Reconfigurable Intelligent Surface Networks

• RIS Networks are wireless systems that leverage programmable metasurfaces with tunable elements to actively control electromagnetic wave propagation.
• They incorporate diverse architectures—including passive, active, and hybrid designs—to optimize beamforming, mitigate interference, and improve signal quality.
• RIS solutions use advanced optimization and learning-based methods for channel estimation and dynamic configuration, enabling enhanced 6G communications and sensing.

Reconfigurable Intelligent Surface (RIS) Networks are wireless networks that leverage programmable electromagnetic metasurfaces comprising large numbers of electrically tunable elements to manipulate the propagation characteristics of radio signals with fine spatial resolution. By dynamically adjusting the phase, amplitude, and, in some architectures, the polarization or transmission/reflection behavior of each meta-atom, RISs transform the physical wireless environment from a static and uncontrollable entity into a software-defined, virtualized component of the network. RIS networks are emerging as key enablers for 6G, offering new degrees of freedom in wireless system design, from green beamforming and energy efficiency to integrated sensing and communications.

1. Architectural Principles and System Modeling

RISs are usually implemented as planar arrays of sub-wavelength unit cells, each of which can be configured to impart a desired phase shift $e^{j\theta_n}$ and, optionally, variable amplitude $\beta_n$ on the incident electromagnetic wave. The canonical network architecture may involve a base station (BS), mobile user equipments (UEs), and one or more RISs, each with $N$ controllable elements. In the narrowband case, the end-to-end received signal at a single-antenna UE is modeled as

$$y = \left( h_d + \mathbf{h}_r^T \mathbf{\Phi} \mathbf{g} \right) x + n$$

where $h_d$ is the direct BS–UE channel, $\mathbf{g} \in \mathbb{C}^N$ is the BS–RIS channel, $\mathbf{h}_r \in \mathbb{C}^N$ is the RIS–UE channel, $$\mathbf{\Phi} = \mathrm{diag}(e^{j\theta_1}, \ldots, e^{j\theta_N})$$ is the RIS phase shift matrix, $x$ is the transmitted symbol, and $n \sim \mathcal{CN}(0,\sigma^2)$ is AWGN (Liu et al., 2024). For MIMO and/or multi-user settings, composite channels are constructed—often as $\beta_n$0 plus direct links and additive noise (Peng et al., 2023).

The path-loss for a cascaded BS–RIS–UE link in the far field scales as

$\beta_n$1

where $\beta_n$2 is the RIS element count, and $\beta_n$3, $\beta_n$4 are the BS–RIS and RIS–UE distances respectively. This quadratic array gain (incoherent power addition) is a signature feature of RIS-assisted links (Liu et al., 2024).

2. Hardware Architectures and Operating Modes

RIS hardware exists in several forms:

• Passive reflectarrays: Each meta-atom provides tunable phase shifts with negligible amplification ($\beta_n$5). Key for ultra-low power operation, serving as large, nearly lossless mirrors.
• Active/amplified RIS: Selected meta-atoms include low-noise amplifiers ($\beta_n$6) to compensate for severe double-fading in long cascaded paths, at the cost of higher power and amplified noise (Basar et al., 2023).
• Transmissive/STAR-RIS: Elements can simultaneously transmit (refract) and reflect waves, realized via programmable amplitude splitting and dual-polarized designs.
• Semi-passive/Hybrid RIS: Some elements switch to absorption, feeding sparse RF chains or low-rate ADCs to assist in CSI acquisition (Basar et al., 2023, Liu et al., 2024).

Operating modes span:

• Fully passive (no sensing—external control only),
• Semi-passive (partial on-board sensing/action),
• Network- or UE-controlled operation (central or distributed command architectures),
• Hybrid/STAR modes for simultaneous transmission, reflection, or absorption (Liu et al., 2024).

3. Channel Estimation, Configuration, and Optimization

RIS configuration typically seeks to maximize SNR or SINR by jointly optimizing the BS precoder and the RIS phases under discrete (finite-bit) or continuous phase constraint (ElMossallamy et al., 2020). The base station may employ WMMSE-based precoding (Peng et al., 2023), while the RIS phase design is rendered non-convex due to unit-modulus constraints. Algorithmic solutions include:

• Alternating Optimization (AO): Iteratively optimizes the precoder and RIS phases, leveraging convexity in each (ElMossallamy et al., 2020).
• Semidefinite Relaxation (SDR): Lifts non-convex phase constraints to the positive semidefinite cone, followed by randomization for feasibility (ElMossallamy et al., 2020).
• Heuristic/Greedy/Codebook approaches: Including element-wise greedy updates or codebook scans, which are practical albeit sub-optimal.
• Learning-based methods: Neural architectures (e.g., RISnet) that map partial or low-dimensional CSI to high-dimensional phase configurations, often via unsupervised learning on sum-rate gradients (Peng et al., 2023).

Channel estimation techniques fall into pilot-based (sparse sampling via sensors on a subset of elements), blind or semi-blind receiver-side feedback, and compressed sensing exploiting the angular sparsity of the cascaded channel (ElMossallamy et al., 2020, Basar et al., 2023).

