AN ALGORITHM TO COMBINE LINK ADAPTATION AND TRANSMIT POWER CONTROL IN HIPERLAN TYPE 2 Markus Radmrsch Inst. f. Allgem. Nachrchtentechnk, Unv. Hannover, Appelstr. 9a, 3167 Hannover, Germany Tel.: +49-511-762 2835; Fax: -33; e-mal: radmrsch@ant.un-hannover.de Abstract: The HIPERLAN type 2 (HIgh PERformance Rado Local Area Network) standard provdes means for lnk adaptaton (LA) and transmt control (TPC). Both mechansms are not ndependent n a cellular network envronment. Ths paper analyses the basc requrements for jont LA and TPC and proposes a sutable algorthm whch s based on the empty part n MAC frames. The mplementaton n a smulator and smulaton results are presented. Keywords: Rado LAN, wreless, network, control, lnk adaptaton, adaptve modulaton, HIPERLAN I. INTRODUCTION HIPERLAN type 2 (H/2) whch has been standardsed by the ETSI Project Broadband Rado Access Networks (BRAN), uses OFDM as ts modulaton scheme. It defnes a centralsed archtecture where one or more moble termnals (MT) communcate wth an access pont (AP), and a decentralsed archtecture where MTs can communcate drectly wth each other. For the purpose of ths paper, the centralsed archtecture s assumed, although the results generally apply to the decentralsed mode as well. H/2 provdes means for lnk adaptaton (LA),.e. the adaptaton of the data rate by varaton of the modulaton scheme and the channel code rate. Ths makes t possble to transmt wth hgh data rate at good channel qualty and wth lower data rate at low channel qualty. The Phy modes gven n the standard are shown n table 1, [1]: Table 1: H/2 Phy modes Modulaton Codng rate R Nomnal bt rate r [Mbt/s] BPSK 1/2 6 BPSK 3/4 9 QPSK 1/2 12 QPSK 3/4 18 16QAM 9/16 27 16QAM 3/4 36 64QAM 3/4 54 Moreover, the standard defnes rules for an open-loop transmt control (TPC). The AP nforms the MTs about ts own transmt and ndcates at whch level t expects to receve. Note that ths nformaton s vald for all MTs,.e. the control s not ndvdual for sngle MTs. In a cellular network, nter-cell nterference s nevtable snce frequences can be reused after a certan dstance. Gven the rules for LA and TPC, an AP can ncrease the data throughput n ts cell by ncreasng the transmt, whch enables the use of hgher data rates. Ths, however, rases the nter-cell nterference of other cells whch, n turn, may ncrease ther transmt to mprove ther carrer to nterference rato (C/I). Such behavour may lead to stuatons where all devces transmt at hghest but have no gan. Therefore, LA and TPC need to be consdered jontly n order to optmse network throughput. Clause II gves a short ntroducton to the functons and features that further nfluence LA and TPC. Clause III provdes an analyss of the nfluencng factors on LA and TPC and a descrpton of the algorthm used. The smulaton envronment and results are the subject of clause IV, followed by a short summary and some conclusons. II. IMPORTANT FEATURES OF HIPERLAN/2 Ths clause pcks out only a few specfc ssues. The nterested reader s referred to [2] and [3] for more detaled nformaton about HIPERLAN/2. MAC Frame MAC Frame MAC Frame MAC Frame BC phase DL phase UL phase RA phase flexble unused part Fgure 1: Basc MAC frame structure The H/2 MAC (Medum Access Control) supports only two types of packets, [4]. The long packets, also called Long Channel (LCH), have a sze of 54 Bytes, the short packets (SCH) have a sze of 9 bytes. We wll consder LCHs only n the sequel. The medum access control (MAC) scheme s based upon a centrally controlled TDMA/TDD scheme, [4]. Ths means that the AP controls all transmssons over the ar nterface. Ths concerns uplnk and downlnk equally. The MAC frame structure s depcted n fgure 1. Each MAC frame wth duraton 2 ms has a number of phases, and rules exst how to compose a MAC frame [4]. The mportant ponts for our purposes are: -783-7589-/2/$17. 22 IEEE PIMRC 22
