1. August 2003 Daniel Nissani (Nissensohn)1 dnissani@bezeqint.net A Novel MIMO Transmission Method proposed herein as 802.11 TGn PHY element

2. August 2003 Daniel Nissani (Nissensohn)2 Synopsis VCXC: A novel MIMO Transmission method is proposed –based on fundamentally different principles than those of Interference Cancellation, Adaptive Modulation VCXC: Virtual Channel Cross-talk Control EXCELLENT BER PERFORMANCE –e.g. 20 db performance gain @ 1E-6 BER for 1 st sub-stream relative to ZF/ LSE/ MMSE EXTREME REAL TIME SIMPLICITY –single matrix-vector multiplication LOW OVERHEAD –e.g. just 6 symbols Pre-Amble for L = R = 3

3. August 2003 Daniel Nissani (Nissensohn)3 Motivation Wireless MIMO based systems promise significant Capacity enhancement To keep this promise MIMO systems have to overcome an inherent cross-talk interference problem –due to imperfect Channel Matrix estimation no similar phenomenon is observed in scalar (SISO) or vector (SIMO/ MISO/ diversity) wireless channels

4. August 2003 Daniel Nissani (Nissensohn)4 Motivation (cont.) Plain ZF, LSE or MMSE methods do not solve this cross- talk problem –very poor BER performance BLAST, other (Sequential/ Parallel) Interference Cancellation schemes –‘brute force’ approach, complex, iterative Structures Adaptive Modulation techniques –‘live with it’ approach, limited performance improvement A ‘Gordian Knot Cut’ approach is herein proposed

5. August 2003 Daniel Nissani (Nissensohn)5 MIMO Basic Model The Cross-Talk Interference Component y = H A s + n = H x + n r = (H n ’ C -1 H n ) -1 H n ’ C -1 y e = r – x = ((H n ’ C -1 H n ) -1 H n ’ C -1 H – I) x + ((H n ’ C -1 H n ) -1 H n ’ C -1 n

6. August 2003 Daniel Nissani (Nissensohn)6 Motivation (cont.) In a broad class of MIMO systems Channel Matrix information is, or can be available, at both the Transmitting and Receiving sides –a natural situation in TDD based systems (like WLAN) Channel Reciprocity –possible in FDD based systems (like most Cellular) by means of return channel feedback a natural extension to Closed Loop Tx Diversity Procedures (3GPP TS 25.214) We’ll focus on these hereon

7. August 2003 Daniel Nissani (Nissensohn)7 MIMO Model, Channel Info at Tx and Rx H n = U n D n V n ’, H = U D V’ x = V n A s y = H x + n r = U n ’ y = U n ’ H x + U n ’ n = (U n ’ U) D (V’ V n ) A s + U n ’ n = B s + U n ’ n

8. August 2003 Daniel Nissani (Nissensohn)8 In the ‘naïve’, perfect channel estimate case H n = H, r = (D A) s + U’ n –r is simply a scaled version of the transmitted data –no Cross-Talk Interference In the realistic case H n H, U n U, D n D, etc. –B is a non-diagonal perturbation of of the diagonal matrix (D A) –all components of r include cross-talk energy from each other MIMO Model, Channel Info at Tx and Rx The ideal (‘naïve’) and realistic Channel Estimation cases

9. August 2003 Daniel Nissani (Nissensohn)9 Motivation (cont.) We propose herein a novel solution to this cross-talk interference problem –based on Channel Matrix transformation –fundamentally different than Interference Cancellation, Adaptive Modulation, etc.

10. August 2003 Daniel Nissani (Nissensohn)10 A Channel Matrix Figure Of Merit We define a set of Channel Matrix Figures of Merit, S i –reflect Matrix Quality w.r.t. the Cross-Talk effect –e.g. S i = S i (H; cov(H n ) = S i (H ;  a ) = E [  xi ] over a population of perturbations H n i = 1,2,…M; M <= rank(H) is the number of user data sub- streams s i other similar definitions of S i are possible –S i (H) is a Random Variable since H is a Random Matrix

11. August 2003 Daniel Nissani (Nissensohn)11 S i Cumulative Distribution Function L = R = 3,  a = 20db, s 1 and s 2 data sub-streams x x NOT ALL CHANNEL MATRICES ARE BORN EQUAL!

12. August 2003 Daniel Nissani (Nissensohn)12 S i representation It can be shown that S i (H) = S i (D(H)) –a smooth, continuous function in D-space, the Singular Values of H (= U D V’) –this is a convenient, lower dimensionality (and much more efficient) representation

13. August 2003 Daniel Nissani (Nissensohn)13 D2D2 D1D1 D3D3 Example: S 2 (D(H)) scatter plot, L = R = 3,  a = 20db

14. August 2003 Daniel Nissani (Nissensohn)14 Typical, Schematic plots of S i (D(H)) in D-space F i j (D) = 0 are ‘iso-merit’ curves in D-space Can be approximated by low order polynomials in D G 2 3 is a ‘Good’ region S1S1 S2S2

