{ Cannel Estimation Filter Using Sinc-Interpolation for UTA FDD Downlink KLAUS KNOCHE, JÜGEN INAS and KAL-DIK KAMMEYE Department of Communications Engineering, FB- University of Bremen P.O. Box 33 4 4, D-8334 Bremen, Germany, Fax: +(49)-4/8-334 GEMANY Abstract: Tis paper introduces a new Cannel Estimation (CE) filter called Polypase SInc-Interpolation ()-filter for downlink coerent ake-combining in a DS-CDMA mobile environment. In tis case some UTA FDD scenarios are taken for sowing te feasibility of tis approac. It uses known periodically time-multiplexed pilot symbols for interpolating te cannel coefficients in between. Main advantage is te aptitude of te -CE especially for ig mobile environments. Unfortunately tis advantage is paid by te use of at least six pilot sequences resulting in a tree slot delay and poor noise reduction capabilities. Some simulation results of tis -filter compared wit oter well known CE-filters are presented. Keywords: CDMA, UTA FDD, Cannel-Estimation, Sinc-Interpolation, CE, UMTS Introduction UMTS is te 3rd generation mobile cellular communication system of te (near) future. Upon oters it defines a FDD-Wide-Band-CDMA sceme using a coerent akereceiver as sown in figure. Te Signal y ID after te I&D-Operation is denoted as y ID;l (i) = (i+), x l (j)p c scr (j)c c(j) () j=i wit c c (j) is te real valued cannelization code also known as OV-code wit spreading factor and c scr denotes te complex scrambling code. Te received signal after maximum ratio combining for a L -finger ake-receiver is denoted by y MC (i)= L l= y ID;l (i) l (i) () wit l (i) is te estimated cannel coefficient for te l-t finger. In UMTS-FDD downlink a Dedicated Pysical Cannel (DPCH) consists of user data also called Dedicated Pysical Data Cannel (DPDCH), time-multiplexed wit control information witin te so called Dedicated Pysical Control Cannel (DPCCH) as depicted in figure or in []. Te DPCCH itself is divided in Transmit Power Control (TPC) Symbols, Transport Format Combination Indicator (TFCI), and a pilot sequence. Fifteen slots wit M slot = 6 cips using tis structure are combined in one Frame of ms lengt. Te pilot sequence canging for every slot witin a frame is used for cannel estimation by correlating te incoming signal wit te pilot sequence. (*) denotes conjugate complex z z z xl-( j) x L-( j) x L-3( j) * * * * c scr( j) c scr () j c scr () j c scr () j c c () j xj) ( c c () j c c () j Σ Σ Σ Σ + c() c j xj) ( y y ( i) y ( i) ID, L - ID, L -3 * * y i) ID, ( i) L ID, ( - () i () i * () i * () i L- L- L-3 DPDCH Data L -ake - Fingers η η η L - DPCCH ymc( i) Fig. : ake-receiver DPDCH TPC TFCI Data Pilot k M slot= 6 cips = * bits (k=..7) DPCH ( Slot) Slot Slot Slot i Slot4 frame, T = ms f DPCCH DPCH: Dedicated PysicalCanel TPC: Transmit Power Control DPDCH: Dedicated Pysical Data Canel TFCI: Transport Format Combination Indicator DPCCH: Dedicated Pysical Control Canel Pilot: Pilotsequence Denote Fig. : Slot and frame structure N D = N slot, N pilot wit N slot = M slot = (3) and N pilot denotes te number of pilot symbol used in
DPCCH we can estimate a cannel coefficient in centre of te pilot sequence in te -t slot: l; = N slot, N pilot k=n D y ID;l (k) Wit te pilot sequence p (k) defined as: 6= : k<npilot p (k) = : oterwise p p (k, N D): (4) () Ten, tese estimates l; are processed furter using CE-filters. In general all furter signal processing as to cope wit noise and te time variant cange of cannel coefficients. Unfortunately bot effects are contradictory to eac oter. To antagonise noise averaging is mandatory but tis will deteriorate te performance in a ig mobile environment. Terefore a compromise as to be found. Tere are two standard linear CE-filters discussed in [] wic will be used as reference in tis paper. Te averaging filter also sown in figure 3a) tries to reduce te noise and is not following te cannel coefficient witin a slot at all. Te linear interpolationfilter is just drawing a line between two neigbouring estimates to follow te cange of te time variant cannel coefficient of one ake-finger as sown in figure 3b). Terefore noise reduction of te linear interpolation CE depends on te position witin te slot. From no reduction at te position of te pilot up to tree db in te middle between two pilot sequences. l,ν ( l,t) int= l, ν+ l, ν = konst ν--t pilotsequence ν-t slot l,ν t ν-t pilotsequence l,ν ( l,t) l,ν pol = l, ν- l, ν k+ N pilot +l,ν- t ν--t pilotsequence ν-t pilotsequence ν-t slot N Slo t ( ) Fig. 3: Classical CE, (a):, (b): Linear interpolation To examine te performance of tese two filters a bit furter te Mean Squared Error (MSE) between te real cannel coefficient and its estimation can be expressed wit: MSE (i):=e xxx; (i), (i) (6) In tis case expectation is substituted by time average due to ergodicity. MSE(i) = lim N! N N, = xxx; (i), (i) (7) wit i [; 6=, ] and xxx P is te estimate of te i+, considered CE and wit (i) = k=i (k) and (k) is te real cannel coefficient calculated for every cip. Using a one tap ayleig cannel at a velocity v wit an =N of 8 db and te UMTS-transmission slot format #8 (=8, pilot symbols per sequence) te MSE depends on te position witin slot. Tis is depicted for te two slot averaging CE in figure 4. MSE in [db] Symbol i Fig. 4: MSE of averaging CE v in [km/] A global minimum in te middle between two pilot sequences can be seen in figure 4 according to te intersection of te estimated and te real cannel coefficient in figure 3a. Especially for iger velocities tis metod degrades fast, but for lower velocities it does effectuate a 3 db gain against a single slot CE. In figure te MSE for te linear interpolation is displayed. MSE in [db] Symbol i v in [km/] Fig. : MSE of linear interpolation CE Wile at te edges (near te pilot sequences) and for low velocities tis metod provides a tree db loss against te averaging CE, it does reac it in te middle between two pilot sequences as expected. On te oter and it does provide a muc better beaviour in a igly 3 3
mobile environment. For iger velocities te advantage of 3 db noise reduction in te middle will be countered by ascending nonlinearity of te real cannel coefficient resulting in a global maximum at te centre, tat can also be deduced from figure 3b). g m (i)..8.6.4. Sinc-Interpolation Looking on te cannel estimation done so far from anoter point of view, we are trying to reconstruct te time variant cannel coefficient from a set of samples. Using te 3GPP UTA FDD DL specification [], te sampling teorem is fulfilled till doppler-sift of 7 Hz. Terefore te teoretical upper bound for te maximum velocity is 4 km/. f d;max = f s = = 7 Hz (8) ms To reconstruct an original signal out of a sampled version of it, a Sinc-interpolation (ideal lowpass) is needed. Like te WMSA cannel estimation filter introduced by [3] pilot sequences of up to six slots are used to obtain te cannel coefficients. Te reconstruction (moter) filter is terefore cosen as: wit g(i) = sin( ( (i, :), 3) 6 ( 6 (i (9), :), 3) i<6 6 and g(i) = oterwise. Note tat te delay of alf a cip is only needed if te number of pilot symbols is even. To overcome problems associated wit te leakage effect, g(i) is additionally weigted using a Hamming window. f m (i) =:4 + :46 cos i 6 6=, wit () g m (i) =g(i)f m (i) () i<6 6 and g m(i) =;f m (i) =oterwise. Te filter g m (i) in time domain is depicted in figure 6 and in frequency domain in figure 7. Polypases of g m are know taken as follows. g m; (i) =g m (i + 6= ) in[; ] () Wit tis an estimate psii(;l) (i) for te -t slot can be calculated. psii(;l) (i) = =,+;l g m; (i) (3) Te MSE of te -CE using UTA FDD slot structure #9 is sown in 8. log G m (f). 4 6 8 i Fig. 6: Polypase moter filter g m; (i) wit = 8 and one polypase g m;(9) 3 ideal lowpass polypase moter filter 3 4 6 7 8 9 3 4 f in Hz Fig. 7: Polypase moter filter ( = 8) in frequency domain [;f n]and ideal lowpass wit fg = 7 Hz MSE in [db] Symbol i Fig. 8: MSE of -CE v in [km/] Because of te non perfect lowpass caracteristic, and te non ideal sampling of te cannel coefficients, performance of -CE degrades wit velocities iger tan 3 km/ on te one and and wit very ig on te oter. For lower velocities te MSE is relatively flat indicating good beaviour, neverteless for velocities lower tan 3 km/ te MSE of te linear interpolation filter for an =N =8dB and a spreading factor of 8 is sligtly better due to te better noise reduction in te centre of a slot. 3 Simulation esults In tis section simulation results are presented. Te slot formats taken for simulation are sown in table 3
