Robust iterative decoding of turbo codes in heavy-tailed noise

The advancements in channel coding theory over the past decades have been accomplished by considering of additive white Gaussian noise (AWGN) channels. Much less is known about the consequences when the standard AWGN assumption is not fulfilled in realistic environments and, more importantly, the appropriate countermeasures. The paper investigates the robustness of turbo codes decoded by existing quadratic-type algorithms in heavy-tailed, non-Gaussian noise channels. It illustrates that impulsive noise constitutes a major impairment in turbo decoding by studying the a posteriori probabilities computed by the constituent decoders, in addition to a formal account in terms of error probability performance. It is found that the performance of turbo codes is extremely sensitive to the shape of the underlying noise density function, being considerably degraded when this function departs from Gaussianity into a heavy-tailed distribution. A robust variant of existing decoders for reliable decoding of turbo codes in heavy-tailed noise is proposed and studied. The robustness is achieved by the enforcement of a non-quadratic soft metric into the decoder for good estimation of the transition probabilities and reliable extraction of extrinsic information.