1 Introduction
2 Algorithm description
2.1 Variable definition
2.2 Lower bounds of the net path metrics
Lemma 1.
Proof
2.3 CVA-based ML decoder on tail-biting trellis
Example 1
3 Simulation results
SNR | |||||
---|---|---|---|---|---|
BLER performance | 0.0 dB | 1.0 dB | 2.0 dB | 3.0 dB | 4.0 dB |
WAVA | 3.95×10−1 | 1.49×10−1 | 2.80×10−2 | 2.70×10−3 | 1.06×10−4 |
ML decoders | 3.80×10−1 | 1.48×10−2 | 2.66×10−2 | 2.50×10−3 | 0.96×10−4 |
SNR | |||||
---|---|---|---|---|---|
BLER performance | 1.0 dB | 2.0 dB | 3.0 dB | 4.0 dB | 5.0 dB |
WAVA | 1.60×10−1 | 3.16×10−2 | 3.10×10−3 | 1.45×10−4 | 3.50×10−6 |
ML decoders | 1.59×10−1 | 3.05×10−2 | 3.10×10−3 | 1.44×10−4 | 3.50×10−6 |
4 Conclusion
Appendices
Appendix 1: decoding process of the two-phase ML decoder
Step | Open stack | Close table | Complexity |
---|---|---|---|
1 | {(1, 1, 0, 0.291)} |
∅
| {(6, 4)} |
2 | {(1, 2, 1, 0.291)} | {(1, 1, 0)} | {(6, 7)} |
3 | {(1, 3, 2, 0.986), (1, 1, 2, 1.234)} | {(1, 1, 0), (1, 2, 1)} | {(6, 10)} |
4 | {(1, 1, 2, 1.234), (1, 3, 3, 1.277)} | {(1, 1, 0), (1, 2, 1), (1, 3, 2)} | {(6, 13)} |
5 | {(1, 2, 3, 1.234), (1, 3, 3, 1.277)} | {(1, 1, 0), (1, 2, 1), (1, 3, 2), (1, 1, 2)} | {(6, 16)} |
6 | {(1, 3, 3, 1.277)} | {(1, 1, 0), (1, 2, 1), (1, 3, 2), (1, 1, 2), (1, 2, 3)} | {(6, 19)} |
Appendix 2: improvements of the BEAST ML decoder with upper-bounding technique
SNR | |||||||||
---|---|---|---|---|---|---|---|---|---|
Decoding complexity | 0.0 dB | 1.0 dB | 2.0 dB | 3.0 dB | 4.0 dB | 5.0 dB | 6.0 dB | 7.0 dB | |
Additions | 10,307 | 10,083 | 9,989 | 9,583 | 9,288 | 9,222 | 9,013 | 8,999 | |
BEAST | Comparisons | 20,613 | 20,167 | 19,978 | 19,165 | 18,576 | 18,443 | 18,026 | 17,998 |
Additions | 5,248 | 4,395 | 3,793 | 3,411 | 3,218 | 3,048 | 2,996 | 2,930 | |
Advanced BEAST | Comparisons | 10,497 | 8,790 | 7,585 | 6,821 | 6,436 | 6,096 | 5,993 | 5,860 |