Research Design And Implementation Of Physical Layer Of Communication System Based On Deep Learning

Thanks to the efficient parallel processing capacity of new computers and the massive sample data collected by sensors connected to the Internet of everything,artificial intelligence(AI)is on the fast track again after it was proposed in the 1950 s.As a current research hotspot,deep learning is the direction closest to AI,and has achieved remarkable results in computer vision and other related fields.In recent years,scholars in the field of communication gradually combined deep learning with signal processing in the communication process to optimize the whole communication process by replacing one or several links in the communication system.Drawing on the above ideas,this paper combines deep learning with communication system,starts from bit transmission and recovery of physical layer,and explores how to optimize end-to-end communication system by using deep learning network model.Existing studies show that the method of deep learning has potential research value in the field of communication.The method of deep learning can directly restore the received baseband signal to the bit information of the home,replace the whole receiving link of the communication system with the network,and optimize the performance of the communication system from the perspective of the physical layer.In order to realize the proposed idea,this paper mainly carries out the following four aspects of research.This paper uses the basic architecture of neural network and classical convolutional neural network to construct wireless communication system from end-to-end perspective,and proposes a constant mode constrained autoencoder network structure for wireless communication system.The autoencoder network architecture is used to simulate the transmitter and receiver of the communication system,and the performance advantages of autoencoder communication performance constructed by fully connected network and convolutional network compared with traditional end-to-end systems are explored.Meanwhile,the problem of sending signal distortion caused by the large peak-to-average power ratio of the existing communication system is considered.The proposed network structure can reduce the signal peak-to-average power ratio without losing the transmission performance.Based on the study of the autoencoder end-to-end communication system,a communication receiving scheme based on deep learning is proposed and verified in the simulation of OFDM communication system.Deep learning is used to construct the receiving model of the communication system.In OFDM communication system,the signals after passing through noisy channels are taken as samples,and the bit information generated by the source is labeled,a signal recovery model based on convolutional neural network is constructed to recover the bit stream information from the baseband IQ data.The network model is used to replace the signal detection,channel equalization,demodulation and decoding in the physical layer of the communication system,so as to overcome the influence of channel fading and noise on the receiving system.In order to solve the influence of Doppler phenomenon on communication in high-speed mobile communication,this paper proposes an OTFS signal receiving and recovery method based on deep learning for modulation technology in Doppler scenario.Based on the realization of autoencoder communication system and OFDM system,using the same research idea,the OTFS signal receiving and recovery network structure with adaptive modulation mode is designed,and multi-bit label signal training is realized by designing multiple binary classifiers in the last layer of the network structure.Using OTFS signals at the receiver as samples,the multi-label classifier is trained at the same time to realize the process of restoring baseband OTFS signals to bitstream signals.The feasibility of the proposed receiving scheme is verified by simulation.At the end of the paper,the OFDM communication system is realized by using the general software radio hardware platform USRP-N310.On this basis,the communication receiving technology based on deep learning proposed in the paper is implemented and deployed,and the signals actually collected are used as training network learning samples.The receiving and recovery model of OFDM signals is obtained by the bit information of the sending end based on the tag training.It can be seen from the statistics of the final bit error that the OFDM signal receiving method based on deep learning is realistic and feasible.Its transmission performance can be compared with that of the traditional detection and demodulation system,and it can be used as the research direction of intelligent communication in the future.