Research On Carrier Frequency Offset Estimation And Correction Technology In OTFS System

Since the 1990 s,with the rapid development of wireless communication technology and the improvement of material culture,mobile communication system has played a great role in many fields and has put forward further requirements on the quality of wireless communication system while accessing more hardware devices of mobile network.With the large coverage of China’s high-speed rail network,the high-speed rail industry has been developing toward intelligence and digitalization,and mobile communication in high-speed scenarios has become a research hotspot in academia.In the high-speed environment,the high Doppler frequency shift caused by high-speed movement can make the performance of the Orthogonal Frequency Division Multiplexing(OFDM)system significantly reduced.The Orthogonal Time Frequency Space Modulation(OTFS)technique proposed in recent years is expected to solve the communication problem in high-speed environment,but OTFS system inherits the sensitivity of OFDM system to synchronization error when the channel is in time-varying situation.On the other hand,machine learning has been widely used in the field of wireless communication in recent years,and its excellent performance in the fields of signal estimation and detection has gained the favor of many scholars.Based on the above background,this thesis investigates OTFS systems and their related frequency offset estimation techniques in highspeed mobile scenarios.Firstly,this thesis introduces the OTFS modulation and demodulation technique and its system model,analyzes the synchronization error of OTFS system,studies the Carrier Frequency Offset(CFO)estimation technique based on Cyclic Prefix(CP)and the CFO estimation technique based on training symbols,and conducts simulation and performance analysis.Secondly,combined with the principle of deep learning,a CFO estimation technique based on Shallow Neural Network(SNN)is designed in this thesis.The core idea is to learn the mapping between the received data of OTFS system and the corresponding frequency offset target value by analyzing the SNN structure.Then the correlation algorithm is used to optimize the network parameters and estimate and correct the frequency deviation of the received data of OTFS system.The simulation results show that the Mean Square Error(MSE)performance of the CFO estimation technology based on SNN is improved to a certain extent compared with the traditional CFO estimation technology,and the performance of the technology in the aspect of Bit Error Ratio(BER)has a significant gain in the high-speed environment.Then,considering that the shallow neural network model contains only one implicit layer,it has limited ability to express complex functions.In this thesis,we also design a CFO estimation technique based on Deep Belief Network(DBN),which first pre-trains the Restricted Boltzmann Machine(RBM)layer by layer by unsupervised learning method,which can control the initial weights and bias values required for network training The network parameters are then fine-tuned using supervised learning methods,which can avoid the local optimum problem caused by random initialization of parameters to a certain extent.The experimental analysis shows that the frequency bias estimation accuracy of DBN-based CFO estimation technique is significantly improved compared with SNN-based CFO estimation technique in a high-speed environment,and the BER performance of this technique is also gained.