Low-ish frequency oscillation on bench, but not in console

leigh · Sep 9, 2014

As part of my Trident Series 65 board overhaul, I was looking for ways to clean up the noise in the Aux section. However, there's a strange oscillation happening there that I can't explain, and since it's not a grounding issue (AFAIK), I'm starting a new thread for it separate from the Trident grounding one.

Background: The Trident "echo return" strip handles 8 aux outputs (as well as 4 effects returns). The circuit for each aux output is fairly simple: a virtual earth summing stage (one TL071), followed by a pot for output level and a solo switch for routing, and lastly into a quasi-floating balanced output driver stage (two more TL071s).

Attached is an example schematic of just the balanced output stage.

Here's what's happening: installed in the console, the aux outs all seem to behave OK. On the bench, they mostly behave ok, except for aux 6. Aux 6's output has a 12VAC, (sort of) square wave, running at about 90Hz, showing up between each output leg and ground. When measured differentially (between + and - output legs), this wave is not seen, so it's common mode between the two sides of the output driver.

This oscillation is also not seen at C4, right before the signal splits to the two halves of the output driver circuit. So the trouble is starting after that point.

While the console's power supply is ±18, my bench supply is ±15. This oscillation at 12V would seem a bit on the low side for rail-to-rail oscillation, even for a TL071?

I have swapped out different TL071's, tried better spec'd opamps, and swapped in better IC sockets, all to to avail. Also tried terminating the output with a 680 ohm resistor, which didn't stop the oscillation either. Grounding the output leg that's not hooked up to the oscilloscope does stop the oscillation.

Any ideas? I'm new to understanding this output driver circuit. What's the most likely suspect for turning "cross-coupled feedback" into "oscillator"?

SSLtech · Sep 9, 2014

What ARE you doing with the other (cold) leg referenced to ground? -leaving it unterminated? Don't.

Instead of just measuring one leg and letting the other one float, try terminating the other to ground through some resistive value... Cross-coupled outputs can do very unpredictable things when they 'float freely'. -The "I'm doing my job right" feedback is a differential comparison of the two outputs AFTER the build-out resistances. Anything common mode is therefore ignored, and is not suppressed. Referring to ground tends to provide more predictable results.

The 'preferred' cross-coupled output uses THREE op-amps, not two. The two-opamp version (as used by Soundcraft and SSL for a while) is a little unpredictable.

Keith

leigh · Sep 10, 2014

SSLtech said:
Instead of just measuring one leg and letting the other one float, try terminating the other to ground through some resistive value... Cross-coupled outputs can do very unpredictable things when they 'float freely'.

Thanks for the idea Keith, I'll give that a try tomorrow.

I did try terminating the output, across the + and - outs, with a 680 ohm resistor, but that didn't stop the oscillation. And, a true balanced output into a differential input shouldn't have to reference ground at all... yes?

Or, is that a weakness of a "pseudo-floating" circuit like this, that its output isn't stable unless it gets to reference ground?

leigh · Sep 10, 2014

Terminating the other leg with a 12k resistor to ground did indeed stop the oscillation.

That was some frustrating, time-consuming troubleshooting, and all the more frustrating because this wasn't my first time being fooled into thinking that a pseudo-floating/EBOS circuit would behave well. Dooops.

It also probably explains what those 12k resistors are doing, wired onto the XLR output jacks, between pin 2 & ground and pin 3 & ground. They are there to ensure stability, even if one leg is left hanging.

Thanks again for the idea, Keith.

leigh · Sep 10, 2014

SSLtech said:
The 'preferred' cross-coupled output uses THREE op-amps, not two. The two-opamp version (as used by Soundcraft and SSL for a while) is a little unpredictable.

Like this? (This is from the Allen & Heath Sabre.)