Deep Learning-Aided Trainable Projected Gradient Decoding for LDPC Codes

Abstract

We present a novel optimization-based decoding algorithm for LDPC codes that is suitable for hardware architectures specialized to feed-forward neural networks. The algorithm is based on the projected gradient descent algorithm with a penalty function for solving a non-convex minimization problem. The proposed algorithm has several internal parameters such as step size parameters, a softness parameter, and the penalty coefficients. We use a standard tool set of deep learning, i.e., back propagation and stochastic gradient descent (SGD) type algorithms, to optimize these parameters. Several numerical experiments show that the proposed algorithm outperforms the belief propagation decoding in some cases.