Research On Polar Code Decoding Algorithm Based On BPD And RNN And Its Application In CVQKD

As the only channel code capable of reaching the Shannon limit,polar code has been designated as the forward error correction(FEC)code of fifth generation(5G)communication system.The polar code is conventionally decoded by belief propagation decoding(BPD)and successive cancellation decoding(SCD).Compared with BPD,SCD has better decoding performance,but due to its serial characteristics,its decoding delay is far greater than BPD.Therefore,BPD is widely used in systems with high throughput requirements.This thesis focuses on improving the decoding complexity,decoding performance and decoding speed of BPD algorithm,and applies it to the continuous-variable quantum key distribution(CVQKD)system.Then,neural network based BPD algorithm is studied,by combining recurrent neural network(RNN)with BPD to improve the decoding performance of BPD.In summary,the innovations of this thesis are as follows:1.The scheduling scheme of BPD directly affects its convergence speed and decoding performance.The advantages and disadvantages of existing scheduling schemes are as follows.Round trip scheduling has good decoding performance because of its uniform information distribution,but the decoding speed is slow.Segmented scheduling has a faster decoding speed,but due to its uneven information distribution,the decoding performance is reduced compared with round-trip scheduling.Thus,we proposed a scheduling scheme by combining round-trip scheduling and segmented scheduling to improve decoding speed and ensure decoding performance.The proposed scheduling scheme is applied to polar code based information reconciliation in CVQKD post-processing to further verify its advantages.2.Inspired by the belief propagation list decoding algorithm(BPD with List,BPDL),this thesis proposes a new BPDL decoding algorithm,namely,scalable BPDL(S-BPDL).This decoding scheme first multiplies different coefficients before the BPD information update formula,and uses decoders with different coefficients as multiple decoding paths.Then,the path with the best decoding result is selected as the final decoding result.In addition,this thesis applies S-BPDL to polar code based information reconciliation in CVQKD post-processing and observes its performance.3.Observing the positions of the information and frozen bits of the polar code,this thesis proposes four constitute codes.With specific calculation,the number of nodes on the factor graph can be reduced by removing these constituent codes,resulting in the reduction of the decoding complexity.In addition,this thesis also applies RNN to the BPD algorithm with removed constituent codes to improve the decoding performance of BPD.