Washout with varying time constant

Hi,

I'm trying to implement a hybridisation of two signals added toghether representing fast and slow dynamics. The idea is to use a high-pass filter with a varying time constant that depends on the amount of high frequency content of the signal.

My problem is that the hybrid signal gets worse (higher amplitude) when using a varying time constant than when using a fixed one, even though the varying time constant is larger.

In the image below you can see the hybrid signal in the first scope for a time const of 5s (cyan) and 7s (magenta) as well as the case of a varying time const (yellow). In the second scope is the variation of the time constant shown.

As you can see, the signal with the varying time constant goes wild as soon as the time constant starts to change. Furthermore, the signal behaves "as expected" when the time constant is rather constant.

Some more info: -The system is solved using a fixed step discrete solver with a step of 0.04s. -The feedforward of the time constant is implemented with a fuzzy logic controller. -The filter is a discrete "homebrew" with the scheme shown below.

Is anyone familliar with this kind of problem when using a varying time constant?

Thanks in advance!