RIS-Enabled Smart Wireless Environments: Fundamentals and Distributed Optimization

TL;DR

The chapter addresses how RIS-enabled SWEs can turn the propagation environment into a programmable resource, enabling joint optimization of RIS configurations and transceiver processing. It advances two distributed SWE designs based on Beyond-Diagonal RISs: (i) a multi-user MISO downlink with a scalable hybrid distributed ML framework (MBACNN with CoSyNE training) achieving near-optimal sum-rate with low online complexity, and (ii) a wideband interference broadcast setting with cooperative optimization across multiple BD-RISs and BSs. The work highlights BD-RIS advantages for interference management, spectral efficiency, and scalability, and demonstrates strong performance gains over diagonal RIS benchmarks. Collectively, these results underscore the practical viability of distributed RIS control and ML-driven configuration in next-generation wireless networks, including potential benefits for ISAC and OTA computation paradigms.

Abstract

This chapter overviews the concept of Smart Wireless Environments (SWEs) motivated by the emerging technology of Reconfigurable Intelligent Surfaces (RISs). The operating principles and state-of-the-art hardware architectures of programmable metasurfaces are first introduced. Subsequently, key performance objectives and use cases of RIS-enabled SWEs, including spectral and energy efficiency, physical-layer security, integrated sensing and communications, as well as the emerging paradigm of over-the-air computing, are discussed. Focusing on the recent trend of Beyond-Diagonal (BD) RISs, two distributed designs of respective SWEs are presented. The first deals with a multi-user Multiple-Input Single-Output (MISO) system operating within the area of influence of a SWE comprising multiple BD-RISs. A hybrid distributed and fusion machine learning framework based on multi-branch attention-based convolutional Neural Networks (NNs), NN parameter sharing, and neuroevolutionary training is presented, which enables online mapping of channel realizations to the BD-RIS configurations as well as the multi-user transmit precoder. Performance evaluation results showcase that the distributedly optimized RIS-enabled SWE achieves near-optimal sum-rate performance with low online computational complexity. The second design focuses on the wideband interference MISO broadcast channel, where each base station exclusively controls one BD-RIS to serve its assigned group of users. A cooperative optimization framework that jointly designs the base station transmit precoders as well as the tunable capacitances and switch matrices of all metasurfaces is presented. Numerical results demonstrating the superior sum-rate performance of the designed RIS-enabled SWE for multi-cell MISO networks over benchmark schemes, considering non-cooperative configuration and conventional diagonal metasurfaces, are presented.

Figures (12)

• Figure 1: The architecture of the proposed HDF ML framework. Each $k$-th BD-RIS controller designs, via an appropriately designed NN, the configuration $\boldsymbol{\Phi}_k(t)$ of its metasurface panel as well as a candidate index set $\mathcal{I}_k(t)$ from the BS precoder matrix codebook. Those $K$ index sets are then collected, via control links, and fused, through an adequately designed NN, at the BS side to decide the final multi-UE precoding matrix $\boldsymbol{V}(t)$. The details of the proposed Multi-Branch Attention Convolution NN (MBACNN) and the Feed Forward (FF) NN are presented in Section \ref{['sec:HDF']}.
• Figure 2: The amplitudes of the real and imaginary parts of a $10\times50$ channel matrix $\mathbf{H}_{1,k}(t)$ constituting a realization of the Ricean distribution Alexandropoulos2021_PervasiveML with $7$ dB $\kappa$-factor. It can be observed that adjacent matrix elements have similar values (spatial correlation), a fact that motivates the investigation of attention mechanisms as a means to extract important channel features.
• Figure 3: The proposed MBACNN architecture with its CSI inputs $\tilde{\mathbf{H}}_{\rm D}(t)$, $\tilde{\mathbf{H}}_{1,k}(t)$, and $\tilde{\mathbf{H}}_{\rm I,2,k}(t)$ at each time instance $t$, which is hosted at each $k$-th BD-RIS controller. This architecture comprises a multi-branch attention module, followed by a convolutional NN module, a Multi-Layered Perceptron (MLP) module for the selection of the BD-RIS configuration matrix $\boldsymbol{\Phi}_k(t)$, and an additional MLP module for selecting the set of indices $\mathcal{I}_k(t)$ indicating the candidate BS precoding matrix $\mathbf{V}_k(t)$.
• Figure 4: Block diagram of the scaled-dot-product self-attention layer for feature extraction on the channel matrix $\mathbf{\tilde{H}}_{1,k}$.
• Figure 5: Achievable average rate performance in bits/s/Hz versus the transmit power $P$ in dBm with the presented HDF approach, trained via NE as presented in Section \ref{['sec:NE_training']}, and the considered benchmark schemes, considering a single ($K=1$) conventional RIS ($N_{\rm B}=0$) with $N_{\rm ris}=400$ uncoupled unit elements, a BS with $N_{\rm tx}=16$ transmit antennas, and a single ($N_{\rm ue}=1$) UE with noise level of $-50$ dBm. All RIS-aided wireless fading channels were simulated as Ricean distributed with $\kappa=10$ dB, while the direct BS-UE channel was blocked.