UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Priority CSMA schemes for integrated voice and data transmission Ching, Kai-Sang 1988

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

Item Metadata

Download

Media
831-UBC_1988_A7 C44.pdf [ 3.68MB ]
Metadata
JSON: 831-1.0064778.json
JSON-LD: 831-1.0064778-ld.json
RDF/XML (Pretty): 831-1.0064778-rdf.xml
RDF/JSON: 831-1.0064778-rdf.json
Turtle: 831-1.0064778-turtle.txt
N-Triples: 831-1.0064778-rdf-ntriples.txt
Original Record: 831-1.0064778-source.json
Full Text
831-1.0064778-fulltext.txt
Citation
831-1.0064778.ris

Full Text

P R I O R I T Y C S M A S C H E M E S F O R I N T E G R A T E D V O I C E A N D D A T A T R A N S M I S S I O N by K A I - S A N G C H I N G B . A . S c , University of Windsor, 1986 A T H E S I S S U B M I T T E D IN P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F A P P L I E D S C I E N C E in T H E F A C U L T Y O F G R A D U A T E S T U D I E S Department of Electrical Engineering We accept this thesis as conforming to the required standard T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A August, 1988 © K a i - S a n g Ching, 1988 In presenting this thesis in partial fulfilment 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. Department of "^£&jiJ&r]t**JZ. The University of British Columbia Vancouver, Canada Date £Xc^3 > t f g f DE-6 (2/88) Abstract Priority schemes employing the inherent properties of carrier-sense multiple-access ( C S M A ) schemes are investigated and then applied to the integrated trans-mission of voice and data. A priority scheme composed of 1-persistent and non-persistent C S M A protocols is proposed. T h e throughput and delay characteristics of this protocol are evaluated by mathematical analysis and simulation, respectively. T h e approach of throughput analysis is further extended to another more general case, p-persistent C S M A with two persistency factors, the throughput performance of which had not been analyzed before. Simulations are carried out to study the delay characteristics of this protocol. After careful consideration of the features of the priority schemes studied, two protocols are proposed for integrated voice and data transmission. While their ultimate purpose is for integrated services, they have different application. One of them is applied to local area network; the other is suit-able for packet radio network. The distinctive features of the former are simplicity and flexibility. T h e latter is different from other studies in that collision detection is not required, and that it has small mean and variance of voice packet delay. Performance characteristics of both of these protocols are examined by simulations under various system parameter values. ii Table of Contents Abstract ii List of Figures v Acknowledgements y i i 1 Introduction 1 1.1 Background 1 1.2 Objective and Organization of the Thesis 4 2 Network with 1-persistent and non-persistent C S M A Protocols 6 2.1 Review of Protocols 7 2.2 Throughput Analysis 7 2.2.1 Derivation of 5 X 8 2.2.2 Derivation of S2 9 2.3 Throughput Characteristics 9 2.4 Delay Characteristics 10 2.5 Summary I 7 3 Performance Evaluation of p-persistent C S M A with Two Persistency Factors 18 3.1 Review of Protocol 19 3.2 Throughput Analysis : infinite users 20 3.2.1 Derivation of Si 20 3.2.2 Derivation of S2 26 3.3 Throughput Characteristics : infinite users 26 3.4 Throughput Analysis : finite users 32 iii 3.5 Throughput Characteristics : finite users 33 3.6 Delay Characteristics 36 3.7 Summary 38 4 P r o t o c o l f o r I n t e g r a t e d V o i c e a n d D a t a T r a n s m i s s i o n i n L A N 42 4.1 Channel Access Protocol 43 4.2 Characteristics and Requirements for Voice and Data Traffic 43 4.3 Simulation Model 44 4.4 Simulation Results 46 4.4.1 Voice Transmission 46 4.4.2 Integrated Voice and Data Transmission 48 4.4.3 Comparisions and Discussions 51 4.5 Summary 54 5 I n t e g r a t e d V o i c e a n d D a t a T r a n s m i s s s i o n i n P a c k e t R a d i o N e t w o r k 55 5.1 Description of Protocol 56 5.2 Simulation Model 59 5.3 Simulation Results and Discussions 61 5.3.1 Voice Transmission 61 5.3.2 Integrated Voice and Data Transmission 63 5.4 Comparisons with Reservation A loha 69 5.5 Summary 82 6 C o n c l u s i o n s 83 A p p e n d i x A 86 B i b l i o g r a p h y 94 iv List of Figures 2.1 Throughput characteristics of Si and S2 (a=0.3) 11 2.2 Throughput characteristics of 5X and 5 2 (ct=0.6) 12 2.3 Packet delay of class 1 and class 2 versus throughput for constant a 13 2.4 Simulated delay-throughput performance (without collision detection) . . . 15 2.5 Simulated delay-throughput performance (with collision detection) 16 3.1 Channel state of p-persistent C S M A : busy period and idle period 21 3.2 Throughput characteristics of 5X and S2, infinite users (pi = 0.9, p2 = 0.6, a = 0.2, 0.5, 0.8) 28 3.3 Throughput characteristics of Si and S2, infinite users (pi = 0.8, p2 = 0.3, a = 0.2, 0.5, 0.8) 29 3.4 Throughput characteristics of Si and 52, infinite users (pi = 0.8, p 2 = 0.03, a = 0.2, 0.5, 0.8) 30 3.5 M a x i m u m throughput of p-persistent C S M A with two persistency factors versus a 31 3.6 Throughput characteristics of 5X and 52, finite users (a = 0.1) 34 3.7 Throughput characteristics of Si and 52, finite users (a = 0.3) 35 3.8 Throughput characteristics of Si and 52, finite users (a = 0.5) 37 3.9 Simulated delay-throughput performance (without detection) 39 3.10 Simulated delay-throughput performance (with collision detection) 40 4.1 Voice packet loss rate versus number of calls, n 47 4.2 Normalized mean delay versus data throughput [pd) 49 4.3 Voice packet loss rate versus data throughput (pd) 50 4.4 Standard deviation of delay versus data throughput (pd) 52 v 5.1 Voice packet loss rate versus voice throughput for different p„'s 62 5.2 Comparison of voice packet loss rate of T A S I with that of the proposed scheme 64 5.3 Voice packet loss rate versus data throughput for different p u 's 66 5.4 Mean and standard deviation of first voice packet delay 67 5.5 Mean and standard deviation of subsequent voice packet delay 68 5.6 Mean and standard deviation of data packet delay 70 5.7 Voice packet loss rate versus number of calls without data load 72 5.8 Components of voice packet loss rate versus number of calls 74 5.9 Frequency versus loss of consecutive packets in the proposed scheme, n = 30 75 5.10 Frequency versus loss of consecutive packets in the r-aloha scheme, n = 30 76 5.11 Frequency versus loss of consecutive packets in the proposed scheme, n = 37 77 5.12 Frequency versus loss of consecutive packets in the r-aloha scheme, n = 37 78 5.13 Normalized voice packet delay versus data throughput {pd) 80 5.14 Normalized data packet delay versus data throughput (pd) 81 A . l Components in the j th sub-busy period 88 vi Acknowledgements T h e advice and assistance from my research supervisor, Dr . H . W . Lee, are gratefully acknowledged. vii Chapter 1 Introduction 1.1 Background W i t h the advent of computers, which are ubiquitous nowadays, data traffic has grown tremendously. In order to handle data traffic efficiently, packet transmission networks have been developed [l], A R P A N E T , A L O H A , and E T H E R N E T networks are typical examples. These packet-switched networks have established themselves as an attractive option of data transmission. Traditionally voice and data transmission have been handled by different com-munication networks. But like data, voice traffic has also been increasing steadily [2]. W i t h voice and data traffic growing at this alarming rate, it is inefficient to handle such data using disjoint networks. Thus , packet voice communication has been receiving considerable attention and basic issues associated with packet voice communication are well-elucidated in [3]. Reduction from noise and crosstalk is one of the benefits offered by packet voice. Another more important advantage is that packet voice is compatible with data packet with the consequence that voice and 1 CHAPTER 1. 2 data can be multiplexed together and transmitted in a single network [ 4 ] . Combin-ing voice and data in a single network promises several advantages [ 5 ] : • Cost reduction; transmission and switching facilities can be shared by both types of packets; and cost of maintenance, components, and production are guaranteed to be lower. • Speech alternates between talkspurt and silence periods; data packets can make use of the silence periods which may result in higher utilization. • More sophisticated services can be provided to users. However, integrating voice and data is not a simple task. Because of their completely different characteristics, combining them creates problems. Data traffic is usually characterized as bursty; that is, there is a large peak-to-average data rate. Unlike data, voice packets are generated periodically while in talkspurts. Besides, voice packets demand small average and variance in delay, whereas data packets do not have such stringent demands. T o fulfill these requirements, priority multiaccess protocols are necessary to prioritize voice and data packets. A m o n g the various random access priority protocols available, carrier sense mul-tiple access ( C S M A ) is a very popular one [6,7,8]. There are two main reasons for this. First , it has been shown that compared to other random access schemes, C S M A performs quite efficiently if the propagation delay is short relative to the packet transmission time [9]. Second, an improved version, known as C S M A with collision detection ( C S M A / C D ) , has been successfully applied to local computer networks, typified by the Ethernet [10]. CHAPTER 1. 3 A l l of these protocols [6,7,8] employ a time interval called reservation period. Th is reservation period ensures the higher priority messages having complete access to the channel over the lower priority ones. But this reservation period is an addi-tional cost because users have to transmit this reservation period before sending its own packets. A p-persistent C S M A / C D priority resolution scheme that does not employ a reservation period has been proposed in [17]. In this scheme, the higher the priority of a packet, the higher the probability of transmission. In other words, each priority class has its own probability of transmission. Since the characteristics of voice and data packets are dissimilar, ordinary prior-ity schemes may not be suitable for integrated voice and data transrnission. Proto-cols, therefore, must be well-designed. Probably due to the omnipresence of Ether-net and the advantages of collision detection, many good protocols (e.g. [11,12,13]) have been proposed for such application in local area networks (LANs) . Neverthe-less, these protocols are so specifically devised to suit the characteristics of voice and data packets that they become rather complex. Implementing a complex algo-r i thm, sometimes, may be very costly. In view of this, a protocol which is simple, robust, and has an acceptable performance is undoubtedly desirable. Strangely enough, while there are abundant protocols for integrated applications in L A N s , exceedingly few protocols have been proposed for integrated voice and data transmission in packet radio network. A s a matter of fact, some published works dealt with voice transmission only [14,15]. Since quick collision detection may not be available in a packet radio network, the performance of a protocol can be easily degraded if collision occurs frequently. A s a result, a protocol for integrated services should be designed in such a way that frequent collisions between voice and CHAPTER 1. 4 data packets should be avoided. One simple way of achieving this goal is to take advantage of the characteristics of voice and data traffic. Some of the related works and motivations have been briefly reviewed in this part of the introduction. The objectives and organization of this thesis are described below. 