Discrete Voltage Regulator

[olafmatt]

Yes, the negative regulator uses transistors of the opposite sex. So the pass transistor I use is MJE3055.

Will try your recommendations with C104 and report back. I can't get the MJE171 easily, maybe I'll try BD140 and see whether it changes anything. With my 50mA current draw I could probably also go with BC560.

Olaf

 

[olafmatt]

Small update:

Changed the pass transistor to BD140 and oscillation got worse. Then tried series R with C104 and oscillation got worse. Decreased C104 to 100uF and HF oscillation went away, but I got oscillation in kHz range instead. Finally increased the resistor in the base of Q109 to 470 Ohms and oscillation was gone (with 2200uF for C104 and no series R, but still with BD140). Have to try different loads now to see whether it's really stable into all sorts of loads.

Olaf

 

[bcarso]

Anything with this prodigious amount of loop gain is going to be very sensitive to details of grounding.  It would be helpful if we could see more details of your PC layout.

If things oscillate more vigorously with loading, it is possible that it's a shift in gain bandwidth product of some of the transistors.  But I would suspect current-induced voltage drops in the grounding system that effect positive feedback.

I had a power amplifier once based around some STMicro integrated parts (TDA2007A, notorious for instability as it was) that oscillated with astonishing vigor due to the choice of output Zobel network grounding locations.  It went away with a movement of about a centimeter, and this with a hand-wired prototype using 20 gauge wire.  I was using the part very diferently from the datasheet recommendations btw.

But when it was stable it delivered extraordinary performance.  Harman was buying them for under 30 cents IIRC, which is pretty nice for two times six watts and ~0.01% THD+N.

 

[Samuel Groner]

I don't see anything obviously wrong with the layout. Ideally one would connect R114/R115 directly (i.e. with its own trace) to the terminals of C104 though, not via the output current/ground return path. Brad, any idea?

Samuel

 

[bcarso]

Swamped right now, but I would suggest fattening the big traces with solder and/or some wire and seeing what it does to the load current sensitivity versus oscillation.  Also look at the noise level at high frequencies and look for resonances/correlation times.

You could also simulate with resistances and inductances of the traces to see where the major sensitivity is.  Look for the paths that produce voltage drops at sense points versus the output loading.

Wish I had it on my bench---I'll bet I could locate the problem(s) quickly.

 

[olafmatt]

Simulation shows that the higher the resistance between C104 and R114 the higher the value of the resistor in the base of Q109 has to be in order to avoid oscillation.

I think I will just run a separate trace for the load from the collector of Q109 to avoid additional voltage drop when load current increases.

Olaf

 

[Samuel Groner]

I tried to reproduce the instability problem today, with not much success. Re-routing the feedback connection to something similar to your layout didn't initiate any instability (tested up to 450 mA); same result for a faster output transistor. A heavy TO-220 part showed some instability at 2.3 MHz, but a 120 Ohm base resistor (didn't try any lower values) completely cured things up to 900 mA.

Samuel

 

[bearzbear]

It's only 8 years old now... but wondering if the "final" schematic exists for reference.
The link is no longer "live".

Has this design been supplanted by anything else in the interim?
IC or discrete.

Curious...

                          _-_-