Robust Automated VHF Modulation Recognition Based on Deep Convolutional Neural Networks

Abstract

This letter proposes a novel modulation recognition algorithm for very high frequency (VHF) radio signals, which is based on antinoise processing and deep sparse-filtering convolutional neural network (AN-SF-CNN). First, the cyclic spectra of modulated signals are calculated, and then, low-rank representation is performed on cyclic spectra to reduce disturbances existed in VHF radio signals. After that, before fine tuning the CNN, we propose a sparse-filtering criterion to unsupervised pretrain the network layer-by-layer, which improves generalization effectively. Several experiments are taken on seven kinds of modulated signals, and the simulation results show that, compared with the traditional methods and some renowned deep learning methods, the proposed method can achieve higher or equivalent classification accuracy, and presents robustness against noises.