Max-Min SINR Analysis of STAR-RIS Assisted Massive MIMO Systems With Hardware Impairments

Reconfigurable intelligent surface (RIS) has emerged as a cost-effective solution to improve wireless communication performance through just passive reflection. Recently, the concept of simultaneously transmitting and reflecting RIS (STAR-RIS) has appeared but the study of minimum signal-to-interference-plus-noise ratio (SINR) and the impact of hardware impairments (HWIs) remain open. In addition to previous works on STAR-RIS, we consider a massive multiple-input multiple-output (mMIMO) base station (BS) serving multiple user equipments (UEs) at both sides of the RIS. Specifically, in this work, focusing on the downlink of a single cell, we derive the minimum SINR obtained by the optimal linear precoder (OLP) with HWIs in closed form. The OLP maximises the minimum SINR subject to a given power constraint for any given passive beamforming matrix (PBM). Next, we obtain deterministic equivalents (DEs) for the OLP and the minimum SINR, which are then used to optimise the PBM. Notably, based on the DEs and statistical channel state information (CSI), we optimise simultaneously the amplitude and phase shift by using a projected gradient ascent algorithm (PGAM) for both energy splitting (ES) and mode switching (MS) STAR-RIS operation protocols with reduced feedback, which is quite crucial for STAR-RIS systems that include the double number or variables compared to reflecting only RIS. Simulations verify the analytical results, shed light on the impact of HWIs, and demonstrate the better performance of STAR-RIS compared to conventional RIS. Also, a benchmark full instantaneous CSI (I-CSI) based design is provided and shown to result in higher SINR but lower net achievable sum-rate than the statistical CSI based design because of large overhead associated with the acquisition of full I-CSI acquisition. Thus, not only do we evaluate the impact of HWIs but we also propose a statistical CSI based design that provides higher net sum-rate with low overhead and complexity.