Personalized Recognition for Distributed Jamming in Dynamic Environments

Abstract

In this letter, we investigate distributed jamming recognition based on federated learning (FL). Since the conventional FL always renders slow convergence and reduced accuracy with non-independent and identically distributed (non-IID) jamming, a self-adaptive personalized FL (SPFL) algorithm is proposed. We first develop a personalized FL training architecture to incorporate the independent local learning into the conventional FL with a weighting factor. Then a self-adaptive adjustment mechanism of the weighting factor is proposed to strike a trade-off between generalization and distinctness according to the jamming characters. The simulation results illustrate that the proposed algorithm exhibits superior performance compared with conventional counterparts in dynamic and heterogeneous jamming scenarios involving distributed edge users.