Subtle intrapulse feature extraction based on CEEMDAN for radar signals

Effective signal feature extraction builds the foundation of highly accurate radar emitter identification, a key function of the electronic warfare. Conventional features used in practice such as the pulse description words cannot fulfill the task in complex electromagnetic environments. An effective subtle intrapulse radar feature extraction method based on the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) was proposed. Radar signals were reconstructed by components provided by the CEEMDAN decomposition process, which was highly effective for non-stationary and nonlinear signals; the de-noising effect of the reconstruction on radar signals was validated through 1 000 Monte Carlo experiments, and an intrapulse feature space based on the reconstruction was designed. The identification accuracy of the proposed feature space was compared to that of the mainstream methods in the area, on 3 000 noise-contaminated signal samples supposed to be generated by 6 radar emitters, with different intrapulse modulation. Experimental results show that the samples are totally distinguishable in the proposed feature space, and the proposed method is more accurate in the comparison, especially in highly noisy environment, with accuracy above 90% at 0 dB signal to noise ratio (SNR).