ƒthe broadcast (BC) phase contans all announcements of the AP, statng who shall receve and transmt at whch tme, the knd of data and the Phy mode. The Phy mode s constant and fxed durng the BC phase. ƒthe downlnk (DL) phase contans data from the AP to the MTs, the uplnk (UL) phase from the MTs to the AP. Note that transmt n the DL s constant, whereas t s the receve (Rx) at the AP whch s constant n UL drecton. Ths means that, assumng no nter-cell nterference, the Phy mode n the DL depends on the dstance between MT and AP, whereas the Phy mode s equal for all MTs n the uplnk. ƒthe random access (RA) phase allows termnals to get nto contact wth the AP when they have no resources granted. The Phy mode s constant and fxed durng the RA phase ƒthere may be an unused part n a MAC frame. It appears f not enough data s avalable to fll the whole MAC frame. We shall see that the unused part contrbutes to optmse the nterference stuaton. Buldng the MAC frame s performed by the AP. The buffer states of the MTs are, therefore, sent to the AP. The process of allocatng resources s called schedulng. Schedulng s mportant for farness and QoS support, but s not consdered n further detal n ths paper. PER (LCH) 1.1.1 6 Mbps 12 Mbps 18 Mbps 27 Mbps 36 Mbps 54 Mbps.1 5 1 15 2 25 3 C/I (db) Fgure 2: PER over C/I for H/2 LCHs The nvestgatons are based on the packet error rate (PER) curves and ther propertes for HIPERLAN/2 as publshed n [3]. The throughput of the H/2 rado lnk s gven by the PER curves n combnaton wth the ARQ protocol, [4]. The throughput, D, can be derved from the PER and the data rate r per Phy mode (see table 1) by [3]: D = r( 1 PER) (1) The resultng curves for D correspondng to the PER curves n fgure 2 are shown n fgure 3. Note that the throughput rses monotoncally wth ncreasng C/I. III. PREREQUISITES AND PROPOSED ALGORITHM A. Prerequstes The scheme to jontly handle LA and TPC wll be called JLAP (Jont Lnk Adaptaton and transmt Power control) n the sequel. The nvestgatons are based on a number of assumptons: 1. It has been shown n [6] that, wth respect to the number of erroneously receved packets, t s better to use low at a lower data rate. The remander of ths paper wll be based on ths assumpton. 2. The C/I at the recever can not be measured drectly. Therefore, t s derved from the estmated PER n the recever accordng to fgure 2. The PER estmaton happens n the H/2 error control, [4]. 3. All nvestgatons assume pure Best Effort servce. Throughput D (Mbt/s) 5 4 3 2 1 6 Mbps 12 Mbps 18 Mbps 27 Mbps 36 Mbps 54 Mbps 5 1 15 2 25 3 C/I (db) Fgure 3: Throughput D over C/I for H/2 LCHs JLAP n conjuncton wth the MAC scheduler has to make sure that a number of prerequstes can be fulflled n addton to the above assumptons: The Phy mode s always adjusted such that, for the average C/I gven, t generates maxmum throughput, see fgure 3. All packet arrvals n a rado cell need to be served. The buffer states of the MTs n UL and DL drecton are known to the AP. Dependng on the used Phy modes, the system can carry a certan capacty. So f the buffer states grow, the capacty s too low and, hence, the throughput needs to be ncreased. Ths, consderng the prevous bullet, would requre an ncrease of transmt. It s vtal for MTs that they are able to receve the BC phase, snce t contans all nformaton about the ongong MAC frame. The H/2 standard provdes a means for feedback to ndcate the loss of BC phases to the AP. B. Optmsaton Problem Consderng the assumptons and prerequstes from the prevous secton, the algorthm s based on the followng OPTIMISATION PROBLEM
To formulate the optmsaton problem, we need to ntroduce a number of varables: ƒthe MTs belongng to an AP shall be numbered wth nteger numbers, startng at 1 up to the number of MTs assocated. ƒthe expected Rx of the AP n the uplnk shall be denoted by P e,u, the transmt of the AP shall be named P s,d. They are arranged n a vector P G such that G T P = ( Pe. U, Ps, D ). ƒthe data rate for the uplnk transmsson of MT shall be named r,u, the data rate for downlnk transmsson from the AP to MT by r,d. ƒthe tme granted for transmsson to MT n the downlnk shall be named τ,d, from MT to the AP τ,u. ƒthe overall servce rate of a rado cell, R S, s defned as: R = r τ + r τ ) S (. d. D ƒthe data arrval rate of data n MT destned to the AP shall be denoted by α,u, the arrval rate n the AP destned to MT by α,d. ƒthe overall arrval rate n a rado cell, Rα, s gven by Rα = ( α. U + α. D ) We are now ready to formulate the optmsaton problem: Mnmse P G subject to: { : ( r τ + r τ ) ( α α )} ( Rs Rα ) d. D. U +.. D In prose, the proposed soluton operates such that the transmsson n a rado cell shall be adjusted as low as possble, as long as the assumptons and prerequstes from the prevous secton can be kept. The key to the algorthm s the unused part of a MAC frame, see fgure 1, n conjuncton wth the buffer states. Ths s due to the fact that the throughput functon over C/I s a strctly monotoncally ncreasng functon, see fgure 3. So