Identification of Cellular Signal Measurements Using Machine Learning

Spectrum awareness has a plethora of civilian and defense applications, such as spectrum resource management, adaptive transmissions, interference detection, and identification of threat signals. This article proposes an identification neural network (INN)-based model that identifies cellular signals from three different radio access technologies, namely global system for mobile (GSM) communications, universal mobile telecommunications service, and long-term evolution. The proposed INN identifies whether or not the measured power spectral density belongs to a certain cellular signal type. Two data collection approaches (DCAs) are considered: in-band and multiple-band. The over-the-air measurements for the two DCAs show that with low computational complexity, the proposed INN model provides an identification accuracy between 93% and 100%, with a false alarm (FA) rate between 0% and 10%.