1970s Design Indulgence

Hi Bruce, answering your questions...

1. Yes. Within constraints that's correct.

For example, increasing the VAS local feedback capacitor (Cdom) any further than necessary will not improve phase any further, because being in a NFB loop with substantial transistor base-emitter capacitances (in the output stage), it has already been subjected to phase shift. This is like what happens with an op-amp driving a capacitive load. The only way around that is to drive the output stage with more VAS current, leading to an even-hotter VAS transistor, or making the output stage a triple instead of a double. Even so, we will be faced with the transition frequency of the last transistor.

Decreasing the value steepens the fall-off, increasing phase shift, which might still be considered stable, but anything less than 90 degrees (steeper than 20dB/decade) causes ringing. The output capacitor starts to shift from being "purely" capacitive to becoming inductive at high frequencies (10kHz upwards for example), and excessive ringing can heat it up, which eats into its lifetime, "wearing it out" little by little.

Here bypassing might help, but not to get more HF to pass for subjective improvement/difference, but simply to add some protection to the output capacitor. This might be facilitated by joining the junction of the output capacitor and the output inductor with the junction of the 4u7 capacitor and the 100 ohm resistor. I shall have to give it a try.

2. In theory the input filter should "slow" the signal such that it cannot exceed the available slew-rate, and this reduces the chance of ringing. Getting it to work in practice is the difficult/impossible bit: The value is calculated from the input series resistor combined with the output resistance of the source. 2k2 is the maximum "standards" value from a solid-state source, but what about a valve preamp with a 100k volume pot on its output? Having realised that since my post, I decided against putting much design emphasis on it, and simply seeing it as EMI blocking.

3. The inductor serves the purpose of isolating the output from complex loads (in a similar way to using a 100 ohm resistor on an op-amp output).

The Zobel network, operating before the output inductor, provides a fixed and simpler value of loading at high frequency. It is mainly required for the purpose of stopping the emitter follower pair (the upper part of the output stage, T4 and T6) going into "Colpitt's oscillation". If it operates before the 0dB crossing it adds to the phase shift making ringing worse. If it operates too far beyond the 0dB crossing it allows the onset of "Colpitt's oscillation". The effect can be seen as a slight bellying out (toward the right and upwards) of the gain curve below the 0dB crossing, which at first looks quite benign because the simulator shows improved phase margin. But this is a trap.

Tandberg (and I think Aiwa did it too) got away with it because of the poor HF extension of the germanium output devices which made the difficulties highlighted above less of an issue. Progress has led to improvements which have in-turn led to problems...

So little is written about this power amplifier circuit configuration that it feels like I'm finishing the job started by HC Lin, Tobey and Dinsdale - filling-in the many blanks they left behind.

Take on a slightly different configuration and there's a wealth of published information, and if you decide to go the common route of the long tailed pair input stage you don't even need to be an electronics design engineer! You just copy one of the myriad of published designs.

DC Stability

There is however one article which covers the class-A voltage amplifier to some degree with fig. 11 page 41 here: https://www.americanradiohistory.com/Archive-Wireless-World/70s/Wireless-World-1971-01.pdf - being the closest thing to what I have.

The limited information here helped jog my memory about proportioning supply voltage for DC stability. It helped me understand a "rule of thumb" gleaned from my research nearly 43 years ago.

It is an ambiguous rule of thumb that should have read that the emitter voltage of T2 should be 1/10th the supply. But it discussed a 9V supply and rounded up the emitter voltage to 1V, so the memory might interpret that in any such design only 1V is required. In this design I have proven for myself that 1/10th or thereabouts is the requirement, and the 3-4 volts I had made settling the DC a "regular task", which means not stable.

This design is so close to that of my Proprius that I expected it to sound almost identical, but it took on a "roughness", and I can only put that down to a change in distortion between the test bench and the listening room, and the only difference is in the mains supply voltage: test bench is adjusted to 230V (by means of a giant 'variac'); and listening room is on 245 - 254 volts depending on substation loading.

Rather than recalculating, it appeared logical to simply use the same values as in the Proprius for collector load and emitter, which contains the DC output level trimmer (the Tobey and Dinsdale design doesn't place it here).

This does however increase the VAS current as the HT voltage (offload) is around 73V where it is 48V in the Proprius, and so there is 50% more collector current. The VAS transistor will run hot due to the increased wattage, but we have the "assurances" in Doug Self's book "Audio Power Amplifier Design" that this is normal.

The revised circuit will be published in due course, but it required the "padding" in T1's emitter illustrated by the resistor "R3" and the bypass capacitor shown in the link given above.

