Research On Polar Code Decoding Algorithm Based On Neural Networ

Since polar codes was proposed by Professor Arikan in 2009,the decoding research of polar codes has been a hotspot in the field of channel coding and decoding.Among the traditional decoding algorithms,the Successive Cancelation(SC)algorithm and the Belief Propagation(BP)algorithm are the most commonly used decoding methods.However,both of them are not ideal and cannot meet the requirements of high speed and low delay proposed by future communication.Therefore,it is necessary to conduct in-depth research in this direction to expect a low-complexity and low-latency decoding method.In recent years,neural networks have developed rapidly in various fields,and they have solved many problems by using their advantages in data processing.Considering that decoding can actually be regarded as a multi-classification problem for data,it is of great research value and practical significance to introduce neural networks into the field of channel decoding.In this thesis,by applying the neural network model to the decoding field of polar codes,three different decoding algorithms based on neural networks are designed and implemented,and their performance,advantages and disadvantages are analyzed respectively.Then,two improved algorithms are proposed.The main contents are as follows.Four basic neural network decoder models are built,which include Multilayer Perceptron(MLP)decoder,Convolutional Neural Network(CNN)decoder,Long Short-Term Memory Network(LSTM)decoder and Gated Recurrent Unit(GRU)decoder.These four decoders are simulated under noise-free and noise conditions respectively.Through the analysis of the simulation results,it can be seen that the performance of the MLP decoder is poor and not suitable for decoding.The performance of the LSTM decoder and the GRU decoder is close to and better than the CNN decoder;According to the neural network also has a certain learning ability to noise,a polar code decoding algorithm based on denoising neural network is proposed.To improve the performance,the neural network is used to first denoise the received signal to reduce the influence of noise on the information data,and then the traditional algorithm is used for decoding.The simulation results show that CNN has more advantages in denoising than LSTM and GRU;Based on the simulation analysis of the neural network decoder and neural network denoiser,the neural network concatenated decoding algorithm is researched,and the CNN-RNN decoding model is designed,which combines CNN’s better denoising performance with RNN’s better decoding performance.The simulation results show that the neural network concatenated algorithm can effectively alleviate the problem of poor performance caused by insufficient training set in the neural network decoder under the noise condition.Through the neural network concatenated decoding algorithm,the problem of infinite codebook data set caused by noise can be ignored in practical application,and it is more possible to apply neural networks to the practical condition;Finally,two improved algorithms are proposed for the neural network concatenated decoding algorithm,which are the neural network parallel decoding algorithm and the neural network concatenated decoding algorithm based on the auxiliary denoising module.The former reduces the bit error rate by analyzing the decoding results,and the latter improves the performance of the denoiser by introducing the traditional decoding algorithm for auxiliary denoising,and further improves the performance of the decoding model.The simulation results show that comparing with the neural network concatenated algorithm,the two algorithms have about 0.5dB and 0.3dB performance gain respectively under the simulation condition of this thesis.