assumng that the nterference caused by other rado cells remans constant, transmttng at hgher always means ncreasng C/I at the recever whch, n turn, means hgher throughput. Hgher throughput means that the same amount of data can be transmtted n shorter tme,.e. the sze of the unused part s ncreased. Ths s equvalently vald for decreasng transmt. Note that an ncrease or decrease of ether UL or DL has an equal mpact on the sze of the unused part. However, t seems advsable to change the of the drecton whch has bgger buffers or a bgger buffer ncrease/decrease rate. Gven the above, the unused part ndcates whether the avalable capacty s suffcent to serve all packet arrvals n the rado cell and can be used as a measure on whether the transmt shall be kept, ncreased or decreased. If the total packet arrval rate,.e. the sum of UL and DL arrvals, exceeds the avalable capacty for a gven vector P G, the unused part n a MAC frame s very small and the buffer states grow. In ths case, the transmt needs to be ncreased n order to ncrease capacty. If the unused part s rather large and the buffer states are constant, the avalable capacty s hgher than requred and the transmt can be decreased.. An algorthm based on a smlar bass has already been proposed and nvestgated n [5]. There, however, the TPC s on a per-mt bass whch s n prncple possble accordng to the standard, [1], but orgnally provded to cover producton qualty varatons of RF crcutry. Here, control s mostly used as standardsed for the whole rado cell. The only dfference n ths paper compared to the standard s that there s no lmtaton to ncrease DL transmt by not more than 9 db n every 5 mnutes nterval. C. Proposed Algorthm We can now wrte down the proposed algorthm. The AP starts wth maxmum DL transmt (Tx) and expected UL Rx. JLAP happens n the AP perodcally every th MAC frame, controlled by a tmer, where t performs 2 actons alternatngly: 1. Buffer State and Unused Duraton (BSUD): Calculate the average buffer states n UL and DL over the last MAC frames Calculate the average unused parts of the last MAC frames Check f more than two BC falures have been reported durng the last MAC frames, set the BC falure flag. Increase or decrease transmt as shown n fgure 4. The ncrease or decrease of Tx n the DL happens n 3 db steps, n the UL n 4 db steps. For both, UL and DL, a mnmum and maxmum exsts whch cannot be exceeded. The unused part s checked aganst a hysteress wth a lower and an upper lmt. If the unused part s nsde the hysteress, T s reman unchanged. Tx s always ncreased for ether DL or UL, decrease happens for both UL and DL. The behavour n the case the BC falure flag has been set, s not shown n fgure 4. In ths case, the DL s ncreased by 3 db and the new value s set as the mnmum DL Tx, P DL, mn, for n MAC frames. After that perod, P DL, mn s decreased by 3 db. Fnally, f the UL Tx has been changed, the Phy modes n the UL are recalculated, based on prevous C/I estmates (see step 2 below) plus the UL change. Note that the Phy modes n the DL are calculated by the MTs and ther proposal s accepted by the AP.
massve buffer ncrease? unused part <= lower lmt? UL buffer ncrease > DL buffer ncrease? unused part <= lower lmt? decrease UL and DL L, j = 37 + 3 log 1 d, j + p where d,j s the dstance between staton and j, and p s a factor to smulate slow fadng accordng to the ARMA model, [7]. All APs operate on the same frequency. The scenaro conssts of 5 rado cells whch are arranged n a manhattan grd, see fgure 5. The crcles are rado cells, each wth an AP n the mddle. The APs are located at dstance d from each other. The MTs move randomly and equally dstrbuted nsde the crcles n fgure 5 (radus r=5m around ncrease DL ncrease UL r d AP 5 Fgure 4: Power ncrease or decrease 2. Calculate optmum Phy modes (COP) Calculate the average PER for all actve connectons durng the last MAC frames. Calculate the correspondng C/I value. If the number of samples for the PER estmaton s too low, the C2I estmaton uses addtonally the averaged receved sgnal strength (RSS) values of the last MAC frames. If the RSS values are used, a securty margn of 6 to 12 db s subtracted from the resultng C/I value. Calculate the optmum Phy mode for all actve connectons for the estmated PER, see fgure 3. Note that the step BSUD does not estmate C/I values, whereas the step COP does not change Tx s. The MTs perform the COP step of the AP perodcally every th MAC frame