Best approaches for an accurate four-quadrant multiplier for CV?

[jheronymo]

I’m prototyping some ideas that involve using a DAC to control a bipolar VCA / four-quadrant multiplier.

The analog inputs can be assumed to be CV, although audio range is a bonus.

Ideally, I’d like for the circuit to be accurate enough that “unity” or “negative unity” from the (16-bit) DAC is precise enough to scale pitch CV without tuning issues. Accurate linear response around 0V is also a concern.

Also, I might want to put a bajillion of these VCA channels into a module, so parts cost is a factor.

My question is what type of VCA/RM circuits would people recommend I look into? 2164? 3360? Discrete?

So far I have been checking out:

Mutable Frames - (not bipolar) DAC-controlled 2164 circuit which uses a logarithmic lookup table on the DAC

Mutable Blinds - 2164-based four-quadrant multiplier. Each channel uses two VCAs: one to linearize and one to multiply. Seems like there’s a compromise since whichever input is linearized takes a hit in noise around 0V, and Émilie’s solution is to linearize the “input” as the VC to the 2164 multiplier, and offset/subtract the “CV”.

MOTM-110 - uses the CEM3330, and the MOTM-190 with CA3280. I am eyeing the AS3364 - it’s like the AS/CEM3360 but quad linear, specs seem good enough

Discrete VCA - a la Rene Schmitz ? Who has an RM based on a matched differential pair. Studio H seemed to do something like this on the well-received and pitch-accurate CSR with THAT300 and high-performance op-amps. But that’s Buchla CV, so strictly unipolar.

I am confident on the programming/MCU side, but less so on this analog stuff (I’m trying to change that). I’ll take a deep dive into schematics and simulations, but if anyone here has opinions on what is a good idea and what isn’t, I’d be grateful to hear it as it would save me a lot of (limited) time in testing a bunch of circuits/chips!

[ropesalesman]

I've been impressed by the SSI2164 in a CV-accurate VCA that I designed but I have not tried it in a 4 quadrant multiplier for CV yet. It worked well for audio in the 4Q configuration. I'd at least explore that route if I was looking to design a CV-accurate 4Q, I can't think of any "gotchas" off-hand.

[metamorphmuses]

I'm going to dip my toe in, even though I haven't the slightest clue about chips or circuit design. My only 'qualification' is that I like bipolar VCAs; I generally lean toward bipolar VCAs over strictly unipolar ones.

Make Noise modDemix — one of the best-sounding, best responding bipolar VCAs / ring modulators I have and have used.

Befaco A*B+C — I had one and sold it off because it was one of the batches afflicted with unwanted noise. Otherwise, I'd value it as highly as the modDemix.

Intellijel μMod II — now out of production, but I got one anyway and it's a keeper.

Intellijel Planar 2 — this is a joystick, but it's more than that; at its heart is bipolar voltage control over four quadrant cross-fading / mixing.

Instruō vincâ — this module features two VCAs, but only one of the two VCAs in each unit is bipolar. That's why I have two vincâs side by side.

[jheronymo]

Awesome, thanks. I recently ordered a small batch of them because if nothing else they have a good reputation, are reasonably inexpensive and available at medium-sized bulk discounts. And I figure that, depending on DAC precision, I could get away with linearizing in code.

Thank you! Part of the reason I’m asking here is that I don’t have the funds or time to collect a bunch of bipolar VCAs to see what works best! I’ll do some hunting on these to see if I can glean what type of VCA circuits they’re using.

[guest]

have you considered an MDAC (multiplying dac) - it can do both steps at once.

[fitzgreyve]

AD633 ?

[dslamnig]

AD633 ?

Here's my take on the AD633 4-quadrant multiplier:

I guess this is as simple as it gets. I decided to drop the Z summing input. I've built this as a one-off prototype, works fine as a VCA and ring modulator and everything in between. If I had to do it again, I'd probably make one or both bias pots "thru zero" (connecting the lower end of the pot to -12V instead of ground), because when using it as a "normal" VCA there is a tiny leak due to offset errors.

Note - I initially bought a couple of cheap AD633s on eBay and got fakes. The genuine chip I bought later cost me about $15 - if they cost less, beware.

[al2o3cr]

+1 for considering a multiplying DAC; one benefit of that approach is that those parts have explicit specifications for linearity.

