Help me taming an oscillating audio amplifier board

[dzseki]

I've got a pair of audio amplifier boards, those have been used in a hungarian made active studio monitor (Beag HEC45). Back in the 80's this was a state of the art stuff, at least in the eastern block...
Nevertheless the amplifier cuircuit is also very interesting, and is potentially can be a "good sounding" amplifier, it is a rather complicated circuit but well thought out IMHO. See the schematic diagram attached.
I got the boards defectly, but they are original boards, not a replica or remake (to make sure for you that the PCB layout is as it should be). There were a handful burnt resistors and shorted/open diodes and transistors in each module, all in the power output stage. All components were replaced in pair, replaced transistors got matched for hfe, etc.
After replacing the deffective components now I am strugling with heavy oscillation in both modules. The oscillation frequency is at about 100-500kHz (see later) and a few Volts peak to peak.

I have tried quite few things, but with little success so far:
-As I've mentioned these board are original, I assume these were working properly at some point. All components in the input stage and VAS looks original and identical on both boards.
-The power supply that I'm using is linear, the input stage runs from +/-47V for this I've built a stabilized supply, the power output runs of two 22000uF smoothing caps and two 27R series resistor for current limiting in a possible hiccup.
-The board currently runs on my bench, no case etc.
-The board as is oscillates at ~200kHz. Since the output stage is a compound-darlington I've removed the output transistors (2N3055 and BDX18), then the oscillation frequency increased to 500kHz.
-I've tried to further simplify the output stage by separating T15 and T16 from the rest of the output, and by dividing R44 into two halves, I reconfigured the global feedback network to the middle point of R44. In this case the oscillation frequency got down to ~100kHz, but still there.
-I've tried to play with different grounding schemes -> no difference at all.
-Replaced all electrolytic capacitors related to power supply bypassing, even increased values, added more capacitors etc. ->nothing
-If I short together the inverting and non-inverting inputs of the amplifier, the oscillation stops (at least then...)
-Removed C5 ->nothing
-Even dared to modify C22 to 100nF instead of 10nF, then in steady state the oscillation was absent, but as soon as I put some input signal the oscillation (this time a very distorted one) kicked in and remained afterwards.
-Now I only focus on one board with my experiments, but initially I confirmed that both boards oscillates roughly about the same way, when fully assembled.

I am starting to run out of ideas how to diagonse further...

[bob91343]

Perhaps the original transistors were selected for low gain.

[xavier60]

Maybe try some conventional Miller style compensation around the VA stage.
A small capacitor from the Collector of T9 to the Base of T5, same for T10 and T6. Start with 100pF.
C3 and C4 should possibly be removed.

[xavier60]

Could be a mistake in the schematic. The feedback should be taken from the junction of the Emitter resistors.

On second thoughts, the RC is too short to cause a problem.
And also the dividing of R44 .

[timenutgoblin]

-Replaced all electrolytic capacitors related to power supply bypassing, even increased values, added more capacitors etc. ->nothing

-If I short together the inverting and non-inverting inputs of the amplifier, the oscillation stops (at least then...)

From what I can see, there are 3 power supply rails: +/-47V, +/-34V and +/-15V. The +/-15V is for IC1 (pins 4 and 7). The +/-15V comes from zener diodes D1 and D2 . Did you check C6? It looks like an electrolytic capacitor (100uF). It is located next to IC1.

[dzseki]

Perhaps the original transistors were selected for low gain.

If that would be the case, then when I left T15 and T16 as sole output transistors, the oscillation should have stopped, but it didn't.

[dzseki]

Maybe try some conventional Miller style compensation around the VA stage.
A small capacitor from the Collector of T9 to the Base of T5, same for T10 and T6. Start with 100pF.

Will try.

C3 and C4 should possibly be removed.

Isn't those caps for taming high freq oscillations? Eg. C3 with R9 shunts R5 at high frequency, thus lowering the gain of the stage. Removing those wouldn't make the things even worse?

In the meantime I've increased C5 to 47pF, but that didn't change a thing either.

[dzseki]

-Replaced all electrolytic capacitors related to power supply bypassing, even increased values, added more capacitors etc. ->nothing

-If I short together the inverting and non-inverting inputs of the amplifier, the oscillation stops (at least then...)

