Research On Adaptive Modulation And Coding Technology For Fast-varying Channel Based On Machine Learning

With the gradual popularization of 5G technology and further research on 6G technology,more and more high-speed mobile scenarios have begun to place higher requirements on mobile networks.The communication link not only needs to meet high communication quality with low error codes,but also needs to meet high communication efficiency with high data rate.Link adaptation technologies have become a research hotspot.However,the fast-varying channel is very different.The parameters of the channel change rapidly in a short time,and it is difficult for the traditional channel estimation technology to accurately track the fast-varying channel at a low cost.One of the important foundations of link adaptation technology is to accurately perceive and estimate the channel state.Therefore,channel sensing and link adaptation technology have great joint research value.Meanwhile,machine learning algorithms have attracted more and more communication researchers.The hardware platform performance of machine learning algorithms has become increasingly powerful,and the response time of the algorithm has been greatly shortened,showing great application potential.Therefore,this paper uses neural networks and reinforcement learning algorithms to study channel prediction and adaptive modulation and coding technology,breaking through many problems of traditional communication algorithms under fast-changing channels of 5G system.The main work content of this paper is as follows:For the channel prediction problem,this paper combines traditional channel estimation algorithms with Long Short-Term Memory networks,and proposes a fast-changing channel prediction model based on M₂M₄-LSTM.Since CNN has powerful feature extraction capabilities,this paper converts the received signal into an image-liked format and proposes a fast-varying channel prediction model based on CNN.This paper uses the feature extraction capabilities of CNN and the prediction capabilities of LSTM to further propose a fast-varying channel prediction model based on CNN-LSTM.Compared with the M₂M₄-LSTM model and the CNN model,the proposed CNN-LSTM model has great advantages in predicting signal-to-noise ratio and multipath delay extension.And it has good robustness at a speed of120km/h?500km/h.The range of channel coherence time is 14.6?s?60.9?s.For the adaptive modulation and coding problem,this paper proposes an adaptive modulation and coding method combined with reinforcement learning and neural network algorithms,based on the modulation and coding scheme recommended in the 5G NR standard.This paper uses Dueling DQN as the decision algorithm and CLSTM as the channel tracking algorithm to conduct joint simulation experiments under the fast-varying channel of 5G NR system.Simulations show that the proposed Dueling DQN-based adaptive modulation and coding scheme has better bit error rate performance,spectrum efficiency performance and convergence performance,compared to the CQI-based adaptive modulation and coding scheme and the DQN-based adaptive modulation and coding scheme.Dueling DQN also has good generalization performance in the satellite communication system of DVB-S2 protocol.