Multiple vs. Random Access. Spread Spectrum MAC. Random Access and Scheduling. BPSK Example. Direct Sequence

Transcription

1 7C Cimini-9/97 EE360: Multiuer Wirele Sytem and Networ Lecture 4 Outline Announcement Proect propoal due Feb. ( wee) Maeup lecture Feb, 5-6:5, Gate Preentation chedule finalize Random v. Multiple Acce Random Acce and Scheduling Spread Spectrum Multiuer Detection Multiuer OFDM and OFDM/CDMA Multiple v. Random Acce Multiple Acce Technique Ued to create a dedicated channel for each uer Orthogonal (TD/FD with no interference) or emiorthogonal (CD with interference reduced by the code preading gain) technique may be ued Random Acce No dedicated channel aigned to each uer Uer contend for channel when they have data to end Very effient when uer rarely active; very ineffient when uer have continuou data to end Scheduling and hybrid cheduling ued to combine benefit of multiple and random acce RANDOM ACCESS TECHNIQUES Random Acce and Scheduling Dedicated channel wateful Ue tatitical multiplexing Random Acce Technique Aloha (Pure and Slotted) Carrier ening Can include colliion detection/avoidance If channel buy, determinitic or random delay (non-peritent) Poor performance in heavy loading Reervation protocol Reource reerved for hort tranmiion (overhead) Hybrid Method: Pacet-Reervation Multiple Acce Retranmiion ued for corrupted data (ARQ) Hybrid ARQ partial retranmiion: more coded bit Spread Spectrum MAC Baic Feature ignal pread by a code ynchronization between pair of uer compenation for near-far problem (in MAC channel) compreion and channel coding Spreading Mechanim direct equence multiplication frequency hopping Note: preading i nd modulation (after bit encoded into digital waveform, e.g. BPSK). DS preading code are inherently digital. Direct Sequence BPSK Example Linear Modulation. (PSK,QAM) SS Modulator d(t) S (t) (t) Channel Synchronized Chip time T c i N time the ymbol time T. Bandwidth of (t) i N+ time that of d(t). Channel introduce noie, ISI, narrowband and multiple acce interference. Spreading ha no effect on AWGN noie ISI delayed by more than T c reduced by code autocorrelation narrowband interference reduced by preading gain. MAC interference reduced by code cro correlation. S (t) Linear Demod. SS Demodulator T b T c =T b /0 d(t) (t) (t)

2 8C380.7-Cimini-7/98 Spectral Propertie Code Propertie Modulated Data Data Signal with Spreading Narrowband Interference Receiver Input ISI Other SS Uer Original Data Signal Other SS Uer Demodulator Filtering Narrowband Filter ISI Autocorrelation: T r( t ) ( t t ) dt T 0 Cro Correlation T ri ( t ) c ( t t ) dt T 0 Good code have r(t)=d(t) and r i (t)=0 for all t. r(t)=d(t) remove ISI r i (t)=0 remove interference between uer Hard to get thee propertie imultaneouly. ISI Reection Tranmitted ignal: (t)=d(t) (t). Channel:h(t)=d(t)+d(t-t). Received ignal: (t)+(t-t) Received ignal after depreading: r( t) d( t) d( t t ) ( t t ) d( t) d( t t ) ( t t ) In the demodulator thi ignal i integrated over a ymbol time, o the econd term become d(t-t)r(t). For r(t)=d(t), all ISI i reected. MAC Interference Reection Received ignal from all uer (no multipath): r( t) Received ignal after depreading r( t) d M ( t t ) i M, i d ( t t ) ( t t ) In the demodulator thi ignal i integrated over a ymbol time, o the econd term become M, For r i (t)=0, all MAC interference i reected. M d ( t t ) r ( t ) i d ( t t ) i c c ( t t ) Walh-Hadamard Code For N chip/bit, can get N orthogonal code Bandwidth expanion factor i roughly N. Roughly equivalent to TD or FD from a capaty tandpoint Multipath detroy code orthogonality. Semi-Orthogonal Code Maximal length feedbac hift regiter equence have good propertie In a long equence, equal # of and 0. No DC component A run of length r chip of the ame ign will occur -r l time in l chip. Tranition at chip rate occur often. The autocorrelation i mall except when t i approximately zero ISI reection. The cro correlation between any two equence i mall (roughly r i =G -/, where G=B /B ) Maximize MAC interference reection

