基于FFT-MCC分析的ICA(BSS)盲不确定性消除

A New Method for Recognition of Independent Noise Sources of Complex System under Strong Noisy Environment

为了消除ICA(BSS)估计的幅值、相位及排序等盲不确定性,提出一种基于快速傅里叶变换与最大相关准则分析的ICA(BSS)估计源自适应校正方法。借助对原始传感观测及估计源的频谱分析,近似获得各本底源信号在观测信号中所占的比重———初始放大权值;基于最大相关准则优化调整ICA(BSS)估计源的相位,并对初始放大权值进行微调,从而消除ICA(BSS)估计的盲不确定性,实现源波形的恢复及其混合参数的估计。仿真试验结果证明了该方法的有效性,也表明它在复杂系统源识别或重建方面具有较大的应用潜力。

In order to eliminate blind uncertainty of ICA estimation on amplitudes, phases and orders, and to use them better, a new adaptive method for revision of ICA estimates was proposed based on combinition of fast Fourier transform （FFT） with maximum correlation criterion （MCC）. By using FFT into both observations and ICA estimates, approximate proportions（i, e. the first weight） of every sources among all observations were solved firstly. Then, optimizations of phase and the first weights were done. Thus, blind uncertainty of ICA estimates was eliminated effectively, and the waveform of sources and their mixing coefficient were restored correctly. The simulation results imply that the new method is effective, and of great potential in source recognition and reconstruction of complex systems.