Hidden Markov Model Based Spectrum Sensing for Cognitive Radio

Cognitive radio is an emerging technology for sensing and opportunistic spectrum access in wireless communication networks. It allows a secondary user to detect under-utilized spectrum of a primary user and to dynamically access the spectrum without causing harmful interference to the primary user. A number of spectrum sensing techniques has been proposed in the literature to identify the state of the primary user in the temporal domain. However, most of these techniques make instantaneous decisions based on current measurement received at the cognitive radio, and they do not consider the transmission pattern of the primary user which can be acquired from past measurements. Thus, sensing performance can be improved by incorporating measurement history into the sensing decision. Moreover, using all available data may enable prediction of the primary user activity, which will allow a cognitive radio to better plan for its spectrum usage. In this thesis, we utilize both current and past data to improve temporal spectrum sensing performance for cognitive radio. We focus on sensing of a narrowband channel, and assume that the primary user transmission pattern alternates between idle and active states.;We assume that time is slotted into sensing intervals. We formulate the temporal spectrum sensing problem in the framework of hidden Markov models (HMMs), in which the primary transmission pattern is modeled by a Markov chain and the signal power levels received at the cognitive radio is presented by a state dependent Gaussian process. Furthermore, we develop a new statistical model, namely hidden bivariate Markov chain model (HBMM), and apply it to spectrum sensing for cognitive radio. The main advantage of using an HBMM, compared to a standard HMM, is that it allows a non-geometric distribution of the dwell time of the primary user in each state. This distribution, called phase-type, is far more general than the geometric dwell time distribution of a standard HMM. We develop an expectation-maximization (EM) algorithm, which extends the Baum re-estimation algorithms for HMMs, for maximum likelihood estimation of the parameter of the HBMM. We also develop an online recursion for estimation and prediction of the state of the cognitive radio channel. We analyze the performance of our proposed approach using both real spectrum occupancy data and simulated data derived from the spectrum measurements. Our numerical results show that the proposed spectrum sensing approach, which uses the new HBMM outperforms earlier approaches which rely on the standard HMM or a simple energy detector. Performance enhancement is especially noted in scenarios with high path loss and strong shadowing eects, which are characterized by low signal to noise ratio.