Semiparametric approach to Nonstationary Signal Analysis
Y.G Ku, Masashi Kawasumi · 2008
We suggest a semiparametric approach to analyze nonstationary signal. A Gamma probability density and maximum likelihood is employed to estimate the most model order and model coefficients on the assumption that model parameters are distributed on kernel density estimator with two hyper-paraneters. The innovation noise is no more identically distributed in semiparametric method. The simulated results showed two hyper-parameters alpha=0.15 and beta=0.99 are determined for the most suitable model parameters and had an advantage of both parametric and nonparametric method.