Adding a third input to a differential input stage?

[maudio]

Would it be possible to construct a differential amp that has an additional input which it compares to the 2 standard inputs?

Why do I need that
I have an amplifier that has a differential output, it uses an output transformer with a grounded center tap. Now I need to feed the output signal back to some kind of input stage, where it should be compared to the input signal. The input stage has to provide around 30 dB gain as well, and plenty of bandwidth to prevent instability, but that is all solvable.

My current solution is to use 2 ltp's (top schematic): feed both outputs into ltp 2 to get the differential signal and feed that into a second ltp (ltp 1) comparing it with the input signal. That works allright, but any non-linearity in ltp 2 is not corrected by feedback which leaves me somehow uncomfortable. Also, more stages in the feedback loop isn't good for stability.

I have been thinking about ways to eliminate ltp2 (bottom schematic). Now I have to find a way to inject the input signal in ltp 1, in a way that it will compare this with the differential signal from the ouputs. I could use some resistors at the inputs to do the math, but that would require an inverted version of the input signal and requires buffering of the attenuators which are frequency compensated (they are 1:2000).

I was thinking about modulating the tail current of ltp 1, just wondering how linear that would be. In spice it looks allright. But maybe there are better solutions?

Btw I know it would make sense to add 30 dB gain to the power amplifiers input stage and combine everything there (shortest possible path). But I badly need nested feedback for stability.

[maudio]

Ok, one sunday experimenting further

Seems like the only reasonable way to do it is this. Actually I used this method before in a direct drive esl amp.

Only problem with it is that frequency compensation of the voltage dividers is awkward. I somehow have to compensate c2 and 3 (about 9pf) paralleling r9 and r10 (which are 10 meg). The other ends are attached to the output of a 1:150 stepup trannie with signal levels in the 10kV range. So unfortunately I do need such high values.

It can be done as shown, placing a (rather large) C between the emittors. Not a good idea as this bypasses r7,8.

A RC between the collectors will work a little better. Still not ideal. Neither way will give a perfect flat response for some reasons I don't understand yet.

Maybe I should buffer the output from the dividers first.

[chrispenycate]

Being used to blanced inputs as well as outputs I've done this in the past with a separate winding on the input transformer, This means that the input transformer is also inside the feedback loop (good) but I never found a circuit that didn't give me a virtual earth input (great for a balanced mix amp, less practical for a mic preamp)
Still, I've gone for floating (as against centretapped) outputs for many years now.

[powerbecker]

maudio:
"I have an amplifier that has a differential output, it uses an output transformer with a grounded center tap. Now I need to feed the output signal back to some kind of input stage, where it should be compared to the input signal. The input stage has to provide around 30 dB gain as well, and plenty of bandwidth to prevent instability, but that is all solvable."


You can try to simulate the attachment
Regards
Heinz!

[maudio]

Originally posted by powerbecker
You can try to simulate the attachment
Regards
Heinz!

Thanks for the schematic! In general this is what I built as well, using discrete differential amps instead of opamps.
The difference-amp you propose instead of the attenuator might improve things, I' give that a try.

As you may have guessed, it's a setup for driving electrostatics The clue being that I use a RL network instead of just a resistor before the transformer. This raises the minimum impedance of the transformer from something horrible (~1 ohm) to around 6 ohm over the audio range. While at the same time changing the transfer curve of the transformer into a 6 db/oct falloff, starting at a few khz, without peaking. It won't turn to 12 db before 200 khz. This gives enough margin to use the transformer for frequency compenstation instead of c5.

The happy result after closing the loop is a flat response to >60 khz using a 1:150 transformer, which normally would be impossible at such high stepup. I have a perfect 1khz square over the esl. It should be possbile to raise stepup even further allowing for more output voltage

I can have full output swing up till 2 khz which should be enough (I hope). And since the loading on the amplifier isn't dropping into the one ohm regions distortion is lowered as well. It actually sounds very promising. Still, lot of work to do..

[peufeu]

Remco uses tubes to direct-drive his ESLs :

http://ultranalog.com/9912/p25/

(just for inspiration)

[maudio]

I know his designs, have been working on direct drive myself for some time as well. The problem is that it's impossible to create enough drive voltage that way, for a fullrange panel you'll need something > 8 kV peak to peak and a lot of peak current as well. So that's why I am back at transformers and looking for ways around the problems with them.

Remco's amp is very good looking though

[peufeu]

So, you have one input signal (single ended) and two feedback paths.

Input is single-ended.
Feedback 1 is the "short" feedback path for HF, and it is single-ended
Feedback 2 goes through the transformer and is thus balanced.

You could only take feedback at one half of the transformer, which makes it single-ended.

