1. Optimized MIMO Aryan Saed, Jung Yee, Jim Wight
Month 2002
doc.: IEEE
August 2004
Optimized MIMO
Aryan Saed, Jung Yee, Jim Wight
ICEFYRE Semiconductor Inc.
n-optimized-mimo.ppt
Aryan Saed, ICEFYRE Semiconductor Inc.
John Doe, His Company

2. A Strawman (what could happen)
August 2004
A Strawman (what could happen)
Mandatory: HT-n (High Throughput n)
SMX 2x2
MAC with feedback allowance for fast rate adaptation (SNR based), feedback fields for CSI, calibration, beamforming, with parts of “best practices” for convergence
20MHz for maximum cell density and pan-regulatory coverage
BCC 64QAM (64pt FFT) for proven robustness
performance meets PAR
MAC is ready for TGn-options (below)
TGn label is credible and label is fast to market on the shelves from broad range of vendors
ready for high volume, high yield
pan regulatory
dense cell planning
Aryan Saed, ICEFYRE Semiconductor Inc.

3. A Strawman (what could happen)
August 2004
A Strawman (what could happen)
Optional: for VHT-n (optimizationa for Very High Throughput)
SMX 3x3, 4x4,
STBC NxM (M<N) for heavy-AP and light-STA, e.g. 4x2
Beamforming for low RF cost where doppler allows
40MHz for low RF cost where cell density and regulations allow
High performance FEC (LDPC/TC) for extended reach where latency allows
128QAM & 256QAM where high quality RF (phase noise and distortion) allows
(combinations of 1-6 allowed)
enables very high throughput
flexibility in terms of how very high throughput may be achieved, depending on application and deployment circumstances
ready for future analog and RF design improvements and circuit techniques
Aryan Saed, ICEFYRE Semiconductor Inc.

4. A Strawman (what could happen)
August 2004
A Strawman (what could happen)
Vendor discretionary optimization:
Mandatory mode has MAC level feedback mechanisms in place for
RF level beamforming
Beamforming matrix calculation (SVD, other), includes spatial averaging (Hadamard or Fourier) for broadcast
Leverages 11n into realms of vendor distinction in broad or niche high performance markets
Enables vendors / labs to experiment with future technology for ultra high throughput standards compliant systems
Aryan Saed, ICEFYRE Semiconductor Inc.

5. Proposal for Optimized MIMO
Month 2002
doc.: IEEE
August 2004
Proposal for Optimized MIMO
MIMO using channel diagonalization
Allow vendor discretionary optimization for maximum rate and/or maximum reach without adding complexity and options
Allow vendor discretionary optimization of hardware and overhead through mutual signaling of MIMO complexity without adding options
Aryan Saed, ICEFYRE Semiconductor Inc.
John Doe, His Company

6. Diagonalization (Beam forming)
August 2004
Diagonalization (Beam forming)
Aryan Saed, ICEFYRE Semiconductor Inc.

7. Diagonalization by SVD
August 2004
Diagonalization by SVD
Singular Value Decomposition (SVD):
S is diagonal (contains singular values of H)
Aryan Saed, ICEFYRE Semiconductor Inc.

8. Diagonalization by QR-Eig
August 2004
Diagonalization by QR-Eig
Step 1: QR-Triangularization of H
Step 2: Eigenvector decomposition of Q
Aryan Saed, ICEFYRE Semiconductor Inc.

9. Diagonalization by QR-Eig
August 2004
Diagonalization by QR-Eig
and
Set:
Then:
Aryan Saed, ICEFYRE Semiconductor Inc.

10. Implications of SVD SVD maximizes theoretical (Shannon) capacity
August 2004
Implications of SVD
SVD maximizes theoretical (Shannon) capacity
SNR per stream is ranked according to singular values of channel matrix H
In practical TGn channels the SNR per sub-carrier can differ by 10dB..20dB as a result of singular value spread
Aryan Saed, ICEFYRE Semiconductor Inc.

11. August 2004
Implications of SVD
To maximize practical capacity, rate depends on SNR
Usually only the strongest mode can support maximum rate (e.g. 256-QAM)
rates in all other modes are then backed-off, some only support a fraction (half, quarter, eighth) of the maximum (e.g. 16-QAM)
Aryan Saed, ICEFYRE Semiconductor Inc.

12. August 2004
Implications of SVD
Rate across multiple streams not equal. Example: one stream at 54Mbps, another at 24Mbps. Data-bit multiplexing becomes a PHY function, keeping both streams equal duration (e.g ms)
Large portion of bits on the strongest mode. Remainder of bits on remaining weaker modes.
Aryan Saed, ICEFYRE Semiconductor Inc.

13. Implications of QR-Eig
August 2004
Implications of QR-Eig
QR-Eig has equal SNR across streams
Same rate and length per stream
Same interleaver and FEC
Same latency, and same PER
Data-bit multiplexing can be a MAC function: one MSDU per stream.
Aryan Saed, ICEFYRE Semiconductor Inc.

14. Which to choose? Depends on application: maximum rate or maximum reach
August 2004
Which to choose?
Depends on application: maximum rate or maximum reach
Depends on MAC preference: - individual control over each MIMO stream (separate PER at 10%, separate ACK) or - view all streams as one packet (joint PER at 10%, joint ACK and larger data buffer)
Aryan Saed, ICEFYRE Semiconductor Inc.

