分布式机器学习在RIS辅助的无线信道估计中的应用

无线信道估计是部署可重构智能超表面(Reconfigurable Intelligent Surface, RIS)辅助通信系统的关键与前提，然而下行链路传输环境下信道估计困难且导频开销大是对智能超表面辅助通信的重大挑战。针对以上问题，提出了一种基于分布式机器学习(Distributed Machine Learning, DML)训练模型的区域交集切换方案。首先，建立了一个多用户共享的下行信道估计神经网络，通过DML技术协同用户与基站训练网络模型。其次，搭建分层次神经网络结构对用户区域信道进行分类和特征提取。最后，针对用户处于相邻信道交集位置问题采用特征区域模型融合。仿真结果表明，基于区域交集的DML模型方案能在减少信道训练导频开销的同时最大化信道估计的精准性能。

Distributed Machine Learning for Channel Estimation in RIS-assisted Wireless Channel

Wireless channel estimation is the key and prerequisite for deploying reconfigurable intelligent surface(RIS)-assisted communication system.However,the difficulty of channel estimation and the huge pilot overhead in the downlink transmission environment are major challenges for RIS-assisted communication.In order to solve above problems,a regional intersection switching scheme based on distributed machine learning(DML) technique is proposed.Firstly,a multi-user shared downlink channel estimation network is proposed,which is trained by DML technique in collaboration with users and base stations.Then,a hierarchical neural network structure is built to classify and extract features of user regional channels.Finally,the feature region model fusion is used to solve the problem of users at the intersection position of adjacent channels.The simulation results show that the DML model scheme based on regional intersection can reduce the channel training pilot overhead and maximize the accurate performance of channel estimation.