A modulation recognizer with ITD-based features

Abstract

This paper proposes a modulation recognizer based on the feature vectors obtained by Intrinsic Time-scale Decomposition(ITD) algorithm and Support Vector Machine(SVM). ITD is employed to extract time-frequency information of communication signals and the obtained feature vectors are transformed into lower-dimensional subspace according to Fisher analysis theory. Multiclass SVM is employed to modulation classification, including 7 types of digital modulation such as 2ASK, 4ASK, 2PSK, 4PSK, 16QAM, 2FSK and 4FSK. The ITD-based features can be directly obtained by means of waveform analysis, which not only has very low computational complexity, but also have good discriminability because of dimension reduce based on Fisher analysis. Simulation results show that the modulation recognizer can provide very high recognition accuracy with very low processing complexity.

Distributed Few-Shot Learning for Intelligent Recognition of Communication Jamming

“…From the perspective of efficient machine learning, it is essential to reduce the computational complexity in feature calculation. Various characteristics of communication jamming in both time-and frequency-domains were analyzed in [10], upon which two classifiers of neural network and decision tree were designed to realize jamming recognition. In [11], a broadband communication jamming recognition method based on graphs and neural networks was proposed.…”