Over-the-Air Federated Learning Client Selection in Integrated Sensing, Computing and Communication
Paul Zheng, Yao Pan Zhu, H. Yulin, Anke Schmeink · 2024
In Internet-of-Things (IoT) era, integration in multiple levels is needed to effectively address the increasing demands of communication, computation, and sensing (e.g. data acquisition). Federated learning (FL) is widely regarded as a promising distributed machine learning framework to enable network intelligence in future-generation networks. Using over-the-air computation (AirComp) (already as an integration of communication and computation) for spectral-efficient FL model aggregation requires massive devices to transmit over the same orthogonal resources. For further integration gain, it is considered employing the same signal for coordinated device joint target sensing, which constitutes a fully integrated scenario of sensing, computing and communication (ISCC). This work focuses on the client selection and power control problem as crucial challenges of AirComp-FL in such scenarios while a requirement of a sensing task as target detection needs to be satisfied. A flexible system design has been proposed by allowing three groups of clients to be chosen: those participating in communication and communication, those solely for sensing, and the ones that transmit nothing. This work proposes a polynomial-time complexity algorithm. Simulation results corroborate the importance and the performance of the proposed framework.