1.2 Objectives and Organization of the Thesis There are three variants of C S M A protocols, namely, non-persistent, 1-persistent, and p-persistent. E a c h of these protocols has its own inherent properties. W i t h the proper combination, a priority scheme can be implemented. In chapter 2, we suggest that 1-persistent and non-persistent is a good combi-nation, and that it can be implemented as a non-preemptive priority scheme. It is different from other works in that, like the p-persistent priority scheme proposed in [17], it does not use reservation period. T h e throughput expressions are derived, one 'for each type. Although throughput expressions of a similar system 1 has been analyzed in [16], the approach adopted in chapter 2 is much simpler. Most impor-tantly, the approach can be easily extended to a more general case, p-persistent C S M A . Unlike the p-persistent C S M A , there has been no delay-throughput characteris-tics of the p-persistent C S M A with different probabilities of transmission available in the literature. Hence, the analytic approach of chapter 2 is applied to this case. 1The system in [16] consists of non-persistent, 1-persistent, and virtual-time CSMA. CHAPTER 1. 5 T h e analysis shown in chapter 3 is the first piece of work done on the throughput performance of this protocol. Although only two transmission probabilities are in-volved in the analysis, the case of more than two can also be included. T o better understand the throughput performance, the cases of infinite and finite users are considered too. Simulations are performed to observe the delay characteristics. Observations from analytic and simulation results in chapter 3 show that, under data traffic, the protocol of p-persistent C S M A / C D with two transmission proba-bilities is well behaved. Hence, the application of this protocol to the integration of voice and data transmission in L A N s is first proposed in chapter 4. Compared to other works, the distinctive features of this protocol are simplicity and flexibility. Besides, simulation results indicate that it is a viable scheme. Since not many works have been done in the area of integrated voice and data transmission in packet radio network, in chapter 5 we propose a scheme [28] with an acceptable level of performance. One of the interesting properties of our protocol is that no collision detection is required. Furthermore, the mean voice packet delay, in general, is within two voice packets transmission time, and the variance of delay is low. These properties are important if good voice quality is desired. Chapter 6 concludes the thesis. It summarizes the research results presented in previous chapters. Chapter 2 Network with 1-persistent and non-persistent CSMA Protocols This chapter is concerned with the throughput and delay characteristics of a network made up of two groups of users. T h e first group uses the 1-persistent C S M A protocol to access the channel while the second group applies the non-persistent C S M A protocol. Throughput expressions are derived. These will be used to investigate the throughput characteristics of each group. Problems involved in evaluating the resultant throughput are discussed. Besides, simulations are performed to observe the delay characteristics. Results indicate that a simple non-preemptive priority scheme can be implemented as follows: a higher priority group transmits in 1-persistent mode and the lower priority group transmits in non-persistent mode. 6 CHAPTER 2. 7 2.1 Review of Protocols Operation of the 1-persistent and non-persistent in the network is assumed to be slotted. A slot is of r seconds long, where r is the end-to-end propagation delay. Besides, all users are forced to start transmission at the beginning of a slot. Slotted non-persistent C S M A is considered first. When a user is ready the channel is sensed and the following steps are carried out: 1. If the channel is idle, the packet is transmitted; or 2. If the channel is busy, the ready user reschedules the packet according to the retransmission algorithm; and then, go to step 1. T h e slotted 1-persistent protocol is as follows: 1. If the channel is idle, the packet is transmitted ; or 2. If the channel is busy, the ready user persists sensing the channel and transmits the packet as soon as it is sensed idle. 2.2 Throughput Analysis For simplicity, the users who transmit in 1-persistent mode are called class 1 and those who transmit in non-persistent mode are labeled class 2. Hence, the through-put of class 1 is denoted by Si and that of class 2 by 52. It is assumed that the packet arrival process and channel traffic follow the Poisson distribution and that CHAPTER 2. 8 the arrival processes of class 1 and class 2 are independent. Also, the channel traffic G comprises the traffic generated by class 1, Gt and by class 2, G2. Thus G = G i + G2. Throughout the derivations, a represents the normalized end-to-end propagation delay and TP, the transmission period which equals 1+a. Using the renewal theory arguments 1 [9] the average throughput is given by s = Trn <"> where U is average time during a cycle that the channel is used without conflicts, B the mean busy period, and J the mean idle period. Accordingly, Si = Ui/(Bi + Ji ) and S2 = U2/(B2 + l2). 2.2.1 Derivation of Si Since the number of slots in an idle period and a busy period are geometrically distributed with mean 1/(1 - e-aClle-aGa) and l / (e~ ( G l + o ( ? ) ) , respectively, we have and fii = e - ( e t + « c ) ( 2- 3) T o find 17x, one can compute the probability of success over the first TP and the probability of success over any other TP. T h e mean number of TP is l / ( e - ( G l + a G ) ) ; therefore, U 1 - 1 _ c-aC7 + l e !J ! _ c-(G 1 +aC7) VA> 1A Markov chain approach can also be used. In [16], this method is applied to compute the throughput of a system with three different CSMA protocols. CHAPTER 2. 9 Hence, g 1 ( l + e-e-«0)e-t°'+«g) . , 1 (1 + a) (1 - e - a G ) + a e - ( ° » + » G ) 1 ' ; 2.2.2 Derivation of S2 If J denotes the number of idle slots in an idle period, then Pr[I = k] = e - ( O i + « o ) ( e — o j f c - i ^ _ e-a<7) fc > j_ ( 2 6 ) T h u s , a e - ( G l + o G ) 1 - e In the non-persistent mode, £? 2 is simply a transmission period. Therefore, B2 = 1 + a and the average successful transmission is Putt ing U2, f?2 and I 2 together, we obtain o * G 2 e - G  2 (1 + o) (1 - e-«G) + oe-(°»+«c) 1 J 2.3 Throughput Characteristics In the previous section, two expressions for Si and S2 are derived in terms of the normalized propagation delay, a , channel traffics of both classes, G\ and G2. In order to investigate the throughput characteristics of both classes, the approach in [16] is followed: introduce one more parameter, a , which is defined as follows. G i = ctG; G2 = (l- a)G\ 0 < a < 1 (2.10) CHAPTER 2. 10 T h e throughput characteristics of class 1 and class 2 for a = 0.3 and 0.6 are shown in Figures 2.1 and 2.2. T h e value of a is kept at 0.01, and is the same for all the results presented in this chapter. E a c h figure is composed of parts (a) and (b). Part (a) shows the throughput of class 1 and class 2 for a specific value of a. Part (b) compares the resultant throughput (Si + St) of this heterogeneous network with that of a network using either 1-persistent or non-persistent protocol. T h e resultant throughput is depicted by a dotted line which terminates at the maximum of Si or 52, whichever occurs first. The reason for this will be explained in the next section. A s shown in Figure 2.1, when a = 0.3, both classes have their maximum through-put occurred at the light traffic regions, which is similar to the 1-persistent. But as a increases to 0.6, class 2 performs very badly and the maximum throughput is about 10%. T h e reason is that when class 1 traffic is dominant, the chance for class 2 users sensing a busy channel increases, which forces the ready users to be-come backlogged, thereby driving S2 down. Observations from other a's show that when a = 0.23, the maximum of S i and S2 are very close, and that the resultant throughput cannot be greater than the case of the non-persistent protocol. 2.4 Delay Characteristics T h e delay performance is investigated by simulation. T o be confident of the simu-lator, it is validated by comparing the (S,G) relationship obtained from simulation with those obtained from analytic expressions of Si and S 2 ; they match very well. T h e normalized mean packet delay of both classes for a = 0.3 and 0.6 are shown in Figure 2.3. It is a well known fact that the packet delay becomes enormously CHAPTER 2. 1.0 0.9 0.8 0.7 0.6 0.5 0.4 h 0.3 0.2 -0.1 -a = 0.3 class 1 class 2 0.0 1x10 1x10° 1X101 Channel Traffic: G (a) Si and S2 versus G 1x10' non - persistent • • t i . i 1 x 1 0 ° 1 x 1 0 1 Channel Traffic: G 1x10' (b) Throughput comparison Figure 2.1: Throughput characteristics of Si and S2 (a=0.3) CHAPTER 2. 1 .w 0.9 a = 0.6 0.8 class 1 0.7 class 2 0.6 0.5 0.4 0.3 0.2 0.1 0.0 1x10 i i • i • i i 1x10° 1x101 1x10' Channel Traffic: G (a) Si and S2 versus G 1.0 0.9 - a = 0.6 non - persistent C.E -0.7 0.6 - \ Throughpu 0.5 0.4 -/\ \ 1 * persistent 0.3 0.2 -0.1 0.0 1x10 -1 1x10° 1x101 Channel Traffic: G (b) Throughput comparison 1x10* Figure 2.2: Throughput characteristics of Si and S2 (a=0.6) CHAPTER 2. 1x102 CO ? JZ u ca a. IxlO1 -1x10° Throughput: , S2 (a) Packet delay versus throughput for a = 0.3 1x1 o2 Cm i 1 73 1X101 1x10° • 1 o = 0.6 1 1 1 t 1 class 1 class 2 1 I 1 1 I T I i / / / / I / • ,e T" . i i • i . i 0.0 0.1 0.2 0.3 0.4 Throughput: S, , S2 0.5 0.6 (b) Packet delay versus throughput for a = 0.6 Figure 2.3: Packet delay of class 1 and class 2 versus throughput for constant CHAPTER 2. 14 large in the vicinity of maximum throughput. There is no exception in the case of two classes operating in the same network, packet delay approaches infinity as maximum of Sx or S2 is reached. Since these maximum points occur at different values of G, problems arise when adding Si and S2. When Gi and G2 increase in such a way that a is maintained constant, users in one group may still enjoy the low delay, but those in the other group suffer from an infinitely large delay. Simulated results shown in Figure 2.3 support this observation. In consequence, the resultant throughput cannot be added arbitrarily. In fact it is dependent upon when the maximum throughput of each class occurs. Tha t is why the dotted line terminated in part (b) of Figures 2.1 and 2.2. Thus far the delay-throughput performance has been shown for constant value of a. T h e case of various a is shown in Figures 2.4 and 2.5. Figure 2.4 describes the trade-off of packet delay and throughput when the input rate of the other class is fixed at 0.1, 0.2, and 0.3. A s shown, the class 1 delay is low whereas the class 2 delay increases sharply. Th is is largely due to the inherent properties of 1-persistent and non-persistent protocols. T h e case of collision detection is shown in Figure 2.5. Collision recovery time is assumed to be two slots. Compared with Figure 2.4 (the case with no collision detection) one can notice the delay and throughput are greatly improved. A C S M A / C D priority scheme using non-persistent and 1-persistent protocols has been proposed in [6]. However, their scheme requires an initialization signal (a form of reservation period) before a packet is transmitted. But one can see from Figures 2.4 and 2.5 that the class 1 packet delay is lower than that of class 2. In case of collision detection, the throughput-delay performance is even better. In other CHAPTER 2. 15 1x102 2 0 1x101 2 1 1x10° 0.0 O : S 2 «0.1 A : S 2 - 0.2 + : S 2 -0 .3 J . 0.5 0.6 0.7 0.8 0.9 Throughput: S, (a) Packet delay versus throughput, class 1 1x102 I CS Oh 0 1x101 -J 1 1x10° O : 5,-0.1 A : S, = 0.2 + : S, - 0.3 j L 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.0 Throughput: S 2 (b) Packet delay versus throughput, class 2 Figure 2.4: Simulated delay-throughput performance (without collision detection) CHAPTER 2. 16 Throughput: S 2 (b) Packet delay versus throughput, class 2 Figure 2.5: Simulated delay-throughput performance (with collision detection) CHAPTER 2. 17 words, the presence of class 2 users has little effect on the users of class 1, except when Si + S2 approaches maximum throughput. Th is nice feature, therefore, can be used as a non-preemptive priority scheme. Th is scheme is simple because users have to follow the 1-persistent and non-persistent protocols. No further actions or reservation periods are required from the users. 