A fair protocol for fast resource assignment in
wireless PCS networks
∗
Tsan-Pin Wang, Chien-Chao Tseng and Shu-Yuen Hwang
Department of Computer Science and Information Engineering, National Chiao Tung University, Hsinchu, Taiwan, ROC
Efficient sharing of communication resources is essential to PCS networks since the wireless bandwidth is limited. The Resource Auction Multiple Access (RAMA) protocol was recently proposed for fast resource assignment and handover in wireless PCS networks. The RAMA protocol assigns available communication resources (e.g., TDMA time slots or frequency channels) to subscribers one at a time using a collision resolution protocol based on subscriber ID’s. However, the RAMA protocol encounters an unfairness problem; furthermore, performance results also indicate that it is inefficient at transmitting fixed-length subscriber ID’s. Moreover, the emerging services such as teleconferencing have been presenting new challenges to dynamic-priority resource assignment. In this paper, we propose a modification to the RAMA protocol to improve its performance and resolve the unfairness problem. The proposed protocol also adopts dynamic priority assignment to improve the QOS for subscribers in overload environments.
1. Introduction
A personal communication services (PCS) network [5,6] is a digital communication system that enables subscribers to communicate with each other at any time from any lo-cation. To support the user mobility, a wireless link should be established before connection. Efficient sharing of com-munication resources (e.g., TDMA time slots or frequency channels) is essential to PCS networks since the wireless bandwidth is limited and very scarce.
As the demand for new services increases, next-generation wireless PCS networks will need to support in-tegration of various types of data, such as voice, video, and multimedia data in mobile computing environments [7,14]. Consequently, there is a need for fair access and fast re-source assignment for call origination and handoff due to the huge numbers of users and the small sizes of cells (e.g., microcells or picocells [8]) in future wireless PCS networks. Moreover, it is hard to provide acceptable quality of service (QOS) for emerging services such as teleconferencing in fixed-priority resource assignment. The emerging services have been presenting new challenges to dynamic-priority resource assignment [14].
Multiple access is one of the most important issues in communication networks, especially in wireless net-works. In the literature, several categories of multiple access protocols have been studied [2,9,10,13], including
fixed-assignment, random access, and demand-assignment.
Fixed-assignment protocols are inefficient because the as-signed bandwidth is wasted when the user has nothing to transmit. On the other hand, contention-based random ac-cess protocols [13] encounter stability problems in heavy load environments. Under heavy traffic, the throughput of contention protocols decreases rapidly because time slots ∗This research was supported by the National Science Council, ROC,
under grant NSC 85-2213-E-009-063 and NSC 86-2213-E-009-076.
are wasted in collisions between subscribers accessing at random times.
To eliminate these problems, a demand-assignment tocol, the Resource Auction Multiple Access (RAMA) pro-tocol, was proposed [2–4,10]. The RAMA protocol is a deterministic algorithm that can provide good performance even under heavy loads. Each subscriber ID consists of a 9-digit phone number and a priority digit. In each assign-ment cycle, the subscriber with the highest ID value is the unique winner. Using this collision resolution method, the RAMA protocol creates unfairness problems. For example, among subscribers with the same priority, the one with the largest phone number will always ‘win’ auctions. Further-more, even when only one subscriber requests a commu-nication resource, the entire fixed-length subscriber ID is still transmitted one digit at a time in the RAMA protocol. This fixed-length auction cycle significantly degrades the performance of the RAMA protocol.
It is shown that RAMA is one of the most promising multiple access protocol. However, although RAMA pro-vides enough assignments per second for many applica-tions in the cellular environment [4], it still poses us un-fairness and performance inefficiency problems. Further-more, adopting dynamic priority assignment is impossible in RAMA. Therefore, we propose an efficient and fair pro-tocol called random RAMA to adopt dynamic priorities for fast resource assignment in future high-capacity wireless PCS networks.
