Deep Neural Network-based Decoder for Short Linear Block Codes Transmitted via Binary Symmetric Channel

• Autor(a) principal: Kamassury, Jorge Kysnney Santos
• Data de Publicação: 2021
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Remat (Bento Gonçalves)
• Texto Completo: https://periodicos.ifrs.edu.br/index.php/REMAT/article/view/4389
• topic: Neural Network-Based Decoder; Binary Symmetric Channel; Error Correcting Codes; Decodificador Baseado em Rede Neural; Canal Binário Simétrico; Códigos Corretores de Erros

Short-length codes have been the subject of recent studies mainly due to the need for specific communication requirements expressed by emerging technologies. However, for the most promising code class (BCH), decoding is complex when using traditional decoders. In this context, projects that use neural networks for this purpose appear as interesting alternatives. That said, in this article, the decoder project proposed in the literature that applies the neural network to estimate the error pattern considering the received vector syndrome extends to the BCH codes of length n less than or equal to 31. In addition, a new decoder is introduced, one that iteratively estimates the most reliable positions to be the erroneous bits of the error pattern previously predicted by a neural network. The results presented show that, for all the analyzed codes, the new decoder reaches the maximum theoretical performances.