1. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 1 MIMO Mode Table for 802.11n Ravi Mahadevappa, ravi@realtek-us.com Stephan ten Brink, stenbrink@realtek-us.com Realtek Semiconductors, Irvine, CA May 2004 DCN 802.11-04/553r0

2. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 2 Why Different MIMO Modes Preliminaries, MIMO Modes Maximum ratio combining (MRC) „Circular Alamouti“ (CIRCAL) versus orthogonal space/time block codes (OSTBC) MIMO mode table „Circular SMX“ (CIRCSMX) Example, discussion 4x4 Conclusions Appendix, results 1x1 to 4x4 Overview

3. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 3 We should allow both –Small/low cost terminals (moderate speed) with e.g. 2 TX-ant. (for use in handhelds, digital cameras etc.) –And high speed terminals with e.g. up to 4 TX-ant. (for use in laptop computers etc.) Thus, there will be equipment with a variety of number of TX/RX antennas out there operating under the 802.11n-“umbrella” Solution: Switch between MIMO modes depending on available number of TX and RX antennas of the equipment involved –Spatial multiplex (SMX): High data rates, short range; optimal detection tends to be complex –Space time block codes (STBC), e.g. Alamouti 2x1 Long range, lower data rates; optimal detection is simple Should be transparent, 11a to 11n modes From today’s perspective: Max. N T =4 TX antennas, N R =4 RX antennas reasonable (full RF chain TX, RX antennas) Why Different MIMO Modes

4. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 4 General scenario: N T xN R –Obvious: For 1xN R use MRC at receiver –For N T >N R, in particular, for N T x1: Use simple TX diversity schemes based on space/time block codes No general construction known 2x1: Alamouti (perfect; full diversity, “rate 1”) Should 3x1, 4x1, full diversity, “rate 3/4”-codes be used? For N T >N R, N R >1 –use STBC and MRC at receiver –Or, use SMX with subset of TX antennas For N T <=N R : use SMX for high rate Preliminaries, MIMO Modes

5. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 5 1x1 QPSK 16QAM 64QAM Code Rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 1x2 MRC 1x3 MRC 1x4 MRC Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM Performance measure: required SNR for 10% PER Packet length: 1000 bits exponential decay, T rms = 60ns For simulation: Perfect –Channel estimation –Packet detection, synchronization – f off estimation –No clipping DAC/finite precision ADC –No front-end filtering 1x1 up to 1x4 MRC More receive antennas improve SNR (range)

6. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 6 N T =1234 N R =1 1x1 11a 2 1x2 11a MRC 3 1x3 11a MRC 4 1x4 11a MRC TX ant. RX ant. MIMO Mode Table (I) N T =1, simple MRC for any N R Table specifies the allowed MIMO modes depending on N T, N R

7. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 7 N T x1 Space/Time Block Codes (STBC) –Orthogonal STBC (OSTBC) allow simple ML detection –For N T x1, no general construction of OSTBC known with full spatial diversity and full rate –Moreover, it can be proved that no full rate OSTBC exists with N T >2 –2x1, Alamouti: only known OSTBC with full diversity and “rate 1” –3x1, 4x1, full diversity, spatial “rate 3/4” codes known We can easily build orthogonal N T x1 STBC based on the 2x1 Alamouti code with spatial “rate 1”; however, no full diversity –“circular Alamouti” 2(N T )x1 CIRCAL: For each channel symbol, only use 2 TX antennas out of the N T available ones Cycle through all N T !/(2! (N T -2)!) combinations (2 out of N T ) Example: N T =3, cycle period is 2. 3=6 combinations; N T =4, cycle period is 2. 6=12 combinations Experiment: Compare performance of 3x1, 4x1 full diversity rate 3/4 OSTBC with 2(3)x1, 2(4)x1 rate 1 CIRCAL for MIMO-OFDM TX Diversity with “Circular Alamouti”

8. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 8 Provides TX diversity N T =2 TX antennas, N R =1 RX antennas 2 channel uses, 2 symbols transmitted; thus, spatial rate 1 To read as –At time 1, transmit s 1 from antenna 1, s 2 from antenna 2 –At time 2, transmit -s 2 * from antenna 1, s 1 * from antenna 2 –Repeat pattern with new symbols s 1, s 2 2x1 Orthogonal Design (“Alamouti”)

9. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 9 Provides TX diversity N T =3 TX antennas, N R =1 RX antennas 4 channel uses, 3 symbols transmitted; thus, spatial rate 3/4 3x1 Orthogonal Design

10. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 10 Provides TX diversity N T =4 TX antennas, N R =1 RX antennas 4 channel uses, 3 symbols transmitted; thus, rate 3/4 4x1 Orthogonal Design

11. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 11 Based on 2x1 Alamouti (full diversity, spatial rate 1) Cycling for averaging good/bad MIMO subchannels N T =3 TX antennas, 2 used at a time, N R =1 RX antennas 6 channel uses, 6 symbols transmitted; thus, „rate 1“ Example: 2(3)x1 CIRCAL (or any other way of cycling through 2 TX antennas used at a time)

12. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 12 Based on 2x1 Alamouti (full diversity, spatial rate 1) Cycling for averaging good/bad MIMO subchannels N T =4 TX antennas, 2 used at a time, N R =1 RX antennas 12 channel uses, 12 symbols transmitted; thus, „rate 1“ Example: 2(4)x1 CIRCAL (or any other way of cycling through 2 TX antennas used at a time)

13. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 13 Based on 2x1 Alamouti (full diversity, spatial rate 1) Rotation for averaging good/bad MIMO subchannels N T TX antennas, n T =2 used at a time, N R RX antennas Use MRC (maximum ratio combining) for N R receive antennas Cycling through all P CIRCAL = N T !/(n T ! (N T -n T )!) combinations on a per OFDM symbol basis –since two OFDM symbols per Alamouti code are used, the cycle period in OFDM symbols is 2. P CIRCAL For better averaging in OFDM systems, TX antenna cycling can be applied over both, time index and subcarrier index 2(N T )xN R CIRCAL

14. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 14 -50510152025 0 10 20 30 40 50 60 70 SNR for 10 % PER (dB) Rate (Mbps) 3x1 2(3)x1 CIRCAL STBC “full diversity” Rate 3/4 3x1: 2(3)x1 CIRCAL versus OSTBC Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 3x1 STBC: „full diversity“, but (spatial) rate 3/4, rate loss 2(3)x1 CIRCAL outperforms 3x1 STBC –suffers no rate loss –cycling/rotating compensates some of the diversity losses 2(3)x1 CIRCAL better than 3x1 OSTBC

15. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 15 -50510152025 0 10 20 30 40 50 60 70 SNR for 10% PER (dB) Rate (Mbps) 4x1 2(4)x1 CIRCAL STBC “full diversity” Rate 3/4 4x1: 2(4)x1 CIRCAL versus OSTBC 4x1 STBC: „full diversity“, but (spatial) rate 3/4, rate loss 2(4)x1 CIRCAL outperforms 4x1 STBC –suffers no rate loss –cycling/rotating compensates some of the diversity losses 2(N T )x1 CIRCAL better than known OSTBC 2(N T )xN R CIRCAL by using MRC at the RX 2(4)x1 CIRCAL better than 4x1 OSTBC

16. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 16 0510152025 0 10 20 30 40 50 60 70 SNR for 10% PER (dB) Rate (Mbps) QPSK 16QAM 64QAM Code Rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 2(3)x1 CIRCAL 2(4)x1 CIRCAL 2x1 AL 2(3)x1 CIRCAL improves over 2x1 AL by about 0.5dB 2(4)x1 CIRCAL improves over 2x1 AL by about 1dB Only 2 antennas used at a time, but rotated through all combinations The antenna cycling captures more of the available diversity in the system More transmit antennas improve diversity, even if only used in rotation with Alamouti code 2x1, 2(3)x1, 2(4)x1 CIRCAL Gains of 2(N T )x1 CIRCAL over 2x1 AL saturate for N T =4 and more

17. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 17 N T =1234 N R =1 1x1 11a 2x1 AL 3x1 2(3)x1 CIRCAL 4x1 2(4)x1 CIRCAL 2 1x2 11a MRC 2x2 Low rate, 1: 2x2 ALMRC 3x2 1: 2(3)x2 CIRCAL/MRC 4x2 1: 2(4)x2 CIRCAL/MRC 3 1x3 11a MRC 2x3 1: 2x3 ALMRC 3x3 1: 2(3)x3 CIRCAL/MRC 4x3 1: 2(4)x3 CIRCAL/MRC 4 1x4 11a MRC 2x4 1: 2x4 ALMRC 3x4 1: 2(3)x4 CIRCAL/MRC 4x4 1: 2(4)x4 CIRCAL/MRC TX ant. RX ant. MIMO Mode Table (II) NT>=NR, low rate: Use Circular AL (CIRCAL) and MRC Table specifies the allowed MIMO modes depending on N T, N R

18. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 18 When N T >N R, we use same idea as with CIRCAL: –collect diversity by cycling through TX antennas n T (N T ) CIRCSMX: –Use n T TX antennas per channel use for SMX, out of N T >n T available antennas at transmitter –Cycle through all P CIRCSMX =N T !/(n T ! (N T -n T )!) combinations on a per OFDM symbol basis –The cycle period in OFDM symbols is P CIRCSMX For example, say we have N T =3 TX antennas, but only N R =2 RX antennas –Possible cycling for 2(3)x2 CIRCSMX: at time 1, transmit from antennas 1, 2 at time 2, transmit from antennas 1, 3 at time 3, transmit from antennas 2, 3 repeat the pattern Circular SMX

19. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 19 2x2 SMX 2x2 AL/MRC 2x2 2(3)x2 CIRCSMX 2(3)x2 CIRCAL/MRC 2(4)x2 CIRCSMX 2(4)x2 CIRCAL/MRC 64QAM 16QAM QPSK Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM CIRCSMX improves TX diversity 2(4)x2 SMX gains about 1dB over 2x2 SMX CIRCAL curves given as references 2x2 SMX, 2(3)x2, 2(4)x2 CIRCSMX For N T >N R : N R (N T )xN R CIRCSMX better than N R xN R SMX

20. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 20 N T =1234 N R =1 1x1 11a 2x1 AL 3x1 2(3)x1 CIRCAL 4x1 2(4)x1 CIRCAL 2 1x2 11a MRC 2x2 Low rate, 1: 2x2 ALMRC High rate, 2: 2x2 SMX 3x2 1: 2(3)x2 CIRCAL/MRC 2: 2(3)x2 CIRCSMX 4x2 1: 2(4)x2 CIRCAL/MRC 2: 2(4)x2 CIRCSMX 3 1x3 11a MRC 2x3 1: 2x3 ALMRC 2: 2x3 SMX 3x3 1: 2(3)x3 CIRCAL/MRC 2: 2(3)x3 CIRCSMX 3: 3x3 SMX 4x3 1: 2(4)x3 CIRCAL/MRC 2: 2(4)x3 CIRCSMX 3: 3(4)x3 CIRCSMX 4 1x4 11a MRC 2x4 1: 2x4 ALMRC 2: 2x4 SMX 3x4 1: 2(3)x4 CIRCAL/MRC 2: 2(3)x4 CIRCSMX 3: 3x4 SMX 4x4 1: 2(4)x4 CIRCAL/MRC 2: 2(4)x4 CIRCSMX 3: 3(4)x4 CIRCSMX 4: 4x4 SMX TX ant. RX ant. MIMO Mode Table (III) Any N T, N R, high rate: SMX, or Circular SMX (CIRCSMX)

21. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 21 4x4 2(4)x4 CIRCAL/MRC 2(4)x4 CIRCSMX 3(4)x4 CIRCSMX 4x4 SMX Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM The rate/SNR envelope includes –2(4)x4 CIRCAL for lowest rates –2(4)x4 CIRCSMX –3(4)x4 CIRCSMX –4x4 SMX for highest rate To simplify: –Omit 2(4)x4 CIRCSMX –Use 4x4 SMX for highest two rates –Use 2(4)x4 CIRCAL for lowest Example, Discussion of 4x4 For N T >N R : N R (N T )xN R CIRCSMX better than N R xN R SMX

22. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 22 123...NTNT 1 1x1 11a 2x1 AL 3x1 2(3)x1 CIRCAL... N T x1 2(N T )x1 CIRCAL 2 1x2 11a MRC 2x2 Low rate, 1: 2x2 ALMRC High rate, 2: 2x2 SMX 3x2 1: 2(3)x2 CIRCAL/MRC 2: 2(3)x2 CIRCSMX... N T x2 1: 2(N T )x2 CIRCAL/MRC 2: 2(N T )x2 CIRCSMX 3 1x3 11a MRC 2x3 1: 2x3 ALMRC 2: 2x3 SMX 3x3 1: 2(3)x3 CIRCAL/MRC 2: 2(3)x3 CIRCSMX 3: 3x3 SMX... N T x3 1: 2(N T )x3 CIRCAL/MRC 2: 2(N T )x3 CIRCSMX 3: 3(N T )x3 CIRCSMX.............................. NRNR 1xN R 11a MRC 2xN R 1: 2xN R ALMRC 2: 2xN R SMX 3xN R 1: 2(3)xN R CIRCAL/MRC 2: 2(3)xN R CIRCSMX … For N T >N R : N R : N R (N T )xN R CIRCSMX For N T <=N R : N T : N T xN R SMX... N T xN R 1: 2(N T )xN R CIRCAL/MRC 2: 2(N T )xN R CIRCSMX 3: 3(N T )xN R CIRCSMX … For N T >N R : N R : N R (N T )xN R CIRCSMX For N T <=N R : N T : N T xN R SMX TX RX MIMO Mode Table (IV), any N T, N R

23. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 23 Conclusions Use 2(N T )x1 CIRCAL rather than full diversity, spatial rate 3/4 STBC for N T =3, 4 Always use all receive antennas Not always use all transmit antennas –Only when very high rates are desired –For medium rates, it is better to use less antennas, but higher modulation/rate; to have at least one excess antenna at receiver CIRCSMX N R (N T )xN R always better (slightly) than SMX N R xN R For low-complexity, suboptimal MIMO detection, excess antennas pay off –For example 3x3 for medium rate, use 2(3)x3 CIRCSMX (high rate code/modulation) rather than 3x3 SMX (medium code rate/modulation) Only use 3x3 SMX when really high rates are desired

24. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 24 Rate versus SNR-charts for N T xN R = 1x1 to 4x4 Appendix

25. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 25 Performance Criteria/Abbreviations Receiver sensitivity for 10% PER Abbreviations: –SEL: selection diversity at RX –MRC: maximum ratio combining at RX –AL/MRC: Alamouti Space/Time with MRC at RX [7,8] –SMX: spatial multiplexing (i.e. MIMO mode, [4,5,6]) –n T (N T ) CIRCAL: “circular Alamouti”, using n T =2 antennas per channel use, out of N T >n T available antennas at transmitter, and cycling through all P CIRCAL = N T !/(n T ! (N T -n T )!) combinations on a per OFDM symbol basis; since two OFDM symbols per Alamouti code are used, the cycle period in OFDM symbols is 2. P CIRCAL –n T (N T ) CIRCSMX: “circular SMX”, using n T transmit antennas per channel use for SMX, out of N T >n T available antennas at transmitter, and cycling through all P CIRCSMX =N T !/(n T ! (N T -n T )!) combinations on a per OFDM symbol basis; the cycle period in OFDM symbols is P CIRCSMX MIMO detection used in following plots –ZF and APP post processing –Note: ZF with APP post processing provides close to optimal performance for higher order modulation (increasing number of excess antennas); for small constellations (QPSK), i.e., low-rate communication, ZF/APP suffers significant performance loss; that’s the main reason why AL-type of STBC are a good choice for low-rate communication

26. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 26 802.11a PHY simulation environment, plus –Higher order QAM constellations –Higher/lower channel code rates –TX/RX diversity/MIMO OFDM ZF detection and soft post processing (shown in plots) APP and reduced APP detection Perfect channel knowledge/synchronization Idealized multipath MIMO channel –Sub-channels independent; exponential decay, T rms = 60ns –Quasi static (channel stays constant during one packet) Packet length: 1000 bits Perfect –Channel estimation –Packet detection, synchronization – f off estimation –No clipping DAC/finite precision ADC –No front-end filtering Simulation Environment/Assumptions

27. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 27 1x1 QPSK 16QAM 64QAM Code Rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 1x2 MRC 1x3 MRC 1x4 MRC Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM More receive antennas improve SNR (range) 1x1 up to 1x4 MRC

28. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 28 0510152025 0 10 20 30 40 50 60 70 SNR for 10% PER (dB) Rate (Mbps) 2x1 AL QPSK 16QAM 64QAM Code Rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 2x1 Alamouti STBC

29. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 29 -5051015202530 0 20 40 60 80 100 120 140 SNR for 10% PER (dB) Rate (Mbps) 2x2 SMX 2x2 AL/MRC 2x2 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM Low rate: AL/MRC High rate: SMX 2x2

30. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 30 -10-50510152025 0 20 40 60 80 100 120 140 SNR for 10 % PER (dB) Rate (Mbps) 2x3 2x3 SMX 2x3 AL/MRC Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM Low rate AL/MRC High rate SMX 2x3

31. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 31 -10-505101520 0 40 60 80 100 120 140 SNR for 10 % PER (dB) Rate (Mbps) 2x4 2x4 SMX 2x4 AL/MRC Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM Low rate AL/MRC High rate SMX Almost all rates can be covered with SMX only 2x4

32. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 32 -50510152025 0 10 20 30 40 50 60 70 SNR for 10 % PER (dB) Rate (Mbps) 3x1 2(3)x1 CIRCAL STBC “full diversity” Rate 3/4 3x1 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 3x1 STBC: „full diversity“, but (spatial) rate 3/4, rate loss 2(3)x1 CIRCAL outperforms 3x1 STBC (suffers no rate loss; no significant loss due to smaller diversity; circulating/rotating compensates some of the diversity losses, see next slide)

33. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 33 -5051015202530 0 20 40 60 80 100 120 140 SNR for 10 % PER (dB) Rate (Mbps) 3x2 2(3)x2 CIRCAL/MRC 2(3)x2 CIRCSMX 3x2 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 2(3)x2 CIRCAL/MRC better for low data rates 2(3)x2 CIRCSMX better for higher data rates (no rate loss); for low data rates: loss due to ZF/APP detection for small constellation size

34. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 34 -10-505101520253035 0 20 40 60 80 100 120 140 160 180 200 SNR for 10 % PER (dB) Rate (Mbps) 3x3 3x3 SMX 2(3)x3 CIRCAL/MRC 2(3)x3 CIRCSMX2 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 3x3 SMX only for high rate 2(3)x3 CIRCSMX better for medium rate (easier detection since excess antenna) 3x3

35. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 35 -10-50510152025 0 20 40 60 80 100 120 140 160 180 200 SNR for 10 % PER (dB) Rate (Mbps) 3x4 3x4 SMX 2(3)x4 CIRCAL/MRC 2(3)x4 CIRCSMX Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 3x4: highest rate 2(3)x4: medium rate 2(3)x4 CIRCAL/MRC: lowest rate 3x4

36. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 36 -50510152025 0 10 20 30 40 50 60 70 SNR for 10% PER (dB) Rate (Mbps) 4x1 2(4)x1 CIRCAL STBC “full diverity” Rate 3/4 4x1 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 4x1 STBC: „full diversity“, but (spatial) rate 3/4, rate loss (as with 3x1 STBC case) 2(4)x1 CIRCAL outperforms 4x1 STBC (suffers no rate loss; no significant loss due to smaller diversity; circulating/rotating compensates some of the diversity losses, see next slide)

37. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 37 -5051015202530 0 20 40 60 80 100 120 140 SNR for 10% PER (dB) Rate (Mbps) 4x2 2(4)x2 CIRCSMX 2(4)x2 CIRCAL/MRC 4x2 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 2(4)x2 CIRCAL/MRC better for low data rates

38. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 38 -10-505101520253035 0 20 40 60 80 100 120 140 160 180 200 SNR for 10% PER (dB) Rate (Mbps) 2(4)x3 CIRCAL/MRC 4x3 2(4)x3 CIRCSMX 3(4)x3 CIRCSMX 4x3 Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM 2(4)x3 CIRCAL/MRC better for low data rates (hardly) 2(4)x3 CIRCSMX better for medium data rates 3(4)x3 CIRCSMX better for high data rates Note for all SMX at low data rates: loss due to ZF/APP detection for small constellation size; that is why 2(4)x3 CIRCSMX better than 3(4)x3 CIRCSMX at medium data rates

39. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 39 4x4 2(4)x4 CIRCAL/MRC 2(4)x4 CIRCSMX 3(4)x4 CIRCSMX 4x4 SMX Code rates 1/4, 1/3, 1/2, 2/3, 3/4, 7/8 Modulation QPSK, 16QAM, 64QAM Similar situation as for 3x3 4x4 SMX only for highest rates 4x4

40. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 40 Linear dispersion (LD) codes [9] –Very general set-up –Powerful codes designed using an optimization algorithm –Good performance for any N T, N R –However, generally requires ML/APP detection (e.g. Sphere detection); complexity is an issue Delay diversity –good for MIMO-OFDM since guard interval eliminates need for multi-tap channel equalizer –However, it has been shown that to obtain „full diversity“, the delay needs to be bigger than the guard interval (and, in turn, would require a multi-tap channel equalizer, counteracting the benefits of OFDM transmission) Other Designs Considered

41. doc.: IEEE 802.11-04/553r0 Submission May 2004 Ravi Mahadevappa, Stephan ten Brink, Realtek Slide 41 Some References [1]IEEE Std 802.11a-1999, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, High-speed Physical Layer in the 5 GHz Band [2]S. Sandhu, M. Ho, “Transmit diversity for MIMO OFDM”, IEEE 802.11-03/847r0, Nov. 2003 [3]H. Sampath, R. Narasimhan, “Advantages and drawbacks of circular delay diversity for MIMO-OFDM”, IEEE 802.11-04/075r1 [4]J. H. Winters, J. Salz, R. D. Gitlin, “The impact of antenna diversity on the capacity of wireless communication systems”, IEEE Trans. Commun., vol. 42, no. 2/3/4, pp. 1740- 1751, Feb./Mar./Apr. 1994 [5]G. J. Foschini, “Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas”,Bell Labs. Tech. J., vol. 1, no. 2, pp. 41- 59, 1996 [6]H. Sampath, S. Talwar, J. Tellado, V. Erceg, A. Paulraj, “A fourth-generation MIMO-OFDM broadband wireless system: Design, performance, and field trial results”, IEEE Commun. Mag., pp. 143-149, Sept. 2002 [7]S. M. Alamouti, “A simple transmit diversity technique for wireless communications”, IEEE J. on Select. Areas in Commun., vol. 16, pp. 1451-1458, Oct. 1998 [8]V. Tarokh, H. Jafarkani, A. R. Calderbank, “Space-time block codes from orthogonal designs”, IEEE Trans. Inform. Theory, vol. 45, pp. 1456-1467, July 1999 [9]B. Hassibi, B. M. Hochwald, “High-rate codes that are linear in space and time,” IEEE Transactions on Information Theory, July 2002.