The remainder of this paper is organized as follows. In section 2, we present a brief introduction to the RAMA protocol. In section 3, we propose the random RAMA pro-tocol. Then in section 4, we present a performance analysis of the random RAMA protocol. Section 5 shows a scheme for dynamic priority assignment. Section 6 discusses some issues for random RAMA. Finally, we conclude this paper in section 7.
Figure 1. Example of RAMA auction. 2. Resource auction multiple access protocol
The resource auction multiple access (RAMA) protocol that is a demand-assignment access protocol facilitates fast access and resource assignment to spatially distributed sub-scribers in a deterministic manner irrespective of loading.
In RAMA, each subscriber has a unique ID that consists of a priority digit and a nine-digit phone number. Available communication resources are ‘auctioned’ one at a time us-ing a collision resolution algorithm based on subscriber’s IDs. The subscriber with the highest ID value “wins” in an auction cycle. After each auction cycle, the base assigns an available communication resource to the winner. This assignment cycle is repeated for other available resources until either all requests are satisfied, or no more resources are available.
Subscribers requesting communication resources trans-mit their ID’s one digit at a time. The set of values that can be assumed by a digit is represented by a set of orthog-onal signals such as M -ary FSK or binary ASK. Figure 1 shows an example of a RAMA auction. In this example, we assume that each subscriber ID (d3d2d1d0) is represented
by a 8-ary FSK (i.e., 0 6 di 6 7), and subscribers with
ID’s 3421, 6313, 6422 and 6634 are seeking communica-tion resources. In the first auccommunica-tion time slot, all subscribers transmit their most significant digit d3, i.e., subscriber 3421
transmits a ‘3’ by transmitting an F3 FSK signal, and
sub-scribers 6313, 6422 and 6634 transmit ‘6’s by transmitting F6 FSK signals. The base detects these orthogonal
sig-nals, and feeds back the largest digit, ‘6’, to the subscribers by transmitting an F6 FSK signal. Upon receiving this
feedback, all subscribers with most significant digits lower than ‘6’ drop out of the auction, and wait for the next as-signment cycle. Those subscribers with d3 = 6 continue
by transmitting the next digit (d2). In this example,
sub-scribers 6313, 6422 and 6634 transmit ‘3’, ‘4’ and ‘6’,
respectively. After the base announces that ‘6’ is the large d2digit transmitted by this group of remaining subscribers,
subscribers 6313 and 6422 drop out of the auction. Sub-scriber 6634 continues by transmitting the remaining digits (d1 and d0) one digit at a time, and the base feeds back
the corresponding digits. When the entire fixed-length ID has been transmitted, the base broadcasts a resource assign-ment for subscriber 6634, the unique winner of the auction. In the next assignment cycle, subscribers that dropped out of the previous cycle (i.e., 3421, 6313 and 6422) partici-pate in a new auction along with requests from other new subscribers.
Note that although the winner 6634, in this example, has been uniquely identified by the base after transmitting the d2 digit, the entire fixed-length ID still must be
transmit-ted in the RAMA protocol. Moreover, if subscriber 6634 requests an additional resource in the next assignment cy-cle, subscribers with the same priority but smaller ID val-ues (i.e., 6313 and 6422) will drop out again. In the next section, we propose a fair protocol for fast resource as-signment. Improvement is achieved with a modification of the RAMA protocol, and a thorough analysis shows this improvement is significant.
3. Random RAMA protocol
In this section we propose a novel method, random RAMA protocol, for fair and fast resource access in fu-ture wireless PCS networks. For simplicity, it is assumed that base stations can detect whether more than one sub-scriber is transmitting orthogonal signals. The detection of multiple users is beyond the scope of this paper. Related work can be found in [1].
