Paralleled swinging op amp

[atavacron]

This bootstrapped-rails output is based on two main things: the King/Watkins article in EDN, and the @KA-Electonics.com update of the MCI 2003 with a current-boosted back end, detailed at Pro Audio Design.

With this, I’m basically saying “you’re already surrounding an amp with parts, just do it with a dual and split the increased load.” The goal is to provide a fairly cheap, very low noise +24dBu single ended output, with a low Iq and parts count, in an otherwise dual 15V environment. @thor.zmt lobbied me successfully toward this in my thread about loading and THD.



The amp doesn’t have to be an OPA1602 — most any dual whose amps can push 22.4mA at full output will do. It could easily be an OPA1692 or OPA1678. ADA4075-2 would perform nearly as well as drawn, but maybe Vcc and Vee would need to be a bit higher. The Rf/Rg are chosen for 2.5nV/rtHz, but are of course scalable. I forgot the BAV99s at the output, just imagine ‘em there.

Questions:

• Could anything be done better/simpler?

• What would be the best choice in SMT for the NPN and PNP? SOT. I am not sure we need such a high performer as the ZTX851/951 here. But if those do have SOT equivalents, I’d love to know.

• What voltage do the amps see - +/-15.3V or thereabouts?

• If you’re doing a board with a bunch of these outputs on it, you’re bound to start loading the supplies. What happens if you multiply the supply dividers by (say) 5x or 10x in ohmic value? Higher source resistance at the bases, surely. Does the noise matter? At what point do you start to have to take a closer look at the PSRR of the amp? Any advantage to using MOSFETs at these low currents?

Thanks!!

 

[thor.zmt]

• Could anything be done better/simpler?

Simpler zener diodes instead of resistor & series diode?

More complex, use a folded pre-driver to minimise non-linear loading of output

• What would be the best choice in SMT for the NPN and PNP? SOT. I am not sure we need such a high performer as the ZTX851/951 here. But if those do have SOT equivalents, I’d love to know.

I'd go BCP53/56, that is in effect BD139/140 in SOT223.

• What voltage do the amps see - +/-15.3V or thereabouts?

Put it into TINA TI. If you use folded pre drivers to compensate the series transistors and zener diodes, then always the Zenner Voltage.

• If you’re doing a board with a bunch of these outputs on it, you’re bound to start loading the supplies. What happens if you multiply the supply dividers by (say) 5x or 10x in ohmic value? Higher source resistance at the bases, surely. Does the noise matter? At what point do you start to have to take a closer look at the PSRR of the amp?

You could Mosfets instead of BJT's and then use very high impedance dividers.In fact the circuit could be much simplified:



Typical Mosfets have ~ 3V Threshold voltage, so the Op-Amp's get +/-12V and all cap's can be ceramic, SMD, I think X7R, despite being microphonic will do. Mosfet's look for low reverse transfer (miller) capacitance as Cin is essentially boot-strapped.

Thor

 

[atavacron]

I'd go BCP53/56, that is in effect BD139/140 in SOT223.

Thank you!

You could Mosfets instead of BJT's and then use very high impedance dividers.In fact the circuit could be much simplified:


View attachment 109201

Damn dude, you draw fast. What Mosfets do you like there?

Typical Mosfets have ~ 3V Threshold voltage, so the Op-Amp's get +/-12V and all cap's can be ceramic, SMD, I think X7R, despite being microphonic will do.

GRM C0G, up to 470n in 25V. 100n in 50V — sweet spot. 1206.

I didn’t know about the threshold voltage. So with that, the package is cooler, but the amps are being swung more often. Any downside to relying on that for the bulk of (strong) program material, rather than just peaks? [edit: I guess that’s the last 3.7dB in this case, so still just peaks.]

Mosfet's look for low reverse transfer (miller) capacitance as Cin is essentially boot-strapped.

2.5pF Ccm, 2pF Cdiff on the 1602, avoids compensation up to nearly 2K i think. Handy.

 

[atavacron]

Speaking of zeners, kudos to Graeme in 1973(?) for this no-resistor unity gain thingy.



Could be a good starting point for a 24dBu input, unless this Mosfet thing pans out and it’s fine for unity gain too. Either way, I’m gonna need a differential buffer that can accept that level on one side before it hits a diff amp to knock it down to 21.5dBu. I put the cart before the horse but they need to travel together regardless.

Bummer that each amp will have to be a single; so it goes. Could parallel OPA1692 sections though, -124.7dBu I think, two packages per input and still cheap.

I remember that zeners were not preferred from signal to rails in The Phantom Menace Returns…

Also Thor makes a good point, the 15V dual is already there, it’s advantageous to use it. No zener dissipation.

 

[thor.zmt]

Damn dude, you draw fast. What Mosfets do you like there?

I have a sketchbook on my desk.

What MOSFET's depends on supply chain and what you can get in P-Channel. Have a look at Mouser, it gives a good idea.

GRM C0G, up to 470n in 25V. 100n in 50V — sweet spot. 1206.

Sure, overkill IMNSHO, X7R will do.

I didn’t know about the threshold voltage.

