Improving Efficiency of Timeslot Assignment for Non-realtime Data in a DVB-RCS Return Link: Modeling and Algorithm

Transcription

1 Iproving Eiciency o Tieslot Assignent or Non-realtie Data in a DVB-RCS Return Link: Modeling and Algorith Ki-Dong Lee, Yong-Hoon Cho, Ho-Jin Lee, and Deock-Gil Oh This paper presents a dynaic resource allocation algorith with ulti-requency tie-division ultiple access or the return link o interactive satellite ultiedia networks such as digital video broadcasting return channel via satellite systes. The proposed tieslot assignent algorith, called the very eicient dynaic tieslot assignent (VEDTA) algorith, gives an optial assignent plan within a very short period. The optiality and coputational eiciency o this algorith deonstrate that it will be useul in ield applications. Manuscript received Sept. 11, 2002; revised Apr. 8, Ki-Dong Lee (phone: , eail: kdlee@etri.re.kr), Yong-Hoon Cho (eail: ladder@coesta.co), Ho-Jin Lee (eail: hlee@etri.re.kr) and Deock-Gil Oh (eail: dgoh@etri.re.kr) are with Radio & Broadcasting Research Laboratory, ETRI, Daeeon, Korea. I. Introduction The digital video broadcasting (DVB) return channel via satellite (RCS) syste is a geostationary earth orbit (GEO) satellite interactive network that provides ultiedia services, including Internet traic service [1], [2]. Copanies and industries worldwide are developing broadband interactive satellite systes [2]-[4]. Recently, they have becoe coercially available [4] and network access deand is expected to increase. Thereore, strategies or achieving axiu use and iniu cost o the liited available radio resources have taken on increased iportance. Figure 1 shows a coniguration o a DVB-RCS syste called the Broadband Satellite Access Network (BSAN) syste developed by the Electronics and Telecounications Research Institute in In the igure, two kinds o satellite links are shown: the orward link and the return link. Satellite terinals attept to log on and send capacity request essages and data via the return link at a transission rate up to 4 Msps (ega sybols per second) whereas they are authoried by the hub and assigned additional aount o capacity upon their respective capacity requests, and they receive individual data and/or ulticast data ro the Internet and/or other servers via the orward link at a rate up to 45 Msps. Because the available radio resources are very liited, one o the ost iportant probles is to iniie the scheduling tie and axiie the radio link throughout [4], [5], [8], [9]. For the return link in DVB-RCS systes, since there is ETRI Journal, Volue 25, Nuber 4, August 2003 Ki-Dong Lee et al. 211

2 RCST Satellite Terinal Forward Link Return Link Ka/Ku band HUB Tieslot Scheduling RCST Server Realtie Resource Manager (RRM) Tx Data Processor Router Internet Rx Data Processor MF-TDMA Deodulator Rx IF Module (RIFM) Fig. 1. An exaple o a DVB-RCS syste coniguration: the Broadband Satellite Access Network (BSAN) syste developed by the ETRI, Korea ( neither a broadcasting eect as in the orward link nor high reuse eiciency as in the present and eerging cellular systes, achieving high capacity with liited available radio resources is an iportant ocus o investigation. The European Telecounications Standards Institute (ETSI) s standard [1] calls or a return link using a ulti-requency tie-division ultiple access (MF-TDMA) schee. Thus, we are otivated to ind an optial tieslot schedule or each superrae in a ixed MF-TDMA return link so that the return link throughput is axiied [6], [7]. Introducing an econoic concept-based penalty weight vector, we orulate the tieslot assignent proble as a binary integer-prograing proble with a vast nuber o decision variables (e.g., ore than 8,000,000 binary integer variables or the superrae pattern presented in [6]). To solve this proble with coputational eiciency, we decopose the original binary integer-prograing proble into two sub-probles, where the optial assignent aount vector is deterined in the irst phase and a terinal burst tie plan (TBTP) is deterined in the second phase. Experiental results show that our ethod perors very well, i.e., it is very eicient. The optiality o the solution obtained by our proposed very eicient dynaic tieslot assignent (VEDTA) algorith is shown in the appendix. Our experiental and theoretical results deonstrate that this ethod succesully provides both solution eiciency and optiality. Thanks to the eiciency, which is