Conceptually, the random RAMA protocol can be viewed as a RAMA protocol in which each subscriber has
Base
while any subscriber requests are pending begin{assignment cycle}
broadcasts the begin-auction symbol
repeat{auction cycle}
receives all active orthogonal signals Fi Fmax= max Fi
if more than one requesting subscriber then feedbacks Fmax to the subscribers
else
feedback [ack] to subscribers
until feedback is [ack]
resource assignment for the ‘winner’
end
Figure 2. Random RAMA protocol for a base.
a virtual ID. A virtual ID consists of a priority digit P and a variable number of randomly-generated digits with P as the most significant digit (MSD). The priority digit is used to designate the service priority, and the variable length of the virtual ID is used to uniquely identify the winner. Note that the length of virtual subscriber’s IDs are variably de-pendent on the numbers of requesting subscribers (e.g., call originations or handoffs) in a cell and the randomness char-acteristics of transmitting digits. Each requesting subscriber transmits its virtual ID one digit at a time until it drops out or becomes the winner. The winner is the subscriber with the longest length of virtual ID in an auction cycle. After each auction, the base assigns an available communication resource to the winner. Like RAMA, this cycle is then re-peated for other available resources until either all requests are satisfied or no more resources are available.
Figure 2 depicts how a base uses the random RAMA protocol. Whenever there are available communication re-sources and subscribers requesting, the base broadcasts a
begin-auction symbol to inform the subscribers that a new
auction cycle has begun. In an auction cycle, the base lis-tens to all active orthogonal signals. If there is only one subscriber requesting a communication resource, the base feeds back an acknowledgement symbol [ack] to the sub-scriber, and assigns a resource to the subscriber; otherwise, the base feedbacks the maximum active orthogonal signal to the subscribers. This procedure is repeated until the winner is uniquely identified. Then the base broadcasts a
begin-auction symbol to start the next auction cycle.
Figure 3 depicts the random RAMA protocol for sub-scribers. After receiving the begin-auction symbol, the re-questing subscribers transmit their priority digits in orthog-onal signals simultaneously. If the base feeds back an ac-knowledgement, the auction cycle is completed and the subscriber waits for resource assignment from the base. If the base feeds back any symbol other than its own, the subscriber drops out of further participation in this assign-ment cycle. The remaining subscribers continue in this auc-tion cycle by transmitting a randomly-generated digit, and
Subscriber
while additional resource is required begin
wait for the begin-auction symbol d← P {the priority digit}
repeat{auction cycle}
transmit an orthogonal signal Fd to the base
receive a feedback F from the base
if F = [ack]
then waits for resource assignment else if Fd6= F
then drop out
generate a random number d
until F = [ack] or Fd6= F
end
Figure 3. Random RAMA protocol for subscribers.
Figure 4. Example of random RAMA auction.
reacting according to the feedback. This transmit-reaction process is repeated until one subscriber becomes the unique winner or drops out of further participation in this assign-ment cycle. The winner is then assigned an available re-source by the base, and those dropout subscribers partici-pate in a new auction during the next assignment cycle.
Figure 4 shows an example of a random RAMA auc-tion. In this example, we assume that each subscriber ‘ID’ (either the priority digit P or a random number d) is rep-resented using radix 7 notation (i.e., 06 P , d 6 6, the ac-knowledgement symbol is denoted by 7), and subscribers A, B, C, and D with respective priority digits P = 3, 6, 6, and 6 are seeking for communication resources. In the first auction time slot, all subscribers transmit their priority digits, i.e., subscriber A transmits a ‘3’ by transmitting an F3 FSK signal, and subscribers B, C, and D transmit ‘6’s
by transmitting F6 FSK signals. The base detects these
orthogonal signals, and feeds back the largest digit ‘6’ to the subscribers by transmitting an F6 FSK signal. After
dig-its P lower than ‘6’ drop out of the auction, and wait for the next assignment cycle. Subscribers with priority digit P = 6 continue by transmitting a randomly-generated num-ber d. In this example, subscrinum-bers B, C and D randomly generate and transmit 3, 4, and 6, respectively. After the base announces that ‘6’ was the largest digit transmitted by this group of remaining subscribers, subscribers B and C drop out of the auction, and subscriber D continues by ran-domly generating and transmitting a ‘3’. Finally, the base feeds back the acknowledgement symbol [ack] by trans-mitting, for example, an F7 FSK signal, and broadcasts a
resource assignment to subscriber D, the unique winner of the auction. In the next assignment cycle, subscribers that dropped out of the previous cycle (i.e., subscribers A, B, and C) participate in a new auction along with any new subscribers requesting service.
