基于时间结构的盲信号提取不动点算法

Fixed- Point Algorithm for Blind Signal Extraction Based on Time Structure

为了降低具有时间结构的拟牛顿盲信号提取计算复杂度。通过每次更新后对权向量的归一化处理,利用梯度下降算法在收敛点处的性质,根据经典的Kuhn-Tucker条件提出了一种时间结构的盲信号提取不动点算法,避免了增广拉格朗日函数二阶导数的计算,并简化了一阶导数表达式,同时避免了人为选取步长参数。针对合成数据和实际的心电图数据的计算机仿真表明,提出算法具有良好的提取期望源信号的性能和更快的收敛速度。

To reduce the computional complexity of the Newton-like algorithm for blind signal extraction based on time structure, a fixed-point algorithm was presented through normalizing the weight vector and utilizing the property of the convergence point for gradient descent algorithm with classical Kuhn-Tucker condition. The proposed algorithm can avoid computing the second derivative of the augmented Lagrangian function and simplify the first derivative. The step parameter can also be avoied. Computer simulations with synthetic signals and real electrocardiogram data dem- onstrate good separation performance and better convergence of the proposed algorithm.