本文選題：盲源分離　切入點：獨立成分分析　出處：《信號處理》2014年12期 　論文類型：期刊論文

【摘要】：在盲源分離和獨立成分分析中,峭度是衡量隨機信號非高斯性的常用對比準則,通過不同類型的算法對其進行優(yōu)化,找到非高斯性極大值點,即實現(xiàn)了源信號的提取或分離。例如,基于峭度的快速不動點算法,它是一種收斂速度很快的算法。最近,Marc Castella等人提出了一類基于所謂"參考信號"的對比準則,以及對應的梯度最大化優(yōu)化算法,這些算法具有很好的收斂性能。受其啟發(fā),文章以一種類似的方式將"參考信號"思想應用到峭度中,得到一種新穎的對比函數(shù),并基于該新峭度對比函數(shù),提出了一種新的快速不動點算法。與經典的基于峭度的快速不動點算法相比,該算法極大地提高了收斂速度,尤其是隨著信號樣值點數(shù)的增加,該算法的優(yōu)勢會更加明顯。文章分析和證明了該新峭度對比函數(shù)的局部收斂性,給出了新算法的詳細推導過程,仿真實驗驗證了該算法的性能,并與經典算法進行了比較分析。
[Abstract]:In blind source separation and independent component analysis (ICA), kurtosis is a common contrast criterion to measure the non-#china_person0# character of random signals. For example, the fast fixed point algorithm based on kurtosis is a fast convergence algorithm. Recently, Marc Castella et al proposed a kind of comparison criterion based on so-called "reference signal". These algorithms have good convergence performance. Inspired by these algorithms, this paper applies the idea of "reference signal" to kurtosis in a similar way, and obtains a novel contrast function. Based on the new kurtosis contrast function, a new fast fixed point algorithm is proposed. Compared with the classical fast fixed point algorithm based on kurtosis, this algorithm greatly improves the convergence rate, especially with the increase of signal sample points. This paper analyzes and proves the local convergence of the new kurtosis contrast function, gives the detailed derivation process of the new algorithm, and verifies the performance of the algorithm by simulation, and compares it with the classical algorithm.
【作者單位】： 解放軍理工大學通信工程學院;國防信息學院;
【基金】：國家自然科學基金資助項目(61172061) 江蘇省自然科學基金資助項目(BK2011117)
【分類號】：TN911.7
，

本文編號：1648779