Robust Neural Network-Based Spectrum Occupancy Mapping
Abbas Termos, Bertrand M. Hochwald · 2021
We present a neural network decision system for determining if spectrum is occupied in a region. Given a threshold, we wish to determine if power at a given frequency exceeds the threshold, thus determining if that frequency is “occupied”. The emitting sources are unknown in number, locations, and powers. The sensors, which measure the signal power, are random in number and location. The measurements are aggregated as log-likelihood ratios into a fixed-resolution image suitable as input to a neural network. The network is trained to produce an occupancy map over a wide area, even where there are no sensors, and achieves excellent accuracy at determining occupancy. The system is robust to the number of sensors, and occupancy threshold in a variety of environments.