Enabling Large Intelligent Surfaces with Compressive Sensing and Deep Learning

Employing large intelligent surfaces (LISs) is a promising solution for the coverage and rate of future wireless systems. These surfaces a massive number of nearly-passive elements that interact with the signals, for example by reflecting them, in a smart way that improves wireless system performance. Prior work focused on the design of the LIS matrices assuming full knowledge of the channels. Estimating these at the LIS, however, is a key challenging problem, and is associated large training overhead given the massive number of LIS elements. This proposes efficient solutions for these problems by leveraging tools from sensing and deep learning. First, a novel LIS architecture based on channel sensors is proposed. In this architecture, all the LIS elements passive except for a few elements that are active (connected to the of the LIS controller). We then develop two solutions that design the reflection matrices with negligible training overhead. In the first, we leverage compressive sensing tools to construct the channels at the LIS elements from the channels seen only at the active elements. These channels can then be used to design the LIS reflection matrices with no overhead. In the second approach, we develop a deep learning based where the LIS learns how to optimally interact with the incident given the channels at the active elements, which represent the current of the environment and transmitter/receiver locations. We show that the rates of the proposed compressive sensing and deep learning approach the upper bound, that assumes perfect channel knowledge, negligible training overhead and with less than 1% of the elements being.