Neural Mutual Information Estimation for Channel Coding: State-of-the-Art Estimators, Analysis, and Performance Comparison

Deep learning based physical layer design, i.e., using dense neural networks as encoders and decoders, has received considerable interest recently. However, while such an approach is naturally training data-driven, actions of the wireless channel are mimicked using standard channel models from the literature, which only partially reflect the physical ground truth. Very recently, neural network based mutual information estimators have been proposed that directly extract channel actions from the input-output measurements and feed these outputs into the channel encoder. This direction is promising as such a new design paradigm is fully adaptive and training data-based. In this paper, we take forward this approach and implement further recent improvements of such mutual information estimators, analyze theoretically their suitability for the channel coding problem, and compare their performance.