Anomaly Detection-based Spectrum Sensing using the k-Nearest Neighbors Algorithm

The efficient utilization of the scarce spectrum is essential to satisfy the requirements of future 6G mobile networks. Spectrum sensing allows secondary users to detect unused spectrum portions as a first stage to enable dynamic spectrum access. Artificial intelligence algorithms have been applied to solve the spectrum use (occupied or vacant) as a classification problem. However, they required vast information set for training. In this paper, spectrum sensing is addressed as an anomaly detection problem. The normal behavior is defined as the inactivity of the primary (licensed) user in a specific frequency band. Thus, an anomaly is the presence of the primary users’ signals. The k-nearest neighbors algorithm is implemented to detect the anomalies in the frequency band. The obtained results show an improvement in detection performance compared to the conventional energy-based spectrum sensing technique.