MIND: Maximum Mutual Information Based Neural Decoder

We are assisting at a growing interest in the development of learning architectures with application to digital communication systems. Herein, we consider detection/decoding problem. aim developing an optimal neural architecture for such task. The definition criterion is fundamental step. propose use mutual information (MI) channel input-output signal pair, which yields minimization a-posteriori transmitted codeword given output observation. computation formidable task, and majority channels it unknown. Therefore, has be learned. For objective, novel estimator based on discriminative formulation. This leads derivation decoder (MIND). developed capable not only solve decoding problem unknown channels, but also return estimate average MI achieved coding scheme, as well error probability. Several numerical results reported compared maximum likelihood strategies.