New estimates for the performance of soft-decision sequential decoders

A recent new method analyzes sequential decoding under the sole constraint that a set error probability Pe is achieved. The result is a good estimate of the number of paths searched and the definition of the minimum possible search region in the code tree or trellis. This work extends the analysis to AWGN soft-decision channels. We find the AWGN channel output quantization that leads to the minimum decoding effort. Large reductions in path searching are possible with three-level quantization, whereas octal quantization achieves most of the savings. Furthermore, the channel quantization that optimises the decoder work is not the one that maximizes the capacity. Overall, savings from soft decision are more dramatic at high code rates and in poor channels, in many cases being a must to achieve decoder feasibility