A Hybrid MAC Protocol for Multimedia Trafc in Wireless Networks

Mainak Chatterjee and Sajal K. Das Center for Research in Wireless Mobility and Networking (CReWMaN) Department of Computer Science and Engineering University of Texas at Arlington Arlington, TX 76019-0015, USA E-mail: chat,das @cse.uta.edu

Abstract
This paper proposes a new MAC protocol, called request-TDMA/CDMA protocol, for supporting multimedia trafc in wireless networks. The protocol has CDMA laid over TDMA like frame structure which has two kinds of slots, viz., data slots and control slots. The control slot holds the information of the next frames data slot allocation. Whenever a mobile terminal needs to transmit a message, it sends a request to the base station and enters the contention process. The scheduler at the base station takes into consideration the time of generation of a call, the bit rate requirement and the message length while reserving slots for the entire length of the message generated. Three scheduling algorithms are proposed and their performances are studied for four classes of trafc. We also model our protocol using a two-dimensional Markov chain, and for a given system load we compute the state transition probabilities and derive the average waiting time. By simulation experiments we show that our (hybrid) request-TDMA/CDMA protocol is able to effectively combine the orthogonality of both time and code division multiplexing.

Introduction

In the past two decades there has been a tremendous growth in the eld of wireless communications. The ultimate goal is to provide communication services anywhere, anytime using light-weight portable devices at minimum cost with acceptable delay, quality and security. In general, wireless networks are divided into coverage areas, each of which is served by a central controller. In a typical cellular architecture, these coverage areas are called cells and the central controller is the base station. More recently, there has been a growing interest to provide wireless access to applications that are typically of local area networks giving rise to the concept of wireless LANs (WLAN) [10].
This work is supported by Texas Advanced Research Program grant TARP-003594-013 and by Nokia Research, Irving, Texas.

Since sharing of the limited radio spectrum resource is essential and a need for medium access control (MAC) protocol arises. One of the main considerations in the design of a wireless system is to incorporate multiple access schemes [12] that make efcient use of the allocated bandwidth. With the advent of high bandwidth wireless networks, various kinds of multimedia trafc text, voice, audio and video are expected to be supported. These applications are characterized by quality-of-service (QoS) parameters such as bit-rate (bandwidth) requirement, delay, jitter, etc [8]. In a multimedia environment the real-time applications generally have higher priority than the non-real time trafc and hence they are allocated a signicant portion of the bandwidth. Most of the known wireless MAC protocols are not specically designed to support multimedia applications. Since a single protocol cannot often handle the throughput and latency demands of such applications, hybrid protocols [4, 5, 9] are designed which combine the features of more than one protocols and thus perform better. The cellular radio capacity using special multiple access schemes has been studied [2]. Multiple access having at least CDMA (code division multiple access) or SSMA (spread spectrum multiple access) component are superior to other multiple access schemes because by these techniques, the frequency selectivity of the radio channel, which severely impairs the system performance, can be averaged out [3]. In this paper we propose a request-TDMA/CDMA protocol for supporting multimedia trafc for wireless networks. Our protocol is a combination of xed and random assignments for supporting trafc with various data rates. The resources are managed by the base station (scheduler) and are allocated to the users based on certain criteria. The performance of the protocol is analytically measured in terms of the average waiting time for each message before it gets a slot reservation and also the average channel utilization. We conduct simulation experiments in which three scheduling policies are studied which are specically meant to take care of the variable bit-rate (VBR) requirements of the users.

Proposed Protocol

The two most important aspects of a MAC protocol is the design of the frame structure and the scheduler at the base station. We describe them below for our proposed requestTDMA/CDMA protocol.
FRAME
. . . . . . . . . . . . . .

zero to a maximum, although the sum of the reserved and competitive slots is xed. The user who has reserved slots, will have the reservation till the entire message is transmitted, irrespective of the number of frames required.