From what I can see, there are 3 power supply rails: +/-47V, +/-34V and +/-15V. The +/-15V is for IC1 (pins 4 and 7). The +/-15V comes from zener diodes D1 and D2 . Did you check C6? It looks like an electrolytic capacitor (100uF). It is located next to IC1.

U1, C6 along with R54  forms a DC servo (as there is no capacitor in the feedback path), it is about the calmest part of the circuit, C6 has been replaced already.

[xavier60]

So called "dominant pole" compensation should be applied at only one location in the signal path otherwise the phase lags will accumulate causing oscillation.
It can be applied the way it is now with R9, C3, R10 and R4. As Miller feedback between the output and input of the VA or direct capacitive loading of the VA's output.
The presence of R9 and R10 cause the minimum voltage gain from the input of T1/T2 to the VA output to drop to a minimum of 7 at high frequency by my rough calculation. Maybe this is the cause of the marginal instability.

[xavier60]

The DC conditions of the input stage can also affect stability. I have calculated some approximate voltages.
Voltage drops across R5, R17, R6, R18 should be 4.6V.
Across R23, R24 should be 3.4V.
Also check that the 5.6V Zener diodes are regulating properly.

[dzseki]

So called "dominant pole" compensation should be applied at only one location in the signal path otherwise the phase lags will accumulate causing oscillation.
It can be applied the way it is now with R9, C3, R10 and R4. As Miller feedback between the output and input of the VA or direct capacitive loading of the VA's output.
The presence of R9 and R10 cause the minimum voltage gain from the input of T1/T2 to the VA output to drop to a minimum of 7 at high frequency by my rough calculation. Maybe this is the cause of the marginal instability.

I have removed C3 and C4, but then the oscillation became even wilder, it was changing all the time.

[xavier60]

So called "dominant pole" compensation should be applied at only one location in the signal path otherwise the phase lags will accumulate causing oscillation.
It can be applied the way it is now with R9, C3, R10 and R4. As Miller feedback between the output and input of the VA or direct capacitive loading of the VA's output.
The presence of R9 and R10 cause the minimum voltage gain from the input of T1/T2 to the VA output to drop to a minimum of 7 at high frequency by my rough calculation. Maybe this is the cause of the marginal instability.

I have removed C3 and C4, but then the oscillation became even wilder, it was changing all the time.

Did you also add the 100pF capacitors that I mentioned?

[dzseki]

The DC conditions of the input stage can also affect stability. I have calculated some approximate voltages.
Voltage drops across R5, R17, R6, R18 should be 4.6V.
Across R23, R24 should be 3.4V.
Also check that the 5.6V Zener diodes are regulating properly.

I could only check this by having the inverting and non inverting input shorted, but the the output settled at -5V roughly, therefore there is some slight imbalance (or was that the DC servo in act?).
Nevertheles, the DC voltages are roughly corresponding to what you have calculated.

[StillTrying]

-The power supply that I'm using is linear, the input stage runs from +/-47V for this I've built a stabilized supply, the power output runs of two 22000uF smoothing caps and two 27R series resistor for current limiting in a possible hiccup.

Have you bravely tried the +/- 34V output supply without the 27Rs.
With only 2.2uF on the 34V supplies you might get away with 2 quick blow fuses instead.
Or that might not be anything to do with it.

[dzseki]

So called "dominant pole" compensation should be applied at only one location in the signal path otherwise the phase lags will accumulate causing oscillation.
It can be applied the way it is now with R9, C3, R10 and R4. As Miller feedback between the output and input of the VA or direct capacitive loading of the VA's output.
The presence of R9 and R10 cause the minimum voltage gain from the input of T1/T2 to the VA output to drop to a minimum of 7 at high frequency by my rough calculation. Maybe this is the cause of the marginal instability.

I have removed C3 and C4, but then the oscillation became even wilder, it was changing all the time.

Did you also add the 100pF capacitors that I mentioned?

