Radar signals clustering based on spectrum atom decomposition and kernel method

The recognition and parameters estimation of radar emitter signals are crucial techniques in signal processing of Electronic Intelligence. In this paper, a new method of radar emitter signals recognition and parameters estimation based on time-frequency atom is proposed, which is by introduced into the idea of non-orthogonal decomposition of signals. Firstly, a new dictionary of spectrum atoms is proposed to decompose the signals by using Marching Pursuit based on FFT. Then the atoms characteristics vector can be achieved according to the parameters of decomposed atoms. Finally, the kernelized clustering algorithm is applied to identify radar emitter signals automatically and estimate parameters. Experiment results show that the proposed approach can obtain high accurate recognition rate and accurate parameters estimation result, it is feasible to realize in engineering.