Paralleled swinging op amp

[thor.zmt]

After the NE5534 was released, the engineers wanted to use the 5534 (relabeled in a metal can as MCI2003) for its superior performance and low noise, however, I guess that they wanted to retain the big headroom that they used to obtain with the 2002 (or perhaps as a marketing gimmick, since they actually used to place magazine ads boasting the bootstrapped 5534), so they used the transistor bootstrapped arrangement, which (according to one of the designers) was taken straight out from an 'op-amp cookbook".

NE5534 is subject to phase-inversion when exceeding the moderately limited Common Mode Range. That leads to frying things. When using bootstrapped rails observing CMRR is essential (actually, not only then).

Uncritically copying Textbooks is dangerous, especially in this case for circuits with a gain of > 1.5....

Thor

 

[user 37518]

NE5534 is subject to phase-inversion when exceeding the moderately limited Common Mode Range. That leads to frying things. When using bootstrapped rails observing CMRR is essential (actually, not only then).

Uncritically copying Textbooks is dangerous, especially in this case for circuits with a gain of > 1.5....

Thor

I am guessing they didn't just copy it, I was trying to point out that it wasn't an original idea. One of the designers commented to me when asked that they got it from a cookbook. They probably 'massaged' the design to get it right, though.

 

[atavacron]

Looks ok for the job, watch power dissipation with shorted output.

ah, yup…fine for a dual, not so great for a quad!

edit: still ok with a quad but just marginally - there are 500mA options out there with almost-as-good specs which would be better

 

[atavacron]

@thor.zmt this is specifically a gain of 1.333 at the output, though i would expect some people to want to bump that down a little if they’re using +22dBu-capable amps throughout their chain (like all 1692s for example). [oh whoops, you were talking about MCI]

@Dualflip this is only for a typical differential input buffer prior to a differential amp, and for the line driver at the output. The 24V dual supply is the highest I’d consider. I think it would be foolhardy to go applying this approach to every stage. I’m not even gonna try to make an all-purpose circuit, just one for each of these tasks that is rock solid. Maybe that was clear from the jump, maybe not. Also, the MCI micro-history was great, cheers!

 

[thor.zmt]

[oh whoops, you were talking about MCI]

And in general. Let's imagine a summing Amp.

Inputs sit at 0V all the time. At what rail voltage as pur rails shift do we exceed the CMRR and risk phase inversion?

@Dualflip this is only for a typical differential input buffer prior to a differential amp, and for the line driver at the output. The 24V dual supply is the highest I’d consider. I think it would be foolhardy to go applying this approach to every stage.

MC2 Amplifier use rail bootstrapped 5532/34 with triple emitter followers to make killer whot amplifiers and swing huge voltages... And have high gain. And sound pretty decent too.

As the Cat in "Last Unicorn" said:



There be a trick to it, of course.

Thor

 

[atavacron]

Let's imagine a summing Amp.

Inputs sit at 0V all the time. At what rail voltage as pur rails shift do we exceed the CMRR and risk phase inversion?

Relevant sidenote:

The BB app note about swinging a diff amp up to its own normal rails

Worth a look because +24dBu is 17.4Vpk and one might be able to just use 18V rails instead of 16V & 24V, by doing something like this (looking at you, OPA1602). Clearly blows the noise gain savings, but maybe there’s a way with a paralleled dual and all 806R or 1K pairs, whilst getting the +2.5dB using the same CFA technique.

And while the INA1620 is the spendiest of all the practical audio op amps for line driving, it does indeed have two incredibly burly/stable drivers and four matched “1K” pairs (averaging 965R).

Or remove the limitation of all matched values and give the amp itself a gain of 1.333. Either way, there is something to be said for using an Riso of 10R, which INA1620 or OPA1622 would love. OPA2210 too (3R3), but it can’t hit these kinda levels.

 

[thor.zmt]

Relevant sidenote:

The BB app note about swinging a diff amp up to its own normal rails

I used something like this with a LM3875 as Op-Amp and banks of 2SC5200 / 2SA1943 and of course rail bootstrapped and grounded bridge to boot.

The LM3875 biased into class A (resistor to negative rail instead of GND) acts as error amplifier in a Quad style current dumper up to a few Ampere output current, the big artillery takes over after that.

120V DC rail for grounded bridge and closer to rail to rail means around 80V RMS undistorted output, bridgeable again and indestructible in my tests.

On top, a very "Hifi" Sound, like a really high end class AB amp. But with 800W 8 Ohm (or 3,200W 8R bridged).

