Online Training-based Channel Estimation for RIS-assisted Communication Systems with Low Complexity

‍ ‍In Reconfigurable Intelligent Surface （RIS）-assisted communication systems， channel estimation algorithms should be designed with low complexity and a high estimation accuracy. In this paper， aiming at the uplink of a RIS-assisted Multi-input Single-output （MISO） system， we propose a low-complexity learning channel estimation （LCL-CE） method based on online training. Utilizing the spatial correlation between the cascaded channels of RIS reflection elements， the cascaded channels of the remaining reflection elements are obtained from the Least Squares （LS） estimation results of partial reflection elements. In the training stage， a set of training datasets is first obtained using LS estimation， and then the weight matrix of the linear neural network is obtained through training. In the cascaded channel estimation stage， the cascaded channels of all RIS reflection elements can be obtained by sending a small amount of the pilot signal. This paper first describes the implementation concept of the LCL-CE method， the pilot distribution scheme， and the basic structure of the linear neural network. Subsequently， the online training and training dataset generation methods are elaborated. Finally， we compare the computational complexity and pilot overhead analysis with the channel estimation method based on the traditional neural network. Simulation results showed that compared with the traditional machine learning-based channel estimation method， the low-complexity learning channel estimation method can obtain a higher estimation accuracy with a lower pilot overhead.