Performance analysis of neural network-based polar decoding algorithms with different code rates

• Published in: Iran Journal of Computer Science (Online) Vol. 5; no. 1; pp. 83 - 97
• Main Authors: Devamane, Shridhar B., Itagi, Rajeshwari L.
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.03.2022
• Subjects: Artificial Intelligence; Computer Science; Computer Systems Organization and Communication Networks; Data Structures and Information Theory; Mathematics of Computing; Original Article; Software Engineering/Programming and Operating Systems; Theory of Computation; Polarization; Neural network; Binary phase shift keying; Block error rate; Belief propagation
• ISSN: 2520-8438; 2520-8446
• DOI: 10.1007/s42044-021-00094-2

Neural network approach with deep learning is of interest of late in channel decoding. Polar code is a desirable candidate in 5G and replaces Turbo code with better error correction capacities. In view of this, there is a need to explore polar decoding with neural network techniques. In this paper, the performance of polar decoding using conventional successive cancellation (SC) algorithm and belief propagation (BP) algorithm are evaluated in additive white Gaussian noise (AWGN) in python platform. Also, an effort is made to test the performance of decoding by implementing belief propagation algorithm using neural network, called as belief propagation neural network (BPNN). BPNN is chosen, as it requires a minimum number of iterations to complete the decoding operation. Performance is compared with that of conventional decoding. Performance analysis is carried out in terms of bit error rate (BER) for different input code lengths and code rates. Block length error rate (BLER) analysis of algorithms is also investigated. It is observed that BPNN performs better with improvement in the performance with modified neural network architecture and training sets.