时间独立分量分析模型的新息方法

Innovation Method for Temporal Independent Component Analysis Model

考虑时间独立分量分析(TICA)模型中的时间结构信息,引入新息思想,提出TICA的新息模型和估计方法(ITICA)。研究表明：1)TICA的新息模型仍然满足经典独立分量分析(ICA)的潜在分量统计独立和非高斯假设,TICA模型中的混合矩阵(或分离矩阵)可以从对应的ITICA模型中估计得到；2)随着新息的引入,新模型中的潜在独立分量非高斯性增强,ITICA方法能有效提高估计的效率。实验结果表明,ITICA方法可以改善经典ICA算法的收敛性,解决实际中近似独立源信号估计过程中出现的收敛振荡和速度慢的问题,结合先验知识和选择合适的新息还可以有效提高算法的辨识精度。

For temporal independent component analysis （TICA） model, a novel innovation model and method of TICA （ITICA） is presented when the time-structure information is considered. It shows that the new model also sadsfics the assumptions of statistical independence and nonganssianity for sources, and the mixing（ or demixing）matrix in TICA model can be estimated from the corresponding ITICA model directly. And when the innovation process is employed ,the nonganssianities of latent components are increased ,thus ITICA has a faster convergence rate than classical ICA does. Experimental results demonstrate that the method of ITICA has superior performance of accuracy and convergence rate to that of classical ICA.