Fast Machine Learning-Based Signal Classification in Energy Constrained CRN: FPGA Design and Implementation

Cognitive Radio Networks (CRNs) is positioned as an appealing autonomous system to enhance spectrum scarcity by dynamic spectrum access and spectrum sharing across wireless networks. To operate at the highest performance level, the allocation and vacation process of primary and secondary users need to be accomplished rapidly. This issue motivates us to propose a fast machine learning-based processing algorithm, referred to as the Arithmetic Shifter-Based Support Vector Machine (ASB-SVM) classifier. The novelty of our proposed scheme is to increase the speed of signal classification by employing shift registers in a two multipliers feature mapping method instead of using multiplication blocks in the SVM classifier. The proposed ASB-SVM design is implemented in Xilinx Virtex-6 XC6VLX240T FPGA. By exploiting spectral features for the classifier, an overall accuracy rate of 98:2% is achieved for green modulated signals in CRNs. Experimental results show that given the feature vector, our proposed system is capable of classifying a blind modulated signal within just 3 ns in the classifier block of a CRN while achieving 30% resource reduction and 45% increase in speed compared to the conventional linear SVM implementation.