How much db for an antialiasing filter ?

[Tepeix]

Yep, my idea is maybe not good and hard to implement.. Don't know if i will try as i'm more into fx than synth.
The difficulty being that every note might produce something in relation of any other notes, making a lot of calculation..

For mixing overdrive and clean it's possible, the famous tubescreamer pedal do it.
I think it's possible with distortion also, but with certain setting like 50/50 it could sound strange.
The worst is if the distorded signal is reversed, then mixed with the original, erasing the original frequency of the note.

Not sure myself about the true difference between overdrive and distortion. Lot of overdrive have asymmetric clipping that will sound softer. But i see more the difference as a sound that is almost clean and only break whit high attack volume, and another that will do a lot of harmonic at low volume then even more.

Seeing more the distortion prim, it's more an overdrive for me. It only do 1 harmonic at X3 multiple.
With the 4X oversampling that provide Martin Vicaneck you could get rid of all aliasing i think !

But strangely i don't hear so much difference.. For a better sound i think that other parameter must be take into account. Compression, dynamics, intermodulation...

The bandstop is not really a good solution, because it's really hard to know where would occur the aliasing, it could be close to a desired/original frequency.

[R&R]

With the 4X oversampling that provide Martin Vicaneck you could get rid of all aliasing i think !

Yes I also noticed it does remove most aliasing... The distortion becomes almost usable

But strangely i don't hear so much difference.. For a better sound i think that other parameter must be take into account. Compression, dynamics, intermodulation...

Play higher notes and feed the distortion with a signal already full of alias artifacts and you'll see/hear

I agree... to create a "better" sound more advanced DSP has to be done...

The bandstop is not really a good solution, because it's really hard to know where would occur the aliasing, it could be close to a desired/original frequency.

Yes placing bandstop close by harmonics is a bad idea... But sometimes the aliasing can manifest at frequencies lower than the fundamental. Should be easy to track particular band/range that should be filtered out using played note... Any frequency modulation would be tricky to take into account I guess.

But for FX even tracking clean simple notes/instruments would be difficult. Running into the delay problem of detection... and also tracking a vast amount of fundametals seems undoable. But maybe some FX do just that... I have no idea...

[Tepeix]

Yep in fact i continue the research, i'm sure there must be some secret and mystery about how to make a dsp drive sound like an analog one.

Héhé, i try a very simple and minimalist polydrive like i was thinking.. Sound strange
Not really a drive but i recognize some part of the sound.
It's very simple, it only work for 4 note, then 4 note again.. Using the SSE in the most simple way.
Trying to make the harmonic that occur between 2 notes when using asymmetrical clipping.
(I suppose that simply adding the frequency do the job)
but it's only one of the harmonic created.

[Tepeix]

I just read this interesting paper : https://www.dafx12.york.ac.uk/papers/da ... ion_45.pdf
About different soft clipping.
(i don't understand all and where intermodulation are evaluated)

[R&R]

Trying to make the harmonic that occur between 2 notes when using asymmetrical clipping.
(I suppose that simply adding the frequency do the job)
but it's only one of the harmonic created.

I kind of understand what you are trying to do by adding harmonic but not why there are 3 summed oscillators?

I just read this interesting paper : https://www.dafx12.york.ac.uk/papers/da ... ion_45.pdf
About different soft clipping.

Most math is beyond me...

Part of the conclusion:
"However, this conclusion is based on a limited set of measurements and further investigation is needed."

Pretty much like every schoolpaper i've ever written...

However, their graph, table and the observations made... state that Tanh and Exp5 are in fact the easier controllable options. Taking all variables into consideration (performance and complexity of the task of softclipping itself) THD and instability may be complete dealbreakers. I have no idea. Maybe someone with FS DSP and math skills has exeperimented...
They even used 96khz so the problem of aliasing at low samplerate wasn't adressed.

i don't understand all and where intermodulation are evaluated

By intermodulation you mean cause of aliasing or over all intermodulation? (meaning lower harmonics modulating higher freq)

[Tepeix]

I make time also to understand those equation..
In fact they are not so much complex but the notation is.. Not clear..
Why mathematician lack so much clear explanation ?)

The cubic soft clip is simply : 1 limit the input to the range 0.66667 to -0.66667. (like an hard clip)
Then x = input*2.25 - (input*input*input*1.6875)

But for the exponential i'm not sure because i get good result for exp 2 but not for exp 5... ...

There's 3 operation in my test, because it try to make relation of one note with 3 others (if they exist).
(But not more because it will be more complex to do it..)

In fact my test do not harmonics but intermodulation. I forget that this is also it, because the high intermodulation sound good to my hear.

