Time Series Analysis of Multiple Primary User Environment Using HMM-Based Spectrum Sensing

In this paper, we propose a method to estimate the communication parameters and the number of Primary Users (PUs) by using Hidden Markov Model (HMM) from time series data such as received power in the environment. Many researchers focus on a measurement based Radio Environment Database (RED) that utilizes the actual received signal power obtained by spectrum sensing as an enabler for an efficient frequency sharing. In addition, by estimating average received power and channel occupancy, etc. from radio observation information obtained by spectrum sensing, it is possible to realize high precision RED construction. Conventional research on parameter estimation using machine learning for a system like wireless LAN assumes that the number of PUs existing in the sensing environment is known. This assumption is not realistic in an actual environment. In this research, we estimate the number of PUs and parameters of each PU by using HMM and the Bayesian Information Criterion (BIC) for spectrum sensing when the number of PUs existing in the sensing environment is unknown. Thereby, the parameters related to the each PU in the sensing environment can be separated and stored in RED. Moreover, when estimating by using HMM and BIC, we improve the estimation result of the number of PUs by using the threshold for the estimation result of the transition probability matrix which is the parameter of HMM. The simulation results confirm that the number of PUs can be estimated with high precision when targeting wireless LAN.