D-RMA: A Dynamic Reservation Multiple Access Protocol for Third Generation Cellular Systems

1. D-RMA: A Dynamic Reservation Multiple Access Protocolfor Third Generation Cellular Systems Published in IEEE Transactions On Vehicular Technology,Vol. 49, NO.5, September 2000 by Antonio Iera and Salvatore Marano. Presented by Antonis Stefanogiannis on 16th November 2004, as partial fulfilment of the postgraduate course ‘Special Topics in Communication Networks’ with the School of E.C.E., T.U.C.

2. INTRODUCTION (Towards 3G Wireless Broadband Personal Communication Systems) • Paper was published on September 2000…… • Personal Communication Systems (PCS), based on wireless technologies, have evolved towards the support of a wider range of applications including voice, video, data and multimedia. • Ideal scenario : several Mobile Audio Visual Terminals (MAVTs) by which users, via a Base Station (BT) connecting wired and wireless networks, have access to multimedia services scattered all over a high-speed communication backbone. • Attention focused on the integration between enhanced wireless systems and broadband networks, such as B-ISDN using ATM protocol. Enable wireless PCSs to embrace a wider range of communication capabilities, besides those offered (substantially voice services when this paper was published). • This integration process requires the design of Multiple Access protocols showing high flexibility and efficiency to support future integrated services. A.I.Stefanogiannis

3. MULTIPLE ACCESS PROTOCOLS (Overview) • Multiple Access Protocols proposed for satellite systems or wireless LANs (start point) reconsidered for cellular mobile radio systems. • An interesting output of this activity was the Packet Reservation Multiple Access (PRMA) protocol. • In PRMA the frame rate coincides with the arrival rate of voice packets so that a voice terminal requires exactly one slot per frame. Within the frame the terminal detects an available or reserved slot according to the feedback information stream broadcasted on the down-link from the BT. • An active terminal contends in order to access the channel on the available slots. If the attempt is successful, the transmission begins and the terminal holds a reservation for a slot in subsequent frames until the end of its talkspurt. • Every up-link slot can be used for reservation, thus an acknowledgement at the end of each slot is required on the down-link. • Main inconveniences : An increase in the traffic load causes a decrease in the probability of finding free slots in the frame and an uncontrolled increase in the access delay. Large number of acknowledgement slots are required on the down-link. A.I.Stefanogiannis

4. MULTIPLE ACCESS PROTOCOLS (Overview, cont.) • An adaptive protocol, named Multiple Access (MA) protocol attempts to solve the previous problem. This protocol allows a reservation only on the so-called R-slots, and the transport of information on the I-slots. • In this protocol, a minimum number of R-slots is fixed to guarantee good performance under high loading conditions. When the traffic is low every other free I-slot can be used for reservation. • Main inconveniences : partial separation between control and information channels. While, reservation packets transmitted over the R-slots are strongly protected against interference, this protection is absent on I-slots and the reservation packets can be corrupted by channel impairments. • A new version of PRMA, named PRMA++, has been proposed and analyzed. • Separation of control and information slots permits guarantees of different degrees of quality to control and information channels. • The number of R-slots as well as their positions within the frame are fixed. • Unlike PRMA, a lower number of acknowledgement slots are required on the down-link. • Main inconveniences : fixed number of R-slots has the effect of causing Mobile Terminals (MTs) to experience access delay even at low load. In the presence of multiple slot assignment (i.e. multimedia traffic), the experienced access delay may become a critical factor. A.I.Stefanogiannis

5. THE D-RMA PROTOCOL( A protocol for 3G Cellular Mobile Radio Systems) • A protocol which is explicitly designed to support multimedia traffic. • Based on PRMA++ but, it introduces a flexible dynamic approach in the choice of the percentage of bandwidth to be used for reservation. • It couples the advantage deriving from a complete separation between reservation and information channels , along with dynamic adaptation of the percentage of reservation bandwidth within a frame to traffic condition. It demonstrates better performance when compared to ‘non-dynamic’ protocols in terms of both the offered QoS and number of connections which can be activated in a microcell at one time. • Belongs to the family of protocols, which have been explicitly conceived with the aim of dynamically adapting the traditional Reservation Multiple Access protocols to multimedia traffic needs and sharing of buffer resources and channel bandwidth. • Up-link frame format constituted by a sequence of information slots (I-slots) and Reservation slots (R-slots), whose total number is set to N. • The choice of maintaining a complete separation between I-slots and R-slots, and dynamically adapting the percentage of reservation bandwidth (number of R-slots) within a frame to traffic conditions coupled with a suitable bandwidth allocation strategy, is preferred in order to meet the QoS requirements of multimedia (integrated video/voice) services. • This means that the number of R-slots, Nr varies dynamically. The condition Nrmin≤ Nr ≤ Nrmax< N holds, where Nrmin (Nrmax ) is the optimum number of the R-slots for the highest (lowest) load. Why??? A.I.Stefanogiannis

