Zero Delay Feedback Filter (How to test if your synth has a )- Xils-Lab White Paper -

[AdmiralQuality]

As the technical discussion wore on, it became clear that this was a topic for experts.

You don't need to be an expert to realize that the digital filter designer may have put code in to adjust the resonance level across the cutoff range.

Since you have no way of knowing what's been done here, if anything, you can't make any assumptions based on measurements of the resonance peak levels across the cutoff range.

It's as simple as that. There's myriad other issues as well, but the show stopper is right there at that first assumption -- that they know how the filter is designed. As I've already pointed out, if you know how the filter is designed you don't NEED to test if it's zero delay feedback because you already know.

Oops, I was supposed to be done with this. But it just won't die as the OP and his phantom sponsor won't give it up.

[aciddose]

If it is the second case (for the instance the resonance drop-down in the bottom of the spectrum, bellow around 100 Hz) I think you didn't read what I said/posted/explained several time.

this actually doesn't need to happen. this is with an unbuffered dc blocking capacitor. if you use a buffer here - say you use a quad opamp, one for the expo current source, three for the output you can apply a much lower frequency dc blocker here. if you very closely match VBE (and, use a current mirror to maintain constant voltage) you can also eliminate cv feed-through and adjust for dc directly rather than using any capacitor at all.

it's not very difficult to design a biased-transistor ladder using heaps of extra opamps to eliminate all capacitors from the circuit other than the integrators themselves. you can also adjust the current on both ends and do all sorts of interesting things. the implementation you see in the minimoog, for example, is merely the cheapest one that actually works reasonably well. ...and only reasonably well.

the designers of these circuits were not geniuses. these were often semi-educated tinkers and in most cases they were operating in fairly unexplored territory. it's extremely easy to apply modern knowledge and technology to eliminate the issues from their circuits. sometimes issues that they had never even noticed.

the circuit used in the minimoog wasn't designed to be accurate or to sound best, produce low noise, low feed-through or have high linearity, any of these things really. it was designed to be cheap and efficient. this was at a time when the opamp in IC form was a significant cost. the cv mixers in the minimoog were all discrete!

my implementation of this filter can produce 1hz oscillation no problem. as can my modification of the korg-35, or it's improved implementation which hasn't ever been named, i refer to it as "x1". these can be used as LFOs. these can be used to filter LFOs.

you refer to these things in such a static way. "all" of these filters will had highpass stages in the feedback? admit it, you've never worked with this at all. you're clueless.

[xavier]

I understand your point acciddose.
You tested a lot of analogue filters, measuring them and got flat or not resonance responses.
But with all your examples, are you speaking of "real" implementation for schematics filters ? simulations of them ? or digital implementation at a limited sample frequency ?

All I'm saying is that a zero-delay digital implementation must have the same behaviour of the real model.

So I agree with AQ statement : If you don't know what is the model and then what is the relation between phase and frequency, the test result can be difficult to understand.
But IMO, in most of the cases, it gives you good informations, especially at the very upper frequency range, over 10k (spectrum warp, resonance pick, aliasing, ...)
As AQ pointed for Poly-Ana as it gave some strange dips you can't find in real things.

Is this thread should have been put in the developer forum ? IHMO no, every one, without a lot of knowledge, could want to get some clues about all the marketing blurb they got.
These clues are sometime obvious, sometime more difficult to get.
Some will don't care, some will be interested in separating true and bla-bla.




Best regards
Xavier

[AdmiralQuality]

As AQ pointed for Poly-Ana as it gave some strange dips you can't find in real things.

You're not understanding what the "dip" is. It's not a dip in resonance levels across the cutoff range. It's a dip in the frequency response of the filter, just beyond the resonant peak. Where there should be a nice ski slope heading downward after the peak, instead there's a little valley then it rises back up, then continues in the slope you'd expect. You don't need to sweep cutoff to see this, it happens at any static cutoff setting. That's what the red arrows are pointing at on the (static images) of the spectrum analyzer I posted.

If you remove the resonance, the frequency response looks exactly like it should, a nice 24 dB/oct cutoff slope. So the "dip" is clearly caused by the feedback delay. But increasing oversampling decreases this dip. Another thing that you can't make assumptions about in a strange filter that you know nothing of its inner workings.

For YOUR test, the actual resonance level in Poly-Ana is pretty much flat across the cutoff range at 16X oversampling. And at 1X oversampling the resonance level goes DOWN towards the high end, which is exactly the opposite of what you attest it should do. Hence my joke that it must have a negative delay feedback that looks into the future.

