ROBUST SIGNAL DETECTION USING CORRENTROPY

摘要

A method (200) for detecting a periodic signal (141) in a noisy signal (101) is provided. The method can include applying (210) correntropy to the noisy signal to generate a non-linear mapping, and applying (220) a sub-space projection to the non-linear mapping to produce principal components. A correntropy kernel can be applied to the noisy signal to generate a Gram matrix that is used in a Temporal Principal Component Analysis (TPCA). The correntropy kernel projects nonlinearly the input data to a reproducing kernel Hilbert Space (RKHS) preserving the input time structure and attenuating impulsive noise. The correntropy kernel is data dependent, and the RKHS correlation matrix has the same dimension as the input data correlation matrix. A principal component having a majority of signal energy can be chosen (230) to detect the periodic signal.