带噪的战场声信号盲分离方法研究

Blind Source Separation of Noisy Acoustics Mixtures of Battlefield Sound-taget

提出了一种噪声环境中战场混合声信号盲分离方法。基于含噪的独立分量分析模型,对观测信号进行准白化,去除噪声引起的协方差偏移量;定义观测信号中随机变量的高斯矩为无偏估计的目标函数,最大化此目标函数得到了一种改进的FastICA算法,应用于带噪的战场混合声信号盲分离。仿真实验证明,改进算法能较好改善分离效果,具有很好的鲁棒性。

A new method of BSS（Blind Source Separation） for acoustic mixtures of the battlefield sound-target in the noisy environment is proposed. Based on the noisy ICA （Independent Component Analysis） model, the observations are quasi-whitened to reduce the bias of covariance result from the noise. And the Gaussian moments of random variable in the mixtures are defined as the objective function for no asymptotic bias estimation. The improved FastICA algorithm is proposed by maximizing this objective function, and it is applied successfully to separate the acoustic mixtures in the military field. The simulation demonstrates that the proposed method is robust and can improve the separation effect obviously.