Neural Belief Propagation Auto-Encoder for Linear Block Code Design

The growing number of Internet Thing (IoT) and Ultra-Reliable Low Latency Communications (URLCC) use cases in next generation communication networks calls for the development efficient Forward Error Correction (FEC) mechanisms. These usually imply using short to mid-sized information blocks requires low-complexity and/or fast decoding procedures. This paper investigates joint learning mid block-length coding schemes associated Belief-Propagation (BP) like decoders Machine Learning (ML) techniques. An interpretable auto-encoder (AE) architecture is proposed, ensuring scalability block sizes currently challenging ML-based linear code design approaches. By optimizing a scheme w.r.t. targeted decoder, proposed system offers good complexity/performance trade-off compared various codes from literature with length up 128 bits.