Measurement signal preprocessing consists of several techniques for da
ta preanalysis and modification. The removal of erroneous samples is o
ne of these tasks. This paper presents a predictive median threshold f
ilter with robust prediction estimation. The filter can be applied to
the elimination of impulsive disturbances of short duration. The main
demand for this kind of filter is that the elimination of a disturbanc
e must not distort the signal in the time domain outside the erroneous
region. The filter structure and some results of tests illustrating t
he performance of the filter in both the time and the frequency domain
are presented. The results show that the new filter structure is less
sensitive to changing properties of disturbances than the correspondi
ng former filter structure. This means that the values for the filter
parameters are easy to select and that the filter can be applied to su
ppress high-amplitude disturbances with great reliability. (C) 1997 El
sevier Science B.V.