Discrete OTA?

[Mark Hammer]

Q:  There are so many ways in which OTAs can be used.  In what applications does the drift matter and not matter? To whom would it matter and not matter?

[Earthscum]

Q:  There are so many ways in which OTAs can be used.  In what applications does the drift matter and not matter? To whom would it matter and not matter?

Implying that maybe there's applications where a discrete version would work just fine, kinda like the Compression OP Amp project? That thing is the cause for many of my breadboarding experiments, yet I still haven't built it. 

[merlinb]

I have difficulty seeing how a discrete OTA would have thermal problems. It's nothing more than an LTP??

[Earthscum]

I was thinking about what Mark said earlier, and I sat there staring at the OTA's (instead of looking at synth circuits and looking for the OTA), and all of a sudden it jumped out at me... I've been staring at these for a couple weeks now since Rick's post about a stomp box ladder filter.

Now a bunch of other stuff makes sense about how this works... my brain hurts, lol.

[Cliff Schecht]

Q:  There are so many ways in which OTAs can be used.  In what applications does the drift matter and not matter? To whom would it matter and not matter?

This is simple. The human ear is very sensitive to changes in pitch but not very sensitive to changes in amplitude. With synths this translates to anything that is pitch or frequency related needs to be stable and anything that is amplitude related can be a bit sloppier without you noticing. The exception to the rule here are filters where the cutoff frequency can drift slightly without you noticing.

I have difficulty seeing how a discrete OTA would have thermal problems. It's nothing more than an LTP??

Well, yes. But there are multiple temperature effects we are fighting against. With any long tailor pair, the first-order temperature effects are essentially canceled out by the nature of the circuit. But there is a second-order term that has to do with the temperature coefficient of the forward biased diode in the transistors base-emitter junction. Essentially, Vbe is not a constant term and drifts at about 3300PPM/degC. Negative temp coefficient resistors are usually used to compensate for this and are typically connected thermally to the transistor pair with some heatsink goo. Another neat trick is to actually use a heater to keep the circuit at a constant temperature. The old Curtis VCO chips did this and are known for being pretty damn stable. The negative side of this is having to wait 5-10 minutes for your synth circuit to warm up before you can really play it.

[Eb7+9]

I have difficulty seeing how a discrete OTA would have thermal problems. It's nothing more than an LTP??

!

[merlinb]

Another OTA article for the archives. (Apologies if it has already been posted)
http://www.idea2ic.com/LM13600/Using_OTAs.pdf

[JDoyle]

Q:  There are so many ways in which OTAs can be used.  In what applications does the drift matter and not matter? To whom would it matter and not matter?

If you aren't using direct coupling, - i.e. every single guitar circuit that runs off a 9V battery - and your circuit is competently designed with feedback and offset trim allowance, which is a basic requirement of any uber-high gain stage, I doubt you would even notice drift's existence. It could be argued, after all, that drift is simply the last AC signal in the spectrum before you get to true DC. With direct coupled circuits any offset will be multiplied along with your signal as it procedes through the circuit. In terms of drift's effect upon the individual circuit itself, unless one is amplifying DC values and the accuracy of doing so is important, (such as in synths, analog computers, etc.), in my opinion, there is no need to worry about drift; and if you find that there is, servo bias it out.

The only voltage that matters in an OTA circuit is the difference voltage between the two inputs (+ve) - (-ve), besides that, everything else is current - thus why an OTA doesn't have resistors in it's circuit. And even that differential input voltage is an OTAs largest source of error - the diodes one can bias on second generation OTAs like the 13600 are there to 'pre-distort' the input signal in a manner that is opposite to the error from the diff. amp, due to the intrinsic resistance of the diff. amp transistor's emitters.

