Hidden Markov Model Spectrum Predictor for Poisson Distributed Traffic

Static spectrum allocation policies allied with the increasing demand for higher data rates stimulated the pursuit of alternative spectrum allocation strategies. In this context, Opportunistic Spectrum Access (OSA) has been considered an alternative to allow licensed portions of the spectrum to be shared with unlicensed users. OSA requires unlicensed users to identify unused portions of the spectrum for opportunistic access that minimizes possible interference with the licensed users. Accurate mechanisms to avoid interference and improve the spectrum usage is highly desirable. This work investigates the performance of a traditional Hidden Markov Model (HMM) predictor where the licensed traffic follows a Poisson distribution. The results show that, under the evaluated settings, traditional HMM predictor can improve spectrum usage up to 15% at the expense of a high rate (approximate to 50%) of inaccurate forecasts of idle periods. Based on these results, this paper proposes two techniques to optimize prediction performance and reduce inaccurate forecasts rate. Using the proposed enhancements inaccurate forecasts rate was reduced to only 6%. An additional benefit was observed when reducing collision with the licensed user, that is an improvement in the amount of slots effectively used by the unlicensed users from 15% to 23%.