Distinguished Lecturer Tour: Federated Intelligence Over the Air: From Centralized to Collaborative Sensing

About

The next generation of wireless networks will no longer be confined to moving bits — they will sense, communicate, and learn simultaneously. This convergence is anticipated to enable distributed intelligence across devices, unlocking new capabilities for real-time perception and decision-making in dynamic environments. In this talk, two complementary advances in federated signal processing will be presented. First, an over-the-air federated edge learning (OTA-FEEL) framework with integrated radar sensing will be discussed. By leveraging echoes from the environment, rather than treating them solely as interference, robust model aggregation will be maintained while ensuring high-quality sensing and communication performance. A joint scheduling and beamforming design will be presented, supported by low-complexity optimization techniques, to preserve aggregation accuracy under realistic wireless conditions. Second, FedTrack, a novel federated learning–inspired algorithm for distributed target tracking, will be presented. By treating local log-likelihood functions as loss functions in a distributed optimization framework, FedTrack enables devices to collaboratively estimate a moving target's position and velocity. This communication-efficient method closely approximates centralized maximum likelihood estimation, achieving accuracy near the Cramér–Rao bound while reducing reliance on a central coordinator. Together, these developments illustrate how federated intelligence over the air can transform 6G networks into systems that not only communicate but also sense and learn collaboratively.