Research On Meta-Learning Based Time-Varying Channel Estimation Method For High-Speed Mobile OFDM System

With the mature development of 5th generation wireless systems(5G),the mobile communications in high-speed trains,satellite communication and unmanned aerial vehicles are also developing rapidly.With the booming development of 5G and the emergence of the 6th generation wireless systems(6G),more and more attentions have been paid to the high-speed mobile wireless communication.However,in the high-speed mobile condition based on 5G or future 6G system,the rapid movement of communication terminals will cause larger Doppler shift,which will destroy the orthogonality between subcarriers in the multi-carrier system and cause greater inter-carrier interference(ICI).It also leads to more rapid random changes in the wireless channel,which will affect system performance.As an important part of wireless communication system,channel estimation plays a very important role in improving system performance.However,the traditional channel estimation algorithm can not be well applied to the high-speed mobile communication system based on 5G or 6G in the future due to the need for prior statistical information or low estimation accuracy.Therefore,to ensure the validity of high-speed mobile communication systems,it is urgent to design a time-varying channel estimation method with high accuracy and low complexity.The thesis applies machine learning to channel estimation,and aims to improve system performance and reduce computational complexity to study time-varying channel estimation methods which more suitable for high-speed mobile scenarios in 5G or future 6G systems.The contents and innovations are as follows:(1)To overcome the problems that the existing channel estimation methods based on deep learning(DL)have large training sample and time overhead,and the offline training network cannot adapt to the actual real-time changing channel environment,a novel time-varying channel estimation method based on meta-learning(ML)is proposed.The model-agnostic-meta-learning(MAML)method is used to replace the DL method to train the estimation model.Firstly,the channel task set is set according to the time-varying fading channel characteristics,and few samples are input to the network under each task.To enhance its practicability,the target of the training model is approximate set as the channel estimate.In addition,in offline training and online estimation,the proposed method only uses the received pilots to construct training samples and test samples,which significantly reduces the complexity.Simulation analysis shows that the proposed method has low complexity and high estimation precision,and it can quickly adapt to the new channel environment,which is suitable for time-varying channels acquisition in high-speed mobile communication systems.(2)To reduce the effect of ICI caused by Doppler shift on the accuracy of channel estimation,a time-varying channel estimation method with ICI elimination is proposed.At the sending end,according to the time-frequency relationship of the signal,a new transformation matrix is firstly constructed to transform the transmitted signal with the comb pilot in the frequency domain into the one with the block-type pilot in the time domain,so as to ensure that the data and pilot samples do not overlap in time and reduce the neighboring data symbols to the pilots.At the receiving end,training samples are first constructed based on the the received ICI-free pilots,and then the ML method is used for estimating the time-varying channel,which significantly advances the adaptability of the estimation model to the changing transmission conditions.In addition,in the training stage,using channel estimation instead of ideal channel state information(CSI)as the training target of ML networks can enhance its practicability.Simulation analysis shows that the method has high estimation performance and it is robust to the fast time-varying channel in the high-speed mobile scenarios.