An Channel Estimation Technique Based On Machine Learning In High-speed Mobile Environment

With the continuous development of high speed trains,wireless communication technology used in high-speed mobile scenarios has become a research focus.Among them,channel estimation,as an important part of an Orthogonal Frequency Division Multiplexing(OFDM)communication system,will directly affect the communication quality of the entire communication system.In a high-speed mobile environment,complex mobility of vehicles,obstacles and non-line-of-sight communication may reduce the quality of the communication channel.In this special environment,traditional channel estimation methods are no longer applicable.To this end,this article focuses on the shortcomings of existing channel estimation methods,combining deep learning and reinforcement learning to conduct in-depth research on channel estimation in internet of vehicles.The specific contents are as follows:Firstly,aiming at the existing channel response estimation method without considering the weight between features,a channel response estimation method called CLSTMA based on convolutional neural network(CNN),long short term memory(LSTM)network and self-attention mechanism is proposed.This method uses CNN to extract the features of channel responses.On this basis,combined with the LSTM network to estimate channel,and use the self-attention mechanism to assign weights to the features.This method uses a single-level LSTM network to significantly reduce the number of training parameters and achieve a compromise between performance and complexity.Simulation results show that this method can reduce the number of iterations and save computing resources,and is more accurate than a single neural network model in predicting channel response.Secondly,in the existing channel response regression prediction methods based on deep learning,a large number of target values are usually required for regression,and the screening and processing of these target values becomes a bottleneck for improving the prediction speed.This paper proposes a channel response prediction method CLSTMA-DQN based on the fusion of deep reinforcement learning and regression network to improve the accuracy of regression prediction.The model mainly includes three parts: feature extraction network,DQN network and regression network.Among them,the feature extraction network is a pre-trained CLSTMA network,and the DQNDel network needs to construct an environment state space set,design the interaction rules between the agent and the environment,and design the reward function that is fed back to the agent.In order to realize the joint optimization of the DQNSel network and the regression network,the loss function is set as the multi-task loss function,that is,the weighted sum of the DQNSel network loss and the regression network loss.Compared with the performance of neural network-based channel estimation algorithms and traditional channel estimation algorithms,the experimental results confirm that the model has good effects in terms of bit error rate and prediction accuracy.