Deep Learning Based Identification of Wireless Protocols in the PHY layer

The need for Artificial Intelligence algorithms for future Cognitive Radio (CR) systems is unavoidable. For a CR to operate as best as possible it must identify who is present in spectrum of interest, and what they are doing (jamming, communicating, rogue transmission, etc.). Using this information, a CR can accordingly decide what to do next. Furthermore, being able to determine which wireless protocols are occupying spectrum is an important ability in heterogeneous wireless networks. In this work, we investigate the robustness of various Neural Network (NN) algorithms for classification of wireless protocols when looking at base-band In-phase/Quadrature (IQ) data without needing to decode. We propose a spectrum sensing algorithm based on NNs or other similarly behaved classification algorithms for identifying wireless technologies occupying spectrum. In previous literature, using base-band IQ data, researchers have shown that NN models can classify different modulation formats with promising accuracy. This work explores the potentials, usage, and limitations of using base-band IQ data for classifying various wireless network protocols that employ the same modulation format.