2.5 Summary A network with two groups of users, one using the 1-persistent, the other the non-persistent C S M A protocol, is considered in this chapter. T w o expressions of throughput (without collision detection), one for each group, is derived. It is found that the resultant throughput is dependent on where the maximum throughput of each group occurs. In general, this throughput cannot perform better than the case if both groups apply the non-persistent protocol. However, one can notice, from the simulated delay-throughput performance, that a simple non-preemptive priority scheme can be implemented by simply assigning the higher priority a 1-persistent mode and the lower priority a non-persistent mode. Chapter 3 Performance Evaluation of p-persistent CSMA with Two Persistency Factors Recently, a priority resolution scheme using p-persistent C S M A / C D was proposed in [17]. In this scheme, the higher the priority of a packet, the higher its p-persistent probability for transmission. For example, if there are two classes of packets when the channel is sensed idle, the higher priority packets transmit with probability Pi (persistency factor), while the lower ones transmit with probability p 2 , where Pi > P2- In [17], the protocol was evaluated in terms of non-transmitting slots. 1 However, the throughput-delay performance is also of great interest. T o the best of the author's knowledge, there has been no such performance analysis in the literature to date. Therefore, this chapter provides a throughput analysis of p-persistent C S M A (without collision detection) with two persistent factors. The 1Most of the priority schemes axe implemented by reservations. As a result, time slots are wasted when there are no high priority packets. The protocol proposed in [17] employed probabilities rather than reservation to resolve conflicts; therefore, no time slots are wasted when there are no high priority packets. 18 CHAPTER 3. 19 approach adopted in the analysis can be extended to the case of any number of persistent factors, although the derivations will be very cumbersome. Both the infinite users case and finite users case are considered. A s the delay analysis is of great difficulty and complexity, it is evaluated by simulation. 3.1 Review of Protocol T h e slotted p-persistent C S M A is considered here. T h e time is quantized into slots of T seconds, where t is the end-to-end propagation delay. Besides, all users are forced to start transmission at the beginning of a slot. If a user is ready, it proceeds as follows: 1. If the channel is sensed idle, it transmits with probability p; with probability (1 — p), it defers the transmission to the next slot. 2. A t this new slot, if this user still senses an idle channel, step (l) is repeated; otherwise, this user reschedules its packet according to the retransmission algorithm. 3. If the channel is sensed busy, it persists sensing the channel until it is idle, and then it will go to step (1). CHAPTER 3. 20 3.2 Throughput Analysis : infinite users T h e analysis provided here focuses on two groups of an infinite number of users but the approach can be extended to any number of groups. For discussion purposes, users in the high priority group are named class 1, and those in the low priority group class 2. T h e persistency factors assigned to class 1 and class 2 are pi and P2, respectively. A s in the previous chapter, Si and Gi denote the throughput and channel traffic of class 1; likewise, S2 and G2 denote those of class 2. T h e total channel traffic is Gi + G2 = G. Packet arrival processes of both classes and channel traffic are assumed Poisson. Besides, the arrival processes of class 1 and class 2 are assumed independent. The time axis is normalized to the packet transmission time, where a denotes the normalized propagation delay. T h e derivation of the throughput of p-persistent C S M A with two persistent factors follows the approach in [9] in which the p-persistent C S M A was analyzed. Aga in , from the renewal theory, Si and S2 are given by Ui/(B + I) and U2/(B + T), respectively. T h e definitions of Ui, U2, B, and I are the same as those in the previous chapter. Note that B and J do not need supscripts 1 and 2, because they are identical for both class 1 and class 2. 3.2.1 Derivation of S\ Figure 3.1 shows an example of the busy and idle period of the p-persistent pro-tocol, together with some notations. If N[ and N2 are the number of class 1 and class 2 packets present at the last slot of the previous idle period, and Ni and N2 CHAPTER 3. 21 < TP IRTD « ,1. . I L. -I l_ J I 1 I , . I I L t : class 1 arrival 't : class 2 arrival a : normalized end-to-end propagation delay TP : transmission period IRTD : initial random transmission delay B : busy period / : idle period Figure 3.1: Channel state of p-persistent C S M A : busy period and idle period CHAPTER 3. 22 are the number of class 1 and class 2 packets accumulated at the end of a transmis-sion period {TP), then according to the Poisson arrivals, and assuming the arrival processes for both classes are independent, 7r'nuna = Pr[Nl = nuNl = n2] = : (9-he~A (^e'A , n1 + n2>0 (3.1) where <7i = o G i , g2 = aG2, and, ""m.na = Pr[Ni = nuN2 = n2] _ K i ± ^ e x p { _ ( 1 + a ) G i } [(l + a) G 2 ] n s n2! e x p { - ( l + a)G2}, ni + n 2>0 (3.2) T h e average length of the I R T D between two consecutive T P ' s can be deter-mined as follows. Let 171 = 1—pi, q2 = 1—p2, and <ni,na be the number of slots elapsed until some class 1 or class 2 packet is transmitted, given that Ni — tii,N2 = n2. T h e n , P r [ * n i > n a > *] = exp ( g l ( g l ( l - g l t ) - k)^j . c 7 2 ^ e x p ( c 7 2 ( g 2 ( l - g 2 t ) - f e ) ) (3.3) CHAPTER 3. 23 Therefore, for k > 0, P r [tnun2 = k] = gi kn*q2 kn> [l - q^qf* e x p { - f f l ( l - - g2{l - q2k)}] •exp Pi P2 (3.4) and for k = 0, P r [ t n i ln 2 = 0] = l - g i n i < ? 2 n 2 (3.5) Removing the condition on Ni and N2, the average I R T D is given by n i + n 3 > o where, 2 0 1 — 0^,0 ?n 1 („ 2 = E P r [ ^ , n 2 > A : ] (3.7) k=0 T h e computation of the probability of success over a transmission period (TP) is split into two parts: the first TP; and, then any other T P ' s . Let T P , denotes the i th TP in the busy period, P}(niyn2), the probability of success for class 1 packets over TP{, and L n i t T i 2 , the number of class 1 and class 2 packets present when r e -starts. CHAPTER 3. 24 First consider the case t ^ 1. Given Ni = ni and JV2 = rc2, + ( l-?i n i92 n 2)^ 1,n I 4.n a, ( / i , / 2 ) > ( n i , n 2 ) (3.8) where c5,->;- is the impulse function. Hence, the probability of success for class 1 packets over TP,-, i ^ 1 is given by P}{nun2) = £ £ ^Irl • ^ • P r [ L „ 1 , n 2 = (/1,/2)] (3.9) Unconditioning iVi and 7Y2, we obtain P} = £ P } ( n u n 2 ) ^ - (3.10) Now consider the case t = 1. Let t'nin2 be the initial I R T D , and P ' * ( n i , n 2 ) , the probability of success for class 1 packets over TPi. Al though using different nota-tions, these two quantities have the same distributions as those shown in equations (3.3) and (3.9), except that they are conditioned on N[ and N^. Thus removing these conditions yields, 7 = £ (3-11) ni+nj>0 CHAPTER 3. 25 and, P'\ = £ P'\(nun,)ir'nuni (3.12) T h e task now is to express B in terms of t , t , and U\ in terms of P, 1 , P'\. Notice that the number of T P ' s in a busy period is geometrically distributed w i t h parameter 7r 0 ,o. Thus given that there are m T P ' s , one can write Bm = at + ( m - l)at + m(l + a) (3.13) and, Ulm = P'] + (m-l)P} (3.14) T h e final expressions for B and Hi can be obtained by removing the condition on m. Hence, 00 B = £ S m7r 0 lo(l — T 0 >o) m - 1 m=l = a r + ^ ( i - ^ ) + i + a ( 3 1 5 ) 0^,0 and, tfl = E^lmTo.oU-TTo.or-1 m=l = P>) + Lll2fipi (3.16) CHAPTER 3. 26 T h e average idle period is geometrically distributed with mean 1/(1 — e  Sle  32); therefore, 7 = (3.17) Putt ing equations (3.15), (3.16), and (3.17) into Ui/(B +7) , expression for Si is obtained. 3.2.2 Derivation of 52 T h e expression for S2 can be similarly written by noting that expressions for B and J are the same as those in Su and that the probability of success for class 2 packets over TP{, P , 2 ( n i , n 2 ) , is given by tfKn.) = £ £ ' S i 1 • P r [ L n , n 3 = (/i,/2)] (3.18) T h e procedures for obtaining S 2 would be exactly the same as those for Si. 3.3 Throughput Characteristics: infinite users Recall that Si and S2 are expressed in terms of Gu G2, pi, p2, and a. A s in the previous chapter, another parameter a is required so that Gi and G2 are related to a and G as follows. Gi = aG, G2 = {l-a)G, 0<a<'l (3.19) CHAPTER 3. 27 Strictly speaking, there are innumerable combinations of p i , p 2 , and a. Thus , it is unrealistic to show the effects of these parameters on S i and S 2 . In view of this, three representative cases of different pi 's and P2's are shown from Figures 3.2 to 3.4, in which the values of a are fixed at 0.2, 0.5, and 0.8. Also, a = 0.01 is used in this chapter. Observations from these figures lead to the following points: 1. W h e n pi and p 2 are close, such as the case shown in Figure 3.2, the maximum of Si and S2 occurs almost at the same value of G. Furthermore, the shapes of Si and S2 are very similar. 2. A s observed from Figures 3.3 and 3.4, the peaks of S x and S2 start shifting more when pi and P2 are further apart. In other words, for a specific value of a, the peaks of S i and S2 may occur at different values of G. Consequently, as explained in the previous chapter, the resultant throughput cannot be summed arbitrarily. Also, the shapes of S i and S2 are different, which is typified by a case of a = 0.2 in Figure 3.4. Another interesting thing to look at is the resultant throughput (S x + S 2 ) of this protocol. Figure 3.5 shows the maximum throughput versus a for different values of pi and p2. T h e maximum throughput here is obtained by adding S x and S2 up to the peak value of either S i or S2, whichever occurs first. Another point which should be mentioned is that when a = 0 or a = 1 all users adopt p2-persistent (a = 0) or pi-persistent (a = 1) protocol. Figure 3.5 indicates that the maximum throughput decreases from the maximum point to the lowest point as a varies from 0 to 1. Al though four combinations of p x and p 2 are shown in Figure 3.5, a lot of cases have been tested and they all show this phenomenon. T h e conclusion here is CHAPTER 3. Figure 3.2: Throughput characteristics of Si and iS 2, infinite users (pi = p2 = 0.6, a = 0.2, 0.5, 0.8) CHAPTER 3. 1.0 Figure 3.3: Throughput characteristics of Si and £ 2 , infinite users (pi P2 = 0.3, a = 0.2, 0.5, 0.8) CHAPTER 3. 1.0 Figure 3.4: Throughput characteristics of Si and S2, infinite users (px p2 = 0.03, a = 0.2, 0.5, 0.8) CHAPTER 3. 31 Figure 3.5: M a x i m u m throughput of p-persistent C S M A with two persistency fac-tors versus a CHAPTER 3. 32 that the maximum throughput of this protocol is upper-bounded by the maximum throughput of p2-persistent and lower-bounded by that of pi-persistent. 3.4 Throughput Analysis : finite users This section is mainly motivated by two facts. The first one is the resultant through-put. A s mentioned before, the resultant throughput cannot be arbitrarily added for a specific value of a. T h e second motivation comes from [18]. T h e authors in [18] carried out a throughput analysis of p-persistent when there were a finite number of users, and pointed out that the throughput did not degrade to zero when p was small. Motivated by these two results, one would like to investigate the performance of this protocol under the condition of finite users. Expressions of Si and S 2 can be easily derived by following the approach in [18]. A s shown in Appendix A , S i and S 2 are given by, oo P i M i ] T A • • CMi Si = oo (3.20) 1 + o + aJ2 DMl ' EM* where, A = (1 - pi)* - (1 - gi) l+(l/a) [ P l ( l - P l ) * - g l ( l - g l ) P i - 91 (3.21) CHAPTER 3. 33 B = ( l - P l ) t + 1 - ( l - « 7 i ) 1 + ( 1 / 0 ) P i (1 " P i ) * + 1 " (1 " 9i)k+1 Pi ~ 9i (3.22) C = ( l - p 2 ) t + 1 - ( l - & 2 ) 1 + ( 1 / a ) P 2 (1 " Pl)M - (1 - 92)k+1 Pi - 92 (3.23) D = (1 - P l ) f c - (1 - ffi)1+(1/a) Pi ( l - P l ) k - ( l - f f i ) ' Pi - 9i (3.24) E = (1 - P2)k - (1 - g2)1+Wa) P2 (1-P2)k-(1-92Y P2 - 92 (3.25) Also, d = a G , G 2 = (1 - a ) G , 0 < a < 1, gt = o G i / M i , and g2 = a G 2 / M 2 . M i and M 2 are the number of class 1 and class 2 users, respectively. A l l other notations here are the same as those denned for the infinite users case. Similarly, one can write 5 2 by changing all subscripts 1 in equation (3.20) to 2. 3.5 Throughput Characteristics: finite users T h e throughput characteristics are shown in Figures 3.6 and 3.7, which include part (a) and part (b). Part (a) shows the throughput of each class and the resultant throughput is shown in part (b). Also depicted in part (b) are the throughputs of pi-and p-persistent C S M A with M = 10. A s shown in both figures, the throughputs CHAPTER 3. 