Gradient Descent Decoding of MDPC Codes Optimized with Genetic Algorithm

Bit-flipping (BF) is a very simple algorithm for decoding linear block codes. For the BF to achieve high performances of belief-propagation (BP) algorithms, which are far more complex, we apply several optimizations using the genetic algorithm (GA), to optimize the parameters of our decoder. Adaptive diversity gradient descent bit-flipping with momentum (AD-GDBFwM) algorithm is designed as a cascade of multiple decoders with different parameters, which are optimized with GA. In this paper we consider AD-GDBFwM decoding of mediumdensity parity-check (MDPC) codes. We measure the decoding performance and compare it with the results of various types of decoders, including BP and neural BP.