A 3 point order statistic filter is used. That is, if the time series is
... a b c ...
then the value
b is replaced by the following algorithm:
(a) Sort (a,b,c) into (p,q,r), where p <= q <= r.
(b) Output = 0.7*p + 0.15*(q+r)
This algorithm is applied to each point in the input time series to produce a new time series. It is
not applied to points where any of (a,b,c) are being ignored.
Order statistic filters were originally defined to provide an intermediate type of filter between linear filters and median filters. I chose this particular filter back in 1993 (in the old program FD2) mostly because I liked the results visually. As I recall, these coefficients are optiimal for filtering time series which have Laplacian distributed noise.
bob cox