Adaptive Slot Allocation to Control Queueing Delay in
TDMA Wireless Base Station
*MONG-FONGHORNG1,3,YAU-HWANGKUO1,2, JANG-PONGHSU4
ANDREN-HAO CHENG4
1
Center for Research of E-life Digital Technologies 2
Department of Computer Science and Information Engineering National Cheng Kung University
Tainan, 701 Taiwan 3
Department of Computer Science and Information Engineering Shu-Te University
Kaohsiung, 824 Taiwan
4Advanced Multimedia Internet Technology Inc. Tainan, 710 Taiwan
An Adaptive Slot Allocation (ASA) scheme for controlling the queueing delay of packet delivery in a TDMA base station is presented in this paper. ASA utilizes the multi-queue architecture in a base station to support differentiated services for mobile hosts. The services required by each host are divided into quality-guaranteed type and best-effort type, which are served by separate queues. Another mechanism to realize ser-vice differentiation is the scaling factors used to differentially affect the determination of the outgoing rate for each queue in ASA. Also, the queue status is used as another pa-rameter to determine the outgoing rate so that the adaptation of ASA can be achieved. Based on the allocated data rate and real channel quality, ASA further allocates time slots among the mobile hosts to control their packet delay. Such an adaptive slot alloca-tion scheme can track the variaalloca-tion of input traffic and channel capacity adequately but still controls the queueing delay of target packets.
In this analysis, we illustrate how the proposed ASA controls the queue dynamics. Then we investigate the relationship between queue dynamics and queueing delay. As a result, we can conclude that the queueing delay of services for each host is effectively controlled by the parameters of the ASA queuing model. Moreover, the controllability of the service queues relieves the task of buffer management in a base station. A simulation of real streaming traffic traveling across hops is made to evaluate the queue dynamics and delay performance of ASA. The simulation results confirm the expected properties, even under heavy traffic variation.
Keywords: wireless TDMA base station, delay controllability, quality-of-service,
adap-tive slot allocation, queue dynamics
1. INTRODUCTION
Nowadays, development of quality-guaranteed services over the Internet has at-tracted much research interest. The objective of Quality-of-Service (QoS) technology is to offer network services with different service level agreements (SLAs) through a
dif-Received December 25, 2002; revised April 29, 2003; accepted August 25, 2003. Communicated by Yu-Chee Tseng.
Communica-ferentiated service mechanism. The service level agreement results from the negotiation between user and network, in which the quality guarantees include packet delay, packet loss and packet jitter. According to the negotiated SLA, the routers on the delivery path allocate network resources such as bandwidth, time slots and packet buffer to accomplish the guarantee of network quality.
The QoS problem becomes important when multimedia services are involved. Mul-timedia data transmission has some significant features including in-time delivery, higher tolerance to packet loss, more restricted buffer management and low tolerance to packet jitter. These phenomena of multimedia traffic demand a sophisticated QoS mechanism to ensure the availability of resources of various service classes.
The popularity of wireless communication adds new complexity to the QoS problem, where the frequent variation of a wireless link results in the inefficiency of bandwidth control. Two factors cause the variation of wireless links: radio characteristics and user mobility. The channels of a wireless network operate at the Super High Frequency (SHF). Because of the properties of line-of-sight in SHF, channel fading, multipath effect and radio interference [1-3], the channel characteristics depend on the locations of the mobile host and base station, as well as their distance. The change of location caused by host mobility causes further variation of the channel characteristics. Moreover, the link condi-tions of mobile hosts to a base station are not only time varying but also different from each other. Therefore, to support quality-guaranteed wireless multimedia services, we need a more efficient scheduling mechanism of the base station to allocate resources so that the QoS requirements promised to mobile hosts are accomplished as well as possi-ble.
