EDIT: nevermind. Don't want to derail this thread again

I'm not making any claims one way or the other. Just pointing out Lotuzia's strawman (to my knowledge no one has claimed that non-0df filters can't produce rapid transients).

Hi,

I agree with AQ that a test which can give an "absolute" answer to the question "is my filter" zero-delay (meaning is my filter correctly manage the delay(s) in its digital implementation) would be nice.
This test doesn't give an absolute answer in all cases, but is a tools to give a lot of interesting informations about how the filter is handling the delay.

So what is behind this test ?
Well know for all DSP developers, directly substituting the Z-transform to the p-transform in the analogue filter formula will bring
_ spectrum warped
_ additional phase shift when feedback loops are used.
_ There are other points, but these ones are the most noticeable.

Is that true or is that false ?


So if we take just the feedback loop case, this additional phase shift implies that the add between the input and the output won't be done with the same phase that this would be without the delay.
As the phase is frequency dependent, this add will be frequency dependent.
Then, if the add is used for the resonance this one will be frequency dependent.

In the above statement, nothing were said about the filter, its internal schematics, capacitor, integrator, feedback path. Only that the delay implies this dependence.

CQFD

So the test we suggested is based on the two following results :
_ The frequency and the Q are independent : this means the delay is correctly managed in a way or in an other

_ The frequency and the Q are not independent : the answer is not obvious.
But if this dependent seems to be in the same range than analogue models (drop down in low frequency in the case of moog for instance), then we can't be wrong in saying that this delay is correctly manage.
On the contrary, if the dependence is weird (drop down, then raise again at the upper part of the frequency spectrum for instance) then you can say, hmm something is weird, maybe the daly is not correctly managed.

An other point : The phase shift brings by one delay is 180*f/Fs, with Fs the sample frequency.
At Fs = 44.1kH, and f = 10kHz, p = 40° which is important.
But oversampling to 64, we get 0.63°, quite unnoticeable.

So yes, oversampling a lot is a way to manage the delay.


Best regards
Xavier

What about where there is rapid modulation such that cut off has changed significantly during the one sample delay?

In this case doesn't using the last sample as the input for feedback lead to significant discrepancies between that and the results with 0-delay (ie instantaneous feedback)?

You shouldn't need spectrometer actually. I didn't reply last night, 'cos I figured I'd do a bit of field work before I go making bold claims based on a few cheap filters (that I have at home), but this stuff should be audible with anything I've tested so far.

Now, I didn't get my hands on a model D on such short notice (so maybe those are immune), but I figured the Voyager in the local store should be discrete (and high quality) enough to bother checking out. Unsurprisingly, with oscillators turned off, the self-oscillation pitch shift from the edge of self-oscillation (ie minimum that you get a stable sine) to the maximum position of the resonance knob seems about half a semitone or so (and as far as my ears can tell, more or less independent of cutoff, once you've past the point where DC blockers have significant effect). As far as I can tell without poking probes into the guts of the thing (to see exact feedback levels) it seems to agree with basic 0df models pretty well.

However, (and this actually surprised me) I also tried some DSI stuff just to see what they'd do... and you can observe pretty much exactly the same with Prophet-08 as well (for Mopho I can't tell, 'cos even with the keyboard version I couldn't figure out how to program it for the test setup). It's a little more subtle (don't know if the resonance limit is just lower) but certainly perceptible. That's curious, because I thought that was supposed to be an OTA cascade where this should be a non-issue.

[edit: curiously it seems CEM3328 datasheet doesn't draw buffers between OTA stages.. if that's what's really inside those chips, could it explain the strange behavior?]

[edit2: stupid me, I just realize that it probably should happen with an OTA cascade like CEM3328 too, even assuming the OTA inputs draw no current, assuming ofcourse that the OTA inputs are non-linear]

Strawman ??? Well ... as always innuendos and not the single beginning of a fact I've read this in some threds, and at least in another thread in wich you participated massively with all your .... erf ...... knowledge.

Knowledge : After all weren't you the one that claimed back in November 2011 " We can be sure that Xils didn't implement a zero-delay-feedback filter ".

And now with all your knowledge, you have to ask pityfully in this thread for a method to be able to judge if a synth has 0df filters or not. ( Btw who is us ? Do you have a hamster, or a cat, or is it that you think you can think for all the people ? )

But never mind, your simple common hatred, as well as your deep knowledge, just shows so much in this thread that, well ........ as usual people will judge.

@mystran - stupid me, dunno what I thought when I read your reply, but of course they're going flat. All of them do. Just the Steiner-Parker doesn't - it goes sharp first, and then flat again.

I think your point still holds: even with the subtle flattening OTA/transistor cascades are quite close to independent unless pushed to extremes (in terms of feedback or input levels). Closer than what you typically get with extra delays.

Oh and off-topic note: Playing with the Voyager was healthy (for wallet), as I don't really want one anymore. I found the design rather painful in some ways actually.

Yep. I was wrong. Apologies to Xavier.

how it is easy to hate and now to apology ?
most kvrian are just here to count the number of post they made or the number of dll in their vst folder ...
what a sad forum

Okay, I guess that admitting mistakes and apologising for them is a bad thing?

I happily admit to only having a layman's appreciation of the subject - and you're hardly the oracle on all matters DSP, are you?

Well seeing as many people have pointed out that the 'simple test' you outlined in the first post fails to distinguish between 0df and non-0df, it seemed a relevant question to ask. I don't know what's pitiful about that.

I really haven't a clue what your babbling about here.

personally, i think that is easy to apology, for example, people can kill other people in a car accident and apology ...
it's better to take our responsibility before to have to apology
but most people prefer to act impulsively before thinking about their act

A bit of a ridiculous comparison, don't you think?