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 1x10 a = 0.1 M, = M2 = 5 p, = 0.1 p 2 = 0.01 \ \ • i i i i 1111 • 1x10° ix icr Channel Traffic: G (a) Si and 5 2 versus G _l I L__l 1x10-^  p o 0.01 £ = 0 . 1 0.0 1x10 i i i u 1x10° IxlO 1 • • i i i i i 1x10^ Channel Traffic: G (b) Throughput comparison Figure 3.6: Throughput characteristics of Si and S2, finite users (a = 0.1) CHAPTER 3. 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.0 1x10 0.0 1x10 <x = 0.3 M ,°3 M2 = 7 p, •* 0.3 p 2 «c 0.05 1x10' J 1 1x10' 1x10' Channel Traffic: G (a) 5i and S2 versus G 0.05 • 1 < • • • • ! 1x10" 1X101 1x10' Channel Traffic: G (b) Throughput comparison Figure 3.7: Throughput characteristics of S\ and 5 2 , finite users (a = 0.3) CHAPTER 3. 3 6 of both classes do not drop to zero eventually. Part (b) of Figure 3.7 shows the maximum resultant throughput is bounded by the maximum throughputs of 0.05-and 0.3-persistent. However, part (b) of Figure 3.6 reveals the maximum resultant throughput can be slightly higher. Recall that pi is the transmission probability of class 1 users and this value, shown in both figures, is quite low; this may result in an undesirable large delay. If pi (or p^j is raised to a higher value, such as shown in Figure 3.8, both Sx and S2 degrade to zero. It is observed from other results that the throughput of either class, in general, does not degrade to zero ultimately if both persistent factors are small. In summary, if the protocol proposed in [17] is applied to an environment, in which there are finite number of users and no collision detection is involved, one has to take the stability problem into account when evaluating the throughput performance. 3.6 Delay Characteristics T h e complexity of the delay-throughput analysis involved in the p-persistent C S M A with priority can be seen in [8]. It was mentioned there that the procedure is computationally expensive to be of practical use if the system is large (e.g., if the number of users is greater than 5). In view of this, one would like to rely on simulation to study the trade-off of throughput and delay. Furthermore, the primary interest here is to examine the delay performance of p-persistent with two persistent factors, therefore a simulator assuming there are infinitely many users was written. T h e main reason for this assumption is that it usually requires less time to acquire equilibrium results. This simulator, like the one in chapter 2, is tested by comparing the (Si,Gi) and (S^G^) relationships obtained from simulation with CHAPTER 3. S. 00 3 o I 00 g 1.0 0.9 f-0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 1x10 1.0 0.9 0.8 0.7 0.6 0.5 0.4 -0.3 -0.2 -0.1 0.0 1x10 a = 0.5 M1 = M2 = 5 p1 = 0.8 1x10' IxlO1 Channel Traffic: G (a) Si and S2 versus G a = 0.5 M,-M2 = 5 p, = 0.6 1x10 1x101 Channel Traffic: G 1x10' 1x10' (a) Si and 52 versus G Figure 3.8: Throughput characteristics of Si and S2, finite users (a = 0.5) CHAPTER 3. 38 those obtained from analytic expressions of Si and S2. Figures 3.9 and 3.10 show the normalized mean packet delay versus Si and S2 without collision detection and with collision detection, respectively. In both figures, the packet delay of class 1 and class 2 are obtained by fixing the input rate of the other class at 0.1, 0.2, and 0.3. A lso, to study the sensitivity of packet delays to pi and P2, two pairs of persistent factors are compared. They are (0.8 ,0.03) and (0.7,0.1), where (* , *) represents (p x , P 2 ) . One can clearly observe from part(b) of both figures that packet delay for class 2 with persistent factor 0.03 is larger than that with 0.1. But there is hardly any difference in class 1 packet delays in part(a) of Figure 3.9, which is reasonable as 0.8 and 0.7 are very close. However, when there is collision detection, Si can be increased further and the effects coming from class 2 become more conspicuous. A s shown in part (a) of Figure 3.10, the class 1 packet delay with persistent factor 0.8 is lower than that with 0.7, when the system approaches saturation. Th is is mainly because class 2 users with persistent factor 0.03 have little influence on class 1 users who are using 0.8. 3.7 Summary A throughput-delay performance of p-persistent C S M A (without collision detection) with two persistency factors is examined. T h e protocol was originally proposed in [17]; however, no throughput-delay performance was available there. T h e objective of this chapter was to provide such information. T h e throughput of this protocol without collision detection is examined by deriving the throughput expression for each class. Both cases of infinite users and finite users are considered. It is observed CHAPTER 3. 39 I v I O 2 S 1 x 1 ° 1 r 1x10° p1 = 0.8 Pi = 0.7 0.0 O : S 2 = 0.1 A : S 2 = 0.2 + : S 2 = 0.3 _. I i I i I u 0.4 0.5 0.6 Throughput, S, 0.7 0.8 0.9 1.0 (a) Packet delay versus throughput, class 1 1x1 o2 (b) Packet delay versus throughput, class 2 Figure 3.9: Simulated delay-throughput performance (without collision detection) CHAPTER 3. 40 1x102 •o i a S 1 x 1 ° 1 r 1x10° 1x102 •o i Q ^ 1x101 1x10° p, = 0.8 Pi = 0.7 o (a) Packet delay versus throughput, class 1 (b) Packet delay versus throughput, class 2 Figure 3.10: Simulated delay-throughput performance (with collision detection) CHAPTER 3. 41 that when the number of users is infinite the maximum throughput of this protocol is upper-bounded by that of p2-persistent and lower-bounded by that of pi-persistent. T h e trade-off of throughput and delay is investigated by simulation. T h e results indicate that the class 1 packet delay is lower than that of class 2 and that the transmission probability of class 2 has a definite effect on class 1. Chapter 4 Protocol for Integrated Voice and Data Transmission in L A N There have been many studies on voice transmission or integrated voice and data transmission in the local area network ( L A N ) . Results for C S M A / C D based channels have indicated the data load would have definite effects on voice transmission [19] [20]. T o avoid the interference from data packets, protocols are specifically devised so as to achieve good performance for both voice and data packets. However, these protocols require either a complex algorithm [13] or a special packet format [12]. One would, certainly, notice the simplicity of the p-persistent C S M A / C D with two persistency factors. In addition, it has been shown in chapter 3 that its delay-throughput characteristics are well behaved, under data traffic. Th is chapter, therefore, proposes the application of this simple protocol for combined voice and data transmission. Simulation results are presented to evaluate its performance and comparisons with other works, which use C S M A / C D , are also discussed. 42 CHAPTER 4. 43 4.1 Channel Access Protocol B o t h the voice and data packets acquire the channel by following the slotted p-persistent C S M A / C D protocol. Voice packets are transmitted with probability pv and data packets with probability p<j. A s explained below, voice packets should avoid long delay and are of higher priority over data packets; therefore, p„ > p<j. In case of collisions the packets involved are retransmitted according to the binary exponential backoff algorithm. This will be elaborated below. 4.2 Characteristics and Requirements for Voice and Data Traffic T h e traffic characteristics for voice and data are very dissimilar. Al though voice traffic has been regarded as a continuous source, it has been shown [24] that a typical conversation has an on-off characteristic and can be modeled as having alternate "talkspurts" and "silences". T h e lengths of these talkspurts and silences are exponentially distributed. Also, to keep voice intelligible voice packets must be under a very stringent delay constraints. Furthermore, delay variance must be small [21]. G o o d voice quality can still be maintained if 2% of voice packets are discarded [21]. Recent results [22] also reveal that a higher percentage (about 5%) of the voice packets loss rate is acceptable without sacrificing the quality and intelligibility of speech. T h e data traffic, on the other hand, is bursty in nature. D a t a packets arrive sporadically and long delay and large variance in packet delay can be tolerated [21]. However, loss of data packets is generally considered unacceptable. CHAPTER 4. 44 4.3 Simulation Model In this simulation model there are n finite number of calls. E a c h of the n speakers alternates independently between intervals of talkspurt and silence. T h e number of calls in each simulation is maintained constant and the voice throughput, p„, can be adjusted by varying the value of n. Note that variations in the number of speakers due to calls entering and leaving the system are not included in this study because the variations of talkspurt/silence occur much more frequently than the variations of n. T h e number of data sources, on the other hand, is assumed to be infinite. Collec-tively these form an independent Poisson source. Besides, the buffer size is assumed infinitely large; therefore, no data packet is lost. T h e data throughput, pj, can be varied by changing the data packet interarrival time, which has an exponential distribution. In the simulation the durations of talkspurt and silence period are assumed to be exponentially distributed with mean Tt and r „ respectively. A s a call is active less than 50% of the time [23], two speech activity factors, 1 saf, are used; they are 42% and 45%. Note that for saf = 42% (or 45%), rt = 1.0 (or 1.34), and r, = 1.4 (or 1.67) [24] are used. It is assumed that the coding rate is 64 K b p s and that the channel transmission rate is 10 Mbps. T h e effects of voice packet length on the performance of voice transmission will be studied; the packet length chosen to be 5.75 ms, 10 ms, and 28 ms. Assuming 1 speech activity factor, saf, is given by rt/ (rt + T,). CHAPTER 4. 45 there are 208 overhead bits [20] which include 64 synchronization bits, 112 bits of header and 32 bits C R C field, the voice packet length for frame size of 5.75 ms, 10 ms, and 28 ms are 576 bits, 848 bits and 2000 bits, respectively. T o facilitate programming, the length of data packet is the same as that of voice packet. T h e time axis in the simulator is assumed slotted; the size of a slot is the prop-agation delay normalized by the packet transmission time. Due to the propagation delay, the channel remains busy for a certain time after a successful transmission. Th is time is assumed to be 1 slot. If collision occurs, 3 slots are required to change from the busy state to the idle one. Regardless of the type of packet, the packets involved in a collision are rescheduled for a later retransmission by using the same algorithm: binary exponential backoff ( B E B ) . T h e B E B algorithm is as follows. If the first attempt fails, the n th retransmission is scheduled at the point r slots from the time the channel is sensed idle again, where r is chosen at random from the set {0,1,2, • • • ,2 f c — 1} and fc=min{n,8}. If at this new scheduling time the channel is sensed busy, the same procedure is repeated but n is not incremented. Note that if a packet has more than eight collisions, the same interval will be used for selecting a retransmission time. Al though both the data and voice packets adopt same B E B algorithm for retransmission, they stay in the system in a different manner. Voice packets remain in the system until the delay exceeds the voice frame size and thus are discarded. Unlike voice packets, data packets stay in the system until they are successfully transmitted. CHAPTER 4. 46 4.4 Simulation Results A computer simulation, written in C language, was developed to study the per-formance of both voice and data transmission using p-persistent C S M A / C D with two persistency factors. Simulation length of each run is about 5 min. T h e results are divided into two parts. T h e first part examines the performance of voice trans-mission; the second part studies the effects of data load on the voice transmission. Comparisons with other works are also discussed. 