Filtering and smoothing algorithms

Question

I have this raw data set :

I want to filter, in order to get the smooth curve (in this example the noise will always decrease the signal, and rarely increase it)

I implemented a kmean based algorithm to filter it and had this:

But still not smooth enough, because in the raw data we can see clearly the overall curve shape, but in filtered data this shape is damaged.

I tried using moving average and Savitzky-Golay filtering but both give results worst than my kmeans based algorithm.

Do you have any ideas about algorithms best suited for this kind of noise and signals. any matlab examples are welcomed.

Thank you

Answer 1

Median filtering is a technique used to remove similar kinds of noise. Since downward noise is much more likely than upward, you could use the 75th percentile (or what have you) instead of the median on each window.