研究了卷积混合信号的频率域盲分离问题,为解决不同频率点分离输出信号的次序与幅度模糊性对信号分离造成的不利影响,利用高阶累积量谱线增强技术确定特殊频率点,提出了一种基于少数线谱频率点的频域盲分离方法,并提出多频率点问盲分离信号强相似的次序调整准则和幅度规范化。该方法既能有效分离独立源信号,又能节省运算时间,有利于实时信号分离恢复。使用语音信号和实录舰船辐射水声信号对算法进行了仿真检验,结果表明该方法分离效果较好。
The problem on blind signals separation with convolutive mixtures in frequency domain is studied. To solve the disadvantage of blind separation due to indeterminacy of the permutation/amplitude for output signals in different frequency bins, we propose a novel approach to blind separation based on line spectrum frequency bins, which are chose through enhancing technique of higher order cumulant based line spectrum. The criterion is proposed for permutation correction and amplitude adjusting to separation signals with different frequency bins. The proposed approach can both effectively separate independent signal and reduce computation. It is advantageous to the separation and recovery of real-time signal. The validity of the proposed approach is demonstrated by computer simulation with speech and ship radiating underwater acoustic signals. Simulation resuits illustrate that the new method is good.