Latency Minimization for Federated Learning in NOMA-Enabled and RIS-Assisted Industrial IoT

Federated learning (FL) offers a novel approach to safeguarding data privacy in the context of the Industrial Internet of Things (IIoT). Nevertheless, FL encounters a communication bottleneck when the number of IIoT devices increases significantly, hindering the achievement of satisfactory model accuracy. The purpose of this paper is to harness the potential benefits of non-orthogonal multiple access (NOMA) and reconfigurable intelligent surface (RIS) to improve the federated learning (FL) performance in IIoT. We specifically formulate a joint power allocation, receiving beamforming and RIS reflection optimization problem to minimize the latency in local computing and model uploading during FL training. We offer a solution based on alternating optimization for transmission power, receiving beamforming and RIS reflection. Numerical results show that the RIS-NOMA reduces the latency and test error when compared with the baseline techniques.