@article {10.3844/jcssp.2019.78.91,
article_type = {journal},
title = {Direct Sequence and Frequency Hopping Signals Classification Based on Co-Occurrence Matrix and Clustering Techniques},
author = {Fouad, Haidy S. and Elsayed, Hend A. and Girgis, Shawkat K.},
volume = {15},
number = {1},
year = {2019},
month = {Jan},
pages = {78-91},
doi = {10.3844/jcssp.2019.78.91},
url = {https://thescipub.com/abstract/jcssp.2019.78.91},
abstract = {Spread Spectrum techniques (SS) were first developed for military applications, but currently, they have commercial applications. SS provides secure communication and allows multiple accesses for same radio spectrum. So, most Wireless Local Area Network (WLAN) systems use it, as do Cognitive Radios (CR), space systems and Global Positioning Systems (GPS). Direct Sequence Spread Spectrum (DSSS) and frequency hopped spread spectrum (FHSS) are the two most-used techniques today. Nowadays radio spectrum has become very crowded and so now there is a need for spectrum efficiency. Automatic SS classification presents a rather difficult problem, especially if the parameters, such as the signal power, carrier frequency, etc., are unknown. This research takes a new direction; it deploys the Gray Level Co-occurrence Matrix (GLCM) to capture statistical features of SS signals. Using GLCM, 22 features are extracted for each vector of signal. Analyzing the signals is done in the time domain which measures the variation of amplitude of signals with time. Therefore, the main contribution is to apply and show how GLCM improves the identification accuracy of the two signals in presence of noise. The proposed model achieves considerably accurate results even with a low SNR. GLCM features help classifiers to achieve average accuracy 84% and reach 100% signal identification at a zero SNR. To prove the superiority of these features, a variety of clustering methods are applied, such as centroid, connectivity, model-based and message-passing models. Clustering performance results based on GLCM features are compared With Principal Components Analysis (PCA), Kernel-based Principal Components Analysis (KPCA) and fast Independent Components Analysis (Fast-ICA). Clustering results are evaluated with external and internal validity indices. The accuracy was tested over 26 levels of Signal-to-Noise Ratios (SNR).},
journal = {Journal of Computer Science},
publisher = {Science Publications}
}