The third-generation code-division multiple access (CDMA) high data rate (HDR) system is the most promising technology for wide-area wireless multimedia networks. In this kind of system, the mobile hosts in the same cell share the same CDMA channel to connect the base station. The downlink from the base station to the hosts adopts time division multiple access (TDMA), where time is divided into fixed-size time slots to serve the hosts in a cell. The base station allocates the time slots for the served mobile hosts to meet their QoS requirements. Andrews et al. [4, 5] proposed a through-put-optimized scheduling approach for a shared wireless link with variable channel con-ditions. Mirhakkak et al. [6] suggested an approach using dynamic adaptation of QoS levels in which applications can specify their QoS needs as ranges rather than the scalar values and tolerate transient periods of degraded service in a wireless network. In con-trolling resource utilization, queue dynamics is often used to compare with the defined threshold so that the targeted traffic conditions are aware. Hahne and Choudhury [7] es-tablished a threshold mechanism with multiple loss priorities to solve the mem-ory-sharing problem in packet switching. The works presented in [8, 9] use a threshold to detect queue state and to allocate the bandwidth of various service classes. Choosing the best threshold value is difficult. Although the concept of dynamic thresholds [7-9] has been introduced, the calculation of thresholds is time-consuming and complicated. In this paper, we focus on the development of an efficient approach for TDMA data scheduling. Instead of using a threshold, we adopt the queue occupancy as the state variable to adapt the bandwidth and slot allocation in the downlink from the base station. The queue status in the base station is used as a feedback to determine the required data rate of mobile hosts so that their queueing delay is under control.
Some frameworks [4, 10-15] have been proposed to realize the QoS objective on wired and wireless networks. In these frameworks, prioritization is the fundamental issue to construct a QoS mechanism and to offer the functionality of service differentiation. Most frames follow the process of classification, queuing and scheduling to realize the prioritization. All received packets are classified according to some pre-defined criteria such as packet source/destination, packet type and data type. A set of packet buffers, called service class queues, keeps the classified packets until the packet scheduler sends them out. Therefore, the architecture of multiple queues becomes one of the fundamen-tals of differentiated networks for handling incoming packets with their desired service quality. However, this architecture raises the issue of how to allocate each queue with an adequate bandwidth, that is, how to determine the delivery order and bandwidth sharing such that the queueing delay and loss of each class queue are guaranteed. The investiga-tion presented in this paper also adopts a multi-queue architecture as the queuing mecha-nism for multimedia services in a TDMA base station, and then develops a packet sched-uler with delay-controllable slot allocation capability.
The remainder of this paper is organized as follows. Section 2 describes the opera-tional model of the proposed delay-controllable slot allocation scheme. Section 3 ana-lyzes the properties of the proposed model and the effects of the operational parameters. Section 4 illustrates and analyzes the simulation results for real multimedia traffic. Fi-nally, the conclusion is given in section 5.
2. MODEL OF PACKET DELAY-CONTROLLABLE SLOT ALLOCATION SCHEME
Delay control is the major target of this investigation. End-to-End (E2E) packet de-lay is defined as the time taken by a packet traveling from the source host to the destina-tion host, and is usually composed of propagadestina-tion delay, processing delay and queuing delay. Propagation delay is caused by the propagation of packet signals over transmission media. Processing delay is the time spent by networking devices (e.g., routers), to deter-mine how to forward packet, for example, packet routing and packet filtering. Queuing delay is the waiting time of packets stored in networking devices before being transmit-ted.
Propagation delay of a packet depends on transmission media and packet length. For example, the propagation delay of packets carried on an optical link is usually faster than on electrical link. Processing delay of packet depends on the computation power of net-working devices and the applied protocol stack. Low efficiency of interpreting the pro-tocol header in a router degrades the processing speed. The length and hop count of delivery path are two other factors affecting propagation and processing delays. But, for a specific delivery route, the propagation and processing delays of each packet are almost invariant. Only the queuing delay heavily depends on the traffic conditions of the travel-ing route.
