You are probably already using Zero Delay Feedback filters, so let your customers know!

[The_Hidden_Goose]

I've got no DSP knowledge, but I personally think the filters in the better quality modes on Diva are sublime.

With my own tinkerings with synthedit, I've also noticed a significant difference in quality between previous filters (including all the biquad & svf types available) and the ones implemented by xoxes & TD - and I find the so-called ZDF ones to be audibly superior to my ears.

I've never played with The Drop so can't comment on that, although I'd give TheoM's experience the benefit of the doubt on that and assume that it too is a rather good'un.

So in my experience, filters which have been marketed as zdf *thus far* seem to be very good. So as a potential customer, I'll use it as an indicator. It's not going to be an absolute selling point unless the actual filter turns out to be sonically satisfying to me, and that's something that I can only get with playing with the thing and twiddling knobs like a madman. So I'm fine with people using it as a marketing term if they're excited about the sound of it and it conforms to some dsp/mathematical definition to their own satisfaction. It still needs to sound good to get my dosh.

So unless people are buying products entirely based on the blurb and not the sound, then this is a bit of a non-issue to me.

[Lotuzia]

For those who want to find less semantic and more information about 0df :

http://www.xils-lab.com/pages/How-to-te ... ilter.html

http://www.xils-lab.com/pages/Zero-Dela ... lters.html

It certainly doesnt explain it all, and some things could have been added, but at least it will eventually give the *brave* reader a 1st sight of 0df filters.

Nota : Xils-Lab synthesizers were 0df from the start. Now 3rd generation.

[Urs]

Well, I'm very sure that The Drop uses zero delay feedback filters. If I understood correctly then he does "implicit integration" to generate discrete models of analogue circuits. These by definition have delayless feedback paths, or so I was told.

What this is all about is this, in my own interpretation:

- Andy makes great sounding filters, but took long time to do so
- the marketing terminology he helped promote during development is blurred by now
- Andy puts "implicite integration" as new term in place
- Andy slams existing terminology by blurring it even more

#---

We would have loved to find a better way to describe our filters - for the very reason that some average sounding filters were taughted as "zero delay this or that" after Diva's success. Sure, if Andy can place "implicite integration" as new hype terminlogy, we'll happily use that, because to my best knowledge this is what we've done.

That said, while "implicite integration" is probably a more precise term for the purpose of describing the transformation of an analogue circuit into code, I think it lacks catchiness. Can't think of it as a hype-able terminology.

On the other hand, we've learned to just smirk it off when some random competitor uses a term that we coin in a way that we didn't intend. I guess it's a compliment rather than a threat, especially in this very case.

[whyterabbyt]

We would have loved to find a better way to describe our filters

Unfortunately 'Z-1-Delay-less-Feedback Filter' is clumsy to write and pronounce.

[neotec]

Just to write something in this thread here.

For me, it doesn't really matter what those filters are called. More important is the fact that you have the same topology in your digital filters as in the analogue counterparts.

My standard library now has one-pole HPF/LPF, 12/18/24dB SVFs and configurable ladders. All of those don't have any unit delays and are just perfect. The tuning is on spot and they do self-oscillate when they should.

Never had the same experience with any DF biquads. They nearly always have poor resonance behaviour and just don't sound right.

It's also pretty easy to incorporate non linearities.

Another thing that's super easy using this topology preserving stuff is e.g. inserting various filters in the feedback path without sacrificing quality or tuning.

I'm working on new plugins and one of those is a 808/909 emulation. The 808 uses the same building blocks as the real one (speak: resonant 2-poles) and it just works perfect when using the 12dB SVF without unit delays.

If anyone's interested, here's a sound demo (Drums-Emu, Bassline and Overdrive are mine, additional FX come from Renoise):

https://soundcloud.com/neetbit/vst-plugins-prototyping

So, the name doesn't matter. Telling ppl which technique your filters are based on is totally unnecessary. Just build stuff that sounds good, then it will do the marketing on its own.

[andy-cytomic]

This term is fuzzy enough to include almost all digital filters. Any filters that don't fit it are able to be called "topology preserving" so that pretty much covers everyone all around. I am not trying to stop people using the terminology as that is clearly not going to happen, instead I'm encouraging everyone to use it all the time, and hence nullify any possibly distinctions that can be made between products that use such terminology in their descriptions.

That's ridiculous.

As analogue filters don't have unit delays in the feedback path, the term is clearly used to describe a property of a digital filter algorithm. Any claim of software to feature zero delay feedback filters when in fact placing unit delays in feedback paths is a fraud. That naturally includes standard biquad filters, as opposed to your semantic stunt in the first post.

