Sparse Component Analysis Based on Hierarchical Hough Transform

作者：Jin, Yi^*; Qin, Shaoqian; Zhu, Changan

来源：Circuits, Systems, and Signal Processing, 2017, 36(4): 1569-1585.

DOI：10.1007/s00034-016-0374-8

• Sparse component analysis
• Blind source separation
• Hough transform
• Grid resolution

Sparse component analysis (SCA) has been extensively studied to solve undetermined blind source separation problem in various fields over the last decades. This paper proposes a SCA algorithm based on hierarchical Hough transform. The hyperplanes clustering within the mixture space are revealed as local maxima in the parameter space after Hough transform is performed. Then, the local maxima are picked up, and the mixing matrix is calculated. The grid resolution in which the parameter space is divided plays an important role on the estimation error and the computational load. Therefore, the parameter space is divided into hypercubes recursively from low to high resolution, and Hough transform is performed only on the hypercubes with votes exceeding a selected threshold. The grid resolution selection problem is solved, and the computational load is reduced a lot in the meantime. After the mixing matrix is obtained, the sources are recovered with -norm optimization. Numerical simulation and a speech separation application illustrate the superior performance of the proposed algorithm.

• 出版日期2017-4
• 单位中国科学技术大学