An autoregressive approach for spectrum occupancy modeling and prediction based on synchronous measurements

Inefficient spectrum usage is a crucial issue in wireless communications and methods for dynamic spectrum access are proposed based on spectrum sensing methodology of the cognitive radio systems. Beside the detection and estimation methods, spectrum sensing procedures can also benefit from the modeling and prediction of the wireless spectrum usage. Markovian, regressive and other approaches are introduced for time or frequency domain channel modeling however, the research on the spectrum allocation methods indicates that location information has also an important influence on the spectrum occupancy characterization. In this paper, linear autoregressive prediction approach for binary time series is employed to investigate channel occupancy prediction performance based on spectrum measurements conducted in four different locations synchronously. Through the modeling procedure, dependency in frequency domain is also taken into consideration by modeling the adjacent frequency bands together. The model order is selected based on mean residual magnitudes and Akaike information criterion, mode order parameters are tabulated, and comparative prediction analysis considering the observation time is given for each location. The performance of the proposed linear modeling method is also compared with continuous-time Markov chain modeling in one of the locations.