Research On Fading Channel Equalization Technology Based On Deep Learning

With the popularity of 5G technology,the emergence of many new multimedia applications has led to a great increase in the amount of wireless data.High-speed transmission in complex and changeable wireless channels may produce serious inter symbol interference.Therefore,it is necessary to use channel estimation and channel equalization technology at the receiver to suppress or eliminate the interference in addition to channel detection,so as to reduce the bit error rate.Neural network has very strong learning ability and nonlinear characteristics,and often has a good performance in dealing with complex nonlinear scenes.Different from the traditional channel equalizer,the structure of the equalizer based on deep learning is more complex,the optional parameters and opt imization algorithms are also diverse,and it has a better ability to extract the de ep-seated features of the data.In order to eliminate the distortion caused by the high-speed transmission of the signal in the frequency selective fading channel as far as possible and effectively recover the original signal,this paper proposes a fading channel equalization scheme based on deep learning,designs two neural network equalizers with different structures to recover the distorted signal,and makes innovations in the structure and algorithm compared with the traditional channel equalizer to reduce the bit error rate.The generalization and robustness of the system are enhanced.Firstly,this paper proposes a channel equalization scheme based on multi-layer perceptron,constructs the system model of multi-layer perceptron equalizer,introduces the network structure of multi-layer perceptron,and optimizes the algorithm and parameter selection of multi-layer perceptron equalizer.The simulation results show that the multi-layer perceptron equalizer can effectively correct the constellation when the signal has amplitude fading,phase shift and other distortions;In Rayleigh multipath fading channel,the multi-layer perceptron equalizer has better BER performance than the traditional linear adaptive equalizer.Secondly,a channel equalization scheme based on convolutional recurrent neural network is proposed.In this scheme,a network structure with four modules is proposed.The convolution layer is used to extract the deep-seated features of the data,and then the long short-term memory is used to extract the temporal features of the sequence;The network scale and key parameters are selected through simulation;After selecting the best parameters,the constellation correction ability of convolution recurrent neural network equalizer is demonstrated by constellation correction simulation.Through the bit error rate curve simulation,it is proved that the proposed channel equalizer based on convolutional recurrent neural network can obtain smaller convergence error in the training process than the traditional decision feedback equalizer,and can achieve the performance improvement of close to 2d B on the premise of reaching the same bit error rate.