If one studies the internal schematic of an IC OTA like the CA3080, one should be able to deduce quite quickly that the transistors in the circuit do not ALL have to be matched, because there isn't any way they can be in the 3080 - the PNPs are darlington connected, making them immediately different from the NPNs. This is because lateral PNP transistors used in ICs simply suck, especially in early ICs like the 3080. If one uses discrete transistors, a standard two-transistor current source will work quite nicely, as long as the transistors are matched.

Going further one should also realize that the only transistors that need to be matched are the individual pairs - the two NPNs in the diff. amp (which should be the best quality pair as well, or at least, the lowest noise), the two NPNs in the Iabc current mirror that is the 'tail' of the LTP, and then each pair of remaining current mirrors, two PNP and one NPN. The current mirrors also operate solely on the Vbe/Ic principle, except their input signal is a current, not a voltage like that of a diff. amp. The Ic of the diode connected transistor can change in magnitude for several decades of it's value and its Vbe (and that of the 'mirror' transistor's) will change accordingly, but very little in comparison to the decades of current change available (change Vbe by 60mV and the current will change by a factor of ten and vice-versa - note the complete absence of any reference to Beta). In my opinion that last fact, that a very slight change in Vbe causes a rather enormous change in Ic, is why people have long made the mistake of thinking a transistor is a current controlled device, advancing the importance of Beta, rather than what it is - a voltage controlled device. The numbers one sees in the formulas for an OTA (gm=19.2Iabc) are the same as those from the laws of transistor physics (gm=38.4Ic)- except halved because of the presence of TWO input transistors.

It is very much possible to build discrete OTAs that outperform the 3080 in guitar effects.

[PRR]

The OTA is serendipity. If you build it discrete, it is so many parts and so much matching that you'll find another way to do your thing. If you lay it out in an IC process, matching is possible and a bunch of parts overlap saving connections and space.

The OTA is "NOT" an opamp. All you know about an opamp is that gain is "very high", and overall gain can be set with external feedback. The whole point of an OTA is that gain can be varied. Putting NFB around it (usually) defeats the point.

> are the buffers in LM13600/13700's there as current to voltage converters?

No. The core OTA is normally loaded with an impedance (often a resistor). This converts the current to a voltage. Because Iabc can be very low, and output current is about the same, this impedance tends to be large. Larger than is convenient in system design. The Darlington is a voltage to voltage converter with large current gain.

Matching.

If you want to limit an audio level 20dB, that's not hard. 10:1 range of action is not large, and 10% errors go unnoticed.

If you want to tune a musical oscillator or sharp filter over a 1000:1 range, you will get frustrated. 1000:1 is not a long way for a junction, but enough to glimpse both low-level and high-level errors, meanwhile errors have to be far-far-far under 6%, more like 0.1%.

As Cliff said.

Discrete.

You can't match like peas in a pod. You can't easily evade the basic laws which limit dynamic range. It would be hard to improve the low-level performance of the 13600. For some specific cases you "might" be able to gain a slight edge with LARGE area devices working at high current. OTOH commodity large devices are "hammers", not fine screwdrivers, and matching becomes difficult; also large currents get into parasitic resistance errors.

If you love building current-mirrors, build your own 37000. You can poke around "inside" and see what happens, though mostly the voltages "don't change". If you want some easy mid-fi audio level control, just buy 13600.

> VCAs produce ... ... ... a logarithmic gain control.

There are a WIDE variety of "VCA"s.

Many many are linear. There are probably more linear VCAs in all the TV sets in the world than there are log-law VCAs. They control contrast, chroma, and often audio level (TV source and listening levels don't cover a wide range).

dBx/Buff/Allison/THAT VCAs are an awkward serendipity. A simple transistor pair can give a lovely logarithmic current split. For symmetry you want four. Some errors cancel and others remain. There's housekeeping to control the idle currents. This type VCA has been a mainstay of the professional audio world (and now THAT is marketing consumer uses); but there are others. (The several tube VCAs stay popular for limiter work.)

> mistake of thinking a transistor is a current controlled device, ..., rather than what it is - a voltage controlled device.

