摘要
对时变的水下目标声信号进行盲分离时,针对自然梯度算法存在不稳定收敛的不足,提出了一种可对水下目标进行辨识的改进自然梯度算法。通过对观测数据的白化处理、构建实信号和复信号盲分离都适用的非线性函数、快速收敛学习因子和基于功率谱方法的盲分离效果评估函数,实现了水下目标声信号盲分离和目标辨识的改进算法。仿真结果和实船信号试验数据的一致性,表明了改进算法具有更好的收敛性能,而且在频域评估时盲分离效果更加简单直观,验证了改进算法的实用性和有效性。
In our opinion, Natural Gradient Algorithm （NGA）, when applied to the blind identification of underwater target, must be improved. We now present our improved NGA （INGA）. Three improvements are necessary for single-target or double-target identification. For double-target identification, a fourth improvement is also needed. In the full paper we explain the four improvements in much detail; in this abstract, we just list the four improvements.. （1）the application of Principal Component Analysis to whiten observed data in order to enhance stability; （2）the construction of a nonlinear function that can separate useful part of signal from noise and is applicable to both signals mathematically expressed in real numbers and those expressed in complex numbers; （3）the construction of a learning factor for faster convergence; （4）the construction of a function based upon power spectrum for evaluating blind separation effect （needed for double-target or multi target identification）. One achievement particularly worth noticing is that evaluating the effect of blind separation in frequency domain is much simpler and more convenient than in time domain. We performed both simulation and on-site experiments. We obtained results of evaluation functions for double-target identification with INGA as[^1 0.0017417 ^0.0026758 1]（simulation） and[ 0.99747^0.11106 0.1707^0.99987]（on-site） respectively. These results indicate that INGA is effective in double-target identification.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2006年第1期5-9,共5页
Journal of Northwestern Polytechnical University
关键词
盲分离
自然梯度算法
目标辨识
blind identification, Natural Gradient Algorithm （NGA）, double-target identification