In the past, Best-Effort (BE) service is the major kind of Internet services, where all packets in a router compete for being delivered in a First-In-First-Out (FIFO) manner. As a result, the queuing delay is not controllable and thus varies. This scheme leads to no guarantee or control on the delivery quality, particularly when a link congestion occurs. Therefore, we need to improve the controllability of queuing delay of packets to help
control of the E2E delay. In this section, we present an adaptive slot allocation (ASA) scheme to well control the queuing delay of packets forwarded by a TDMA wireless base station so that the QoS requirement is satisfied. In a wireless base station, slot allocation becomes complicated because of the varying channel capacity. How to allocate time slot among the mobile hosts with different and time-variant channel capacity is the focus of our investigation.
2.1 Queue Architecture of ASA Scheme
Basically, the proposed ASA scheme adopts a two-class queuing mechanism in the base station for each mobile host, where the two queues are denoted as the Guaran-teed-Service (GS) class and Best-Effort (BE) class. During delivery, the GS queue has higher priority than the BE queue. Because each mobile host may have a different link condition and QoS requirement, the base station constructs individual GS for each regis-tered mobile host. However, all mobile hosts share the same BE queue. The BE class queue will not be served until all GS queues have been served. In summary, ASA applies a multiclass queuing architecture as shown in Fig. 1.
q1k q2k λ1k λ2k N r1k r2k Slot scheduler Q1 Q2 qm+1k λmk r k m+1 Qm+1
Adaptive Slot Allocation
n1k
n2k
nmk
TDMA wireless Base Station qmk Qm λk m+1 Guaranteed Service rmk Best-Effort Service Packet Classifier Input Traffic
Fig. 1. The multiclass queueing architecture utilized by the ASA scheme.
2.2 Operational Model of ASA Scheme
Since multiclass queuing is the basic component part of differentiated service for realizing QoS-enabled networks, the slot allocation of each class queue directly impacts the scheduling performance and the delivery quality in a base station. The packet sched-uler serves all class queues in a round robin way. In each service round, a four-phase procedure is performed to realize the ASA-based scheduling for the GS queues:
Phase I: Allocate the data rate of each GS queue Qi at the kth round according to k k i i i q r T ω = (1) where rik, , ωi qik and T are the allocated data rate, scaling factor and queue length of Qi
at the kth round and service round time, respectively.
Phase II: Calculate the required number of time slots of each GS queue Qi for the
allo-cated data rate by considering the link capacity:
if 0 0 otherwise k k i i k k i i r T C n C t > = (2) wherenik, k i
C and t are the required number of time slots of Qi, the link capacity between
the mobile host corresponding to Qi and base station, as well as the period of time slots,
respectively. If 1 m k i i n N = ≤
∑
, where N denotes the total number of available time slots (=T/t), go to Phase IV directly. Otherwise, go to next phase.
Phase III: Reallocate the time slots to each GS queue as follows:
/* N : total number of time slots
/* Nunused : the number of unused time slots
/* ni : the number of time slots required by Qi
/*ni: the number of time slot actually allocated for Qi
/* m : number of GS queues sort {ni} in an increasing order
reset variableni Nunused = N for (j = 1; j <= m; j++) { if Nunusd> (m + j − 1)*Min({ni, j ≤ i ≤ m}) { ni = +ni Min n({ , })i j≤ ≤i m for j ≤ i ≤ m
Nunusd = Nunusd− (m + j − 1)*Min({ni, j ≤ i ≤ m})
} else { ( 1) unused i i N n n m j = + − + Nunused = 0 } }
Phase IV: Deliver the packets in GS queues according to the result of time slot
alloca-tion in Phase II or Phase III.
In the procedure described above, the ASA approach uses queue dynamics and a set of scaling factors to control the delivery rate of each queue. The scaling factor is adopted to reflect the service level agreement. For example, a larger scaling factor will be speci-fied when better delay performance is required. Then, the number of required time slots is obtained from the factors of allocated data rate, service round time, slot time period and channel capacity. If the number of available time slots is less than the total allocated time slots, the time slots are reallocated to meet two objectives: guarantee of minimum bandwidth and fair bandwidth sharing. Thus, in Phase III, the accumulated time slots of all queues grow together from zero at the same pace. The procedure progressively fills the time slot requirements of queues and is terminated when the unused slots are ex-hausted. This filling procedure obviously offers fair slot allocation for all queues and ensures the queues allocated with a minimum number of time slots are optimal.
