stability in audio amplifiers - what were they thinking?

[nev23]

I was going to put this in repairs, but I think it is raises some wider questions.
In the picture is an audio power amp from about 1978 by AB Systems, Inc.
I have replaced all the output devices with modern ones and I get a little oscillation near clipping on the positive peaks of a signal. This occurs with or without a dummy load connected.
I looked at the diagram and realised that it has many capacitors that I wasn't used to seeing in audio amps, arrowed in red.
Any suggestions as to what they all do and why they are there, or is it just a bit of a mess?
I can see the 47pF is the so-called Miller capacitance, but the others I have not seen before.
The 0.05 uF caps around the V-I protection transistors are in the wrong place, but removing them didn't help.
I can't see D17 and D18 being much help, either.

Any thoughts?

[T3sl4co1l]

Well, yeah.  "Modern" devices have much higher fT, even for classic types like 2N3055 (which haven't been "classic" since the 80s, I think).

I'd suggest using exact equivalents if at all possible; and if not, then add C-B capacitors until it goes away.  Examples might be ~1nF around output transistors, 100pF around drive transistors, etc.

Looks like they're already controlling poles and zeros carefully, with the combination of C and R+C spread about.

Tim

[German_EE]

You sometimes see RC networks like that on the outputs of audio amplifiers so putting them earlier in the circuit as well just ensures greater stability. Circuits like that often have VERY long speaker leads so there will be all sorts of extra stuff put in 'just in case'.

[Kleinstein]

Faster transistors at the output should not be a problem to stability, it tends to even make a feedback circuit more stable.
The 1 nF cap around Q15 might need adjustment if the negative side shows local instability.

The diodes D17/D18 might help with linearity before feedback a little, if they have a suitable forward voltage.

[Pjotr]

As mentioned, use the same transistors. if it's just for repair. Although modern 2N3055 are not the same as "classic" ones, FT didn't rise much. Stability in those days was usually a matter of trial and error, but carefully crafted. Using modern faster transistors can be done but requires a whole re-evaluation of the stability and where the pitfalls are. It's not simply to say.

[Pjotr]

Faster transistors at the output should not be a problem to stability, it tends to even make a feedback circuit more stable.

True, but OP talks about oscillation near clipping. At that point things go into saturation and loop stability can change in unexpected ways with other transistors.

[dom0]

The 50 nF between the emitters of Q1 and Q2 bypasses D3 and D4, increasing gain at higher frequencies. D3/D4 are meant for, hm, not sure: that's the actual uncommon sight. I see no real reason, if the wanted to improve balance they should have used resistors instead ; they won't cancel BE-nonlinearity since they're in series, they'll increase it instead.

The collector load of Q1 and Q2 is interesting. Q3 is a current source, and Q6 copies that current onto Q2's collector - but only very roughly (on a first glance, at least). Q6 and D8 provide some offset to Q7 at the same time.

The 1 nF across CB of Q15 stabilize the Sziklai configuration of Q15 working on Q17/19/21/29/25. It isn't a long time constant due to the 100 ? in parallel to Rbe of Q15 and the others.

The 470 pF serve a similar purpose, the 36 ? raise the base drive impedance of Q14 to avoid self-oscillation.

The 20 pF in the feedback network are normal feedback technique ("type 2 compensator", in academia speech).

The 47 pF from Q7's collector to Q6's collector add some gain peaking in that stage. misread the schematic. It reduces gain, i.e. miller compensation, since Q7 gets it's input from Q6's emitter.

The 50 nF at Q10 and Q11 are too small for a SOA time constant (~5 µs), so it probably quenches an oscillation when it hits current limit.

[Audioguru]

It seems that the problem only occurs when the amplifier is driven into clipping. Then turn it down a little!
If you like hearing acid rock with the amplifier clipping like crazy then you probably also like a little oscillation added to the noise.

[Pjotr]

It seems that the problem only occurs when the amplifier is driven into clipping. Then turn it down a little!
If you like hearing acid rock with the amplifier clipping like crazy then you probably also like a little oscillation added to the noise.

It is often a design choice: Maximise loop gain and allow some oscillation at clipping (usually at recovery) or overcompensate to avoid oscillation at clipping, but this reduces loop gain.

[T3sl4co1l]

It seems that the problem only occurs when the amplifier is driven into clipping. Then turn it down a little!
If you like hearing acid rock with the amplifier clipping like crazy then you probably also like a little oscillation added to the noise.