suitable or interactive satellite networks, and the solution optiality, we believe that our ethod could be used to iprove data throughput in the practical developent o an interactive satellite ultiedia network. II. Model and Proble Description 1. Syste Model e consider an interactive broadband satellite access network with one earth station (the Hub), a GEO satellite, and a nuber o iobile group terinals called return channel satellite terinals (RCSTs) (Fig. 1). The ultiple access schee in the return link (RCST to the Hub via satellite) is based on MF-TDMA [1]. The radio resources allocated to the return link are shared by ultiple RCSTs. Let R be the set o logon RCSTs. The RCST R sends capacity request essage(s) to the realtie resource anager (RRM). Upon receiving the essages, the RRM generates a TBTP table and sends it to the RCSTs. Upon receiving the TBTP table, each RCST reads the table to deterine what tieslots are assigned. This procedure is executed every superrae. Figure 2 depicts a uniied capacity request and allocation procedure in an interactive satellite network using bandwidthon-deand. Figure 3 shows an exaple o a superrae structure in our MF-TDMA odel [6]. e consider an MF-TDMA odel in which a superrae, which is deined as a speciic tie- 212 Ki-Dong Lee et al. ETRI Journal, Volue 25, Nuber 4, August 2003

3 RCST CR data 125s GEO satellite TBTP Hub public network Fig. 2. Capacity request (CR) and allocation procedure in an interactive satellite access network. assignent (i.e., one tie assignent, repeating assignent, or assignent release); the Start_Slot ield gives the nuber o the irst tieslot in the block; and the Assignent_Count ield gives one less than the nuber o tieslots assigned in the block. There are two kinds o schees: ixed-slot MF-TDMA and dynaic-slot MF-TDMA [1], [2]. A dynaic rae pattern design as presented in [7] belongs to the category o a dynaic-slot TDMA. Because this schee requires a highly coplex coputation to solve an optial rae pattern, any heuristic algoriths, such as the ean ield annealing algorith [7] and siulated annealing algorith, have been proposed. In this paper, however, we develop an algorith or a ixed-slot MF-TDMA schee. This schee can be ipleented ore siply and has a sipler coputational coplexity. t requency T T superrae... rae... CSC ACQ SYNC TRF... TRF T csc T acq T sync n tr. T tr tie 2. Proble Deinition e consider a penalty weight vector v = ( v1, Λ, v R ) or the logon RCSTs to relect the grade o service o each RCST in our optial scheduling. Penalty weights are widely used in atheatical orulation [8]. The penalty weights are deterined by various actors, such as average waiting tie and average raction o packet loss. For a given TBTP table, the total penalty can be calculated with this penalty vector. Thus, the obective can be suaried as iniiing the total penalty. For exaple, i the penalty vector were a unction o the average waiting tie o packets, then the purpose o our scheduling would be expressed to iniie the average waiting tie. According to the unctional relation in deining the penalty vector, the obective will have dierent eaning. CSC: coon signaling channel SYNC: synchroniation ACQ: acquisition TRF: traic Fig. 3. An exaple o a superrae structure. requency block T ( µ s MH) in the tie-requency doain, includes a group o raes less than M. A rae, which is deined as a speciic tie-requency block T, consists o coon signaling channel tieslots ( T csc t ), acquisition tieslots ( T acq t ), synchroniation tieslots ( T sync t ), and traic tieslots ( T tr t ). An RRM, which is a subodule o the Hub, is responsible or traic (TRF) tieslot scheduling o the return link. The RRM generates a TBTP table ro the scheduling result. A TBTP table contains such inoration ields as Logon_ID, Assignent_Type, Start_Slot, and Assignent_Count. The Logon_ID ield shows the identiier assigned to the terinal at logon tie; the Assignent_Type ield deines the repetitive nature o the 3. Proposed Strategy Figure 4 shows a tiing diagra o the proposed TBTP table generation procedure during a superrae, and Fig. 5 CR sg CR sg TSF TBTP 0 t1 t2 t3 t4 t5 TSF RCST: return channel via satellite terinal RRM: realtie resource anager RCST RRM Tie Fig. 4. A diagra o the terinal burst tie plan (TBTP) table generation. ETRI Journal, Volue 25, Nuber 4, August 2003 Ki-Dong Lee et al. 213

