A Pruning Neural Network for Automatic Modulation Classification

Automatic modulation classification (AMC) is a promising technology for non-cooperative communication systems in both military and civilian scenarios. Nowadays, More and more scholars apply deep learning (DL) framework to AMC. However, most of the papers do not consider that the typical deep learning model is difficult to deploy on the resource constrained devices. In this paper, a lightweight DL based Average percentage of zeros (APoZ) is used with pruning neural. We introduce a novel method of generating modulation signals called contour stellar image (CSI). We train the data through some scaling factors in convolution neural network (CNN) especially the AlexNet. It can screen out the inconsequential neurons which can be pruned. Experimental results suggest that using APoZ to prune can not only slim the network but also stabilize the average error about 1.05% compared with the original network.