2.2
Data Slot S
C T S

The Scheduler

R T S

Data Slot 1

Data Slot 2

User 1 User 2 User 3

1 2

User U

No. of data slots

Figure 1. The proposed frame structure

2.1

The Frame Structure

In our protocol, time is divided into xed size frames which repeats itself, i.e, a new frame begins as soon as the previous one ends. A frame is divided into two kinds of slots, the data slots and the control slots. Control slots are further divided into request to send (RTS) and clear to send (CTS) slots as shown in Figure 1. Each frame has data slots of equal length which are used for transmission of data packets from the users to the base station (via uplink). The RTS slots are used by the users to send their requests to the base station. The processing of the requests and the scheduling is done during the CTS slots. The CTS slot is divided into mini slots, each holding information of the corresponding data slot for the next frame. Each mini slot is further divided into grids, where is the maximum number of users who can transmit data simultaneously within a data slot. Each of these grids is initialized with a CDMA code that the scheduler allocates to the user who succeeds in getting a reservation for that slot. In a TDMA system, each data slot can be exclusively used by one user at any point of time, whereas a CDMA system allows multiple users to share the same channel in the time and frequency domain. This is achieved by assigning unique codes to all the users. The codes have the property that the cross-correlation among any two codes is very low. We exploit this orthogonality property [1, 11] to accommodate multiple users in one data slot. The maximum number of users who are allowed to transmit their data in one slot is restricted to as shown in Figure 1. Each user is allocated a code by the CTS which he uses to transmit data in one or more data slots in a frame. The number of data slots allocated per frame depends on the bit-rate requirement of the call. The data slots can be of two types reserved and competitive. The reserved slots are meant for a particular class of trafc and cannot be accessed by another class, whereas the competitive slots can be accessed by any class of trafc. The number of reserved slots is not xed and varies from

The scheduler at the base station scans the request queue which keeps track of all the events occurring in the system including the time of generation of a call and bit-rate requirements. The scheduler allocates data slots depending on the bit rate requirement of the call. Higher bit rate calls are assigned multiple data slots in a frame, thus enabling the users to transmit more data per frame. The lowest bit rate request is treated as a single request and higher bit rates as multiples of the lowest bit rate. For example, if the requested bit rate is twice the lowest bit rate then it is considered as two requests and hence two data slots per frame are allocated. Before processing the requests, the class of the data trafc is checked and a decision is made depending on the availability of reserved and competitive slots.

No. of Simultaneous Users

Analytical Model

For the sake of simplicity, we make certain assumptions for the system under consideration. We focus only on the uplink trafc, the ones from the nodes (users) to the base station. Let us assume that the number of nodes in the system is . The duration of a frame is seconds and the number of data slots per frame is . Therefore, time of each data slot is seconds, where and are respectively duration of the CTS and RTS slots. We also assume the following.

A1. Each node generates messages with a rate per frame, which is Poisson distributed. Also, whenever a message is generated, the entire message for which the connection is requested is ready. A2. In spite of assumption A1, the nodes cannot generate a new message until all packets of the current message are transmitted completely. Moreover, if a node ends transmission in the current frame, it does not generate a new message in the same frame. A3. Each message consists of a number of packets, and the message length (measured in number of packets) follows a negative exponential distribution. The mean message length is . A4. A node generating a message in the current frame cannot access the data slots in the same frame. It has to wait at least for the beginning of the next frame. A5. A node having a message to transmit and yet to establish a connection, stays in the queue and contends for the data slots in every successive frames till it gets a reservation, otherwise it is dropped.

A6. The number of data slots that can be allocated to a node will depend on the bit-rate requirement of the node. The number of data slots allocated per frame remains xed for the entire length of a message. A7. A node after getting a reservation goes into the transmission phase and continues to transmit the entire length of the message. No reserved slots goes empty as a consequence of assumption A1. We consider classes of trafc having bit-rate requirements in the ratio of 1:2: :. Let the number of data slots required by a node per frame be , which can take values between 1 and depending on the bit-rate requirement. So, when a node is allocated data slots, it will be allocated a minimum of 1 and a maximum of slots per frame. As discussed earlier, the CTS slot has columns and rows. So, there are slots in a frame, each of which can transmit a packet. Each of these data slots are either reserved or unreserved. Let be the probability that a data slot is unreserved and be the probability that it is reserved. Each of the nodes in the system can be in one of the three possible states; (i) thinking state, i.e., the node is yet to generate a message; (ii) contention state, i.e., the node has generated a message and is yet to get a reservation; and (iii) reserved state, i.e., the node has obtained a reservation and is under transmission.

