A Performance Comparison Of Spectrum Sensing Exploiting Machine Learning Algorithms

Machine learning is the powerful tool of the artificial intelligence which is popularly implemented in several applications. Spectrum sensing is the important function of a cognitive radio which detects the available channels. Although, energy detection is the simplest technique, its detection performance suffers from several communication environment factors. Since each classifiers of the machine learning has its own advantages/disadvantages and suits for a specific data characteristic. In this paper, we evaluate the performance of energy detection implemented with the machine learning methods including logistic regression, k-nearest neighbor (KNN) and neural network. Then, the simulated performance is compared to the traditional energy detection (CFAR). The simulation results show that the data characteristic of the detected energy under different distances is not linear. Therefore, it is difficult to determine the existence of PU by using linear machine learning model such as logistic regression. On the other hand, KNN presents lower performance than CFAR because the training data is not appropriate. Moreover, KNN suffers from long sensing time. The neural network presents the highest performance since it suits for non-linear distribution of data and it does not consider the pre-determined decision threshold. Moreover, it consumes the shortest sensing time.