3 SINR analyi SINR (for K uer, N chip per ymbol) K N 0 SINR 3 N E Interference limited ytem (ame gain) 3N 3G SIR K K Random preading code Interference limited ytem (near-far) 3N 3G SIR ; ( K ) ( K ) Aume random preading code 3N 3G SIR ( K ) ( K ) Nonrandom preading code CDMA v. TD/FD For a preading gain of G, can accommodate G TD/FD uer in the ame bandwidth SNR depend on tranmit power In CDMA, number of uer i SIR-limited 3G SIR ( K ) 3G K SIR For SIR3/, ame number of uer in TD/FD a in CDMA Fewer uer if larger SIR i required Different analyi in cellular (Gilhouen et. Al.) Frequency Hopping Tradeoff Nonlinear Modulation. (FSK,MSK) d(t) FM Mod (t) Channel FM Demod Nonlinear Demod. Hopping ha no effect on AWGN S (t) FH Modulator VCO VCO FH Demodulator S (t) Spreading code ued to generate a (low or fat) hopping carrier frequency for d(t). Channel BW determined by hopping range. Need not be continuou. Channel introduce ISI, narrowband, and MAC interference No ISI if d(t) narrowband, but channel null affect certain hop. Narrowband interference affect certain hop. MAC uer collide on ome hop. Spectral Propertie Slow v. Fat Hopping D (f-f c ) 3 4 D i (f-f c ) 4 3 Fat Hopping - hop on every ymbol NB interference, MAC interference, and channel null affect ut one ymbol. Correct uing coding Slow Hopping - hop after everal ymbol NB interference, MAC interference, and channel null affect many ymbol. Correct uing coding and interleaving if # ymbol i mall. Slow hopping ued in cellular to average interference from other cell 3

4 FH v. DS Linear v. Nonlinear DS i a linear modulation (pectrally effient) while FH i nonlinear Wideband interference/amming Raie noie pectral denity, affect both technique equally. Narrowband interference/amming DS: interfering ignal pread over pread BW, power reduced by preading gain in demodulator FH: interference affect certain hop, compenate by coding (fat hopping) or coding and interleaving (low hopping). FH v. DS Tone interference DS: tone i wideband, raie noie floor for duration of the tone. Compenate by coding (tone duration=ymbol time) or coding and interleaving (tone duration>ymbol time). Similar affect a NB interference in FH. FH: Tone affect certain hop. Compenate by coding or coding and interleaving. ISI Reection DS: ISI reduced by code autocorrelation. FH: ISI motly eliminated. FH v. DS MAC interference DS: MAC interference reduced by cro correlation of preading code. Each additional uer raie noie floor. Overall SNR reduced FH: MAC interference affect certain hop. Each additional uer caue more hop to be affected. More bit liely to be received in error. Overlay ytem: high-power NB interferer Similar impact a with regular interferer DS: Noie floor raied ignificantly FH: Hop colliding with interferer are lot Can notch out interfering ignal Evolution of a Sentit turned Entrepreneur Spread pectrum communication - myth and realitie, A.J. Viterbi, IEEE Comm. Magazine, May 79 (Linabit 5 year old - TDMA company). When not to pread pectrum - a equel, A.J. Viterbi, IEEE Comm. Magazine, April 985 (Linabit old to M/A-Com in 98) Wirele digital communication: a view baed on three leon learned, A.J. Viterbi, IEEE Comm. Magazine, Sept. 9. (Qualcomm CDMA adopted a tandard). Myth and Realitie Myth : Redundancy in error correction code pread ignal bandwidth and thereby reduce proceing gain Reality: Effective proceing gain increaed by coding by conidering ymbol rate and energy Reality today: coded modulation more effient even without ymbol argument. But tradeoff between coding and preading an open iue. Myth : Error correction code only good againt uniform interference Reality: Not true when coding combined with pread pectrum, ince SS average interference. Reality today: Unchanged. Myth 3: Interleaving detroy memory which can be ued to correct error, hence interleaving i bad Reality: Memory preerved by oft-deion even with an interleaver Reality today: Unchanged, but interleaver may require exceive delay for ome application. Myth 4: Direct equence twice a effient a frequency hopping Myth=Reality. Argument i that DS i coherent and that account for 3dB difference. Analyi how that higher level ignaling alphabet doe not help FH performance with partial band ammer. Reality today: A true effiency tradeoff of FH veru DS ha not been done under more general aumption. FH typically ued to average interference. Appealing when continuou preading BW not available. 4

