Predicting Spectrum Occupancies Using a Non-Stationary Hidden Markov Model

One of the critical challenges for secondary use of licensed spectrum is the accurate modeling of primary users' (PUs') stochastic behavior. However, the conventional hidden Markov models (HMMs) assume stationary state transition probability and fail to adequately describe PUs' dwell time distributions. In this letter, we propose a non-stationary hidden Markov model (NS-HMM), in which the time-varying property of PU behavior is realized. A variant of the Baum-Welch algorithm is developed to estimate the parameters of an NS-HMM. Finally, the performance of the proposed model is evaluated through experiments using real spectrum measurement data. The results show that the NS-HMM outperforms existing HMM-based approaches.