This paper proposes the deep neural network (DNN) based on a belief propagation flip (BPF) decoding algorithm. The conventional BPF decoding does not determine which bits to flip and exhaustively flips the bits of a critical set (CS). This paper uses a DNN to decide which bits to flip. In addition, to reduce the training complexity of DNN, the codeword segmentation and the classes consisting of CS were used. As a result, the proposed BP-DNN-Flip algorithm shows a performance gain of 0.3dB at frame error rate (FER) 10−4compared to the conventional BPF decoding algorithm. In addition, it has a lower average time complexity of at least 55% compared to traditional BPF decoding algorithms at a signal-to-noise ratio (SNR) of 1.5dB.
This paper proposes the deep neural network (DNN) based on a belief propagation flip (BPF) decoding algorithm. The conventional BPF decoding does not determine which bits to flip and exhaustively flips the bits of a critical set (CS). This paper uses a DNN to decide which bits to flip. In addition, to reduce the training complexity of DNN, the codeword segmentation and the classes consisting of CS were used. As a result, the proposed BP-DNN-Flip algorithm shows a performance gain of 0.3dB at frame error rate (FER) 10−4compared to the conventional BPF decoding algorithm. In addition, it has a lower average time complexity of at least 55% compared to traditional BPF decoding algorithms at a signal-to-noise ratio (SNR) of 1.5dB.
The belief propagation (BP) decoding has the advantage of providing soft information iteratively, which allows polar codes to be applicable in scenarios requiring high throughput. However, the unsatisfactory error-correction performance makes the BP polar decoding unable to compete with the successive cancellation list (SCL) decoding that is the baseline algorithm in 3GPP. The BP flip (BPF) decoding offers one option to address this issue through additional decoding attempts with a few bits flipped. However, current BPF decoders yield less efficient flipping accuracy. Considering the complex bit correlation and unanalyzable erroneous bit probability of BPF decoding, we propose a long short-term memory (LSTM) network-assisted BPF (LSTM-BPF) decoding algorithm. Given the soft information output by the original BP decoding, the trained LSTM network achieves more accurate erroneous bit prediction ability. The simulation results show that compared with the state-of-the-art BPF decoders, the LSTM-BPF decoder improves the hit rate of the error bits by 10%, thereby obtaining more than 0.1 dB gains at a frame error rate (FER) of 10-4for polar codes with length 128 and code rate 0.5. The efficient hardware architecture of the proposed LSTM-BPF decoder is designed. By introducing a critical set to reduce the size of the network, the LSTM accelerator could significantly reduce the required computation and storage resources, thereby reducing the inference time.
Deep learning recently shows strong competitiveness to improve polar code decoding. However, suffering from prohibitive training and computation complexity, the conventional deep neural network (DNN) is only possible for very short code length. In this paper, the main problems of deep learning in decoding are well solved. We first present the multiple scaled belief propagation (BP) algorithm, aiming at obtaining faster convergence and better performance. Based on this, deep neural network decoder (NND) with low complexity and latency, is proposed for any code length. The training only requires a small set of zero codewords. Besides, its computation complexity is close to the original BP. Experiment results show that the proposed (64,32) NND with 5 iterations achieves even lower bit error rate (BER) than the 30-iteration conventional BP and (512, 256) NND also outperforms conventional BP decoder with same iterations. The hardware architecture of basic computation block is given and folding technique is also considered, saving about 50% hardware cost.
Polar codes is the only coding scheme in the world that has been proved that the Shannon threshold is reachable. Because of the high time delay caused by the serial characteristic of Successive Cancellation (SC) decoding, the Belief Propagation (BP) decoding has received more attention because of its higher throughput. However, because its decoding performance is not as good as SC decoding, Bit-Flip (BF) decoding is used in BP. Aiming at the inaccurate construction of flip set in current bit flipping decoding, this paper proposes a new bit error rate BP BF (BPF) decoding algorithm, which is used to construct flip set. We also propose rough flip set and fine flip set Concept of flip set. The rough flip set is constructed by the variance of the characteristics of data stability. The range of the flip set is reduced by the difference in bit error rate, that is, the fine flip set. The research results show that the algorithm proposed in this paper is superior to the current best decoding algorithms such as BP, Enhanced BPF (EBPF) and Log Likelihood Ratio BPF (BPF-LLR). At the same time, the number of iterations of the algorithm is significantly reduced compared with other bit-flipping decoding algorithms.
The bit-flip method has been successfully applied to the successive cancellation (SC) decoder to improve the block error rate (BLER) performance for polar codes in the finite code length region. However, due to the sequential decoding, the SC decoder inherently suffers from longer decoding latency than that of the belief propagation (BP) decoder with efficient early stopping criterion. It is natural to ask how to perform bit-flip in a polar BP decoder. In this paper, bit-flip is introduced into the BP decoder for polar codes. The idea of critical set (CS), that is, originally proposed by Zhang et al. for identifying unreliable bits in a SC bit-flip decoder, is extended to the BP decoder here. After revealing the relationship between CS and the incorrect BP decoding results, critical set with order ω (CS-ω) is constructed to identify unreliable bit decisions in polar BP decoding. The simulation results demonstrate that compared with the conventional BP decoder, the BLER of the proposed bit-flip decoder can achieve significant signal-to-noise ratio (SNR) gain which is comparable to that of a cyclic redundancy check-aided SC list decoder with a moderate list size. In addition, the decoding latency of the proposed BP bit-flip decoder is only slightly higher than that of the conventional BP decoder in the medium and high SNR regions.
Known for their capacity-achieving abilities, 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, in general, the error performance of BP is worse than that of enhanced successive cancellation (SC). Recently, critical-set bit-flipping (CS-BF) is applied to BP decoding to lower the error rate. However, its trial and error process result in even longer latency. In this work, we propose a convolutional neural network-assisted bit-flipping (CNN-BF) mechanism to further enhance BP decoding of polar codes. With carefully designed input data and model architecture, our proposed CNN-BF can achieve much higher prediction accuracy and better error correction capability than CS-BF but with only half latency. It also achieves a lower block error rate (BLER) than SC list (CA-SCL).
Bit flipping was first used to improve the decoding performance of successive cancellation (SC) decoder. Since SC decoding process is serial, so the first bit that goes wrong in the decoding process will cause several subsequent bits to go wrong, which will suffer from higher decoding delay than belief propagation (BP) parallel decoding. In order to meet the performance of 5G low latency and high reliability, BP decoding algorithm have received a lot of attention. However, at medium and high signal-to-noise ratios (SNR), BP decoding does not perform as well as the SC list (SCL) decoding algorithm. Hence, BP decoding algorithm borrows bit-flip strategies from SC bit flip (SCF) decoder to improve the performance. The performance of BP bit flip (BPF) decoding depends on whether the construction of the flip set is correct, so this letter proposes a flip set construction approach based on the standard deviation of the log-likelihood ratio (LLR) and the reliability weight of polarized channels, which is comparable to the existing BPF decoding, the improved algorithm can identify the error-prone bits more efficiently and improve the decoding performance.