Unlike RAMA, an auction cycle lasts until a winner is uniquely identified in the random RAMA protocol. A var-iable-length ID in a random RAMA auction yields lower delay, particularly under light loads. When there is only one requesting subscriber, each auction cycle requires only two slots: one slot is the priority digit from the subscriber to the base, and the other is the acknowledgement from the base to the subscriber. In contrast, the entire fixed-length auction cycle must be completed in the RAMA protocol, wasting valuable slot time. Furthermore, the random RAMA proto-col is a fair protoproto-col because subscriber’s IDs are randomly generated here.
4. Performance analysis
In this section, we present preliminary performance re-sults for the random RAMA protocol. In our model, perfor-mance metrics include mean service time and mean waiting time (access delay). In the real world, the amount of avail-able resource units might affect the waiting time of a cus-tomer. However, the wireless bandwidth in future wireless PCS networks is much larger than that in current wireless networks. We assume that the amount of resource units is enough for assignment so that the waiting time in the re-source assignment periods is negligible, i.e., the period of resource assignment is short for handoff or initial access. Moreover, there is no difference in the cost of resource as-signment and message transmission between RAMA and random RAMA.
For fair comparison with RAMA, we focus on the auc-tion cost and model the waiting time as the time in aucauc-tion instead of the waiting time until a customer gets service. In the steady state, the number of requesting subscribers is derived in equation (1) and the service time of a requesting subscriber is equal to the ID length. The time for transmit-ting an M -ary symbol is the time unit in this analysis. Thus mean service time for random RAMA can be considered as the mean ID length. In our study we considered only mo-bile subscribers, and excluded fixed-network subscribers. With the mobile subscribers, two types of requested re-sources, call setups and handoffs, were investigated. To
simplify our model, we did not distinguish between call setups and handoffs. Furthermore, error-free transmission was assumed.
Suppose that there are N mobiles in a cell on average. Let Poc denote the probability of mobiles originating calls
and Phandoff denote the probability of mobile handoffs. In
the steady state, the number of active mobiles n is the sum of the number of mobiles originating calls and that of hand-offs as derived below. The number of mobiles originating calls is equal to N·Poc. A mobile only requests its handoff
to a specific cell and the cell will inform the counterpart of the handover via the wireline network. For example, a mobile requests its handoff to the new base station with the mobile-controlled handover scheme [12]. Let each cell have nc neighboring cells and the direction of handover to each neighboring cell is uniform for a mobile. The number of handover is equal to N·nc·(1/nc)·Phandoff= N·Phandoff.
Thus,
n = N· (Poc+ Phandoff). (1)
In RAMA, each subscriber has a unique ID that consists of 10 decimal digits (a priority digit and a nine-digit phone number). Let M denote the number of orthogonal signals, e.g., M -FSK. Thus the length of the ID (in M -ary symbols) is
L = 10· logM10. (2)
The mean waiting time (access delay) wi (in M -ary
sym-bols) for one mobile when there are i active mobiles is equal to the total waiting time of i mobiles divided by i
wi= 10· logM10 i i X j=2 (j− 1). (3)
Note that the first mobile gets service in time equal to the ID length, the second gets service in two ID lengths, etc., and that this is the source of the total waiting time.