Is this thread should have been put in the developer forum ? IHMO no, every one, without a lot of knowledge, could want to get some clues about all the marketing blurb they got.
These clues are sometime obvious, sometime more difficult to get.
Some will don't care, some will be interested in separating true and bla-bla.




Best regards
Xavier

Do you really think 17 pages of false suggestions, refutations and "bla-bla" are helping any laypersons "get some clues"?

It's self-serving spam. And worst of all, your "white paper's" suggestions are just plain wrong. Putting it in the developer board would make a tiny bit more sense, as you clearly need some help. Though you've received that help and it doesn't seem to be sinking in.

[ariston]

Is this thread should have been put in the developer forum ? IHMO no, every one, without a lot of knowledge, could want to get some clues about all the marketing blurb they got.
These clues are sometime obvious, sometime more difficult to get.
Some will don't care, some will be interested in separating true and bla-bla.

Well, yes, some will care and some won't, some say tomayto and some say tomahto. My point, however, was that we don't NECESSARILY need the technical info to separate the chaff from the wheat: our ears and experience are enough. Of course some people like to know things in detail, at least more than others. But your "test" is flawed, and to point out why it's flawed takes a little time and effort; to make the kind of constructive criticisms you're requesting takes a lot more than that: knowledge of filter behaviour and DSP, plus a lot of hands-on experience with original equipment. Which brings me right back to the "expert" argument.

I'd be interested to know whose marketing blurb and bla-bla you're referring to, by the way. Tread carefully, though.

I still can't understand why one thing is not blatantly obvious: don't you see that there is a certain haut goût about a developer posting this kind of thing? I've been here a coupla years now, and this entire topic has only come up ever since a certain synth was released. In effect: why should we trust information that comes from a competitor? Because it's innocently presented? Because it's peppered with faux sincerity and mock humility? Because you send in one of your colleagues instead of posting yourself? Said colleague, who by turns claims to be a part of your company ("we at Xils") and by turns says he's a private individual whose affiliation with you is merely superficial - whenever it suits his intentions? Wouldn't that reek of manipulation to you? I don't think I can put it more clearly than that.

If it had come from somebody neutral, all of the hubbub in this thread would have concentrated on the merits of the test itself. And the thread would have been over in three pages.

[aciddose]

it should have been clear i was testing real electronic circuits. what do you mean "real implementations" ? there is no such thing as a simulation of an analog filter. the thing is either analog, or not. the fact it is being simulated makes it automatically discrete.

you get similar results in simulation but you have to do a hell of a lot of work to get there. the basic spice features do not automatically provide for any real-world testing. you have to understand the function of the circuits, all influences and effects and try to model these using the features you have available.

to plug a circuit into a spice simulation and expect it to provide real results is idiotic at best. a simulation has a sample rate too, which generally is variable. the integration methods used are variable as well and there are many complex features. the claim actually that certain methods are applied to produce "zero delay" is ridiculous. these are never "zero delay", they're only approximations. the simulation is still discrete. any activity which occurs between two samples is only approximated, if accounted for at all.

in some cases it's simply impossible to get accurate results without bench-testing because even if you can model the effects you want to examine, the cost is prohibitive.

All I'm saying is that a zero-delay digital implementation must have the same behaviour of the real model.

no. the "zero delay" implementation of which there are many does not always reflect the behavior of the circuit itself. in fact there are more cases where these methods fail than cases in which they succeed. that we can barely produce audio-rate results after a ridiculous investment of effort into tweaking the simulation is laughable.

strange dips?


there are many dips you can find in real things. you obviously haven't experienced many real things.



...and here is a 24db/o slope with no dip. no oversampling was not involved and it is not zero-delay. is this because it was analog? or was it magic?

[Ingonator]

I'd be interested to know whose marketing blurb and bla-bla you're referring to, by the way. Tread carefully, though.

This is easy too. It was first started with Diva, before nobody including me didn't care about zero delay feedback filters. Later that as that "blurb" seemed to get more popular it was "adapted" by Tone2 for Saurus. I am quite sure Tone2 would have not mentioned it without the "blurb" produced by the release of Diva.

AFAIK this thread was done as someone here at KVR (hakey ?) stated that the Xils synths did not feature a 0df filter design which seemed to be wrong.

I appreciated that someone finally started to post a kind of test for this feature and another thank you goes to AQ for his own test (after i asked for another way to test this as the first one does not seem to be 100% safe).

It also seems to be the case that filters with high oversampling rates could get a similar behavior than 0df designs. Which could be nicely tested with AQ's test and PolyAna when using different quality settings.