4. Spectrum Learning, Control, and Identification

A central RIS network challenge is dynamic interference from non-intended users. Intelligent Spectrum Learning (ISL) introduces convolutional neural networks embedded at RIS controllers, operating directly on baseband I/Q samples to infer the set of active interferers and support adaptive ON/OFF control and phase-optimization (Yang et al., 2021). The ISL-aided distributed control achieves:

• Real-time identification ($\beta_n$795% CNN accuracy for interfering user detection in testbed settings),
• Drastic SINR gains (up to $\beta_n$8300% over naive policies),
• $\beta_n$9 control-plane complexity per iteration (Yang et al., 2021).

RIS identification in networks with multiple panels is crucial for resource allocation and interference management. Approaches include pilot-based code watermarking, power/amplitude modulation sequences, spectral fingerprinting, statistical model-driven, and supervised or deep learning classifiers. Performance metrics include probability of detection/false alarm, ROC curves, and complexity-throughput trade-offs (Arslan et al., 2024).

5. Frequency-Domain, Multi-Access, and Information-Theoretic Perspectives

RISs radically alter the structure of wireless multiple access:

• RIS-NOMA (Non-Orthogonal Multiple Access): Over-the-air NOMA via RIS partitioning (each partition serving a different user with over-the-air PSK/QAM modulation), eliminating SIC at the UE, outperforming TDMA/OMA in sum-rate and error probability, provided accurate CSI and tight synchronization (Dogukan et al., 2024).
• RIS-aided NOMA with static/dynamic surfaces: Dynamic RIS configurations (time-varying phases) strictly enlarge the achievable capacity region relative to OMA and static RIS, with asymmetric RIS placement (favoring strong users) preferred for NOMA (Liu et al., 2020).
• Index and Reflection Modulation: RIS element or regional selection encodes information bits; RIS time-variant phases transform unmodulated carriers into multi-level constellations, enabling new forms of physical-layer multiple access and over-the-air modulation (Basar et al., 2023).

6. Phase Alignment, Cascaded/Routing, and Large-Scale Network Considerations

Performance of RIS networks is extremely sensitive to phase alignment conditions.

• Phase Alignment Categories: Outage probability and diversity order depend on four alignment regimes—perfect (diversity $N$0), coherent (quantized errors, diversity $N$1), random (diversity $N$2), and destructive alignment (outage floor). Discrete phase quantization ($N$3) preserves full diversity, while random alignment or destructive nulls drastically degrade SNR (Xu et al., 2020).
• Cascaded RIS networks: Multiple cascaded RIS units (multi-hop reflecting paths) may extend coverage and boost SNR ($N$4 scaling), but are practical only when extra path loss is small and phase estimation errors are low (Tyrovolas et al., 2022).
• RIS Selection in Dense Deployments: With multiple candidate RISs, stochastic geometry models (PPP for RIS locations) yield closed-form selection rules and outage/rate expressions. Product-scaling ($N$5) and sum-scaling ($N$6) path-loss models dictate optimum selection policies—pick the RIS minimizing the respective metric. Even limited local feedback strategies closely approach full-CSI centralized performance (Fang et al., 2020).

7. Applications, Sustainability, and Future Directions

RIS networks are foundational for ultra-reliable, green, and adaptable wireless systems:

• Energy and Sustainability: Per-element power consumption is in the μW–mW regime, favoring ultra-dense deployments. Control circuitry and phase quantization introduce linear energy overheads with $N$7, with diminishing returns in far field for $N$8 (Liu et al., 2024).
• Sensing, Localization, and ISAC: RIS-enabled hybrid beamforming/beam-focusing supports joint positioning (sub-degree/centimeter accuracy), environment mapping, and integrated radar–communications (Stratidakis et al., 2023).
• Non-terrestrial and Satellite Networks: RISs match size, weight, and power (SWaP) constraints in LEO/HAPS/NTN networks, enable virtual LoS restoration and interference shaping, and are critical for coverage under blockage/mobility (Zheng et al., 24 Dec 2025, Tekbıyık et al., 2020).
• Vehicular and Aerial Networks: On-vehicle and roadside RISs improve V2X link reliability, enable rapid beam reconfiguration under mobility, and reduce pilot overhead by $N$960% via sparse channel tracking (Chen et al., 2022).
• Standardization: ETSI ISG RIS and 3GPP Rel-18/19 are defining architectures, signaling, and EM compatibility criteria. Key challenges remain in real-time control, channel estimation, EM compliance, multi-operator coexistence, and security (Liu et al., 2024).

---

RIS networks present a paradigm shift in both wireless theory and practice by embedding computational and electromagnetic programmability into the medium itself. Challenges at the intersection of hardware, control, and algorithmic scalability persist, but empirical and analytical evidence across deployment scenarios shows substantial, often order-of-magnitude improvements in coverage, efficiency, and network adaptability (Liu et al., 2024, Basar et al., 2023, ElMossallamy et al., 2020, Yang et al., 2021, Stratidakis et al., 2023). Ongoing efforts in machine learning-based control, massive MIMO integration, and robust standardization will define the trajectory and impact of RIS networks in the coming decade.