, and be the number of states in the reserved, thinking and contention states respectively. So, we can write the following constraints

. For any number of nodes N, the system has a nite number of states which can be given by the tuple . Let

and

where = is the number of data slots. The cardinality of the state space can be obtained by all the possible combinations of the tuple with the constraint that . It can be shown that for a xed value of and varying between 0 and Y, the possible number of states is .

3.2

Transition Probabilities

In a Markov chain, if the system is in state , then there is a xed probability that it will next be in state regardless of the process history prior to arriving at [6]. We refer to as the transition probabilities, which satisfy and where To calculate the elements of the state transition matrix, we need to calculate the . The trantransition probabilities sition between the different modes is as shown in Figure 2.

nc

3.1

The state of the system with nodes will not be completely determined by only one variable because there can be various combinations of the number of nodes waiting in the queue and the number of nodes in the reserved state. We model our system as a two-dimensional Markov chain in which a state can be completely represented by a tuple ( ), where : Number of nodes in contention mode : Number of nodes in reserved mode : Number of nodes in thinking mode. The performance of the system can be evaluated by studying the distribution of the state variables. The system is fully known at the end of the CTS, i.e., at the end of each frame. So, we can take this point as the embedded Markov point. The state of the system that is reached at the end of each frame depends only on the arrivals, the departures and the result of the contention process during the interval of that frame only. Hence, the frame boundary is an ideal place to represent the system evolution, as a Markov chain. Let us assume that the state of the system after the frame be given by , and after the frame be given by . Let be the number of requests for reservation, be the number of nodes departing after transmitting their message and be the number of nodes that are successful in getting a reservation during the inter frame. So, we have, val between frame and

Markov Model

Thinking Mode

Contention Mode

Reserved Mode

nt

nr

Figure 2. The transition probabilities 3.2.1 Probability of new arrivals ( ) We have assumed that the arrival rate is per node per frame. So, the probability that there are arrivals, given the number of nodes in the thinking mode is , the framelength is , is equal to .

3.2.2 Probability of obtaining reservation ( ) In each frame there are slots which can either be in a reserved state or in an unreserved state. Let be the probability that a slot is in an unreserved state. Consider a request which has a requirement of slots per frame. A single code has to be assigned to the slots, which effectively means that the slots should be from the same row. For the request to get a reservation, there must be at least unreserved slots in the same row. Therefore, the probability of a node in the contention state going to the reserved state is equal to the probability that there is at least one row with at least unreserved slots. The complement of this will be to nd the probability that all the rows have strictly less than unreserved slots, which is given by

where is the probability that the number of unreserved slots in the row is less than . As the status of each row is inde-

pendent of that of the other, the above expression reduces . The probability of having unto reserved slots out of slots is given by . The probability that a node does not get a reservation in any of the rows is obtained by summing the probabilities of having 0 to unreserved slots which is given by . Therefore, the probability that a request with requirement of data slots gets a reser vation is then given by . Thus, for nodes to go from the contention mode to the reserved mode, given that there are nodes in the con . tention mode, is given by

probability of waiting for the second epoch is independent of the fact that it waited for the rst epoch, therefore we can regard the various epochs as separate but similar experiments in the spirit of Bernoulli trials. A known result in Bernoulli trials states that the expected number of experiments to be performed before success is reached is . Therefore, it follows that the number of epochs a request waits before leaving the queue is given by . It is not necessary that the arrivals occur only at the beginning of a frame. The expected time of arrival in a frame is and not . So the correct expected waiting time is given by .

