Discussion of article "Creating Non-Lagging Digital Filters" - page 5

[[Deleted]]

Since the filter is nonlinear, the impulse response at different sections is different. Therefore, it will be better to use deconvolution - an operation of inverse convolution on the necessary section, for this purpose you can use ALGLIB library.

And plot the spectrum for the resulting impulse response. An ideal filter should have a vertical line between suppression and transmission.

[Konstantin Gruzdev]

GT788:

Since the filter is nonlinear, the impulse response at different sections is different. Therefore, it will be better to use deconvolution - an operation of inverse convolution on the necessary section, for this purpose you can use ALGLIB library.

And plot the spectrum for the resulting impulse response. An ideal filter should have a vertical line between suppression and transmission.

You can look at it too. I marked it for myself, I will do it, but not now.

[David Raine]

Heisenberg's uncertainty principle!

[Евгений]

So why not use the local approximation method https://chaos.phys.msu.ru/loskutov/PDF/Los_Kotl_Zhur.pdf ?

It allows not only predicting data, but also filtering. And, theoretically, such a filter should have no delay at all!

The idea is very similar. Only here you don't need to switch between different filters for different parts of the time series. Here we search for areas in the history similar to the last counts of the series, and then these found areas are averaged over the ensemble. This is the result of filtering. The main thing is that there should be enough data in the history, otherwise there may be no similar plots.