Mixed Signal Modulation Classification Based on Deep Convolutional Neural Networks

Aiming at the modulation classification of mixed signals, a modulation classification method based on Deep Learning (DL) is proposed in this paper. First, the baseband complex time-domain signal is split into two parts, the real part and the imaginary part, through In-phase and Quadrature (I&Q) modulation to form the input of the DL network. Subsequently, the design and optimization of the structural parameters of the deep convolutional neural network. Finally, accurate modulation classification of two-component signals is achieved by a multi-class loss function. The simulation results show that compared with other two DL-based two-component modulation classification methods, the detection performance of this method is significantly improved.