In ASA, the scaling factorω ∈i (0, 1]associated with each queue is a user-supplied operational parameter to control the queue dynamics. If ωi is equal to zero, the associated
queue is prohibited from delivering data. On the contrary, the queue with ωi = 1 will be
served with a full delivery of data in each service round. Between the two extremes, ac-cording to Eq. (1), the delivered byte volume in a service round is proportional to the occupancy of each queue. The queue occupancy is related to the waiting time spent by the packets in a queue. Thus, for any queue, if the occupancy is regulated through the adaptation of bandwidth allocation, the queueing delay of each queue would be well con-trolled. The detailed analysis of the relationship among queue occupancy, bandwidth allocation and queueing delay will be illustrated in the following sections. It is noted that an empty queue shares no bandwidth or time slot. ASA suspends the bandwidth alloca-tion for an inactive connecalloca-tion, so that the utilizaalloca-tion of bandwidth is greatly improved in comparison with schedulers using constant bandwidth reservation.
3. THEORETICAL ANALYSIS OF ASA SCHEME
To perform qualitative and quantitative analyses, we assume that ASA is applied to the TDMA base station under the following conditions:
1. The channel capacity of connecting each mobile host is available to the base station in each round.
2. During a service round, the channel capacity remains constant. 3. The computation time needed for slot allocation is ignored.
4. The overhead time taken in service switching between GS queues (that is, mobile hosts) is ignored.
5. The packets in a queue will not be lost except for the failure of the channel to mobile host; in this case, the queue is removed.
6. The admission control is applied before establishing a connection to ensure the suffi-ciency of network resources at the base station.
Based on the assumptions mentioned above, we first analyze the queue dynamics and convergence of ASA under normal conditions, in which the time slots of each ser-vice round are sufficient to support the required time slots of all queues in the TDMA base station. In normal operation of ASA, we will show that the queue occupancy is ex-plicitly formulated as a function of the scaling factor, service round time and input traffic rate. Second, the convergence of bandwidth allocation of ASA is analyzed. Both random traffic and constant traffic are of concern. Finally, the controllability of queueing delay in ASA is proven. When the queue is at the steady state, the scaling factor and service round time concisely control the delay of each packet in the cases of constant input traf-fic or slightly varying traftraf-fic. Even when the input traftraf-fic is regarded as random, the av-erage delay of packets is guaranteed by statistics. In summary, the mean of the queueing delay is independent of the queue occupancy when the queue is at the steady state.
3.1 Queue Dynamics and Convergence of ASA
Based on the fluid model [16], the queue dynamics of ASA is modeled as 1
( )
k k k k
i i i i
q + =q + λ −r T (3) where λ is the data arrival rate of Qik i at the end of the kth service round. Let us start the
performance analysis from the queue state. The convergence property of ASA is stated as follows:
Theorem 1 In ASA, for a given service round time T, scaling factor ϖi and input rate λi,
the state of each queue Qi is obtained by
1 1 (1 ) k k m k m i i i m q ω −Tλ − = =
∑
− (4)Proof: Applying ASA bandwidth allocation formula of Eqs. (1) to (3), we have
1 ( ) k k k k i i i i i q q q T T ω λ + = + − (5)
Z-transforming Eq. (5) yields ( ) ( ) ( 1 ) i i i z T q z z λ ω = − + (6)
where qi(z) and λi(z) are the z-transforms of q and ik k i
λ , respectively. To obtain the time
sequence of q , we apply the inverse z-transform on Eq. (6) and ik
1 1 ( ) [ ( )] [ ] ( 1 ) k i i i i z T q Z q z Z z λ ω − − = = − + (7)
1 2 2 3 1 0 1 1 (1 ) (1 ) ... (1 ) (1 ) k k k k k i i i i i i i i k m k m i i m q T T T T T λ ω λ ω λ ω λ ω λ − − − − − − = = + − + − + + − =
∑
− (8)Eq. (4) illustrates the dynamics of queue Qi in ASA, which is related to the input
rate, service round time and its scaling factor.Tλk m− is the number of bytes received at the (k − m)th round. Therefore, the queue state of k
i
q is regarded as the weighted average
of the input traffic rate at all past rounds. The term (1 − ωi)m−1 converges to zero for any
ω ∈ (0, 1]. Moreover, the convergence of k
i
q depends on the value of ω; the queue state
depends on shorter traffic history when ωi is close to 1. On the other hand, as ω decreases
to near zero, the queue state depends on a longer range of traffic history.
