Deep neural network-based automatic modulation classification technique
ByeoungDo Kim, Jaekyum Kim, Hyunmin Chae, Dongweon Yoon, Jun Won Choi · 2016
Deep neural network (DNN) has recently received much attention due to its superior performance in classifying data with complex structure. In this paper, we investigate application of DNN technique to automatic classification of modulation classes for digitally modulated signals. First, we select twenty one statistical features which exhibit good separation in empirical distributions for all modulation formats considered (i.e., BPSK, QPSK, 8PSK, 16QAM, and 64QAM). These features are extracted from the received signal samples and used as the input to the fully connected DNN with three hidden layer. The training data containing 25,000 feature vectors is generated by the computer simulation under both additive Gaussian white noise (AWGN) and Rician fading channels. Our test results show that the proposed method brings dramatic performance improvement over the existing classifier especially for high Doppler fading channels.