whch leads to Phy mode proposals to the AP. If the AP changes DL Tx Power, the MTs go through an extraordnary COP step where they add the DL Tx change to the estmated C/I. IV. SIMULATIONS AND RESULTS The HIPERLAN/2 protocol has been mplemented n an SDL (Specfcaton and Descrpton Language) based smulaton tool. It contans all convergence layer (CL) and data lnk control (DLC) data transmsson functons of H/2, [3]. Each staton has a moton profle whch s read from a fle at smulaton start or can be generated randomly. The rado channel operates such that each transmtter passes ts data down together wth the transmt duraton and transmt. After the transmt duraton has passed, the whole data s transferred to the recever n one step, ncludng the calculated C/I value. The nterference between dfferent transmssons are based on the pathloss, weghted wth the overlap duraton. The pathloss L,j between statons and j n db s calculated as: Fgure 5: Spatal dstrbuton of APs and MTs the respectve AP). In the smulaton runs, 8 MTs are allocated to each AP, each wth one connecton n UL and one n DL. All connectons are served wthout Qualty of Servce dfferentaton. They have equal data arrval rates accordng to a posson process wth a packet length of 1484 bytes above the CL and nterarrval tmes adjusted to the data rate. The throughput n the smulaton results s measured n the recever on top of the CL layer,.e. after reassembly. All graphs show the results of the rado cell belongng to AP 3 surrounded by four other cells, thus avodng edge effects. Frst smulaton results are shown n fgure 6. It shows the total throughput over the total offered load wth the dstance d between APs as parameter. The throughput follows the offered load up to a maxmum value and then goes nto saturaton. The saturaton throughput grows wth ncreasng dstance d. The curve for a sngle rado cell wthout surroundng nterferers s shown as an upper throughput lmt. total throughput rado cell 3 [Mbt/s] 35 3 25 2 15 1 5 d = 1 m d = 2 m d = 3 m d = 4 m d = 6 m d = 8 m sngle cell AP 2 AP 3 AP 4 AP 1 5 1 15 2 25 3 35 4 total offered load n rado cell 3 [Mbt/s] Fgure 6: Throughput for dfferent dstances d
AP transmt [dbm] Average MT expected UL [dbm] 3 25 2 15 1 d=1m d=2m 5 d=3m d=4m d=6m d=8m 5 1 15 2 25 3 35 4 Total Offered Load AP 3 [Mbt/s Fgure 7: AP 3 s transmt wth devaton bars -4-45 -5-55 -6-65 d=1m d=2m d=3m -7 d=4m d=6m d=8m -75 5 1 15 2 25 3 35 4 Total Offered Load AP 3 [Mbt/s] Fgure 8: Expected uplnk Rx at AP 3 wth devaton bars The AP transmt and ts standard devaton over the total offered load of rado cell 3 s shown n fgure 7. The Tx stays at ts maxmum when the throughput lmt s reached. Smlar results can be seen from fgure 8 where the expected UL Rx at the AP s shown. The AP Tx generally ncreases wth ncreasng load for a gven dstance d. It has, however, a mnmum just before t goes to maxmum Tx. It seems that ths s the pont where the algorthm starts to ncrease uplnk Rx whch affects DL. VII. REFERENCES [1] TS 11 475, "BRAN; HIPERLAN Type 2; Physcal (PHY) Layer", ETSI, March 2 [2] M. Radmrsch, V. Vollmer, "HIPERLAN Type 2 Standardsaton an Overvew", European Wreless Conference, Munch, Germany, Oct. 1999 [3] J. Khun-Jush, et al., HIPERLAN type 2 for Broadband Wreless Communcaton", Ercsson Revew No. 2, 2 (http://ercsson.com/revew) [4] TS 11 761-1, "BRAN; HIPERLAN Type 2; Data Lnk Control (DLC) Layer; Part 1: Basc Data Transport Functon", ETSI, March 2 [5] A. Krämlng, A Power Control Strategy for HIPERLAN/2, 1 th Aachen Symposum on Sgnal Theory, Sep. 21, Aachen,,Germany [6] M. Radmrsch, Analyss of the rado lnk propertes of HIPERLAN/2, COST 273 TD(1)3, Bologna, Italy, Oct. 21 [7] D. Huo, Smulatng Slow Fadng by means of one dmensonal stochastcal process, proc. 46 th IEEE VTC, Vol. 2, Atlanta, 1996 V. SUMMARY AND OUTLOOK The presented algorthm for jont control of transmt and lnk adaptaton n HIPERLAN/2 s able to dmnsh both UL and DL transmt s sgnfcantly, as long as the offered load stays below the overall capacty of a rado cell. It can be expected, however, that even wth hgh nterference, a hgher throughput can be acheved when the DL Tx and the expected UL Rx are not at ther maxmum. For ths purpose, a method s requred whch rewards the decrease of Tx but stll tres to maxmse throughput. Such a scheme s currently under nvestgaton by the author VI. ACKNOWLEDGEMENTS The results have been funded by the Deutsche Forschungsgemenschaft (DFG).