Just trying to clear up some potential confusion there. To me a theoretically perfect saw would be one with infinite harmonics that followed the 1/n amplitude rule. Naive saws are indeed geometrically perfect, but not harmonically perfect.Aroused by JarJar wrote:Don't know what you mean by "harmonically perfect". I just meant for example that a theoretically perfect saw would have zero rise time (instantaneous change from -1 to 1) and an absolutely straight direct ramp between 1 and -1. This doesn't happen in reality, as the finest analog doesn't operate at infinite speed and bandwidth. In digital this waveform is represented by a "naive" line from 1 to -1. This aliases like a hell.hakey wrote:theoretically perfect = harmonically perfect?Aroused by JarJar wrote:"naive" (aka theoretically perfect) saws and such.
Do hardware VA's alias as badly as software?
- KVRAF
- 5233 posts since 25 Feb, 2008
- u-he
- 28042 posts since 8 Aug, 2002 from Berlin
If I was spiteful, I'd have my lawyer send cease and desist letters to any company that claims to have "aliasing free oscillators" in their digital synths. Our lawyer could actually make a living from that.
- KVRAF
- 5233 posts since 25 Feb, 2008
What you said was confusing. And any occurrence of the '48kHz is insufficient to capture the harmonics' canard will inevitably draw comments.Aroused by JarJar wrote:I think it's rather a matter of you're being confused by your own intent to pick at whatever I say rather than actually try to understand it.
And you still seem to be suffering from the delusion that all VA variably generate waveforms.
Edit: fair enough, I did say "VA oscillators aren't sampled and repitched".I should have said that "good quality VA oscillators are generated" or similar.
- KVRAF
- 5233 posts since 25 Feb, 2008
So the look-up table method, does this, as AJJ suggest, simply involve sampling a waveform at a single frequency and then repitching it, with the undesired effects (missing harmonics for low pitches aliasing for high) following on from that?
Are there any modern VA's that use such a crude method?
Are there any modern VA's that use such a crude method?
- KVRAF
- 23077 posts since 7 Jan, 2009 from Croatia
Roland Gaia uses such method, its oscillators are sampled. O.o'
-
david.beholder david.beholder https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=159839
- KVRAF
- 1865 posts since 13 Sep, 2007
Wait a second, but what about representation as sum of sines: k1*sin(x) + k2*sin(x*2)+ ... + kn*sin(x*n) - it's perfect aliasing free osc, ain't it?Urs wrote:If I was spiteful, I'd have my lawyer send cease and desist letters to any company that claims to have "aliasing free oscillators" in their digital synths. Our lawyer could actually make a living from that.
Murderous duck!
- KVRAF
- 23077 posts since 7 Jan, 2009 from Croatia
Not a single sample IIRC. I don't remember correctly, but it might be a set of single-cycles for each key. Or something like that.
-
Music Engineer Music Engineer https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=15959
- KVRAF
- 4264 posts since 8 Mar, 2004 from Berlin, Germany
yes - and that's the way, i rendered the example file for the anti-aliased sawtooth. it's not very practical for a realtime-synth, though. in a realtime-synth, you will typically have some kind of trade-off between between computational cost and the level of aliasing.david.beholder wrote:Wait a second, but what about representation as sum of sines: k1*sin(x) + k2*sin(x*2)+ ... + kn*sin(x*n) - it's perfect aliasing free osc, ain't it?Urs wrote:If I was spiteful, I'd have my lawyer send cease and desist letters to any company that claims to have "aliasing free oscillators" in their digital synths. Our lawyer could actually make a living from that.
-
david.beholder david.beholder https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=159839
- KVRAF
- 1865 posts since 13 Sep, 2007
DIVA doesn't know any cpu tradeoffs Altho i guess it's not using sum of octave sinesRobin from www.rs-met.com wrote: it's not very practical for a realtime-synth, though. in a realtime-synth, you will typically have some kind of trade-off between between computational cost and the level of aliasing.
Frankly, I can't imagine how fast are modern FPUs - might be very realtimish for say 10-20 harmonics
Murderous duck!
-
- KVRian
- 877 posts since 24 Jan, 2006 from Universe #5346198720
nm
Last edited by dreamkeeper on Tue Nov 20, 2012 1:59 pm, edited 1 time in total.
The hole is deeper than the hum of its farts
- KVRAF
- 5233 posts since 25 Feb, 2008
Just to be clear, waveforms generated using a band-limiting method must necessarily exhibit some aliasing, therefore anyone claiming that their (advanced) band limited impulse train oscillator is completely alias free is telling porkies?
-
Music Engineer Music Engineer https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=15959
- KVRAF
- 4264 posts since 8 Mar, 2004 from Berlin, Germany
hmm - well, not necessarily. it is theoretically possible to obtain perfect results. but it is not always practical. additive synthesis was already mentioned as a means to generate perfectly alias-free waveforms. discrete-summation-formulas may also give you perfect results. it's only that both of those have other drawbacks: the former is computationally expensive, especially in the low registers and the latter has some problems with transients and/or DC offsets. the more practical methods like (Min)BLEP and mip-mapped wavetables will not give perfect results, but arbitrarily good approximations and the goodness of the approximation will depend on the computational cost, you are willing to spend. i found my own sweet spot somewhere, where the aliasing was pushed somewhere below 100 dB (where i can't see it anymore on my spectral analyzer ). a simple one-line change in the code could push it even further down, but i found it not to be worth the increased CPU usage.hakey wrote:Just to be clear, waveforms generated using a band-limiting method must necessarily exhibit some aliasing, therefore anyone claiming that their (advanced) band limited impulse train oscillator is completely alias free is telling porkies?
BTW.: what are porkies?
Last edited by Music Engineer on Tue Nov 20, 2012 2:16 pm, edited 1 time in total.
- u-he
- 28042 posts since 8 Aug, 2002 from Berlin
For a static sawtooth waveform, yes.david.beholder wrote:Wait a second, but what about representation as sum of sines: k1*sin(x) + k2*sin(x*2)+ ... + kn*sin(x*n) - it's perfect aliasing free osc, ain't it?Urs wrote:If I was spiteful, I'd have my lawyer send cease and desist letters to any company that claims to have "aliasing free oscillators" in their digital synths. Our lawyer could actually make a living from that.
But if you add vibrato or any more drastic kind of pitch modulation, then you would have to e.g. insert BLEPs or similar means on each change of derivate, e.g. per sample.
This isn't viable and it's also not necessary to reduce aliasing below perceptible levels, but to believe that any digital oscillator with a common feature set is "aliasing free" is naive.
Even worse, making oscillators "aliasing free" doesn't mean that filters etc. do not alias. This is usually met by oversampling. Which involves filtering that reduces aliasing, but can't eliminate it.
All digital synths I tested in the past year or two have relatively aliasing-free oscillators and quite audibly aliasing filters and/or waveshapers.