The harmonic is when an original frequency is multiplied. 800hz would make 1600hz then 2400hz then 3200hz..
But when more than one note is produced before a non-linear/overdrive process they will be intermodulation between the note.
They would produce f1+f2, the high intermodulation that i like. (But maybe it could be cool to attenuate it in some way)
But also f1-f2 and f2-f1. This give the bass intermodulation. Those one sound pretty bad to my hear.
That's why i try to use a high pass to attenuate them.
I get it from a video of Wampler pedal where he explain that most overdrive pedal use a highpass filter.
This one will be fixed in pedal. This do a great job, but i'd like to find an adaptive filter that could change when we play higher note.

Well it's true that i don't understand all from the paper.. Maybe instability would say that the harmonic and intermodulation would oscillate more and give more tremolo effect that ruin a little the overdrive ?

Now i get another paper.. https://simulanalog.org/clip.pdf
Seams to suggest that the clipping must more occur whit transient.
I suppose that using a peak detector that give more gain could be good.

Finally the initial question get no answer..
Not sure but i think that 100db might be to much but at least i'm sure that no alias could happen.
At 60-70db it seams most common.. Almost sure that no alias could be audible (if in the range of the filter).
But maybe even 20 or 40 db could be sufficient ?

[R&R]

The harmonic stuff is much to advanced for me maybe in time... further on...

I don't know how to. But two things I would experiment with myself (in my ignorance) if I could... Maybe you can try if you can. Don't know if it's a good route but atleast it's a fun experiment if you're trying to create custom FX maybe.

1. Creating a linear filter (with a slope across the whole 20-20khz range). Then feeding that linear LP or HP with the clean (incoming) signal and using some kind of RMS average code to output average as a variable controller output for other modules (overdrive and other filters in particular).

2. Then altering/writing an custom overdrive... so as when the signal content contains alot of low freq the overdrive softens the curve calculation (iow saturation). That in my mind would act a bit like analogue, since we then are simulating overdriving the input in a manner of speaking. And maybe I would also try and feed/mix in a harder clipped HF signal as well to preserve the hi freq. The "mixing in" of the HF clipped signal maybe also can be adjusted with the average controller output.
I feel that the sound becomes a bit dull (lacking clean alias-free HF content) when using for example the toolbox overdrive/distortion.

[Tepeix]

Interesting idea, but for me, it's the high that must have less overdrive.
I would prefer to try this : Separate the signal in 2 with the filter mentioned (long slope)
Overdrive the bass part then clean boost the high and compress it.

I made this little filter/separator. But once again i don't know if it is good, it's a little experimental..
(there's a sample rate adapt inside, but i'm not sure he will always do exactly the same in each
sample rate; but it's almost the case specially with low cutoff.. )
Also i'm not sure for the group delay.. I'm experimenting with very simple filter with only one delay mix..
Using them in series, the group delay seams not so bad but i suppose might be affected.
(the group delay make some frequency have more latency that other's which could make the sound different, specially with percussive sound)

[R&R]

Overdrive the bass part then clean boost the high and compress it.

I guess it retains the transients and only affects the dynamics when used mildly. But can you compress to needed level without crushing the sound too much... considering you wan't to boost the HF before. Doesn't compression become a bit overdrivey (in a "digital domain sense") as well? I'm guessing you're talking about a static compression?

Also i'm not sure for the group delay.. I'm experimenting with very simple filter with only one delay mix..
Using them in series, the group delay seams not so bad but i suppose might be affected.
(the group delay make some frequency have more latency that other's which could make the sound different, specially with percussive sound)

I don't understand the delay part since I don't work with FX. The compression is a point of delay?
Or... where does delay otherwise occur in your scenario?

[Tepeix]

That some of my assumption, i always have to reduce the level of overdrive when i go to +1 octave.
(I think that my filter does not have enough attenuation for this. Not sure how much is needed)
But hearing distortion and overdrive they have lot of treble. Without those dissonance that make me want
to diminish the level.
So i suppose that the treble are more compressed than distorted ?

For the group delay, it's for the filter, but finally i don't know..
That's a very light filter so i don't think it could add so much latency..
Also the principle is to subtract some bass from the signal, treat it in a different way then add it again..
I don't know what happen for the latency in this case.

Filters add some latency, the more fast slope, higher order, the more latency you get.
In this picture i try to measure latency and group delay.
The green is the original signal. The white is the bass part of my long slope filter.
The latency is very short as the slope is long.
The blue is a more common filter with faster slope. You get more latency.
At some point if you subtract the filter from the original signal it could be better to also delay the original or you will get strange frequency curve..
The red is a filter from hell i made there's a lot of latency but not only. Here the group delay are completely messed up. The pattern that make multiple frequency change a lot.
That's because some frequency get a very different latency than other's.
From what i read messing with the group delay could be bad especially for percussive sound.
Sometime it could be a desirable effect in some way like in spring reverb, sometime it could mess with the sound. (From what i understand.)