4.4.1 Voice Transmission Figure 4.1 shows the voice packet loss rate (in %) as a function of number of calls for various frame sizes and values of a, the normalized propagation delay. For 5.75 ms and 10 ms frame sizes, saf is 45%, whereas saf is 42% for 28 ms frame size. Results are obtained without any data load; only voice packets are allowed in the system and they are transmitted with pv = 0.8. Recall that voice packets are discarded if they are delayed to a time longer than the frame size. A s shown in Figure 4.1, the number of calls that can be accommodated in the system increases as the voice frame size increases. Th is is understandable as there are more slots in a larger frame. If 2% is the maximum allowable loss rate, the system can support approximately 130 and 170 calls for frame size of 5.75 ms and 10 ms, respectively. T h e effects of propagation delay can also be observed. For the 28 ms frame size, if the value of a is doubled, the number of calls drops from 290 to 262, approximately. CHAPTER 4. 10.0 Figure 4.1: Voice packet loss rate versus number of calls, n CHAPTER 4. 48 4.4.2 Integrated Voice and Data Transmission T h e primary objective here is to examine the effects of data load on the performance of voice transmission. In this simulation study, the number of calls (n) is fixed (i.e. p„ is a constant) and the data throughput (pd) is varied gradually until the system approaches saturation. T h e parameters in the simulation are pv — 0.8, Pd = 0.03, saf = 42%, a = 0.0125, and voice frame size is 28 ms. A lso , the delay of voice packet (or data packet) has been normalized to the voice packet (or data packet) transmission time. T h e mean delay of voice packet and data packet versus data throughput when n = 50, 100, 150, and 200 is described in Figure 4.2. A s can be observed, the delay of voice packet is well below that of data packet. W h e n n is small, voice packet delay increases almost linearly whereas data delay increases exponentially. Note that there is a limitation on the level of data throughput for a specific n. Th is can be observed from the delay of data packets; this becomes very large at higher data throughput. T h e loss of voice packets resulted from the combined voice and data transmission is described in Figure 4.3. For n = 50, 100, 150 and various values of pd, very few voice packets are discarded. Loss of voice packets becomes appreciable when n = 200 and increase sharply as pd is further increased. Th is phenomenon, to a large extent, is attributed to the parameters pv and pd. Recall that pv = 0.8 and Pd = 0.03 are used in the simulation. A s values of Pd is very small compared with that of pv, the chance for both voice and data packets to transmit at the same slot is very rare. Th is reduces significantly the frequency of collisions between voice packet CHAPTER 4. 2x101 IxlO 1 -data packet delay voice packet delay o : pv = 0.15 (50 calls ) A : pv « 0-30 ( 100 calls ) + :p v = 0.45 (150calls) <> : pv = 0.60 (200 calls ) 1 x10° ' 1 ' ' 1 1 ' ' ' 1 ' ' ' ' 1 1 ' ' ' 1 ' ' ' ' 1 ' ' ' ' 1 1 ' ' 1 1 ' 1 ' ' 1 ' ' ' ' 1 ' 1 ' 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Data Throughput Figure 4.2: Normalized mean delay versus data throughput (/><*) CHAPTER 4. Figure 4.3: Voice packet loss rate versus data throughput (pd) CHAPTER 4. 51 and data packet. Loss of voice packets, therefore, is mainly due to collision among voice packets. When n is large, for example 200, chance of collisions increases. More collisions require more reschedules and voice packets are further delayed. As a result, voice packets which have been delayed to a time longer than the voice frame, are discarded. Figure 4.4 shows the standard deviation of delay for both voice and data pack-ets. A s shown in the figure, voice packet has a low standard deviation of delay. It increases only at large n. Data packet, however, has a large standard deviation of delay. Th is large value should not pose a problem as data packet, as mentioned above, can tolerate high delay variance. A l l these results indicate that the protocol can be used for combined voice and data transmission. Moreover, two basic re-quirements for voice traffic, namely small delay and low delay variance, have been achieved. 4.4.3 Comparisons and Discussions It would be interesting to compare these with other schemes that use C S M A / C D . One should notice that because of different values of parameters and model as-sumptions, a strict comparison is inappropriate. For example, some schemes do not discard voice packet if the delay exceeds the voice frame size. Besides, some works reported in the literature do not describe how a conversation is modeled. In the following, it is assumed that two calls represent one conversation. Some measurements of voice packet transmission in the Ethernet has been re-ported in [25]; however, the results obtained there were under different parameters. CHAPTER 4. o : pv = 0.15 (50 calls ) A : pv " 0.30 (100 calls ) + :pv = 0.45 (150calls) O : Pv = 0.60 (200 calls ) i v I O ' 1 ' ' ' * ' ' 1 ' 1 ' ' ' 1 1 ' ' ' ' 1 ' ' ' ' 1 ' ' ' ' 1 ' ' ' ' 1 ' ' ' ' 1 ' ' 1 ' 1 ' ' ' ' 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Data Throughput Figure 4.4: Standard deviation of delay versus data throughput (pd) CHAPTER 4. 53 Models and parameters that are similar to those used in this study are found in [20] and [13]. In [20], simulation results of voice transmission with frame size of 5.75ms are obtained in the Ethernet, in which 1-persistent C S M A / C D is adopted. Results from [20] show that about 50 to 55 conversations (100 to 110 calls) can be supported, assuming 2% of voice packets are discarded. It has been shown in the previous section that with 5.75ms frame size the system can support about 130 calls. This suggests an observation. Compared with the 1-persistent, 0.8-persistent C S M A / C D can resolve conflicts better. This results in higher utilization (at the expense of higher voice packet delay but not large enough to increase the voice packet loss rate). A complicated protocol 2 is proposed in [13]. W i t h 6 ms (slightly larger than 5.75 ms) frame size, that scheme can support about 75 conversations (150 calls), which is higher than 130 calls reported here. However, considering the simplicity of 0.8-persistent C S M A / C D and the complexity of the protocol in [13], it may be justified to have a lower number of calls. Since different frame sizes (5.75 ms and 6 ms) are used for combined voice and data transmission in [20] and [13], comparison with their results is not appropriate. Nevertheless, observing the throughput, mean, and standard deviation of delay of voice and data packets, in addition to the simplicity of the protocol, it is reason-able to state that p-persistent C S M A / C D , with two persistency factors, is a viable scheme for interated voice and data transmission in L A N . 2Each user, in this scheme, is equipped with two clocks. One of the clocks, if permitted to run, runs i] times faster than the other, where T) is a system parameter. Transmission of voice packets is activated by an algorithm. CHAPTER 4. 54 Obviously, the parameters p„ and pd play a crucial part in the protocol. Through-out the simulation pv = 0.8 and pd = 0.03 are used. One may, therefore, wonder if these are the optimal values. More simulations can be carried out to determine the answer but these simulations are too time consuming and expensive, especially at the transmission rate of 10 Mbps. Consequently, these attempts are avoided. 4.5 Summary In this chapter, the application of p-persistent C S M A / C D protocol in the integrated environment is proposed. T w o types of traffic are studied, namely voice and data. Priority is achieved by assigning a higher persistency factor to the voice traffic, and a lower one to the data traffic. Performance of this scheme is evaluated through simulation. Evaluations are based on the voice packet loss rate, delay, variance of delay, and number of calls that can be supported. Results indicated that both the requirements of voice traffic and data traffic are satisfied. Chapter 5 Integrated Voice and Data Transmission in Packet Radio Network A s briefly outlined in chapter 1, combining voice with data for transmission offers many advantages. One of them is cost savings. Because of these potential benefits, we propose a scheme for voice only or combined voice and data transmission which is applicable to the packet radio network. T h e proposed scheme does not require collision detection, attains an acceptable level of throughput if about 1% of voice packets are allowed to be discarded, limits the delay of voice packet to the voice frame size, and, most importantly, satisfies the stringent transmission requirements of voice packets which demand small mean and variance in delay. T h e proposed scheme is a variation of movable slot T D M [12] and p-persistent C S M A [9]. It takes advantages of the periodicity of voice packet generations while a voice user is in talkspurt, the bursty nature of data packets, and the random transmission of p-persistent C S M A . T h e performance of this scheme is evaluated by simulation. 55 CHAPTER 5. 56 5.1 Description of Protocol In this section we first describe the access protocol for voice and data users and then explain the characteristics of the protocol. T h e operation of the protocol is assumed slotted. Th is requires all users to start transmissions only at the beginning of a slot. T h e size of a slot is equal to the end-to-end propagation delay. Dur ing a talkspurt, voice packets are generated at a fixed period of time (T). For discussion purposes, we call the first packet of a talkspurt the "first voice packet" and subsequent packets other than the first one the "subsequent voice packet". We use the term "voice packet" to refer both of them. We assume that voice packet size is fixed and that data packet size is not greater than that of a voice packet. If a voice user has a ready first voice packet it proceeds as follows. 1. If the channel is sensed idle it transmits the packet with probability pv or defers the transmission to the next slot with probability (1 - p„) . If this first voice packet is successfully transmitted at time t, the subsequent voice packet is scheduled for transmission at time t + T. However, if this first voice packet involves in a collision, or its delay exceeds the packet generation time (T), it is discarded. T h e subsequent voice packet immediately following the discarded one is regarded as the first voice packet and the algorithm is repeated. 2. If the channel is sensed busy it keeps on sensing the channel until the channel is idle. A t this idle time, it further defers the transmission for one slot and senses the channel again. Dependent upon the state of the channel, step (l) or (2) is repeated. CHAPTER 5. 57 T h e following steps are carried out for the subsequent voice packets. 1. If the channel is sensed idle, the subsequent voice packet is transmitted. 2. If the channel is sensed busy, the subsequent voice packet is transmitted as soon as the channel is idle; that is, the transmission time of this subsequent voice packet has been shifted to a new time (t f). If it suffers a collision, it is discarded. Regardless of the outcome of this packet, the next subsequent voice packet is scheduled at time t' + T. A ready data user operates as follows after the channel is sensed. 1. If the channel is idle, it transmits the data packet with probability p<j, or with probability (1 - p<i), the user defers the transmission to the next slot. 2. If the channel is busy, it senses the channel one slot after the channel becomes idle and repeats the algorithm. 3. If the data packet suffers a collision, it is retransmitted according to a retrans-mission algorithm (discussed in section 3). We should note that the above access protocols for first voice packet and data packet are the variations of p-persistent C S M A and subsequent voice packet follows the 1-persistent C S M A . T h e idea of shifting transmission time of subsequent voice packets is adopted from [12]. In the following we describe some characteristics of the proposed scheme. Due to the shifting of transmission time of subsequent voice packets, they would not CHAPTER 5. 