Application of Wavelet Scattering Network and Ensemble Learning on Deception Jamming Recognition for Ultra-Wideband Detectors

Ultra-wideband (UWB) radio technique, due to its potential applications and great detection ability, is vulnerable to electromagnetic interference that cannot be ignored. With the improvement and maturity of digital radio frequency memory (DRFM) technology, a new type of active deception jamming, which has similar features in the time or frequency domain with transmitted signals, poses a greater risk to the UWB detectors compared with the conventional suppressive jamming. Therefore, antijamming has become a critical study, and how to recognize jamming signals becomes one of the vital steps. In this article, the geometric models of three ground vehicles are established first to obtain the echoes with multiple scattering centers model under the UWB system. Then, the wavelet scattering network (WSN) is used to extract the differentiable features of target echo and deception jamming. Multiple base classifiers are trained by the wavelet scattering features at different scales, and ensemble learning is adopted to improve the classification rate under the minor sample conditions. The WSN has better interpretability than deep neural networks and can effectively handle the environmental white noise. The proposed method proved efficient based on the results obtained from both measured and simulated data despite the low signal-to-noise ratio (SNR).