Improving energy detection in cognitive radio systems using machine learning

Research has shown that a huge portion of the electromagnetic spectrum is underutilized. Over the years, cognitive radio has been demonstrated as an efficient dynamic spectrum management technique. Energy detection is one of the widely used spectrum sensing techniques. However, its performance is limited by factors such as multipath fading and shadowing, which makes it prone to errors, particularly in low signal-to-noise ratio conditions. Yet, it still has a low computational cost, which reduces communication overhead. This paper aims to improve the detection accuracy of the energy detector through the use of machine learning (ML) techniques. In this research, ML models were trained using the energy characteristics of the primary user and other users present within the system. Weighted KNN produced the highest overall accuracy with an average of 91.88% accuracy at various SNR conditions. However, complex tree algorithm gave the most accurate detection (99% accuracy) of the primary user across all the channel conditions tested. This detection also helped to differentiate between the identity of the primary or secondary user from interference.