Communication-Efficient Personalized Federated Edge Learning for Decentralized Sensing in ISAC

As an emerging key technology for next-generation wireless networks, Integrated Sensing and Communications (ISAC) has received much attention. Ubiquitous off-the-shelf Wi-Fi devices have been used for ISAC to enable the dual function of transmitting data and monitoring the indoor environment which, however, still has many problems to solve, such as data privacy and model personalization. Motivated by this, this paper proposes a decentralized sensing framework called Cep-FEL for ISAC-based gesture recognition systems with real Wi-Fi data. Particularly, we design a feature fusion method for model training, which can not only speed up the accurate and personalized model training process but also protect user' privacy. Extensive experiments on a real-world ISAC dataset are conducted to evaluate the effectiveness of Cep-FEL and demonstrate that our method outperforms both the non-collaborative local training and Federated Averaging (FedAvg) method.