Improving the Performance of Wireless Signal Recognition Using Dimensionality Reduction–Based Classifiers for Low SNR Signals

Abstract

Future communication networks face the predicament of scarce spectrum resources in order to accommodate the exponential proliferation of heterogeneous wireless devices. The significance of wireless signal recognition (WSR) is steadily growing, especially concerning spectrum monitoring, spectrum management, and secure communications, among other crucial domains. Numerous techniques for conducting WSR were presented in the literature; however, the majority of them demonstrate limited efficacy in low SNR scenarios. This work proposes a dimensionality reduction (DR)–based machine learning classifier (DR-MLC) that simultaneously performs modulation and signal classification in heterogenous waveform scenarios. Additionally, the model employs a projection technique (PT) for noise removal in input data. The proposed framework is the first of its kind in the wireless signal processing domain and its performance is extensively evaluated in low SNR conditions using DR techniques including principal component analysis (PCA), linear discriminant analysis (LDA), and kernel LDA (KLDA) in combination with various machine learning techniques such as support vector machine (SVM), $K$-nearest neighbor (KNN), and nearest mean (NM) methods. The publicly available RadComDynamic dataset is utilized for the experiment. The paper demonstrates the improvement in classification accuracy of the proposed model over the reference architectures taken, specifically in low SNR scenarios for both single and dual classification tasks.