15. Simulation details TGn Comparison Criteria IEEE 802.11-03/814r31
August 2004
Simulation details
TGn Comparison Criteria IEEE /814r31
IM1 RAPP PA model at 14dBm output
IM4 Phase Noise
IM5 Noise Figure
IM6 half-wavelength
2x2 MIMO
Aryan Saed, ICEFYRE Semiconductor Inc.

16. August 2004
Comparison
Aryan Saed, ICEFYRE Semiconductor Inc.

17. August 2004
Rate-pair Selection
Aryan Saed, ICEFYRE Semiconductor Inc.

18. August 2004
Objective
To allow vendor choice of algorithm : SVD, QR-Eig or other vendor algorithm
optimized for particular usage scenario (reach, rate, latency, low cost RF, rapid 11n deployment). Consider that IM4 Phase Noise is 0.5deg rms IM1 RAPP PA model is -40dB EVM
Aryan Saed, ICEFYRE Semiconductor Inc.

19. August 2004
Objective
Ensure mechanisms are in place in TGn to allow calculation of T and R matrix at one side of the link, and
to communicate matrix from STA to AP or from AP to STA so that matrix can be operated at both sides (Tx and Rx side) of link
Aryan Saed, ICEFYRE Semiconductor Inc.

20. Per-sub-carrier Diagonalization
August 2004
Per-sub-carrier Diagonalization
52 n-by-n (e.g. 2x2) complex matrix coefficients. Overhead in feedback.
Aryan Saed, ICEFYRE Semiconductor Inc.

21. Grouped Diagonalization
August 2004
Grouped Diagonalization
Grouping: fewer n-by-n (egg 2x2) complex matrix coefficients. Lower overhead.
Aryan Saed, ICEFYRE Semiconductor Inc.

22. August 2004
Objective
To ensure mechanisms are in place in standard to allow reduced overhead from coefficient exchange
Example: AP requires grouping of 4 sub-carriers due to its reduced hardware complexity.
Aryan Saed, ICEFYRE Semiconductor Inc.

23. August 2004
Objective
Groups of 4 or 8 sub-carriers reduces T and R matrix calculation and update overhead by a factor of 4 and 8 respectively.
STA and AP must communicate their capabilities
Aryan Saed, ICEFYRE Semiconductor Inc.

24. Overview Some further implications of MIMO with diagonalization:
August 2004
Overview
Some further implications of MIMO with diagonalization:
Stream splitting: FEC, interleaving
MIMO Header
Aryan Saed, ICEFYRE Semiconductor Inc.

25. MIMO Stream splitting August 2004
Aryan Saed, ICEFYRE Semiconductor Inc.

26. Diagonalization in an 11n packet
August 2004
Diagonalization in an 11n packet
Aryan Saed, ICEFYRE Semiconductor Inc.

27. MIMO Headers Service Field signifies
August 2004
MIMO Headers
Service Field signifies
During Link Set-up, sub-carrier grouping: no grouping (52 matrices), groups of four (16 matrices), groups of eight (8 matrices).
During Link Set-up, supported rates, equal rate across all streams or stream dependent rate
Aryan Saed, ICEFYRE Semiconductor Inc.

28. MIMO Headers Service Field (STA side calculation)
August 2004
MIMO Headers
Service Field (STA side calculation)
During Link operation: regular feedback of channel estimation from AP to STA for matrix calculations
During Link operation: regular update of Tx side diagonalization coefficients from STA to AP
Aryan Saed, ICEFYRE Semiconductor Inc.

29. MIMO Link Set-up AP to STA: Transmit intent for 11n Link
August 2004
MIMO Link Set-up
AP to STA: Transmit intent for 11n Link
STA to AP: Acknowledge with 11n enabled packet. Includes 11n header for channel estimation and multiple branch utilization
AP to STA: send back AP-side estimated channel matrix and AP-side diagonalization capability (sub-carrier grouping)
Aryan Saed, ICEFYRE Semiconductor Inc.

30. August 2004
MIMO Link Set-up
STA to AP: AP-side diagonalization coefficients using STA-side discretionary algorithm
AP to STA: AP-side (Tx) and STA-side (Rx) diagonalization applied
STA to AP: AP-side (Rx) and STA-side (Tx) diagonalization applied
Aryan Saed, ICEFYRE Semiconductor Inc.

31. MIMO Link Maintenance Continual feedback per packet
August 2004
MIMO Link Maintenance
Continual feedback per packet
Regular rejuvenation of AP channel estimate that is mirrored at STA side
Regular recalculation of AP diagonalization matrix at STA side.
Aryan Saed, ICEFYRE Semiconductor Inc.

32. Proposal Summary Frequency domain diagonalization
August 2004
Proposal Summary
Frequency domain diagonalization
Vendor discretionary algorithm (e.g. SVD, QR with Eigenvector, other)
Orthogonal LTS for MIMO Channel Estimation
Vendor discretionary grouping of sub carriers (no grouping, grouping with 4, or 8 adjacent sub carriers)
Aryan Saed, ICEFYRE Semiconductor Inc.

33. August 2004
Thank you.
Aryan Saed, ICEFYRE Semiconductor Inc.