DFT signal detection and channelization with a deep neural network modulation classifier
Nathan West, Kellen Harwell, Ben McCall · 2017
A system capable of detecting and classifying narrowband signals transmitted over the air at radio frequency is described. The system is composed of two parts: (1) a signal detector and channelizer; (2) a radio-frequency modulation classifier. The signal detector uses an FFT for band edge detection. The channelizer uses the estimated bands and FFT vector to create a variable number of resampled time-domain streams (1 for each band detected) that are put in a queue for classification. The classifier is a deep neural network trained to classify the modulations expected. Overall system architecture consisting of a GNU Radio front-end, a message queue, and a Tensorflow-based neural network is explained along with individual algorithms and training of the modulation classifier.