Proposed Scheduling Algorithms

3.2.3 Probability of departures ( ) To nd the probability that there are departures, given that there are nodes in the reserved mode, we should take into consideration the probability of departures of all classes of trafc separately. Let us assume that there are departures from nodes in the reserved mode where is the number . So we of nodes belonging to class , for have . For a node in transmission mode belonging to class , the probability of departure is , where is the mean message length. The probability that there are departures from nodes in the reserved mode is given by . Therefore, the probability of a total of departures from all the classes combined is given by with the constraint that . We denote the prob ability of departures from nodes in the reserved mode as We will now consider all the possible ways in which the state can go to the state with all the constraints satised. Let us say, and where and can be either positive or negative. Depending upon the values of and , four cases may arise for which can be written in the form .

The scheduler at the base station scans through the request queue and tries to allocate slots as requested by the nodes. If the scheduler nds enough empty slots to accommodate a new request with its bit-rate requirement, it reserves slots for the entire length of the message. The nonserviced request are again considered for the next frame and this process continues till it gets a reservation or forcefully dropped from the queue. A node can only transmit if it gets a reservation for the data slots. The reservation is made for the entire length of the message. Once a reservation is made, the node has to transmit packets and no free slots will be allowed. This is due to the assumption of bulk arrival of the entire message. We rst present two simple schemes and study their performance with respect to average waiting time and channel utilization. Then we propose a third scheme as a hybrid of the rst two schemes. These three schemes mainly deal with how reservation is done for various classes of trafc. For brevity, we consider four classes of trafc. The ratios of their bit-rate requirements are 1:2:3:4, which means that the requests in each class require 1,2,3 or 4 data slots per frame, respectively.
Data Slot 1 Code 1 Code 2 Code 3 Code 4 Data Slot 2 Data Slot 3 Data Slot 4 Data Slot 5

Resv. for Class 1 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 2 Resv. for Class 2 Resv. for Class 3 Resv. for Class 3 Resv. for Class 3 Resv. for Class 1 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 2 Resv. for Class 2 Resv. for Class 3 Resv. for Class 3 Resv. for Class 3 Resv. for Class 1 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 2 Resv. for Class 2 Resv. for Class 3 Resv. for Class 3 Resv. for Class 3

3.3

Expected Waiting Time

Code 5

The expected waiting time in the contention mode will

a request given by , where is the duration of a time-frame. We know, , where is the probability that it waits for epochs. We regard the probability of a request leaving the queue after waiting for one epoch (one frame) as and think of it as the probability of success in Bernoulli trials. Note that the

of be

Code 6

Code U-1 Resv. for Class 1 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Resv. for Class 4 Code U Resv. for Class 2 Resv. for Class 2 Resv. for Class 3 Resv. for Class 3 Resv. for Class 3

Figure 3. Reservation for various types

Scheme 1 : No reservation for any class of trafc In this scheme, none of the slots are reserved for any class of trafc and any node requesting for data slots can contend for an unreserved slot. If a node is allocated multiple data slots in a frame, then the same code is used, i.e, the data slots must belong to the same row. Scheme 2 : Complete reservation for each class of trafc In this reservation scheme, data slots are identied for each class of trafc and only the intended class can use the corresponding reserved data slots. The reservation of each class is based on the assumption that the arrival rate of each class is equal, i.e, requests from all the classes are equally probable. The reservation made for the four classes considered is as shown in Figure 3. It can be seen that the number of data slots reserved for the four classes is in the ratio 1:2:3:4. The scheduler after checking the bit rate requirement (say ) of the request tries to allocate slots from those already reserved for that particular class. If it nds so, then it reserves slots for the next frames, where is the length of the message. The request will not be allocated slots if all data slots meant for that class are already reserved, even if data slots for other classes are available. Scheme 3 : Partial reservation for each class of trafc The previous two schemes have certain drawbacks when working independently. To eliminate those limitations, we derive a hybrid scheme in which not all the data slots are reserved, nor are they left unreserved. Instead, some data slots are reserved for each class of trafc depending on their relative arrival rates and some data slots are not reserved for any class. The unreserved slots can be assigned to any class of trafc which does not nd slots from the reserved slots for that class. On the arrival of a request, the scheduler rst tries to allocate slots which have been reserved for that class. If it cannot be accommodated, then the scheduler look for unreserved slots from the common pool.

age waiting time and channel utilization as the performance measures.