6. THE D-RMA PROTOCOL (Transmission Frame Format) • The transmission frame format of the proposed MAC protocol is shown in the figure below. • The positions for Nrmax R-slots within the frame are fixed. They are scattered in a homogeneous manner along the frame, so that a new burst will not have to wait a long time to intercept an available R-slot. • All non-activated R-slots, are used for transporting information, like I-slots. A.I.Stefanogiannis

7. THE D-RMA PROTOCOL (Access Mechanism) • When a terminal has packets to transmit (i.e. start of burst/spurt is detected by VAD), with probability p(permission probability), it sends a reservation packet on the first available R-slot. • If the reservation packet collides or is corrupted by channel impairments, a negative acknowledgement is transmitted from the BT (Base Station) on the down-link. Unsuccessful terminals retry to transmit their reservation packets on the next free R-slot, with probability p (generally pt ≥ pr ). • On the contrary, if a success occurs, the BT sends an acknowledgement on the down-link and assigns the requested number of I-slots to the calling terminal. If that number is not available, the reservation packet is buffered in a reservation queue within the BT, and waits for I-slots to become available. • The described contention mechanism is similar to other reservation protocols. • What really makes the D-RMA protocol really different from other reservation protocols??? A.I.Stefanogiannis

8. THE D-RMA PROTOCOL (Access Mechanism, cont.) • What really makes the D-RMA protocol really different from other reservation protocols, is the dynamic management of reservation bandwidth. • In response to traffic load fluctuations (in terms of number of transmitting terminals i.e. originating/terminating calls or hand-off events), the BT is allowed to modify the number of active R-slots within the transmission frame. • It will be demonstrated, later on, that better performances compared to fixed number of active R-slots (‘fixed Nr’ policy) can be obtained by properly varying Nr. It will also be shown that a reduction of Nr is preferred when traffic load increases and an increase in Nr is desired when the load decreases. • The optimum number of active R-slots for a given channel load value, has to be chosen according to a quality index, which is an ad-hoc defined QoS metric, and will be presented later on. A.I.Stefanogiannis

9. THE D-RMA PROTOCOL (Access Mechanism, cont.) • A block diagram for D-RMA protocol (model of the access mechanism) is given in the figure below. A.I.Stefanogiannis

10. THE D-RMA PROTOCOL (Access Mechanism, cont.) • As mentioned before, the proposed protocol is able to dynamically adapt itself, on a frame basis, when traffic load variations occur. Particularly, at the end of each frame the BT runs an algorithm which examines the channel load (by controlling the number of terminals within the microcell, which have a connection established) and determines a suitable Nr value. This algorithm is computed by the Reservation Bandwidth Controller, as shown in the previous figure. • Bandwidth requests which cannot be served are buffered and handled according to a specific rule (FIFO, Scan & Serve) implemented by the Access Controller. The latter also allocates the requested Basic Bandwidth Units (BBUs), I-slots, to each terminal according to suitable strategies, such as Complete Sharing (CS), Complete Partitioning (CP), or Mutually Restricted Access (MRA). Therefore, the Access Controller performs both reservation queue management and information bandwidth allocation management, following specific policies for each one. Both of these types of management have a strong effect on the performance of D-RMA protocol. • An overview of FIFO and Scan & Serve reservation queue (buffer) management policies and CS,CP,MRA information bandwidth allocation management strategies will be given later on. A.I.Stefanogiannis

11. THE D-RMA PROTOCOL (Access Mechanism, cont.) • An efficient and simple technique to manage the variation of Nr by traffic load fluctuations, is a ‘threshold-based strategy'. According to this strategy, the BT (Reservation Bandwidth Controller) decides to change Nr when traffic load exceeds one of a set of fixed threshold values. Specifically loadto-threshold mapping is available as a look-up table whose figures are a-priori evaluated, according to a quality index which will be defined later on, in this presentation. • Is only a threshold value enough??? • With the above approach, a persistent need to vary the active R-slot number, Nr, in subsequent frames will arise, when the load rapidly fluctuates around the threshold value. This phenomenon would create an undesired overload of signalling traffic on the downlink. In order to overcome this problem an hysteresis control mechanism is introduced. Precisely, each threshold value is provided with an hysteresis margin, as illustrated in the figure below (optimum number of active R-slots, Nr, vs. traffic load), which permits a change in Nr only when offered load shows a convincing (actual) trend of rising (when crossing Lr) or falling (when crossing Lf). A.I.Stefanogiannis

