Twin Servo Mic preamp - oscillation at maximum gain

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
squarewave said:
So taking that into consideration you are right about R6 returning to A1 COM because most of the output current of A1 is coming form FB and not LOAD.
This is not the only reason. One needs to create a reference node that is used by the input signal, the NFB loop and the servo. The second stage must seek its remote ground where the signal comes from; again this is the same as the input stage reference, so the NFB loop and the servo of the 2nd stage must use this remote ground. Now the load must be returned to the decoupling caps common point, because that's the point of least impedance for the output current loop. Parasitic currents are by nature spiky, contain high HF content. returning them via PCB traces or wires is inefficient.

But there's no reason for A2 LOAD to be connected to any grounds of A1.
Did I suggest that?

EDIT: OK I made a wrong shortcut. It's the return of the OT that must be tied to the 2nd COM.

But still, in this particular circuit, I believe it might actually be better to use a separate ground for R3, R6, A1 COM and BLK because A2 LOAD throws a lot of current
It does, but where? The most important factor is to make sure the output current loop is as short as possible and does not interfere with the 1st stage input.

You say connect transformer BLK to A2 COM? That makes zero sense to me. BLK is just for shunting common mode noise to ground. Capacitively coupling the +80dB output of A2 LOAD to A1 +IN through the transformer shield does not seem like a good idea to me.
Agree. I should edit my previous post.
 
I must admit every time I think about this sort of issue I end up being slightly uncertain.

There is very little current flowing between A1 and A2. The input impedance of A2 at R10 is 300K. So if separate grounds are used, it follows that all currents going into A1 must return through it's ground and that all currents going into A2 must return on it's ground. So send follows return and all is well. Then, by my possibly flawed rationale, there could be some benefit from decoupling the +80dB output from the input by using separate grounds.

On the other hand (and what I believe you are saying is that) any parasitics or more generally "noise" from A2 will be dropped across the separate ground for A2 and therefore create a voltage difference between +IN and -IN of A2 because one is sourced from the output of A1 and the other is sourced from output of A2.

So I don't know. This is one of those cases where I would have to actually try it and measure the noise carefully before I make up my mind. It could also be that it just doesn't make a difference either way. And every case is unique. This circuit in particular is special in that it's two stages so it's not clear that generalizations can be derived anyways.
 
squarewave said:
There is very little current flowing between A1 and A2. The input impedance of A2 at R10 is 300K.
  Indeed; I'm more concerned about currents that result from the operation of the DOA. They need to be constrained to the shortest loop.

So if separate grounds are used, it follows that all currents going into A1 must return through it's ground and that all currents going into A2 must return on it's ground. So send follows return and all is well. Then, by my possibly flawed rationale, there could be some benefit from decoupling the +80dB output from the input by using separate grounds.
These two separate "grounds" will have to be connected galvanically. How can you do it in a way that ensures there is no differential voltage between these two "grounds"?
I hear quite often that the magic receipe is "star ground". The system ends up being two separate stages, with significant differential voltage between the two references. The common answer is using a differential connection for signal. Unless the output stage is reconfigured as a diff amp, the actual solution is to connect the input reference point of the 2nd stage (R12 & R14) to COM of the 1st stage in the shortest manner.

On the other hand (and what I believe you are saying is that) any parasitics or more generally "noise" from A2 will be dropped across the separate ground for A2 and therefore create a voltage difference between +IN and -IN of A2 because one is sourced from the output of A1 and the other is sourced from output of A2.
That's correct.

it's not clear that generalizations can be derived anyways.
Rabbit-out-of-a-hat solutions are a common disease in most disciplines. Nothing beats analysis.
 
I believe the reply n.o. 7 from Audio1man has the highest S/N ratio 'till now. 
Nobody mentioned ground planes on top and bottom layers, bonded with  minimum 10 vias per square inch ;)?
The real question to me is is the first servo really needed in the original schematic ....
 
A lot of great infos guys! Thanks

I've seen a few schematics where they added a large capacitor before R6 & R12 ground connection... anyone could share some light on this?

Thanks

Twin-Servo-Cap.jpg
 
elskardio said:
A lot of great infos guys! Thanks

I've seen a few schematics where they added a large capacitor before R6 & R12 ground connection... anyone could share some light on this?

Thanks
These caps reduce the gain at DC to unity, so basically output offset is similar to input offset, instead of being increased by the gain factor. It defeats somewhat the raison d'être of servos, which is to avoid electrolytic capacitors in the signal path. NFB loop is in the signal path. One may argue that level in the NFB loop is smaller than at the output, though.
 
Audio1Man said:
I also would add a local RC to each of A1, A2 & A23 on the PS volt lines. Typ decoupling of 10 to 100 ohms and 47uf to 100uf and .1uf bypass on each leg are good practice.
Indeed, it's a good recommandation; it actually does not rely on PSU rails stiffness, rather on closing a very short loop for output currents. Pls note that 0.1uF caps are already in the 990 (joined at COM).
 
abbey road d enfer said:
These caps reduce the gain at DC to unity, so basically output offset is similar to input offset... It defeats somewhat the raison d'être of servos

Hence my confusion  ???

Might as well ditch the servo and add input & output capacitors.

I started revising my pcb and will follow Audio1man good advice. I'll post more information when I'm almost done
 
Hi All
Power Supply rails stiffness tests.
I DEBUG DUT’s using several tests. One of my best is that I drive an AC current into the power supplies and look @ FFT of the output signal of the device. I do this by using the AP generator with a blocking cap and a series resistor to limit current signal to drive the power supply. I repeat this for all supplies and use different current signal frequencies.
Duke
 
elskardio said:
Hence my confusion  ???

Might as well ditch the servo and add input & output capacitors.

I started revising my pcb and will follow Audio1man good advice. I'll post more information when I'm almost done
Actually the 1000uF caps in the NFB loop are submitted to a current that is Vout/[Gain.43]. When they are in the output, the current is Vout/Zpri; worst case Zpri is about 150r. Could be about 15 times, distortion will follow. However in many cases load is 10k, so Zpri is 2.5k, and distortion similar.
But offset is small anyway, as long as resistors are kept small enough, so servos are not a strict necessity.
I would think caps+servos is belt and braces.

The reason for wanting to minimize output DC offset is to minimize the effects of DC in the xfmr's primary. In the absence of servos, but with caps in the NFB loop, Output DC offset can be estimated at a few mV (probably under 10), resulting in less than 1mA. This is not enough to change significantly the op point on the B-H curve of the magnetic core (capable of more than 100mA peak).
 
Back
Top