Testing Global feedback stability and question about practical componsation for outputstage

[SoulFetish]

I started doing some square wave testing on my output stage to for any instability, measure open loop gain, etc., and see if it's really worth having any global feedback anyway. I have a function generator on my phone which I used initially for the square wave source, and monitored the output at the load.
With the feedback engaged, I snapped a pic of the waveform. It was looking ugly and seemed to have some nasty ringing in the loop. Here is the image:


Then I though I should check the signal coming out of my phone to double check. This was the output:


That is entirely useless, and quite pathetic. So, I wired up a quick 1458 square wave oscillator with a buffered output and was much happier. Here is the circuit (schematic, build, and output):






not too bad, eh?

Here is a schematic of the output stage and feedback loop.



I'm going to run out of attachment room so I'm going to post scope traces in the next one....

[SoulFetish]

One of the reasons I wanted to test this was I was getting some shrill high frequency dominating the sound with the presence control at max settings.
Starting with the GFB engaged, I tested the output with the presence turned all the way down and here is the shot:


Here is the presence at full:


Here is the output with no global feedback:



As an aside, I'm becoming really underwhelmed with the resolution on my scope and wonder if I should be doing this on an analog scope.
But my questions are; first, since this is a guitar amp, what kind of realistic phase compensation do I need to concern myself with? I was thinking of adding the 8k2/200pF circuit shown in parenthesis on my schematic.
Second, what kind of linearity should I be looking for? What can be inferred from the wave forms above?
Last, any thoughts on taming the HF in the presence circuit? I think more useful would be some kind of actual small bandpass filter to shut actual mid frequencies for an active mid boost.

[nickb]

Square wave testing is great for a for an instant overall picture or what is going on but it's hard to get useful numbers out of it. Therefore you should consider running a sine wave frequency sweep through it. Say 100hz to 100Khz.

One thing that jumps out at me from the iPhone test, is that ringing is due to a sharp low pass filter. It might be the one after the DAC in which case it's going t be ringing at something like 20Khz. The point it that it's very high (ultra sonic for me!) frequency and you probably don't what that coming out of your power amp anyway, I'm thinking intermodulation and nasty high harmonics. The sweep will tell you how much and where. The issue shows up as the tilt on the square wave test.

The traditional way of looking at the loop gain to design the feedback network is open the loop and to drive the feedback node (5.6k resistor) from a signal generator and then look at the response where the loop was broken. It's gets a little tricky as you need to choose source and load impedance so the loop doesn't know it;s been opened. It's even trickier if there is DC feedback. There are other ways too. To really make it fly you need phase info too and now the test gear s getting expensive. At this point it subject turns into a book so I suggest you search using suitable terms such as "open loop testing". I think the guys at Cleverscope have a nice white paper on it IIRC.

[loudthud]

You have to watch the open loop gain with EL84s. Did you measure it? What is the OL gain and gain after feedback? Try 1K or 470 in series with the Presence cap.

[Helmholtz]

Tube PA gain varies with load impedance. So (open loop) gain and phase response will be different at a real speaker load. Preferably stability testing should be done with a load that emulates speaker impedance at high frequencies up to the 100kHz range, where instabilities often show.

[uneumann]

... I was thinking of adding the 8k2/200pF circuit shown in parenthesis on my schematic.
Second, what kind of linearity should I be looking for? What can be inferred from the wave forms above?
Last, any thoughts on taming the HF in the presence circuit? I think more useful would be some kind of actual small bandpass filter to shut actual mid frequencies for an active mid boost.

Rather than the 8k/200pf you show - you might try bypassing the 120K NFB resistor with a small cap - try something between 500pf and 1000pf. That provides additional NFB at very high freq's - likely above what you can hear from the spkr, while still allowing the presence control to work in the audible range. Besides removing highs you can't hear, it reduces the stage gain at these HFs and may help stabilize the amp, if it has a stability problem.

[Helmholtz]

Rather than the 8k/200pf you show - you might try bypassing the 120K NFB resistor with a small cap - try something between 500pf and 1000pf. That provides additional NFB at very high freq's - likely above what you can hear from the spkr, while still allowing the presence control to work in the audible range. Besides removing highs you can't hear, it reduces the stage gain at these HFs and may help stabilize the amp, if it has a stability problem.