Mosfet's exist with very low threshold voltages, as low as 0.5V. Common ones you might find will tend to be something like IRF610/9610 or comparable. These tend to be rated for Vth = 2...4V with design center at 3V.

So with that, the package is cooler, but the amps are being swung more often.

The Op-Amp's are ALWAYS being swung. All these circuits provide bootstrap. This can actually be a good thing for Op-Amp performance.

Supply bootstrapping reduces distortion in op-amp circuits

Thor

 

[JohnRoberts]

When paralleling op amps I prefer to only make one negative feedback loop control the output. Still use the 2.2 ohm current sharing resistors but feed the input to the second op amp from output of the first,,, That way DC offsets will not cause them to fight each other. Of course good modern op amps may have good enough DC performance for this not be a concern.

JR

 

[thor.zmt]

When paralleling op amps I prefer to only make one negative feedback loop control the output. Still use the 2.2 ohm current sharing resistors but feed the input to the second op amp from output of the first,,, That way DC offsets will not cause them to fight each other. Of course good modern op amps may have good enough DC performance for this not be a concern.

JR

This version of paralleling is to reduce noise, primarily.

Thor

 

[atavacron]

This version of paralleling is to reduce noise, primarily.

it’s source and sink first (here), followed closely by noise. i really do want it to be applicable across a wide range of modern amps, without requiring an output BD139/140 pair or insisting on a specialty driver such as the OPA1622.

@JohnRoberts your point is well taken. the output stage that all the modern TI bipolar amps are based on is pretty consistent, and i’m inferring that most people are going to be using amps with max Vos below 2mV, typically below .5mV.

but 2R2 is low, truly, considering how these types of parallels are almost always drawn with 10R ballast. would 4R7 or 10R be more comfortable, or does the specific value not matter so much w/r/t Vos?

Relevant to our interests:

How to Solve Analog High Voltage Delivery Challenges with a Bootstrap Approach​

 

[JohnRoberts]

it’s source and sink first (here), followed closely by noise. i really do want it to be applicable across a wide range of modern amps, without requiring an output BD139/140 pair or insisting on a specialty driver such as the OPA1622.

@JohnRoberts your point is well taken. the output stage that all the modern TI bipolar amps are based on is pretty consistent, and i’m inferring that most people are going to be using amps with max Vos below 2mV, typically below .5mV.

but 2R2 is low, truly, considering how these types of parallels are almost always drawn with 10R ballast. would 4R7 or 10R be more comfortable, or does the specific value not matter so much w/r/t Vos?

If you are only doing onsey-twosey, do what works... you can tell with a finger touch test if the two op amps are fighting each other enough to do harm. Ohms law will predict the current from I=E/R so likely to only be mA from mV of offset. Those mA will be times the op amp rail voltage for power dissipation.

I have only paralleled op amps together one time and that was because I had quad bifets (TL07x) already in the design with one unused section sitting there. Putting it in parallel was cheaper and easier than any other way to buffer the output (and just felt right) . TL07x are notorious for weak drive capability (nominally 2k). I was dealing with a low noise, gain stage and wanted to drop the feedback resistor lower than 2k for it's noise contribution. In decades of circuit design, that was the only one time I ever did that.

JR

 

[thor.zmt]

but 2R2 is low, truly, considering how these types of parallels are almost always drawn with 10R ballast. would 4R7 or 10R be more comfortable, or does the specific value not matter so much w/r/t Vos?

If you have guaranteed +/-10mV offset and use DC Blockers in the feedback network the worst case offset is 20mV across the two resistors in series.

With 10 Ohm we get 1mA, with 2.2 Ohm we get 4.55mA cross current. Both I think are acceptable.

If the Offset is better, we can lower resistors relative to permitted cross current.

it’s source and sink first (here), followed closely by noise. i really do want it to be applicable across a wide range of modern amps, without requiring an output BD139/140 pair

Funny, I'd be more interested how I can do this with a NJM5534 and a pair of BD139/140 or their SMD equivalents. And perhaps use a 2SK2145 to replace the bipolar inputs on the NJM5534. It could even run on +/-22V and drive the output stage from the Comp Pin.

Thor

 

[thor.zmt]

Damn dude, you draw fast. What Mosfets do you like there?

Slightly more elaborated. Assumes FET/JFET input op-amp - unsuitable for bipolar input OPA's without some changes:



Advantage, you can use those 100nF C0G capacitors and have guaranteed offset for paralleling and you could PROBABLY use a single feedback divider after the current sharing resistors.

Thor

 

[atavacron]

View attachment 109235

Personally I’m dialed in to optimizing for BJT, but this is great too. That IRF610/9610 is THT and the article you posted is about using an LME49710 with bias current relief via the bootstrap for source impedances up to 1M, so I do wonder if you might not be veering towards a very flexible network to surround a paralleled LM4562 with 1M/220n or something. Y’know, breadboardable. Polypropylene in that case I suppose, though I am recalling one affordable PPS series with leads, maybe Kemet.