Anyway personally i am not a really big friend of such technical discussions as first filters without 0df design are not bad per se and second there are many more facts that make a filter sound bad or good which seems to be the case for Diva too.
The settings which you need to detect the 0df behavior are settings which will not be used for most of the "usual" presets/sounds anyway. It's mostly about having quite high resonance at a quite high Cutoff frequency which would maybe apply to e.g. emulations of TB-303 like bass sounds. For "typical" analog sounds like e.g. Synth Strings or Synth Brass sounds you need only a small or no amount of Resonance so the 0df filter design is useless for those.
What makes the special sound of e.g. Diva is more about authentic behavior of the filter and Osc modules and about a good low end of the sound which to my knowledge is not based on the 0df filter design alone.

So while i own synths like the 3 Xils synths, Saurus, Diva and also PolyAna and like them all my opinion is that we did not really need this discussion about zero delay feedback design at all, incuding this thread. On the the hand like already mentioned it was NOT Xils Lab who started this discussion.

I also agree that none of the tests provided here are 100% foolproof but they give an idea about what happens with the resonance of certain filters which is also interesting for sound design purposes so maybe finally it has a use/sense at least for me anyway.

Just my 2 cents...


Ingo

[Urs]

I feel strange that a lot of you are posting statements without any images, videos, explanations.

Well, taking videos etc. isn't exactly part of the test, and not all of us have that much time on their hands (otoh we're preparing a paper too, a scientific one, but that deals with subjects beyond delayless feedback - we've moved on already).

However, believe me or not, we thoroughly analysed more than 30 analogue filters in the past 3 years. Hardly any of them exhibits any constant q over frequency at any level. Thus we don't believe that this is a desirable property to model in any emulation of analogue filters. It would be a waste of time to go through this again and I don't see why we need to prove anything.

When you put up a thread like this without doing your homework first, then you shouldn't blame others for not understanding it, not conducting it properly, not using the right tools and what not. It's your own responsibility if it deflates.

Also, quite frankly, I'm very over the spite that swings in a lot of threads lately. Provocations like constantly referring to cpu usage - as seen in many posts - won't change a thing. It's just a poison that most of us have developed immunity to - one that stings for a few moments, then goes away.

[aciddose]

... Thus we don't believe that this is a desirable property to model in any emulation of analogue filters.

definitely not in emulation and of course i understand if that is the goal.

it is useful in some cases to eliminate these issues however as often times as for example in the korg-35 filter which requires compensation in-circuit which leads to excessive feedback at high frequencies.

the results are far less desirable than what you would get if you simply used an opamp without compensation and more accurately matched capacitor values, and thus never needed to "correct" the issue you never introduced in the first place.

it's possible to do a korg-35 without needing that POS capacitor and the horrid effects it introduces. of course some would call that character. at times it's extremely annoying though, as well.

[Urs]

... Thus we don't believe that this is a desirable property to model in any emulation of analogue filters.

definitely not in emulation and of course i understand if that is the goal.

it is useful in some cases to eliminate these issues however as often times as for example in the korg-35 filter which requires compensation in-circuit which leads to excessive feedback at high frequencies.

the results are far less desirable than what you would get if you simply used an opamp without compensation and more accurately matched capacitor values, and thus never needed to "correct" the issue you never introduced in the first place.

it's possible to do a korg-35 without needing that POS capacitor and the horrid effects it introduces. of course some would call that character. at times it's extremely annoying though, as well.

Well, the Korg 35 is also the one that was a bit beyond us. We might give it another go later. Andy Simper seems to have nailed it, maybe he'll share some insights with us when he's at our office next month.

[Richard_Synapse]

Ihmo this thread should've been in the dev corner first for feedback (no pun intended)

Then other devs can suggest possible improvements and modifications to the test, and won't gun it down...basically the same procedure you use in beta testing plugins.

Richard

[Lotuzia]

....things ..... ... Ooops So I promised at the beginning of this thread that i would not post again in it, but I'm so trusty I now eat my word several times per day ........ (Here its me talking for you Ariston so that it can be clear for everybody reading that )....other things ....

I still can't understand why one thing is not blatantly obvious: don't you see that there is a certain haut goût about a developer posting this kind of thing? I've been here a coupla years now, and this entire topic has only come up ever since a certain synth was released. In effect: why should we trust information that comes from a competitor? Because it's innocently presented? Because it's peppered with faux sincerity and mock humility? Because you send in one of your colleagues instead of posting yourself? Said colleague, who by turns claims to be a part of your company ("we at Xils") and by turns says he's a private individual whose affiliation with you is merely superficial - whenever it suits his intentions? Wouldn't that reek of manipulation to you? I don't think I can put it more clearly than that.