15. August 2003 Daniel Nissani (Nissensohn)15 Taking Advantage of this Insight A Pre-Equalizer P

16. August 2003 Daniel Nissani (Nissensohn)16 Taking Advantage of this Insight (cont.) We modify our original channel H so that a more favorable channel H m (in the sense of better cross-talk SNR) is observed between Tx and Rx sides P = V n D n -1 D m H m = H n P = (U n D n V n ’ ) (V n D n -1 D m ) = U n D m I V m = I U m = U n

17. August 2003 Daniel Nissani (Nissensohn)17 Taking Advantage of this Insight A simple, Sub-Optimal scheme Solve for D m so that is satisfied, subject to This minimizes the SNR loss incurred in P application –A is re-scaled by so that transmission power is preserved to (say) unity –other, simple and optimal problem formulations exist

18. August 2003 Daniel Nissani (Nissensohn)18 Taking Advantage of this Insight (cont.) P equalizes the channel for cross-talk noise resultant from the non-diagonal elements of B above A Post-Equalizer Q is also added to equalize residual noise, due to the diagonal elements of B We have to estimate H n (and resultant H m, A, V m, P, U m ) and to calculate the Post-Equalizer Q –2 short (e.g. L x L) Training Matrices are transmitted as burst pre-amble

19. August 2003 Daniel Nissani (Nissensohn)19 Our Proposed MIMO Model x = (P V m A) s y = H x + n and z = (Q U m ’) y A SINGLE MATRIX-VECTOR MULTIPLICATION AT EACH RX AND TX SIDES !!

20. August 2003 Daniel Nissani (Nissensohn)20 VCXC Preliminary Simulation Results L = R = 3, QPSK, Raleigh, flat fading, uncorrelated H, 3.5% Channel Exception ‘proposed’ vs. ‘naïve’ vs. ‘ideal’ models ~ 20 db s 1 sub-stream gain vs. ‘naïve’ ZF @ 1E-6 BER 5 db s 1 gap vs. ‘ideal’ @ 1E-6 BER Preliminary initial results, sub- optimal, no parameters fine tuning

21. August 2003 Daniel Nissani (Nissensohn)21 Preliminary Simulation Results (cont.)

22. August 2003 Daniel Nissani (Nissensohn)22 A broader perspective Cross-Talk Interference Precedents These results may seem new (and even a bit strange) to us all We have met with Cross-Talk Interference before, e.g. –CDMA Orthogonality loss due to Synchronization Miss-alignment (reverse) and Inter-Cell Interference (forward) –xDSL Near and Far End Parasitic bundle coupling

23. August 2003 Daniel Nissani (Nissensohn)23 In both CDMA and xDSL cases Cross-Talk Interference is –externally imposed –beyond our model control –within our model we can only attempt to fight against it –(Sequential or Parallel) Interference Cancellation live with it –Adaptive Modulation A broader perspective (cont.) Cross-Talk Interference Precedents

24. August 2003 Daniel Nissani (Nissensohn)24 In Wireless MIMO we also experience Cross-Talk Interference, but this time it is –intrinsically inherent to our model, and –(to a major extent) well under our control A broader perspective (cont.) Cross-Talk Interference Precedents

25. August 2003 Daniel Nissani (Nissensohn)25 WLAN Modeled Expected Rate-Range Performance L = R = 3, Path Loss = 27 db/ decade, f o = 5.3 GHz

26. August 2003 Daniel Nissani (Nissensohn)26 Summary EXCELLENT BER PERFORMANCE –20 db performance gain @ 1E-6 BER for 1 st sub-stream relative to LSE –0.2 db performance gap @ 1E-3 BER for 3 rd sub-stream relative to utopian perfect channel estimate case –even with sub-optimal implementation, initial results EXTREME REAL TIME SIMPLICITY –single matrix-vector multiplication at each side –easily adapted to FDD schemes by appropriate Channel Information Feedback TDD schemes by inherent Channel Reciprocity OFDM schemes, at sub-carrier level LOW OVERHEAD –e.g. just 6 symbols Pre-Amble for L = R = 3

27. August 2003 Daniel Nissani (Nissensohn)27 Summary (cont.) VCXC-- proposed herein as PHYsical LAYER element of 802.11 TGn PATENT PENDING Sorry, this is NOT a self-sufficient document

28. August 2003 Daniel Nissani (Nissensohn)28 References [1] G.D. Golden, G.J. Foschini, R.A. Valenzuela, and P.W. Wolniasky, ‘Detection algorithm and initial laboratory results using the V-BLAST space-time communication architecture’, Electronics Letters, Vol. 35, No. 1, pp. 14-15, 1999 [2] Andersen, J.B. ‘Array gain and capacity for known random channels with multiple element arrays at both ends’, Selected Areas in Communications, IEEE Journal on, Volume: 18 Issue: 11, Nov 2000, Page(s): 2172 -2178 [3] Telatar, E. I., ‘Capacity of Multi-antenna Gaussian Channels’, Technical Memorandum, Bell Laboratories, October 1995 [4] Stewart, G.W., ‘Perturbation Theory for the Singular Value Decomposition’ UMIACS-TR-90-124, September 1999 [5] Nissani (Nissensohn), D.N., ’The MIMO Cross-Talk Interference Problem- a Novel Solution’, Internal Technical Report, March 2003