In random RAMA, let P (i, j) denote the probability that j mobiles transmit the relative maximal signal within i ac-tive mobiles, and xidenote the mean service time in M -ary
symbols, i.e., the mean ID length, for one mobile when there are i active mobiles. xiis recursively defined as
cur-rent transmissions plus the time servicing j mobiles that transmit the relevant maximal digit, as shown below:
xi = i X j=1 P (i, j)· (1 + xj), 1 < i6 n, 1, i = 1. (4)
Note that for any positive integer i, Pij=1P (i, j) = 1. Also, P (i, 1) = 0 for i > 1. Then, eliminating the xi term
on the right-hand side of the above equation, we obtain the following equation: xi= 1 + i−1 X j=2 xjP (i, j) 1− P (i, i) , 1 < i6 n, 1, i = 1. (5)
Now we derive the probability P (i, j). LetD represent the relevant maximal digit of transmitted signals (0 6 D 6 M− 1). The probability P (i, j) is equal to the summation of the conditional probabilities of all possible values ofD. Imagine the number of ways to place i balls of the same color in M numbered boxes. This is the total number of ways i mobiles can transmit M -FSK signals, and is equal to M +ii−1. The total number of ways that j mobiles can transmit the relevant maximal digit D within i mobiles is the same as the number of ways i−j balls of the same color can be placed in D numbered boxes. Thus the following equations are derived:
P (i, j) = MX−1 k=0 P (i, j| D = k) = MX−1 k=0 k + i− j − 1 i− j ! M + i− 1 i ! = MX−1 k=1 M− k + i − j − 1 i− j ! M + i− 1 i ! , i6= j, M M + i− 1 i , i = j. (6)
Similar to RAMA, the mean waiting time wi when there
are i active mobiles using random RAMA is computed as follows: wi= 1 i i X j=2 (j− 1)xi. (7)
Numerical results are shown in figures 5 and 6. Figure 5 depicts a comparison of mean ID lengths between RAMA and random RAMA. It shows that the mean ID length of the random RAMA protocol is much shorter than that of the RAMA protocol. The curves also indicate that mean ID length of random RAMA increases slowly (a log-like function) according to the number of active mobiles. The number of active mobiles can be estimated by using equa-tion (1). In the comparison, we considered the number of active mobiles ranging from 0 to 500. Comparison of mean waiting times (access delays) between RAMA and random RAMA is shown in figure 6. It shows that mean delays are also greatly improved by the random RAMA protocol. Additionally, the mean waiting time is almost proportional to the number of active mobiles, i.e., linear growth.
Figure 5. Comparison of mean ID length between RAMA and random RAMA.
Figure 6. Comparison of mean waiting time between RAMA and random RAMA.
5. Dynamic priority assignment
In the random RAMA protocol, communication re-sources are fairly assigned to subscribers according to their service priorities. When communication resources (radio bandwidth) are not sufficient to satisfy the requirements of all active subscribers (applications), the ones with lower priorities may not retain services. This forced termination is inconvenient and sometimes unacceptable to subscribers since it is more harmful to the quality of service (QOS) than initial access blocking. Therefore, it is desirable when al-locating channel capacities to assign priorities dynamically to various applications or subscribers according to their rel-ative urgency.
Assigning priority according to relative urgency, on the one hand, can alleviate the problem of forced termination for low-priority subscribers; on the other hand, it might de-grade the QOS for high-priority subscribers under overload conditions. Fortunately, a graceful degradation of QOS is allowable for voice and multimedia data that can tolerate some loss of information. It is appropriate to transmit these types of data on sub-rating channels [11] or to directly dis-card some frames. In the following, we present a dynamic priority assignment scheme to improve the quality of ser-vice for subscribers under overload conditions.
Typical values of the reliability requirements and tolerance.
Voice Video Image Text
Requirement (r) 0.98 0.90 1.0 1.0
Tolerance (1− r) 0.02 0.10 0.0 0.0
Our scheme is based on the fact that each PCS subscriber must define his or her QOS requirement which is an agree-ment between the PCS subscriber and service provider, in order to obtain services. The QOS information includes ser-vice priority, reliability tolerance, deadline constraints, and other related parameters. The reliability tolerance indicates the maximum percentage of multimedia data that can be dropped when the wireless bandwidth is insufficient. Typi-cal reliability requirement values and various types of data tolerance are listed in table 1 [14]. The deadline constraint specifies the real-time characteristics of voice and multi-media services. It depends on the types of transmitted data and resources in mobile units such as the sizes of buffers. Note that the larger the buffers, the looser the deadline con-straint will be. To satisfy these QOS requirements, each subscriber must measure his or her current reliability tol-erance value and deadline urgency. The current reliability tolerance value is the percentage of multimedia data that have been dropped and the deadline urgency is a count-down value that represents the remaining time to meet the deadline constraint.
