A Survey of Deep Learning Architectures for Intelligent Reflecting Surfaces

TL;DR

The paper addresses the challenge of deploying intelligent reflecting surfaces (IRS) in multi-link, dynamic wireless environments by surveying deep learning (DL) approaches for signal detection, channel estimation, and beamforming. It compares supervised, unsupervised, reinforcement, and federated learning paradigms, highlighting the trade-offs between labeled data requirements and training efficiency. Key findings show the effectiveness of end-to-end DL for detection, CNN-based DL for CSI estimation, and RL/FL-based beamforming strategies, while outlining the data, training, and hardware challenges that accompany IRS deployments. The work emphasizes practical considerations for outdoor/indoor IRS, THz channels, and integrated sensing and communications (ISAC), and suggests hybrid learning frameworks as a promising path forward.

Abstract

Intelligent reflecting surfaces (IRSs) have recently received significant attention for 6G wireless communications as they enable the control of the wireless propagation environment. The use of IRS also provides reducing the hardware complexity, physical size, weight as well as cost of conventional large antenna arrays. However, deployment of the IRS entails dealing with multiple channel links between the base station (BS) and the users. Further, the BS and IRS beamformers require a joint design, wherein the IRS elements must be rapidly reconfigured. Data-driven techniques, such as deep learning (DL), are critical in addressing these challenges. The lower computation time and model-free nature of DL make it robust against data imperfections and environmental changes. At the physical layer, DL has been shown to be effective for IRS signal detection, channel estimation, and active/passive beamforming using architectures such as supervised, unsupervised, and reinforcement learning. This article provides a synopsis of these techniques for designing DL-based IRS-assisted wireless systems.

Figures (5)

• Figure 1: IRS-assisted wireless communications for outdoor and indoor deployments. A BS on top of the infrastructure (left) communicates with the users on the ground through an intermediate IRS mounted on other buildings (center). The BS also serves users (right) inside the apartment building through an IRS placed on the wall of the room.
• Figure 2: Model-based versus learning-based frameworks for signal detection and channel estimation. Model-based approach (top) comprises multiple subsystems to process the received signal. Learning-based signal detection (bottom, left) provides an end-to-end data mapping from the corrupted symbols under the channel effects at the receiver to the transmit symbols. Learning-based channel estimation (bottom, right) maps the input received signals to the channel estimate as output labels.
• Figure 3: The mean-squared error of channel estimates normalized against ground truth channel, obtained using CNN in centralized and federated learning frameworks, MMSE and LS. The BS consists of $64$ antennas and IRS employed $64$ passive reflecting elements elbir_LISelbir2020_FL_CE.
• Figure 4: In RL, the DQN and DDPG architectures accept the same state (channel data and received SNR) and environment data (beamformers to be evaluated). The DQN involves training a single neural network based on the reward determined by the environment. On the other hand, the DDPG has multiple neural networks, where actor-critic architectures are used to compute actions and target values, respectively.
• Figure 5: Beamforming gain for IRS with discrete phase shifts, when $L=32$.