Cooperative Spectrum Sensing based on Anomaly Detection and K Nearest Neighbors

Abstract

Spectrum sensing (SS) is an enabling task for efficient spectrum utilization, which is required by future 6G mobile networks. SS allows cognitive users to dynamically identify the status (present or absent) of primary users (PU) to access the underutilized frequency bands. The use of artificial intelligence algorithms has been proposed to solve the SS problem by treating it as a classification problem. However, this implies the need for vast information for training, which might not be available for all types of PU signals. This paper addresses SS as an anomaly detection (AD) problem, where the “normal” behavior is defined as the absence of the PU. Thus, the presence of the PU is considered an anomaly. The K-nearest neighbors algorithm is implemented after a pre-processing signal stage based on a Decomposition and Recombination algorithm to improve the performance in a cooperative SS (CSS) scenario. Results exhibit the gain in detection performance compared to the conventional energy detector.