新颖的能量函数准则下的主分量分析算法

Aprincipal component analysis algorithm based on a novel energy function

本文通过一个新颖的能量函数把Oja规则与准确的梯度搜索联系起来 ,从而证明了Oja规则可以通过梯度搜索而获得。推导了相应的梯度算法和递归最小二乘算法 ,根据Lyapunov稳定性原理和随机扰动理论分析了算法的全局渐近收敛性能。最后 ,给出了跟踪时变DOA的计算机模拟结果。

The direct connection between the Oja rule and the exact gradient search rule is established with a novel energy function It is shown that the Oja rule can be exaxtly interpreted as a gradient rule of an energy function The gradient based and recursive least squares learning algorithms are derived The global asymptotic convergence is analyzed with the Lyapunov stability theorem and the stochastic disturbance theory Finally,The computer simulation for estimating time varying directions of arrival(DOA)is given using the proposed principal component extraction algorithm