Yeah, I think this is called lead (as opposed to lag) compensation. And original proposal seems to work mainly on one half-cycle.
Don't forget the option of a Zobel/Boucherot network across the OT primary.

[Helmholtz]

From the scope shots it appears that there is hardly any increase in gain when disconnecting NFB. Is this real or did you change/adjust input level?

[SoulFetish]

Sorry for the delay in my responses. I've been laid up, fighting a cold for the better part of the last few days.

You have to watch the open loop gain with EL84s. Did you measure it? What is the OL gain and gain after feedback?

I did measure it... at least attempt to measure it. I was getting what looked like conflicting measurements. On my multimeter which claims true RMS, I was getting a figure which seemed quite a bit off from looking at the p-p sine wave on my scope crudely dividing V_peak/1.414.
So, I'm going to measure it again, and confirm it using another true rms meter. I'll probably do that tonight.

Try 1K or 470 in series with the Presence cap.

Thanks, I'll try that.

[SoulFetish]

From the scope shots it appears that there is hardly any increase in gain when disconnecting NFB. Is this real or did you change/adjust input level?

right, that was by design. Looking at the schematic, I put in a potentiometer at the NFB switch to equalize the gain so I can get a more accurate feel for any differences in touch dynamics, sound, etc.
The pot will ultimately be replaced by a fixed voltage divider.

If helpful, I can remove the attenuation and show traces of true open loop gain and closed loop gain.

[SoulFetish]

Rather than the 8k/200pf you show - you might try bypassing the 120K NFB resistor with a small cap - try something between 500pf and 1000pf. That provides additional NFB at very high freq's - likely above what you can hear from the spkr, while still allowing the presence control to work in the audible range. Besides removing highs you can't hear, it reduces the stage gain at these HFs and may help stabilize the amp, if it has a stability problem.

I was thinking about this as an option as well. I believe this is a more common approach. I wasn't sure about what values to try. I tried bypassing it with a 100pF and 200pF, but saw no visible change in the leading edge of the square wave. I'll experiment with the values you suggested.

Yeah, I think this is called lead (as opposed to lag) compensation. And original proposal seems to work mainly on one half-cycle.
Don't forget the option of a Zobel/Boucherot network across the OT primary.

I was wondering if slugging the dominant pole may be the way to go. I got this idea reading Turner's Audio site http://www.turneraudio.com.au/ (or it may have been Morgan Jones). In any case, the suggestion was to connect this network from plate to ground to compensate from the HF phase shift created by the capacitance to ground in the loop. But, the half cycle observation is a good one. I believe in his example feedback was inserted in a 12AX7 stage feeding a cathodyne phase inverter. In my case, I would probably have to add it to the non-inverting plate as well.
I considered the Zoble network across the OT primary, but didn't know how to implement it with the MOV over voltage protection I have from each plate to the CT.

[SoulFetish]

Square wave testing is great for a for an instant overall picture or what is going on but it's hard to get useful numbers out of it. Therefore you should consider running a sine wave frequency sweep through it. Say 100hz to 100Khz.

I don't thing I'll have any problem at 100Hz, but I'm wondering if Ill run into slew limiting distortion at 100kHz with my humble 1458. Oh, wait, you said sine wave. I can find something to do that.

The traditional way of looking at the loop gain to design the feedback network is open the loop and to drive the feedback node (5.6k resistor) from a signal generator and then look at the response where the loop was broken. It's gets a little tricky as you need to choose source and load impedance so the loop doesn't know it;s been opened. It's even trickier if there is DC feedback. There are other ways too. To really make it fly you need phase info too and now the test gear s getting expensive. At this point it subject turns into a book so I suggest you search using suitable terms such as "open loop testing". I think the guys at Cleverscope have a nice white paper on it IIRC.

I don't know If I understand what you mean by driving the 5k6 node with the signal. I'll read up on it though..
Do you see any problem from where I'm inserting the signal currently? (marked on the schematic)

[trobbins]

It can be difficult for others to help if the water is muddy for starters.