Then TI cancelled the LM3875 and for whatever reason the remaining LM3886, while sounding good it never had the "magic" of the LM3875 is terms sound. So the Amp project died with the LM3875.

Honestly, driving passive ATC SCM20 in bridged mode was something else.

Thor

 

[ccaudle]

the remaining LM3886, while sounding good it never had the "magic" of the LM3875

LM3886 also has a pin labeled GND. I assume it would have to be driven from a divider that moves with the bootstrapped power supplies. Not a big deal, but an additional annoyance when you don't want to switch parts anyway.

 

[thor.zmt]

LM3886 also has a pin labeled GND.

It is needed for the mute function. If mute is permanently unmute, it can be kinda ignored. Tie to 1/2 Vcc-Vss with resistors for safety.

What it actually is, is a unity gain input when mute is asserted.

I wanted to try how much swing that handles (probably all the LM3886 can do) and if it can be used as unity gain buffer following a tube gain Stage. Somehow I never found the time.



The mute pin switches between the normal differential input and a differential that is heavily degenerated and has one input tied to the "GND" pin via 10k and the other via 1k to output.

So, drive the GND pin and you have a unity gain power buffer.

Needs to be charaterised of course. Only a thought, never tried...

Thor

 

[atavacron]

The “CYA Version,” below. The “KISS version” (for direct coupled BJT amps) is forthcoming, but I want to lock in the bootstrap network first.



220n is the largest value for those positions that allows either 25V PPS (ECHU-X) or 25V C0G (GRM), affordably.

Note the shortlist of duals it should work with, covering a range of price points, eins, and 1/fs. Feedback attenuator values take advantage of the 1656’s noise characteristics if chosen. 17.7mA seems like easy driving for all, into 600 ohms. Again, no DC protection drawn at the output.

Questions:

1. Thor, did I get your amp architecture description right?

2. (a) Any potential issues with using such a high value - 6M8 - for gate bias, and (b) I had previously put in X7R for the sense caps at the output because we’re not listening to the result, but is it smart to use C0G there just because we want the values to remain accurate under full rail bias?

3. (a) Does the LPF created by the 1K/10p at the gates effectively address note 2, subnote 2 in the quote of post #38, (b) is that even a necessary LPF with Mosfets rather than the original architecture from post #1, and (c) is a gate cap to ground not the best way to accomplish this?

4. How’s 47u for 600 ohm loads here - more needed, or less, or just right? I’ve never really dialed in calcs for that cap position, but I am aware that it relates to rail voltage, possible output level, transient speed, and of course load.

5. Shall we look at potential failure modes now? One dual supply starts up before the other, one side of one supply fails, various latchup concerns (addressed I think!), anything else that comes to mind that should be addressed?

6. If an OPA1656 is chosen and the circuit output is shorted to ground, what is the source-drain current on each side of the complementary pair? That Diodes part with max 1V threshold should be considered 200mA maximum, but there are 400-500mA parts with higher Vth and similar capacitances available.

@Dualflip , @KA-Electonics.com , thoughts?

 

[thor.zmt]

The “CYA Version,” below. The “KISS version” (for direct coupled BJT amps) is forthcoming, but I want to lock in the bootstrap network first.

View attachment 109430

No caps on the gates, just a 1k resistor. Rest looks fine. CM range also looks ok.

Thor

 

[atavacron]

No caps on the gates, just a 1k resistor. Rest looks fine. CM range also looks ok.

Thor

 

[thor.zmt]

2. (a) Any potential issues with using such a high value - 6M8 - for gate bias, and (b) I had previously put in X7R for the sense caps at the output because we’re not listening to the result, but is it smart to use C0G there just because we want the values to remain accurate under full rail bias?

No, all good. Mosfet's have essentially gohm level.gate resistance. The capacitance it the problem.

3. (a) Does the LPF created by the 1K/10p at the gates effectively address note 2, subnote 2 in the quote of post #38, (b) is that even a necessary LPF with Mosfets rather than the original architecture from post #1, and (c) is a gate cap to ground not the best way to accomplish this?

As said, leave that off.

4. How’s 47u for 600 ohm loads here - more needed, or less, or just right?

This has more to do with the regulators that feed the circuit.

I think 100uF low ESR would keep current loops local if you add a few ohm in the power lines.

5. Shall we look at potential failure modes now? One dual supply starts up before the other, one side of one supply fails, various latchup concerns (addressed I think!), anything else that comes to mind that should be addressed?

If all parts are correctly sized, a missing PSU or a short should be tolerated well, but a missing Supply may cause DC on the output.

