摘要 |
PROBLEM TO BE SOLVED: To allow for extraction of a weak signal buried in noise by means of a nonlinear filter, even if the probability density function of noise is unknown.SOLUTION: By kernel density estimation using Epanechnikov kernel, an optimum filter function Ffor maximizing the SN ratio of the output from a nonlinear filter is estimated as follows. At first, a noise sample preparation unit 12 prepares a sample of noise n. A band width calculation unit 13 then estimates an optimum band width h. Subsequently, a noise average calculation unit 11 calculates the average of noise samples existing in the region extending from (x-h+ε) to (x+h-ε) out of the noise sample (a range of ±(h-ε) with xas a center, where xis an input signal to the nonlinear filter) as noise average μ. Thereafter, the optimum filter function Fis calculated using the optimum band width hthus estimated and the noise average μ.SELECTED DRAWING: Figure 1 |