Generalized Statistical Spectrum Occupancy Modelling and its Learning based Predictive Validation

Modeling of spectrum occupancy is important for better channel utilization, accurate spectrum sensing, and enhanced Quality of Service (QoS) to the primary user (PU) in a cognitive radio (CR) system. Existing models are highly dependent on the spatio-temporal variations of the PU activity as the statistical behavior of the PU changes with respect to the location, spectrum band, and the varying load time. In this work, a generalized Gaussian Mixture model (GMM) has been investigated for characterizing the spectrum occupancy of the PU in three spectrally different CR scenarios, viz. VHF/UHF band, GSM band, and ISM band. The goodness of fit performance of GMM is compared with the widely used spectrum occupancy model based on Beta distribution. Further, the robustness of GMM has been validated through learning based prediction via Recurrent Neural Networks (RNN), thereby proposing a hybrid approach of statistical and predictive modeling of spectrum occupancy for enhanced dynamic spectrum access.