Quad II ringing

[PaulM]

Capacitance will obviously slow things down and limit the square wave rise/fall times, but there are resonances too. At the end of the day, a loudspeaker is a current driven device and a current probe might be a more realistic way of testing? I can't help but think that voltage tests across a resistor are not <<real>>.

Best regards,

Paul M

[trobbins]

Originally Posted by cathoderay57

From a bit of reading I think the 135 and 45 degree points are maybe referring to phase margin which is 180 degrees minus the end-to-end phase shift of the open loop amp but that still doesn't help me make much sense of my measurements.... J

Those phase shifts of 45 and 135 don't directly relate to phase margin. For open loop conditions they relate to benchmark phase changes on the way to (and beyond) 180 deg, as a way to assess the basic forward open-loop gain-phase behaviour of the amp. For a closed loop measurement, they still don't indicate phase margin, or gain margin, as those two margins relate to the conditions/frequency when phase = 180deg, and gain = 0dB.

Crowhurst has many many articles that touch on the stability topic - with a simple description of margins in "stabilising feedback amplifiers' article in Audio Dec 1956. Some of his articles get in to polar diagrams, which are tougher to interpret for many, but well worth searching out his many articles/books.

Originally Posted by PaulM

I just wonder whether this is 'real world'? Using a resistor as a dummy load is all well and good, but a loudspeaker is anything but a pure resistor. Are these observations of ringing something that happens in reality? Paul M

Using a resistive-only load at the nominal secondary impedance value allows a baseline measurement that can be easily replicated by others, as well as interpreted against other results and to the internal combination of parasitic resonances.

Using a range of non-ideal loads, like open-circuit and different value capacitances, can provide confidence that no external conditions would lead to unstable operation. However, some loads can indicate how severe any ringing could become if the right conditions were applied to the amplifier - such as if you provided a specific high frequency signal to the amplifier and the load (eg. tweeter) presented such an impedance at the frequency.

[cathoderay57]

Thanks for the clarification in phase margin. The Quad II output signal starts at 180 degrees out of phase with the input (at low frequency) and the phase lag increases with frequency, so I'm afraid I still don't get it as to how you determine the reference points at which the phase shift is 135 or 45. J

[Gabe001]

Originally Posted by cathoderay57

Thanks for the clarification in phase margin. The Quad II output signal starts at 180 degrees out of phase with the input (at low frequency) and the phase lag increases with frequency, so I'm afraid I still don't get it as to how you determine the reference points at which the phase shift is 135 or 45. J

Yes, the phase shift is 180 because in the link I circulated earlier the signal is injected at the cathode. So a -45 degree shift (-1 slope) is 135 degrees.

I quickly tried doing this myself yesterday on one of my amps. Ignoring the phase shift at the low end, this is my curve at 18khz. If you measure the width, it's 3.6cm. the distance between lines at the centre is 2cm. The phase shift works out at sin(angle)=2/3.5 =34 degrees. Unfortunately my signal generator only starts at 100mhz so I cannot generate sine waves >20khz but <100khz, but I'd estimate that my 45 degree breakpoint (slope of -1) is in the low 20khz range. Hope this helps.

The phase and gain margin can easily be worked out from bode plot

[Radio Wrangler]

In the gain and phase margin calculations, you must go round the entire loop, and you can break the loop anywhere. So, say you break at one end of the feedback path series resistor - imagine that the output transformer end of R11 is lifted, and the test signal is fed into R11, while you look at the signal coming out of the output transformer pin that R11 previously used.

Note that you are not using the amplifier's normal input. Your signal travels through everything around the loop. It gets attenuated by the feedback resistors, it gets amoplified by the forwards amplifier, it comes out of the output transformer at the point the feedback needs to use.

You ideally want to feed your signal in from a source impedance similar to what the output impedance was from the output transformer pin. You ideally want to load that transformer pin with an impedance similar to what it saw when it was driving the feedback resistor.

NOW, you're comparing apples with apples. You see the actual gain and phase shift which matter in terms of making a loop stable and well-behaved.

So now you see that changing the feedback resistors so as to increase the closed-loop gain from the normal input to the normal output means the feedback resistors produce more attenuation, So you get more gain margin around the loop. This seems good, right? But you now have less feedback in play, so the feedback has a lower factor of improving distortion, improving output impedance and improving flatness. So this is not good.

So you choose to change the feedback resistors to set the amplifier up for less closed loop gain from normal input to normal output. T~he feedback resistors no give less attenuation to the signal passing around the full loop. This detracts directly from your gain margin. The amplifier is less stable so it rings more, it shows bigger ultrasonic peaking, the peak lowers in frequency and the whole thing might sustain oscillation if you go far enough, and that might not need to be very far.

