A Machine Learning based Spectrum-Sensing Algorithm Using Sample Covariance Matrix

In this paper, we propose a machine learning based spectrum sensing method using the sample covariance matrix of the received signal vector from multiple antennas. Before sensing, the cognitive radio (CR) will first apply the unsupervised learning algorithm (e.g., K-means Clustering) to discover primary user's (PU) transmission patterns. Then, the supervised learning algorithm (e.g., Support Vector Machine) is used to train CR to distinguish PU's status. These two learning phases are implemented using the feature vector that is formed by two parameters of the sample covariance matrix. One parameter is the ratio between the maximum eigenvalue and the minimum eigenvalue; the other is the ratio between the absolute sum of all matrix elements and absolute sum of the diagonal elements. The proposed method does not need any information about the signal, channel, and the noise power a priori. Simulations clearly demonstrate the effectiveness of the proposed method.