Hyper-Graph-Network Decoders for Block Codes
This work addresses decoding challenges in communication systems for engineers and researchers, but it is incremental as it extends existing neural decoder methods to more code types.
The paper tackled the problem of decoding large families of algebraic block codes, such as BCH, LDPC, and Polar codes, by using graph neural networks with hypernetworks and a stabilized activation function, resulting in improved decoding performance compared to vanilla belief propagation and other learning techniques.
Neural decoders were shown to outperform classical message passing techniques for short BCH codes. In this work, we extend these results to much larger families of algebraic block codes, by performing message passing with graph neural networks. The parameters of the sub-network at each variable-node in the Tanner graph are obtained from a hypernetwork that receives the absolute values of the current message as input. To add stability, we employ a simplified version of the arctanh activation that is based on a high order Taylor approximation of this activation function. Our results show that for a large number of algebraic block codes, from diverse families of codes (BCH, LDPC, Polar), the decoding obtained with our method outperforms the vanilla belief propagation method as well as other learning techniques from the literature.