30

Average Waiting Time (frame)

25

20

15

10

0 0.1

0.2

0.3

0.4

0.5 Arrival Rate ()

0.6

0.7

0.8

0.9

Figure 4. Performance of Scheme 1

350

300

Average Waiting Time (frame)

250

200

150

100

50

0 0.1

0.2

0.3

0.4

0.5 Arrival Rate ()

0.6

0.7

0.8

0.9

Figure 5. Performance of Scheme 2

35

30

Average Waiting Time (frame)

25

20

15

10

0 0.1

0.2

0.3

0.4

0.5 Arrival Rate ()

0.6

0.7

0.8

0.9

Simulation Results

The wireless system considered in our simulation experiments consists of only one cell in which there are active nodes (or users) which are generating messages to be transmitted to another node. The rate of generation of messages is Poisson distributed with a mean of messages per node per frame. Each message has a certain length which is exponentially distributed with the mean size of packets. We use frame-time as the unit of time. of We have considered ve data slots which occupy the frame duration and the two control slots (RTS and CTS) occupy the remaining of the frame duration. One data slot time is the time taken for the transmission of one data packet. Maximal Length codes of length 15 were used and there were 15 such codes [7] which were allocated to every data slot. Therefore, at any time a maximum of nodes can transmit data simultaneously. We consider the aver-

Figure 6. Performance of Scheme 3 The performance of Scheme 1 is shown in Figure 4. It is observed that the trafc with higher bit-rate requirement were delayed more than the trafc with lower bit-rate requirement. This is due to the fact that it is less probable to accommodate requests with larger bit rate since the same code has to be used for all the data slots. The performance of Scheme 2 is depicted in Figure 5 where trafc of class 1 had a considerable amount of delay. This is due to the assumption that the mean message lengths of all the types are same and class 1 required more time to vacate the reserved slots since it could only transmit one packet per frame, whereas the other types transmitted more number of packets in a frame and vacated the reserved slots after the transmission was complete. So, for the same mean message length, the nodes with higher bit rates needed less time for the transmission. As all the 64 nodes randomly generated trafc belonging to all four classes, more class 1 trafc got

0.9

Scheme 1 Scheme 2 Scheme 3

Average Channel Utilization

0.8

0.7

for different values of channel utilization. Figure 8 shows the waiting time for class 2 and 4 only. We observe that for class 2 the waiting time is appreciably low even when the system is loaded but for class 4 the waiting time blows up when the channel utilization exceeds 0.7.

0.6

0.5

6
0.2 0.3 0.4 0.5 0.6 Arrival Rate () 0.7 0.8 0.9 1

Conclusion

0.4 0.1

Figure 7. Comparison of the three schemes

queued due to its low probability of departure, resulting in very high waiting time. For Scheme 3, a total of 8 codes (rows) were reserved for the various trafc types. The ratio of reservation of each class was based on the average requirement for that class. The remaining 7 codes were left unreserved and any class could access it. The relative performance of the three schemes are shown in Figure 7. Though the hybrid scheme has the same waiting time as Scheme 1, it performs better in terms of channel utilization. In fact the channel utilization of Scheme 3 is very , which means that there were very few data close to slots which were not allocated to any node and thus were wasted. The reserved slots catered the minimum steady ow, whereas the unreserved slots were used for the uctuations in the arrivals of the requests of different classes.

