January 2004 Turbo Codes for IEEE 802.11n Marie-Helene Hamon, Olivier Seller France Telecom R&D Claude Berrou, Catherine Douillard, Sylvie Kerouedan ENST Bretagne Brian Edmonston iCODING Technology Inc.

Contents TC for 802.11n Performance Granularity (Flexibility) January 2004 Contents TC for 802.11n Performance Granularity (Flexibility) Complexity & latency Conclusion

Turbo Codes: Iterative FEC for 802.11n January 2004 Turbo Codes: Iterative FEC for 802.11n Revolutionary form of error correcting, relying on soft iterative decoding to achieve high coding gains Very good performance, near channel capacity for long blocks TC advantages led to adoption in several recent digital communication standards: 3GPP UMTS (WCDMA), DVB-RCS, DVB-RCT, cdma2000 and consumer sattelite broadcast… Hardware development and complexity well controlled

Turbo Codes: Iterative FEC for 802.11n January 2004 Turbo Codes: Iterative FEC for 802.11n High coding gains over classical convolutional code: - Turbo Codes enable the use of more efficient transmission modes (coding rate and modulation) more often, to increase throughput - Turbo Codes yield a lower PER: better system efficiency, as ARQ algorithm could be used less frequently - Reasonable memory requirements

January 2004 Duo-Binary CTC

Advantages of this TC Duo-binary: January 2004 Advantages of this TC Duo-binary: - reduction of the latency and complexity per decoded bit - reduction of the path error density better convergence Circular Revursive Systematic Codes as constituent codes no trellis termination overhead Original permuter scheme larger minimum distance, better asymptotic performance

Performance on AWGN channel January 2004 Performance on AWGN channel Duo-Binary 8-state CTC K=1600 bit, R=4/5 Max-Log-MAP decoding 8 iterations At BER=10-5, 8-state CTC is 1.305 dB from capacity

January 2004 Performance on AWGN

Block size Very good performance whatever the block size January 2004 Block size Very good performance whatever the block size Two Duo-Binary CTC: 8-state for PER>10-4, 16-state for PER>10-7 Better performance than LDPC codes - for any code rate - for any block size <10000 bits - for any BER <10-9 - for any associated modulation Block sizes as small as 12 bytes (DVB-RCS) and 18 bytes (DVB-RCT) have been standardized

High Flexibility With the same encoder/decoder: January 2004 High Flexibility With the same encoder/decoder: - several coding rates allowed through simple adaptation of the puncturing pattern (DVB-RCS: 7 coding rates) - different block sizes enabled just by adjusting permutation parameters (DVB-RCS: 12 block sizes) For the duo-binary Turbo Code, a set of 4 permutation parameters needs to be modified. Each set of parameters defines an interleaver for one block size, all coding rates.

High Flexibility Better granularity in block size and coding rate: January 2004 High Flexibility Better granularity in block size and coding rate: FER (hardware measurements) for 8-state CTC, 8 iterations, 4-bit quantization at decoder input, Max-Log-MAP decoding

January 2004 High Flexibility Turbo codes can adjust to all kinds of coding rates, block sizes and modulations

January 2004 Granularity Modulation and Code rate are adjusted to keep actual performance close to capacity for any given SNR The more granularity the greater the OVERALL performance Low granularity substantially reduces the benefit of the advanced coding

January 2004 Granularity

January 2004 Complexity The decoder of DVB-RCS is not more complex than any LDPC decoder Terminals are equiped with both encoder and decoder To enable high bit rates, decoding can be restricted to 4 iterations without significantly altering performance

TC Memory Requirements January 2004 TC Memory Requirements Rate 4/5, 1600 information bits (1600,2000) 6 bit samples 70,000 bits (approx) of RAM 160,000 soft information reads per frame @ 8 iterations 80,000 soft information reads per frame @ 4 iterations Power consumption ~ total memory x iterations Lower memory requirements and iterations -> less power consumption

LDPC Memory Comparison January 2004 LDPC Memory Comparison 120 Kbits of memory 8 iterations minimum 274,285 soft decision read/write per frame

January 2004 Latency The decoder has an inherent possibility of parallelism, thanks to the simple generic permutation. For example, with: - parallelism of degree 4 (4 backward/forward processors) - 250 MHz circuit clock - k=500 information bits per block - 4 iterations about 250 Mbits/s data rate can be achieved with 2.5 μs decoding latency

January 2004 Conclusion Duo-Binary Turbo Code enables large performance gains, for all block sizes and coding rates Highly flexible solution Minimizes memory access requirements Mature technology, implemented in 3rd generation mobile phones & volume consumer satellite

January 2004 References C. Berrou, A. Glavieux, P. Thitimajshima, "Near Shannon limit error-correcting coding and decoding: Turbo Codes", ICC93, vol. 2, pp. 1064-1070, May 93. C. Berrou, "The ten-year-old turbo codes are entering into service", IEEE Communications Magazine, vol. 41, pp. 110-116, August 03. TS25.212 : 3rd Generation Partnership Project (3GPP) ; Technical Specification Group (TSG) ; Radio Access Network (RAN) ; Working Group 1 (WG1); "Multiplexing and channel coding (FDD)". October 1999. EN 301 790 : Digital Video Broadcasting (DVB) "Interaction channel or satellite distribution systems". December 2000. EN 301 958 : Digital Video Broadcasting (DVB) "Specification of interaction channel for digital terrestrial TV including multiple access OFDM". March 2002.