Time and space complexity reduction of KFDA-based LTE modulation classification

Abstract

Kernel Fisher discriminant analysis (KFDA) is a nonlinear discrimination technique for improving automatic modulation classification (AMC) accuracy. Our study showed that the higher-order cumulants (HOCs) of the Long-term evolution (LTE) modulation types are nonlinearly separable, so the KFDA technique is a good solution for its modulation classification problem. Still, research papers showed that the KFDA suffers from high time and space computational complexity. Some studies concentrated on reducing the KFDA time complexity while preserving the AMC performance accuracy by finding faster calculation techniques, but unfortunately, they couldn't reduce the space complexity. This study aims to reduce the time and space computational complexity of the KFDA algorithm while preserving the AMC performance accuracy. Two new time and space complexity reduction algorithms have been proposed. The first algorithm is the most discriminative dataset points (MDDP) algorithm, while the second is the k-nearest neighbors-based clustering (KNN-C) algorithm. The simulation results show that these algorithms could reduce the time and space complexities, but their complexity reduction is a function of signal-to-noise ratio (SNR) values. On the other hand, the KNN-C-based KFDA algorithm has less complexity than the MDDP-based KFDA algorithm. The time and space computation complexity of the KFDA could be effectively reduced using MDDP and KNN-C algorithms; as a result, its calculation became much faster and had less storage size.