Again Urs you are missing the point. A DF1 biquad from the RBJ cookbook is a zero delay feedback filter. It uses trapezoidal integrators (the bi-linear transform) to discretize the continuous equations (eg from the active state variable filter) without adding delays, and it does this via the Linear Time Invariant laplace transform, which is completely valid way to model the system. It uses implicit integration which solves things without the need for adding unit delays to feedback paths. Go and go that math to work it out yourself and see. Better yet, get Clements to do the math and get him to ask me about any clarifications he needs.

[Urs]

This term is fuzzy enough to include almost all digital filters. Any filters that don't fit it are able to be called "topology preserving" so that pretty much covers everyone all around. I am not trying to stop people using the terminology as that is clearly not going to happen, instead I'm encouraging everyone to use it all the time, and hence nullify any possibly distinctions that can be made between products that use such terminology in their descriptions.

That's ridiculous.

As analogue filters don't have unit delays in the feedback path, the term is clearly used to describe a property of a digital filter algorithm. Any claim of software to feature zero delay feedback filters when in fact placing unit delays in feedback paths is a fraud. That naturally includes standard biquad filters, as opposed to your semantic stunt in the first post.

Again Urs you are missing the point. A DF1 biquad from the RBJ cookbook is a zero delay feedback filter. It uses trapezoidal integrators (the bi-linear transform) to discretize the continuous equations (eg from the active state variable filter) without adding delays, and it does this via the Linear Time Invariant laplace transform, which is completely valid way to model the system. It uses implicit integration which solves things without the need for adding unit delays to feedback paths. Go and go that math to work it out yourself and see. Better yet, get Clements to do the math and get him to ask me about any clarifications he needs.

You're missing the point too. There's no need to do any maths. Any 2nd order biquad in any form has up to two unit delays in its feedback path(s) - no matter what analogue response it perfectly models. Claiming otherwise is purely ridiculous.

[Urs]

Let me expand: Biquads can be great unless you modulate any coefficients. If you do latter, the unit delay(s) may have a negative effect. If you don't then they're perfect. This is why we can use trapezoidal integrators to represent a filter state.

If we're talking "states" then we might as well say: Adding a unit delay to the feedback path in a model of an analogue circuit is like adding an extra state into the feedback path.

So here it is. Let's call them "filters with a stateless feedback path". Then we're good, because that's neither biquads nor analogue-style prototypes with a unit delay.

[aciddose]

http://en.wikipedia.org/wiki/Standing_o ... _of_giants

I don't have much problem with whatever claims people want to make about what they invented or how great it is. That is easy to call bullshit on, and move on. If it isn't useful or novel to you such as any form of 12db tpt state-variable is to me a completely useless senseless and wasteful expense with no practical benefit what-so-ever (as I want to apply it in particular cases), use the method that works.

The only thing that does bother me is the stupidity of people in general. You get used to that. People are stupid. I'm a moron, you're a moron, that's how things are. We're all basically shit-flinging monkeys. Imagine how we look to some super-advanced race of robots or whatever. No surprise they always want to massacre us in movies.

http://www.dailymail.co.uk/sciencetech/ ... GENCE.html

Chimps that throw the most - usually, sadly, their own filth - tend to be more sociable, and have more highly developed speech centres

Explains this whole thread.

That said, it obviously is still a bothersome issue. People will jump into a field they know nothing about and make assertions like tpt = better. You can't blame them though, we've all gone through this process at some point and we've come to the conclusions we were led to, not those we choose on our own. It isn't something I hold people personally responsible for unless they become excessively assertive even in the face of evidence which points obviously to their belief being incorrect.

Even so, this is the human condition.

[andy-cytomic]

You're missing the point too. There's no need to do any maths. Any 2nd order biquad in any form has up to two unit delays in its feedback path(s) - no matter what analogue response it perfectly models. Claiming otherwise is purely ridiculous.

Urs, please get Clements on this, you are out of your depth here.

All recursive digital filters need to store some state, this has nothing to do with zero delay feedback, obviously there is a delay there since you store the state of the previous output! Go ahead and show me an iir that has no state, please go on.

So, back to the proper conversation, we are not talking about delays in the implementation, they are always required for an iir filter. What we are talking about here is if a delay is added to a feedback path to make the filter realisable or not. The RBJ DF1s add no such delay, they use a trapezoidal integration via the bi-linear transform which doesn't add a delay, it solves for the filter implicitly, and in doing so it just doesn't preserve the original topology, but if you care to read the topic of this thread we are talking about "zero delay feedback filters", so the DF1 RBJ qualifies on all counts as this.