4 Store R (at tie-t SF ) scheduler Build D [t 1 -T SF,t 1 ) Find [t 1, t 2 ) Build TBTP [t 2, t 3 ) Packetie & Send TBTP Sub-proble1 (phase I) Sub-proble2 (phase II) Fig. 5. The terinal burst tie plan (TBTP) table generation procedure. shows the proposed TBTP table generation procedure, including the tieslot scheduling. The RCST R sends a capacity request essage to the RRM anytie it is necessary. The RRM accuulates the respective deands or the RCSTs R or the superrae duration T and builds a deand vector D during [ t0 T, t0 ), where t 0 = 0 or the sake o convenience. hen the RRM receives the last TRF tieslot o the current superrae (at tie t 0 ), then the scheduler starts inding the optial assignent aount vector, where eleent denotes the assigned aount o TRF tieslots to the RCST R. The optial assignent aount vector ust be ound by tie t 2. Tieslot scheduling is executed (TBTP table generation) with this vector by tie t 3. During the interval [ t 3, t4), the TBTP is packetied in a transportstrea packet orat. Then the packets are transitted with a packet identiier value to the RCSTs via a satellite. I an RCST receives the TBTP table (at tie t 5 ), it analyes the received TBTP table and waits or the tieslot(s) assigned to it. I the tie interval [ t5, T + t0) is too short or an RCST to read the schedule, the schedule cannot be used in the next superrae. To provide suicient tie or RCSTs R to read the schedule encoded in the received TBTP table, we ust reduce the scheduling tie, i.e., t, ). [ 1 t3 III. Proble Forulation 1. Input Paraeters The resource allocation scheduler periodically requires updated inoration: the nuber o logon RCSTs (denoted by set R ), the nuber o active return link deodulators (denoted by set A ), and the capacity deands o logon RCSTs (denoted by vector D = D, Λ, D ). The nuber o ( 1 R logon RCSTs ust be reported to the scheduler every superrae, and the status o each return link deodulator ust be checked beore aking every schedule. In addition, the capacity deand o each RCST is accuulated during a period equal to superrae duration T. 2. Decision Variables Our obective is to ind an optial layout o tieslot assignent in a superrae. The scheduling period is thus T. In order to denote the assignent o each tieslot, we introduce a decision variable atrix x = [ x i ], where x i is unity i tieslot i is assigned to RCST, and ero otherwise. For exaple, the nuber o binary decision variables is given by R A S and its upper bound is equal to 8,388,608 i we have 2,048 tieslots per rae, 32 available raes per superrae, and 128 logon RCSTs. 3. Proble Forulation Reerence [1] recoended ive types o capacity requests. e consider continuous rate assignent and volue-based dynaic capacity, in which an RCST requests the volue units o payload sie * scaling actor. The rate-based dynaic capacity, in which an RCST requests a bit rate in units o 2 kbps * scaling actor, is siply tranored to a certain capacity as i it had been requested as volue-based dynaic capacity. Free capacity assignent is no ore than a proble in which ree TRF tieslots, i there are any, are assigned. ith a given sybol rate, a rate-based dynaic capacity request could be tranored into a volue-based dynaic capacity request with a capacity equivalent to the resource request. The total capacity deand o the respective types o capacity requests [1] generated by an RCST can be tranored into a certain nuber denoting the nuber o TRF tieslots required or its data transission. This nuber or each RCST is the input paraeter o our capacity assignent proble. Thus, in our capacity assignent proble, the total capacity deand or each RCST (in tieslots) is suicient inoration but the respective deands or the ive types are not required. I an RCST requests ore capacity than the capacity available, a portion o the requested capacity will not be aditted and the residual capacity ust wait to be aditted. However, i the capacity request essages o the logon RCSTs are dependent upon the buer state (the sie o packets waiting or transission), it is not necessary or the RRM to know the aount o deand which is not satiied every superrae. The RRM only has to know the deand generated or the previous superrae duration. Let Y be the inial capacity that ust be assigned to a 214 Ki-Dong Lee et al. ETRI Journal, Volue 25, Nuber 4, August 2003

