Using cascaded VCAs in compressor designs?

[mr coffee]

Has anyone here used analog cascaded current mode VCAs and felt they improved s/n ratios over paralleling VCAs (or in combination)?

I have been considering tinkering around with such a topology using modern VCAs. I did try it eons ago with SSM 2020s in a feedback configuration and it seemed to provide some improvement, but my skills then were pretty limited as was my instrumentation <LOL>

Thanks in advance for any wisdom you can share.

 

[JohnRoberts]

Not sure what you mean by cascaded? Do you mean in series?

Paralleling old school VCAs have been known to provide a couple dB improvement, not unlike paralleling discrete transistors, But there are diminishing returns.

Depending on what you are trying to accomplish I have experimented with alternate topologies, like subtracting the VCA current from a unity path for zero noise/distortion at unity gain. I've seen this used to make clean limiters.

Modern VCA technology is quite good, and multiplying audio in the digital domain is even better.

JR

 

[mr coffee]

Hi JR,
Thanks for reply. Yes, I mean in series.

Current in to VCA1>Current out from VCA1>Current in to VCA2>Current out from VCA2>

I used it before in the feedback loop of an op amp, so the input to the first VCA was always at a good level for s/n.

The goal is for a time-variable-gain, always-above-threshold limiter and swell for a dedicated electric guitar with very low noise, low impedance pickups.

The idea is to keep noise and distortion down as the guitar signal decays. Clearer?

 

[user 37518]

Not sure what you mean by cascaded? Do you mean in series?

Paralleling old school VCAs have been known to provide a couple dB improvement, not unlike paralleling discrete transistors, But there are diminishing returns.

Depending on what you are trying to accomplish I have experimented with alternate topologies, like subtracting the VCA current from a unity path for zero noise/distortion at unity gain. I've seen this used to make clean limiters.

Modern VCA technology is quite good, and multiplying audio in the digital domain is even better.

JR

Just a small intervention, and I don't want to sound like an ass, but technically speaking cascading is not the same as series. In two port theory cascading is one thing, paralleling is another, and series is yet another thing. From the following diagrams, cascading is indeed what the OP is asking, and the term is the most correct.

From Wikipedia,

Series:

Cascade:

 

[JohnRoberts]

I do not expect any benefit from putting them in series

JR

PS; I do not understand what Dualflip is suggesting... most VCAs I am familiar with have a single current input and single current output.

 

[user 37518]

I do not expect any benefit from putting them in series

JR

PS; I do not understand what Dualflip is suggesting... most VCAs I am familiar with have a single current input and single current output.

I am just being a snob about the proper term for the configuration you guys are talking about, when the output of a two port system is connected to the input of another two port system like you described, it is called a "cascade" connection (Cascade connection - Wikipedia), not a series connection, regardless of whether it is a current, voltage, etc... One terminal can be considered the current output and the other ground, it is still a two port network.

 

[JohnRoberts]

I guess I would need to see a schematic with "cascaded" VCAs to understand...

Tag you can help the OP from here.

JR

 

[user 37518]

I guess I would need to see a schematic with "cascaded" VCAs to understand...

Tag you can help the OP from here.

JR

Sorry, John. I wasn't trying to upset you.

 

[JohnRoberts]

Sorry, John. I wasn't trying to upset you.

You didn't upset me, but I am confused about how to cascade two VCAs for the OP. based on your images...

I have never used them that way, or seen it done. (I have seen VCAs paralleled).

JR

 

[abbey road d enfer]

I agree with John, I can't even envision how cascading two or more VCA's could improve performance significantly, although operating a VCA in substracting mode has a potential noise and distortion advantage, particularly with relatively low GR. However, the difference becomes moot when GR exceeds 6dB.
There may be a case to be made for a structure where 10dB of GR can be achieved with 10 VCA stages operating at 1dB GR or 5 operating at 2dB GR vs. 1 stage operating at 10dB GR. I'm not crazy enough to prototype that.
Parallelling VCA's is a well-known, well documented and efficient implementation.

 

[mr coffee]

Talking about 2 VCAs connected out-to-in so both work within their optimum range at high attenuation 40+ db.

Put the cascade in the feedback loop of a low noise amplifier. The first VCA in the feedback loop sees a high amplitude input for good s/n performance, the second VCA sees a lower input amplitude, but it attenuates the distortion along with the signal. And since each VCA is inverting, the control voltage feed through tends to cancel each other out, i.e., the first pedestal is sorta cancelled by the second VCA pedestal.

The intended outcome is, with a parallel ‘dry’ signal, a swell or ‘negative compression’ effect to sustain the guitar signal with decent s/n and low distortion and fast gain reduction’pop’.

Clearer?

Hawksford did something with cascading VCAs IIRC way back when, though not in a feedback loop.

