A novel modulation classification method in cognitive radios based on features clustering of time-frequency
Xu Zhu, Takeo Fujii · 2016
The demand of spectrum sensing can be met by a reliable modulation classification (MC) scheme. This paper proposed a novel features clustering algorithm based on the joint distribution of time and frequency. It uses Pseudo Wigner-Ville Distribution (PWVD) as feature extraction approach. Density-Based Spatial Clustering of Applications with Noise (DBSCAN), alternatively, is utilized as classifier for single carrier modulation classification. Unlike ALRT approach, which is conventional method of likelihood based, this novel one is free from carrier phase offset. In addition, it has no variance in features, which is a huge advantage over Cumulant-based approach. The latter suffers from its variance in features, which degrades its performance in complex scheme badly. Moreover, training, which is an incredible time-consuming step for Support Vector Machine (SVM), is not necessary for DBSCAN. This enables DB-SCAN a faster processing. Simulation results indicate an overwhelming advantage over Cumulant-based classifier in performance. Moreover, carrier phase offset does not influent its performance at all.