Simulation Research Of Wireless Communication Physical Layer Algorithm Based On Deep Learning Autoencoder

For traditional communication systems,a large number of algorithms have been optimized for fixed modules.Although the modular optimization design method can improve the performance of a single module,it does not mean that the performance of the whole system is optimal.In addition,because modular-based systems require certain functions to be performed based on well-designed algorithms,such systems often lack flexibility.Deep learning is structured and intelligent,and can better adapt to the needs of next-generation communication systems or certain specific application scenarios.Through the idea of autoencoder in deep learning,an end-to-end communication system can be realized by deep neural network,which can be jointly optimized end-to-end to achieve better communication performance.As a result,in this thesis,we investigate the feasibility and performance of the end-to-end communication system implemented based on deep-learning through simulations.The first part is mainly about the use of deep learning autoencoder to implement endto-end communication systems under Gaussian and Rayleigh fading channels.For the communication system under Gaussian channel,the influence on the model exerted by the number of hidden layers in the model construction and the model hyperparameter in the training phase are studied,on the basis of which,the model structure and parameters are determined to obtian the system BER performance.The simulation results show that compared with the traditional systems,the neural network-based system possesses lower bit error rate.Further,under the Rayleigh channel,the impact on the model exerted by the model batch size and learning rate is investigated to obtain the best model parameter setting.Finally,we simulate the two-user interference channel and the influence of different loss weight factors on the system is studied.Considering that MIMO technology plays an important role in the existing communication system,we simulate the MIMO system based on deep learning algorithm in the second part of the system,the results of which show that deep learning still performs well in MIMO system.We consider different antenna configurations here.Firstly,we compare the autoencoder based SIMO system with the traditional maximum combining system,and the former exhibits better BER performance.Subsequently,we simulate and compare the MISO system based on autoencoder and Alamouti transmit diversity respectively,the results of which show that the system based on neural network possesses a lower bit error rate.Finally,the simulation results show that the MIMO system based on neural network has better BER performance than the SVD precoding system.