摘要
由于缺乏有色信源的时域相关性等先验信息,在设计有色信源卷积混合盲分离算法时难以利用源信号的特性,造成“有色源”的卷积混合盲分离比“白色源”的卷积混合盲分离更具挑战性。通过挖掘信源有色特性对卷积混合多项式矩阵的作用机制,提出了一种新的有色信源卷积混合盲源分离算法。首先将有色信源建模为白色信源激励有限冲击响应(Finite Impulse Response,FIR)滤波器的响应,将有色信源卷积混合模型转化为等效的白色源卷积混合模型,此时等效的卷积混合多项式矩阵即为FIR滤波器和原混合多项式矩阵的乘积;其次使用基于白化的方法获得白色激励源和等效卷积混合多项式矩阵的估计;最后利用最大公因式提取方法从等效卷积混合多项式矩阵中提取出FIR滤波器,进而恢复出原始的有色信源。仿真结果表明,所提算法在不同信噪比下的分离性能均优于现有算法,具有良好的实用性。
Due to the lack of priori information such as temporal correlation of colored sources,it is difficult to utilize the properties of the source signals when designing the blind source separation of convolutive mixture with"colored source",which makes the blind separation of convolutive mixtures with colored sources more challenging than its counterpart with"white sources".Through exploiting the interaction mechanism of source signals colouredness on convolutive MPM(Mixing Polynomial Matrix),this paper proposes a new blind source separation algorithm for convolutive mixture with colored source.First,the colored source signal is modeled as the response of a FIR(Finite Impulse Response)filter excited by a white source signal such that the convolutive mixture model with colored source is transformed into an equivalent convolutive mixture model with white source,where the equivalent MPM is the product of the FIR flter and the original MPM.Then,whitening-based method is used to estimate the white excitation source and the equivalent convolutive MPM.Finally,an approximate greatest common divisor algorithm is applied to extract the FIR filter from the equivalent convolutive MPM,which in turn recovers the original colored source.Simulation results indicate that the separation performance of the proposed algorithm is better than the existing algorithms on all SNR cases,which obviously boosts its applicability in practice.
作者
莫海歌
胡显智
戴旭初
MO Haige;HU Xianzhi;DAI Xuchu(University of Science and Technology of China,Hefei Anhui 230026,China)
出处
《通信技术》
2023年第12期1350-1358,共9页
Communications Technology
基金
国家自然科学基金(61971391)。
关键词
盲信号处理
盲解卷积
最大公因式
有色信源
blind signal processing
blind deconvolution
greatest common divisor
colored source