12. THE D-RMA PROTOCOL (Simulation Study, Parameters Definition) • A flexible discrete event simulator has been implemented and used to study the D-RMA protocol behaviour. • The achieved performance has been analyzed and compared by evaluating the following parameters : DHvo : the voice holding time - the time spent from the instant when a talkspurt is generated until its first packet is transmitted on the channel (upper bound 32ms). DHvo consists of two components, Dvo, DPROCvo and it holds that DHvo= Dvo+ DPROCvo , where the first term is the access delay (i.e. the time required to gain a channel reservation from the BT), and the second term is the time required to process the data before the transmission over the air interface. DHvi :the video holding time – the analogous definition of DHvo for video (upper bound of 100ms for time-critical traffic, like video). PdropVO : the average packet dropping probability for a voice terminal, computed as dropped packets/arrived packets (upper bound 0.01). PdropVI : the average packet dropping probability for a video terminal (upper bound 0.001). A.I.Stefanogiannis

13. THE D-RMA PROTOCOL (Simulation Study, Quality Index Definition) • The following quality index, Q, is defined : • Where PdropVO, PdropVI , Dvo , Dvi , were defined before, α is the percentage of voice terminals over the total number of transmitting terminals and TON is the average talkspurt duration. • Actually, quality index, Q, is an ad-hoc defined QoS metric. The quality index is the sum of two terms, one concerning video traffic and the other one concerning voice traffic. Both the access delays and the packet dropping probabilities of the two examined types of traffic are considered to define the quality index, Q. • Previewing the sum of these two terms, we obtain that Q is a dimensionless number taking values in the interval [0,1].The maximum admissible value of Q (i.e. Q=1) is obtained in ideal conditions when both dropping probabilities and access delays are zero. • For a given load, the optimum number of R-slots which can be activated, is given by the following criterion : ‘Optimum Nr is the one which maximizes Q’ A.I.Stefanogiannis

14. THE D-RMA PROTOCOL (Simulation Study, Quality Index Definition, cont.) • Performance index, Q, considers even small additional access delays undesirable. The rationale of this can be easily understood by focusing on the concept of holding time. • Holding time takes into account the time spent within the system to cross thewhole transmitting chain (contention, queuing, data processing : channel coding, interleaving, modulation, etc.). For some services the time required for data processing can be high , this reducing the value of the allowable access delay. For this reason it becomes very important to minimize Dvo (Dvi ) as much as possible to leave more time available for performing other transport layer functions. The greater the margin, the more effective the techniques can be (i.e. deeper interleaving). A.I.Stefanogiannis

15. THE D-RMA PROTOCOL (Simulation Study, Speech Traffic Model) • The speech traffic model used during the simulation tests considers a voice source as a sequence of talkspurts and gaps. By assuming that a Voice Activity Detector (VAD) can be used to differentiate between principle talkspurts (ON) and principle gaps (OFF), voice traffic can be characterized by the two-state Markov chain model displayed in the figure below. • In the ON state, voice packets are generated at a constant rate. No packets are generated in the OFF state. Time spent in each state is exponentially distributed with means α-1 (TOFF) for the OFF state and β-1(TON) for the ON state. • Reservation required when in the ON state and released when in the OFF state (no packets available for transmission). • If a packet is queued in a terminal buffer for a time longer than Dmaxvo it is dropped. A.I.Stefanogiannis

16. THE D-RMA PROTOCOL (Simulation Study, Video/Data Traffic Model) • Video or data terminals are modelled as bursty sources, which generate frames of variable length, exponentially distributed with mean L (Kb). The arrival rate of a new burst is assumed equal to λm. • Since a video service is a ‘time critical service’, it is also assumed that video packets are dropped if their waiting time within the terminal queue exceeds the maximum delay value Dmaxvi. A.I.Stefanogiannis

17. THE D-RMA PROTOCOL (Simulation Study, List of Nominal Values Utilized in Simulations) • Reported simulations have been run under the conditions listed in the following table. • A voice source generates data @ 8Kbps thus requiring one BBU (I-slot) per frame. On the contrary, considered video source generation rates are 32Kbps and 64Kbps, which require four and eight BBUs (I-slots) respectively. A.I.Stefanogiannis

18. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Voice-Only Traffic Performance) • First, we examine the D-RMA performance results when in the presence of voice-only traffic (homogeneous traffic). • The illustrated figure (optimum Nr vs. voice channel load ρ) shows that the optimum number of active R-slots, Nr, has a linearly decreasing relationship with the channel load, ρ, and consequently with the number of active terminals. Why??? A.I.Stefanogiannis

19. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Voice-Only Traffic Performance, cont.) • In Reservation Multiple Access (RMA) protocols two contributions affect access delay. • The first is due to the access contention process, referred as contention delay Dcon . • The second is due to the waiting time of the reservation packet within the BT reservation queue, referred as queue delay, Dque . • The access delay Dvo= Dcon + Dque. • Under low traffic load conditions, the access contention delay is dominant. Therefore, the presence in the frame of a greater number of available R-slots contributes to reduce the access delay. • Conversely, for great values of traffic load, the queue delay dominates. Thus, it is rational to reduce the number of active R-slots in order to increase the information channel bandwidth. A.I.Stefanogiannis

20. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Voice-Only Traffic Performance, cont.) • The figure below (average voice access delay, Dvo , vs. voice offered channel load ρ), shows the access delay Dvo as a function of the offered channel load ρ, for both D-RMA and a protocol which exploits fixed number of R-slots, ‘Nr-fixed’ protocol. • The lower the access delay, the higher the number of terminals that can be simultaneously activated. A.I.Stefanogiannis

21. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Voice-Only Traffic Performance, cont.) • The figure below, shows the average packet dropping probability Pdrop for voice terminal as a function of the offered channel load ρ, for both D-RMA and a protocol which exploits fixed number of R-slots, ‘Nr-fixed’ protocol. • The lower the packet dropping probability, the higher the number of terminals that can be simultaneously activated. A.I.Stefanogiannis

22. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Voice-Only Traffic Performance, cont.) • In order to better understand the improvement deriving from the adoption of D-RMA instead of a ‘Nr-fixed’ solution, we refer to the previous figure (which is illustrated again below for convenience) and focus on the worst case of Pdrop=10-2. • If Pdrop has to be kept below this upper bound, then the maximum value of admissible load when D-RMA is adopted is about 0.89.In the same condition, the values obtainable when Nr=6 and Nr=8 are adopted, are 0.84 and 0.80 respectively. Thus, the achieved gains are about 5% and 10%.Are these achieved gains low??? A.I.Stefanogiannis

23. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Voice-Only Traffic Performance, cont.) • Even greater margins are obtainable under different, not shown, simulated network conditions. • These percentages refer to one carrier only! As each microcell contains a set of carriers, a considerable increment in the number of simultaneously active terminals can be obtained. A.I.Stefanogiannis

24. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Multimedia Traffic Performance, cont.) • The protocol behaviour analysis is extended to the case of heterogeneous traffic. Reported simulations have been run under the nominal conditions listed in the table given before. Furthermore, it is assumed that the nature of the heterogeneous traffic is 75% voice and 25% video, the bandwidth allocation strategy applied is Complete Sharing (CS) and the reservation buffer policy is FIFO. A.I.Stefanogiannis

25. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Multimedia Traffic Performance, cont.) • The figure shown below illustrates the average voice access delay Dvo as a function of the total (voice+video) offered channel load ρ, for both D-RMA and a ‘Nr-fixed’ protocol. See nominal parameter values in previous table (Rvi=32Kbps, L=5.12Kb, ρ=75%voice + 25%video, bandwidth allocation strategy=CS). • Curves in the figure above, show that D-RMA can always guarantee a lower voice access delay when compared to ‘fixed-Nr’ algorithms. D-RMA achieves a greater delay performance if it is compared to algorithms exploiting a lower Nr. A.I.Stefanogiannis

26. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Multimedia Traffic Performance, cont.) • The figure shown below illustrates the average video packet dropping probability Pdropvi as a function of the total (voice+video) offered channel load ρ, for both D-RMA and a ‘Nr-fixed’ protocol. See nominal parameter values in previous table (Rvi=32Kbps,L=5.12Kb,ρ=75%voice + 25%video, bandwidth allocation strategy=CS). • If we focus on the above figure, for Pdrop=10-3 (upper bound value for the considered type of traffic), we notice a gain in the acceptable load which is about 6.7% when compared to Nr=6, about 11-12% when compared to Nr=8, and about 17% when compared to Nr=10. • These are greater margins than those achievable in the voice-only traffic case and they prove that D-RMA is particularly adequate for supporting multimedia application traffic in a heterogeneous mobile environment. A.I.Stefanogiannis

27. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Multimedia Traffic Performance, cont.) • This last figure illustrates the quality index Q as a function of the total (voice + video) offered channel traffic load ρ. See nominal parameter values in previous table (Rvi=32Kbps,L=5.12Kb,ρ=75%voice + 25%video, bandwidth allocation strategy=CS). • The improvement in the quality index Q, when optimizing an ‘optimum Nr based’ instead of a ‘fixed Nr based’ strategy, is obvious. An increase in the maximum number of simultaneously activated terminals with respect to the ‘Nr-fixed’ algorithm is always present. • But, the above figure reveals good Q measure performance under low traffic conditions, irrespective of the value of Nr. Why??? Is D-RMA protocol ineffective at low loading??? A.I.Stefanogiannis

28. THE D-RMA PROTOCOL PERFORMANCE RESULTS(Multimedia Traffic Performance, cont.) • At low loading, the adoption of a constant small Nr value would correspond to an increase in the voice access delay, as observed in the figures Dvo vs. ρ for both types of traffic (voice-only and multimedia). • Focusing on the figure Dvo vs. ρ for multimedia traffic type, it is obtained that a gain of about 10ms is achievable by substituting the Nr=4 technique with D-RMA. This amount of time can be differently exploited (to perform a deeper interleaving for example), during the transport on the air interface, while remaining within the same overall allowed transmission delay. • This is an advantage which cannot be evaluated through the Q index (see Q index definition), and it consists in a greater robustness of the successfully transmitted information against channel errors. A.I.Stefanogiannis

29. THE D-RMA PROTOCOL PERFORMANCE RESULTS (Coupling D-RMA Protocol with different Bandwidth Allocation Strategies and different Reservation Buffer Management Policies) • A detailed test campaign has been also performed by coupling D-RMA with different bandwidth allocation strategies (CS,CP,MRA) and different reservation buffer management policies (FIFO, Scan & Serve). As mentioned before, both of these types of management have a strong effect on the performance of D-RMA protocol. • Performance curves for various combinations of bandwidth allocation strategies and reservation buffer management policies are skipped in the paper due to length constraints. Nevertheless, the obtained results can be summarized by stating that the adoption of both the MRA strategy and the Scan & Serve policy, allow the achievement of the best performance when compared to simple CS strategy or MRA policy with FIFO queue management. • We will try to provide an explanation for the above result by overviewing the aforementioned bandwidth allocation strategies along with the reservation buffer management policies. A.I.Stefanogiannis

30. RESERVATION BUFFER MANAGEMENT POLICIES (FIFO, Scan & Serve) • Consider slot allocation procedure with two types of traffic – wideband (WB) traffic (requiring multi-slot allocation, i.e. video or graphics) and narrowband (NB) traffic (requiring single-slot allocation i.e. voice and data) – as a multi-server queuing system making use of a FIFO queue with reservation requests, with N identical, parallel servers where requests are served to completion. • Arriving request which cannot be served immediately is ‘put on hold’ in the queue for delayed service and can wait for a specific time and is dropped thereafter. • A given request can enter service only when all its requested number of servers are available. This means that servers maybe idle while requests are waiting! • Need for efficient utilization of servers while providing equitable access to different types of requests. A.I.Stefanogiannis

31. RESERVATION BUFFER MANAGEMENT POLICIES (FIFO, Scan & Serve, cont.) • Different traffic types access the N-server channel in their order of arrival or FIFO order. • A ‘large’ request (WB call) waiting at the head-of-the-line (HOL), may adversely block a large number of ‘smaller’ (NB calls) requests which could get service sooner! Therefore, FIFO policy is inefficient as far as the servers’ utilization/throughput is concerned. • However, any attempt to allow NB requests to overtake WB requests will result in indefinitely postponing the WB requests from accessing the channel. • Suppose we ignore this fairness issue. How can we overcome the FIFO disadvantage of having a large request wait at HOL while servers are unutilized??? • Scan the queue of waiting requests and insert into service those sets of ‘smaller’ requests, if any, that best fit into the currently available portion of the channel. This policy is known as Scan & Serve policy. A.I.Stefanogiannis