58 collide with one another. A lso, the maximum time-shift is one packet transmission time because at the new transmission time the subsequent voice packet is either successfully transmitted or collided with other voice packets. T h e transmission of the first voice packet, as we propose in the scheme, is probabilistic; it may still collide with other voice packets if they transmit at the same slot with a first voice packet. Should collisions occur, voice packets continue colliding on successive transmissions as they are generated periodically. However, the proposed scheme stipulates that if the first voice packet is discarded, the next subsequent voice packet is treated as a first voice packet, acquiring the channel with probability pv. Th is strategy, therefore, reduces the chance of collision significantly. A s will be seen from the simulation results, the performance of this protocol is dependent on a judicious choice of pv. W i t h the proposed scheme, a considerable number of voice packets that are discarded are at the front parts of the talkspurts, which may cause speech clipping. However, it has been observed from a listening test that this clipping does not degrade the voice quality if the percentage of voice packets that are discarded is small [15]. Note that when a subsequent voice packet is waiting for transmission, voice bits are arriving which cannot be placed in the speech bits of voice packet. Therefore, an overflow area is required for carrying these extra bits. A s a result, the format of voice packet is composed of header bits, speech bits, and overflow bits, whereas the data packet consists of only header bits and data bits. When data traffic is combined with voice, the former would inevitably affect the latter. T o mitigate this interference, we require Pd < Pv CHAPTER 5. 59 5.2 Simulation Model T h e approaches of simulation are very similar to those used in chapter 4. In the simulation model, there are n finite number of calls. E a c h of the n speakers alter-nates independently between intervals of talkspurt and silence. T h e number of calls, in each simulation, is maintained constant and the voice throughput, pv, can be ad-justed by varying the value of n. T h e number of data sources, on the other hand, is assumed to be infinite. These collectively form an independent Poisson source. Besides, the buffer size is assumed infinitely large; therefore, no data packet is lost. T h e data throughput, pa, can be varied by changing the data packet interarrival time, which has an exponential distribution. T h e durations of talkspurt and silence period are assumed to be exponentially distributed with mean 1.0 sec and 1.4 sec, respectively. Thus , the speech activity factor is 42% [23]. If the first voice packet is delayed, to a time longer than the voice frame size, it is assumed lost. We first study the performance of voice transmission under the proposed scheme. In particular, we are interested in the voice packet loss rate versus the voice through-put with different p„'s. T h e n we examine the effects of data load on the voice traffic. Since the channel rate, voice packet length, coding rate, and pv are important to the system performance, several values are used in this simulation study. We use the notation (tr, vf, cr, pr) to describe them, where tr, vf, cr, and pr represent channel transmission rate in K b p s , voice frame size in msec, coding rate in Kbps , and pv, respectively. CHAPTER 5. 60 In the simulation of voice transmission we use (10 3, 28, 32, pv): the first set, (32, 100, 2.4, pv): the second set, and (16, 20, 2.4, pv): the third set, where pv is varied to optimize the performance. We assume there are 172 header bits for the first set and 20 header bits for the second and third sets. Recall that a voice packet comprises header bits, speech bits, and overflow bits. T o compute the required overflow bits in each set of parameters, we solve the following equation ^(H + S + 0) = 0 (5.1) where cr is the coding rate, tr the channel transmission rate, H the header bits, S the speech bits, and O the overflow bits. Therefore, O equals 36, 22, and 12 bits for the first, second, and third set, respectively. Also, the normalized propagation delay (a) used in the simulation is 0.0125 for the first set and 0.01 is used for the remaining sets. If c denotes the number of slots available in a frame, then c=25.0 slots for 28 msec, 11.2 slots for 100 msec, and 4.0 slots for 20 msec. T o evaluate the effects of data load on voice traffic, we choose the second set of parameters, and in the simulation we assume the size of data packet is the same as that of voice packet. If a data packet becomes involved in a collision, it is retransmitted as follows. If the first attempt fails, the n th retransmission is scheduled at the point r slots from the time the channel is sensed idle again, where r is chosen at random from the set {0, J , 2 J , • • • , (2* — 1) J } , where fc=min{n,16}, and J is a constant which increases with the data throughput to ensure stable operation of the system. A l l of the afore-mentioned parameters do not correspond to a particular system; nonetheless, they provide an indication of how the proposed scheme operates. CHAPTER 5. 61 5.3 Simulation Results and Discussions T h e results presented here are obtained from a simulator written in C language. T h e value of each point in our figures is the average of three simulation runs. Each run requires at least 100,000 voice packets and data packets in the system. 5.3.1 Voice Transmission We first examine the performance of voice transmission. T h e parameters chosen are (10 3, 28, 32, p„) . Figure 5.1 not only shows the voice packet loss rate (in %) versus voice throughput (p„) , it also reveals the important role played by pv. As can be observed from the figure, when pv is small the loss rates for various values of pv are no different from others. A s pv increases from 0.3 to 0.75, loss rate for p „ = 0 . 3 is the highest, while the loss rates for other p„'s are still very close. However, as pv is further increased from 0.75 to 0.87, p „ = 0 . 0 4 has the lowest discard rate, followed by p „ = 0 . 1 and p „ = 0 . 0 1 . Such phenomenon is no surprise to us. Recall that c=25.0 in 28 msec frame size; therefore, the number of calls supported in the system is large. W i t h n becoming larger, p „ = 0 . 3 is obviously not able to resolve conflicts, resulting in many collisions which lead to higher loss rate. In the medium load (0.3 to 0.75), other p„'s perform equally well. However, as pv is in the high range (0.75 to 0.87), p „ = 0 . 0 1 has a higher discard rate than that of p „ = 0 . 1 , which is quite contrary to one's expectation because the lower the values of p„, the better the chance of resolving conflicts. But notice that at such a high voice throughput, the voice packets are already very tight-packed, with very few idle slots left in between. If the first voice packet is transmitted with low probability, such as 0.01, it misses CHAPTER 5. 10.0 I ( 103, 28, 32, pv) 9.0 - A : Pv = 0.3 o : Pv - 0.1 + : Pv = 0.04 8.0 -V : Pv = 0.01 7.0 h Voice Throughput Figure 5.1: Voice packet loss rate versus voice throughput for different p„'s CHAPTER 5. 63 these slots very easily. Once the chance for transmission is not available, it has to wait for an idle channel, which incurs more delay. Consequently, voice packets are discarded as delay is larger than the frame size. We should note that if 1% is the maximum allowable loss rate, the achievable voice throughput is over 80% with T o better evaluate the performance of the proposed scheme, we assume all n voice users operate in T A S I mode [23], which is an ideal scheduling, and compare the loss rate obtained from simulation with the analytic results of freeze-out fraction [26], which is given by where 7Y is the number of voice paths, p is the speech activity factor, and c' the number of channels. Figure 5.2 depicts the comparisons, together with two more sets of parameters. Since T A S I represents an ideal scheduling, the loss rate of T A S I provides a lower bound. We can see from the figure that as pv is small, the simulated results are quite close to those of T A S I . However, as it increases, simulated results deviate further from ideal ones. T o keep voice intelligible, loss rate must be low, therefore our scheme performs quite efficiently. One point is worth mentioning. W h e n the number of slots is small, for example c=4, the voice throughput is less than 60% at the 1% loss rate. 5.3.2 Integrated Voice and Data Transmission p „ = 0 . 0 4 . (5.2) T h e primary objective here is to examine the effects of data load on the performance of voice traffic by utilizing the proposed scheme. In this simulation study the number Figure 5.2: Comparison of voice packet loss rate of T A S I with that of the proposed scheme CHAPTER 5. 65 of calls (n) is fixed (i.e., pv is a constant) and the data throughput (pd) is increased gradually. Throughout this subsection, the parameters used are (32, 100, 2.4, 0.1). We used pd=0.01 to reduce the interference from data load. Figure 5.3 shows the voice packet loss rate versus data throughput (pd) for a fixed number of calls. Recall from Figure 5.2 that when there is no data load, the pv is about 70% at the 1% limit. T h e system throughput (pv + pd) in Figure 5.3 is no less than 70% at the same loss rate. Hence, the data packets, under these parameters, can make efficient use of silence intervals of voice calls by accessing the channel with 0.01-persistent C S M A . We next investigate the mean delay and standard deviation of delay for first voice packets, subsequent voice packets, and data packets. T h e delay of a packet is defined as the difference between the packet arrival time and the time that the packet is successfully transmitted. T o facilitate presentation of the results, the delay has been normalized to the packet transmission time. We can see from Figure 5.4 that both the first voice packet delay and its standard deviation increase almost linearly when Pd is small and increase rapidly as pd becomes larger. T h e reason is that when there are more data packets in the system the first voice packets have to wait longer for an idle channel before they can be transmitted. T h e delay of subsequent voice packets and its standard deviation are in sharp contrast with those of the first voice packets. Figure 5.5 shows the delay increases linearly, while the standard deviation decreases at high data throughput. Also, the waiting time (i.e., the amount of time that has been shifted) is well below 1.0, which is reasonable because the maximum amount of time that can be shifted is one packet transmission time. Due to this upper limit, the standard deviation drops slightly at high channel utilization. As to Figure 5.3: Voice packet loss rate versus data throughput for different pv's CHAPTER 5. 4.0 3.5 3.0 2.5 2.0 1.5 0 2.5 2.0 1.5 -1.0 0.5 1 1 1 1 1 1 • • • • 1 • • 1 1 1 0.1 0.2 0.3 0.4 0.5 0.6 Data Throughput o : pv = 0.224 (6 calls ) A :p v = 0.372 (10 calls) + :p v = 0.520 (14calls) V :p v = 0.664 (18 calls) 0.7 0.8 0.9 1.0 (a) Mean first voice packet delay O : pv = 0.224 (6 calls) A :p v = 0.372 (10 calls) + :p v = 0.520 (14calls) V :p v = 0.664 (18 calls) 1 • • • • 1 1 • • 1 1 1 , 1 1 • i i • • • i • • • .0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Data Throughput (b) Standard deviation of first voice packet delay Figure 5.4: Mean and standard deviation of first voice packet delay CHAPTER 5. 1.5 1.4 1.3 1.2 1.1 0 0.5 0.4 0.3 0.2 0.1 '•o 1 • 1 • 1 • • 1 • • • • 1 1 • • • 1 o : pv = 0.224 (6 calls) A :p v = 0.372 (10calls) + :p v = 0.520 (14calls) V :p v = 0.664 (18calls) i i i i • i i t . . . i i • i i i i i i .0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Data Throughput (a) Mean subsequent voice packet delay o : pv - 0.224 (6 calls ) A :p v = 0.372 (10 calls) + :p v = 0.520 (14calls) V :p v = 0.664 (18 calls) ' •• l t i i i I i i i i I 1 1 1 • 1 1 • 1 ' ' 1 ' • t I i i i i 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Data Throughput (b) Standard deviation of subsequent voice packet delay Figure 5.5: Mean and standard deviation of subsequent voice packet delay CHAPTER 5. 69 data packets, both delay and standard deviation increase exponentially, which can be observed from Figure 5.6. 5.1 Comparisons with Reservation Aloha T h e scheme, as proposed in this chapter, requires carrier sensing. Under bursty data traffic, it is a well-known fact that C S M A can attain a higher throughput than the slotted A loha one. Nevertheless, it has been shown [27] that a variant of slotted A loha , called reservation A loha (r-aloha) scheme, is able to achieve a throughput beyond the 1/c limit if there is more than one packet in a message. Recall that voice packets arrive periodically while in talkspurts. In general there are many packets in a talkspurt. One may, therefore, surmise that the r-aloha scheme may be suitable for packet voice transmission or even combined voice and data transmission. A comparison of the performance of this alternative for integrated services with that of the proposed scheme is the prime objective of this section. T h e following describes the protocols for voice and data users. In r-aloha, the channel time is assumed slotted and slots are organized into frames with c slots per frame. If voice user t has a successful transmission in slot k (acquiring the channel with slotted A loha mode), this slot k in the next frame is reserved for user % as long as it still has packets to transmit. If it transmits its last packet, the slot k in the next frame is available to other users. If the first attempt of user i fails, the packet is retransmitted in the next unreserved slot with probability prt. Data users transmit their packets with slotted Aloha mode and data message is assumed to be composed of one packet. Retransmission algorithm of data packets is the same as CHAPTER 5. 