基于变分模态分解的侵彻过载信号盲分离

Blind separation of penetration overload signals based on VMD

侵彻过载信号包含复杂的信号分量,传统的信号处理方法无法有效提取弹体的侵彻过载特征。提出一种将变分模态分解与盲源分离相结合的侵彻过载信号特征分离方法,首先由变分模态分解将源信号分解成一系列本征模态函数;然后将本征模态函数与源信号组成多维观测信号,对其自相关矩阵进行奇异值分解估计源信号数目,并计算各本征模态函数与源信号的相关系数,根据源信号数目和相关系数,选择相应的本征模态函数与源信号重构多通道观测信号;最后采用特征矩阵联合近似对角化法对多通道观测信号进行盲源分离。与传统信号处理方法相比,该方法能够有效分离出侵彻过载信号,积分结果较好地反映了弹体的实际侵彻深度,为引信系统的结构设计提供依据。

Penetration overload signal contains complex signal components, and the traditional signal processing methods can’t effectively extract penetration overload features of a projectile. Here, a feature separation method of penetration overload signal was proposed to combine variational modal decomposition(SVD) with blind source separation. Firstly, a source signal was decomposed into a series of intrinsic mode functions(IMFs) with variational modal decomposition(VMD). Then, IMFs and source signals were used to form a multi-dimensional observation signal, its autocorrelation matrix’s singular value decomposition(SVD) was performed to estimate the number of source signals, and correlation coefficients between IMFs and source signals were calculated, respectively. According to the number of source signals and correlation coefficients, the corresponding IMFs and source signals were selected to reconstruct a multi-channel observation signal. Finally, the characteristic matrix joint approximate diagonalization method was used to do blind source separation of the multi-channel observation signal. It was shown that compared with the traditional signal processing method, the proposed method can effectively splinter the penetration overload signal, and its integration results can better reflect the actual penetration depth of the projectile to provide a basis for structural design of detonator systems.