Sorry, I have misunderstood this test.
Indeed adding the two 100pF caps to the cascode stages, and removing C3 and C4 tamed the oscillation! Interestingly T5 and T9 (or T6 and T10 for that matter) are quite far away on the PCB so it is rather difficult to safely install such small cap.
What do you think, can I get away simply playing with C3-R9 and C4-R10 as well?

[xavier60]

So called "dominant pole" compensation should be applied at only one location in the signal path otherwise the phase lags will accumulate causing oscillation.
It can be applied the way it is now with R9, C3, R10 and R4. As Miller feedback between the output and input of the VA or direct capacitive loading of the VA's output.
The presence of R9 and R10 cause the minimum voltage gain from the input of T1/T2 to the VA output to drop to a minimum of 7 at high frequency by my rough calculation. Maybe this is the cause of the marginal instability.

I have removed C3 and C4, but then the oscillation became even wilder, it was changing all the time.

Did you also add the 100pF capacitors that I mentioned?

Sorry, I have misunderstood this test.
Indeed adding the two 100pF caps to the cascode stages, and removing C3 and C4 tamed the oscillation! Interestingly T5 and T9 (or T6 and T10 for that matter) are quite far away on the PCB so it is rather difficult to safely install such small cap.
What do you think, can I get away simply playing with C3-R9 and C4-R10 as well?

Try removing the 100pF caps and installing C3 and C4 again with the series resistors bridged out. If it's still oscillating, increase the size of C3 and C4.
I would like to see the layout.

[Kleinstein]

It is not a surprise to see oscillation in this circuit. I would be more suprised if such a circuit would not oscillate. It would probably be a good idea to enter the ciruit to a simulation like LTspice to see the trouble - it likely takes more than just changing a compensation cap, more like redo the calculation for the compensation. A main weak point is slowing down the loop at multiple points: There is C3/C4 - at least with series caps, so it would not be effective at the very highest frequencies.  R53 and C22 also form a low pass - usually such a combination is used to tame the load impedance, but not in the feedback. So my first guess would be moving the feedback to the other side of R53.

Besides the main loop, there is also a chance the Sizlaki type output stages may show local instability.

Overall it looks like a combination of some good ideas and some bad ideas - more like a beginners try. Usually the designs from the east were much better, using way fewer expensive parts and based on solid math, not pure guesswork. To me this looks more like a multiple choice US design.

Besides the oscillation, having only 2 power transistors, but a +-34 V main supply is a path to thermal or SOA failure. The amplifier may need a high impedance load, like 16 Ohms.

[dzseki]

Try removing the 100pF caps and installing C3 and C4 again with the series resistors bridged out. If it's still oscillating, increase the size of C3 and C4.
I would like to see the layout.

-Bridge out -> no joy
-Added +3.3nF to C3 and C4 initially looked stable, but later (with signal) ticked into oscillation.
-Added +10nF instead, remianed stable. Now I am not sure why, but with this the sine looked  kinda distorted at higher freqs, but granted this board was still running without the 2N3055 and BDX18, so might need further investigation, or have some faulty components that I missed before. but one way or another the oscillation looks tameable now.

I tried my other board which is fully assembled (but also had the oscillation problem) and installed the +100pF Miller cap modification (with removing C3-4) for starter, with that I could not see any apparent distortion on this board on sine and looked stable. The -3dB frequency was about 150kHz which is still plenty.

Please find the layout attached. It is reverse engineered from the original PCB (not my work), and may contain faults, but should be extremely close to the original.

[dzseki]

It is not a surprise to see oscillation in this circuit. I would be more suprised if such a circuit would not oscillate. It would probably be a good idea to enter the ciruit to a simulation like LTspice to see the trouble - it likely takes more than just changing a compensation cap, more like redo the calculation for the compensation. A main weak point is slowing down the loop at multiple points: There is C3/C4 - at least with series caps, so it would not be effective at the very highest frequencies.  R53 and C22 also form a low pass - usually such a combination is used to tame the load impedance, but not in the feedback. So my first guess would be moving the feedback to the other side of R53.

Of course I have entered the circuit into a simulator (TINA), without the input LF and HF filters in the model it shows a rather small peaking at about 500kHz, but this of course does not take into account the PCB parasitics.

Red trace is the midpoint of R44, green is the output.

