Adaptive Riemannian Manifold Learning for Real-Time Spectrum Sensing in Cognitive Radio Networks

This paper introduces an Adaptive Riemannian Manifold Learning (ARML) framework designed for realtime spectrum sensing in Cognitive Radio Networks (CRNs). Our method essentially exploits the underlying pattern of Symmetric Positive Definite (SPD) manifolds to visually encode the covariance of the received signals thus facilitating a better sensing performance especially when signal to noise ratio (SNR) is very low. To name a few, the fundamental difference of ARML with traditional Euclidean based ways and other standard Riemannian classifiers like R-KNN and R-SVM, is that ARML is able to reconfigure itself in response to drastically varying spectrum scenarios while still keeping in step with the limited computation power. Experimental results reveal how ARML substantially beats those tops, notch techniques in detection accuracy as well as in speed of processing. To give an example, while operating at a signal to noise ratio (SNR) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 0 ~ d B}$</tex>, ARML records a detection rate as high as 72.3 %, way above that of Energy Detection (58.2%), R, KNN (61.5%), R, SVM (64.5%), and SPDNet (65.7 %). Considering its average sensing delay of about 12.4 milliseconds, ARML can be said to meet the realtime demands of dynamic CRN settings, and this makes it a reliable, sophisticated solution for wireless networks of the future. The confusion matrix for primary user (PU) detection assesses how well the classification is done on the test set, and at the same time, ROC curves further demonstrate ARML's excellent detection capability over a range of SNR levels.