doc.: IEEE /794r1 Submission Slide 1 André Bourdoux (IMEC) July 2004 Preambles for MIMO channel estimation André Bourdoux Bart Van Poucke Liesbet Van der Perre IMEC, Wireless Research

doc.: IEEE /794r1 Submission Slide 2 André Bourdoux (IMEC) July 2004 Motivation MIMO-OFDM is key to achieve 100 Mbps at the MAC SAP Conventional SISO preamble (11.a, g) is not sufficient MIMO channel estimation requires a new preamble

doc.: IEEE /794r1 Submission Slide 3 André Bourdoux (IMEC) July 2004 SISO Preamble (1) STS used for  AGC, Packet detection(Power measurement)  Coarse timing acquisition(Auto-correlation)  Coarse Carrier freq. acquisition(Auto-correlation) SISO Preamble BBBBBBBBBB CP CC SIG CP Data LTSSTS LTS used for  Fine timing acquisition(Auto/cross-correlation)  Fine Carrier freq. acquisition(Auto-correlation)  Channel estimation(direct, least-square)  IQ imbalance estimation(specific algorithm)

doc.: IEEE /794r1 Submission Slide 4 André Bourdoux (IMEC) July 2004 SISO Preamble (2)  Desirable properties for STS  Short periodicity: CFO acquisition range = 1/2T B =  625 kHz  Long periodicity: > max excess delay (T B = 800 ns  240 m.)  Low PAPR  Desirable properties for LTS  Low auto-correlation sidelobes  Double-length CP to accommodate coarse timing estimation  repeated C sequence allows  Long auto-correlation for accurate CFO estimation  3 dB SNR improvement for Channel estimation input data  Low PAPR

doc.: IEEE /794r1 Submission Slide 5 André Bourdoux (IMEC) July 2004 MIMO Preamble Requirements:  SISO requirements:  AGC, packet detection  CFO estimation  Timing estimation  MIMO requirements  Detect number of TX antennas (N T )  On each RX antenna, differentiate and Estimate N T channels from one received signal  Low cross-correlation between TX antenna signals  Legacy requirements  When N T =1, compatible with SISO transmission (11a,g)

doc.: IEEE /794r1 Submission Slide 6 André Bourdoux (IMEC) July 2004 Assumptions for MIMO preamble  Reuse of SISO preamble (STS, LTS, SIG) for legacy  Coarse/fine timing and CFO is achieved before channel estimation  AGC from TX1 only cannot be reused, second AGC needed  # TX antennas is known before channel estimation  AGC is settled before channel estimation  CP for MIMO channel estimation can be 16 samples long  Total energy available per “SISO” channel is constant STS TX 1 TX 2 TX 3 TX 4 Data 2 Data 3 Data 1 Data 4 LTSSIGSIG2STS1LTS1 LTS2 LTS3 LTS4 Legacy preamble - # TX antennas - MIMO mode - … - Second AGC- Multi TX antenna Channel estimation STS2 STS3 STS4

doc.: IEEE /794r1 Submission Slide 7 André Bourdoux (IMEC) July 2004 Orthogonality between TX antennas  We focus on the part of the preamble for Multi-TX antenna channel estimation  LTS sequences from different TX antennas must be differentiated  LTS sequences can be made orthogonal in -Time:TDM -Frequency:FDM -Code:CDM -Hybrid (for N T > 2):TDM-FDM TDM-CDM FDM-CDM

doc.: IEEE /794r1 Submission Slide 8 André Bourdoux (IMEC) July 2004 TDM preamble LTS CP C CData 1 2 x 3.2 µs0.8 µs TX 1 TX 2 Data 2 CP C C C 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 Data 2 Data 3 Data 1 CP C C C C 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 TX 4 Data 2 Data 3 Data 1 Data 4

doc.: IEEE /794r1 Submission Slide 9 André Bourdoux (IMEC) July 2004 TDM preamble  Minimum duration: N T x (16+128) samples  N T x 7.2µs  Processing (per RX antenna):  Estimate = measurement: N T x SISO Channel estim.  Least square : smoothes freq-domain channel estimate with time-domain constraint; N T x 2 x N c x L complex MACs  Reuse of existing blocks (IP)  Allows IQ Imbalance compensation based on preamble  Requires higher average power per antenna during LTS  10log 10 (N T ) dB more TX power per TX antennas  RX AGC is a problem (1 TX antenna active at a time)  AGC values must be the same as during payload transmission

doc.: IEEE /794r1 Submission Slide 10 André Bourdoux (IMEC) July 2004 FDM preamble 2 x 3.2 µs0.8 µs C1 C2 C3 CP C3 CP C2 CP C1Data 1 Data 2 Data 3 TX 1 TX 2 TX 3 C1 CP C1 2 x 3.2 µs0.8 µs C1 C2 C3 C4 CP C4 CP C3 CP C2Data 2 Data 3 Data 1 Data 4 TX 1 TX 2 TX 3 TX 4 TX 1 TX 2 2 x 3.2 µs0.8 µs C1 C2 CP C2 CP C1Data 1 Data 2