In this dynamic scheme, the assigned priority of an ap-plicant (subscriber) is based on service priority, and can be dynamically adjusted according to other QOS metrics such as reliability tolerance and deadline urgency. Let s denote a subscriber’s service priority, and wr and wd represent
the relative weights of reliability tolerance and deadline ur-gency, respectively, to service priority. The priority p of an applicant (subscriber) transmitted by the proposed protocol is dynamically assigned as follows:
p = s +wr·∆r· δ(∆r) + wd· δ(d)
, (8)
where∆r is the difference between current drop percentage and reliability tolerance, d is the value of deadline urgency, bXc represents the largest integer less than or equal to X, and δ is a step function:
δ(x) = (1
d, if x > 0, 0, otherwise.
(9) Note that the value of deadline urgency d is larger than zero. When d = 0, the request will be given up since it fails to meet the deadline. As stated in equation (8), when either the reliability tolerance is violated or the transmission is near deadline, the priority of the subscriber is dynami-cally increased to prevent forced termination under heavy loading. The level of increment is proportional to the level of reliability tolerance violation and inversely proportional to the value of deadline urgency. Note that although the QOS parameters considered in our scheme are reliability
Time t1 t2 t3 t4
Requesting A(5) B(4) C(3) C fails
subscribers B(4) C(3) D(4) to meet
C(3) the deadline
Winner A(5) B(4) D(4)
(a) Static priority assignment
Time t1 t2 t3 t4
Requesting A(5,5,3) B(5,4,2) C(5,3,1) D(5,4,2) subscribers B(4,4,3) C(4,3,2) D(4,4,3)
C(3,3,3)
Winner A(5,5,3) B(5,4,2) C(5,3,1) (b) Dynamic priority assignment Figure 7. Effect of deadline urgency.
requirements and deadline urgency, it is easy to adopt other QOS parameters.
The example shown in figure 7 illustrates the proposed scheme. Because the effect of reliability tolerance is simi-lar to that of deadline urgency, only the effect of deadline urgency is concerned for demonstration. In this example, we assume that the deadline constraint for each request is 3, that is, subscribers should be able to gain the required com-munication resources in 3 auction cycles after requesting. Subscribers A, B, and C request communication resources at time t1and subscriber D requests at time t3. The service
priorities of subscribers A, B, C and D are 5, 4, 3 and 4, respectively.
In a static priority assignment scheme (cf. figure 7(a)), subscribers A, B and D become the winners at times t1, t2
and t3, respectively. Even though subscriber D is not an
urgent subscriber at time t3, he or she gets the
communi-cation resources immediately regardless of the urgency of subscriber C. Thus subscriber C fails to meet the deadline constraint. This scenario shows an example of the QOS degradation which can be alleviated by a dynamic priority assignment scheme.
In a dynamic priority assignment scheme (cf. fig-ure 7(b)), the deadline urgency of a subscriber is an integral part of his/her deadline constraint. Thus subscriber A with service priority s and deadline urgency d can be represented by a triple A(p, s, d), where p is the assigned priority in equation (8). After each auction cycle, the subscriber either becomes the winner or counts down its deadline urgency until it fails to meet the deadline constraints (i.e., it counts down to 0). In this example, we assume wr is equal to 0
and wd is equal to 2. Thus the winners are the same as in
the static assignment scheme at time t1and t2, as shown in
figure 7. At time t3, however, subscriber C counts down
its deadline urgency to one and thus has an assigned pri-ority of p = 3 +b2 · δ(1)/1c = 5, while subscriber D has an assigned priority p = 4 +b2 · δ(3)/3c = 4. Conse-quently, subscriber C becomes the winner at time t3, thus
meeting his/her deadline constraint. Also subscriber D will get his/her communication resources at time t4 without
advantage of dynamic priority assignment for deadline ur-gency. In a similar manner, we can improve the QOS using reliability tolerance and other QOS parameters in dynamic priority assignments under heavy load conditions.
