IEEE COMMUNICATIONS LETTERS, VOL. 7, NO. 6, JUNE 2003
263
Throughput-Maximizing Timeslot Scheduling for Interactive Satellite Multiclass Services Ki-Dong Lee, Member, IEEE, Ho-Jin Lee, Yong-Hoon Cho, and Deock Gil Oh
Abstract—We develop an efficient method for optimal timeslot scheduling in an interactive satellite multimedia network. We formulate the timeslot assignment problem as a binary integer programming (BIP) problem, where the throughput is maximized, and decompose this BIP problem into two sub-problems. With this decomposition, we promote the computational efficiency in finding the optimal solution of the original BIP problem. Index Terms—DVB-RCS, MF-TDMA, satellite, throughput, timeslot scheduling.
I. INTRODUCTION
D
EVELOPING an interactive satellite multimedia (ISM) network, such as digital video broadcasting (DVB) return channel via satellite (RCS) systems, has become one of hot issues. The DVB-RCS system is a geostationary earth orbit (GEO) satellite interactive network providing multimedia, including Internet traffic service [1], [2]. Worldwide companies and industries are developing broadband interactive satellite systems, and its commercial availability has been announced recently [2]–[4]. In order to accommodate increasing network access demand at the lowest possible cost, it is imperative to achieve high bandwidth utilization. In the return link of DVB-RCS systems, since there is neither a broadcasting effect as in the forward link nor high reuse efficiency as in the present and emerging cellular systems, achieving high capacity with limited available radio resources is an important focus of investigation [5]–[7]. The European Telecommunications Standards Institute (ETSI)’s standard [1] calls for a return link (terminal to hub via satellite) using an MF-TDMA scheme. Thus, we are motivated to develop a practical method for making an optimal timeslot schedule for each superframe in a fixed MF-TDMA return link so that the (weighted) throughput is maximized [5], [8]. Introducing a penalty weight matrix for active terminals and multiple service classes, we formulate the timeslot assignment problem (TAP) as a binary integer programming (BIP) problem. The penalty weight matrix can be dynamically specified according to the quality-of-service condition of each terminal, such as delayed service time, buffer overflow status, and so on. In order to solve the BIP problem with computational efficiency, we use the well-known problem decomposition technique [9]. As a result, the BIP problem is decomposed into two Manuscript received December 14, 2002. The associate editor coordinating the review of this letter and approving it for publication was Dr. C. Douligeris. The authors are with Broadband Wireless Communication Research Department, Radio & Broadcasting Lab, ETRI, Yuseong, Daejeon 305-350, Korea (e-mail:
[email protected]). Digital Object Identifier 10.1109/LCOMM.2003.813997
Fig. 1.
An example of superframe structure in a DVB-RCS return link.
sub-problems, where the optimal assignment amount vector is determined in the first phase (solving the first sub-problem) and a terminal burst time plan (TBTP) is determined in the second phase (solving the second sub-problem). Performance analysis shows that the proposed method provides both solution efficiency and optimality. Thus, we believe that the proposed method can improve the return link throughput in practical ISM systems. II. MATHEMATICAL FORMULATION OF TAP Our objective is to maximize the return link throughput for each superframe. The computational complexity is one of the most important requirements to meet the time constraint. Because the round-trip time in an interactive satellite network is about 500 ms (return up/down links and forward up/down links), it is preferred to have as short scheduling time as possible in order to sensitively reflect fluctuating demands onto each TBTP. A. Return Link Model We consider an interactive satellite multimedia network with one earth station (hub), a GEO satellite, and a number of group terminals called return channel satellite terminals (RCST’s). As shown in Fig. 1, we consider an MF-TDMA model [1], where a superframe, which is defined as a specific time-frequency s MHz in the time-frequency domain, block includes a group of frames, and each frame, a specific time-fre, consists of common signaling channel quency block
1089-7798/03$17.00 © 2003 IEEE
264
IEEE COMMUNICATIONS LETTERS, VOL. 7, NO. 6, JUNE 2003
(CSC) timeslots , acquisition (ACQ) timeslots , synchronization (SYNC) timeslots , . The resources in and traffic (TRF) timeslots our MF-TDMA return link, denoted by set , are defined as available TRF timeslots. B. Input Parameters and Control Variables The scheduler periodically requires updated information such as the set of active RCST’s (denoted by set , where the number ), and the capacity demands of of elements is limited by (denoted by matrix ). active RCST’s during be the set of service classes. A three dimensional Let denotes the timeslot assignment: is unity array of RCST , if timeslot is assigned to service class ’s are binary control (decision) variables zero otherwise. of (TAP). C. Problem Formulation of RCST has a capacity Each service class . The problem of interest is how to allocate the upper bound available resources per superframe to the RCST’s in order to minimize the total penalty. (1)
As shown in (1), the objective of (TAP) means a weighted penalty, where the respective penalty increases proportional by factor if the assignment amount is less to . Constraint (2) implies that than the requested amount the number of TRF timeslots assigned to class of RCST is not greater than the maximum capacity and than . Each class of RCST must be the requested amount assigned a certain amount of capacity greater than or equal to the minimum capacity. Constraint (4) means that every TRF . timeslot cannot be assigned to more than one (TAP) Minimize Subject to (2) (3) (4)
where is a given threshold value denoting the minimal requirement on the fraction of assigned capacity out of requested capacity. For example, consider a case with . It means that at least 50% of the demand must be assigned to the class in RCST . These threshold values will be specified according to service providers’ policies.
