Integrating Aircomp with Isac Over Federated-Learning 6G Mobile Wireless Networks

To satisfy Quality-of-Service (QoS) requirements and reduce core server overload in distributed-computing-based $\mathbf{6 G}$ mobile wireless networks, federated learning is developed as the promising techniques. However, the data collected by mobile users varies geographically, making it essential to take the diverse mobile user locations into account in federated learning algorithms. On the other hand, the Integrated Sensing and Communication (ISAC) technique has been proposed to sense the mobile user locations and wireless-channel status through superposition signals in the same frequency band. To reduce core server computation load and achieve efficient model aggregation in federated learning systems, Over-the-Air Computing (AirComp)based federated learning is widely investigated. However, how to efficiently integrate sensing, communication, and computation techniques still remains the challenging problems. To overcome these difficulties, in this paper we propose a new schemes called Integrated Sensing, Communication, and Computation (ISCC) by integrating AirComp with ISAC to support highly-efficient model training and aggregations over federated learning 6 G mobile wireless networks. First, we develop the system architectures models for our proposed ISCC scheme and use superposition signals in AirComp as sensing functions to train a deep-learning based mobile-user perception model. Then, we develop the new federated learning algorithms to mitigate the impact of heterogeneity of datasets collected from different mobile agents located at diverse regions. Finally, we validate and evaluate our proposed schemes through simulations.