Testing Global feedback stability and question about practical componsation for outputstage

[SoulFetish]

I didn't want to post too much maths.
To calculate the closed loop gain (G) I used the classic equation for closed the loop gain in a feedback amplifier i.e. G=A/(1+AB)
Where A is the open loop gain and B is the feedback factor.

After I posted wondering if I could calculate the closed loop gain, I remembered the Aiken white paper on his site. I used that info as a guide when first designing the amp. But at the time the phase inverter was designed for a 12AY7 anyway and had a different gain altogether. So whatever I settled on as a good feedback factor at the time would be mute for this version.
I changed the LTP to a 12AX7 because I wanted a higher input sensitivity for this stage. For a 12AX7, I needed to rebias the stage and change the tail values (lowering the available voltage across the tail resistor) to allow for the same output voltage swing as the 12AY7. The trade off was at the expense of sacrificing some balance in the inverting and non-inverting gain.
C'est la vie.

(why was I telling you this again??....) Oh yeah. Aiken's site.
Anyway, I checked it out briefly late yesterday and using the simplified equations in summary:
Acl (neglecting the 8Ω Ro) = A / (1 + A* Ri / (Ri + Rf))
I came up with a similar answer as you did.

For me, the difficult thing in trying to accurately calculate this on paper has always been finding some definitive resource for understanding and determining pentode gain in the output stage. I have a "better" understanding of this now, thanks to conversations on this forum. But starting out, I didn't understand why the hell all the values for mu seemed to disappear in all the pentode datasheets, when it seemed like kind of a big deal for triodes.
Of course, I was able to understand this much better from talking to folks who could answer this in this thread here:
https://music-electronics-forum.com/...cathode+bypass

[SoulFetish]

Finally!!
I was able to solve the HF oscillation and stabilize the global feedback. I was almost gave up on being able to incorporate global feedback at all. I feared the output transformer I installed had would not allow me effectively stabilize the HF phase shifts enough to prevent ultrasonic oscillation.
There were a couple of areas in the circuit where I suspected might be contributing to the oscillations and wanted to try some suggestions for common fixes.
The first thing I did was remove any shunt capacitance or compensating components from the feedback circuit so I could isolate each area and start troubleshoot. The first thing I wanted to look into was whether the capacitance in the MOVs across the primary were problematic. According to the data sheet, each MOV was contributing 300pF of capacitance, across the primary. trobbins, I know you like to string series combinations of smaller diameter MOVs to lower the added capacitance. However, in this particular case, 300pF is pretty much the exact capacitance I need for a reactance of 2560Ω @ 100MHz to create a Zobel network for my primary load impedance.
So after I added the necessary resistance, I powered it back up and tested the the output with feedback and little change I could see and the problem persisted. A little discouraged, I moved on and tried Malcom's suggestion of bypassing the feedback resistor with a 560pF build-out capacitor. After that didn't change anything, I increased the value to 1200pF and still saw no improvement, so I removed them altogether.
There was one technique I wanted to try, although I was not optimistic at that point. It was something I learned from reading Jones, Turner, and Max Robinson on the topic of amplifier compensation. Jones calls it "slugging the dominant pole" and it involves a shunt RC network on the plate of the input tube in the feedback loop. Turner writes: "V1 anode to 0V, R = 1/10 of total ac and dc RL in parallel, C should have its reactance in ohms = 1/10 V1 RL at 100kHz".
I my case this worked out to be 200pF and 8k2. Helmholtz correctly made the observation that I would need to in a LTP input stage, both plates would need to be fitted with this for it to work properly. So, I powered down to install the added components and powered back on and the sine wave was clean as a whistle. I mean, it was too clean. All that noise was gone. I checked twice to make sure the feedback resistor was still connected.
Now I get to actually really test the output, and run some square wave analysis....
But I want to put this into a speaker and just play some guitar for a couple of hours.

Oh, one other thing I noticed;
I began to see a roll off in measured RMS voltage running frequency sweeps, decreasing sharply around 3kHz to 5kHz. But I was looking at my scope and there was no real noticeable change in amplitude on my scope. I'm thinking this might be read errors because I've exceeded the bandwidth of my meters. So do you guys think it is worth building a bench meter for measuring true RMS, using Analog Devices's AD8436 chip?
Anyone want to help me design a PCB for the project
AD8436.pdf
AN-1341.pdf

[trobbins]

The OPT is the dominant cause of the high frequency gain and phase shift that is going to cause instability when feedback is applied. No matter how poor the transformer response is, global feedback can always be added to achieve a stable amp - its just how you go about it.