Fig. 2. Job flow diagram in the proposed scheduler.
III. SOLUTION METHOD A. Exact Solution Algorithm We develop a two-phase exact solution algorithm (TEXSA) for (TAP) and prove that TEXSA successfully finds the optimal solution. Fig. 2 shows a detailed procedure of timeslot assignment in the proposed scheduler. Fig. 3 presents a detailed procedure of TEXSA. Proposition 1-Optimality: TEXSA finds the optimal solution. Proof: (Feasibility) TEXSA always gives a feasible solu, which is defined as , denotes the number tion. of assigned TRF timeslots to service class of RCST . In Phase I, following Step 2 gives a feasible solution which does not violate constraint (2) and following Step 3 a better feasible solution which does not violate constraint (3). In Phase II, since the timeslot index is increased if timeslot is once assigned to , there is no assignment violating one-to-one mapsome , which shows the meaning of ping from set to set constraint (4). (Optimality) Since the assignment is done in a manner of “highest penalty first served”, there is no better solution [9], [10]. Proposition 2-Computational Efficiency: TEXSA has com, , putational complexities of , and , respectively in Steps 1, 2, 3, and 4.
LEE et al.: THROUGHPUT-MAXIMIZING TIMESLOT SCHEDULING FOR INTERACTIVE SATELLITE MULTICLASS SERVICES
265
TABLE II UPPER BOUND OF ELAPSED TIME (
ms)
Pentium III PC 1.0 GHz. Source code in C++
IV. CONCLUDING REMARKS
Fig. 3. Timeslot assignment procedure in TEXSA.
We developed an efficient method for optimal timeslot scheduling in an interactive satellite multimedia network so that the system (weighted) throughput is maximized. The timeslot assignment problem was formulated as a binary integer programming problem, which has vast numbers of decision variables. We employed a problem decomposition technique so that a remarkable decrease in computational burden might be achieved. Extensive computational results show that the proposed algorithm solves a throughput-maximizing timeslot assignment problem within a short period of time, much shorter than the designed superframe duration. Because of a fast convergence speed to the global optimum, we believe that the proposed optimization approach can be used for throughput performance improvement in practical interactive satellite multimedia networks. ACKNOWLEDGMENT
TABLE I PARAMETER VALUES USED IN OUR EXAMPLE
This work is a result of the Broadband Satellite Access Network (BSAN) system, a DVD-RCS system developed by the Electronics and Telecommunications Research Institute (ETRI), Korea. The constructive effort of all BSAN development members was helpful to the successfull working of this scheduling algorithm in the BSAN system. REFERENCES
In a practical system, and are much less than . For and in an optimally example, we have is less than the average designed superframe pattern [5], and because there are not so many classes. Proposition 2 implies that TEXSA has a linear function of computational complexity on . It is expected that this efficiency is very beneficial to the capacity request and assignment function in interactive satellite communications. B. Computational Results and Discussions We show extensive computational results of our algorithms using randomly generated demand vectors. Table I shows the superframe pattern, the available resources, the number of active RCST’s, and so on. Table II presents the upper bounds of ) and computing times (ms) elapsed in Phase I (finding ), which shows the computational effiPhase II (finding ciency of our method.
[1] Digital Video Broadcasting (DVB); Interaction channel for Satellite Distribution Systems, ETSI EN 301 790 (V1.2.2), 2000. [2] J. Neale, R. Green, and J. Landovskis, “Interactive channel for multimedia satellite networks,” IEEE Commun. Mag., Mar. 2001. [3] Y. H. Cho and H.-J. Lee, “Broadband Satellite Access Network (BSAN) system for interactive multimedia services,” in 2nd ETRI-CRL Joint Conf., Japan, 2001. [4] S, atNews Online. (2001). http://www.satnews.com/stories2/4nov20013.html [Online] [5] K.-D. Lee, Y.-H. Cho, S. J. Lee, and H.-J. Lee, “Optimal design of superframe pattern for DVB-RCS return link,” ETRI J., vol. 24, no. 3, pp. 251–254, 2002. [6] S. H. Kim and S. Kim, “Time slot assignment in a heterogeneous encironment of a SS/TDMA systems,” Int. J. Satellite Commun., vol. 15, pp. 197–203, 1997. [7] T. Lee and S. Park, “An integer programming approach to the time slot assignment problem in SS/TDMA systems with intersatellite links,” Eur. J. Oper. Res., vol. 135, pp. 57–66, 2001. [8] G. Wang and N. Ansari, “Searching for optimal frame patterns in an integrated TDMA communication system using mean field annealing,” IEEE Trans. Neural Net., vol. 9, no. 6, pp. 1292–1300, 1998. [9] K. Murty, Linear and Combinatorial Programming. New York: Wiley, 1976. [10] K.-D. Lee, Y.-H. Cho, H.-J. Lee, and H. Jeong, “Optimal scheduling for timeslot assignment in MF-TDMA broadband satellite communications,” in Proc. IEEE VTC Fall 2002, Vancouver, Canada, 2002, pp. 1560–1564.