6. Discussion
The issues of fairness, fault-tolerance, and dynamic pri-ority are all critical to resource assignment in wireless PCS. In this section we discuss these design issues and some practical issues for the random RAMA protocol.
• Fairness.
We define that an auction-based protocol is fair if and only if with the protocol each portable with the same pri-ority has the same ID distribution. As a result, RAMA is unfair since each portable has a fixed, but different ID. Conceptually, with Random RAMA, each portable has a random-generated ID. We assume that the random number generator is fair so that each portable with the same priority has the same ID distribution. Therefore, we conclude that Random RAMA is a fair protocol. • Fault-tolerance and error-recovery.
Because wireless networks are prone to error, fault-tolerance and error-recovery are critical when design-ing a wireless multiple access protocol. Fault-tolerance and error-recovery can be easily be achieved in the ran-dom RAMA protocol as follows. In ranran-dom RAMA, if channel errors lead to exclusion of all subscribers (mo-biles) from transmission of their next random IDs, the mobiles will drop out. If channel errors cause multiple mobiles to each think they have received an acknowl-edgement symbol [ack], then the mobiles will wait for resource assignment from the base. In both cases, the base receives no active orthogonal signals from the mo-biles in the subsequent transmission cycle, and thus de-tects these errors. Consequently, the base broadcasts a
begin-auction symbol to initiate another auction cycle.
Besides, the random RAMA protocol is fault tolerant to other kinds of errors because the subscriber ID’s are randomly generated.
• Dynamic priority assignment.
Static priority assignment is inherent in the RAMA pro-tocol since each subscriber has a fixed ID that designates its priority. In contrast, random RAMA adopts dynamic priority assignment. The basic idea of dynamic priority assignment for random RAMA is to use a priority that is dynamically assigned according to the QOS information to replace the random-generated digit. The advantage of dynamic priority assignment is to improve the quality of service for subscriber, especially for the subscriber that transmits multimedia data under overload conditions. We then conclude this section with some practical issues of the random RAMA protocol. First, a random number (digit) can be generated by the system time clock. There is no need to embed a real random number generator (RNG)
in mobile units. For the sake of resource and power con-sumption, we use the system clock to replace the RNG even though random and pseudo-random numbers can be gener-ated easily. This advantage is based on the assumption that the system clocks of mobiles are not synchronized in the last n LSD (Least Significant Digit) digits. On the contrary, RNGs are needed if the system clocks are well synchro-nized. Secondly, the memory requirement is small. Only a one-digit memory is required to store either the priority digit or a random digit. Thirdly, the acknowledgement sym-bol [ack] can be represented by a reserved M -FSK signal, for example, the FM−1 signal. In contrast, it is not
nec-essary to reserve an M -FSK signal for the begin-auction symbol since the begin-auction symbol is not transmitted simultaneously with the feedbacks.
7. Conclusions
To meet the need for fair and fast resource assignment in future wireless applications and services with integrated traffic (e.g., in mobile computing environments), we have proposed an extension of the RAMA protocol. The ran-dom RAMA protocol offers fast and fair access to avail-able communication resources using a randomly-generated virtual ID. Fairness is achieved by the randomness of vir-tual IDs. The variable length of the virvir-tual ID in a random RAMA auction also yields lower delay, particularly un-der light loading. The features of small resource require-ments, inherent fault tolerance, and high performance make the random RAMA protocol attractive for low-cost mobile units. Moreover, the proposed protocol also uses dynamic priority assignment to improve the quality of service for subscribers under overload conditions. Although our dy-namic priority assignment scheme is based on service pri-ority, reliability tolerance, and deadline urgency, adopting other QOS parameters would be quite straightforward.