Next, we discuss the convergence of the queue state in ASA, for which a queue reaches and stays within a specific zone around the final state. Such a zone is specified by a percentage value of the final state to indicate the tolerance of a steady state.
Corollary 1 In ASA, if the input rate of queue Qi isconstant, then it will maintain a
steady state whose value is given by
ss ss i i i T q λ ω = (9)
Proof: According to Eq. (4), when a queue Qi has a constant input rate,λiss, the queue
state can be written as
2 1 (1 (1 ) (1 ) ... (1 ) ) (1 (1 ) ) 1 (1 ) (1 (1 ) ) k ss k i i i i i ss k i i i ss k i i i q T T T λ ω ω ω λ ω ω λ ω ω − = + − + − + + − − − = − − − − = (10)
Moreover, since ωi∈ (0, 1], the term (1 − (1 − ωi)k) approaches 1 as k increases with time.
Consequently, we have 1
/
k k ss
i i i i
q + =q =Tλ ω when qi exponentially approaches the steady
state. That is, the queue Qi reaches a steady state whose value is determined by the input
rate, service round time and scaling factor. When Qi has the scaling factor ωi = 1, its
queue state reaches a steady state of Tλikin a service round. From Corollary 1, it is obvious that ASA has some advantages in buffer manage-ment since we just need to control two coefficients. If the input traffic is generalized to a random process with varying rate, ASA also exhibits good statistical properties.
Corollary 2 In ASA, if the input rate
λ
iis a random process with a distribution func-tion Pλ and its mean is finite, then the average queue state of Qi with ωi = 1 is equalProof: From the generic dynamics model of class queues given by Eq. (4), to derive the
mean of the queue state Qi, we apply the mean operation to both sides of Eq. (4) to yield
1 1 [ ] [ ] [ (1 ) ] k k m k m i i i i m E q E q E ω −Tλ − = = =
∑
− (11) In Eq. (11), except for the input rate, both the service round time and scaling factor are constants. Thus, if the mean input rate is finite, we finally have the approximation:1 1 1 1 [ ] [ (1 ) ] ( ) (1 ) 1 (1 ) ( ) k m k m i i i m k m i i m k i i i E q E T E T E T ω λ λ ω ω λ ω − − = − = = − = − − − =
∑
∑
(12) Obviously, when the scaling factor is equal to one, the queue state arrives exactly at [ ] ,iEλ T regardless of the service round index, k. Otherwise, the queue state of Qi
expo-nentially approaches the steady state,E[ ]λi T ω given by Eq. (12). i, Therefore, in the case of random traffic, ASA still offers a guarantee of queue dy-namics controlled by the service round time and the scaling factor. In other words, ASA make buffer management easy. In Theorem 1 and associated corollaries, the queue dy-namics and steady states in various conditions are specified. In the following, we analyze the convergence time from initial queue state to the steady state. When the traffic rate is a random process, because of the traffic locality, the traffic can be approximated as lots of segments of piece-wise constant traffic. Usually, the duration of each segment is larger than the service round time of ASA. If the input rate is a piece-wise constant, k
i
λ, in the short interval (t0, t0 + τ), combining Eqs. (1) and (4), we have
(1 (1 ) ) k k k i i i r = − −ω λ for all 0 0 [ t , t ] T T k∈ +τ (13)
By the allocation of outgoing data rate according to Eq. (13), the queue Qi converges
exponentially to the steady state in the duration τ. According to Eq. (10), the conver-gence of Qi to steady state is determined by ωi and T. Table 1 shows the required time of
ASA converging to the steady state under different ωi. For the cases of ωi = 0.5, the
Table 1. Convergence time of ASA for different scaling factors and state tolerances.