5 When not to Spread Spectrum - A Sequel (85) Concluion : When power i limited, don t contribute to the noie by having uer am one another. Concluion : Networ control i a mall price to pay for the effiency afforded by TDMA or FDMA Power control i a big control requirement. Concluion 3: Interference from adacent cell affect the effiency of TDMA or FDMA le everely than in CDMA. Concluion 4: Treating bandwidth a an inexpenive commodity and proceing a an expenive commodity i bung current technology trend. Application wa mall earth terminal for commeral atellite. Three Leon Learned (9) Never dicard information prematurely Compreion can be eparated from channel tranmiion with no lo of optimality Gauian noie i wort cae. Optimal ignal in preence of Gauian noie ha Gauian ditribution. So elf-interference hould be deigned a Gauian. Standard i.e. pread pectrum optimal for G/3G Realitie (0) Never dicard information prematurely Ue oft-deion and equence detector Compreion can be eparated from channel tranmiion For time-invariant ingle-uer channel only. Self-interference hould be Gauian Baed on Viterbi argument, thi repreent a addle (not optimal) point. If the elf-interference i treated a noie, not interference, then Gauian ignaling i uboptimal (by Shannon theory). pread pectrum lot out to OFDM in 4G Multiuer Detection In all CDMA ytem and in TD/FD/CD cellular ytem, uer interfere with each other. In mot of thee ytem the interference i treated a noie. Sytem become interference-limited Often ue complex mechanim to minimize impact of interference (power control, mart antenna, etc.) Multiuer detection exploit the fact that the tructure of the interference i nown Interference can be detected and ubtracted out Better have a darn good etimate of the interference MUD Sytem Model MUD Algorithm Synchronou Cae y(t)= (t)+ (t)+ 3 (t)+ n(t) c (t) c (t) MF MF MF 3 y +I y +I y 3 +I 3 Multiuer Detector Optimal MLSE Linear Multiuer Receiver Suboptimal Non-linear c3 (t) Matched filter integrate over a ymbol time and ample Decorrelator MMSE Multitage Deion -feedbac Succeive interference cancellation 5

