【摘要】 高阶收敛的复值FastICA(CFICA)算法是一种有效的独立分量分析算法,具有收敛速度快、形式简单的特点。其对初始值的选择要求较高,如果初始值选择不当,不仅会影响收敛的效果,甚至有可能造成不收敛的结果。针对这一问题,在高阶收敛的CFICA的基础之上,通过引入超松弛因子对随机产生的初始值进行处理,放宽算法对初始值的要求。在保证收敛速度的前提下,获得了能有效克服初值敏感性的高阶CFICA改进算法。仿真实验结果表明:改进后的算法不依赖于初始值的选择,避免了收敛速度不均衡的现象,提高了算法的收敛性能。更多还原

【Abstract】 High order convergence Complex-valued Fast ICA(CFICA) algorithm is an effective independent component analysis algorithm. CFICA has the characteristics of quick convergence and simple form, but the demand of choosing the initial value is higher, it will affect convergence effect and even results in non-convergence if the initial value is not chosen appropriately. In order to solve this problem, successive over relaxation factor was introduced into high order convergence CFICA to process randomly generated initial, and relax the request of choosing the initial value. In the premise of guarantee convergence speed, the high order improved CFICA algorithm was obtained, which could effectively overcome the initial value sensitivity. Simulation result shows that the improved algorithm does not depend on initial value choice, and avoids the uneven phenomenon of convergence speed, and improves the convergence property.更多还原