Deep Learning-Based Mobile User Localization with Reconfigurable Intelligent Surface

Reconfigurable intelligent surface (RIS) is a promising technique for positioning systems. A large number of reference signals is needed for traditional base station (BS) with RIS localization system which occupy communication resources. To solve this problem, we directly use the DMRS (Demodulation Reference Signal), which does not consume additional communication resources to estimate the user equipment (UE) position. In this paper, we consider a RIS-assisted downlink localization system that takes account for the channel time variability caused by UE mobility. To estimate the UE position, DPSCN (Depthwise Separable Convolution) network consisting of three stages is adopted to capture the relationship between the received signal and the UE position.Based on the estimated position, the RIS phase is dynamically adjusted, to achieve accurate and continuous tracking of the UE. Extensive simulation results are also presented to demonstrate that the proposed DPSCN learning algorithm can estimate the UE position accurately and outperforms the traditional algorithms.