As precision parts they aren't cheap though - for instance, if you can find a DAC8801 you'll probably pay $12 apiece.

[KSS]

AD633 or MDAC were my first thoughts too.

[jheronymo]

Thanks for the replies everyone.

I hadn't heard of an MDAC! That is essentially what I had in mind. They seem to have a pretty straightfoward bipolar application with the addition of a single op-amp, which is great.

Unfortunately, this is where my cost restriction comes in... I'm aiming for 64 VCAs in one module

It looks like a good 14-/16-bit MDAC (ex. AD5554 /AD5544, LTC2754) is about 3-4x the cost of a 14-/16-bit voltage-output DAC (AD5360, LTC2688) per channel. And the MDACs generally max out at quad packages, whereas voltage-output are easily available in 16- or 32-channel packages.

I'm not trying to make a commercially-viable module or anything... but $800 of MDACs alone is probably not going to be doable.

And that puts the AD633 beyond the pale price-wise too :( as much as I'd like to use them.

[imrae]

I’ve got an AS3363 chip in my “to play with on breadboard” box, which provides a couple of multiplier + VCA channels and a crossfader (which could be turned into a multiplier by multing the input via an inverting amplifier?)

Any opinions on this one? Is it ok for both CV and audio?

https://www.alfarzpp.lv/eng/sc/AS3363.php

[al2o3cr]

Another possibility (though less accurate) could be digital potentiometers.

For instance, the AD5144 has 4x "pots" controlled via serial data. Reasonably inexpensive (Digikey shows them at $8.43 each, so about $2 per pot) Downside: the resolution is much lower (8 to 10 bits) compared to actual DACs.

If you need a LOT of VCAs but don't want to spend a lot, consider applying an old telecommunications trick and time-sharing the parts.

For instance, let's say you wanted to make an 8x8 matrix mixer with every control point under voltage control. A straightforward approach will need 64 VCAs / MDACs. An alternative approach would be to use 8 VCAs / MDACs (one for each input) and sum together their output, then sample & hold that value to each of 8 outputs in succession at a high rate.

[jheronymo]

Yeah I’m afraid I’ve given up on digital pots for not being accurate enough.

For instance, let's say you wanted to make an 8x8 matrix mixer with every control point under voltage control.

Bingo! I mean… wait, what? No, my prototype idea is far more obscure and mysterious than that!

A straightforward approach will need 64 VCAs / MDACs. An alternative approach would be to use 8 VCAs / MDACs (one for each input) and sum together their output, then sample & hold that value to each of 8 outputs in succession at a high rate.

Interesting. I’ll look into this.

To make sure I understand what you’re describing — 8 sample and holds, i.e., one per output?

This was my original plan, and my thoughts were:
(a) needs a lot sample & holds (however I thought—erroneously, I believe— that I’d need one S&H per node, not per output)
(b) introduces more work to control noise
(c) a good enough VCA could handle audio and CV, whereas trying to implement this S&H approach with audio seems very hard or impossible

[KSS]

There was a 2U rack unit that did this back in the day. Syn-something <--Of course it's Syn- How original
Had 8x8 matrix LEDs on the panel, some cursor control.

I'm sure it's on Matrixsynth somewhere.

[Mungo]

For instance, let's say you wanted to make an 8x8 matrix mixer with every control point under voltage control.

Bingo! I mean… wait, what? No, my prototype idea is far more obscure and mysterious than that!

64x "CV" DACs + multipliers
vs
8 ADCs + 8 DACs and a DSP/Microcontroller

Somethings are just better done in digital.

[jheronymo]

Hmm, yeah Mungo you may be right. Maybe it’s just got to be ADCs. I never looked into DC precision with ADCs.

Although 8 MDACs, a demux, and 8 S&Hs is not terrible either…

What I’d love is to be able to scale and mix complex CV shapes and pitch controls accurately. And throw a +5V or +10V reference on each (switched) input jack.

Well this hasn’t helped narrow down choices but it’s almost surely going to lead to a better overall outcome.

Thanks everyone for the input so far!

[wsy]

Yeah, if you want a full 8x8 crossbar four-quadrant mixer...... do it digitally. Parts count will totally kill you otherwise.