32. BANDWIDTH ALLOCATION STRATEGIES(CS, CP, MRA) • We consider again the integration of WB and NB types of traffic with a multiserver queuing system model. • Among many strategies, the simplest one is the Complete Sharing (CS), in which the total servers are shared on the first-come-first-service (FCFS) basis whenever sufficient servers exist. • Another strategy is the Complete Partitioning (CP), where the total servers are partitioned into distinct server groups, and they are assigned exclusively to each customer group. • Heuristic optimization of the above two strategies provides other strategies. • Let’s investigate MRA strategy. • Basic assumptions : 1. Consider broadband (BB) network serving two types of traffic, NB and WB. Assume that a common transport capacity of this network is divided by m BBUs. (In the wireless context, each uplink frame can be seen us a channel in which each slot represents one BBU). 2. The NB traffic requires one BBU, the WB traffic requires n BBUs (n≤m), and all BBUs assigned to a WB call are occupied and released together. 3. The WB traffic is non-queueable, whereas the NB traffic may be queued in this system (infinite queue length). A.I.Stefanogiannis

33. BANDWIDTH ALLOCATION STRATEGIES(CS, CP, MRA, cont.) • If no traffic access control is applied, random fluctuations in the NB traffic may cause excessive blocking of the WB traffic while some channels are idle. This is severe when the bandwidth ratio n for the two types of traffic is large, because the NB traffic leaves insufficient bandwidth for accepting a WB call request. • Solution : restrict the maximum number of NB calls that can be in service simultaneously in order to reserve bandwidth for the WB traffic. Similarly, excessive delay of the NB traffic caused by the overloaded WB traffic can be protected by setting some restriction on the WB traffic. • Denote these reference values of restrictions for the NB and WB traffics as the cutoff valuesc1and c2respectively. • Upon arrivals, an NB call is queued if 1. the number of NB calls in service is equal to c1 2. there is no free channel • Upon arrivals, a WB call is blocked and cleared if 1. the number of WB calls in service is equal to c2 2. there are not enough free channels A.I.Stefanogiannis

34. BANDWIDTH ALLOCATION STRATEGIES(CS, CP, MRA, cont.) • The cutoff value c1 , is chosen in a way that makes the reserved channels for the WB traffic a multiple of the bandwidth requirement of a WB call. • By varying c1and c2 we achieve an effective control to adapt to the varying traffic load. • Set c1=s and c2=r (the maximum number of WB calls that can be in service at the same time if the WB traffic can use the whole bandwidth, i.e. m= r∙ n +s, 0≤s<n), this MRA strategy becomes the pure CS strategy. • When c1+ n∙ c2=m , it becomes the pure CP strategy. • Therefore, MRA strategy is a generalization that includes CS and CP strategies. • Other strategies can be obtained by varying the cutoff values c1and c2. • Complete Sharing (CS) of bandwidth for non-multimedia, and Mutually Restricted Access (MRA) for multimedia services seem to be promising strategies in a multimedia wireless environment. A.I.Stefanogiannis

35. CONCLUSION • Design of 3G cellular systems requires the adoption of radio channel access schemes, efficient in supporting multimedia services. • D-RMA both permits wireless networks to embrace a wider range of capabilities and contributes to the research work toward the definition of the Personal Communication Systems (PCS). • Its behaviour has been observed under homogeneous and heterogeneous traffic conditions. • In order to compare D-RMA behaviour with that of other traditional access protocols, a new quality index has been defined. • Better performance resulted from the adoption of the dynamic approach when compared to a ‘Nr fixed’ policy. • Trade-off : complexity vs. protocol performance (in terms of both the offered QoS to multimedia services and number of connections that can be activated simultaneously in a microcell at one time). A.I.Stefanogiannis

36. REFERENCES • Antonio Iera, Salvatore Marano, ‘D-RMA : A Dynamic Reservation Multiple Access Protocol for Third Generation Cellular Systems’, IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 49, NO. 5, SEPTEMBER 2000. • Pallapa Venkataram, Anirban Roy, A. Chockalingam, ‘Performance of a Link Control Protocol for Local Wireless Multimedia Communications’, DEPT. OF ELECTRICAL COMMUNICATION ENGINEERING, INDIAN INSTITUTE OF SCIENCE, BANGALORE 560012,INDIA. • Young Han Kim, Chong Kwan Un, ‘Analysis of Bandwidth Allocation Strategies with Access Restrictions in Broadband ISDN’, IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 41, NO. 5, MAY 1993. A.I.Stefanogiannis