Bandlimit or not?

MegaHurtz · 5 Apr 2010

I think yes, if you got the vocabulary to program your own osc`s for less cpu. You can be better off doing it yourself. Aliassing is also removed by the Pi and the moog filter. And it`s mainly the way how the controll is wired inside your code what the output of the osc will do. In synthmaker the oscs`s sound pretty sweet for common use, but they might get a little dirty on lower frequency`s. The pi filter on the other hand is a real peice of work. I beleive this is truely one of the best filters around. You may well use it to remove some aliasses.

Twisted_Chrylle · 5 Apr 2010

Thanks MegaHurtz

Yes, it gets blury because its 256*2 sample wave table, thats why I asked what size is best to use in wave table. 🙂

where do I find a Pi filter?

MegaHurtz · 5 Apr 2010

in your toolbox, the state variable filter.

Twisted_Chrylle · 5 Apr 2010

Thanks again MegaHurtz.

Now I feel stupid.. 😊 😀 But now I know, almost.

So I can simply just use the SVF as normal and most or all Aliasing is gone? (I suppose its the same with all those optimized SVF and Moog filters around)

Do you also know what size of a wave table is the best overall? 😇

MegaHurtz · 5 Apr 2010

for the most part it`s gone then. And.. Any power of 2 will do fine.

Twisted_Chrylle · 5 Apr 2010

Oh yes, Im strong enough to take 2 at the time... *kidding* Sorry but you lost me again... 😊

do you mean times 2 like 256 * 2 or do we still speaking about the filters? 😀

sorry, I have been up all night but Im almost there..

EDIT: Sorry, the coin just dropped, yes ofcourse, x2 like 2,4,8,16 and so on... But is 2048 better than 1024? or you don't know for sure?

MegaHurtz · 5 Apr 2010

It`s the tablesize that`s best to be any power of two. so 256,512,1024,2048 and so on. This to enshure a good clean loop.

There is some sceince in removing aliassing head on, and then theres taste. If on a synth you use your filter, it`s gone for the most part. Thats one thing you can rely on. The other is that you have to feel if you like the sound that comes out. If you like it, theres no aliassing police waiting to apprehend you. It`s one of the nicer forms of distortion.

Edit: I do a lot of wavetables, one thing I noticed is the way I do my interpolation and square windows. is that the sound is on the soft side already. So I wouldnt soft it up by that much. Also the way I calculate pitches dont do much harm to my sound. Since it doesnt jump to various positions, but flows there. The only slight aliassing filter i might use is on the FM in line. Becouse it gives a nice organic quality to the sweeps. So none to very little is the answer.

Twisted_Chrylle · 5 Apr 2010

You have been a great help MegaHurtz.

I suddenly feel there is no limits.. *whatch your ears* 😈 Now I wounder if old classical synths (analogue) have something called aliasing... 😇

I know Nord Lead 1 and 2 had, but they are not classical synths, not for me...

MegaHurtz · 5 Apr 2010

analog ones dont. digital ones do. But you can see the popularity on the nords, the sounds are built like a house.

Twisted_Chrylle · 5 Apr 2010

I personally don't like Nords at all, very thin sounding compare to most VAs/ANs. But thats a matter of taste.

So they dont build their sounds simular as we do in SM? They maybe use Concrete... 😉

rl · 6 Apr 2010

Twisted_Chrylle wrote: Thanks again MegaHurtz.

Now I feel stupid.. 😊 😀 But now I know, almost.

So I can simply just use the SVF as normal and most or all Aliasing is gone? ...

Sorry, it's not that simple. Aliasing means that frequency artifacts are "folded" back into the audible spectrum. This could even be at lower frequency range (20..800 Hz), so filtering would be of limited use. Better try to avoid aliasing at the origin (the oscillator coding). There are various techniques, like "band limited step", phase distortion, inverse FFT or MIP-mapping (as used by SM's oscillators). Sometimes maybe you want aliasing, as it could add "character" to the sound (The PPG wave was famous because of it's aliased oscillators).

But what happends if I use an Ramp to do a phase mod in a Sine or Triangle?

Because of discontinuities of the ramp signal (jump from max to min) the phase modulated signal will also be aliased.

MegaHurtz · 8 Apr 2010

Yes it`s indeed folding of the frequency domain, but it only happens at and above the niquist frequency. becouse the computer starts to make lower frequencys with higher ones. typically a 3rd order fir is used for this, say a biquad located at the niquist frequency. But yes. it really is completely gone if you run it through any lowpass.

Alternatively- what they did in the L2, Is generate some extremely high throughput, say 64/88. use some interpolation, and then sample down with some noise shaping and/or dither.

But.. good luck getting these bitrates 🙂

unkargherth · 16 Jul 2010

Twisted_Chrylle wrote: Is the end resault Bandlimited or Non-Bandlimited then? 😕

Even using the stock "bandlimited" osc's you could get lots of aliasing doing certain operations. The typical issue is FM. Take a sine osc and connect output to phase input. Voila. you get lots of aliasing. and the problem is not in the sine wavetable or in bandlimiting. The problem comes from the fact taht doing "digital phase modulation" is , from mathematics point of view, INHERENTLY a NON-LINEAR PROCESSING, an that's produce aliasing in a ny digital oscillator

Then your only solution, is to apply the standard oversample->block filter->downsample. Oversample by 8x grade, blokfilter with a 10th polyphase filter ( 108 dB or so) and down sample. It is not cpu friendly, but it is digitally clean enough

Twisted_Chrylle wrote: I think I know that everything "past thru" wave table gets bandlimited, Am I right?

As far as i know, no, in the sense you are asking. if the waveform you provide to wavetable is not alias-free, the result won't be alias free, but, it's true the result won't generate "NEW" aliasing when used on higher frequencies