So we can't muck about much with the gain factors before we run into trouble.

This leaves phase.

Say we bridge R11 with a small capacitor? This puts an added phase lead in the loop. Because the capacitor increases the amount of feedback, it reduces the closed loop (normal input to normal output) gain at frequencies where the capacitor starts to be significant. In other words, it deliberately rolls off the amplifier frequency response as-used. Provided this roll off is kept to frequencies above where real people with non-golden ears can hear (and you don't tell the golden eared brigade what you've done, because that only starts them thinking) who will notice? This phase lead improves your phase margin and your gain margin (the two interact) your amplifier becomes more stable and ringing improves.

So if you look at lots of different amplifier circuits, you'll find that a great many of them have capacitors in similar places. Even transistor ones!

You have added a zero in the feedback path, but when the loop is closed this is inverted to become an added pole in the closed-loop gain. Beta is inverted in the classic feedback equation (Black's)

David

[cathoderay57]

Thanks Gabriel and David for the detailed explanation - clearer now, thanks. Before reading this, at Gabriel's suggestion I had one last play around. First I tried putting capacitors in parallel with the 470 Ohm NFB resistor. 6.8nF caused the amp to go into HF oscillation. Around 1nf reduced the first overshoot peak but the ringing amplitude stayed the same. I found that increasing the value of a capacitor between V3 grid1 (equivalent to V1 anode to ground) from 30pF to 82pF was beneficial. Adding capacitance between V4 grid1 and ground either did nothing or made the ringing worse. The results with and without are shown in the photos below, with a sketch of the filter added. J

[Valvepower]

Hello,

Whilst ferreting through some schematics I came upon the circuit and the parts list for the QC12, the first incarnation of the QUAD II. I felt it worth posting given the subject of the thread.

I noticed there was a series resistor network (C3 – 47pF, R8 – 100K) in parallel to the 680K grid 1 leak resistor (R9) on the ‘inverted’ KT66 (V2).

Looking at the detailed winding arrangement of the output transformer it appears to differ to that of the QUAD II amplifier, with 7Ω and 15Ω taps, which may have necessitated some compensation in the driver?

Finally, I was going to post some measurements I did of the phase and gain of the QUAD using the previously mentioned method. I’m now in two minds as to whether to post these results, as there is a tad too many measurement uncertainties.

Terry

[Gabe001]

Since you've gone through the trouble of taking them, please do post them.

[Valvepower]

Originally Posted by Gabe001

Since you've gone through the trouble of taking them, please do post them.

Righty-Ho, here we go I’ve bitten my lip and gone hell for leather, warts, uncertainties ‘an all

I had a hasty attempt at looking at the phase and amplitude relationship with the signal being fed into the feedback (inverting) input.

The initial problem was, I was getting quite a lot of hum, which was making the setting up and reading of the measuring equipment rather arduous, this was cured by placing a 1K terminator resistor on the input of the amplifier to eliminate hum pick-up from the high impedance input of 1.5Meg.

I inverted one of the two scope inputs, on both scopes, to counter the inverted nature of the signal. I monitored the signal between the 470Ω and 100Ω resistors in the feedback network as I feel this better represents the signal directly on the input.

I displayed the Lissajous pattern (not centred on graticule) using an Hameg analogue scope and the two individual waveforms using a Digital Tektronix scope.

Looking at the Tektronix scope I could ascertain I was getting a 57° phase relationship between the two waveforms at 260.3KHz where the amplifier has reached unity gain (both traces were reading roughly 520mV pk-to-pk). As I said I was monitoring the input signal between the 470Ω and 100Ω resistors, with one of the two inputs on the scope inverted.

I must be honest here, and I feel there are some reading uncertainties surrounding inverting the signal(s) on the scope and the point where to monitor the input signal to give sensible readings, and because of this I decided to abandon measuring the Quad as I really wanted to get on with the wiring of an amplifier I’m currently building.

Finally, I did a Phase and Amplitude plot up to 200KHz using an audio analyser, however, with a caveat based on the uncertainties mentioned above.

Terry

[Gabe001]

Interesting, thanks terry.135 degree breakpoint 110k/140k, so lead comp cap across 470 ohm feedback resistor calculates to around 2-3nf (based on the article linked previously)

45 degree breakpoint 10k (a bit low). Lag compensation calculated at 88pf + 180k between the ef86 anodes (if I'm not mistaken, again using the guide above, see example 3) - I note quad used 47pf+180k across one of the two ef86 valves in the schematic you posted above, so the I don't think you're far off the mark Terry. Anyway, an interesting academic exercise at this point as Jerry sorted the ringing out.