The Maximum-Likelihood Soft-Decision Sequential Decoding Algorithms for Convolutional Codes

In this thesis, we study the performance of the maximum-likelihood soft-decision sequential decoding algorithm (MLSDA) for convolutional codes, which is previously proposed in [1]. Instead of using the conventional Fano metric, the proposed algorithm employs a new metric based on a variation of the Wagner rule, which is referred to as the second form of the maximum-likelihood decoding rule. The original analysis in [1] assumed infinite stack size. This assumption, however, may not be feasible in practice, and the performance of MLSDA is expected to degrade if finite stack constraint is applied. Our simulation results, however, indicate that the BER (Bit Error Rate) remains almost intact for a moderate stack size (e.g., 512). Yet, a further decrease in stack size (e.g., 64) may significantly degrade the BER. We also empirically investigate the tightness of the theoretical upper bound for the computational efforts of MLSDA derived in [1]. Simulation results show that the difference between the curves obtained from simulations and from the theoretical upper bound is at most 0.67; and therefore, the room for improvement to the theoretical upper bound seems prohibitively limited. Finally, a more complex metrics is derived to obtain the performance of the MLSDA under the fading channel. We also provide the performance comparison between the MLSDA using the former metrics for the additive white Gaussian noise (AWGN) channel and that using the new metrics under the same fading channel. Surprisingly, the simulation results show that the former metrics for the AWGN channel is robust for the BER, but the new metrics