Testing the open-loop response is one way to appreciate how an amp's forward path gain-phase changes, which would then allow a compensation scheme to be designed and added, and then final testing with feedback applied used to confirm that all is ok.

You have gone straight to adding in a compensation scheme, in the form of a shelf-network, which is pretty much a standard approach for the last 70 years for feedback amps so will be described in many articles. The shelf network starts attenuating gain at about 8kHz with your component values, and the gain levels off at a low value at about 100kHz. At some frequency in between, the total loop gain will drop through 1, and hopefully the phase shift won't get too close to 180deg in that vicinity. Generally that means your level of feedback has to fall within the audio range (ie. distortion increases for signal frequencies above a few kHz) and the phase shift of the step network added to the phase shift from the OPT doesn't cause poor transient response in the vicinity of unity loop gain.

Higher frequency compensation networks, such as a shunt capacitor across the feedback resistor, or parasitic capacitance from Miller effects or from MOVs, are typically only influential at quite high frequencies and would only be used if needed to subtly tweak the phase margin (eg. to flatten high frequency performance and reduce square wave twiddles). These types of compensation are typically never meant to be the main means to achieve feedback stability.

I think some amps with LTP's have used a plate-to-plate compensation network (as per your original 200pF part, or as a shelf style with a series resistor) to provide a simpler network than separate plate shelf networks.

Many multimeters have a bandwidth only up to about 1kHz. Only lab style meters typically have higher bandwidths, so yes that is the value of a scope and especially for X-Y display of input ans output signals to show up gain and phase change.

[SoulFetish]

Testing the open-loop response is one way to appreciate how an amp's forward path gain-phase changes, which would then allow a compensation scheme to be designed and added, and then final testing with feedback applied used to confirm that all is ok.

I can see how this would make sense. I suppose I was anxious to start throwing "solutions" at the problem to make it stop, rather than test and observer gain and phase so I know where the problem starts to occur in the upper bandwidth. Without knowing any of that, all I really know is 50-100kHz was being a real problem for me.
Plus, I'm not going to lie, I was just kind of hoping this would all just work out. I've never done this kind of testing, or compensation design. In order to do it properly, I fear I'm probably going to have to step by step do something like this:
http://www.angelfire.com/electronic/...pensation.html
and when you've never done it, that just really seems like a lot. A lot of plotting and boding... and then some more.
But the XY mode for checking phase is pretty cool. It's a little less intimidating that I imagined.

You have gone straight to adding in a compensation scheme, in the form of a shelf-network, which is pretty much a standard approach for the last 70 years for feedback amps so will be described in many articles. The shelf network starts attenuating gain at about 8kHz with your component values, and the gain levels off at a low value at about 100kHz. At some frequency in between, the total loop gain will drop through 1, and hopefully the phase shift won't get too close to 180deg in that vicinity. Generally that means your level of feedback has to fall within the audio range (ie. distortion increases for signal frequencies above a few kHz) and the phase shift of the step network added to the phase shift from the OPT doesn't cause poor transient response in the vicinity of unity loop gain.

I'll have to get back to it, then.. but it will have to wait til tomorrow. I'm going to bed

peace.

[SoulFetish]

*UPDATE (had to repair the vertical attenuator switches on my 465 in order to continue this....)

The OPT is the dominant cause of the high frequency gain and phase shift that is going to cause instability when feedback is applied. No matter how poor the transformer response is, global feedback can always be added to achieve a stable amp - its just how you go about it.

Testing the open-loop response is one way to appreciate how an amp's forward path gain-phase changes, which would then allow a compensation scheme to be designed and added, and then final testing with feedback applied used to confirm that all is ok.

