Optimizing Intelligent Reflecting Surfaces with Discrete Phase Shifts and Pilot Overhead Reduction Using Deep Learning

Intelligent Reflecting Surface (IRS) is a cost-effective technology for future wireless communication systems to achieve high spectrum and energy efficiency. If the IRS elements are adjusted properly, the surface can provide a power gain proportional to the square of the number of IRS elements. Optimizing the phase shifts of IRS elements is a challenging task since they are not equipped with active RF chains. Previous studies have mostly assumed that the phase shifts of the elements take continuous values, which is practically difficult to implement due to hardware constraints. In addition, a large number of elements at the IRS may result in substantial training overhead. The study presents a Deep Learning approach for configuring discrete phase shifts of IRS elements. The proposed method uses the pilot signals reflected by the IRS as the model input and provides the optimum discrete phase shift values as its output. The model is trained using different lengths of pilot symbol blocks, and numerical results are presented.