Robust Automatic modulation Classi cation using Deep Neural Networks

Abstract

Automatic Modulation classi cation plays key role in Cognitive Radios and Electronic warfare. is thesis looks into the bene ts of using deep learning to classify wireless signal modulations, which include ten di erent types of pulse compression radar modulations, as well as analogue and digital communication modulations. For SNR ranges ranging from -25dB to 25dB, a data set of these waveforms for three separate IF categories has been created.For three separate neural networks, the e ect of adjusting the IF value from low to medium and then to high on modulation classi cation has been observed. In the competition of modulation classi - cation between the three, the Med IF value comes out on top.Hybrid Neural Network i.e. CNN and GRU-64 called SGCNN(Sequential Gated Convolutional Neural Network) has been proposed for be er modulation classi cation performance. e proposed hybrid neural network, SGCNN (CNN+GRU), outperforms the other neural networks considered in the thesis, such as CNN, LSTM, GRU, and CNN+LSTM.Finally, it was observed that changing the activation function from relu to elu has a negative impact on the model’s classi cation accuracy