Author |
Message |
Rojo
Joined: May 10, 2005 Posts: 14 Location: Boston
|
Posted: Wed Aug 05, 2009 8:55 am Post subject:
Pacarana aliasing at high pitches? |
|
|
I'd like to know, please, how well the Pacarana can make conventional synthesizer sounds at high pitches without aliasing.
The test of this would be something like this: take a conventional patch based on a sawtooth wave and a lowpass filter, play a very high note, then add a little pitch bend to make any aliasing more evident. How high can the base note go before you hear aliasing?
In my experience with VAs (I haven't tried anything newer than the Nord G2), there will be a couple octaves or more where the base note is audible, but aliasing is audible too.
If you know of some digital synthesizer or system other than the Pacarana that's good for this scenario, I'd be interested to know about that too. Even if it's one of the systems that does its DSP on a general-purpose computer.
Thanks. |
|
Back to top
|
|
|
pdj
Joined: Jun 02, 2009 Posts: 4 Location: UK
|
Posted: Thu Aug 06, 2009 12:57 pm Post subject:
|
|
|
Hi Rojo
The reason you get aliasing when generating repeating tones in the digital domain is because if the pitch you are trying to generate doesn't match a sub harmonic of the sample frequency you get beating.
For example if your sample rate was 48K and you wanted to make a square wave at 1 khz then you would have 24 high level samples followed by 24 low level samples. This would sound perfect. but if the frequency was 1.1Khz the you would have to have some cycles with 24 highs but from time to time you would have a cycle with only 23 to get the pitch right. This gives the aliasing.
But what about an analogue synth recorded on a CD? This sounds fine. the reason is that there is an anti aliasing lo pass filter before the A to D convertor which makes a signal that smears the 23/24 borders.
The problem is that you cannot put a low pass filter on a internally generated oscillator because the aliasing is already part of the audio and in the same or even lower frequencies as the oscillator fundamental pitch. It's too late for any filter to have any cleansing effect.
Kyma has at least three different method to get around the problem Ranging from pre filtered wavetables, additive synthesis and the Surbiton oscillator.
These can be processor intensive but the Pacarana has so much processing power that you can have plenty of these non aliasing oscillators and still have loads of power left.
Hope this makes sense
Pete |
|
Back to top
|
|
|
DrJustice
Joined: Sep 13, 2004 Posts: 2114 Location: Morokulien
Audio files: 4
|
Posted: Thu Aug 06, 2009 6:16 pm Post subject:
|
|
|
pdj wrote: | The reason you get aliasing when generating repeating tones in the digital domain is because if the pitch you are trying to generate doesn't match a sub harmonic of the sample frequency you get beating.
For example if your sample rate was 48K and you wanted to make a square wave at 1 khz then you would have 24 high level samples followed by 24 low level samples. This would sound perfect. but if the frequency was 1.1Khz the you would have to have some cycles with 24 highs but from time to time you would have a cycle with only 23 to get the pitch right. This gives the aliasing.
But what about an analogue synth recorded on a CD? This sounds fine. the reason is that there is an anti aliasing lo pass filter before the A to D convertor which makes a signal that smears the 23/24 borders.
... |
pdj, aliasing occurs when a sampled signal has frequency content above 1/2 the sampling frequency. It has nothing to with the lack of symmetry you are describing.
Overly simply put, if the signal changes too fast for the sampling frequency used, you will get artefacts in the form of additional frequency content below 1/2 of the sampling frequency. A square wave like you describe, consisting of max -ve samples followed by max +ve samples will have too fast changes (actually to fast lack of change in that you cant have perfectly square edges, they have to wiggle a bit and settle on the max -ve and +ve values).
That's not a very good description - Unfortunately I'm too tired to type out a longer explanation in layman's terms right now, but there are plenty of references on the net. Otherwise a search for e.g. "sampling theorem" will give you a concise description of the aliasing phenomenon.
DJ
Edit: typo
-- Last edited by DrJustice on Fri Aug 07, 2009 2:29 am; edited 1 time in total |
|
Back to top
|
|
|
BobTheDog
Joined: Feb 28, 2005 Posts: 4044 Location: England
Audio files: 32
G2 patch files: 15
|
Posted: Fri Aug 07, 2009 2:21 am Post subject:
|
|
|
P.S. Are you Peter of Surbiton Oscillator fame? |
|
Back to top
|
|
|
pdj
Joined: Jun 02, 2009 Posts: 4 Location: UK
|
Posted: Fri Aug 07, 2009 12:32 pm Post subject:
|
|
|
Hi Bob
Yes the Surbiton Oscillator is one of mine.
Hi DJ
I often have to point out that just because you don't look at something in a convectional way, it does not make it wrong, and often things are really a lot simpler if you look at it from a different view point.
We tend to think of signals as a collection of sine waves for convenience but it's not the only way to analyze waveforms. You can think of a square wave as a collection of sine waves (harmonics) in this case odd harmonics only. It is true what you say in as much as any harmonics that fall above the half sample rate will produce aliases.
It just so happens that the aliases it produces are like a mirror image in the frequency domain with the mirror placed at half sample rate. There is also a second mirror at DC (0Hz) as well. If the square wave is exactly a sub division of the sample rate all the reflected harmonics fall exactly on the true harmonics that were already there and you will hear a clean repeating wave form.
If you really feel inclined you can calculate where all the aliased harmonics lay when the square wave is not a sub division of the sample rate and you will find that the aliased harmonics will make a de-tuned harmonic series with a specific beat frequency. It's no coincidence that this frequency is the same as the beat frequency of the original square wave and the sample rate. So you will get the same beating frequency content wether you calculate it the simple way or the complicated way.
I do agree that if the signals we were working with were more complex than a square wave and if we wanted more detailed info of the harmonic content of the aliasing you would be better using the conventional sine wave analysis.
I hope this makes sense
Pete |
|
Back to top
|
|
|
|