6 Optimal Multiuer Detection Maximum Lielihood Sequence Etimation Detect bit of all uer imultaneouly ( M poibilitie) Matched filter ban followed by the VA (Verdu 86) VA ue fact that I i =f(b, i) Complexity till high: ( M- tate) In aynchronou cae, algorithm extend over 3 bit time (t)+ (t)+ 3 (t) VA ample MF in round robin faion c (t) c (t) c3 (t) MF MF MF 3 y +I y +I y 3 +I 3 Viterbi Algorithm Searche for ML bit equence Suboptimal Detector Main goal: reduced complexity Deign tradeoff Near far reitance Aynchronou veru ynchronou Linear veru nonlinear Performance veru complexity Limitation under practical operating condition Common method Decorrelator MMSE Multitage Deion Feedbac Succeive Interference Cancellation Mathematical Model Simplified ytem model (BPSK) Baeband ignal for the th uer i: t x i c i t it t i0 (i) i the i th input ymbol of the th uer c (i) i the real, poitive channel gain (t) i the ignature waveform containing the PN equence t i the tranmiion delay; for ynchronou CDMA, t =0 for all uer Received ignal at baeband K yt t nt K number of uer n(t) i the complex AWGN proce Matched Filter Output Sampled output of matched filter for the th uer: T y yt t dt 0 c x x c t t dt tn t dt t term - deired information nd term - MAI 3 rd term - noie K Aume two-uer cae (K=), and r T 0 T t t T 0 0 dt Symbol Detection Output of the matched filter are: y cx rc x z y cx rc x z Detected ymbol for uer : xˆ gn y If uer much tronger than uer (near/far problem), the MAI rc x of uer i very large Matrix repreentation Decorrelator y RW x z where y=[y,y,,y K ] T, R and W are KxK matrice Component of R are cro-correlation between code W i diagonal with W, given by the channel gain c z i a colored Gauian noie vector Solve for x by inverting R ~ y R y W x R z xˆ gn ~ y Analogou to zero-forng equalizer for ISI Pro: Doe not require nowledge of uer power Con: Noie enhancement 6

7 Multitage Detector Deion produced by t tage are nd x tage: gn y rcx x gn y rcx and o on x x, Succeive Interference Canceller Succeively ubtract off tronget detected bit MF output: b c x rc x z b cx rc x z Deion made for tronget uer: xˆ gn b Subtract thi MAI from the weaer uer: xˆ gn y rc xˆ gn c x rc x xˆ z all MAI can be ubtracted i uer decoded correctly MAI i reduced and near/far problem alleviated Cancelling the tronget ignal ha the mot benefit Cancelling the tronget ignal i the mot reliable cancellation Parallel Interference Cancellation Performance of MUD: AWGN Similarly ue all MF output Simultaneouly ubtract off all of the uer ignal from all of the other wor better than SIC when all of the uer are received with equal trength (e.g. under power control) Performance of MUD Rayleigh Fading Near-Far Problem and Traditional Power Control On uplin, uer have different channel gain If all uer tranmit at ame power (P i =P), interference from near uer drown out far uer Traditional power control force each ignal to have the ame received power Channel inverion: P i =P/h i P 3 P Increae interference to other cell Decreae capaty Degrade performance of ucceive h interference cancellation and MUD P Can t get a good etimate of any ignal h h 3 7

8 Near Far Reitance Synchronou v. Aynchronou Received ignal are received at different power MUD hould be inenitive to near-far problem Linear receiver typically near-far reitant Diparate power in received ignal doen t affect performance Nonlinear MUD mut typically tae into account the received power of each uer Optimal power pread for ome detector (Viterbi 9) Linear MUD don t need ynchronization Baically proect received vector onto tate pace orthogonal to the interferer Timing of interference irrelevant Nonlinear MUD typically detect interference to ubtract it out If only detect over a one bit time, uer mut be ynchronou Can detect over multiple bit time for aynch. uer Significantly increae complexity Channel Etimation (Flat Fading) Nonlinear MUD typically require the channel gain of each uer Channel etimate difficult to obtain: Channel changing over time Mut determine channel before MUD, o etimate i made in preence of interferer Imperfect etimate can ignificantly degrade detector performance Much recent wor addreing thi iue Blind multiuer detector Simultaneouly etimate channel and ignal State Space Method Antenna technique can alo be ued to remove interference (mart antenna) Combining antenna and MUD in a powerful technique for interference reection Optimal oint deign remain an open problem, epeally in practical cenario Multipath Channel Summary In channel with N multipath component, each interferer create N interfering ignal Multipath ignal typically aynchronou MUD mut detect and ubtract out N(M-) ignal Deired ignal alo ha N component, which hould be combined via a RAKE. MUD in multipath greatly increaed Channel etimation a nightmare Current wor focued on complexity reduction and blind MUD in multipath channel (Wang/Poor 99) MUD a powerful technique to reduce interference Optimal under ideal condition High complexity: hard to implement Proceing delay a problem for delay-contrained app Degrade in real operating condition Much reearch focued on complexity reduction, practical contraint, and real channel Smart antenna eem to be more practical and provide greater capaty increae for real ytem 8