You have gone straight to adding in a compensation scheme, in the form of a shelf-network, which is pretty much a standard approach for the last 70 years for feedback amps so will be described in many articles. The shelf network starts attenuating gain at about 8kHz with your component values, and the gain levels off at a low value at about 100kHz. At some frequency in between, the total loop gain will drop through 1, and hopefully the phase shift won't get too close to 180deg in that vicinity. Generally that means your level of feedback has to fall within the audio range (ie. distortion increases for signal frequencies above a few kHz) and the phase shift of the step network added to the phase shift from the OPT doesn't cause poor transient response in the vicinity of unity loop gain.

Higher frequency compensation networks, such as a shunt capacitor across the feedback resistor, or parasitic capacitance from Miller effects or from MOVs, are typically only influential at quite high frequencies and would only be used if needed to subtly tweak the phase margin (eg. to flatten high frequency performance and reduce square wave twiddles). These types of compensation are typically never meant to be the main means to achieve feedback stability.

I think some amps with LTP's have used a plate-to-plate compensation network (as per your original 200pF part, or as a shelf style with a series resistor) to provide a simpler network than separate plate shelf networks.

Many multimeters have a bandwidth only up to about 1kHz. Only lab style meters typically have higher bandwidths, so yes that is the value of a scope and especially for X-Y display of input ans output signals to show up gain and phase change.

I had to wait to test the phase response in order to service the vertical attenuator cam switches on my Tek 465. I bought some 99% isopropyl alcohol and syringes to clean the switch contacts, although I had a bit of a scare. After reassembly, I was trying to get a little bit of the noise out of the channel two V/div switch. So I removed the center knob and went to squirt the alcohol into the space between the panel and shafts, and the needle came off from the pressure. With nothing to govern the pressure I was applying, all the isopropyl bounced right back into both eyes. It was a perfect bank shot. I don't know how that much ended up in both my eyes, but you'd of thought I was Minnesota Fats in the Hustler it was such a good shot. Anyways, luckily I didn't go blind, but it hurt like a mutha' f*k'r. (eye rinsing. cold water for several minutes)

back to the amp..
Resetting up the test, I removed all feedback and compensation to test the open loop gain and frequency response. Injecting a 1kHz sinewave, with an amplitude of 319mV RMS into the amplifier input, yielded ≈6.16V at the output for an open loop gain of 19.31. Then I set up and tested the phase response in XY. Firstly, I noticed a discrepancy looking at the shape of the sine wave at the output. Compared to the input, the tops of the wave were quite rounded.
I took note of this, and switched over to XY. Starting at 1kHz, I could see that there was the beginning of a shift in phase, with a slight "S" shape at the corners. I increased the frequency until I reached a 45 degree shift at 20kHz.
Before go on, doesn't this seem not... good? Could what I'm observing, indicate that the leakage inductance might be problematic?
Since this is my first go at this, I'm not sure what I should be expecting, so I would appreciate any kind of feedback on this so far.
Thanks

[trobbins]

Pleased there was no lasting visual damage - just goes to show the variety of hazards on the bench.

Even the Williamson amp showed about 45 deg phase shift at circa 40kHz, which keeps increasing to about 120 deg at circa 70kHz, and then a dip/rise as the first main resonance is passed through, and finally approaching 180 deg at circa 150-200kHz.

That early phase shift is from the output stage plate resistance interaction with the output transformer's leakage inductance and shunt capacitance and reflected output loading.

I'd guess you will find some form of resonant twiddle occurring around 40kHz and perhaps 180 deg phase shift circa 80-100kHz, although there could be further resonant twiddles showing up depending on the output transformer.

It can be fun watching the XY plot rotate and morph its shape as frequency is both increased, and reduced. Keeping the output level not too high may help minimise the influence of output stage distortion on waveform shape, although the input then gets smaller and perhaps noisier. You also have to swap gain steps as output level falls at bandwidth extremes, to keep a sort of balanced X and Y excursion waveform on the screen. I recall Partridge showed some X-Y images at low frequencies to show just how much distortion and morphing was happening in an output transformer at low frequency, and hence how to mitigate it - and that was back in the 1930-40's.

[Helmholtz]

...and reflected output loading

This is where speaker inductance/impedance gets in with a speaker load.

[SoulFetish]

You also have to swap gain steps as output level falls at bandwidth extremes, to keep a sort of balanced X and Y excursion waveform on the screen. I recall Partridge showed some X-Y images at low frequencies to show just how much distortion and morphing was happening in an output transformer at low frequency, and hence how to mitigate it - and that was back in the 1930-40's.

trobbins, I'm not sure how you mean swap gain steps. Could you elaborate?


