Abstract:

To overcome the problem of high computational complexity in existing deep learning channel estimation methods in reconfigurable intelligent surface (RIS)-assisted orthogonal frequency division multiplexing (OFDM) system, the base extension model (BEM) is used to model the time-varying channel under RIS, and a time-varying channel estimation method based on residual link super-resolution convolutional neural network is proposed. Specifically, the proposed method firstly converts the channel coefficient estimation with more parameters to the base coefficient estimation with fewer parameters to reduce the computational complexity of the proposed method. In offline training process, the neural network is trained with low-resolution basis coefficient estimation, where only a small amount of input is required to obtain high-resolution channel estimation. In order to improve the practicability of the proposed method, the training network label is set to have high-precision channel estimation value instead of ideal channel information. The proposed method is verified by simulation test, which proves that it can accurately obtain time-varying channel information in RIS-assisted mobile communication system, and has higher estimation accuracy and lower computational complexity.

Key words: reconfigurable intelligent surface (RIS), orthogonal frequency division multiplexing (OFDM), time-varying channel estimation, basis extension model (BEM), residual-linked super-resolution convolutional neural network (ResSRCNN)