1. Turbo-equalization for 802.11n/ac
May 2006
doc.: IEEE /0528r0
July 2012
Turbo-equalization for n/ac
Date:
Authors:
Laurent Cariou, Orange
Bruce Kraemer, Marvell

2. Context presentation title
July 2012
Context
802.11n/ac are heavily using non-orthogonal MIMO schemes which require co-antenna interference processing in the receiver
Laurent Cariou, Orange
presentation title

3. July 2012
Context
The quality of the MIMO receivers have a very strong impact on the performance
Optimum solution : joint (MIMO and channel) decoding
Maximum Likelyhood based on a « super treillis »
Sub-optimal solutions:
Disjoint decoding : MIMO detection followed by channel decoding
ML detection (hard output, soft output)
Detection with interference cancellation (SIC, …)
Equalization with linear filters (MMSE, ZF, MRC)
Iterative decoding : MIMO detection  channel decoding
ML with a priori information
Equalization with interferance cancellation (MMSE-IC)
Laurent Cariou, Orange

4. July 2012
Context
Turbo-equalization principle allows a very efficient processing of the interference
whatever interferers nature
Including MIMO co-antenna interferences
Laurent Cariou, Orange
presentation title

5. July 2012
Iterative reception allows interference cancellation through turbo-equalization
MIMO non-orthogonal scheme with estimation of transmitted symbols
Estimation of transmitted symbols in order to substract generated interferences
Turbo-equalization principle with interferences cancellation
Laurent Cariou, Orange
presentation title

6. July 2012
Introduction of feedback loop between channel decoding and MIMO detection functions
Turbo-equalization principle with interferences cancellation
Laurent Cariou, Orange
presentation title

7. Performance modeling of iterative reception
July 2012
Performance modeling of iterative reception
SISO transmission on AWGN channel (limit) depends on channel coding and mapping
Genie receiver – ideal knowledge of transmitted symbols
Offset with AWGN limit depends on diversity order
Iterative receiver
Trigger point mainly depends on the amount of interference, channel coding and MIMO detector type
Offset with genie receiver depends on MIMO detector type
Laurent Cariou, Orange
presentation title

8. July 2012
MMSE-IC (Minimum Mean Square Error – Interference Cancellation) principle
Glavieux et al. 97, Wang et al. 99, Tüchler et al. 02, Laot et al. 05
Laurent Cariou, Orange
presentation title

9. Classical MMSE at the first iteration
July 2012
Classical MMSE at the first iteration
As no a priori information is available at the first iteration, equalization is done via a classical MMSE filtering
with and the noise and signal variances
The equalized signal expressed in function of the transmitted signal s
Note that the equalized signal is biased compared to s
Laurent Cariou, Orange
presentation title

10. July 2012
For next iterations, the exact solution is associated to a prohibitive complexity
Exact solution for MMSE-IC for iterations > 1
Such a treatment requires that for each iteration, and for each data symbol in a space-time bloc code:
1 matrix inversion of size NRxNR is performed
Laurent Cariou, Orange
presentation title

11. Approximations have been found to reach suitable complexity
July 2012
Approximations have been found to reach suitable complexity
1st approximation for MMSE-IC: per block power invariance of the estimated symbols (average estimation of symbols)
Iteration 1
Iteration i (> 1)
This allows to reduce the complexity by reusing the same filter for each data symbol within a space-time block code
With this approximation, only one channel inversion is required for each iteration
Laurent Cariou, Orange
presentation title

12. Approximations have been found to reach suitable complexity
July 2012
Approximations have been found to reach suitable complexity
2nd approximation for MMSE-IC: matched filter (MF-IC) which considers that the a priori information is perfect)
Iteration i (> 1): we apply a match filter instead of a MMSE filter
Equalized symbols after iteration l can now be obtained by:
with ddiag(A) corresponding to A matrix without diagonal elements
No channel inversions are required for iterations > 1
The filters are the same for all iterations > 1 and can be stored after their calculation in the first iteration
Laurent Cariou, Orange
presentation title

13. Some theoretical results on a MIMO Rayleigh channel
July 2012
Some theoretical results on a MIMO Rayleigh channel
Parameters
SDM 4x4
QPSK
CC K=7 r=1/2
5 iterations
Random interleaving
2000 bits
Genie bound is reached
Gains at 10-4
4dB on MMSE
2.2dB on perfect ML
disjoint
itérative
Laurent Cariou, Orange

