www.hotchips.org/archive/hcl3/hcl3pres pdfZ13agere.pdf. Stanford Univ, Aug 19-21,2001. [19] E. Casas, Vivato Inc, personal communications, February 2003. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0091330/manifest

Comment

Related Items