AI-enhanced intelligent surfaces set to transform 6G coverage and energy efficiency

The author finds that hybrid models maintain several benefits simultaneously: improved path gains compared to baseline analytical approaches, generalizability across different user locations, and significantly lower computational burden compared to full optimization.

A new investigation unveils how advanced reconfigurable intelligent surfaces could transform the performance, sustainability, and reliability of future wireless networks. The study evaluates analytical, optimization-driven, and AI-enhanced RIS strategies in realistic environments and warns that the next generation of wireless systems will depend on adaptive, intelligent radio environments rather than incremental improvements to base stations alone.

The findings are detailed in the paper “Toward Intelligent and Sustainable Wireless Environments with Hybrid and AI-Enhanced RIS Strategies,” published in Electronics. The research compares five major RIS approaches, two analytical, one fully optimization-based, and two hybrid artificial intelligence–supported methods, under identical physical conditions using detailed ray-tracing simulations.

Through extensive modeling, the study identifies where each RIS strategy succeeds, where it breaks down, and why AI-driven hybrid methods may become essential in building sustainable 6G intelligent environments that balance performance, reliability, and computational cost.

How the study defines RIS as the architectural backbone of 6G environments

According to the author, RIS is a crucial architectural layer that controls radio propagation rather than leaving it to chance. The surface, composed of a grid of tunable elements, can manipulate electromagnetic waves by adjusting amplitude and phase to focus, redirect, or enhance signal strength.

The study defines RIS as a sustainable, low-power alternative to traditional relay systems because it requires no active amplification. Instead, it uses intelligent reflection to strengthen paths that would otherwise be weak due to blockages or long-distance propagation. This makes RIS central to addressing limitations in high-frequency 6G communications, where signals face severe attenuation.

The paper categorizes RIS designs into three classes. The first comprises purely analytical models: the Phase Gradient Reflector (PGR) and Focusing Lens (FL). These are simple to compute and follow well-defined mathematical structures but lack the agility needed for complex or dynamic environments. The second category is fully optimization-based, represented by the Gradient-Based Optimization (GBO) method, which seeks the ideal amplitude and phase configuration for each RIS element through iterative calculations. The third category includes hybrid strategies that blend analytical formulas with AI-driven adaptability, including a Mixture-of-Experts (MoE) framework and a CNN-based gating model.

All five strategies are tested in a controlled campus scenario using Sionna RT, a realistic 3D ray-tracing environment that simulates electromagnetic propagation with high fidelity. The scenario includes varied transmitter and receiver positions, obstructed and open regions, and both near-field and far-field user locations. This controlled environment enables a direct and fair comparison across all methods, allowing the research to parse out the strengths and weaknesses of each RIS category.

Under this model, the PGR and FL methods deliver predictable and interpretable behavior. PGR behaves as a simple beam deflector, achieving average path gains around −112 dB. FL achieves stronger focusing, with gains around −97 dB. These values provide baselines for comparing more advanced approaches.

On the other hand, the fully optimized GBO method achieves the strongest path gain at approximately −92 dB, produced by carefully tuned amplitude and phase values across the entire surface. The simulations reveal that GBO can substantially enhance signal strength but at the cost of high computational complexity. It is also highly localized: it optimizes coverage for a single user at a time, which reduces its ability to generalize across large areas or multiple receivers.

The hybrid AI-supported methods fall between these extremes. The Mixture-of-Experts model reaches roughly −93.5 dB on average and demonstrates the ability to create highly localized coverage improvements of more than 40 dB. It uses analytical experts for baseline predictions and a learned gating function that selects which expert to favor based on position. The approach offers strong performance without the heavy computational load of full optimization.

The CNN-based gating framework achieves smoother coverage patterns, improving signal strength across wider regions and boosting weaker areas by up to 20 dB. Its output resembles a learned segmentation of the environment: binary-like decisions that strengthen some areas while deprioritizing others. This offers better general usability for multi-user scenarios where broad improvement is more valuable than extreme localization.

Taken together, the RIS strategies represent a continuum: from interpretable but limited models to advanced AI-driven configurations that deliver high improvements while remaining computationally manageable. The author argues that hybrid and AI-enhanced models provide the most promising blueprint for future intelligent wireless design.

Where hybrid AI RIS strategies outperform and where they introduce new constraints

The author finds that hybrid models maintain several benefits simultaneously: improved path gains compared to baseline analytical approaches, generalizability across different user locations, and significantly lower computational burden compared to full optimization.

One of the clearest advantages is the ability to leverage the strengths of both learning-based and analytical approaches. In the Mixture-of-Experts system, analytical models provide stable baseline predictions while the AI component adjusts behavior in response to user locations. This combination produces high-quality coverage without the susceptibility to overfitting or instability observed in purely learned systems. It also means that prediction time is extremely fast, enabling adaptive real-time reconfiguration.

Meanwhile, the CNN-gating method excels at identifying key regions in the environment where reinforcement of the signal has the largest system-level impact. In the simulated campus scenario, this produces more equitable enhancement, allowing the RIS to serve multiple receivers more reliably. The system avoids extreme focusing, which is useful in multi-user environments where narrow beams could leave many users underserved.

The study warns, however, that hybrid methods add complexity to design and deployment. They require pre-trained gating networks, curated datasets, and thoughtful integration with analytical models. In addition, because they operate on learned perceptions of spatial patterns, hybrid RIS solutions may require retraining when deployed in significantly different physical environments.

The research further highlights that even hybrid RIS strategies cannot escape certain physical constraints. For instance, near-field users receive better highly localized improvements, but these gains diminish as users move farther away from the surface. Likewise, since RIS is passive, it cannot amplify signals beyond what the physical surface permits. Its energy efficiency becomes a major advantage, but the ability to enhance links remains bounded by the electromagnetic properties of the system.

The author notes that these constraints must be managed through system-level design. Future networks must consider where RIS tiles are placed, how they interact with buildings, how often configurations change, and how multiple RIS surfaces communicate or coordinate with base stations.

Why intelligent and sustainable wireless environments depend on hybrid RIS for 6G

The third major issue explored in the study concerns the sustainability and scalability of RIS solutions in the long-term evolution of wireless communication. As global data demand increases, the energy footprint of wireless networks is projected to grow substantially. Conventional solutions, such as deploying more base stations, would significantly increase energy consumption and infrastructure costs.

RIS offers a low-power alternative because the surfaces operate passively, reflecting waves without power-hungry amplifiers. The study emphasizes that RIS can become central to sustainable 6G systems by replacing inefficient relay nodes and providing wide-area improvements with minimal energy draw.

However, sustainability in this context requires more than energy savings. It demands efficient configuration algorithms, scalable deployment strategies, and flexible systems that respond dynamically to user mobility and environmental shifts. Analytical models alone lack the adaptability to achieve this. Optimization-only models consume too much computing power and cannot scale to dynamic multi-user networks.

This is where AI-enhanced RIS strategies become indispensable. The hybrid models demonstrated in the study offer both improved performance and computational practicality. They strike a balance between sustainability and adaptability, enabling real-time decision-making with minimal energy use.

The author envisions future intelligent wireless environments where RIS panels are integrated into building surfaces, public infrastructure, indoor spaces, transportation corridors, and industrial settings. In this scenario, hybrid and AI-driven strategies allow these surfaces to coordinate, redirect signals, recover obstructed paths, and handle user mobility with minimal overhead.