Convolutional Neural Network-Aided Tree-Based Bit-Flipping Framework for Polar Decoder Using Imitation Learning
Journal
IEEE Transactions on Signal Processing
Journal Volume
69
Pages
300-313
Date Issued
2021
Author(s)
DOI
ITPRE
Abstract
Known for their capacity-achieving abilities and low complexity for both encoding and decoding, polar codes have been selected as the control channel coding scheme for 5G communications. To satisfy the needs of high throughput and low latency, belief propagation (BP) is chosen as the decoding algorithm. However, it suffers from worse error performance than that of cyclic redundancy check (CRC)-aided successive cancellation list (CA-SCL). Recently, convolutional neural network-aided bit-flipping (CNN-BF) is applied to BP decoding, which can accurately identify the erroneous bits to achieve a better error rate and lower decoding latency than prior critical-set bit-flipping (CS-BF) mechanism. However, successive BF, having better error correction capability, has not been explored in CNN-BF since the more complicated flipping strategy is out of the scope of supervised learning. In this work, by using imitation learning, a convolutional neural network-aided tree-based multiple-bits BF (CNN-Tree-MBF) mechanism is proposed to explore the benefits of multiple-bits BF. With the CRC information as additional input data, the proposed CNN-BF model can further reduce 5 flipping attempts. Besides, a tree-based flipping strategy is proposed to avoid useless flipping attempts caused by wrongly flipped bits. From the simulation results, our approach can outperform CS-BF and reduce flipping attempts by 89% when code length is 64, code rate is 0.5 and SNR is 1 dB. It also achieves a comparable block error rate (BLER) as CA-SCL. © 1991-2012 IEEE.
Subjects
belief propagation; bit-flipping; convolutional neural network; imitation learning; Polar codes
Other Subjects
5G mobile communication systems; Backpropagation; Bit error rate; Channel coding; Convolution; Decoding; Error correction; Signal encoding; Signal to noise ratio; Belief propagation; Block error rates; Cyclic redundancy check; Decoding algorithm; Encoding and decoding; Error correction capability; Imitation learning; Successive cancellation; Convolutional neural networks
Type
journal article