5 logon RCST. In our practical syste, Y is used or assigning a certain capacity to a continuous rate assignent request with a higher priority. Let Q be the axial capacity that can be assigned to a logon RCST. The proble o interest is how to allocate the available resources per superrae to the RCSTs in order to iniie the total penalty. (CAP) g (x (1 xi ) (1) R Miniie ) = v subect to constraints x i x i x i R x i in { Q, X + Y }, R, Y, R, 1, i S, A, {0,1}, where v s are positive. As shown in (1), the obective o the capacity allocation proble (CAP) calls or a weighted penalty when a certain aount o capacity is not assigned, where each RCST ay have a dierent value o penalty weight v according to several actors, such as the registered service grade (e.g., a low priority service class or a high priority service class). Since n Ai S n x denotes the nuber o TRF tieslots assigned i to RCST, and n A i S ( 1 xi ) denotes the nuber o n TRF tieslots that are not assigned to RCST, thus, v n A i S ( 1 xi ) denotes the penalty caused by the n aount not assigned to RCST, and (1) denotes the su o the respective penalties. Constraint (2) iplies that the nuber o TRF tieslots assigned to RCST is not greater than the axiu capacity and the su o the requested capacity and the iniu capacity. The nuber o TRF tieslots assigned to RCST is greater than or equal to the iniu capacity. Constraint (4) eans that no TRF tieslot can be assigned to ore than one RCST. Since we can separate the constant ter ro the obective o (CAP) as ollows: (2) (3) (4) a constant ter g 1 ( x) ( x) v S = n v xi, R (5) R g (CAP) is reorulated as (CAP ). (CAP ) x x (6) 1 i R Maxiie g ( ) = v subect to the sae constraints as (CAP). IV. Solution Method Table 1 shows the nuber o binary decision variables or each case o paraeter values. A vast nuber o binary decision variables ay give rise to a very long coputation tie, causing ineiciency o tieslot scheduling. Thus, it is necessary to investigate a siple and eicient ethod to solve (CAP ). t Table 1. Nuber o decision variables. T T n tr R Nuber o binary decision variables , ,200, ,502, ,323, Decoposition o (CAP ) (CAP ) s vast nuber o decision variables causes a heavy coputational load. Each TBTP table per superrae ust be generated within a desired tie. To alleviate the processing burden, we decopose the original proble (CAP ) into two sub-probles [8]. One proble is to ind an optial assignent aount vector. The other proble is to ind a TBTP table with the vector. According to (CAP ), the aount o TRF tieslots assigned to RCST is = Ai x S i. Thus, the proble to ind an optial can be orulated as (P( )). (P()) Maxiie subect to constraints ( ) = v (7) R { Q, X + Y }, R, in (8) R Y, R, (9) S, (10) A + Z {0}. The additional capacity requireent o RCST is given ETRI Journal, Volue 25, Nuber 4, August 2003 Ki-Dong Lee et al. 215

6 by y = Y. ith this vector, we rewrite (P()) as (P 1 (y)). (P 1 (y)) Maxiie subect to constraints y ( y ) = v y (11) 1 R { Q Y, X } R in, (12) R y S Y, (13) y A Z 2. VEDTA Algorith or (P 1 (y)) + R {0}. e present the VEDTA algorith or (P 1 (y)) and prove that the VEDTA algorith inds an optial solution o (P 1 (y)). The procedure o the VEDTA algorith is shown below: Step 1 (Sort {v }) J 0 : = {}. FOR(k :=1; k <= R ; k ++) { k arg ax{ v, R J J k : = J k 1 { k}; } : = k 1 Step 2 (Find an optial y * ) c n : = ax{ c y }; = 1 k S A Y k R * I n = R, then y : = in{ Q Y, X } R; Else, * y : = in{ Q Y, X } J n ; * y : = S Y in{ Q Y, X }; 1 n + A R J * + 1 : = 0, n; y > Step 3 (Find an optial *) *: = y * + ( Y1, Λ, Y R ) e relax the integer constraint on vector y. e show that without the integer constraint, the VEDTA algorith inds an integer optial solution given that Q, X, and Y are integers. Proposition 1: The VEDTA algorith inds an optial solution. Proo: See the appendix. I there are ree TRF tieslots, they ay be additionally assigned to logon RCSTs as ree capacity assignents. Capacity assigned in this category is intended as a bonus }; capacity, which can be used to reduce delays in any traic that can tolerate delay itter. V. TBTP Table Generation Procedure In the ETSI DVB-RCS standard [1], raes o a superrae are nubered ro 0 (lowest requency, irst in tie) to N (highest requency, last in tie), ordered irst in tie, then in requency, where N is less than or equal to 31. In a rae, tie slots are nubered ro 0 (lowest requency, irst in tie) to M (highest requency, last in tie), ordered irst in tie, then in requency, where M is less than or equal to 2,047. In our TBTP table generation procedure, we use a siilar nubering syste as ollows. In a superrae, TRF tieslots are nubered ro 0 (lowest requency, irst