6. If an OPA1656 is chosen and the circuit output is shorted to ground, what is the source-drain current on each side of the complementary pair?

Double the short circuit current in the datasheet? Or Supply/max out Voltage divided by build out resistance times two, whichever is lower.

Thor

 

[user 37518]

So, drive the GND pin and you have a unity gain power buffer.

Needs to be charaterised of course. Only a thought, never tried...

Thor

Those parallel diodes are also connected to GND (I am assuming that the "triangle" ground is the ground pin you guys are referring to), so driving the GND pin would inject the same signal to both bases of the diff pair.

 

[thor.zmt]

Those parallel diodes are also connected to GND (I am assuming that the "triangle" ground is the ground pin you guys are referring to), so driving the GND pin would inject the same signal to both bases of the diff pair.

A differential pair with a VAS and Feedback loop will maintain the same voltage on both inputs, within a few mV at the most. Let me illustrate with a photoshopped version of the LM3886 schematic with the disabled input "faded out" and the "GND as input" illustrated:



You see a classic 3-Stage "Lin" type amplifier that operates closed loop, with unity gain.
Input differential, heavily degenerated, VAS, Output Stage.

Of course, +/- in would need to be connected to a voltage divider biasing them to:

1/2 +V - (-V)

So the diodes are off, normally. They will clamp the difference between input and output to ~ 0.5V.

They are included in the circuit as a protective measure, without them one side of this differential Amp for "mute" would swing almost rail to rail (without current flowing in it), results, especially in an IC would likely be interesting.

Thor

 

[user 37518]

A differential pair with a VAS and Feedback loop will maintain the same voltage on both inputs, within a few mV at the most. Let me illustrate with a photoshopped version of the LM3886 schematic with the disabled input "faded out" and the "GND as input" illustrated:

View attachment 109513

You see a classic 3-Stage "Lin" type amplifier that operates closed loop, with unity gain.
Input differential, heavily degenerated, VAS, Output Stage.

Of course, +/- in would need to be connected to a voltage divider biasing them to:

1/2 +V - (-V)

So the diodes are off, normally. They will clamp the difference between input and output to ~ 0.5V.

They are included in the circuit as a protective measure, without them one side of this differential Amp for "mute" would swing almost rail to rail (without current flowing in it), results, especially in an IC would likely be interesting.

Thor

You didn't have to blur out the first stage, I know what you meant. And yes, with unity gain, the output will be near the same value as the input and those diodes will be off, that is correct. However, I still don't like it. Any failure for the output to correctly follow the input in all cases, would produce unpredictable results, the diodes might never turn on, or maybe they will under certain circumstances. I don't want to suggest something similar to TIM, but I still don't like the fact that there is a potential path there. It will definitely work though.

 

[JohnRoberts]

A differential pair with a VAS and Feedback loop will maintain the same voltage on both inputs, within a few mV at the most. Let me illustrate with a photoshopped version of the LM3886 schematic with the disabled input "faded out" and the "GND as input" illustrated:

View attachment 109513

You see a classic 3-Stage "Lin" type amplifier that operates closed loop, with unity gain.
Input differential, heavily degenerated, VAS, Output Stage.

Of course, +/- in would need to be connected to a voltage divider biasing them to:

1/2 +V - (-V)

So the diodes are off, normally. They will clamp the difference between input and output to ~ 0.5V.

They are included in the circuit as a protective measure, without them one side of this differential Amp for "mute" would swing almost rail to rail (without current flowing in it), results, especially in an IC would likely be interesting.

Thor

the anti-sat diode clamps across LTP and even some discrete transistor inputs is to keep them from ever zenering under reverse bias (roughly 7V reverse bias). Low noise transistors can degrade after such zenering incidents.

More practically in a NF application the amplifier will recover more quickly after the outlier event that caused the unclamped LTP to saturate ends.

JR

 

[thor.zmt]

You didn't have to blur out the first stage, I know what you meant. And yes, with unity gain, the output will be near the same value as the input and those diodes will be off, that is correct. However, I still don't like it. Any failure for the output to correctly follow the input in all cases, would produce unpredictable results, the diodes might never turn on, or maybe they will under certain circumstances. I don't want to suggest something similar to TIM, but I still don't like the fact that there is a potential path there. It will definitely work though.

Similar protection diodes are present in many Op-Amps.

And "failure for the output to correctly follow the input" means either clipping or slewing. And both need to be avoided.

I could work, as said, I never got around to testing.

Thor

 

[user 37518]

And "failure for the output to correctly follow the input" means either clipping or slewing. And both need to be avoided.

