A newly developed multi-objective optimization framework utilizes deep reinforcement learning (DRL) to concurrently enhance energy efficiency, spectral efficiency, and environmental sensing accuracy within integrated sensing and communication (ISAC) systems for hybrid sixth-generation (6G) cellular and Wi-Fi 8 networks. The research addresses the inherent challenges of high-density and heterogeneous wireless network environments, where the demand for ultra-high data rates and low-latency communication necessitates a sophisticated integration of radar-like sensing and data transmission. By employing reconfigurable intelligent surfaces (RIS) to manipulate the wireless propagation channel, the proposed system model enables an intelligent, software-defined environment that manages not only information transmission but also real-time environmental perception, providing a roadmap for the future of wireless infrastructure.
The network design employs a hybrid architecture that coordinates 6G base stations and Wi-Fi 8 access points, allowing for seamless and directional data communication across cellular and wireless local area network (WLAN) standards. The system relies on an AI-based central controller that utilizes DRL to dynamically perceive the network state, including channel state information, traffic demand, user mobility, and resource availability. This controller adaptively selects decisions regarding RIS phase shifts, beamforming strategies, and power allocation to optimize a multi-objective reward function. The framework represents a shift from conventional methods, which have typically focused on energy or spectral efficiency independently, to an all-inclusive solution for the effective coexistence of diverse radio access technologies.
In formulating the optimization problem, the research identifies a non-convex objective function that integrates energy efficiency (EE), spectral efficiency (SE), and sensing performance metrics under specific quality of service (QoS) constraints. The decision variables include transmit power at the base stations and access points, beamforming vectors, and the RIS phase-shift matrix. The reward function is defined as a weighted sum of the three objectives, with specific weighting factors assigned to each: 0.3 for energy efficiency, 0.4 for spectral efficiency, and 0.3 for sensing performance. This multi-objective approach ensures that the DRL agent can balance competing demands in dynamic environments, with the framework adhering to constraints on data rates, transmit power, and discrete phase shift limitations inherent in practical RIS hardware.
The system model implements adaptive RIS phase control as a mechanism to improve link quality and reduce overall energy dissipation, leveraging direct and reflected signals to enhance communication and sensing capabilities. The receiver baseband signal is mathematically modeled to account for direct channels, RIS-to-user channels, and additive white Gaussian noise. By employing the DRL controller to dynamically optimize the phase shifts of the RIS and beamforming vectors, the framework is capable of suppressing multi-user interference and effectively countering obstructions in millimeter wave and terahertz frequency bands. This ability to steer reflected signals towards intended users maintains a stable communication link even under complex propagation conditions.
Simulation results confirm that increasing the number of RIS elements significantly improves the performance metrics of the system. For a configuration with 32 RIS elements, the spectral efficiency reaches 7.1 bits/s/Hz, while increasing the array size to 128 RIS elements boosts spectral efficiency to 12.3 bits/s/Hz. Furthermore, the energy efficiency values demonstrate a notable enhancement, rising from 0.6 bits/J to 1.55 bits/J as the number of RIS elements increases. These results indicate that larger RIS arrays provide superior passive beamforming capabilities, enabling the system to achieve its targets with lower transmission power from base stations and access points, thereby contributing to the overall sustainability of the network.
Despite the synergistic scaling observed with larger RIS configurations, the research identifies a critical trade-off between system performance and computational complexity. As the number of RIS elements expands, the feedback overhead associated with channel state information and the complexity of the DRL algorithm increase. For hybrid Wi-Fi 8 and 6G deployments, the study identifies 64 RIS elements as an optimal balance that captures substantial spectral efficiency and sensing gains without the exponential increase in control signaling required for larger arrays. This finding is particularly relevant for the practical implementation of RIS technologies, where managing the state-action space is a primary constraint on real-time edge deployment.
Sensitivity analysis conducted on the weighting factors reveals that prioritizing sensing accuracy introduces a quantifiable trade-off with spectral efficiency, as system resources are diverted to optimize environmental perception. When the weight assigned to sensing is increased, there is a slight drop in spectral efficiency due to a shift in resource allocation toward sensing nulls. This trade-off underscores the necessity of the multi-objective DRL approach, which allows for dynamic adaptation to shifting network priorities. The researchers found that the chosen weighting factors of 0.3, 0.4, and 0.3 represent a balanced operating point that maintains reliable communication while ensuring accurate environmental sensing capabilities.
The implementation of the deep reinforcement learning framework involves significant training overhead, with the Deep Deterministic Policy Gradient (DDPG) method requiring approximately 400 episodes to reach stable multi-objective convergence. The authors note that the high-dimensional state-action space makes this process computationally intensive, indicating that the framework requires offline pre-training before it can be effectively deployed in real-time edge scenarios. Furthermore, the performance results rely on the assumption of perfect channel state information, which serves as an upper bound performance metric. In ultra-dense networks, obtaining this information in real-time could pose considerable signaling challenges, and future iterations of the work will need to incorporate robust optimization methods to handle potential channel estimation errors.
