Multiple Sensing Resources in Cognitive Radio Systems

Abstract

This paper aims to improve the detection accuracy of the energy detection-based cognitive radio system using machine learning while varying the number of cognitive radios used for sensing. It has been an exploratory study in order to identify a suitable number of machine learning-enabled cognitive radios that could be used to improve the performance of the system in terms of sensitivity. Several machine learning techniques have been tested at a low SNR (-25dB) in order to simulate conditions where energy detectors perform poorly. The number of cognitive radios has been varied from three to eighteen in order to explore the potential of increasing the data available for machine learning. Results have revealed that with three cognitive radios, the overall detection accuracy of 97% could be achieved in the energy-detection based system despite the low SNR condition. It has also been discovered that increasing the number of cognitive radios has improved the True positive rate for primary user detection from 98% to >99% with Fine Tree algorithm but at the expense of the overall accuracy of the system. Therefore, demands of specific applications are required in order to identify the number of cognitive radios to employ. Overall, the research presents the potential of using a minimal resource-demanding cognitive radio sensing method (energy detection) for sensing without compromising the detection accuracy of the system. It also improves the spectrum utilization of the system by reducing the probability of false alarms considerably. It highlights improvements that can be made to the conventional energy-detection system through machine learning while retaining the simplicity of the cognitive radio system