Read the book. Amplifying Devices and Low-Pass Amplifier Design, 1968, Ed Cherry and Daryl Hooper

Tubes, FETs, junction transistors are CHARGE-controlled devices.

In low-frequency work, we may pretend that charge=voltage and get the same answers. This assumption fails at high frequency; in audio we merely shrug.

In vacuum triodes, the input charge includes a component of plate voltage.

In junction transistors (and positive-grid tubes), the charge storage is "leaky". This leakage is significant and roughly predictable and it is indeed a handy short-cut to pretend they are current-controlled current sources (a "transfer res[/b/istor[/b]").

I hesitate to bring-up AD&LPAD because it is very expensive and far over most hobbyists' heads. But if you have thoroughly absorbed Ohm's Law, can jiffy-design amplifiers to various uses, and can chew on concepts of "charge", it is a whopper of a thought-provoker.

[Cliff Schecht]

Lots of good points made here. Some comments on a few..

Matching

The most important elements to match in an OTA are the diff amp and the current mirrors. For the best accuracy, it's actually best to match each pair of current mirrors. This is hard to do with discrete but it can of course be done. I'm not a fan of having to sort transistors though, this is bad engineering practice in my book (especially if you are considering any scale of production).

OTA based VCO's

This is a solid concept for a simple reason. With an OTA, you can obviously control the output current with a programmable current pin. If you implement the classic square-triangle style oscillator with an OTA as the comparator, you essentially end up with a voltage controlled square and triangle wave oscillator. Implementing this with the standard 13700 gives you a relatively pitch stable oscillator without too much fuss (assuming you have an accurate Ibias configuration). You get two very clean waveforms to base the sawtooth and sine wave off of which rids this oscillator of the problems associated with a typical saw-based VCO.

The big problem with this is range, the OTA's fart out after about an octave and a half. I played around with this quite a bit for something I was working on for PAiA and kept running into dead ends. After trying to go discrete with 2% matched transistor pairs (the THAT stuff is prohibitively expensive in my case) and ending up with some pretty bad results there as well, I gave up on this concept (for the time being). I eventually plan on revisiting this design with a current buffer after the OTA to solve the small range issues.

charge controlled devices

This is an interesting point. Yes we do typically look at transistors as voltage controlled current devices but each device operates on a different set of principles. With a BJT (let's assume AF range for now) you are essentially treating the device as a Vbe/Ic converter. With the Vbe however, there is a certain amount of current that is required to actually get the device to conduct and it turns out to be proportional to the Vbe (exponentially in nature because it's a diode). So for a given voltage, you have to feed a certain current as well to get your hot carrier and depletion region action. With FET's, the gate current is typically negligible because small signal FET's have TINY gate capacitances. This is the capacitance that you have to charge up to get that field-effect which modulates your channel width and controls the current flow through the drain. This doesn't hold so true for power FET's with nF+ gate caps (usually require AMPS to charge in bigger supplies).

I guess you can look at the base current of a transistor as "leaky" but in reality, it's just the energy over time necessary to cause a depletion region to form and to cause the electrons to go out of equilibrium allowing current to flow out of the collector. It's just something we deal with for AF amplifiers (they're pretty linear if you bias them right) but actually becomes a problem again in higher power applications where it is necessary to actually pull current out of the base to turn off a BJT or IGBT completely. But since we aren't dealing with any kind of power in our effects (or RF hopefully), this discussion kinda gets beyond the scope of DIY guitar effects .

[JDoyle]

The OTA is serendipity. If you build it discrete, it is so many parts and so much matching that you'll find another way to do your thing. You can't match like peas in a pod...It would be hard to improve the low-level performance of the 13600.

I guess I should have been more clear - I didn't need to sort through transistors, I had found a source of Japanese-made duals with high-gain and low noise. So with five parts I could easily make an OTA that exhibited much less noise than a 13600. I guess it is also a stretch of the word 'discrete'.

