My shaper waveshaper has a tanh without aliasing. It’s not hard. In my experience maybe half the waveshaper have aliasing, half don’t.
As far as the VCO, as everyone already said, the basic waveforms are clean, the wave folder and FM are not.
I never noticed tanh aliasing until these posts had me looking for it. I actually bought it on the strength of my experiments in VCV, and it is in constant use.
I was and still am a bit obsessed with anlog oscs tubes etc. for sound design subtil saturations, timbres and harmonics. I instantly heard the aliasing thing which felt veeeery unpleasant to my ears. I tested some waveshapers, distorsions modules etc. and was surpsied by the unpleasent saturations but I’m also picky with analog modules so…
However now I’m curious to know which vcv rack 2 modules sound decent? I’m new to vcv rack so don’t know much about the insights of these modules. Im more interested in polyphonic but I’m curious to test some oscs to my ears 
Shaper, when set to “emitter coupled” is a tanh, you have a choice of no oversampling, 4X, or 16x. I’m old, and can’t hear the difference between 4 and 16, but I don’t like the aliasing of 1x.
Looking back now, I see the manual for shaper is pretty informative. It had some nice pictures of all the shapes, although it is a bit coy about “emitter coupled”. Yes, what it says there is true, but I handily omitted “it’s just the tanh shape that everyone uses for soft clipping”.
1 Like
I can’t tell you what “sounds decent”, but I do have a few things. One is that I have a code repo that has a lot of articles aimed at general nerds. The article on measuring aliasing shows how trivially easy it is to do with free VCV modules.
Here’s my repo - there is a table of contents with links to the articles. GitHub - squinkylabs/Demo: A collection of code and articles of interest to VCV users and developers.
My recommendation would be to test out some modules yourself, and see if for you “sounds good” correlates with aliasing at all.
As I mentioned, last time I measured the wave shapers I happened to have installed it was about 50/50 for whether the aliased a lot.
One I’ve looked at more than once is “West Coast” from @heapdump. When it first came out I looked at it and said “this thing is a joke! It aliases all over the place”. I promptly wrote Shaper.
But probably before I was done with Shaper West Coast had an update that got rid of the aliasing, and least in most of the models. So it’s a fun module in that it emulates some classic wavewhapsers, and sounds really good.
1 Like
Interesting without any testing from all the waveshapers I used I liked the “west coast” most. Thanks for the links and for the context. I now understand better what my ears were trying to tell me…
1 Like
Oh, thanks for the compliment 
I really like Shaper, it sounds pretty cool 
I fight again aliasing all the time; especially if you using non-linear elements like shaper, folder, distortion etc. because they always cause infinite harmonic’s resulting in aliasing when they exceed nyquist (this is always half of the samplerate). These harmonics get reflected back to the lower frequency range where we hear it as annoying interference. You could only lower that effect by oversampling and smart filtering, but you couldn’t remove it completely… sorry I started chattering, nevermind, have a nice weekend gentlemen! 
1 Like
Oh, for sure everything @heapdump says here is true. But, while technically correct, I don’t think “eliminate” vs. “reduce” matters in the real world. I think most of us DSP folks are pretty “ok” with garbage that is like 96 db below full scale. After all, back when we used to have round spinning things, CD audio (at 16 bits uncompressed) did have “junk” at -96db. True, people used dither and such to make that junk be closer noise than to breaking glass distortion, but still. If you look at almost any VCV module on a spectrum analyzer, and turn the vertical range to 96db, your are going to see some crap down there from almost any module.
So, people like to argue about whether humans can perceive artifacts that low. I think “golden ears” like to say you can hear it, and internet trolls like to say that’s ridiculous. I don’t actually have an opinion, or care, but I do think that going for anything better than -96db in VCV is pointless just given all the other modules, the converters, etc…
So, no, you can’t eliminate aliasing in non linear processing. But with oversampling and filters (even dumb ones), you can get that stuff down to the level where you can’t hear it.
