1970s Design Indulgence

I had better get on with it then...

Remember all those amplifier tests deemed necessary in the "new age"? To ensure the amplifier maintained its stability?

Things like driving a 1uF capacitor directly across the speaker terminals... as if a loudspeaker manufacturer would do that

And if the amp was capable, how would the capacitor suffer? And then would they use the same capacitor on the next test?

Just how do you test for stability, which if it goes wrong, the pre-catastrophic effects happen above 1MHz - beyond the range of all audio analysers?

The answer is a load of RF test gear, and the ability to think up tests which might indicate what is happening. Note, I said might. Why?

Well, either SPICE modelling has it wrong, or there should be a lot of red faces.

For stability the output signal potential must decay to less than its input potential (indicating a gain of one - unity - or less) at a rate where its phase is displaced less than 180 degrees, and this usually happens somewhere around 1MHz and above, or in most 70s designs, at less than 1MHz.

45 degrees less than 180 degrees is considered stable for op-amps, but stable in this case doesn't sound good because there is overshoot and ripple if a square wave is examined (if the circuit is of a type which will allow such examination).

A power amp which drives a loudspeaker with 45 degrees of phase margin off-load can go beyond 180 degrees on-load, and result in the speaker wires acting like antennae, and you should not be surprised to hear an analogue radio playing the same music as you at exactly the same time!

Therefore a wider phase margin is required and I for one always want to see 90 degrees. Some argue that 90 degrees limits slew rate too much, and prefer to risk 60 degrees... "because it sounds better". I disagree.

Also, when the phase is 180 degrees out, the gain of the circuit needs to be well below unity, not just below. A golden rule or rule of thumb in analogue electronics is the factor of 10. If it's one tenth of unity gain where it hits 180 degrees then it would have a job to oscillate!

My stability criteria is therefore 90 degrees phase margin and 20dB or greater gain margin, regardless of output conditions. And that includes the 1uF capacitor, but a simulated one in SPICE which cannot be harmed.

Unless you hate your music so much that you need to add "filth" (fuzz effects), you would use an amplifier that has a negative feedback loop to reduce distortion. Harold Black of Bell Labs invented negative feedback (NFB) a long time ago, and told the world that a factor of 100 or 40dB of NFB was required. It was seldom achieved in valve amps but it was the aim of the then designers. Oh so different today....

In any amplifier using NFB (which it should) all phase shifts can add together and make the NFB turn into PFB - positive feedback - and then it will oscillate and do damage.

All the way through there will be phase shifts because as we progress from small transistor to big transistor to get the power handling, the high frequency performance comes down, and each transistor attenuates the high frequency performance of the preceding transistor.

So we limit what the amplifier as a whole can do by placing a capacitor in the voltage amplifier to set its dominant pole. We set it so that with a resistive load, and negative feedback applied, it provides 90 degrees (as near as we can get) of phase margin and 20dB (or better) of gain margin.

Thus we ensure NFB is NFB.

It should be blindingly obvious that if we stick a capacitor across its output that at a particular frequency it represents a short circuit, and therefore there is no feed to the NFB, and so the amplifier goes to full open loop gain with catastrophic consequences.

So what we need is something in series with the output, after the NFB take-off point, to "soak up" the capacitor load.

When dealing with op-amps (or transistor circuits) and low level signals it is apparent that designers have their brains switched on, because in most circuits you see a series resistor, which is there to protect against capacitive loading and preserve stability.

But a series resistor is no use in the output of a power amp because it would severely attenuate the output power to the loudspeaker.

Therefore we need to use a "tuned resistor" - one that changes from zero to some resistance as frequency increases, and that is an inductor.

Ignoring those who say inductors mess-up the sound, because in my opinion you cannot teach the ignorant, we will now explore output inductors for a moment.

If the amplifier must be stable with a 1uF capacitive load, then the inductor must be large enough to replace the lost resistance. However, what we see in numerous designs is a few turns of copper wire round a former or even a resistor. A few turns of wire are not going to be of much use against a 1uF capacitor. This tends to suggest the designer partly understood the problem, but didn't grasp the answer.

I am assuming amplifiers from these great companies were not designed by ear and a bit of math only. That there was more to it than just the basic audio analyser tests. But on looking at their schematics, and in one or two author's reasoning, I can see this could have been how it was done. I feel let down, because I could have been on that gravy train! The reason I wasn't was because I thought there was more to it - and actually there is!

The value of inductor I found that cancelled the effect of a 1uF capacitor was 5uH (5 micro-Henries), which when unwound produces sufficient copper wire to make several of the inductors I have seen in other designs. So what "magic" were they using? (or still are?)

And then we have amps to this day which do not have an output inductor, and the instructions warn to use a low capacitance cable or it will damage the amplifier. Just how can a group of people maintain a myth?

There is one possible answer which is the dominant pole, which could be set so low - as it would be in the days of slow transistors such as the 2N3055 - meaning that the signal is near or below unity at the frequency capacitance becomes problematic, but such an amplifier would not have the "PRAT" its followers claim...

But back to the inductor, and the resistor the more enlightened use serves the purpose of damping the inductor's effect which is to "spike" the response at some high frequency where gain can be above unity. This is because all inductor-capacitor networks have a resonant frequency, and the NFB plus all the phase shifts in an amplifier complicate matters. But if we think of the capacitor test as a short and the amplifier is designed to drive around 8 Ohms, then a resistor in that region will curtail the inductor's effect such that the load is similar all the way.

Also interestingly, and earlier on, I found the gain margin would not go sufficiently low before the 180 degree point was reached. This is most likely what I think of as Colpitt's effect, which is where emitter followers tend to want to be oscillators due to a combination of intrinsic and external capacitance, and inductance (ref: Small Signal Audio Design, Douglas Self, 2010).

I have only ever found one cure, and that is to provide further compensation elsewhere which results in the same transform, but not always where it is expected to be. This being something which SPICE modelling can do, and that my imagination is incapable of. Normally such compensation would be achieved using the equivalent of grid stoppers, but modelling exposes that transistor base impedances aren't easily calculated.

As it was with the Proprius design, so it is with this one, and although the circuit topology differs greatly, the same cure works. And that is to add a small value capacitor between T3 collector and T1 emitter.

Next we shall look at power and current.

Edited by Graham Slee - 27 Jul 2018 at 5:47pm