A Simple Neural-Network-Based Decoder for Short Binary Linear Block Codes

• Published in: Applied sciences Vol. 13; no. 7; p. 4371
• Main Authors: Hsieh, Kunta, Lin, Yan-Wei, Chu, Shao-I, Chang, Hsin-Chiu, Cho, Ming-Yuan
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.04.2023
• Subjects: Algorithms; Analysis; binary linear block code; Bit error rate; Block codes; Codes; Computer architecture; Decoders; Deep learning; Error correction; Error correction & detection; Error-correcting codes; Gaussian elimination; Methods; neural network; Neural networks; Soft decision decoding; Statistical analysis
• ISSN: 2076-3417; 2076-3417
• DOI: 10.3390/app13074371

The conventional soft decision decoding (SDD) methods require various hard decision decoders (HDDs) based on different codes or re-manipulate the generator matrix by the complicated Gaussian elimination technique according to the bit reliability. This paper presents a general multi-class neural network (NN)-based decoder for the short linear block codes, where no HDD and Gaussian elimination are required once the NN is constructed. This network architecture performs multi-classification to select the messages with high occurrence probabilities and chooses the best codeword on a maximum likelihood basis. Simulation results show that the developed approach outperforms the existing deep neural network (DNN)-based decoders in terms of decoding time and bit error rate (BER). The error-correcting performance is also superior to the conventional Chase-II algorithm and is close to the ordered statistics decoding (OSD) in most cases. For Bose–Chaudhuri–Hocquenghem (BCH) codes, the SNR is improved by 1dB to 4dB as the BER is 10−4. For the (23, 12) quadratic residue (QR) code, the SNR is improved by 2dB when the BER is 10−3. The developed NN-based decoder is quite general and applicable to various short linear block codes with good BER performance.