4x101 & 1v101 I u <a a, 3 Q 1x10° o :p v = 0.224 (6 calls) jy,— A :p v = 0.372 (10 calls) • + :p v = 0.520 (14calls) V i i i • 1 . i • . I i i • :p v = 0.664 (18 calls) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Data Throughput 0.7 0.8 0.9 1.0 (a) Mean data packet delay 1x102 2 I a, a to Q •g 1X101 -3 > 1 C/5 1x10° 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Data Throughput (b) Standard deviation of data packet delay Figure 5.6: Mean and standard deviation of data packet delay CHAPTER 5. 71 that used in the proposed scheme. Following the previous notations, parameters chosen for the simulation are: tr=720 K b p s , v / = 1 6 msec, cr=32 Kbps , and p „ = p r t = 0 . 1 . Assuming there are 64 header bits, the overflow bits required for the proposed scheme is 27. Hence the number of slots available per frame is 19 and 20 for the proposed and the r-aloha schemes, respectively. In the simulation, the normalized propagation delay is 0.01 and speech activity factor of 42% is used. Also, voice packets are discarded if delay exceeds 32 msec. W h e n both schemes are applied to voice transmission, the voice packet loss rate versus the number of calls is shown in Figure 5.7, together with the freeze-out fraction of T A S I (equation 5.2). Figure 5.7 indicates the voice packets loss rates for both schemes are very similar. In comparing the results, one should observe the number of slots available per frame for each scheme. Due to the overflow bits, the proposed scheme has 19 slots per frame, which is one less than that in the r-aloha. Al though voice packet loss rates for both schemes are almost the same, the way that voice packets are discarded is quite different. F r o m the description of the r-aloha scheme, one notices that voice packets are lost only at the beginning of talkspurts. However, due to the probabilistic transmission of first voice packet in the proposed scheme, the first voice packet may collide with either the first voice packets or the subsequent voice packets (both from other users). Therefore, loss of voice packets may occur at the beginning of talkspurts as well as other parts of talkspurts. T o acquire a deeper understanding of how the voice packets are lost in the proposed scheme, voice packets are split into two components: first voice packet CHAPTER 5. Figure 5.7: Voice packet loss rate versus number of calls without data load CHAPTER 5. 73 and subsequent voice packet. Their loss rates are plotted in Figure 5.8. A s can be observed, loss of subsequent voice packets is insignificant compared to that of first voice packets. In other words, the chance of collision between these two types is quite rare and most of the loss occurs at the front parts of talkspurts. If a large number of voice packets are discarded at the front parts of talkspurts, speech may be severely clipped and voice quality degraded. It is interesting to note that packet voice transmission using the r-aloha has been examined by simulation recently [15]. They performed a listening test and concluded that at the 1% loss rate voice quality is still good and intelligible. Hence, it is very useful to compare the patterns of lost voice packets in both schemes. Figures 5.9 to 5.12 show the frequency (that is the number of times it occurs) versus the number of consecutive packets that are discarded. T h e frequency values are represented by thick black vertical lines. T h e symbol circle in these figures represents the number as located occurs only once. (Other thinner vertical lines are for the purpose of marking the x-axis only.) A lso depicted in these figures are the number of calls (n), packets that are lost or successfully transmitted, and lost rate in a simulation run. Comparison of the lost patterns in both schemes can be done by fixing the value n. We can observe from these figures that lost patterns are very similar. When n = 30, the loss of consecutive packets is small. A s n increases to 37 (the loss rate is about 1%), the shapes in Figures 5.11 and 5.12 are quite spread out, and consecutive lost packets can go beyond 30, however, this frequency is small. Since a listening test has been carried out in [15] and the lost patterns in both schemes are similar, it is reasonable to say that the criterion of 1% loss rate in the proposed scheme is acceptable. CHAPTER 5. o . o * 1.0 2.0 h 3.0 4.0 5.0 h 6.0 7.0 8.0 9.0 voice packet first voice packet subsequent voice packet 10.0 1—1—1—1—1—1—1—1—1 20.0 - I I I I I I I I L _ 25.0 30.0 35.0 40.0 Number of Calls, n 45.0 50.0 Figure 5.8: Components of voice packet loss rate versus number of calls CHAPTER 5. 75 6x102 proposed scheme : n = 30 loss of voice packets = 537 successful voice packets = 477265 loss rate = 0.112 % 1X102 ->> o c 3 1x101 1x10° 0.0 5.0 1 I OO—1—9-l a o 25^ .0 _L 15.0 20.0 T 30.0 35.0 40.0 Loss of Consecutive Voice Packets _L 45.0 50.0 Figure 5.9: Frequency versus loss of consecutive packets in the proposed scheme, n = 30 CHAPTER 5. 76 1x10J 1x102 o c u 3 1x101 r-aloha scheme : n = 30 loss of voice packets = 927 successful voice packets = 1417901 loss rate = 0.065 % 1x10° M I 11 1 M I 11 I oo 6 I—L_L©e—A_ 0.0 5X)10.0 1?.0 20.5 2S.0 30.0 35.0 40.0 45.0 50.0 55.0 Loss of Consecutive Voice Packets Figure 5.10: Frequency versus loss of consecutive packets in the r-aloha scheme, n = 30 CHAPTER 5. 77 6x102 proposed scheme : n = 37 loss of voice packets = 5717 successful voice packets = 575056 loss rate = 0.984 % 1x102 o c 3 I 1x101 -1x10° 0.0 5.0 _L 10.0 15.0 20.0 25.0 30.0 Loss of Consecutive Voice Packets 40.0 45.0 50.0 Figure 5.11: Frequency versus loss of consecutive packets in the proposed scheme, n = 37 CHAPTER 5. 78 1x103 r-aloha scheme : n = 37 loss of voice packets = 14821 successful voice packets = 1725678 loss rate = 0.852 % IxlO2 -o c I 1x101 1x10° 0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.8 Adt 45.0 50.0 '55.0 Loss of Consecutive Voice Packets Figure 5.12: Frequency versus loss of consecutive packets in the r-aloha scheme, n = 37 CHAPTER 5. 79 T h e performances of both schemes vary significantly when data load is mixed with voice traffic. T h e normalized delay of voice and data packets versus data throughput (pd) are shown in Figures 5.13 and 5.14. Note that "voice packets" implies both first voice and subsequent voice packets and that number of calls, as shown, are fixed at 10 and 20 calls. Observations from both figures show the mean voice packet delay of the proposed scheme is maintained within two packets transmission time. O n the other hand, delay of voice packet in the r-aloha increases rapidly. T h e data packet delay of the proposed scheme is also well below that of the r-aloha one. T h e dramatic increase in delay in the r-aloha scheme is mainly due to the influence from data load and also the protocols adopted by both voice and data users. Recall that data users simply use the slotted Aloha mode and, with data traffic only, the upper bound of pd is l/e(=* 0.36). Since each voice user reserves a slot after a successful transmission, these reserved slots are off limits to data users who have packets to send. A s soon as an unreserved slot is available, data packets which have waited for transmission and those new voice and data packets would all transmit into this slot. Consequently, collisions occur which result in a long delay for both types of packets. Another disadvantage of r-aloha scheme is the small value of pd. If there are even 10 calls in the system, data packets cannot take advantage of the remaining 10 slots available in every frame. In the proposed scheme, however, Pd can go as high as 0.5 without saturating the system. Al though the proposed scheme can achieve a higher system throughput, it does have disadvantages. The simulation model assumes each terminal is in line of sight with others; that is, no hidden terminals problem exits in the system. T h e performance of the proposed scheme will be certainly deteriorated if this assumption is relaxed. CHAPTER 5. 10.0, 9.0 h — r-aloha scheme - - proposed scheme o : 10 calls A : 20 calls 8.0 h 7.0 6.0 / / 5.0 4.0 / 3.0 2.0 h • • • • JLL 3 - O ' 1 ' 1 .0 0.1 0.2 0.3 0.4 0.5 0.6 Data Throughput Figure 5.13: Normalized voice packet delay versus data throughput (pd) CHAPTER 5. 7x102 1x10Z IxlO1 1x10° ? i / 1 / 6 L i 1 i ! 1 4 I 1 1 1 1 ' ' I I I 0.0 0.1 _L — r-aloha scheme - - proposed scheme o : 10 calls A : 20 calls J—i i i I i i i i 0.2 0.3 0.4 Data Throughput 0.5 0.6 Figure 5.14: Normalized data packet delay versus data throughput (pd) CHAPTER 5. 82 5.5 Summary A multiaccess scheme for packetized voice and data transmission in a packet radio network is proposed. T h e performance of the proposed scheme is evaluated by simulation. In this simulation study, three different sets of parameters are selected for voice transmission, whereas a particular set of parameters is used for integrated voice and data transmission. A t the 1% voice packet loss rate, results reveal that for voice transmission, under some system parameters, the channel throughput (pv) is above 80% and that for combined voice and data transmission, under the chosen system parameters, the channel throughput (pv + p^) is higher than the case of voice only transmission. Another option for integration of voice and data traffic is also considered, namely, the r-aloha. Owing to the limited throughput of the slotted A loha , the system throughput of the r-aloha is small and delays for both voice and data packets are large, compared to those of the proposed scheme. Chapter 6 Conclusion T h e research results of the thesis are composed of two areas. T h e first part is concerned with the throughput and delay characteristics of some combinations of C S M A protocols, namely non-persistent with 1-persistent and p-persistent with various probabilities of transmission. Protocols suitable for integrated voice and data transmission in local area network and packet radio network are proposed in the second part. In chapter 2, the delay-throughput performance of a system using 1-persistent and non-persistent C S M A protocols is examined. We derive the throughput expres-sions of each protocol by using a simple renewal theory argument. Observing the simulated delay characteristics, we find the resultant throughput cannot be added arbitrarily. However, due to the inherent properties of these two protocols, it is noticed from the results that a non-preemptive priority scheme can be implemented by assigning 1-persistent C S M A to the higher priority and non-persistent C S M A to the lower priority. 83 CHAPTER 6. 84 A more general case of C S M A scheme, p-persistent, is considered in chapter 3. T h e protocol of p-persistent C S M A / C D with various transmission probabilities used as a priority resolution scheme is proposed in [17]. Since no delay-throughput performance is given in [17], we extend the analysis of chapter 2 to this protocol, and derive the throughput expressions, and simulate the delay characteristics. T o limit mathematical complexity, only two transmission probabilities (px and p 2) are considered in the throughput analysis but the approach can be extended to the case of various transmission probabilities. T h e analysis includes both cases of infinite and finite users. T h e resultant throughput is found to be bounded by the throughput of pi-persistent and that of p2-persistent C S M A protocols. It is also observed from the delay characteristics that a good priority scheme is dependent on a judicious choice of p x and P2. Intrigued by the flexibility and simplicity of the protocol examined in chapter 3, we apply this p-persistent C S M A / C D protocols with two persistency factors to the integrated voice and data transmission in a local area network ( L A N ) . In the proto-col, we stipulate the transmission probability of voice packets be greater than that used by data packets and voice packets are dropped if delay exceeds the voice frame size. T h e performance measures are the voice packet loss rate, mean, and variance of data and voice packets, as well as the system throughput. T h e simulated results suggest that this protocol is a viable scheme for integrated services applications because both the traffic requirements of voice and data are satisfied. In addition, compared to other proposed protocols published in the literature, it is less struc-tured in the sense that users do not need to follow a complex algorithm and thus the scheme can be implemented easily. CHAPTER 6. 