Serum Superiority #1 - Sub Fundamental Aliasing and Non-Harmonic Distortion

[DorianDay]

Hey, I've been thinking about doing a series for a while that looks a deeper technical aspects of synthesizers. With the release of Serum I finally found the right angle, it seemed to successfully add all the features I appreciate in the most popular synths while adding something to the table.

Here's my pseudo-technical analysis of Serum and some of the most popular synthesizers around. I did a sound test, checked the waveforms for asymmetry and assessed Sub Fundamental Aliasing and Non-Harmonic Distortion.

If anyone knows better about this stuff and can enlighten my layman analysis please let me know.

Also I started an IRC channel for serum where I sit while I mess around and hope to build a small community there.

server: irc.mozilla.org
channel: #serum

https://www.youtube.com/watch?v=MtvpAxTe75E

[lnikj]

Interesting comparative analysis. Thanks for this.

At a guess, are we not seeing the differences between analog modelling and purely digital synthesis?

Looking at Zebra, which I think uses a similar approach to Serum, we see a relative lack of harmonic distortion, but Diva, which is modelled on analog synths, creates a lot of non-harmonic distortion which is what, I guess, is giving its oscillators 'character'.

Neither good nor bad, just different?

[DorianDay]

Yeah, I mean based on an ear test I think all those synths sound good and very similar. That's sort of my motivation for this assessment, I don't personally see hear an significant difference between the waves, so what is the end result/cost of this extra potentially unnecessary harmonic information. When you consider down the line processing, especially distortion, the extra harmonic content could have quite an impact on the over all sound as well as potentially increasing aliasing and the non-harmonic distortion.

Perhaps my ears aren't as well trained as they could be, and I'd love to hear from someone with more knowledge on this subject because it seems like the extra information is more likely to cause harm than good. I don't know, maybe I'm over thinking it.

[lnikj]

Again, hazarding a guess, would any audible differences not be more likely to show up at higher frequencies (approaching Nyquist)?

Regarding the knock on effects of further processing, you would have to test for this (obviously using an effects plugin not any onboard effects), but this could again be seen as a difference between digital 'purity' and analog 'character', no?

Maybe I am misunderstanding your motivation though?

[DorianDay]

The motivation is mostly curiosity, I'm mostly trying to pose questions that perhaps someone with more knowledge could shed some light on. I don't really know what I'm asking or looking at or if I'm even using the wrong words but I haven't seen the discussion around and I'm attempting to start it. I could imagine this stuff causing problems (dissonance, harshness in the high end, problems for lower quality plugins who don't have adequate over sampling or something) but maybe I don't know enough, as I have no familiarity with DSP.

So hopefully the conversation continues.

[bitcrusher]

The correct term is aliasing / nyquist reflections. http://www.earlevel.com/main/1996/10/20 ... -aliasing/

The high end is usually more of a problem because the aliasing is louder there, but often are usually masked (often at least somewhat) by louder intended frequencies, so often you don't perceive them. With that said you can very often hear the clarity difference when they are removed.

and a comprehensive "shootout" among host samplerate conversions here:
http://src.infinitewave.ca/

The more important thing watching your video briefly is indeed the high end, both if there are reflections and at what level, and also if the frequency response is indeed "flat" to 20 khz (many synths will band-limit to minimize the amount of nyquist reflection, but this often has a filtered-highs consequence in the audible spectrum).

-Steve

[bmrzycki]

I recommend watching Monty's excellent digital media primer video:
http://xiph.org/video/vid1.shtml
at about 11:18 he digs into aliasing.

Digital noise makers (oscillators, filters, effects) have to combat aliasing with special filters. These filters are hidden from the end user. They consume CPU so it's often a trade-off of how effective these filters are vs how much CPU they use. Older VA synths tend to alias more simply because the developers didn't have the CPU power we take for granted today to remove the unwanted noise. This is what oversampling is meant to eliminate: sample a bigger frequency domain so that aliasing folds back far,far above 22KHz and is safely removed from the signal when band-limiting takes place.

It's important to note: once digital aliasing makes it into the audible spectrum of your signal there is no way to easily remove it. You can't high-pass or low-pass the signal to remove all of it without also removing some of the desired signal. This is why it's so important to get it gone in the first place.

Tired now, will post more tomorrow (unless one of the super-smart devs around here beat me to it)

[Fred_Abstract]

do you use oversampling on strobe? at max?

[tony tony chopper]

If anyone knows better about this stuff and can enlighten my layman analysis please let me know.

I see you looking at the spectrogram at the "wrong" scale, -96dB is already not in the realm of audible anymore.
& as pointed out, it's at the end of the spectrum that you're gonna find the most audible aliasing.

You also say "if a signal is clean, there shouldn't be any information in-between the harmonics", but I hope that you realize that the "bulges" that you're seeing in-between harmonics, are unprecisions of the (FFT-based) spectrogram (& the windowing selected). There are ways to improve spectrogram precision, & thus better out there, for this purpose.


You're gonna get the purest sawtooth out of an additive synthesizer, that's what you can use to calibrate your monitoring.
An additive synthesizer will also give you the purest phases, something that you can't get in the time domain. This said, nothing of this is really audible & pointless to argue about, but really, try an additive synthesizer if you wanna hear what a pure sawtooth sounds like.
Time domain synthesis still has the advantage of being a lot more CPU efficient for very low tones, though.


Here's heavy aliasing vs no audible one, on a 120dB range.
https://drive.google.com/file/d/0B6Cr7w ... sp=sharing

[Ingonator]

Concerning Aliasing:

Waldorf PPG Wave 3.V actually contains an emulation (!!) of the aliasing found in the original PPG Wave 2.2/2.3 synths and that emulation actually eats some additional CPU amount.
The different modes (e.g. 2.2 and 2.3) are avaiable from the TruePPG modes. It is also possible to swicth to 2.V (= old PPG 2.V synth" or switch True PPG off for a more clean sound. Those True PPG modes also include other changes to the sound like e.g. a reduced sample quality (8-bit for the 2.2 mode and 12-bit for the 2.3 mode and full quality with TruePPG off).
This emulation is done to replicate the character of the original synth which in this specific case many people seem to prefer over a 100% "clean" sound.

This does not mean i generally like aliasing. In a good analog emulation i usually prefer to have aliasing free oscillators too.

[DorianDay]

Hey guys, thanks for all the responses. I made a follow up video addressing some of the comments I've gotten with an updated test trying to reflect the suggestions others have made.

https://www.youtube.com/watch?v=l18HOo7vivg