Blind Wideband Spectrum Sensing using Cluster Analysis

This paper presents a spectrum sensing algorithm for wideband interferers to be used within a cognitive radio. A two staged approach is proposed for the problem of wideband interferer detection. The first step requires that the frequency domain samples received are classified probabilistically amongst a number of possible statistical distributions, i.e. each sample is given a probability rating that reflects the likelihood that a specific sample was obtained using some specified distribution. The second stage of interferer detection uses these probabilities in a hidden Markov model to group highly likely samples into common interferer groups. This means that given a sample is highly likely to be an interferer, the surrounding frequency bins are also very likely to be an interferer since the interferers are wideband. Similarly, frequency bins contiguously surrounding a bin which is classified as noise only are more likely to be noise only bins as well. It is shown through simulation that this technique exhibits Receiver Operating Characteristics which are superior to other methods for wideband spectrum sensing.