[jheronymo]

Yeah… the more I think about it, full digital seems like the way to go. Hell, I’d love to go 8x16 or 16x16 and this is suddenly seeming possible.

[KSS]

Found the one I couldn't remember yesterday. ASol Synapse.
16x16 crosspoint switcher only with MIDI. No VCAs.

I've attached the manual also.

[jheronymo]

Found the one I couldn't remember yesterday. ASol Synapse.
16x16 crosspoint switcher only with MIDI. No VCAs.

Aha! I was having a hard time finding anything.

These crosspoint-type modules are still around: Alyseum Matrix II and Erica Desktop Matrix Mixer, which I’d bet are based on the AD75019. The Erica has 3 gain settings (0.3x, 0.7x. 1x). And the SSSR Labs SM010 Matrixarchate.

I’m interested in making a hardware platform for animating the gain of these nodes in ways that would be hard/expensive to do with analog utilities, so these switched designs are a bit orthogonal to what I’m aiming for.

I do have some SSI2164 and AS3364 on the way, I’m still going to test them out since both are ~$1.00-1.50 per VCA which is feasible for an 8x8 grid I figure. The DACs and buffers will be the limiting factor at that point. If either works well, I will write an update here.

[KSS]

There was a guy here a few years ago went by dikkietron or something like that. He was developing a substantial system that became the Anatol XBay.
Near constant derision from Hinton. He had a really good tech from NL in his corner and it seems that the system does deliver. I have no personal experience but it seems like something you would want to look up and into.

Here's one of the early PCB designs that wer intended to fit in memory module connectors.

[jheronymo]

this, I’m guessing :
https://www.modwiggler.com/forum/viewtopic.php?t=170006
https://www.modwiggler.com/forum/viewtopic.php?t=191230

Well that’s some wild shit that probably would have made Don Buchla proud (or equally likely made him cringe as Don himself seemed to take the digital route after a certain point).

I have no such grandiose schemes, I just want a morphing 2-D sequencer/mixer with presets and nobody that I’m aware of has built a convincing one yet.

I abandoned the idea of the AD75019 in favor of attempting to use good enough VCAs that in the worst case (1 “on”, 7 “off”) shouldn’t generate enough noise/offset to be audible or affect pitch accuracy. Whether that’s feasible is still a mystery (to me).

Sadly I guess Hinton isn’t around MW anymore to deride this thread!

[cygmu]

I have no such grandiose schemes, I just want a morphing 2-D sequencer/mixer with presets and nobody that I’m aware of has built a convincing one yet.

I am not sure I 100% understand the target you have in mind but I like the thoughts this is giving me

Do you need 4QM for this? It seems to me that two quadrant would be enough: bipolar signals at each node but unipolar control over them (0 to unity gain)? I probably misunderstood.

[jheronymo]

I have no such grandiose schemes, I just want a morphing 2-D sequencer/mixer with presets and nobody that I’m aware of has built a convincing one yet.

I am not sure I 100% understand the target you have in mind but I like the thoughts this is giving me

Do you need 4QM for this? It seems to me that two quadrant would be enough: bipolar signals at each node but unipolar control over them (0 to unity gain)? I probably misunderstood.

Yeah, me too! That's why I want to build one. I'm just hoping I can pull it off to a degree that I could confidently open-source it or something.

The benefit of bipolar control is being able to invert the signal at each node. Is it strictly necessary? No. But very useful I think, both for CV and audio. It could possibly be done some other way than a 4QM, but I'm wary about introducing a bunch of switches, for instance. And having a smooth transition through 0V would be great.

Anyways, I don't want to make this a thread about my hotshot vaporware ideas... But if it helps clarify this problem, that's the plan.

I'm probably going to try analog (starting small) and digital implementations. Other than the drivers, the code should remain largely the same in either case which is nice.

[devinw1]

I gotta agree with doing it digital here if you can. MDAC is good for this.

If you had to do it pure analog, the least parts count for an accurate design which doens't need a bunch of external trimmers would be 64X AD633 but even in bulk you'd be looking at about $600 in AD chips alone. You could do it with SSI264 but that's gonna have a lot of parts. 64 quad op amps, each 2164 has 4x RC networks hanging off inputs, etc.. SO MANY PARTS. But, it would be epic. Lol.