9 Multiuer OFDM MCM/OFDM divide a wideband channel into narrowband ubchannel to mitigate ISI In multiuer ytem thee ubchannel can be allocated among different uer Orthogonal allocation: Multiuer OFDM Semiorthogonal allocation: Multicarrier CDMA Adaptive technique increae the pectral effiency of the ubchannel. Spatial technique help to mitigate interference between uer OFDM OFDM overlap ubtream Subtream eparated in receiver Minimum ubtream eparation i B/N, total BW i B B/N f 0 Effient IFFT tructure at tranmitter Similar FFT tructure at receiver Subcarrier orthogonality mut be preerved Impaired by timing itter, frequency offet, and fading. f N OFDM-FDMA (a..a. OFDMA) Ued by the CATV community Ued to end uptream data from ubcriber to cable head-end. Aign a ubet of available carrier to each uer f Adaptive OFDM-FDMA Different ubcarrier aigned to different uer Aignment can be orthogonal or emiorthogonal f 0 The fading on each individual ubchannel i independent from uer to uer Adaptive reource allocation give each their bet ubchannel and adapt optimally to thee channel Multiple antenna reduce interference when multiple uer are aigned the ame ubchannel f N Adaptive Reource Allocation Orthogonal Subcarrier Allocation Degree of freedom Subcarrier allocation Power Rate Coding BER Optimization goal (ubect to power contraint): Maximize the um of average uer rate Find all poible average rate vector ( capaty region) Find average rate vector with minimum rate contraint Minimize power for ome average rate vector Minimize outage probability for ome contant rate vector. OFDM-TDMA Each uer equentially end one or more OFDM ymbol per frame A ingle OFDM-TDMA frame:... Uer Uer... Uer N- Uer N- Uer N... 9

10 Multiuer OFDM with Multiple Antenna Multiple antenna at the tranmitter and receiver can greatly increae channel capaty Multiple antenna alo ued for patial multiple acce: Uer eparated by patial ignature (veru CDMA time ignature) Spatial ignature are typically not orthogonal May require interference reduction (MUD, cancellation, etc.) Method of patial multiple acce Singular value decompoition Space-time equalization Beamteering OFDM required to remove ISI ISI degrade patial ignature and interference mitigation CDMA-baed cheme Can combine concept of CDMA and OFDM Reap the benefit of both technique In 993, three lightly different cheme were independently propoed: MC-CDMA (Yee, Linnartz, Fettwei, and other)* Multicarrier DS-CDMA (DaSilva and Soua)* MT-CDMA (Vandendorpe) *Stephan tal Multicarrier CDMA Multicarrier CDMA combine OFDM and CDMA Idea i to ue DSSS to pread a narrowband ignal and then end each chip over a different ubcarrier DSSS time operation converted to frequency domain Greatly reduce complexity of SS ytem FFT/IFFT replace ynchronization and depreading More pectrally effient than CDMA due to the overlapped ubcarrier in OFDM Multiple uer aigned different preading code Similar interference propertie a in CDMA Multicarrier DS-CDMA The data i erial-to-parallel converted. Symbol on each branch pread in time. Spread ignal tranmitted via OFDM Get preading in both time and frequency c(t) (t) S/P convert c(t) IFFT P/S convert Summary OFDM i a well-nown technique to combat ISI Alo very powerful in a multiuer etting Some form of multiuer OFDM lend themelve well to adaptive technique Many high-performance multiuer wirele ytem today are baed on OFDM technique. 0