扩展的快速独立分量分析算法

An extended algorithm for independent component analysis

针对复值快速独立分量分析算法仅适用于非高斯圆信号,而对常用的非高斯非圆信号不适用的问题,提出了扩展的复值快速独立分量分析算法.该算法通过解除原算法假设条件构造新的代价函数,采用现有的近似复数牛顿迭代方法优化该代价函数,推导出适用范围更广的复数快速独立分量分析算法.该算法同原算法一样都是固定点算法,都有很快的收敛速度,而且该算法不但适用于原算法所适用的非高斯圆信号,对原算法所不适用的非高斯非圆信号也是有效的.理论分析和仿真实验验证了算法的有效性.

The complex-valued independent component analysis algorithm is effective for non-Gaussian circular signals, but is not valid for non-Gaussian non-circular signals. To solve this, the authors proposed an extended complex-valued independent component analysis algorithm. This algorithm constructed its cost function by releasing the assumed condition of the original algorithm, and thus deduced an improved complex-valued independent component analysis with larger application scope by using a Newtonian-like iterative algorithm to optimize the new cost function. Like the original algorithm, the proposed algorithm shares a fast convergence rate. It is not only valid for nonGaussian circular signals which the original algorithm is valid for, but also for non-Gaussian non-circular signals which the original algorithm is not valid for. Theoretical analysis and simulations prove the efficiency and correctness of the algorithm.