However this isn't the best solution...

So either you convert the "long feedback" signal to single ended (via a differential amplifier) and everything works in single-ended mode ; or you convert everything to balanced.

Remember that LTPs work well (ie. low distortion) only when the differential voltage is very small (ie. when feedback keeps the differential voltage small and there is only common mode).

I guess either you use resistors for summing (rather, substracting) ; but then the impedance of your attenuator has to be considered (is it this hard ?) ; or you'll have a full serving of opamps : a differential amplifier (instrumentation amp) for LTP2 and a conventional discrete amp for the power amplification whose input stage is LTP1.

I'd use resistor arithmetic but you will need careful calculation of RC networks for compensation and shifting of feedback from one network to the next one.

Ideally the power amp + transformer should be seen as an integrator by the outer feedback loop for easy compensation.

[janneman]

Originally posted by maudio
Would it be possible to construct a differential amp that has an additional input which it compares to the 2 standard inputs?

Why do I need that
I have an amplifier that has a differential output, it uses an output transformer with a grounded center tap. Now I need to feed the output signal back to some kind of input stage, where it should be compared to the input signal. The input stage has to provide around 30 dB gain as well, and plenty of bandwidth to prevent instability, but that is all solvable.

My current solution is to use 2 ltp's (top schematic): feed both outputs into ltp 2 to get the differential signal and feed that into a second ltp (ltp 1) comparing it with the input signal. That works allright, but any non-linearity in ltp 2 is not corrected by feedback which leaves me somehow uncomfortable. Also, more stages in the feedback loop isn't good for stability.

I have been thinking about ways to eliminate ltp2 (bottom schematic). Now I have to find a way to inject the input signal in ltp 1, in a way that it will compare this with the differential signal from the ouputs. I could use some resistors at the inputs to do the math, but that would require an inverted version of the input signal and requires buffering of the attenuators which are frequency compensated (they are 1:2000).

I was thinking about modulating the tail current of ltp 1, just wondering how linear that would be. In spice it looks allright. But maybe there are better solutions?

Btw I know it would make sense to add 30 dB gain to the power amplifiers input stage and combine everything there (shortest possible path). But I badly need nested feedback for stability.

Seems you are looking for something like this:

Jan Didden

[maudio]

Hi Peufeu,

thanks for your suggestions!

So, you have one input signal (single ended) and two feedback paths.

Input is single-ended.
Feedback 1 is the "short" feedback path for HF, and it is single-ended
Feedback 2 goes through the transformer and is thus balanced.

Yes, and even a 3rd path for LF (transformer can't transfer DC). Which also takes carefull tweaking to prevent motorboating...
On the other side I can get away without a seperate HF feedback path so far by maximizing bandwidth of the power amp to around 350khz. The difference between first pole and 2nd pole in the transformer is just enough to allow sufficient amounts of feedback. The price being less local feedback around the VAS in the power amp as I have to reduce Cdom. Maybe in the future I'll try adding a seperate HF path, might give some further improvements

You could only take feedback at one half of the transformer, which makes it single-ended.

However this isn't the best solution...

agreed! As most esl-transformers are not very symmetrical at all.. (especially the EI types)

So either you convert the "long feedback" signal to single ended (via a differential amplifier) and everything works in single-ended mode ; or you convert everything to balanced.

I guess the balanced way would be better in terms of distortion cancelation, but also more difficult because there are 2 feedback paths which may not be 100% equal?

Remember that LTPs work well (ie. low distortion) only when the differential voltage is very small (ie. when feedback keeps the differential voltage small and there is only common mode).

That's exactly what I run into with my attenuator/ltp setup. Powerbecker's solution is better in that respect.

I guess either you use resistors for summing (rather, substracting) ; but then the impedance of your attenuator has to be considered (is it this hard ?) ; or you'll have a full serving of opamps : a differential amplifier (instrumentation amp) for LTP2 and a conventional discrete amp for the power amplification whose input stage is LTP1.

The attenuator has given me quite some troubles, it must withstand very high voltage while remaining very linear. Those special high voltage rated resistors are anything but that. It has to have high resistance to keep current draw and dissipation low. Which is where stray capacitances comes in. I tried several things, even home-made capacitors, before settling to 20 ordinary 499k metal film resistors in series, each paralleled with a 180pf/630V capacitor, resulting in 10Mohm//~10pf. Bulky but reproducable. But at 1:2000 I need a 5k//20nf impedance to hook it up to.
My concern with instrumentation amps is the large amount of active devices in the feedback loop. But maybe I shouldn't worry about that too much.

I'd use resistor arithmetic but you will need careful calculation of RC networks for compensation and shifting of feedback from one network to the next one.

I think I hav some spice-ing to do