Acknowledgement
The authors would like to thank Dr. J.R. Cruz for his assistance in preparing this paper.
References
[1] D.-H. Alexandra, J. Holtzman and Z. Zvonar, Multiuser detection for CDMA systems, IEEE Personal Communications (April 1995) 46–58.
[2] N. Amitay, Resource Auction Multiple Access (RAMA): Efficient method for fast resource assignment in decentralized wireless PCS, Electronics Letters 28(8) (1992) 799–801.
[3] N. Amitay, Distributed switching and control with fast resource as-signment/handoff for personal communications systems, IEEE Jour-nal on Selected Areas in Communications 11(6) (1993) 842–849. [4] N. Amitay, Resource Auction Multiple Access (RAMA) in the
cellu-lar environment, IEEE Transactions on Vehicucellu-lar Technology 43(4) (1994) 1101–1110.
[5] D.C. Cox, Personal communications – A viewpoint, IEEE Commu-nications Magazine 128(11) (1990).
[6] D.C. Cox, Wireless personal communications: What is it? IEEE Personal Communications Magazine 2(2) (1995) 20–35.
[7] G.H. Forman and J. Zahorjan, The challenges of mobile computing, IEEE Computer (April 1994) 38–47.
[8] R. Ghai and S. Singh, An architecture and communication proto-col for picocellular networks, IEEE Personal Communication (Third Quarter 1994) 36–46.
[9] D.J. Goodman et al., Packet reservation multiple access for local wireless communications, IEEE Transactions on Communications 37(8) (1989) 885–890.
[10] M.J. Karol and C.H. I, A protocol for fast resource assignment in wireless PCS, IEEE Transactions on Vehicular Technology 43(3) (1994) 727–732.
[11] Y.B. Lin, S. Mohan and A. Noerpel, PCS channel assignment strate-gies for hand-off and initial access, IEEE Personal Communications (Third Quarter 1994) 2–11.
[12] A. Noerpel and Y.B. Lin, Handover management for a PCS network, submitted for publication (1995).
[13] A.S. Tanenbaum, Computer Networks, 2nd ed. (Prentice-Hall, En-glewood Cliffs, NJ, 1988).
[14] M. Woo, N. Prabhu and A. Ghafoor, Dynamic resource allocation for multimedia services in mobile communications environments, IEEE Journal on Selected Areas in Communications 13(5) (1995) 913–922.
Tsan-Pin Wang received the B.S. degree in
ap-plied mathematics and the M.S. degree in com-puter science and information engineering, both from National Chiao Tung University, Taiwan, ROC, in 1990 and 1992, respectively. From 1992 to 1993, he was a system engineer in the R&D Division of Taiwan NEC Ltd. He is currently working toward his Ph.D. degree at National Chiao Tung University. His research interests include PCS networks and mobile computing.
E-mail: tpwang@csie.nctu.edu.tw
Chien-Chao Tseng is currently a professor in the
Department of Computer Science and Information Engineering at National Chiao Tung University, Hsinchu, Taiwan. He received the B.S. degree in industrial engineering from National Tsing-Hua University, Hsinchu, Taiwan, in 1981, and M.S. and Ph.D. in computer science from the South-ern Methodist University, Dallas, Texas, USA, in 1986 and 1989, respectively. His research interests include mobile computing and parallel and distrib-uted processing.
E-mail: cctseng@csie.nctu.edu.tw
Shu-Yuen Hwang was a professor in the
Depart-ment of Computer Science and Information En-gineering, National Chiao Tung University. He received the B.S. and M.S. degrees in electrical engineering from National Taiwan University in 1981 and 1983, respectively, and the Ph.D. de-gree in computer science from the University of Washington in 1989. His current research interests include artificial intelligence, computer simulation and mobile computing.