Converge to steady state ω = 0.01 ω = 0.1 ω = 0.2 ω = 0.5 ω = 1
Within 1% 418T 41T 20T 8T 1T
5. CONCLUSION AND DISCUSSION
In this paper, we propose a new scheme of adaptive slot allocation, the ASA scheme, to offer QoS-based differentiated services on datagram networks. According to the op-erational model presented in this paper, ASA utilizes an adaptive time slot allocation mechanism. The adaptation is realized by using the queue state as feedback to track the variation of input traffic. A scaling factor is adopted in each GS queue to reflect the ser-vice level agreement that the mobile host has. Therefore, ASA has the functions of input traffic tracking and differentiated services. When the base station has sufficient slots to meet all requirements of GS queues, the queueing delay and the queue occupancy are well controlled by ASA. If time slots are insufficient for all GS queues, ASA serves the queues on the principle of maximizing the satisfaction of the minimum slot requirement among all mobile hosts. In this condition, ASA also exhibits good robustness. In sum-mary, ASA features the properties of delay/queue controllability, robustness and differ-entiation.
The above-mentioned properties have been analyzed theoretically in this paper. The relationships between the operational parameters and the QoS indices are derived. The analyses show that the queue occupancy and queueing delay are effectively controlled by the scaling factor and service round time.
Finally, we perform real case simulations with real video streaming traffic caused by VoD service provided by an ISP to demonstrate the real performance of ASA in the con-dition of highly varying traffic. The simulation shows that ASA rapidly and precisely tracks the variation of the input traffic so that the results are well-controlled queue dy-namic and strongly guaranteed packet delay. The well-controlled queue dydy-namics further eases the buffer management and reduces the possibility of packet loss.
REFERENCES
1. B. A. Forouzan, Data Communications and Networking, McGraw-Hill, Singapore, 2001.
2. A. S. Tanenbaum, Computer Networks, Prentice Hall, New Jersey, 1996.
3. W. Turin, R. Jana, C. Martin, and J. Winters, “Modeling wireless channel fading,” in
Proceedings of International Conference on Vehicular Technology, 2001, pp.
1740-1744.
4. R. Braden, L. Zhang, S. Berson, S. Herzog, and S. Jamin, “Resource reservation pro-tocol (RSVP) version 1 functional specification,” RFC 2205, 1997.
5. S. Blake et al., “An architecture for differentiated services,” RFC 2475, 1998. 6. M. Andrews et al., “Providing quality of service over a shared wireless link,” IEEE
Communication Magazine, Vol. 39, 2001, pp. 150-154.
7. S. Kapp, “802.11: leaving the wire behind,” IEEE Internet Computing, Vol. 6, 2002, pp. 82-85.
8. IEEE 802.11 WG, Draft Supplement to STANDARD FOR Telecommunications and Information Exchange Between Systems -LAN/MAN Specific Requirements - Part 11:
Wireless Medium Access Control (MAC) and physical layer (PHY) specifications:
802.11e/D2.0, 2001.
9. P. Bender, P. Black, M. Grob, R. Padovani, N. Sindhushayana, and A. Viterbi, “CDMA/HDR: a bandwidth-efficient high-speed wireless data service for nomadic users,” IEEE Communication Magazine, Vol. 39, 2001, pp. 70-77.
10. S. Mangold, S. Choi, P. May, O. Klein, G. Hiertz, and L. Stibor, “IEEE 802.11e wireless LAN for quality of service,” in Proceedings of International Conference on
European Wireless (EW ’2002), 2002, pp. 32-39.