The control voltage sensitivity is different from the normal deci-linear relationship, but I’m not going for that. The control voltage is more a time-variable-gain in this application and is shaped for the desired character of swell/sustain.

Sound a little less weird ?

 

[abbey road d enfer]

Talking about 2 VCAs connected out-to-in so both work within their optimum range at high attenuation 40+ db.

Why would -40dB be the optimum range?

Put the cascade in the feedback loop of a low noise amplifier. The first VCA in the feedback loop sees a high amplitude input for good s/n performance, the second VCA sees a lower input amplitude, but it attenuates the distortion along with the signal. And since each VCA is inverting, the control voltage feed through tends to cancel each other out, i.e., the first pedestal is sorta cancelled by the second VCA pedestal.

IIUC the resulting signal would be fedback to an input amp. Having both VCA's at -40dB results in a gain drop of 100ppm or 0.008dB. What's the point?

The intended outcome is, with a parallel ‘dry’ signal, a swell or ‘negative compression’ effect to sustain the guitar signal with decent s/n and low distortion and fast gain reduction’pop’.

Clearer?

Not really. A block dgm would help here.

Hawksford did something with cascading VCAs IIRC way back when, though not in a feedback loop.

Do you have a link to his experiments?

The control voltage sensitivity is different from the normal deci-linear relationship, but I’m not going for that. The control voltage is more a time-variable-gain in this application and is shaped for the desired character of swell/sustain.

Sound a little less weird ?

Not weird but still puzzling.

 

[JohnRoberts]

Talking about 2 VCAs connected out-to-in so both work within their optimum range at high attenuation 40+ db.

What metric are you considering to determine optimum? Current ratioing VCAs (like DBX/THAT) will attenuate their self noise along with signal.

Put the cascade in the feedback loop of a low noise amplifier. The first VCA in the feedback loop sees a high amplitude input for good s/n performance, the second VCA sees a lower input amplitude, but it attenuates the distortion along with the signal.

Putting single VCAs in the negative feedback path is a common limiter topology. Back in the day working with inferior gain control elements the topology reduced the signal level the gain cell had to handle by the amount of gain reduction. This topology allowed inexpensive tape noise reduction compandors to deliver impressive dynamic range.

And since each VCA is inverting, the control voltage feed through tends to cancel each other out, i.e., the first pedestal is sorta cancelled by the second VCA pedestal.

I don't know that control voltage feed through is much of an issue with premium modern VCAs. It was a huge concern with OTA and cheap IC gain cells (like NE570 series).

The intended outcome is, with a parallel ‘dry’ signal, a swell or ‘negative compression’ effect to sustain the guitar signal with decent s/n and low distortion and fast gain reduction’pop’.

negative compression (expansion?) could be accomplished with side chain configuration.

Clearer?

no

Hawksford did something with cascading VCAs IIRC way back when, though not in a feedback loop.

IIRC Hawksford published work with distortion cancellation mechanisms related to power amps. I don't recall him messing with VCAs, perhaps applying a similar technique.

The control voltage sensitivity is different from the normal deci-linear relationship, but I’m not going for that. The control voltage is more a time-variable-gain in this application and is shaped for the desired character of swell/sustain.

most dynamic processors use time varying side chain circuits.

Sound a little less weird ?

EFX design opens up the field to whatever works... If it sounds good it is good.

Still not clear what exactly you are proposing from a hardware perspective.

JR

 

[mr coffee]

Sorry I’m on the road and can’t draw a diagram now. Tonight or tomorrow.

Here’s a link to Hawksfords paper. It is number 18 in the collection.

https://www.researchgate.net/profil...638/Audio-Amplifier-Systems-A-Compilation.pdf
The output of the second current out VCA goes to the inverting input of a low noise amplifier with a resistor to ground so it functions as a voltage divider. 40db attenuation or more means a gain of 40 db for the low noise amplifier, with the VCAs still operating in a region with good s/n and low distortion.

It is like the limiter configuration John mentioned, but designed to be way below threshold much of the time.

The 40 db was a general number I pulled out of my hat. Modern VCAs are designed to perform best around 0db.
The guitar signal will be low amplitude as the strings decay and the gain increases, so the VCAs need to contribute minimal noise at higher gains.

I hope this makes the app a little clearer, at least until I can post a block diagram.

 

[Bo Deadly]

That paper is from the 80's. VCA tech has improved greatly over the years. In particular the THAT 218x VCAs use a special dielectric isolation fab that yields very well matched complementary transistors which is critical to low noise and distortion in the gain cell. The 2181A achieve double-zero THD which is considered good enough for fader replacement in consoles.