I swapped the output transformer and installed the one I originally bought for this project. The one I'm using now is an Edcor CXPP30-10K(see link for specs):
CXPP30-10K

So, I got back to testing the Open Loop Gain and phase response again. This time I used my Rigol Digital Scope. The Rigol was a good choice to get Amplitude measurements of the input and output signals. But it was near useless in XY mode.
Even so, I took some measurements and screenshots.
I'm not sure if this is correct-- but from what I could see looking at the lissajous patterns, it seemed that the phase shifted 45° at ~13kHz, 23kHz, and 33kHz. But I'm not sure if that's correct? It certainly doesn't seem like it should shift 180° that quickly. I'm going to try again on the 465, but I was able to trace the signal amplitude at 1kHz, 20kHz, and 33kHz.
Edcor's specs state that -1dB extends to 20kHz, and it looks like the -3dB corner frequency is right around 33kHz.

here is the input and output showing the open loop gain at 1kHz



Here is the output at 20kHz


and here is the input and output traces at 33kHz



edit: I should have stated that the open loop gain at 1kHz is 23.12 (27.27dB)

[Helmholtz]

Edcor's specs state that -1dB extends to 20kHz, and it looks like the -3dB corner frequency is right around 33kHz.

With such limited bandwidth also the scope will produce some frequency dependent phase shift between its channels.

[SoulFetish]

With such limited bandwidth also the scope will produce some frequency dependent phase shift between its channels.

Still, I'm not sure.. I think I'm doing something wrong. The output maintains a pretty steady gain approaching 20kHz. So, if there was there was a 45° phase shift at 13kHz, wouldn't that be where I would measure a -3dB drop in voltage?

[SoulFetish]

One other thing – I'm using Max Robinson's paper on Amplifier Compensation (found here:Design Stable Feedback Loops). He indicates that the first breakpoint on his Newcomb D10 amplifier was at 9.5kHz where the phase shifted 135°.
Why is he counting down from 135°-90°-45°?

[trobbins]

Max implies (I didn't have time to read everything so he may explain it somewhere) that the amp has in inverted output phase (ie. -180deg) for mid-band response, so that the output can be connected back to the feedback junction to provide negative feedback (not positive in-phase feedback).

[trobbins]

I would recommend the aim is to display an X-Y plot, where output signal is say Y axis and input signal is X-axis, as that form of display allows the phase difference to be much more easily discerned than using a display format where the two signals are on separate displays (or even the same dual display) and where X-axis is time. One of the photos on Max's webpage you link to has an X-Y plot on the screen. Crowhurst shows some XY shapes in his Hi-Fidelity circuit design book.

[SoulFetish]

I would recommend the aim is to display an X-Y plot, where output signal is say Y axis and input signal is X-axis, as that form of display allows the phase difference to be much more easily discerned than using a display format where the two signals are on separate displays (or even the same dual display) and where X-axis is time.

Right. That's how I've been trying to check the phase shift. But I've never done it before. So when I get results I'm not expecting, it makes me question whether or not I'm doing it correctly.
I've been putting the output on the X axis, but I'll switch it up and see what I get.

One of the photos on Max's webpage you link to has an X-Y plot on the screen. Crowhurst shows some XY shapes in his Hi-Fidelity circuit design book.

One of the things which had me worried, was the diagonal line on the scope which is supposed to show a straight line with 0° phase shift had an "S" curve. I figured this indicated an operational problem in the amp, but in the Crowhurst plots you mentioned, this can be caused by the tube characteristic. See the red box below:

[SoulFetish]

okay, there's only so many times I can get the same results before I have to admit I'm in denial and the bandwidth and phase response of my amplifier is total shit.
Using my venerable 465, I got a pretty clear trace of my first break point at ~10kHz with a phase shift of 45°. It's tough to argue with this -



2 different output transformers produced similar results. I couldn't get a clean image of 90°, but it seemed to confirm that there was a phase shift of 135° around 31/32kHz.
What can I take from these results? Is there something in the design I should have done differently with the loop containing 3 stages? Should I care?