11. M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, R. Vijayakumar, and P. Whiting, “CDMA data QoS scheduling on the forward link with variable channel condition,” Bell Labs Technical Memorandum, 2000.
12. M. Mirhakkak, N. Schult, and D. Thomson, “Dynamic bandwidth management and adaptive application for a variable bandwidth wireless environment,” IEEE Journal
on Selected Areas Communications, Vol. 19, 2001, pp. 1984-1997.
13. E. L. Hahne and A. K. Choudhury , “Dynamic queue length thresholds for multiple loss priorities,” IEEE/ACM Translations on Networking, Vol. 10, 2002, pp. 368-380. 14. C. G. Park, D. H. Han, and Y. Lee, “Performance analysis of threshold based band-width allocation scheme for IP traffic on ATM networks,” IEE Proceeding
Commu-nication, Vol. 149, 2002, pp. 29-33.
15. H. H. Yoon, H. Kim, C. Oh, and K. Kim, “A queue length-based scheduling scheme in ATM networks,” in Proceedings of IEEE Region 10 International Conference on
Electrical and Electronic Technology, Vol. 1, 1999, pp. 234-237.
16. J. Filipiak, Modeling and Control of Dynamic Flows in Communication Networks, Springer-Verlag, Heidelberg, 1988.
17. L. Kleinrock, Queuing System, Wiley, New York, 1975.
18. P. Ji, B. Liu, D. Towsley, and J. Kurose, “Modeling frame-level errors in GSM chan-nels,” in Proceedingsof IEEE Global Telecommunications Conference, Vol. 3, 2002, pp. 2483-2487.
Mong-Fong Horng (洪盟峰) received the B.S and M.S. degrees in Control Engineering from National Chiao Tung Uni-versity, Hsinchu, Taiwan, in 1989 and 1991, respectively, and the Ph.D. degree in Computer Science and Information Engineering from National Cheng Kung University in 2003. From 1993 to 2003, he was with the Department of Electrical Engineering, Kao Yuan Institute of Technology, Kaohsiung, Taiwan. From 1999 to 2001, he was the vice chairman of the Computer Center at Kao Yuan Institute of Technology. Since 2003, he has been an Asso-ciate Professor with the Department of Computer Science and Information Engineering, Shu-Te University, Kaohsiung, Taiwan. His research interests include network QoS, broadband communication, and Artificial Intelligence.
Yau-Hwang Kuo (郭耀煌) was born in Tainan, Taiwan in 1959. He received M.S. and Ph.D. degrees in Computer Engi-neering from National Cheng Kung University in 1984 and 1988. He was the President of Taiwanese AI Association from 1999 to 2000, the Director of Research Center for Computer System Technology from 1997 to 2000, and the Managing Director of Chinese Fuzzy System Association from 1996-2000. He is cur-rently Professor with the Department of Computer Science and Information Engineering, National Cheng Kung University. He is also the Director of Center for Research of E-life Digital Tech-nology, and the coordinator of Computer Science & Information Engineering Program of National Science Council, R.O.C. His research interests include intelligent computing, knowledge management, broadband communication, information retrieval, pattern rec-ognition and VLSI design.
Jang-Pong Hsu (許振鵬) received the M.S. and Ph.D. de-grees in Electrical Engineering and Information Engineering from National Cheng Kung University in 1987 and 1998, respectively. Since 2002, he has been the director of RD department at Ad-vance Multimedia Internet Technology Inc. in Tainan, Taiwan, R.O.C. His research interests include pattern recognition, fuzzy neural network systems, intrusion detection systems, and virtual private network.
Ren-Hao Cheng (鄭人豪) received the M.S. degree in Computer Science and Information Engineering from National Cheng Kung University in 2003. Since 2003, he has been a RD engineer at Advance Multimedia Internet Technology Inc. in Tainan, Taiwan, R.O.C. His research interests include wireless networks, packet scheduling algorithms, Quality of Service and VoIP systems.
