<acronym id="2gwac"><center id="2gwac"></center></acronym>
<acronym id="2gwac"><center id="2gwac"></center></acronym>
華中科技大學學報(自然科學版) 2020, Vol. 48 Issue (10): 26-30 DOI10.13245/j.hust.201005

欄目:控制與信息工程
極化碼的低復雜度神經BP譯碼
陶志勇 , 李 艷
遼寧工程技術大學電子與信息工程學院,遼寧 葫蘆島 125105
摘要 提出一種用于極化碼的NBP(神經置信度傳播)譯碼算法的低復雜度替代方案.首先,利用有效的聚類技術,減少權重參數的數量;然后,通過在時間上共享權重,進一步消除大量權重參數中存在的不必要冗余;最后,從均勻量化和非均勻量化兩方面對浮點權重參數進行量化,進一步減少權重參數的存儲需求.仿真結果和復雜度分析表明:通過在時間和空間上應用有效的權重共享策略,將權重參數量化為4位定點形式,可以減少至少80%的網絡權重,并且降低了權重參數的內存消耗,大幅度壓縮了NBP譯碼器,同時保持良好的譯碼性能;在高信噪比區域,對權重參數進行A律非均勻量化時譯碼器的譯碼性能較浮點權重參數譯碼器最高約有0.2 dB的提升.
關鍵詞 極化碼 ;神經置信度傳播 ;權重共享 ;聚類 ;量化
Low-complexity neural BP decoding of polar codes
TAO Zhiyong , LI Yan
School of Electronic and Information Engineering,Liaoning Technical University,Huludao 125105,Liaoning China
Abstract A low complexity alternative scheme of neural belief propagation (NBP) decoding algorithm for polar codes was proposed.First,the effective clustering technique was adopted to reduce the number of weight parameters.Then,the unnecessary redundancy in a large number of weight parameters was further eliminated by sharing weights in time.Finally,the floating-point weight parameters were quantized from the two aspects of uniform quantization and non-uniform quantization to further reduce the storage requirements of weight parameters.Simulation results and complexity analysis demonstrate that by applying the effective weight sharing strategy in time and space and quantizing the weight parameters to a 4-bit fixed-point form,the network weight can be reduced by at least 80%,and the memory consumption of the weight parameters is effectively reduced,which greatly compress the NBP decoder while maintaining good decoding performance. In high signal-noise ratio region,the decoding performance of the decoder when performing A-law non-uniform quantization of the weight parameters is about 0.2 dB higher than that of the floating-point weight parameters decoder.
Keywords polar codes ; neural belief propagation ; weight sharing ; clustering ; quantization
基金資助國家重點研發計劃資助項目(2018YFB1403303);遼寧省博士啟動基金資助項目(20170520098).

中圖分類號TN911.22;TP183
文獻標志碼A
文章編號1671-4512(2020)10-0026-05
參考文獻
[1] ARIKAN E.Channel polarization:a method for constructing capacity-achieving codes for symmetric binary-input memoryless channels[J].IEEE Transactions on Information Theory,2009,55(7):3051-3073.
[2] ZHANG C,PARHI K K.Low-latency sequential and overlapped architectures for successive cancellation polar decoder[J].IEEE Transactions on Signal Processing,2013,61(10):2429-2441.
[3] ARIKAN E.A performance comparison of polar codes and reed-muller codes[J].IEEE Communications Letters,2008,12(6):447-449.
[4] 盧濤,汪家明,李曉林,等.基于中繼循環殘差網絡的人臉超分辨率重建[J].華中科技大學學報(自然科學版),2018,46(12):100-105.
[5] 李如瑋,孫曉月,劉亞楠,等.基于深度學習的聽覺倒譜系數語音增強算法[J].華中科技大學學報(自然科學版),2019,47(9):78-83.
[6] 房立金,劉博,萬應才.基于深度學習的動態場景語義SLAM[J].華中科技大學學報(自然科學版),2020,48(1):121-126.
[7] SIMEONE O.A very brief introduction to machine learning with applications to communication systems [J].IEEE Transactions on Cognitive Communications and Networking,2018,4(4):648-664.
[8] GRUBER T,CAMMERER S,HOYDIS J,et al.On deep learning-based channel decoding[C]// Proc of 51st Annual Conference on Information Sciences and Systems.Balti-more:IEEE,2017:1-6.
[9] CAMMER S,GRUBER T,HOYDIS J,et al.Scaling deep learning-based decoding of polar codes via partitioning [C]// Proc of IEEE Global Communications Conference.Singapore:IEEE,2017:1-6.
[10] DOAN N,HASHEMI S A,GROSS W J.Neural successive cancellation decoding of polar codes[C]// Proc of IEEE 19th International Workshop on Signal Processing Advances in Wireless Communications.Kalamata:IEEE,2018:1-5.
[11] NACHMANI E,MARCIANO E,LUGOSCH L,et al.Deep learning methods for improved decoding of linear codes[J].IEEE Journal of Selected Topics in Signal Processing,2017,12(1):119-131.
[12] LUGOSCH L,GROSS W J.Neural offset min-sum decoding[C]// Proc of IEEE International Symposium on Information Theory.Aachen:IEEE,2017:1361-1365.
[13] KIM H,JIANG Y,RANA R,et al.Communication algorithms via deep learning[C]// Proc of 6th International Conference on Learning Representations.Vancouver:s.n.,2018:1-17.
[14] XU W,WU Z,UENG Y L,et al.Improved polar decoder based on deep learning[C]// Proc of IEEE International Workshop on Signal Processing Systems.Lorient:IEEE,2017:1-6.
[15] ZHOU A,YAO A,GUO Y,et al.Incremental network quantization:towards lossless CNNs with low-precision weights[C]// Proc of 5th International Conference on Learning Representations.Toulon:s.n.,2017:1-14.[16]JACOB B,KLIGYS S,CHEN B,et al.Quantization and training of neural networks for efficient integerarithmetic-only inference[C]// Proc of IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018:2704-2713.
[17] TENG C F,CHEN H,HO K S,et al.Low-complexity recurrent neural network-based polar decoder with weight quantization mechanism[C]// Proc of IEEE International Conference on Acoustics,Speech and Signal Processing.Brighton:IEEE,2019:1413-1417.
[18] WODIANY I,POP A.Low-precision neural network decoding of polar codes[C]// Proc of IEEE 20th International Workshop on Signal Processing Advances in Wireless Communications.Cannes:IEEE,2019:1-5.
[19] KSCHISCHANG F R,FREY B J,LOELIGER H A.Factor graphs and the sum-product algorithm[J].IEEE Transactions on Information Theory,2001,47(2):498-519.
[20] ABBAS S M,FAN Y Z,CHEN J,et al.Low complexity belief propagation polar code decoder[C]// Proc of IEEE Workshop on Signal Processing Systems.Hangzhou:IEEE,2015:1-6.
[21] DENG C,LIAO S,YUAN B.Reduced-complexity deep neural network-aided channel code decoder:a case study for BCH decoder[C]// Proc of IEEE International Confer- ence on Acoustics,Speech and Signal Processing.Bright-on:IEEE,2019:1468-1472.
[22] LECUN Y,BENGIO Y,HINTON G.Deep learning[J].Nature,2015,521(7553):436-444.
文獻來源
陶志勇, 李 艷. 極化碼的低復雜度神經BP譯碼[J]. 華中科技大學學報(自然科學版), 2020, 48(10): 26-30
DOI:10.13245/j.hust.201005
时时彩票