Oh, and thanks for the compliment, @heapdump 
In honour of Bowie’s birthday yesterday
Golden (y)ears, Gold, whop whop whop
1 Like
Hey Bruce, I was re-reading Roger Reynold’s Mind Models (1975) and ran across this statement:
Foldover involves the reflection in the direction of lower pitch of any desired frequency exceeding one-half the sampling rate. As a result, requested higher harmonics may be reflected back down in such a way as to produce unwanted complexity or dissonance in the analog result. Naturally, this property can be advantageous.
Page 115, footnote. Italics his. The last statement is interesting, it reminds of me of a line from William Gibson’s novels:
The street finds its own uses for such things.
Me, I can’t find a place for aliasing in my own work, but I won’t rule out the possibility (a la Reynolds) that it could be a useful effect for a specific musical purpose.
Best regards,
dp
1 Like
I suspect what happened with the Instruo modules is that they were done by grad students who focused really heavily on the analog modeling. My impression was that they ended up incorporating something like SPICE circuit simulations to try to make them as accurate as possible and worked that into their masters thesis.
With such a focus on the analog circuit modeling, I can imagine they were running everything at 192khz, and completely forgot about aliasing issues.
That’s just my most polite guess based on hazy memories of the interviews I saw at the time.
2 Likes
that’s probably true. I will say a “pet peeve” is that tanh module. It’s literally just the C “tanh” function call. No hint of any serious modelling, and of course no alias reduction.
Regarding the instruo tanh, is there any chance that is what the physical module does? It sounds like they are fairly up front about it in the manual for the module:
Description The Instruō tanh[3] is a three channel waveshaper, limiter, overdrive, and feedback controller. Each channel of the tanh[3] takes a signal input and outputs the hyperbolic tangent function (tanh (x)) for that input signal. … In more musical terms, tangential distortion is a form of soft clipping which can produce signal amplitude limiting via analogue waveshaping. Driving a hot audio signal into it will add warm overdrive. Raw waveforms will be subtly shaped; softening or adding harmonic content based on input amplitude. Understanding what the tanh[3] does to a signal, both mathematically and sonically can be hard to explain, but it essentially adds curves to signals. For many Max/MSP and/or PD users, the [tanh~] object may be familiar and is likely a much used secret weapon in many contexts.
Are they up front about the aliasing? Because there are a lot of waveshapers that have the tanh shape, but don’t alias like that.
And, yes, tanh is the standard approximation for the large signal response of an emitter coupled pair of transistors. But of course the analog ones don’t alias.
Here’s a quote from the interview I was remembering when these were released:
Lim: Many of my designs are fully analogue. Some are hybrid, using microcontrollers to handle UI, controls, routing, and even calibration in some instances. The oscillator cores and all signal paths are analogue, other than the obvious DSP-based modules. I didn’t really know too much about the process or practicalities of analogue modelling so it’s something we explored as development got underway.
Much of my circuit work is derived from classic reference design schematics and from classic synth DIY publications.
So it’s my assumption that the physical tanh module is fully analog.
Ah, also this:
My friend and collaborator, Dr Sebastian Lexer (…) gave recommendation to me for a couple of Master’s students last year who were approaching their final years of study. Their program requires an industry work placement in which they would work on, document, and develop a project within the field of music technology. This was of course all put in place pre-COVID and their start dates were planned for mid-June 2020. As the time approached, we had to rather quickly shift gears and figure out a new project that would be better suited for remote working.
Their internship placements were able to go ahead despite the global pandemic. We had some good Zoom chats with Sebastian and myself. Exploring the possibilities of producing analogue models of maybe a couple of the Instruō modules seemed like an ideal task. We now had a new plan and everyone was onboard. I’d like to add that everything here was new to these guys. Not just digging into a new development environment, tool chains and getting acquainted with systems like GitHub, etc. I believe many elements of audio synthesis were new to them as well. They got started and I pretty much left them to it while maintaining my usual day to day operations in-house.
(bolded line was my addition)
So, anyway, my feeling it that it’s just something they all missed, likely half out of naiveté and half out of getting hyper focused on the analog-modeling.
It seems like fixing the aliasing would be a good project for an enthusiastic undergraduate intern.
2 Likes
Well, hm, later in the article he specifically talks about them working on anti-aliasing on the oscillator modules. But also says, ‘I pretty much left them to it’.
So, who knows, I guess!
2 Likes