Joint User Localization, Channel Estimation, and Pilot Optimization for RIS-ISAC

Reconfigurable intelligent surface (RIS), a large array of passive scattering elements, is able to control the properties of electromagnetic waves, thereby enhancing the channel capacity, reducing the bit error rate, and enabling novel signal modulation methods. However, the promising gain of RIS depends on the precision of channel estimation. In this paper, we propose a three-step channel reconstruction framework to improve the channel estimation accuracy inspired by the concept of integrated sensing and communication scenario. Firstly, based on the coarse channel state information (CSI), the proposed dual one-dimensional multiple signal classification (D1D-MUSIC) algorithm improves the localization precision with a reduced complexity. Secondly, expectation maximization-based refined estimation (EMRE) algorithms are proposed to refine the CSI and estimate channel statistical properties (CSP), i. e. , the shadow fading, the power of line-of-sight paths, and that of non-line-of-sight components. Thirdly, a gradient descent-based pilot optimization (GDPO) algorithm is further derived to improve the channel estimation precision on the basis of estimated CSPs. Finally, simulation results demonstrate that the developed D1D-MUSIC algorithm has lower localization error and complexity compared with conventional two-dimensional MUSIC algorithm. Moreover, the EMRE algorithms achieve the identical normalized mean square error (NMSE) performances as the ideal minimum mean square error estimator, while possessing robust resistance to the channel model mismatch. Furthermore, the developed GDPO technique is capable of providing an over 11 dB signal-to-noise ratio gain for channel estimation performance at NMSE = 10^-2.