Research On Channel Estimation Algorithm Based On Deep Learning In 5G High-speed Mobile Scenarios

In recent years,the huge increase in the mileage of high-speed railways and expressways has brought about a significant reduction in people’s work and leisure travel time,and has also aroused people’s attention to wireless communication in a high-speed mobile environment.However,in the high-speed mobile environment supported by 5th Generation Mobile Communication Technology(5G),higher vehicle speeds,more frequent handovers and wider bandwidths make the design of high-speed mobile wireless communication systems more challenging.Therefore,the high-performance wireless communication technology is urgently needed to support the future high-speed mobile scenarios to achieve ultra-relaible and low latency communication,where the anti-Doppler frequency shift technology is the key.Among many anti-Doppler frequency shift schemes,the time-varying channel estimation is an important technique.This is due to the fact that in the high-speed mobile environment in the future,the Doppler frequency shift will be larger and the bandwidth will be wider,which will bring about the characteristics of time-frequency double fading of the channel,which makes it more difficult to realize the correct transmission of the signal.In order to improve the communication quality in high-speed mobile scenarios,high-precision and low-complexity time-varying channel estimation methods are studied in this thesis for 5G system.The main contents and innovations are as follow:(1)For high-speed mobile orthogonal frequency division multiplexing system(OFDM),an improved time-varying channel estimation method based on back-propagation(BP)neural network is proposed.By rationally constructing network input samples,the proposed method can not only make full use of the channel variation features in historical channel information and other features in the received pilot signal,but also take advantage of least squares(LS)estimation to further improve the performance of channel estimation.The auto-regressive(AR)model is adopted to obtain the current channel estimation from historical information,which is averaged with the channel information estimated by the current LS method to obtain high-precision channel information,and construct the input samples.Then the constructed samples are used to first train the BP neural network offline,and finally obtain the time-varying channel information online.Only the received pilot signals and pilot sub-channels are used for time-varying channel estimation to cut down the computational complexity.The simulation results show that a good trade-off between system performance and computational complexity the proposed scheme owns which is suitable for efficient acquisition of time-varying channel information in high-speed mobile scenarios.(2)For high-speed mobile multiple input multiple output-orthogonal frequency division multiple OFDM system,a low-complexity deep learning based time-varying channel estimation scheme is proposed.To reduce the number of parameters to be estimated,the basis expansion model(BEM)is employed to model the time-varying channel,and it converts the channel estimation into the estimation of the basis coefficient.Specifically,the initial basis coefficient estimation is firstly used to train the neural network in an offline manner,and then the high-precision channel estimation can be obtained by small number of inputs.Moreover,the high-precision estimated channel is considered for the target in training phase,which makes the proposed method more practical.Simulation results show that the proposed method has a better performance and lower computational complexity compared with the available schemes,and it is robust to the fast time-varying channel in the high-speed mobile scenarios.(3)Considering the interference caused by the residual Doppler frequency offset in the received signal,a time-varying channel estimation method based on meta-learning is proposed for single input single output OFDM system with residual Doppler frequency offset.In order to overcome the problem that the existing deep learning-based channel estimation methods have too much training data and time overhead,and the offline training network cannot adapt to the real-time changing channel environment in practice,the channels with multiple parameters are considered as one task set,and then the model-agnostic meta-learning(MAML)is adopted to train different subtasks offline,so that the deep learning network can learn the channel transmission characteristics and have the ability to quickly adapt to new tasks.The received signals are used to perform online time-varying channel estimation.Furthermore,the more accessible and high-precision estimated channel estimation is used as the target during network training,making the proposed method more practical.The simulation results show that this method has higher estimation accuracy than the traditional deep learning method,and can quickly adapt to the new channel environment,and is still applicable in the channel environment after the propagation characteristics change.The simulation results show that the new method has higher estimation accuracy than the traditional deep learning method,and can quickly adapt to the new channel environment,and is still suitable for communication systems with residual frequency offset effects.