I'm working on a 1D time series data problem where I need a filter to be robust to a very wide range of types of signal noise - basically any type of noise where the mean is approximately zero and the magnitude is not too large. We're talking white noise, gaussian noise, funky distributions, random pulses and stutters, multiple frequencies, periodic, and anything else that you can imagine.
I'm currently testing this by trying to synthesise different types of noise by just mixing various permutations of standard distributions. Is there a dataset of different real world noises seen in a wide range of real world signals that I could sample from instead?
A dataset built for this purpose would be ideal, but any dataset containing a large range of time series signals with many different characteristics could be adapted to this purpose.
This feels to me like a relatively common requirement, but I haven't yet found the standard approach to solving it.