According to the simulation by increasing C5 to 10pF would fully remove the peaking, while in reality I've even tried 47pF, but the oscillation was still there.

R53 and C22 is set to 1.5Mhz, it is indeed odd to have it there, but it hardly change anything since the cut off frequency is so high...

Overall it looks like a combination of some good ideas and some bad ideas - more like a beginners try. Usually the designs from the east were much better, using way fewer expensive parts and based on solid math, not pure guesswork. To me this looks more like a multiple choice US design.

Besides the oscillation, having only 2 power transistors, but a +-34 V main supply is a path to thermal or SOA failure. The amplifier may need a high impedance load, like 16 Ohms.

OK, but this circuit and board was in production, so perhaps they had this circuit working properly.
The manufacturer of this stuff was a well respected studio gear manufacturer in the eastern block. The whole speaker was a little bit of tour de force having a lot of fancy features.

The module is rated for 50W @ 4Ohm, in my experience with +/-34V of supply and a pair of "115W" transistor this is just about right.
The original speaker contained four of these amplifiers, two bridged for the woofer, and the mid and high had one board each.

[Kleinstein]

The layout also deserves a : the 2 transistors for temperature compensation should be couples to T15-T18, not the ouput power transistors.

When warm the current limit part for protection may engage from some 2 A on - this is why I suggested high impedance. With 4 Ohms one would see clipping from the current limit quite early. 50 W with 4 Ohms. 50 W sine with 4 ohms would only need some 20 V peak output voltage, but allread some 5 A. 8 Ohms may be OK.

The amplifier design has a weakness with the standing current in the voltage amplifier stage. It depends on the offsets of the NPN and PNP input stages. With too low a biasing current the speed would be lower and the amplifier may oscillate. With too much current, chances are the BC640/BC639 would not last very long with the high supply. There is an extra DC servo for the total input offset (indicating that offset may be a problem), but not for the individual offsets.

From a quite view, the loop gain looks rather low and thus relatively high distortion is not a surprise. A problem could be the adjustment of the standing current. The ouput staged are engaging hard and the cross over distortion is thus relatively locallized and thus diffcult to get small with just the right current - especially as the biasing current will tend to drift with the shown layout. At least it would not run away, but more like a reduced bias when the amplifier gets warm.

The simple AC simulation usually only test one half of the output stages. So one should do the simulation with a few DC offsets, like 1 K , 0 V , -1 V at the output.

1.5 MHz cross over means the phase shift may alread start from some 200 kHz on - well early enough to favor oscillation at some 500 kHz. 

[xavier60]

Although it still leaves a big mystery, seems that it's the happiest with Miller compensation. The cap values can be experimented with and I'd suggest putting C5 back to original.
The idea is to have good stability margin while avoiding slew limiting. Slew limiting makes high frequency sine waves start to look triangular.

[xavier60]

The SOA protection limits at about 2A at 0V output and 8A at 30V. Barely adequate for 4Ω single driver.
I agree about the temperature compensation.

[Zoli]

Change T9 to BF421, T10 to BF420; I've fought this design faullt 25 years ago on different amps, and that was the solution.
Details: BC639/40 are rated for 80V, and in this case they are running from 94V(+-47V); after some years, they get tired, and start to protest by oscillating .
Or, in more straighforward wording: they are shitty choice for this function; if I see something like that, I routinly change them, without any second thought.
Good luck to find some BF420/21(or I meant BF620/21? Is too late to remember...).

[magic]

Am I reading correctly between the lines that people have cloned this design and it worked for them?

I could only check this by having the inverting and non inverting input shorted, but the the output settled at -5V roughly, therefore there is some slight imbalance (or was that the DC servo in act?).
Nevertheles, the DC voltages are roughly corresponding to what you have calculated.

That doesn't seem right. The DC servo should bring it to zero.

Even if you magically cure oscillation and re-enable the main loop, I think some offset would remain. The servo has full authority over output DC level and ±13V output swing applied right to the noninverting input.

Perhaps start with sorting out the DC condition?

[xavier60]

With the inputs shorted to each other, the DC servo can only supply common mode voltage to the inputs.
The output on most designs would swing hard + or -, this design doesn't because the VA's outputs are resistively loaded to ground.