信号估计中的核回归方法

核回归方法是比较常用的一种非参数估计方法。讨论了核回归方法在一维信号估计中的理论与应用,实验比较了高斯核函数的平滑参数h及多项式阶数N对估计效果的影响。结果是在相同阶数N下,较小的h使所有的估计点都收敛到观察值,反之则是一个N阶多项式拟合。在相同h下,阶数N越大,误差越小,计算量也较大,但重构效果的提升并不明显。

Kernel Regression in Signal Estimation

The kernel regression method now is the most popular non-parametric estimation method.This paper discusses the theory and application of the kernel regression method in the 1-D signal estimating,and compares the influence how the smoothing parameter h of Gaussian kernel function and the polynomial order N impact the estimation effect through experiments.Result shows that,under same order N,estimators converge to observations as h decreases and on the contrary,they form a polynomial with order N,under same h,higher order N leads to small errors with more computational cost,but effect is not obvious.