 

[atavacron]

Funny, I'd be more interested how I can do this with a NJM5534 and a pair of BD139/140 or their SMD equivalents. And perhaps use a 2SK2145 to replace the bipolar inputs on the NJM5534. It could even run on +/-22V and drive the output stage from the Comp Pin.

That’s cool too! More in line with Wayne’s work. Maybe he will chime in if i keep citing papers & designs

 

[Newmarket]

All interesting from a technical viewpoint. esp as I deal with swinging rails riding on a high CM voltage in the (non-audio) day job. But I have to ask what is the audio application / benefit for this higher voltage operation ?

 

[JohnRoberts]

All interesting from a technical viewpoint. esp as I deal with swinging rails riding on a high CM voltage in the (non-audio) day job. But I have to ask what is the audio application / benefit for this higher voltage operation ?

Several... I did one back in the 60s to make a cheap and easy audio power amp... not much power maybe 30-40 W but more than a single op amp can handle by itself. I converted an old broken hifi/record player , with a functional power supply, into something that didn't suck for my younger sister to play records on.
===
another old white whale being pursued for some time is a DC coupled mic preamp capable of phantom supply input voltages. There are a number of workable topologies that would work, I did a lot of drunken scribbles but never made one, Wayne melted solder and made his work.
===
Higher power audio amps using op amp technology front ends have been in wide use for decades. Not sure about the appropriate terminology, I've heard them called driven rail, but similar net result, bigger output voltage swing from only 36V op amp technology.

JR

 

[Newmarket]

Fair enough. I wasn't really thinking about "power applications" tbh although appreciating that "opamp" technology can be scaled.

 

[atavacron]

@Newmarket a few reasons:

1. +24dBu single ended & impedance balanced, without a DOA or a transformer or a current-boosted output. Granted with Riso of 100R total, you get 1.33dB loss into 600R, but some amps require less Riso, or see #2:

2. Scalable to higher levels for folks who want to throw higher rails in there and do some fine calcs at the edges of their preferred amp’s source & sink curves.

3. +2.5dB noise gain total rather than the +6dB noise gain of an inverter added to an output buffer in a dual-driven scenario. This could get shaved to just +2dB noise gain, but A) 1:0.75 is a very easy ratio to manage at an input, and B) a lot of general purpose amps won’t quite hit +22dBu as voltage followers. THAT1243 is a front end asset, but high ohmic values there blow the noise savings. You can parallel inverters in a driver and cut the noise gain to +3dB, but this output architecture takes just one dual amp package.

4. CM distortion benefits as per the Danyuk article (not a main concern for me but certainly an asset).

5. Everything we’re discussing is compatible with 51X rails of dual 24V and dual 16V, so anything we come up with here already has a regulated PSU framework ready to go. Easy breezy proto.

Actually that reminds me, the next version I draw should be 24V & 16V, and we’ll see how many more 0.1dBs of headroom I can scrape. Maybe 24.5dBu for a single ended chain that is truly capable of 22dBu. I just wanted to make a point about only needing 22V rails for this application.

[EDIT: Oh yeah, 24.5dBu / 13Vrms / 18.4Vpk is totally doable like this. With OPA1692, certainly. Dunno about OPA1602.]

 

[atavacron]

and you could PROBABLY use a single feedback divider after the current sharing resistors.

You’re a 1678 user. Would it like 3M3 for each inner loop and 470n between each inverting input and a shared attenuator? Sort of a fancy way to save parts but I see that it would ensure dead nuts AC accuracy. Values chosen to avoid any phase shift at 20Hz, not because I’m a subsonic stan. That’s that 470n GRM I mentioned, which is surprisingly not so spendy.

 

[marcuswilson007]

It's a good way to improve the common mode problems of cheaper OP amps. Doug Self goes into this in one of his books (can't remember which one) and has some performance measurements of reduced distortion.

 

[Newmarket]

@Newmarket a few reasons:

1. +24dBu single ended & impedance balanced, without a DOA or a transformer or a current-boosted output. Granted with Riso of 100R total, you get 1.33dB loss into 600R, but some amps require less Riso, or see #2:

Understood. But my question is really whether this actually offers a benefit (rather than a feature) to a user. Don't get me wrong - I like it technically. Just not sure about the merit of higher voltage operation in general. Esp when that level is unlikely to be supported throughout the signal chain.

2. Scalable to higher levels for folks who want to throw higher rails in there and do some fine calcs at the edges of their preferred amp’s source & sink curves.

Yes - what I'm questioning is the reason for wanting those higher voltage rails (beyond technical curiosity)

3. +2.5dB noise gain total rather than the +6dB noise gain of an inverter added to an output buffer in a dual-driven scenario. This could get shaved to just +2dB noise gain, but A) 1:0.75 is a very easy ratio to manage at an input, and B) a lot of general purpose amps won’t quite hit +22dBu as voltage followers. THAT1243 is a front end asset, but high ohmic values there blow the noise savings. You can parallel inverters in a driver and cut the noise gain to +3dB, but this output architecture takes just one dual amp package.

+1

4. CM distortion benefits as per the Danyuk article (not a main concern for me but certainly an asset).

Well "Better is Better" wrt CM distortion