........final things ...

Simply that relative positions can evolve over time. So I'm more involved with Xils-Lab, and very happy to be, than before, and its not by turns, and btw its clearly mentioned in my sig ........ for a certain time now.

So when I talk about Xils-lab I simply represent Xils-lab. When I talk about soundbanks for example, I'm representing Le Lotus Bleu. And when I try to help an user for a technical subject on Kvr, I guess I'm just a nobody. I know its difficult for you to understand this, but well I just see it like that.

I wont insist of your hidden motivations by attacking Xils-Lab, or its representatives, in repeated numerous ways, numerous posts simply because you have exposed them quite clearly a few posts before in this thread, so I'd say they are barely hidden. ( you wrote " And still, somehow, despite all attempts to obfuscate the obvious: [a particular synth] is the best-sounding analogue emulation on the market right now )") You will certainly have noted that we did not refer during all this thread to any particular synth, and that the supposed "attempts to obfuscate the obvious" only exist in your -fertile- imagination, wich is certainly full of dramatic plots and conspiracy stories ......

I have nothing against fanboys, actually I'm myself fan of a certain number of things, or at least afficionado, or lets just say that when I love something, I can take a real pleasure to try to communicate this love to other people. So, nothing against fanboys when they talk in a positive and passionate way of what they like. However when a great part of their energy is used to try to destroy what they see as competitors to their favorite toys, I think its both very counterproductive for the people they support, and a global negative attitude wich only brings "noise" in many threads. Honnestly I'm glad you're not fan of Xils-Lab musical instruments, cos, if you were, I would probaly suggest you, in private of course, eventually with little hope, to try to be a bit less less vocal about other developpers who make "different" musical instruments and to try to respect them ( Well sorry, I know you can prefer by far absolute terms like best/worst than "different" but well I tried )

I dont think I can put it more clearly than that. ---->

(here's just a reminder of what you posted in this thread on May 11th, page 5 )

" I never mentioned "truth", I'm just giving my POV; I'm sure you don't need my help for anything. You don't see it my way, I'm fine with that. I see BS, I call BS. Simple.

Last post in this thread, promise. I'll let you get on with your fascinating technical discussion - may it help you make better music. I also solemnly vow never to post anything regarding Mr. Shrug or his dealings here ever again. Best for all concerned, I'm sure."

[ariston]

@Lotuzia: I just noticed you posted something, no doubt a lengthy, shruggy reply to what I said. Take note that I have finally muted you; days ago, in fact. I've got better things to do than read your feeble attempts at justification. So please, don't bother. You are non-existent to me.

Which, sadly, doesn't extend towards others quoting you, so I guess I'll have frequent reminders. They do nothing but make me smile, so that's okay. You're only hurting yourself and digging the hole a little deeper.

"Bye now", I'd say, but I'm sure you can't resist commenting on this post as well. So fire away, if it makes you feel better. You might as well be talking to a blank wall, because I just won't read it. But I assume you enjoy hearing yourself blather on, so that's probably alright then.

@Ingo: well said.

@AQ: I think your posts have been a great contribution in helping to clear up this mess. Thanks a lot for your input.

[Lotuzia]

it's possible to do a korg-35 without needing that POS capacitor and the horrid effects it introduces. of course some would call that character. at times it's extremely annoying though, as well.

Well imo an emulation aims to replicate in the closest possible way the character of the original unit.

Else it is a less acurate emulation, while it can be a nice musical instrument in se, no doubt about this.

Maybe an intermediate solution could be to try a "best of both worlds" approach, offering to users both the original, and an alternative mode with what the devs think is "a better/more musical version" than the original. But as it can be so time consuming and /or difficult to implement, as it can add confusion in the users perception of the instrument, I agree its not obvious to decide if even this is desirable, or less desirable.

Well to each his own.

[AdmiralQuality]

@Lotuzia: I just noticed you posted something, no doubt a lengthy, shruggy reply to what I said. Take note that I have finally muted you; days ago, in fact. I've got better things to do than read your feeble attempts at justification. So please, don't bother. You are non-existent to me.

You're not missing much. He's just rambling senselessly now. And yes, at great length.

@AQ: I think your posts have been a great contribution in helping to clear up this mess. Thanks a lot for your input.

You're welcome. This could have been a good topic if it didn't start off with completely incorrect information right from the first post. And it is rather obviously spammy. If I tried to pull something like that on the Instruments board the immediate massive pile-on would make this thread look tame.