[Urs]

Ok, I can accept that, thanks for the lesson.

(I might take the liberty to write an article "where zero delay feedback counts" to make *my* point)

[DaveGamble]

You're missing the point too. There's no need to do any maths. Any 2nd order biquad in any form has up to two unit delays in its feedback path(s) - no matter what analogue response it perfectly models. Claiming otherwise is purely ridiculous.

Urs, please get Clements on this, you are out of your depth here.

I see you're not taking my hint about how you're coming across. Honestly, I think language like that deserves an apology.

As I've said directly to you before, we are struggling with language.

The phrase "zero delay feedback filters" originated to describe a strategy of avoidance of the very poor Stilson&Smith "Let's just stick a delay in here and hope no-one notices" trick when implementing nonlinear filters.

You are making a category error. It can always be argued that you can apply labels outside their category.
For instance, DF1 filters are also arsenic-free.

Whilst there's no doubt confusion has arisen as to the precise applicability of this phrase, this thread has become something else.

Dear non-technical reader, please allow me to give you some commentary which I hope will allow you to understand this scenario entirely.

In 1996, two very clever chaps, called Stilson and Smith, both HUGELY respected by everyone here, wrote a paper on how one could emulate a Moog filter digitally.

That paper contained some excellent analysis. However, at the end, they found themselves with an extremely complex equation which referred to itself, and they couldn't see a way to solve it or simplify it without spending an awful amount of CPU. So what did they do? They cheated.

They made what could -very- politely be described as an "approximation". Actually they broke the emulation. But for it's time, it was still LOADS better than a lot of what was common. And they included code, which a lot of people added to things, and it was well received.

But the maths was wrong. Broken. Still sounded pretty good most of the time though. But when you turned up the resonance, or pushed up the frequency, it stopped sounding like the Moog.

Well, some years later, people started looking at the paper, and some people (notably Vadim / NI) spotted the "approximation" and set about fixing it. There was more CPU around, and Vadim brought in some maths that Stilson and Smith didn't. And suddenly Vadim had a filter that distorted great, even with high resonance.

So he wrote it up in a book that's now pretty well known. And people started reading this book and using it. Notably Urs, who used it to build the filters in DIVA, which sound spectacular because of the work Vadim did. Vadim made it accessible.

Vadim wasn't the only person looking at this, and Urs wasn't the only person to implement the maths, but they both did it well, and you've heard of them.

Users were asking Urs how the DIVA filters sounded so good, and why they used more CPU. Well, what was the explanation? It was because his maths DIDN'T introduce a random, +EXTRA+ incorrect delay into the filter circuit.

Specifically, Stilson and Smith introduced their delay into the "feedback path". That term "feedback path" really only means anything if you're used to looking at schematics and can make sense of the circuitry; it's being used in a specific technical sense referring to the Moog ladder filter.

Nonetheless, Urs' version based on Vadim's book had *zero* additional *delay* in the *feedback* path of the Moog ladder *filter*.

zero delay feedback filter. A phrase is born.

[Urs]

The RBJ DF1s add no such delay, they use a trapezoidal integration via the bi-linear transform which doesn't add a delay, it solves for the filter implicitly, and in doing so it just doesn't preserve the original topology

So if "implicit integration" does not necessarily maintain the topology then I just had an

It means that "implicit integration" isn't useful as a buzzword either, while, e.g. "TPT" is

[2DaT]

it solves for the filter implicitly...

Not implicilty , but explicitly. When you do RBJ DF1 , you actually
construct a filter that has same transfer function.The requirement for such a construction is Linear time invariance and for analog filter it is equal to static cutoff case.

However , we want to modulate our filter and preserve non-LTI behaviour of analog filter.In case of DF1 when cutoff is modulated , states of the filter get processed
"uncorrectly" , producing different effects like decreace in resonance (difficult to explain this in buzzwords but those differences are really hearable). In case of TPT , we actually transform an original filter, just replacing the integrators. When cutoff is modulated in TPT filter , states get processed "better" but still slightly different from analog.

[HobbyCore]

Anyone remember that scene in Big Lebowski where Walter Sobchak pulls a gun in the bowling alley?

A minute later they're sitting in the care with Walter asking the dude 'Am I wrong?' followed by the dude's iconic response.

This thread is exactly that with andy pulling the gun in the bowling alley.