Real-Time Embedded Adaptive Learning for Wireless Communications

Objective

Wireless communications play a critical role in today’s digital society, with growing demands for speed, reliability, adaptability, and energy efficiency. Yet, modern wireless transceivers still rely on fixed model-based designs developed decades ago, limiting their ability to cope with the increasingly complex and dynamic environments in which they are expected to operate in future generations. Artificial intelligence (AI) offers a promising alternative, enabling communication devices to learn and adapt autonomously. However, current AI methods are typically resource-intensive, static, and ill-suited for real-time operation on low-cost embedded hardware. REALCOM proposes a breakthrough: the development and prototyping of a new class of lightweight, continuously adaptive AI-powered wireless transceivers. These systems will operate in real time on software-defined radios (SDRs), demonstrating autonomous adaptation to varying environments without relying on hand-crafted models. Building on the scientific foundation of the ERC Starting Grant “FLAIR,” this project will extend our methodology beyond simulation to a working prototype operating in the widely used WiFi band. The project will culminate in a fully functional end-to-end system, where transmitter and receiver jointly learn to communicate more efficiently over time as practical realization of semantic communication. REALCOM will deliver a tangible proof-of-concept for a disruptive AI-driven communication paradigm, laying the groundwork for industrial adoption. It will explore commercial potential through early business development, IP protection, and stakeholder engagement, aiming to position REALCOM as a key enabler for next-generation wireless technologies.