Applications of Machine Learning in Spectrum Sensing for Cognitive Radios

Spectrum sensing is an essential component in cognitive radios. The machine learning (ML) approach is part of artificial intelligence which develops systems capable of learning and improving from experience. ML algorithms are promising techniques for spectrum sensing as a favored solution to tackle the limitations of conventional spectrum sensing techniques while improving detection performance. The supervised ML algorithms, support vector machine (SVM), k-nearest neighbor (kNN), decision tree (DT), and ensemble are applied to detect the existence of primary users (PUs) in the TV spectrum band. This is accomplished by building classifiers using the collected data for the TV spectrum over different locations in the city of Windsor, Ontario. Then, the dimensionality reduction technique named Principal Component Analysis (PCA) is incorporated to reduce the duration of training and testing of the model, as well as reduce the risk of overfitting. This is achieved by transforming the input data into a lower-dimensional representation, which is known as the principal components. The Ensemble classification-based approach is employed to enhance the classifier predictivity and performance. Furthermore, the performance of the Ensemble classification method is compared with SVM, kNN, and DT classifiers. Simulation results have shown that the highest performance is achieved by combining multiple classifiers, i.e., the Ensemble, therefore, the detection performance has significantly improved. Simulation results have shown the impact of employing PCA on lowering the duration of training while maintaining the performance