14. Some theoretical results on a MIMO Rayleigh channel
July 2012
Some theoretical results on a MIMO Rayleigh channel
Parameters
SDM 4x4
16QAM
CC K=7 r=1/2
5 itérations
Random interleaving
2000 bits
More sensitivities to interference in case of 16QAM
The gains of iterative receivers are preserved
disjoint
itérative
Laurent Cariou, Orange

15. July 2012
4x4 MIMO transmission test bench for n iterative receiver hardware evaluation
Laurent Cariou, Orange

16. Parameters July 2012 Parameters Values Number of subcarriers 128
Number of data subcarriers
108
Total symbol duration
4µs
Useful symbol duration
3.2µs
Cyclic prefix duration
0.8µs
Channel coding
Convolutional code (7,133,171)
Coding rate
1/2
Modulation
QPSK
MIMO scheme
Spatial Data Multiplexing
NTX x NRX
4x4
Channel models
TGn B/C/D/E/F
Laurent Cariou, Orange

17. Performance results over Channel B
July 2012
Performance results over Channel B
MIMO 4x4, Chan B
Simulation parameters: QPSK, 1/2, perfect CSI
Results for:
1 iteration = MMSE only
more than 1 itertion = turbo-equalization
High gain with iterative interference cancellation scheme
Laurent Cariou, Orange
presentation title

18. Performance results over Channel E
July 2012
Performance results over Channel E
MIMO 4x4, Chan E
Simulation parameters: QPSK, 1/2, perfect CSI
Results for:
1 iteration = MMSE only
more than 1 itertion = turbo-equalization
High gain with iterative interference cancellation scheme
Laurent Cariou, Orange
presentation title

19. Performance results General comments
July 2012
Performance results
General comments
iterative receiver for interference cancellation is bringing significant gain (about 5dB gain at 10-4 BER for 3 iterations).
the gains are bigger when the amount of interference increases, better with 4x4 than with 2x2 or 2x3.
helps you to reduce the required number of receive antennas (no need for 4x5 or 4x6), even for 4 SS
as 11n and 11ac go toward an increase of SS, iterative receivers become more and more interesting
3 iterations is already getting the most part of the iterative gain. This is important as it lowers the constraints regarding the latency.
Laurent Cariou, Orange

20. Performance results A complementary solution to beamforming technique
July 2012
Performance results
A complementary solution to beamforming technique
beamforming is strongly improving the transmission performance but its benefits are very different depending on the MCSs. Beamforming doesn’t improve significantly the performance of 4 spatial streams MCSs in case of a 4x4 system, while iterative receivers do. Iterative receivers clearly increases the range of very high throughputs.
beamforming works well if the transmitter and the receiver are beamforming-capable, and currently if they are from the same vendor. In all other cases, iterative receivers will complement beamforming.
Laurent Cariou, Orange

21. July 2012
Conclusion
We believe that iterative receivers are a very competitive solution for MIMO-OFDM,
It outperforms ML receivers
It has reached maturity regarding implementation complexity and respect of latency constraints
We believe that it is a good time for its integration into products, especially for WIFI
Laurent Cariou, Orange
presentation title

22. July 2012
Combination with LDPC: Two kind of iterations are now combined: LDPC and turbo-equalization
Iterative channel decoding
Iterative equalization and iterative channel decoding
Combination of channel decoding loops and MIMO interference cancellation loops is flexible
Laurent Cariou, Orange
presentation title

23. July 2012
Combination with LDPC: Two kind of iterations are now combined: turbo decoding and turbo-equalization
We have optimized the number of inner (LDPC) and outer iterations (turbo-equalization) in order to improve the performance and reduce the number of total iterations
We have evaluated different static and dynamic stopping criteria for the inner iterations of the LDPC
We concluded that the total number of iterations of LDPC with or without turbo-equalization is kept unchanged. With turbo-equalization, this total number of iterations corresponds to the summation of the LDPC iterations of each outer turbo-equalization iteration.
Iterative receivers are therefore very well suited to a combination with LDPC
Laurent Cariou, Orange
presentation title