AirComp-Based Federated Learning Enhanced by IRS for Secure Transmission in IIoT

To ensure the secure transmission in Industrial Internet-of-Things (IIoT) networks, we investigate an intelligent reflecting surface (IRS)-enahnced over-the-air computation (aircomp)-based federated learning ($F L$) scheme. In such scheme, the aircomp-based $F L$ is utilized to preserve the privacy of each user in IIoT, and the IRS is deployed to improve the propagation environment from users to a base station (BS). To detect malicious users who send corrupted local models to the BS to further guarantee the secure transmission of authorized users, we employ a successive interference cancellation algorithm to decode the local models of participating users in the aggregation process of the aircomp-based FL. Furthermore, to distinguish malicious users while obtaining the aggregated model with small estimation error, we formulate the aggregation minimization problem under the quality of service constraint of each user by jointly designing the phase shifts at the IRS, the decoding vector at the BS, and the transmitting powers of participating users. We propose an alternating optimization (AO) algorithm to tackle such a complex problem where the original problem is decomposed into three subproblems and solved in an iterative manner. Simulation results not only verify the efficiency of the IRS, but also show that the proposed $A O$ algorithm can significantly decrease the meansquare-error (MSE) of the aggregated model compared with the existing methods.