Deep Learning Method for Generalized Modulation Classification under Varying Noise Condition
Modulation signal classification (MSC) is an indispensable technique to make the possible applications of non-cooperative communications. Currently, convolutional neural network (CNN) based MSC techniques can achieve an outstanding performance at a fixed noise regime. However, they are hard to generalize to all of noise scenarios. Because these conventional methods are trained on specific signal samples with fixed SNR and they only perform well under corresponding noise condition. Unlike the conventional methods, in this paper, we propose a robust CNN based generalized MSC (GMSC) method with powerful generality capability. This capability stems from the mixed dataset, containing in-phase and quadrature (IQ) samples under various SNR regimes. Experimental results show that the proposed method is robust under varying noise conditions, while merely losing a slight performance with comparing with conventional methods.