Aliasing

[briandc]

Hi all,
I've seen where some softsynths have "anti-aliasing" oscillators. --But isn't the filter key follow control supposed to work to reduce/eliminate aliasing in the first place?

brian

[Jinsai]

Having the filter roll off doesn't reduce aliasing. It reduces high end, which can camouflage aliasing.

Key follow will just make the filter cut-off track with keys. That actually won't help hide aliasing in most cases, unless you have the filter rolled down.

There's no reason to accept audible aliasing in any synth these days, unless it's part of the "vintage character" you're looking for. Generally it's undesirable.

[briandc]

I guess that makes sense. So then the aliasing itself always comes from the sound source (oscillators)? Is the aliasing due to having too much high-end signal?


brian

[Big Gnome]

Often, but it can also be caused by any process that generates harmonics--distortion, ring modulation, etc. Aliasing occurs when a signal contains partials that exceed 1/2 the sample rate, so yes, it's related to high-frequency content.

[briandc]

Ok. Do applications exist that can detect/indicate if the audio signal I'm working with contains any aliasing? (other than just what's obvious with my ears, of course... )


brian

[madtheory]

A spectrum analyser is what you need to observe and measure. It will usually stick out in a obvious way. Voxengo make a very good one. There's a long running thread on gearslutz testing aliasing in plugins, great info there.

This is an excellent brief explanation of sampling. You'd need to understand that before you understand aliasing- they go hand in hand really.

It's not quite true what was said about the filter above. An anti-aliasing filter is a standard feature of a DAC and an ADC. In software (i.e. between the converters), it gets complicated because everything is a calculation, including the oscillator and the filter. If you get the calculations wrong, aliasing is often a side-effect. Basically, it's what happens if you violate the sampling theorem. You are throwing information away by being innaccurate. So any part of your virtual synth can be the cause of aliasing, because each "part" is just another calculation that can go wrong if the programmer is not careful.

Another complication is that almost all modern converter chips have very gentle filter slopes, so they're a bit rubbish when it comes to aliasing! Just look up the data sheets and you will see. A brute force solution is to use a high sampling rate like 96kHz, but that's a waste of bandwidth and uses up DSP cycles. Better just to preserve information in the first place:)

[knolan]

Digital PWM also suffers from Aliasing and is a chronic problem the Jupiter 80, 50, Integra and JD-XA; rendering PWM sounds on all of those models essentially useless above C4. I own an Jupiter 80 and it is truly dire - the aliasing is about as lould as the the PWM wave form from C4 up, meaning it just can be used.

By contrast, some synthesisers handle it very well. The Yamaha AN1x has very low aliasing, meaning that it sounds very 'natural' even at high frequencies (though it does start to kick in at the highest notes if you do any sort of modulation on the oscillator). But it's exceedingly good for an 1998 synthesizer, and contributes to the continued reputation of the AN1x as a really excellent VA.

Korg's OASYS and Kronos claim low-aliasign oscillators and indeed I own OASYS and find its oscillators on the the AL-1, Poly6, MS20, MOD7 and STR-1 synth engines to be very good in this department, even at the highest frequencies. I believe the KingKorg also claims to have low-aliasing oscillators.

[madtheory]

Yes this is a good example of how it's not just about raw processing power, but how well the thing has been programmed. PWM generates a lot of extra harmonics, so it seems to be quite challenging to do digitally. Some designers do it better than others.

Above C5, my 2002 vintage Novation KSR starts to exhibit aliasing. It gets gradually worse and is really bad at C6. This is not perfect, but it's pretty good for a VA of its era, and I don't use pitches that high very often so it's not a deal breaker for me

Then again, an extreme test like this would indicate that it is more likely that compromises have been made in other areas that are more subtle. So you might find weird stuff happening in your mix, or to your speakers, when you least expect it.

Check out that gearslutz thread, it gives you a lot of insider info on the realities of making audio and music making tools in the digital domain.

[briandc]

Thanks for the tips, madtheory. I've been looking at a thread on Gearslutz about aliasing-- certainly lots of info there. Probably more than what I'm in need of, as my consideration is more of a "performance level" interest more than a technical level.
But a spectrum analyser is a good idea and I have a native one that should do the trick, so I'll give that a shot. What you said about calculations being made with various elements, and the resulting possible errors and consequent aliasing, is another thing to keep in mind. I hadn't considered distortion and ring modulation as examples, but now I'm starting to get the overall picture, I think.

And without a doubt, some synths handle the aliasing phenomenon better than others. So I'll just keep notes on which instruments I come across deal with aliasing better than others, for the record, so to speak.


brian

[madtheory]

Those errors can be nice too! For example, it's what folk like about different samplers. The transposing especially.