The researchers also identified hardware imperfections as a factor that may cause deviations between simulated results and experimental outcomes. The current design utilizes 2-bit RIS phase precision to reduce complexity, but factors such as phase quantization error and atmospheric absorption at high frequencies remain significant considerations. While the system demonstrates promising scalability and performance in a controlled hybrid environment, the rapid growth in the state-action space size for larger arrays provides further justification for the selection of moderate array configurations. These limitations highlight the ongoing need for research into more efficient beamforming techniques and improved hardware integration for next-generation software-defined wireless systems.
Looking toward future developments, the study suggests that the integration of advanced variants such as light-emitting reflective intelligent surfaces (LeRIS) could extend the capabilities of the proposed methodology beyond the simple reflector function of traditional RIS. These active elements offer increased degrees of freedom for positioning and mapping, which may necessitate more complex control approaches. Research directions include reducing the computational complexity of the DRL algorithms and exploring the integration of the current framework with physics-informed metaheuristics to achieve faster codebook compilation, ultimately aiming to provide a robust, sustainable solution for the demands of hybrid 6G and Wi-Fi 8 networks.
In establishing the simulation framework, the researchers modeled network coexistence across high-frequency 6G bands and sub-6 GHz Wi-Fi 8 channels over a 100 MHz system bandwidth. Multiple users were randomly distributed within the service region, allowing the multi-RAT framework to steer devices between cellular base stations and wireless local access points based on local channel states. Specific transmission power thresholds constrained base station emissions to a maximum of 30 dBm for 6G transmitters and 23 dBm for Wi-Fi 8 access points, while user requirements enforced a minimum data rate constraint. Propagation modeling incorporated large-scale path loss and small-scale fading characteristic of millimeter wave and sub-terahertz frequencies, alongside additive white Gaussian noise. To validate the proposed methodology, the study benchmarked system outcomes against four comparative baselines: a conventional multi-RAT architecture lacking RIS and ISAC functionality, a RIS-assisted network omitting sensing integration, a cognitive resource block allocation scheme utilizing RIS support without joint sensing optimization, and a single-objective model configured exclusively for spectral efficiency.
Numerical evaluations across array sizes demonstrated marked progressions in transmission capacity and perception fidelity. At the lower baseline of 16 RIS elements, the system attained a spectral efficiency of 4.5 bits/s/Hz and an environmental sensing accuracy of 0.4. Scaling the surface configuration to 128 elements expanded sensing accuracy to 0.95 alongside the peak spectral efficiency of 12.3 bits/s/Hz. To translate these physical gains into functional network operations, the architecture coordinates metasurface phase shifts through standard network protocols, executing packet-level synchronization between cellular base stations and local access points. Mathematical modeling defined the achievable data rate as a logarithmic function of the signal-to-interference-plus-noise ratio, enabling the controller to dynamically adjust beamforming vectors and suppress co-channel interference across shared frequencies.
Adaptive control within the hybrid architecture depends on closed-loop sensing exchanges between the radio environment and the cognitive controller. The integrated sensing module continuously updates a dedicated perception vector containing channel estimations, detected obstacles, and ambient interference levels. This sensing feedback feeds directly into the reinforcement learning state space alongside channel state information, traffic demands, and user positions. In optimizing the reward function, weighting parameters are strictly normalized to prevent any single operational objective from dominating policy updates. Furthermore, because theoretical continuous phase shifts cannot be deployed on discrete hardware, the continuous action outputs produced by the learning algorithm are quantized to the nearest states in a 2-bit codebook, substantially curtailing control signaling overhead while maintaining robust environmental awareness.
Subject of Research: Multi-objective optimization of integrated sensing and communication in hybrid wireless networks using reconfigurable intelligent surfaces.
Article Title: Intelligent ISAC-Enabled RIS-Assisted Hybrid 6G and Wi-Fi 8 Networks: A Multi-Objective EE–SE–Sensing Optimization Framework
Article References: Aasa, Z. (2026). Intelligent ISAC-Enabled RIS-Assisted Hybrid 6G and Wi-Fi 8 Networks: A Multi-Objective EE–SE–Sensing Optimization Framework. International Journal of Networked and Distributed Computing, 14(2), Article 30. https://doi.org/10.1007/s44227-026-00115-1
Image Credits: AI Generated
DOI: 10.1007/s44227-026-00115-1
Keywords: 6G Communications, Wi-Fi 8, ISAC, Reconfigurable Intelligent Surfaces, Deep Reinforcement Learning, Energy Efficiency, Spectral Efficiency, Wireless Sensing, Network Optimization, Multi-objective Optimization, Beamforming, Heterogeneous Networks, Signal Processing, Radio Access Technology, Channel State Information
