Transient-Based Channel Classification with Chirplet Transform and Convolutional Neural Networks

The identification of the channel propagation environment is a valuable function in wireless communication systems because mitigation techniques could be applied to mitigate the negative effects of the environment like fading or attenuation. Recent studies have investigated the use of machine learning and more specifically deep learning for channel classification with considerable success either in the time domain or spectral domain. This paper proposes an analysis of the application of chirplet transforms to the classification of the propagation environment in combination with a Convolutional Neural Network (CNN). A novel aspect of this study is that it exploits a specific component of the transmitted signal (e.g., the transient rampup portion) rather than relying on ad-hoc signals (i.e., channel sounder) or the preamble as commonly done. The results of the approach applied to a public data set of ZigBee devices show that the use of the chirplet transform supports a classification performance with accuracy higher than 95% and outperforms the use of other time frequency transforms like the spectrogram even in presence of noise.

BALDINI Gianmarco; 

2025-11-24

IEEE

JRC142077

https://ieeexplore.ieee.org/document/11104362,    https://publications.jrc.ec.europa.eu/repository/handle/JRC142077,   

10.1109/MeditCom64437.2025.11104362 (online),