Slot DPCH DPDCH Pilot Format in Symbols per Slot #8 8 7 #3 3 8 7 8 #6 4 64 64 6 Table : Used slot formats witin te simulations BE CE using UMTS Mode 8 and Veicualar A Cannel wit v = km/ 3 or te complete list can be found in []. A Veicular-A cannel is taken, described in [4] wit a defined number of delays for te cannel-taps. Tis knowledge is used in te CE instead of searcing for te best suitable taps by correlation for some time, due to tis only a correlation is done for te six defined taps. Tis is not a significant drawback, because te correlation over te wole time-axis will provide all possible taps but will increase simulation time witout giving new insigts of te feasibility of te CEs. Te ake-receiver got six fingers but is only using taps for MC tat ave at least % of te power of te most powerful tap, because very small taps are very ard to estimate due to crosstalk and wrong selection of delay is also possible wen using a long time correlation to find it but tis is common to all CEs. Power control and cannel coding are not in use. Beside te following simulation always te classical -curve is displayed. In figure 9 a Veicular-A cannel for a velocity of km/ and slot format # 8 is taken. Te velocity of km/ is still muc to ig for te averaging to provide good results. Te linear interpolationon te oter and does provide te best results because te interpolation itself is good enoug to cope wit te time variance of te cannel. -CE is nearly as good as te linear interpolation wic benefits from te ligt noise reduction in te centre of eac slot resulting in a better overall performance of te system wit linear interpolation CE. Using a iger velocity, in tis case 3 km/ will cange te performance especially for iger =N as can be seen in figure. Wile averaging is furter degrading so is linear interpolation because tis CE cannot compensate time variant effects any longer resulting in an error floor at about,4. Wile for low =N, noise is te dominating effect te -CE does not distinguised itself from oter CEs. As iger =N gets, as better becomes te performance of -CE. In figure and 3 simulations for slot format # 3 and # 6 are taken for a velocity of km/. Wile in figure and 4 te same slot formats are taken for a velocity of 3 km/. In general te measures do not cange. is out of touc for tis kind of velocity, wile linear interpolation is best for km/ and -CE does perform superior for 3 km/. As iger te data rate gets (or as lower te spreading factor is set) te overall performance does naturally degrade. In te case of slot format # 6 wit its spreading factor =4interference becomes a problem because of te Tis problem will not occur ere but in field testing BE 3 4 6 8 4 6 8 /N Fig. 9: Veicular-A slot format # 8, v=km/ CE using UMTS Mode 8 and Veicualar A Cannel wit 3 km/ 4 6 8 4 6 8 /N Fig. : Veicular-A slot-format # 8, v=3km/ lower separation capability of te Gold-Code. Meaning tat noise is te dominating issue also for a velocity of km/ terefore in a ig noise environment averaging does ave an advantage over -CE. Neverteless te performance is bad and all CE-scemes are to close togeter in tese areas tat by using a cannel coding results may be different. 4 Conclusion In general CE as to cope wit noise and time variance of te cannel. Unfortunately bot problems are contradictory. Beginning wit WMSA-CE by [3] on te time invariant and noisy side and ending up wit te ere proposed -CE on te fast but low noise site. Tis paper sows te basic idea of te -CE and provides some clues of te utilizability of suc a sceme. Tis sceme may be improved by some points in te future. For example instead of looking for Sinc-Interpolation we may try some sort of remez-syntesis for a filter wit a lower bandwidt meaning lower velocities but a better noise reduction. Anoter point is to use more
CE using UMTS Mode 3 and Veicualar A Cannel wit km/ CE using UMTS Mode 6 and Veicualar A Cannel wit km/ BE 3 BE 4 6 8 4 6 8 /N 3 4 6 8 4 6 8 /N Fig. : Veicular-A slot format # 3, v=km/ Fig. 3: Veicular-A slot format # 6, v=km/ CE using UMTS Mode 3 and Veicualar A Cannel wit 3 km/ CE using UMTS Mode 6 and Veicualar A Cannel wit 3 km/ BE BE 3 4 6 8 4 6 8 /N 4 6 8 4 6 8 /N Fig. : Veicular-A slot format # 3, v=3km/ Fig. 4: Veicular-A slot format # 6, v=3km/ estimations instead of one l (i) for te wole pilot sequence. For example for every pilot symbol tere may be one estimation, using tis ere proposed algoritm as described and averaging all N pilot results at te end. eferences: [] 3rd Generation Partnersip Project, TS. Pysical cannels and mapping of transport cannels onto pysical cannels (FDD), October 999. [] Y. Honda and K. Jamal, Cannel estimation based on Time-Multiplexed Pilot Symbols, Tec. ep. CS96-7, IEICE, August 996. [3] H. Ando, M. Sawaasi, and F. Adaci, Cannel Estimation Filter Using Time-Multiplexed Pilot Cannel for Coerent AKE Combining in DS- CDMA Mobile adio, IEICE Trans. on Communications, vol. E8-B, pp. 7 6, July 998. [4] European Telecommunications Standards Institute (ETSI), DT/SMG-4, Selection procedures for te coice of radio transmission tecnologies of te Universal Mobile Telecommunications System UMTS, 997. Version.9.4. [] K. D. Kammeyer, Nacrictenűbertragung. Stuttgart: Teubner, second ed., 996. [6]J.G.Proakis,Digital Communications. McGraw- Hill, tird ed., 99.