Principle component analysis based cooperative spectrum sensing in cognitive radio

This paper introduces an improved cooperative spectrum sensing (CSS) algorithm for cognitive radio (CR) networks based on the machine learning technique. In this scheme, spectrum sensing consists of two steps. The offline training is performed by K-means clustering method and the threshold is defined by the classification result towards unsupervised data. As for the online classification stage, the similarity between the received signal and the cluster in training data is exploited for channel availability decision. The features of both the received signal and the training samples are extracted by principle component analysis (PCA). The fusion center will make final decision according to the local sensing result from differently sensors sequentially. Each sensor is responsible for the analysis towards certain period of the received signal. In this way, the initial sensing result (which will be updated due to the latest input signal) is available within short period of time. If the primary signal existed, the initial sensing result can stop further detection which improves the efficiency of spectrum sensing and avoid more interference to primary users. Compare to the existed methods of spectrum sensing, the threshold of the proposed scheme is measured by feature similarity in multi-dimension rather than a fixed value. In simulation, we evaluate the performance by making comparison between different method and the result reveals that our scheme can achieve better performance and work efficiently.