Intelligent PHY/MAC design for WiFi: Agentic AI and Wireless Physical Layer Foundation Model with Joint Communication and Sensing

The rapid growth of Wi-Fi density in homes, enterprises, and public spaces is pushing the limits of current PHY/MAC designs. As more devices contend for spectrum, networks must cope with dynamic interference, spectrum contention, variable channel conditions, and coexistence with heterogeneous technologies. Traditional model-based approaches to PHY/MAC optimization — such as analytical link adaptation, interference cancellation, interference detection, and scheduling — have been effective under controlled assumptions, but their performance deteriorates in real-world environments that are non-stationary, multipath-rich, and interference-dominated. While effective in controlled or idealized scenarios, these models struggle in real-world dense Wi-Fi deployments, where channels are highly dynamic due to mobility and multipath propagation, interference is unpredictable, and contention patterns are complex. As a result, they often fail to generalize beyond their assumptions, leading to suboptimal performance, poor adaptability, and limited robustness.

The PhD will investigate developing an agentic AI framework with interaction with a Wireless Physical Layer Foundation Model [1], a large-scale, pre-trained model with knowledge of wireless environments, traffic patterns, and situational awareness, to enable Joint Communication and Sensing-driven WiFi optimization. Unlike static model-based techniques, agentic AI establishes a perception-reasoning-action loop using reinforcement learning, allowing the system to sense the environment and adapt on the MAC/PHY stack in a decentralized way. This includes not only waveform design, modulation and coding scheme (MCS) adaptation, and scheduling, but also intelligent grouping and steering of traffic flows across multiple APs in dense deployments.

To validate the framework in real-world conditions, the research will incorporate the openwifi platform [2] as an experimental testbed. openwifi offers a fully open-source, programmable Wi-Fi PHY/MAC stack, enabling fine-grained control and observability of the wireless link. This allows the deployment and testing of agent behaviors, real-time WPFM interaction, and cross-layer policy enforcement, bridging the gap between theoretical models and practical wireless systems.

The proposed PhD topic is strategically aligned with imec’s connectivity department and aims to enhance and expand the existing openwifi and time-sensitive networking portfolio. By focusing on generalized concepts, the research is designed to be scalable and applicable to future imec technologies, including those operating in high-frequency bands. Furthermore, the topic intersects with imec’s AI Systems division, as it explores the integration of machine learning techniques into network components such as access points and potentially end devices. This integration demands a thoughtful approach, balancing computational and communication constraints to ensure efficient and practical deployment. Overall, the topic represents a forward-looking initiative that bridges connectivity and AI, reinforcing imec’s long-term innovation strategy.

Objectives:

• Design and implement agentic AI architectures tailored for decentralized Wi-Fi RL.
• Develop interfaces and interaction protocols between RL agents and the WPFM
• Integrate and adapt the WFM for real-time use in heterogeneous Wi-Fi environments.
• Design and evaluate AI-driven mechanisms for MCS adaptation, waveform design, and scheduling through JCAS features
• Investigate intelligent decentralized traffic flow grouping and steering across multiple APs, enabling adaptive coordination in dense deployments.
• Implement and evaluate agent-based policies on openwifi, focusing on throughput, interference mitigation, mobility support

Expected Results:

• A novel AI-driven multi-agent system for dynamic Wi-Fi coordination.
• A lightweight, generalizable interface between RL agents and WFMs.
• Demonstrated improvements in MCS selection, waveform adaptability, and scheduling efficiency compared to static or rule-based methods.
• Demonstrated performance improvements using openwifi in realistic communication and sensing conditions

[1] Fontaine, J., Shahid, A., & De Poorter, E. (2024, June). Towards a wireless physical-layer foundation model: Challenges and strategies. In 2024 IEEE International Conference on Communications Workshops (ICC Workshops) (pp. 1-7). IEEE.

[2] https://github.com/open-sdr/openwifi