We have proposed a a new MAC protocol for wireless access systems, called the request-TDMA/CDMA protocol. This protocol has a feature that the central station (scheduler) allocates one or more slots depending on the requests made by various nodes, so that it can support the variable bit-rate trafc. Three scheduling schemes have been proposed and simulation experiments conducted. The partial reservation scheme outperforms the no-reservation and complete-reservation schemes in terms of delay and throughput. Furthermore, the proposed protocol has a high channel utilization compared to [13], which was a purely TDMA based protocol.

References
[1] A.J.Viterbi, CDMA principles of Spread Spectrum Communications. Addison-Wesley, 1995. [2] K.S.Gilhousen, I.M.Jacobs, R.Padovani, A.J.Viterbi, L.A.Weaver, and C.E.Wheatly, On the capacity of a cellular CDMA system, IEEE Trans. on Veh. Tech, Vol. 40, No 2, pp.303-312, May 1991. [3] P.Jung, P.W.Baier and A.Steil, Advantages of CDMA and spread spectrum techniques over FDMA and TDMA in cellular mobile radio applications IEEE Trans. on Vehicular Technology, Vol.42, No.3, pp.357-364, Aug 1993. [4] H. van Roosmalen, J. Nijhof and R. Prasad, Performance Analysis of a Hybrid CDMA/ISMA Protocol for Indoor Wireless Computer Communications, IEEE J. on Selected Areas of Communications. Vol.12, No 5, pp.909-916, June 1994. [5] A.K.Elhakeem, R. Di Giralamo, I.B.Bdira and M.Talla, Delay and Throughput Characteristics of TH, CDMA, TDMA, and Hybrid Networks for Multipath faded Data Transmission Channels, IEEE J. on Selected Areas of Commun. Vol.12, No 4, pp.622-637, May 1994. [6] D. Bersekas and R. Gallager, Data Networks, Prentice Hall, 1996. [7] W.C.Y. Lee, Overview of Cellular CDMA, IEEE Trans. on Vehicular Technology, Vol.40, No 2, pp.291-302, May 1991. [8] K. Ravindran and V. Bansal, Delay Compensation Protocols for Synchronization of Multimedia Data Streams, IEEE Trans. on Knowledge and Data Engineering.Vol.5, No 4, pp.574-589, Aug 1993. [9] J. Blanz, A. Klein, M. Na han, and A. Steil, Performance of a Cellular Hybrid C/TDMA Mobile Radio System Applying Joint Detection and Coherent Receiver Antenna Diversity, IEEE J. on Selected Areas in Communications.Vol.12, No 4, pp.568-579, May 1994. [10] R.O. LaMaire, A. Krishna, P. Bhagwat, and J. Panian, Wireless LANs and Mobile Networking : Standards and Future Directions, IEEE Communications Magazine. pp.86-94, August 1996. [11] G.E. Bottomley, Signature Sequence Selection in a CDMA System with Orthogonal Coding, IEEE Transactions on Vehicular Technology, Vol.42, No 1, pp.62-68, Feb 1993. [12] H.S. Chhaya and S. Gupta, Performance of Asynchronous Data Transfer Methods of IEEE 802.11 MAC Protocol, IEEE Personal Communications, pp.8-15, Oct 1996. [13] G.R.J. Linnenbank, P. Venkataraman, P.J.M. Havinga, S.J. Mullender, G.J.M. Smit, A Request-TDMA Multiple-Access Scheme for Wireless Multimedia Networks, Intl. Workshop on Mobile Multimedia Communications MoMuc-3, New Jersey, Sept 25-27, 1996.

10 Class 2 Analysis Simulation

Waiting Time (frame)

0.1

0.2

0.3

0.4 0.5 0.6 Channel Utilization

0.7

0.8

0.9

10 Class 4 Analysis Simulation

Waiting Time (frame)

0.1

0.2

0.3

0.4 0.5 0.6 Channel Utilization

0.7

0.8

0.9

Figure 8. Performance of class 2 and 4 To validate the correctness of the analytical model, we compare the analytical and simulation results. For all the four classes consider, we obtain the average waiting time