doc.: IEEE /794r1 Submission Slide 11 André Bourdoux (IMEC) July 2004 FDM preamble  Different subsets of sub-carriers used on the TX antennas  For 52 sub-carriers and 4 TX antennas, only 13 sub-carriers per training symbol. … … IFFT C1 C2 IFFT TX 1 TX 2 ………

doc.: IEEE /794r1 Submission Slide 12 André Bourdoux (IMEC) July 2004 FDM preamble  Minimum duration: 1x(16+128) samples  1x 7.2µs  Duration for same energy as TDM: 16 + N T x128 samples  N T x6.4µs  Processing (per RX antenna):  Freq domain interpolator: sensitivity to phase slope  Least square : 2 x N C x (N T + 1) complex MACs  Same average RX power as during payload reception

doc.: IEEE /794r1 Submission Slide 13 André Bourdoux (IMEC) July 2004 CDM preamble LTS CP -C CP CData 1 2 x 3.2 µs0.8 µs TX 1 TX 2 Data 2 CP C C C C C C 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 TX 4 Data 2 Data 3 Data 1 Data 4 CP C C C C C C -C CP -C CP -C CP -C CP -C CP -C CP C C C 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 Data 2 Data 3 Data 1 CP C C C C C -C CP -C CP -C CP -C

doc.: IEEE /794r1 Submission Slide 14 André Bourdoux (IMEC) July 2004 CDM preamble  Minimum duration: 2 TX: 2x(16+128) samples  2 x 7.2µs 3 or 4 TX: 4x(16+128) samples  4 x 7.2µs  Processing (per RX antenna):  Complex additions/substractions for “despreading”  The rest is same as TDM  Same average RX power as during payload reception  LTS for N T =3 must be same length as for N T =4

doc.: IEEE /794r1 Submission Slide 15 André Bourdoux (IMEC) July 2004 TDM-FDM preamble CP C2 CP C1 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 TX 4 Data 2 Data 3 Data 1 Data 4 C1 CP C2 CP C1 C2 C1 C2  Duration for 4 TX antennas: 2x(16+256) samples  2 x 13.6µs  Processing (per RX antenna ):  Least square for the FDM part, the rest is same as TDM  Problem of average RX power (in TDM) not completely eliminated

doc.: IEEE /794r1 Submission Slide 16 André Bourdoux (IMEC) July 2004 CP C C C 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 TX 4 Data 2 Data 3 Data 1 Data 4 CP C -C CP -C CP -C CP -C  Duration for 4 TX antennas: 4x(16+128) samples  4 x 7.2µs  Processing (per RX antenna ):  Complex additions/substractions for “despreading”  The rest is same as TDM  Problem of average RX power (in TDM) not completely eliminated TDM-CDM preamble

doc.: IEEE /794r1 Submission Slide 17 André Bourdoux (IMEC) July 2004 FDM-CDM preamble CP C2 CP C1 2 x 3.2 µs0.8 µs TX 1 TX 2 TX 3 TX 4 Data 2 Data 3 Data 1 Data 4 C1 CP -C2 CP -C1 C2 -C1 -C2 CP C2 CP C1 C2 CP C2 CP C1 C2  Duration for 4 TX antennas: 2x(16+256) samples  2 x 13.6µs  Processing (per RX antenna ):  Complex additions/substractions for “despreading”  The rest is same as TDM  Can also be used for N T =3

doc.: IEEE /794r1 Submission Slide 18 André Bourdoux (IMEC) July 2004 Performance of the various preambles  In principle, TDM and CDM have the same performance  FDM performance degrades for N T =4 because of coarser frequency sampling  Simulations show Channel Estimation Mean-squared Error for preamble options and N T =2, 4  In all simulations  total power / N T is constant  total energy / N T is constant (except for CPs)

doc.: IEEE /794r1 Submission Slide 19 André Bourdoux (IMEC) July 2004 Channel estimation error, N T =2 Impact of zero- carriers on least- square Worse estimation without least- square

doc.: IEEE /794r1 Submission Slide 20 André Bourdoux (IMEC) July 2004 Channel estimation error, N T =4 Impact of zero- carriers on least- square Impact of coarse frequency sampling (FDM) Worse estimation without least- square CDM-FDM: Best performance no problem with AGC

doc.: IEEE /794r1 Submission Slide 21 André Bourdoux (IMEC) July 2004 Our advice for n  Several preamble structures are possible for MIMO channel estimation  Preambles with simultaneous transmission from all TX antennas are mandatory  no problem from AGC  Least-square solution provides better estimate, is mandatory for FDM-based preambles