Hold on to your Complex Phase Estimation Algorithms for Reconfigurable Intelligent Surfaces

Reconfigurable Intelligent Surfaces (RIS) are currently seen as one of the strong candidates for the next generation communication technologies. This paper considered RIS-enabled communication in a multipath channel with random scatterers. In an ideal scenario, the RIS allows phase corrections of infinite resolution to enhance the controllable part of the channel, but in practice, the phase corrections are limited by the tunability of the circuit elements. This paper discusses a practical approach where we utilize quantization to categorize the phase values into a set of known values. We propose a deep learning model to predict the RIS configuration for a given position of the RIS element. Then we compare it with a computationally less loaded interpolation method, which estimates the RIS configuration based on the configuration of the nearest reference points available. By varying the dataset ratio, we also compare the spatial density of the reference points required to estimate the phase at the RIS for both methodologies correctly. Then we compare both methods to evaluate the effect of the dataset variation on the achievable channel capacity. Finally, we conclude that in an environment with no strong LOS channel, the maximum deviation in the channel capacity by using both methods is only around 3.5%.