85 Although there are plenty of protocols proposed for integrated services in L A N , there are very few suggested for the packet radio network. In view of this, we propose a protocol which can be applied to a packet radio network. T h e proposed protocol actually resulted from the one considered in chapter 4 and other published works. However, it turns out to perform well. Comparison of the voice packet loss rate of the proposed scheme with that of the ideal case suggests this observation. Besides, it also possesses two interesting properties. Small mean and variance of voice packet delay is one of these; the other is that no collision detection is required. Appendix A This appendix outlines the deriviations of the throughput expressions of p-persistent C S M A ( without collision detection ) with two persistent factors, assuming finite number of users. T h e number of users are assumed to be divided into two classes, class 1 and class 2. Also, the number of users in class 1 and class 2 are Mi and M2, respectively. T h e throughput analysis of p-persistent C S M A with one persistent factor is available in [18]. T h e approaches shown in [18] can be easily extended to the case of two persistent factors, mainly because of the independent arrival processes assumption. In fact, many of the intermediate results, during the derivations, can be obtained by simply adding similar terms, and distinguishing them with subscripts 1 and 2. Hence, some of the intermediate steps will not be shown here. Because many notations used here have the same definitions as those in the case of infinite users, only those that are new will be defined. T h e arrival process, in this case, is geometrically distributed. Packets in class 1 and 2 arrive in each slot with probability gi and g2, respectively. A s in the case of infinite users, channel state is divided into busy period (B) and idle period (/). 86 APPENDIX A. 87 However, each busy period, in this case, is further divided into J sub-busy periods and each sub-busy period consists of a transmission delay and a transmission time (also see Figure A . l ) . The following describes how S i is obtained. F r o m Figure A . l , one can conclude that BU) = R0) + i + a j = 1,2, •• • (A. l ) and, B = J2BU) (A.2) i f u[j) denotes the successful transmission time of class 1 packets in the jth sub-busy period, then tfi = X>i(i) (A.3) Note that J is geometrically distributed with parameter (1 _ ft)(l+(l/.))Wx ( 1 _ &2)(l+(l/a))M2 ( A 4 ) and that and u['\ j = 2,3, • • • , J are identically and independently distributed. For j = 1, they are also independent from others but have different distributions. Since J is independent of B^ and , then and, B = E[BW] + (J-I)E[BW] TT^Elul^ + iJ-VE [Ui2)] (A.5) (A.6) APPENDIX A. 88 -*1 1 a ' • 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 — i — i — i — i < * f — rw —* < £ M * a : normalized end-to-end propagation delay : retransmission delay of the j th sub-busy period : transmission period of the j th sub-busy period : the j th sub-busy period Figure A . l : Components in the j th sub-busy period APPENDIX A. 89 where, J = (1 _ ffl)(l+(l/a))Jk/, (! _ y 2 ) ( l+( l /«) )Aft ( A , T ) A s in the infinite users case, J is geometrically distributed with mean 1 = 1 - (1 - ft)* (1 - 92)M> (A'8) Since Sx is given by U\I(B + 7), the task now is to find Ui and B. If 7r„in2(x) denotes the probability that there are ni and n 2 arrivals out of M i and M 2 users in x slots, then 7 r ' U , n a ( * ) = I" ( 1 - 0 1 ) ^ ( 1 - 0 2 ) ^ ' • j [1 - (1 - 0O1 n i • (1 " <7 i ) l ( M l " n i ) . ( ^ ) [l - (1 - ff2)']na • (1 -n i = 0 , l , - " , M i ; n 2 = 0 , 1 , - - - , M 2 ; n i + n 2 > 0 (A .9 ) Now, let NQ^ be the number of class 1 and class 2 arrivals at the starting time of Therefore, Pr[JV,j = (n 1 ,n 2 ) ] = { 7 r n i - n a ( l ) n 2 ( l + ( l /a)) J = l y = 2,3, (A.IO) APPENDIX A. 90 Following the approach in [3], and after some algebraic manipulations, one can write the following. E[R^\N^ = (nun2)] = a ^ l - p j k=l (1-P2)kn> Pi ( l - 9i)k ~ 9i(l ~ Pl)k]Ml-ni ( A . l l ) Pi - 9i p 2 ( l - <72)fc - 92(1 - p2)k]M2~n2 P2 ~ 92 E[ul»\NW = (nun2j\ = n i P i E ( l - P i ) ( t + 1 ) n i _ 1 ( l - P 2 ) ( f c + 1 ) n 2 k=0 'pi(l-9i)k+1-9i(l-Pi)k+1"Ml~ni Pi - 9i P2(l ~ 92)k+1 ~ 92(1 - P2)k+1] P2 - 92 A^ 2— + ( M x - ni)glPl E(l - P i ) ( t + 1 ) n i (1 " p 2 ) ( t + 1 ) n 2 Pi ~ 9i \k+l „ (i _ \ fc+11 M l ~ n i _ 1 P i U - f f i r ^ - ^ c i - P i r Pi - 9i p 2 ( l - 92)k+1 ~ 92(1 - p2)k^ P2 - 92 (A.12) Removing the conditions on ni and n 2 , equation ( A . l l ) then becomes E lR{})] - { r ( l + l ( l / a ) ) j = 2,3, . (A.13) APPENDIX A. 91 where, r(x) = 1 - (1 - ( / i) M '*( l - g2)M>* •£,l((i-pi) k-(i-9irpi k=l (1 - Pi)* - (1 - 0i) kl {1-P2)k-{1-92)'P2 Pi-9i a - p 2 y - a - 92) k]\ Ma Mi P2 - 92 - ( i - 9lyM> • ( i - 92) tMi and equations (A.12) can be written as E where, u(x) = P i ( l - 0 i ) * - 0 i ( l - P i V Pi - 0i P2(l ~ 02)* ~ 02(1 -P2) k P2 - 02 Mi (A.14) r r 7(i)i _ J u i ( i ) i = 1 lUl J - \ u x ( l + l ( l / a ) ) J = 2,3, (A.15) P 1 M 1 1 - (1 - 0 i ) l M l (1 - 0 2 ) ^ (1 - PL)FC - (1 - 00 s PI(I-PI)*-0I(I-0I)A PI - 01 (1 - P i ) f c + 1 -(1 -01)* -PI (1-P2) T + 1-(1-0 2) X -P» • 0 X ^ ( 1 - 0 0 ^ ( 1 - 0 2 ) ^ 1- (1-01)^(1-02)^ ( l - p x ) * * 1 - ^ - ^ ) * - " -Pi - 01 ( l - P 2 ) t + 1 - ( l - 0 2 ) t + 1 ' P2 - 02 Mi-1 M2 k=l I Pi ~ 01 P l ( l - 0 l ) t + 1 - 0 l ( l - P l ) f c + 1 l Pi ~ 01 P 2 ( l - 02)* + 1 - 0 2 ( l - P 2 ) t + 1 Af ,-1 M2 APPENDIX A. 92 Using equations (A . l ) , (A.5), (A.7), (A.8), (A.13) and (A.14), one then get B + I = a (1 _ ^(l+tl/aJJM, (! _ fl2)(l+(l/a))Jlfa - ± l ( ( l - P l ) k - ( l - g 1 ) ^ 1 ^ p 1 ( i - P l ) k - ( i - giy Pi - 9i Mi + P.) ' - (1 " ft)™1/-)) p, 1 + a (1 _ ^ ( l+U /a ) )^ (! _ ff2)(l+(l/o))M2 ( l - p 2 ) f c - ( l - e 7 2 ) ' P2 - 92 M2 ' (A.17) Substituting equations (A.7) and (A.15) into (A.6), the mean successful trans-misssion time of class 1 is given by <7i = PiMi  1 _ (1 _ ^(l+Cl/aJjMx (! _ ff2)(l+(l/a))M3 (1 _ p l } * - (1 - fc)(!+(!/.)) . ft(l-Pl)*-ft(l-ft)* fc=0 Pi - 01 (1 - Pl)»+i - (! - 9l)WM) . p i ( l - p O ^ - U - f f x ) ^1 T M, - l Pi -01 ( i - P2)T+1 - ( i - y 2 ) ( 1 + ( 1 / o ) ) • P. • ( 1" FT)FC+1 ~ ( 1" G 2 ) p 2 - 02 (A.18) T h e expressions of Si is obtained by putting equations (A.17) and (A.18) into Ui/(B + 7), which is shown as follows. 5, = 1 + a + o £ DMl • £Ms fc=i (A.19) APPENDIX A. 93 where, A = ( l - p i ) * - ( l _ f f l ) l + ( V a ) P i ( l - P i ) * - 0 i ( l - 0 i ) ' Pi -01 (A.20) B = ( l - p i ) t + 1 - ( l - 0 i ) 1 + ( 1 / o ) P i ( l - P i ) *+ 1 - ( l - 0 i ) * + 1 ' Pi - 0 i (A.21) C = ( 1 - P 2 ) t + 1 - ( 1 - 0 2 ) 1 + ( 1 / O ) P2 ( l - f t ) * + i _ ( i _ f c ) * + r P2 — 02 (A.22) I? = (1 - Pi)' - (1 - 0 i ) 1 + ( 1 / o ) Pi ( l - P l )k - ( l - 0 i ) f c Pi -01 (A.23) P2 (l-P2)k-(l-g2y P2 - 02 (A.24) T h e throughput of class 2 can be similarly derived. Its expresssion is the same as that of Si, except that all subscripts 1 are replaced by 2. Bibliography [1] F . A . Tobagi , "Multiaccess Protocols in Packet Communication Systems", IEEE Trans, on Communications, vol. C O M - 2 8 , No . 4, pp. 468-488, Apr i l 1980. [2] M . Malek, "Integrated Voice and Data Communications Overview", IEEE Communications Magazine, vol . 26, No . 6, pp. 5-15, June 1988. [3] C . J . Weinstein and J . W . Forgie, "Experience with Speech Communication in Packet Networks", IEEE J. Selected Areas Communications, vol. S A C - 1 , No. 6, pp. 963-980, December 1983. [4] D. Minol i , "Issues in Packet Voice Communicat ion", Proc. IEE, vol. 126, No. 8, pp. 729-740, August 1979. [5] T . V . Russotto, "The Integration of Voice and Data Communicat ion", IEEE Network, vol . 1, No. 4, pp. 21-29, October 1987. [6] G . L . Choudhury and S. S. Rappaport , "Priority Access Schemes Using C S M A -C D " , IEEE Trans, on Communications, vol. C O M - 3 3 , No. 7, pp. 620-626, July 1985. [7] I. Iida, M . Ishizuka, Y . Yasuda, and M . Onoe, "Random Access Packet Switched Local Computer Network with Priority Funct ion", Proc. NTC, Hous-ton, T X , pp. 37.4.1-37.4.6, December 1980. 94 BIBLIOGRAPHY 95 [8] F . A . Tobagi , "Carrier Sense Multiple Access with Message-Based Priority Functions", IEEE Trans, on Communications, vol . C O M - 3 0 , No . 1, pp. 185-200, January 1982. [9] L . Kleinrock and F . A . Tobagi, "Packet Switching in Radio Channels: Part I -Carrier Sense Multiple-Access Modes and Their Throughput- Delay Character-istics", IEEE Trans, on Communications, vol. C O M - 2 3 , No . 12, pp. 1400-1416, December 1975. [10] R. M . Metcalf and D. R. Boggs, "Ethernet: Distributed Packet Switching for Local Computer Networks," ACM, vol. 19 no. 7, pp. 395-404, July 1976. [11] J . DeTrevil le and W . D. Sincoskie, " A Distributed Experimental Communica-tion System," IEEE J. Select. Areas Communication, vol . S A C - 1 , pp. 1070-1075, December 1983. [12] N . F . Maxemchuk, " A Variation on C S M A / C D T h a t Yields Movable T D M Slots in Integrated Vo ice /Data Local Networks", B S T J , vol . 61, No. 7, pp. 1527-1551, September 1982. [13] C . H . Lea and J . S. Meditch, " A Channel Access Protocol for Integrated V o i c e / D a t a Applications", IEEE J. Selected Areas Communications, vol. S A C -5, No. 6, pp. 939-947, July 1987. [14] P. Spilling and E . Craighil l , "Digital Voice Communications in T h e Packet Radio Network", ICC, pp. 21.4.1-21.4.7, June 1980. [15] D . J . Goodman, R. A . Valenzuela, K . T . Gayliard and B . Ramamurthi , "Packet Reservation Mult iple Access for Local Wireless Communicat ions", Vehicular Technology Conference, pp. 701-706, June 1988. [16] G . V . Guardalben and M . L. Molle, " O n the Coexistence of Different C S M A Protocols on the Same Network", I C C , pp. 603-609, 1986. BIBLIOGRAPHY 96 [17] S. H . Shen and H . T . L i u , "Dynamic C S M A / C D with a Priority Scheme", IEEE INFOCOM, pp. 258-263, 1986. [18] H . Takagi and L. Kleinrock, "Throughput Analysis for Persistent C S M A Sys-tems" IEEE Trans, on Communications, vol. C O M - 3 3 , No . 7, pp. 627-638, July 1985. [19] G . J . Nutt and D. L. Bayer, "Performance of C S M A / C D Networks Under C o m -bined Voice and Data Loads" , IEEE Trans, on Communications, vol . C O M - 3 0 , No. 1, pp. 6-11, January 1982. [20] J . D . DeTrevil le, " A Simulation-Based Comparison of Voice Transmission on C S M A / C D Networks and on Token Buses", B S T J , vol . 63, No. 1, pp. 33-55, January 1984. [21] T . Bially, A . J . McLaughl in and C . J . Weinstein, "Voice Communicat ion in In-tegrated Digital Voice and Data Networks", IEEE Trans, on Communications, vol. C O M - 2 8 , No. 12, pp. 1478-1490, September 1980. [22] A . Hills and K . Scott, "Perceived Degradation Effects in Packet Speech Sys-tems", IEEE Trans, on Acoustics, Speech, and Signal Processing, vol . A S S P - 3 5 , No. 5, pp. 699-701, M a y 1987. [23] K . Bull ignton and J . M . Fraser, "Engineering Aspects of T A S I " , B S T J , vol. 38, pp. 353-364, March 1959. [24] P. T . Brady, " A Statistical Analysis of On-Of f Pattern in 16 Conversations", B S T J , vol . 47, pp.73-91, January 1968. [25] T . A . Gonsalves, "Packet Voice Communicat ion on an Ethernet Local C o m -puter Network: an Experimental Study", in Proc. ACMSIGCOM '83, 1983, pp. 178-185. BIBLIOGRAPHY 97 [26] C . J . Weinstein, "Fractional Speech Loss and Talker Act iv i ty Model for T A S I and for Packet-Switched Speech", IEEE Trans, on Communications, vol. C O M - 2 6 , No. 8, pp. 1253-1257, August 1978. [27] S. S. L a m , "Packet Broadcast Networks — A Performance Analysis of the R-A L O H A Protocol" , IEEE Trans, on Computers, vol . C-29, No. 7, pp. 596-603, July 1980. [28] K . S. Ch ing and H. W . Lee, " A Multiaccess Protocol for Integrated Voice and Data Transmission In a Packet Radio Network," submitted to IEEE INFO-COM '89 for publication. 

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.831.1-0064778/manifest

Comment

Related Items