基于弹性网正则化的广义逆波束形成

广义逆波束形成是一种高效的声源识别方法。然而受限于较低的算法稳健性,使得其难以实现高精度的声源识别定位。为提高广义逆波束形成声源识别性能,结合弹性网正则化方法和广义逆理论提出一种基于弹性网正则化的广义逆波束形成。首先从广义逆理论出发介绍了特征向量求解以及阈值截断滤波过程,并结合弹性网正则化思想全面阐述了基于弹性网正则化的广义逆波束形成基本理论;其次建立了数值仿真模型,以单极子和多声源识为研究对象,对比其他波束形成算法详细分析了声源类型与频率等因素对其声源成像性能的影响。最后以单极子、不相干以及相干声源为研究对象进行实验分析,结果表明由于基于弹性网正则化的广义逆波束形成的波束输出解具有较强的稀疏性和稳健性,使得其相比传统广义逆波束形成,能更精准地识别定位声源。

Generalized inverse beamforming via elastic net regularizatiom

Generalized inverse beamforming(GIB) is a kind of high effective and widely used sound source identification technology. However, due to its weak algorithm robustness, it is hard to locate the sound source with high accuracy. To improve the performance of GIB, a new source detection algorithm is developed based on GIB, which is called generalized inverse beamforming via elastic net regularization. Firstly, the?solution?procedure?for eigenmode and the threshold de noising method is introduced. Then the basic principle of?the generalized inverse beamforming via elastic net regularization is extended in detail based on the elastic net regularization. Through numerical simulations on single source, coherent and incoherent sources, a comparison between the proposed new method and the other beamforming algorithm is carried to check their qualities of sound source images affected by factors such as source type and frequency. Finally, the relevant experiment on single source, coherent and incoherent sources is implemented. The experimental result shows that the elastic net regularization based GIB realize an output with stronger sparsity and robustness, which can locate sound source more accurately compared with other conventional generalized inverse beamforming.