in tie) to A S 1 (highest requency, last in tie), ordered irst in tie, then in requency. The TBTP table is then built up iteratively. Step 1 (Initialiation) slot_counter := 0; x : = 0; Step 2 (Iteration) FOR( k := 1; k <= R ; k ++) { FOR( i : = slot_counter; i <= k ; i ++) { x : 1; } i, k = } slot_counter += ; k VI. Perorance Analysis and Discussions 1. Optiality The obective shown in (1) denotes the (weighted) throughput (in tieslots), where a certain RCST ay have a higher priority than the others so that it can have ore TRF tieslots than the others. In this sense, our optiiation proble on CAP is to axiie the throughput. Thus, we can saely conclude that our algorith attains the axiu throughput by Proposition Coputational Eiciency Our algorith has a linear coplexity with respect to R, A, and S, respectively. This eans that each actor aects the coputational coplexity within a range o a linear coplexity. In addition, our algorith requires a sall aount o eory in a coputing achine. This is a erit o our algorith or practical ipleentation. 216 Ki-Dong Lee et al. ETRI Journal, Volue 25, Nuber 4, August 2003

7 3. Coputational Results and Discussions Figure 6 shows the clusters o TRF tieslots assigned to the respective RCSTs. As shown in the igure, the cluster o each RCST is nubered ro 1 (lowest requency, irst in tie) to R (highest requency, last in tie), ordered irst in tie, then in requency. According to the characteristics o the hub, an RCST cannot use ore than one carrier at the sae tie (it uses ultiple carriers, but a single carrier at a given tieslot). I Q ntr ( T / T ), it can be siply shown that our TBTP table generation algorith always inds a easible TBTP table that does not violate the certain constraint o using ultiple carriers, but a single carrier at a given tieslot. T SF R -1 R -1 R R SF Fig. 6. A visualied exaple o a terinal burst tie plan (TBTP) table or a superrae (nuber: RCST index). e show the coputational results o our algorith using randoly generated deand vectors. Table 2 shows the superrae pattern, the available resource status, and the nuber o logon RCSTs. Table 3 presents the ties elapsed in Phase 1 (inding ) and Phase 2 (inding x), which deonstrates the coputational eiciency o our ethod. VII. Concluding Rearks e developed a ethod or eicient tieslot scheduling in an interactive broadband satellite access network [1], [2] so that the syste throughput is axiied. The tieslot assignent proble was orulated as a binary integer prograing proble [9], which has ore than 8,000,000 decision variables. e eployed a proble decoposition technique [9] that achieved a rearkable decrease in coputational burden. In an interactive satellite access network, realtie resource allocation scheduling is ipossible because o round-trip delay, i.e., it takes a certain period o tie (e.g., about 500 s in GEO satellite networks) or a terinal to receive a capacity allocation essage generated by a Hub station. Coputational results show that the proposed algorith solves the orulated proble within a short period o tie, uch shorter than the designed superrae duration. Owing to this eiciency, the proposed optiiation approach can be used or throughput perorance iproveent in interactive satellite access networks. Table 2. Superrae pattern used in our exaple. Ite Value / 2 / t 4 T / T 16 n tr 508 A 3 4 R S 508*2*16 Table 3. Coputational results (upper bound o elapsed tie). A R x Tie in seconds. Pentiu III PC 1.0 GH. Appendix e briely describe how any easible solution ound by the VEDTA algorith has optiality. The optiality o the VEDTA algorith can be proved by applying the concept o the siplex ethod or by using the copleentary slackness theore [8]. Thus, we try to prove it along a siilar but dierent way. The outline o our proo is as ollows. Step 1: Linear prograing (LP)-relaxation. Relax the integer constraint on y and consider an LP-relaxed proble. Step 2: Feasibility check. Check i the solution obtained by the VEDTA algorith is easible. Step 3: Local optiality check. Check i that solution is ETRI Journal, Volue 25, Nuber 4, August 2003 Ki-Dong Lee et al. 217

