Pilot Aided Channel Estimation for MIMO MC-CDMA Stephan Sand (DLR) Fabrice Portier CNRS/IETR NEWCOM Dept. 1, SWP 2, Barcelona, Spain, 3 rd November, 2005
Outline System model Frame structure MIMO Pilot aided channel estimation (PACE) Simulation scenario Simulation results Summary & outlook 2
System Model: Downlink MC-CDMA Transmitter Pilot Symb Source COD π MOD S/P 1 Multiple Access Scheme ST-COD Pilot MUX OFDM + T G 1 M Pilot MUX OFDM + T G N TX Pilot Symb 3
System Model: Receiver with Pilot Aided Channel Estimation (PACE) CSI MIMO Channel Estimator Sink DECOD π -1 CSI DMOD P/S M 1 D E S P ST- DECOD + EQX MIMO Pilot DEMUX -T G IOFDM -T G IOFDM 1 N RX 4
Frame Structure TDD transmission 32 OFDM symbols per downlink slot Full OFDM pilot symbols Pilots are Alamouti space-time coded time 1 N time s 32 synchronization symbol pilot symbol data symbol guard symbol 5
MIMO Pilot Aided Channel Estimation (PACE) PACE: Pilot symbols yield initial estimates for the channel transfer function at pilot symbol positions, i.e., localized channel estimate: ( ) 1 1 ( ) { } H H H H Hnl ', ' = Rnl ', ' Pnl ', ' Pnl ', ' Pnl ', ' = Hnl ', ' + Znl ', ' Pnl ', ' Pnl ', ' Pnl ', ', n', l' P, where P denotes the set of pilot symbols. Filtering pilot symbols yields final estimates for the complete channel transfer function: Hˆ = ω H, T P, n= 1,, N, l= 1,, N, nl, n', l', nl, n', l' nl, c s { nl ', '} T nl, where ω n,l,n,l is the shift-variant 2-D impulse response of the filter. T n,l is the set of initial estimates that are actually used for filtering. 6
Pilot Aided Channel Estimation (PACE) Filter design: Knowledge of the Doppler and time delay power spectral densities (PSDs) Optimal 2D FIR Wiener filter Separable Doppler and time delay PSDs Two cascaded 1-D FIR filters In practice, Doppler and time delay PSDs are not perfectly known Robust design assuming rectangular Doppler and time delay PSDs Frequency direction (FD): averaging Time direction (TD): linear interpolation or robust Wiener filter Set of filter coefficients can be pre-computed and stored in the receiver 7
Simulation Results: Scenario Carrier frequency 5.2 GHz Bandwidth / Occupied FFT / Subcarriers 1024 / 768 61.44 MHz / 46.14 MHz Channel model Sampling duration T spl Guard interval T GI OFDM symbol duration Subcarrier spacing Δf 16.27 ns 256 T spl = 4.166 μs 20.833 μs 60 khz time OFDM symbols / Frame 32 Modulation Coding Detection Spreading factor 32 Frequency Interleaving QPSK, 64 QAM Conv. code R=1/2, R=3/4 Alamouti-MRC (SUB) Alamouti-MMSE None APDP Bran E τ max / τ rms 1.76 μ s / 0.250 μs N p 17 f D,max N p: number of non-zero taps 0.005Δf=289 Hz @ 60 km/h 0.015Δf=867 Hz @180 km/h 8
Simulation Results: MIMO Channel Model Channel model developed within the European project IST-MATRICE Semi-geometric model ( deterministic subrays from statistic distributions) based on 3GPP specifications Sum of subrays exploiting angular characteristics 90 120 BS 60 150 30 180 0 210 330 240 270 300 90 120 MS 60 150 30 180 0 1 0.5 0 BS: Antenna Spacing (in wavelengths) 0 2 4 6 8 10 1 0.5 Spatial Correlation Spatial Correlation 210 240 270 300 330 0 MS: Antenna Spacing (in wavelengths) 0 2 4 6 8 10 9
Simulation Results: Perfect Localized Estimation BER vs. E b /N 0, QPSK (R=1/2) Default Values: f D,max : 0.015Δf @180 km/h Spreading Length: L=32 Users: K=1,32 Interpolation error: small BER degradation for linear interpolation Pilot Overhead: 1dB E b /N 0 loss 10
Simulation Results: Perfect Localized Estimation BER vs. E b /N 0, 64-QAM (R=3/4) Default Values: f D,max : 0.015Δf @180 km/h Spreading Length: L=32 Users: K=1,32 Interpolation error: large BER degradation for linear interpolation, but small for Wiener filter Pilot Overhead: 1dB E b /N 0 loss 11
Simulation Results: Imperfect Localized Estimation BER vs. E b /N 0, QPSK (R=1/2) Default Values: f D,max : 0.015Δf @180 km/h Spreading Length: L=32 Users: K=32 Imperfect localized estimate: E b /N 0 loss 1.5dB Pilot Overhead: 1dB E b /N 0 loss 12
Simulation Results: Imperfect Localized Estimation BER vs. E b /N 0, 64-QAM (R=3/4) Default Values: f D,max : 0.015Δf @180 km/h Spreading Length: L=32 Users: K=32 Imperfect localized estimate: E b /N 0 loss 1.5dB Pilot Overhead: 1dB E b /N 0 loss 13
Simulation Results: Comparison different scenarios BER vs. E b /N 0, Default Values: f D,max : 0.015Δf @180 km/h Spreading Length: L=32 Users: K=32 Loc. Estimate: mean Alamouti FD filter: averaging 3 subcarriers Pilot Overhead: 1dB E b /N 0 loss 14
Summary & Outlook PACE for MIMO MC-CDMA: localized estimate for pilots only Filter in frequency direction to reduce noise Interpolate in time direction Simulation results indicate: Linear interpolation: large BER degradation only for high data rate, high speed scenario Robust Wiener filter: Smaller BER degradation for all scenarios Frequency averaging: reduces noise and smoothes CSI Outlook: Reduce pilot overhead Iterative channel estimation for high data rate, high speed scenario Thank you! 15