A Generic Reliability Based Direct Decoding Algorithm for Turbo Codes

Interest for communication with short block length messages has gained much attention recently in many of the current and emerging wireless applications. Classical capacity approaching codes exhibit performance degradation with short block length codewords and hence are not suitable for applications that demand short block length communication. A novel performance enhanced reliability based direct decoding algorithm, that makes use of reliability values and encoder structure, has been proposed recently by the authors for short block length turbo codes. The proposed algorithm is a replacement for the conventional MAP based iterative turbo decoding algorithm. This is the first attempt (to the best of our knowledge) that used the reliability values directly instead of the log likelihood ratio (LLR) values that are commonly used in the conventional decoding algorithm. The proposed algorithm has shown a conspicuous performance improvement as well. Typically it has 2.45 dB coding gain at a BER of \(10^{-3}\) and channel adaptive complexity over AWGN channel with BPSK modulation for a code rate of \(\frac{1}{4}\). The conventional iterative algorithm formulation is based on a Gaussian noise model assumption. Any deviation from Gaussian assumption calls for changes in the branch metric computation since it uses the channel reliability of Gaussian noise. This modified branch metric reflects as modifications in the forward metric, backward metric and hence in LLR computation. In contrast to this, the direct decoding algorithm is a generic scheme and can be adapted for both Gaussian and different non-Gaussian distributions as well. In this paper, a generalization of this generic reliability based direct decoding algorithm without the Gaussian restriction, is presented.