HF Stability

Using modern transistors in this vintage configuration exposes its weakness, which is high frequency instability, and it wasn't fully understood, or necessary to be understood, by its early 1960s authors. The power transistors of that day would simply not do the frequencies this design will do. Here they extend beyond 10MHz.

Soon after the publication of this configuration attention turned to the "singleton" input stage referred to very much earlier in this topic. It can be seen that the "singleton" is a current feedback amplifier, and the authors must have seen their configuration as being much the same.

Theirs therefore has frequency compensation to T1's emitter, and in the Leak Stereo 30 (an almost direct copy) "advantage" of the "inherent stability" of current feedback was taken by not having any compensation at all (thanks to Rod Elliott of ESP - http://sound.whsites.net - for pointing these things out).

Simulating zero compensation in this circuit however, it can be seen that even though phase remains well within 90 degrees, the rate of fall suggests oscillation, and is preceded by a rather large spike.

In practice the circuit is perfectly stable as long as signal is never applied! With only a few millivolts of stimulus the current limiting resistor sensibly placed where the HT fuse would go, emitted smoke!

This circuit is therefore not simply a current feedback design. True, it is current feedback by definition, because NFB is applied to the input transistor's emitter, but that is also a route through to T2's base, and could be conceived as being voltage feedback when it is realised that T1 is not a current source for T2's base, as conventional thinking would have it.

It is best seen as HC Lin's (RCA) original circuit, or even better as Mitsubishi's Teleton SAQ300, where both have a single transistor "preamp" before the power amp "for real" and use voltage feedback compensation. In the Teleton the NFB is extended to incorporate the "preamp", and so is current feedback of sorts, but definitely not at DC. Both circuits use the same 470pF value.

The Tobey & Dinsdale uses current feedback compensation, but the only real difference between this and the RCA/Mitsubishi circuits is that T1 is DC coupled. And as that has no bearing on high frequencies it can be seen that they are all really the same.

So we have voltage feedback compensation; current feedback compensation; and no compensation at all if we include the Leak Stereo 30. Who is right?

Obviously the answer is none of them. These were empirical solutions which were the "glue" that held each particular circuit stable - which do not apply universally.

Neither does the global NFB compensation offered by John Linsley Hood. By his own argument of delays, the transistors will have already gone unstable by the time the global compensation "stabilises" them. Simulate it and you will see.

The only conclusion we can draw is that we have both current and voltage feedback at the same time in this circuit configuration. And that this wasn't grasped all those years ago through the lack of simulation software - and the computer operating systems required to run it - and also the lack of high speed transistors such as are being used here.

Settling on 47pf for the voltage feedback (Cdom, T2 collector to base), and 22pf for the current feedback (T2 collector to T1 emitter), with both applied before the delays (phase shifts) of the output stage - and with the output zobel network of the values chosen - we get approximately 90 degrees phase margin, and better than 10dB (12.5dB actually) gain margin.

Remembering that 45 degrees phase margin and 10dB gain margin is stable, we have bettered that. But by obtaining 90 degrees phase margin there should now be no ringing.

By using 47pF T2 collector to base, with the 2mA standing current of T1's collector, we can argue that slew rate is 2/0.047 = 42.5V/uS. Input slew rate calculates as 33V/uS. And with the input filter (1k and 220pf) driven by a zero output impedance source, we have cut from 773kHz. It should not sound distorted at all.

Measurably it is 0.02% THD at 1kHz to 0.1% THD at 20kHz so there is no audible distortion there.

We are however AC coupled at the output, and using 4700uF the capacitor should not introduce much if any low frequency distortion. And little registers on the distortion sweep. But there is one other source which is bridge rectifier RF, which still requires snubbing, but that will have to wait for the new PCBs which have the facility to mount individual rectifier bypass capacitors.


Edited by Graham Slee - 18 Jan 2019 at 6:57am

Progress, what a wonderful thing, but not for AC coupled power amplifiers...

Power supply smoothing and decoupling capacitors must be rated for the entire HT voltage, not just half of it as in DC coupled amplifiers.

This is probably why good-for-audio electrolytic capacitors stop at 63 volts. That places any AC coupled amplifier greater than 30 watts at a disadvantage.

If the maximum is 63 volts the HT must never exceed that voltage, and that limits the HT at 230V (115V) to 56 volts. Why? To allow for mains voltage variations and the amplifier idling when the HT voltage rises to its off-load value.

If we only have 56 volts of HT and use 6 volts for this design's DC stability we have only 50 volts left for the signal. We halve that and take away output stage bias and saturation voltages. The upper EF pair uses 1.2 volts for biasing and we can guess the saturation voltage as being around 1.3 volts, so we lose 2.5 volts.