Yes, at gain extremes THD and noise increase. However, for a compressor, you're generally not using large positive gain if at all. And negative gain is both limited to maybe 20dB and noise decreases with gain. So my guess would be that if you created a device as you describe, you would find that it's noise and THD was actually higher than with just one device. Especially at 0dB gain.

If you want to achieve the best possible noise and distortion performance, you should build a PCB for one THAT 2181A, characterize noise and distortion performance of the control circuitry and experiment with mutliple designs to determine what works best. Noise and distortion in the control circuity is the primary source of noise and distortion in the signal. This is frequently overlooked.

 

[Was known as VacuumVoodoo]

I do not expect any benefit from putting them in series

JR

PS; I do not understand what Dualflip is suggesting... most VCAs I am familiar with have a single current input and single current output.

Then there is cascode.... how you stack voltage or current sources.
As for VCAs a lot can be done with log circuits and current mirrors.
Fun with electronics at its best.

 

[Script]

Could drive E+ and E- at the same time thru inverting opamp. App notes say it improves noise. Not sure if this is still the case when E+ is used/driven for compression and E- for limiting at the same time -- as opposed to summing two sidechain signals before hitting the VCA.

 

[abbey road d enfer]

Could drive E+ and E- at the same time thru inverting opamp. App notes say it improves noise.

It's a case of a noise voltage modulating noise. The math are very complex but yes, it does reduce noise somewhat. It just follows the rule that uncorrelated noises combine quadratically, not algebraically, but the improvement is not 3dB for doubling.

Not sure if this is still the case when E+ is used/driven for compression and E- for limiting at the same time -- as opposed to summing two sidechain signals before hitting the VCA.

The artefacts introduced by the side-chain are a couple of orders of magnitude over the noise considerations. I wouldn't expect any significant difference there, except for control voltage throughput.

 

[JohnRoberts]

Could drive E+ and E- at the same time thru inverting opamp. App notes say it improves noise. Not sure if this is still the case when E+ is used/driven for compression and E- for limiting at the same time -- as opposed to summing two sidechain signals before hitting the VCA.

ASSuming E+ and E- are the gain control ports, best practice for commanding large gain changes (like deep fader attenuation ) is to drive the control ports differentially. This effectively cuts the max gain control voltage in half reducing voltage potentials between nodes inside the VCA. I never put this on the bench but I suspect it reduces VCA distortion at extreme attenuation settings. I am not sure how to even measure VCA distortion at -60dB gain.

Perhaps not obvious, any noise on the gain control ports will multiply the audio signal and show up in the output signal, in an ideal world these gain control ports would be driven by zero ohm source impedance. Op amp outputs are as close as we will get to that. For typical modest amounts of gain control, grounding the unused control port, is a fair trade off. I have seen applications where IC output are shunted with caps to ground to maintain low impedance above the audio band but of course the op amp must remain stable loaded with a capacitance. Caveat, this is arguably over design.

As I recall the AES journal paper authored by Gary Hebert is a good reference for modern VCA application and performance.

===

+1 to Abbeys observation... side chain manipulations will be far more audible than these subtle performance tweaks to VCA path linearity.

JR

 

[mr coffee]

Thanks for all the useful posts. I remember a high end board that used driving both VCA control ports via an inverter and I was never sure why that added complexity was included. Must have missed that it improves noise in the app notes. Thanks for sharing that bit.

Sorry for the delay in getting a diagram up, but I am traveling and my laptop went south and had to be fixed first. Murphy is always around, especially when you use an older computer.

Here is a sketch of the idea I am trying to ask for others experience with. I'm not looking for justn20 db of compression or so - this is for much higher range, and the idea is getting the amplification from the op amp, not the VCAs, by using the VCAs as attenuators in the feedback loop.

It's OK to tell me what's wrong with the idea - I used it eons ago (around 1975-80) when, as several folks have pointed out, VCAs were not as sophisticated as they are now. I agree that is quite accurate. I was thinking about a more up to date version of that circuit.

And before you dismiss the Hawksford paper out of hand because it dates back to the 1980s, take a look at the tests Duncan did of various VCAs circa 1980 something. He tested lots of VCAs, including 202 modules, various Blackmer cell designs (dbx, sliding bias, and the Valley People variation), the Aphex 1537a and VCA1001, later Allison versions, and even the old PMI databook VCA circuit using MAT04s with 5532s instead of the originally specified op amps. Hawksford also sent Duncan has demo version and he included measurements of it as well in his findings. It performed quite well using LM394s duals in a class A design.

The Hawksford paper I was referring to also speculated (near the end) that a cascaded VCA design would be advantageous to try with Blackmer cells operating in class AB in terms of performance at high levels of attenuation. NOTE: Hawksford's design was operating in class A and still had very low noise and distortion, both when tested at 0db, and tested at 60db of attenuation.

Again thanks for all the ideas and experience shared.