8 locally optial. I there is no easible direction iproving the obective value o that solution, then the solution is locally optial. There is no easible direction in that solution, and a atheatical proo o an equivalent proble can be ound in [10]. Step 4: Global optiality check. Since an LP is convex, a local optiu is a global optiu. Since the easible set o the original proble is a subset o the LP-relaxed proble, the global optiu o the LP-relaxed proble is the global optiu o the original proble. Acknowledgeents The irst author would like to thank the anonyous reviewers or their valuable coents to iprove this work. This work is a partial result o the Broadband Satellite Access Network (BSAN) syste developed by the ETRI (URL The every eort o all BSAN syste developing ebers was helpul to the succesul operation o the proposed algorith in the BSAN syste. Reerences [1] ETSI EN (V1.2.2): Digital Video Broadcasting (DVB); Interaction Channel or Satellite Distribution Systes,(also known as the DVB-RCS speciication). [2] J. Neale, R. Green, and J. Landovskis, Interactive Channel or Multiedia Satellite Networks, IEEE Co. Mag., Mar. 2001, pp [3] Y.-H. Cho and H.-J. Lee, Broadband Satellite Access Network (BSAN) Syste or Interactive Multiedia Services, presented in ETRI-CRL 2nd Joint Con., Japan, Aug [4] SatNews Online, URL htl. [5] Heyung Sub Lee, Un Gi Joo, Hyeong Ho Lee, and han oo Ki, Optial Tie Slot Assignent Algorith or Cobined Unicast and Multicast Packets, ETRI J., vol. 24, no. 2, Apr. 2002, pp [6] K.-D. Lee, Y.-H. Cho, S.J. Lee, and H.-J. Lee, Optial Design o Superrae Pattern or DVB-RCS Return Link, ETRI J., vol. 24, no. 3, June 2002, pp [7] G. ang and N. Ansari, Searching or Optial Frae Patterns in an Integrated TDMA Counication Syste Using Mean Field Annealing, IEEE Trans. Neural Net., vol. 9, no. 6, 1998, pp [8] K. Murty, Linear and Cobinatorial Prograing, iley, [9] K.-D. Lee, H.-J. Lee, Y.-H. Cho, and D.G. Oh, Throughput- Maxiiing Tieslot Scheduling or Interactive Satellite Multiclass Services, IEEE Coun. Lett., vol. 7, no. 6, June 2003, pp [10] K.-D. Lee, Y.-H. Cho, H.-J. Lee, and H. Jeong, Optial Scheduling or Tieslot Assignent in MF-TDMA Broadband Satellite Counications, Proc. IEEE VTC Fall 2002, vol. 3, Vancouver, Canada, 2002, pp Ki-Dong Lee received his BS and MS degrees in operations research (anageent science) and his PhD degree in industrial engineering (with applications to broadband wireless counications) in 1995, 1997, and 2001 ro Korea Advanced Institute o Science and Technology (KAIST). As a Senior Meber o Engineering Sta, he oined the Radio & Broadcasting Research Laboratory o ETRI ater his graduation ro KAIST. His research interests span perorance odeling and analysis, queueing and optiiation theories, and their applications to radio resource anageent, including algoriths and protocols, or broadband wireless/satellite counications networks. He is a eber o the IEEE. Yong-Hoon Cho was born in Seoul, Korea on June 22, He received the BS, MS, and PhD degrees in electrical engineering ro Yonsei University, Seoul, Korea in 1986, 1988 and In 1989, he oined ETRI, where he was a Principal Meber o the Engineering Sta and the Tea Leader o the Satellite Multiedia Research Tea in the Radio and Broadcasting Research Laboratory. He is a regular eber o the Korean Institute o Counication Sciences (KICS) and the Institute o Electronics Engineers o Korea (IEEK). Since 2002 he has been with COMESTA, Inc., where he is the Chie o Technical Oice. His current research interests include satellite obile counications, CDMA counication systes, and wireless obile counications systes. Ho-Jin Lee received this BS, MS, PhD degrees in electronics engineering ro Seoul National University (SNU), Korea, in 1981, 1983, and He oined ETRI in 1983 and has been involved with TDX, a ull electronic digital switching syste developent proect, satellite ground ission/network control syste developent, KOMPSAT ground control syste developent, and satellite counication earth stations/service developent. He has been with TR, USA, as a Visiting Engineer or 2 years. He worked as the Director o the Satellite Counications Application Departent o ETRI and is now the Proect Manager o a DVB-RCS syste, two-way satellite broadband access syste developent proect. Deock-Gil Oh received BS, MS, and PhD degrees in electronics engineering ro Seoul National University, Korea in 1980, 1984, and Since 1990 he has been with the Radio and Broadcasting Research Laboratory, ETRI, Daeeon, Korea, where he is currently a Principal Meber o Research Sta in the Counication Satellite Center. His research interests are in digital satellite counication/broadcasting systes and wireless counication systes. 218 Ki-Dong Lee et al. ETRI Journal, Volue 25, Nuber 4, August 2003