直接数据域局域联合混响抑制方法研究

A reverberation suppressing method by direct data domain based localized domain joint algorithm

针对非平稳条件下混响抑制问题,提出一种可利用待测数据统计信息的直接数据域局域联合(D^3JDL)空时自适应方法。不同于级联方法,该方法没有独立运行直接数据域空时自适应和局域联合方法,仅在待测空时样本内用直接数据域空时自适应原理构造非期望数据矩阵,用于获取待测样本的统计信息,并作为局域联合方法的输入进行降维空时处理。由于D^3JDL方法不存在待测样本与学习样本之间统计特性失配的问题,且较好地利用了统计信息,因此具备良好的非平稳环境适应能力。仿真和实验数据处理结果表明,D^3JDL方法抗混响效果优于常规波束形成加匹配滤波处理和其它空时自适应处理方法。

A direct-data-domain based joint domain localized（D^3JDL） space-time adaptive processing method is proposed which can use statistical information of testing samples aiming at suppressing reverberation under non-stationary background.Different from cascade algorithms,the proposed method does not run direct data domain space-time adaptive processing and joint domain localized algorithm separately.The D^3JDL uses the unwanted data matrix constructed by the direct data domain space-time adaptive processing within the space-time testing samples to obtain statistical information.The unwanted data matrix is used as the input data of joint domain localized method.For having no the statistical characteristic mismatching problem between the testing samples and training samples,and utilizing statistical information,the D^3JDL method possesses inherent adaptive ability for non-stationary background.The processing results from simulation data and trial data show that the D^3JDL has better effect on reverberation suppression than conventional beamformer plus matched filter method and other space-time adaptive processing methods.