[Paper Review] Enabling Large Intelligent Surfaces with Compressive Sensing and Deep Learning

Abdelrahman Taha, Muhammad Alrabeiah+1 more|arXiv (Cornell University)|Apr 22, 2019

Advanced Wireless Communication Technologies139 citations

TL;DR

The paper proposes a sparse-sensor LIS architecture and uses compressive sensing and deep learning to design reflection matrices with negligible training overhead, achieving near-optimal rates with only a small fraction of active elements.

ABSTRACT

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

Motivation & Objective

• Motivate and design large intelligent surfaces (LIS) to improve coverage and rate.
• Address the challenge of massive training overhead and hardware complexity for LIS channel estimation.
• Propose an energy-efficient LIS architecture with a few active sensors.
• Develop compressive sensing and deep learning solutions to design LIS reflection matrices with negligible training overhead.

Proposed method

• Propose an LIS architecture with M passive elements and Ḿ active channel sensors randomly distributed on the surface.
• Formulate the LIS interaction as a diagonal phase-shifting matrix and a codebook of quantized beamforming vectors.
• Recover full LIS-transmitter/receiver channels from sparsely sampled channels using compressive sensing (CS) and an OMP-like approach.
• Construct full channels via sparse representations using a dictionary of array responses and grid-based azimuth/elevation directions.
• Use offline search over a codebook to find the optimal reflection vector that maximizes the sum rate across subcarriers (no online beam training).
• Share a parallel deep learning approach (described in the paper) that maps sampled channels to optimal reflection configurations without full channel knowledge.

Experimental results

Research questions

• RQ1 How can LIS performance be enhanced with a sparse, energy-efficient hardware architecture?
• RQ2 Can compressed sensing recover full LIS-channel information from a small set of active sensors to enable near-optimal reflection design?
• RQ3 Can deep learning map partial channel observations to near-optimal LIS reflection matrices without full channel knowledge?
• RQ4 What training overhead and energy efficiency gains are achievable with sparse active LIS elements compared to fully-connected LIS architectures?

Key findings

• An LIS architecture with only a small fraction of active sensors (less than 1% active elements) can approach the upper-bound rate with negligible training overhead.
• Compressive sensing based design recovers full channels from sampled channels, enabling near-optimal rate performance using offline reflection codebook search.
• Performance gains are demonstrated on DeepMIMO ray-tracing data at 3.5 GHz and 28 GHz, showing near-optimal rates with sparse sensing.
• The approach yields significant energy efficiency benefits due to the majority of LIS elements remaining passive.
• Limitations include reliance on channel sparsity and array geometry knowledge, and potential degradation in rich scattering environments.