This is a non-solution...

Oscillation under any operating condition is a design fault.

Tim

[uncle_bob]

It seems that the problem only occurs when the amplifier is driven into clipping. Then turn it down a little!
If you like hearing acid rock with the amplifier clipping like crazy then you probably also like a little oscillation added to the noise.

This is a non-solution...

Oscillation under any operating condition is a design fault.

Tim

Hi

... but was it present when the amp was brand new?

The caps in the front end are very conventional high frequency roll off parts. The idea is / was to set the amp bandwidth early so it would not be impacted as much by slew rate limiting in the output stages. The later caps are likely a stab at equalizing the slew rate going positive and negative in the output stage. With faster output devices, that process likely will need to be revisited. The issue is that the negative side of the amp is fed from the collector of the output devices. The positive side is fed from the emitter of that bank. They don't quite behave the same way when you do that.

Bob

[T3sl4co1l]

... but was it present when the amp was brand new?

Hopefully not, but by using imperfect substitutions, the design has been tampered with, so my point stands.

Tim

[Pjotr]

Hi

... but was it present when the amp was brand new?

Good question. Have seen many well regarded amplifiers, also in de higher quality range, that do oscillate a little bit at clipping recovery, or on the edge of clipping when driven with a square wave. Those are not designed/made for clipping as the normal operating condition like guitar amplifiers.

Look if the amp behaves normal at low levels (say 2V_pp out) by driving it with a 1 - 10 kHz square wave. Limit the frequency response (or rise time) of the square wave with a RC filter at 50kHz to avoid slew induced issues. It should have no, or negligible, overshoot or slew limiting. A more extensive test is a 100Hz sine wave with a 10% 1 kHz square wave superimposed. Drive the amp to full load, but below clipping. If there are no signs of overshoot of the square wave or signs of oscillations, the thing is ok.

[nev23]

The amp is stereo and I have tested the other channel and it does not oscillate, but I have only tried a resistive load.
The new devices are bigger, a bit faster and a lot cheaper than the originals. I'm not sure the "original" parts I have seen advertised are not fakes. The new devices have a higher input and output capacitance than the originals, so I didn't want to add any more capacitance there.
The oscillation occurs just under clipping. If the amp clips, it appears just before and just after the flat bit at the top.
I don't think the 20pF is a normal feedback technique: it is connected to the Voltage Amplifier Stage. Normally, there would be a capacitor across the main feedback resistor from the output: the lower 30k resistor.

My reading of it was that instead of making the amplifier as broad band as possible and then reducing the gain at high frequencies using a capacitor in the main negative feedback path (across the lower 30k resistor), they had tried to reduce the inherent gain of the amplifier at high frequencies, and have left the feedback wide open. My guess is that this would give a very poor phase margin. The path with the 20pF capacitor cannot bring the gain down below 1 because of the 5.1k resistor.

So, what shall I tweak first? Shall I add a capacitor across the main feedback resistor (the lower 30k)? I think the 47pF "miller" cap can stay. The .05uF across Q1, Q2 emitters seems fairly normal. The capacitors around Q10 & Q11 have been temporarily removed (see quote below) - nothing changed, and anyway the limiting is not being triggered as I am using a resistive load. The 200pF on Q3 seems odd, and the capacitors around Q14 & Q15 seem rather large. I did once build an amp with a cap in the same position as the 20pF, and varying the value (I put in a 65pF trimmer) sharpened the edge of a square wave, bringing a little ringing if too small. So that may be OK, too. But what is the upper 30k resistor doing???

Q10 & Q11:
A small-value capacitor is
sometimes connected across the base-collector junction of each protection
transistor, with a view to eliminating
benign parasitic oscillation that may
occur sporadically in the network
during the limiting process. These
capacitors appear in parallel at A.C.,
and are entirely unsatisfactory, as they
create an ill-defined and therefore
undesirable feed-forward path around
the output stage, shunting it out of the
global feedback loop at high audio
frequencies, precisely where the
amplifier is most vulnerable with
respect to non-linearity. Such
vulnerability is due to a necessarily
diminished feedback factor at high
audio frequencies in the interest of
Nyquist stability. Connecting the
capacitor, (of the order of 1nF), across
the base-emitter junction of each
protection transistor is the preferred
solution.
Michael Kiwanuka, Transparent V-I protection in Audio Power Amplifers (Part 1), ELECTRONICS WORLD September 2002, page 47.