Zero Delay Feedback Filters A very simple test

The delay feedback problem in synthesis software emulation was first pointed out and named by Vadim Zavalishin, of Native Instruments back in 2008, but it was also mentioned some years before, under different names in several other papers.

What really matters is the effect on the sound. This is most apparent in how it alters the relationship between the filter frequency and the filter resonance (or emphasis)

In hardware analog synthesizers, the filter frequency and its resonance are almost independent. When one of them is being modified the other remains almost the same. Adding a delay in digital filter means computing the output of the filter with an input and the previous output creates a link between the cutoff frequency and the emphasis so modifying the frequency also modifies the emphasis.

Here is a simple test that you can make to figure out if you favorite soft synthesizer has a zero-delay-feedback filter. Apply a Mid Emphasis to the filter and sweep the filter through the full range.

With a 0df, the emphasis will keep almost the same value. In a filter design that does not manage the digital delay, the emphasis will increase as well as the frequency increases.

A commonly held belief is that non-0df filters are unable to produce sounds with rapid onset transients, like percussion because of the delay introduced. This is incorrect, the 1 delay lag is not perceived by the ear. The 0df filter problem affects sounds at more or less mid resonance settings, when the filter frequency is varying (when is generally the case).

For those who want to go find out more, and really get your geek on go here for a more scientific description of the 0df filter case.

The Test: You'll need a synthesizer (software of hardware) with a self-oscillating filter and a noise generator.

1. Open the noise generator level and mute the oscillators.

2. Set the filter frequency to a very noticeable value (500 Hz for instance)

3. Set the emphasis to a mid value (not to high, but enough to be noticeable)

4. Sweep the filter frequency toward the highest value.

In Zero Delay Feedback filters, the emphasis will remain the same, but in filters that do not manage this delay, the emphasis will grow until the filter self-oscillates.

Here you can see this test:

http://www.youtube.com/watch?v=AuR5haBJYss

Yes, It's "that" simple !

Final Note: Synths that are not equipped with 0df filters are definitely NOT "bad" synths. They have their own character, filters, etc, and in a lot of situations, you wont be able to tell the difference. The same applies to synths equipped with 0df filters: They COULD sound bad, as this is only ONE aspect of all the overall character of a synth.

At Xils-Lab we're always trying to better our algorithms, and the 0df part is no exception. There are several possibilities and approaches to achieve 0df filters in the digital world, and we hope that our own way to do things will continue to evolve and will allow us to keep on offering beautiful musical instruments in the future.

We hope that this test will be useful and fun for you.

LtZ

Oh really? That test may be an indication, but it is not foolproof - cutoff and resonance can be linked by other means, resulting in a similar response
Last edited by Howard on Thu May 10, 2012 5:23 am; edited 1 time in total

Read the in-depth article, good read. Now ditch the dongles and you might have a sale or two.

(You do realize that almost no synth plug-ins use dongles anymore...)

Why wouldn't we just read the synth specs/marketing blurb instead of doing tests ?
It seems like it's the latest thing for devs to boast about, so they are bound to mention it. If it doesn't say "zero delay feedback filter "in the specs then it mostly likely doesn't have it.
Last edited by Acid Mitch on Wed May 16, 2012 9:47 am; edited 2 times in total

Ohh Cthulhu, KVR is a funny place these days.
I'll prepare my popcorn

Cheers
Dennis

Going by that test, Zebra's XMF filter exhibits zero-delay-feedback behaviour!

(well, maybe not so shocked)

Actually the question becomes, just what filters out there aren't 0df now? I don't htink I have any! Horray, all my synths just got even better!

This is NOT true in general. It is true for certain specific topologies implemented in certain specific ways, but there are tons of examples where this is NOT the case (ie the analog originals are NOT well-behaved in such a sense). Further, many analog designs that have significantly non-linearities can further have dependencies between input signal levels vs. frequency and/or resonance.

Further, even if you took a specific analog topology that does have this independence property, you can still screw it up even with zero-delay design. Likewise you can make a non-zero-delay design appear to have this independence as far as the user is concerned, at least in simple tests.

It is true that when you model a "well-behaved" analog topology (eg OTA 4-pole cascade or SEM-style SVF or whatever) with zero-delay techniques, you will probably end up preserving the "well-behaved" nature of the original. Now take a transistor ladder instead, and the analog-original is unlikely to be very well-behaved to begin with (though you can rather easily model the resulting behavior as well, whether you're using zero-delay or not). There are also analog designs that don't have such independence even in their linear model.

I'm not against zero-delay design; it does generally do a better job in time-varying cases and it even in time-invariant cases it generally saves you from having to do some magic (=fancy math) to fix the independence (or dependence) properties, where as a zero-delay design usually preserves whatever it is that the original does.

Indeed. Happy days!

You probably have tons. They just use some other methods to get the independence properties. It's not like you need to make a naive model even if there's some extra delays in the feedback paths. The whole thread is a joke.

When researching for an adequate filter for my Stargate project, I ported Vadims filter from Reaktor to SM. It didn't sound better than the stock one. The non-linearities are far more important for the character and the quality.

Since a certain competitor has released a certain synth some developers seem to get a bit paranoid. I, and many others, will judge the sound with their ears. Pretty videos and pictures won't help here.

The saddest thing really is that XILS synths are really good! There is really no need for stuff like this.

The *really* sad thing is that the whole thing is a lie though! This is no accurate test. I don't have synth1 installed at the moment but I'm sure this test would show that it has 0-delay-whatever-latestmarketingphrase-0df filters. Which it doesn't and it sounds great nontheless.

Cheers
Dennis

Doesn't make sense. The whole point about 0DFb filters is that they do a better job of reproducing nonlinear (=feedback) behaviour.