Yes, that is exactly what I was referring to. The consequences of slewing are unpredictable on the behavior. It of course should be avoided, but following the engineering maxim the question becomes what happens if it happens. For example, output inversion is something that happens in some op-amps when the CM range is exceeded, it is something that should be avoided, but it is also that needs to be taken into account. I am not saying it wont work, and protection diodes are present, for example, at the input of the NE5534, but not in this same arrangement.

I am not saying it shouldn't be used, I am just posing questions.

 

[thor.zmt]

Yes, that is exactly what I was referring to. The consequences of slewing are unpredictable on the behavior.

Typical slew rate of LM3886 is 20V/uS. Using +/-36V Rails we need 1.3uS to reach the rail (in reality the Amp clips below rail), So a triangular wave with a cycle time of 5.2uS will be at the edge of slewing, that is 200kHz.

I tend to, as a matter of fact, limit Amplifier to a 5Hz-80kHz passband (-3dB - rule of 400,000), so slewing should be next to impossible.

It of course should be avoided, but following the engineering maxim the question becomes what happens if it happens.

The Amplifier will saturate all stages, but because the diodes are present, saturation will be less deep than without, though a better place for such diodes would be the output of the differential amp, not the input.

For example, output inversion is something that happens in some op-amps when the CM range is exceeded, it is something that should be avoided, but it is also that needs to be taken into account. I am not saying it wont work, and protection diodes are present, for example, at the input of the NE5534, but not in this same arrangement.

This is why I said that this has to be tried and needs to be FULLY CHARATERISED before using. No amount of theorising can substitute. We need to remember that is schematic is very much "principle" with a lot of detail omitted.

It could work, the result could be interesting for hybrid Amplifiers. Think Tube differential, perhaps cascoded, plus bridge/parallel LM3886 (2S/3P) in follower mode for 42V/220W into 8 Ohm. Fairly easy DIY project.

Thor

 

[user 37518]

Typical slew rate of LM3886 is 20V/uS. Using +/-36V Rails we need 1.3uS to reach the rail (in reality the Amp clips below rail), So a triangular wave with a cycle time of 5.2uS will be at the edge of slewing, that is 200kHz.

I tend to, as a matter of fact, limit Amplifier to a 5Hz-80kHz passband (-3dB - rule of 400,000), so slewing should be next to impossible.



The Amplifier will saturate all stages, but because the diodes are present, saturation will be less deep than without, though a better place for such diodes would be the output of the differential amp, not the input.



This is why I said that this has to be tried and needs to be FULLY CHARATERISED before using. No amount of theorising can substitute. We need to remember that is schematic is very much "principle" with a lot of detail omitted.

It could work, the result could be interesting for hybrid Amplifiers. Think Tube differential, perhaps cascoded, plus bridge/parallel LM3886 (2S/3P) in follower mode for 42V/220W into 8 Ohm. Fairly easy DIY project.

Thor

Slew rate is not the only thing to take into account, the signal bandwidth should also be considered. If you are planning on using this as a 36 V follower, that means that if you input is 36 V, an attenuation of 0.1 dB in the closed loop gain could be enough to bring the diodes into the verge of conduction. Which can happen well bellow the 3 dB cutoff point, and very well might happen inside the working or even audible range. I haven't looked at the specs regarding this pole, and there is not enough info on that schematic to determine it. This point might very well be very high, specially considering that it has unity gain, so it most definitely might not be a problem.

I have to check the spec sheet, otherwise, I am just guessing. Experimentation is definitely needed

 

[JohnRoberts]

Slew rate is not the only thing to take into account, the signal bandwidth should also be considered. If you are planning on using this as a 36 V follower, that means that if you input 36 V, an attenuation of 0.1 dB in the closed loop gain could be enough to bring the diodes into the verge of conduction. Which can happen well bellow the 3 dB cutoff point set by the pole of that 10 pF cap, and very well might happen inside the working or even audible range. I haven't looked at the specs regarding this pole, and there is not enough info on that schematic to determine it. This point might very well be very high, specially considering that it has unity gain, so it most definitely might not be a problem.

I have to check the speck sheet, otherwise I am just guessing. Experimentation is definitely needed

Talking about slew rate and bandwidth reminds me of the Marshal Leach paper leach aes paper

enjoy

JR

 

[user 37518]

Talking about slew rate and bandwidth reminds me of the Marshal Leach paper leach aes paper

enjoy

JR

Very nice, seems like a derivation any EE college student could've derived but didn't. You know what they say, sometimes the best ideas are the most simple.