New turbo-ARQ techniques based on estimated reliabilities

We propose a new form of incremental-redundancy hybrid ARQ for use with codes that use soft-input, soft-output decoders, such as turbo codes. The new ARQ scheme determines the reliability of bits in the iterative decoding process based on the magnitude of the soft outputs of the decoder, which are the a posteriori log-likelihood ratios. We show that the bits with the smallest magnitudes are not only likely to be in error, but are linked to a set of weak bits that limit the convergence of the iterative decoding algorithm. The new ARQ scheme requests additional information for some subset of these weak bits. We note that the hybrid ARQ scheme we propose is particularly appropriate for use on channels in which there is some asymmetry, as the negative acknowledgements are considerably larger than those used in conventional ARQ schemes. The hybrid-ARQ schemes presented in this paper provide significant performance gains and can be used to achieve performance close to the channel capacity.