Learning to Decode Linear Block Codes using Adaptive Gradient-Descent Bit-Flipping

Abstract

In this paper we propose a generalization of the recently published adaptive diversity gradient-descent bit flipping (AD-GDBF) decoder, named generalized AD-GDBF (gAD-GDBF) decoder. While the original AD-GDBF decoder was designed for the binary symmetric channel and used mostly to decode regular low-density parity-check codes, the gAD-GDBF algorithm incorporates several improvements which makes it eligible for the additive white Gaussian channel and decoding of arbitrary linear block code. The gAD-GDBF decoder uses the genetic algorithm to optimize a set of learnable parameters, for a targeted linear block code. The effectiveness of the proposed method is verified on short Bose-Chaudhuri-Hocquenghem (BCH) codes, where it was shown that for the same number of decoding iterations the gAD-GDBF decoder outperforms the belief-propagation decoder in terms of bit error rate and at the same time reduces the decoding complexity significantly.