Research on Signal Modulation Recognition in Wireless Communication Network by Deep Learning

The rapid identification of the modulation type of wireless communication signal can improve communication efficiency and quality. This paper briefly introduced modulation signal, constellation diagram and the convolutional neural network (CNN), which was used for quickly identifying the modulation type, and improved CNN with particle swarm optimization (PSO) to overcome the optimal local solution produced in training. Then, the simulation experiment was carried out on the three recognition models, Back Propagation (BP), traditional CNN, and improved CNN. The results showed that the signals of the same modulation type had similar distribution after converting into the constellation diagram, and signals of different modulation types had significantly different constellation diagram; in the process of model training, the improved CNN had the fastest convergence, and the training loss after the convergence stability was the smallest, followed by the traditional CNN, and the BP had the slowest convergence and the most loss after convergence stability; with the increase of signal-tonoise ratio (SNR) of the detection signal, the average accuracy of the three recognition models showed a tendency of increasing first and then being stable; under the same SNR, the recognition accuracy of the improved CNN was the highest, followed by the traditional CNN and BP.