基于原子分解的辐射源信号二次特征提取

Cascade Feature Extraction for Radar Emitter Signals Based on Atomic Decomposition

提出了基于原子分解的辐射源信号二次特征提取方法.在过完备多尺度Chirplet原子库基础上,首先用匹配追踪(MP)方法进行信号时频原子分解,并通过改进的量子遗传算法(IQGA)降低MP搜索过程的时间复杂性,得到表示雷达辐射源信号特征信息的最佳Chirplet原子.在此基础上,降低特征参数的维度,提取最具分类意义的原子特征向量.对5种典型雷达辐射源信号的特征提取实验表明,提取的原子特征类内聚集性强、类间分离度大,证实了本文方法的可行性和有效性.

A cascade feature extraction for radar emitter signals based on atomic decomposition was proposed. Radar emitter signals are decomposed into a linear expansion of atoms by the method of matching pursuit （MP） based on an over-complete dictionary of Gaussian chirplet atoms, and an improved quantum genetic algorithm is applied to reduce the time-complexity of each search step of MP. In this way, optimal chirplet atoms representing features of signals are obtained. Then, the characteristic vectors of atoms are extracted to get strong-discrimination. Experimental results on 5 typical radar emitter signals show that the extracted atoms have good ability to cluster the same radar signals and separate different radar signals, which confirms the validity and feasibility of the proposed approach.