So from our 50/2 which is 25 volts we end up with 22.5 volts peak. The rms is a 0.7 of that making 15.75 volts. We square it which makes 248 volts, and divide by 8 to find we have 31 watts.

And for any higher power we can only look at 4 ohm speakers where we can conceivably reach 62 watts, but seeing most quality speakers are 8 ohms this becomes academic.

Now, if there were good-for-audio electrolytics rated at 80 volts or 100 volts we could finish this job, but we have now found the Achilles' heel.

If we need 2200uF to 4700uF at 80 or 100 volts we can only find snap-in electrolytics which sound awful. They are designed for switched mode power supplies and that is their only fit use. As the amplifier is ran through its frequency sweep tests they can be heard at mid to high frequencies. For audio they are worse than useless. Distortion machines intended for the distortion generators the world needs to sustain its rising population. What a boring future!

What is required is the rubber bung electrolytic. The bung damps the foils to prevent the "switched mode" effect. The flexibility of the rubber bung is key to good bass. Its damping abilities are key to the quality of all frequencies.

Pure and simple, that's what an audio electrolytic is. It's how they used to be made in the day when all electronics was for analogue audio by whatever means it came (radio, TV, telecoms, records and tapes).

The DC coupled amplifier can therefore have 56 volts per rail resulting in more than 100 watts.

Our 70s technology has, through progress, been limited to 30 watts.

I don't mind telling you that this amplifier is/has been (hopefully past tense) the worst ever from a stability point of view.

Not that it overheats, blows fuses, makes outbursts of noises, or screws-up in measurement - one would think it perfect - but on listening something untoward can/could be heard.

The crappy capacitors hid it behind their atrocious contribution. How any designer can use snap-ins which today are solely made for switched-mode power supplies (that's why you can hear that screeching sound from some SMPS) I do not know, but they do.

I have replaced all the snap-ins with wire-ended rubber bung type radials, which improved things no end. But now I don't have as much energy storage as I did due to real-estate limits.

The 5,000uF reservoir I have now is still quite sufficient and can be upped to 6000uF relatively easy, but as mentioned before, the decoupling can only reach 2,000uF, so now the output capacitor is a solitary 2200uF, and it might be replaced by two 1,000uF units in parallel to up the ripple current capability.

This means the low frequency -3dB turnover is now at 10Hz for an 8 ohm load. It should make little difference to the perceived sound using all but the largest of bass horns. Any electrolytic distortion will be buried beneath normal distortion in case this worries you.

But back to the stability issue. The simulator shows a bit of a knee (dot dashed line) which although around -15dB, is tending to rise before it falls off more steeply.



The simulator also shows absolute stability, but it would because there are many other instances where simulation shows the same sort of knee from stable circuits.

But this is not the same, and even having a state of the art radio frequency spectrum analyser is no good (it's gathering dust) because it would be picking up all kinds of electro-magnetic disturbances. I simply had to assume the subtle rise at the knee was oscillation.

Oscillation without gain you may ask? Look, anything is possible with emitter followers (and cathode followers!). Things aren't as neat as the textbook tutors would have it. The output stage transistors are 30MHz (their Ft) just where the knee peaks where negative feedback is reduced thus allowing the gain to increase, then fall.

The extra compensation has to be within the negative feedback because that's why it is happening. This sort of thing cannot be solved by an external filter.

Simulating a 470pF capacitor from T1 collector to ground took away the rising tendency such that it falls away smoothly from around 20MHz. One might think that the RC combination of the 16k collector resistor with this 470pF capacitor would make it 21kHz (the math says so) but T2 does most of the gain - we're just adding a pole.

The same idea is used with this type of stage to prevent radio frequency breakthrough, not that we have any here.

In any circuit containing electrolytic capacitors (don't they all?) where such frequencies exist, even some 40dB below normal signal (and remember anything can set it off: thermal changes for example), the capacitor(s) become(s) inductive (as well as capacitive) and will warm up, or even heat up. This not only reduces the lifespan of the unit, it also can hotspot in one particular part of the foil (the reason why they cast disco smoke machine heating coils in a large aluminium block by the way*). It may or may not recover from this (this being the damage I alluded to in an earlier post).

The long term effect in an always-on circuit is bass loss due to this electrolytic "fade". If an amplifier is only switched on for a couple of hours each listening session it might not be noticed, but I want this old-world design to meet my always-on requirement, even though it has an on-off switch.

It would be inappropriate for me to comment on the sound right now as I know new components change over time. If it sounds good after 24 hours there's a chance it will stay good, meaning the stability has been sufficiently tamed (he says hopefully).

(* I know of a rather novel way of instantly heating a smoke machine coil and preventing it hot-spotting without an aluminium block!)