I also don't think that one needs to match the mirrors - the mirrors in the 3080/13600 are unity gain.

Anyway, I'll check out that book.

Jay Doyle

[Cliff Schecht]

You want to match the mirrors for offset/tempco reasons. The diode-strapped side of a current mirror (the ref side) doesn't have to push any power usually so it stays cool while the other transistor is made to push the majority of the current. So one transistor gets really hot, the other stays cool and they drift apart from each other. Again all we can do without adding more circuitry is to match the current mirror transistors and try to keep them as physically close to each other so they maintain about the same temperature.

[JDoyle]

I guess my point is that with guitar effects, it doesn't matter.

For example, in the ross comp, at max gain with all the current coming from a single mirror, and dropped across the full 9V (which will never happen) my calculations make the dissipation 5mW.

Offset/tempco doesn't come into play either in such a circuit.

At least, I've never had any problems and I've been using it for about 5 years.

[Cliff Schecht]

I guess my point is that with guitar effects, it doesn't matter.

For example, in the ross comp, at max gain with all the current coming from a single mirror, and dropped across the full 9V (which will never happen) my calculations make the dissipation 5mW.

Offset/tempco doesn't come into play either in such a circuit.

At least, I've never had any problems and I've been using it for about 5 years.

This. As I deduced in FreqCentral's Murder One thread, a lot of the problems that occur at higher power levels and/or at higher frequencies aren't evident or even present at all in audio circuits. It's great to know about topics like these and we could discuss the device physics for days on end but for the most part, it doesn't matter for what we do around here. We run into some drift and noise problems with our high gain/nonlinear circuits but these are usually manageable in some way or another.

[valvusmusicus]

Don't know if this is the sort of thing you're looking for. It's an OTA I put together for use in various projects. With a 7.5V control voltage the overall gain is unity, but this can be changed by altering the gain of the opamp, of course. Maximum input is 8Vp-p, which is more than most stompboxes will deliver.

Hi
Regarding the circuit below, Im probably being thick, but what are all the battery symbols with voltages for. Are they what the voltage measures at that point, or dothose points need supplying with that specific voltage?

[garcho]

it's not what you would find in common schematics we see around here, which are often oriented towards component layouts. look at OP1, the op amp on the far right, see how it's negative/ground terminal is going to ground (which is pointing upwards, not a very good graphic design policy)? and how the positive PS terminal is 9V above ground, because of two series 4.5 voltage sources? notice how the non-inverting op amp input is biased for audio at 1/2 PS voltage for the op amp, typical of our guitar pedal circuitry? that's because this is more like a textbook schematic as opposed to a "layout" which is what we call a lot of schematics around here. a DC voltage of 4.5 is needed, but how/why it's produced is left out, as opposed to many of schematics which have the "Vref" voltage divider drawn out, because a lot of us can build but don't understand the circuitry. Try drawing it out by hand, copying it, and change everything to the version of the symbol you're most used to, and maybe instead of connecting all the ground points together, draw out each one with the arrow symbol, it might make more sense.

[ElectricDruid]

Here's a link for the Rene Schmitz version of JH's discrete 2040 filter:

http://www.schmitzbits.de/rs2040.html

I'd say the difference between an OTA and a VCA is that an OTA is a specific type of device (Operational Transconductance Amplifier, like the name says) whereas a VCA is a specific application (Voltage controlled amplifier). As such, you can make a VCA using an OTA, or you can use application-specific VCA chips (which may or may not be based on an OTA-type design internally). Whether a VCA has a log or linear control law is not specified - some have both (CEM3330, for example).

Also, VCA chips can be used for all sorts of things, just like basic OTAs. They're not limited to just audio level control. The synth world is full of circuits based on "VCA" chips - filters, phasers, waveshapers, oscillators, etc etc. For example, over at Intellijel David Dixon has virtually made a career and a full synthesiser using the V2164 quad VCA chip, to the point at which people tease him about it on the SDIY list!