High-Performance Deep Learning Classification for Radio Signals

Ahsen J. Uppal, Michael Hegarty, William Haftel, Phil A. Sallee, H. Brown Cribbs, H. Howie Huang

Abstract: The ability to classify different types of signal modulations in radio transmissions is an important task with applications in defense, networking, and communications. This process has traditionally been done manually by human analysts. Recent advances have shown that applying deep learning methods to this task is feasible. But existing recognition networks are complex, with heavy computational requirements, and poor accuracy on some modulation types and in noisy environmentsWe have built a robust radio frequency signal classifier with a hybrid approach that uses images derived from signal constellation and spectrogram data, combined with an efficient convolutional neural network. Compared to the state-of-the-art deep learning classifier, our system obtains better accuracy, with lower computational requirements.