摘要
在盲源提取中,当所要提取信号的峭度在某一区间时,可以采用基于峭度的方法将期望的信号提取出来。如果采用外点惩罚函数法来求解,理论上要求惩罚因子趋于无穷大时才可能收敛到最优解,但是惩罚因子的增大往往导致代价函数的Hessian矩阵病态化。因此,这种方法在实际中稳健性很差。提出了采用Lagrange乘子法来解决特定信号的提取问题,与采用外点惩罚函数法的算法相比,这种方法在惩罚因子相对较小的情况下也能得到最优解。计算机仿真和实际的胎儿心电试验表明了这种方法在收敛速度和稳健性上要优于采用外点惩罚函数法的算法。
In blind source extraction, when the kurtosis of desired signal lies in a specific range, the desired signal can be extracted based on the kurtosis of the desired signal. If outlier penalty function is used to solve the extraction of desired signal, and this method requires the penalty factor should be infinite for guaranteeing converge to optimization solve. And in this situation, the Hessian matrix of cost function will be ill-conditioned result in poor robust in practice. To solve the unstable of numerical value, Lagrange multiplier was used for the extraction of specific signal. Compared with the algorithm used the outlier penalty function, this algorithm can be converged although when penalty factor is relatively very small. Computer simulation and experiments on real FECG data confirm that the proposed algorithm is more robust and has better converge rate than that of the algorithm using the outlier penalty function
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2008年第22期6097-6099,6102,共4页
Journal of System Simulation
基金
河北省自然科学基金（E2008001257）
关键词
盲源提取
盲源分离
LAGRANGE乘子法
峭度
Blind Source Extraction,Blind Source Separation,Lagrange Multiplier Method,kurtosis
