OTA in the feedback loop

[knutolai]

Long time, no post!

Recently revisited the valvewizard website (great articles Merlin!) and saw the new version Engineers Thumb. Got me interested in OTAs in OpAmp feedback loops as I haven't seen this concept in any other DIY projects/articles. Would it be possible to utilize the concept to build 1. order voltage controlled filters/integrators? The better SNR and headroom at 9V unipolar operation is very intriguing if applicable. In extension it would be interesting to use the concept for a 2. order state variable filter or perhaps a MS-20 filter workalike...

http://www.valvewizard.co.uk/engineersthumb4.html

If anyone is familiar with any other projects that uses this concept I'd love to hear about it

Hope everyone is keeping safe!

[swamphorn]

I haven't seen this before myself but I don't see why it isn't possible. The simplest way would be to use an op amp-OTA pair as an inverted-response VCA to scale the voltage into an integrator. With this scheme, as the OTA gets noiser and louder the cutoff frequency drops, masking the noise (at least in my intuition—I'd be surprised if it were that simple). With exponential current control you would only have to negate the input voltage, but the inverse relationship between current and frequency might give a good enough approximation of exponential control in itself.

If you're good with s-domain filter design (I'm certainly not) you can design a filter using differentiators or high-pass filters (which are approximated an integrator in a negative feedback loop with infinite and small open-loop gain, respectively). I would assume there are good reasons this is never done—differentiators are much more limited by gain-bandwidth product than integrators, for example. The most off-the-wall option would be to invert the s-domain transfer function of a current-controlled differentiator to get an integrator; whether this would be as simple as substituting an inductor (real, gyrator, or NIC) or something more complex is unknown to me.

On a different note, the LM13700 is hamstrung by its own datasheet which gives bad advice on bias current and linearizing diodes. openmusiclabs' "Minimizing Distortion in [OTAs]"¹ offers sundry techniques on maximizing the SNR of the LM13700 but even the simplest of techniques will improve a design, even if you want the soft saturation curve (which is a characteristic part of OTA-based filters). The biggest takeaway is that you should limit I_ABC to 100 μA, not 1mA as the datasheet recommends. If you're using the linearizing diodes, the current into I_D should be about 250 μA and held as constant as possible. I would also add that you might try out pre/deemphasis: boost the treble into the chip with a high-shelving filter and cut it afterwards with an RC network in parallel with the load resistor.

¹ http://www.openmusiclabs.com/files/otadist.pdf

[idy]

Old EH Blackfinger compressor had an (early?) version that used OTA in the feedback loop. Wow, interesting idea to use one to control a filter instead of vactrols...

[knutolai]

I haven't seen this before myself but I don't see why it isn't possible. The simplest way would be to use an op amp-OTA pair as an inverted-response VCA to scale the voltage into an integrator. With this scheme, as the OTA gets noiser and louder the cutoff frequency drops, masking the noise (at least in my intuition—I'd be surprised if it were that simple).

Something similar to this?


I find the negative input node of U1 problematic. Wouldn't the OTA output impedance block any signal from entering through R1? Or does the OTA output impedance change when we adjust the control current? I'm thinking you would need to enter through the negative input of U1 in order to achieve voltage attenuation (unless we add an additional OpAmp attenuator stage between U1 and U2?).

As you might tell I'm not used to working with OTAs 

[swamphorn]

Almost that; U₁ and U₃ are correct but U₂ isn't. For U₂, the output of U₁ is a voltage and you need to convert it to a current by placing a resistor between them.

Some quick simulation reveals that the inverse-response VCA works. I would use the linearizing diodes, personally, but that's just details at this point. The negative input of U₁ is a virtual ground: the U₁ does its best to hold it at (in this case) 4.5 V. Keep in mind that OTAs output a current, not a voltage, and as such ideally have an infinite output impedance. Naturally, reality falls short of this but when the output of an OTA is held at a constant voltage as it is here it acts much more ideally.

There is another way to invert the response: place an inverting OTA stage in the feedback loop of a buffer. You can see this in the Lab Series L5 Preamp¹, Q₃ and IC_5B. I want to say this would work with the built-in Darlington buffer instead of the JFET, though this would require biasing changes among other things. I'm just not sure if the Darlington buffer has a sufficiently high enough input impedance to work at that. This should give an inverted transfer function of the form Y(s) = 1 / (1 + B(s)), or at least approximately so given the far from unity gain of a JFET buffer. I would expect something more like Y(s) = 0.95 / (1 + 0.95 B(s)).

¹ https://www.diystompboxes.com/smfforum/index.php?topic=124008.msg1174919

[knutolai]

Almost that; U₁ and U₃ are correct but U₂ isn't. For U₂, the output of U₁ is a voltage and you need to convert it to a current by placing a resistor between them.

Ah that makes sense. Had so time to expand it into a state variable filter. Hope it's readable with the values and references. Had some trouble with kicad.



Still very much a concept drawing. Am I correct in thinking U2B and U2D need to have a below unity gain range (0 to 1)?

[swamphorn]

That looks about what I was thinking. I can't speak as to the actual part values, unfortunately--that would take more calculation on my part. I would like to do some calculations and simulation of an inverted high-pass filter as a pseudo-integrator using the high-pass filter in the datasheet. This would give us approximately a (1 + s) / s response and not the 1 / s we so desire from an integrator, however.

[knutolai]

I'm assuming you are referring to Figure 30?
https://www.ti.com/lit/ds/symlink/lm13700.pdf

That looks very similar to the filter blocks used in the korg MS-20 filter. I might be misunderstanding, but wouldn't such a circuit have a (regular) non-inverse response?

[amz-fx]

Paia Hot Lyx:

https://www.paia.com/talk/viewtopic.php?f=6&t=524

IC2 is the CA3080 OTA.

regards, Jack

[swamphorn]

I'm assuming you are referring to Figure 30?

...

Nearly so; it would be Figure 30 but placed in a negative feedback loop. I don't have part values worked out, but the resistors surrounding the op amp should be such that the gain of the feedback path should be much greater than the gain of the input signal.