Can you find a nice sinewave signal, such as make up a wav or mp3 file and burn it on a cd or add to your phone or a USB memory player or ipod. Just 100-200Hz is fine as your meter may have that bandwidth (have you looked at your meter specs?) and if there is no hum loop (ie. a battery powered player) then it could be connected to the 'input' (before the 4n7 and to ground). That needs to be at a level that doesn't get near output clipping with or without feedback.

Can you disconnect the tone shaping presence network, and do you have a 10-20W power resistor of about 6-10 ohm to connect to the output? Is 'com' taken directly to the ground symbol of the Marshally LTP ?

What is the bandwidth of your Rigol and are you using a 10:1 probe that is compensated ok?

You should then be able to compare output signal level on meter and scope when feedback is in and out, and then calculate feedback drop in dB.

That may give us a better idea when the 1-2kHz square wave is applied to the input (as per the sine source connection) and you show us a scope plot.

If you don't have a sinewave generator, or a 192kHz sampling soundcard or USB interface of some kind, then its a bit difficult to do a frequency sweep and see if there are any quirky peaks, or what you base frequency response is with and without feedback. If you got that far, then you could look further at low and high frequency performance issues.

[SoulFetish]

It can be difficult for others to help if the water is muddy for starters.

okay, lets see if I can clear things up

Can you find a nice sinewave signal, such as make up a wav or mp3 file and burn it on a cd or add to your phone or a USB memory player or ipod. Just 100-200Hz is fine as your meter may have that bandwidth (have you looked at your meter specs?) and if there is no hum loop (ie. a battery powered player) then it could be connected to the 'input' (before the 4n7 and to ground). That needs to be at a level that doesn't get near output clipping with or without feedback.

I can get an actual signal generator, but I've been using a signal generator on my phone (battery powered, and isolated). For sine waves, it seems adequate, but as show above, useless for square waves. I can set and adjust the output amplitude to either clip the input, or not. But before the 4n7 cap is a DC potential of the preceding stage's plate voltage and there's no way I'm putting that across the output of my phone. I can buffer it and capacitively couple it if that is where to inject the signal.

Can you disconnect the tone shaping presence network, and do you have a 10-20W power resistor of about 6-10 ohm to connect to the output? Is 'com' taken directly to the ground symbol of the Marshally LTP ?

It is a bit more "Marshally" than "Fendery" no doubt.
Here is my home shop bench equipment I'm working with:
20A variac with digital rms readout.
Incandescent current limiter (various values)
4Ω/300W, 8Ω/300W, and 16Ω300W resistive loads.
100MHz rigol
10:1 probe compensated to the FFT calibration port at the front of the scope
2 DMMs both true RMS

What is the bandwidth of your Rigol and are you using a 10:1 probe that is compensated ok?

see above

You should then be able to compare output signal level on meter and scope when feedback is in and out, and then calculate feedback drop in dB.
That may give us a better idea when the 1-2kHz square wave is applied to the input (as per the sine source connection) and you show us a scope plot.

yes, I will measure the open loop gain/closed loop gain, and post it. I would have done it last night, but it was 4˚F outside and my garage was too damn cold

If you don't have a sinewave generator, or a 192kHz sampling soundcard or USB interface of some kind, then its a bit difficult to do a frequency sweep and see if there are any quirky peaks, or what you base frequency response is with and without feedback. If you got that far, then you could look further at low and high frequency performance issues.

If I need to, I can bring it into work where I have access to more equipment.
cool, thanks for your questions and recommendation.

[Helmholtz]

One of the reasons I wanted to test this was I was getting some shrill high frequency dominating the sound with the presence control at max settings.
Starting with the GFB engaged, I tested the output with the presence turned all the way down and here is the shot:
[ATTACH=CONFIG]52016[/ATTACH]

Here is the presence at full:
[ATTACH=CONFIG]52017[/ATTACH]

Here is the output with no global feedback:
[ATTACH=CONFIG]52018[/ATTACH]

.

As the square wave test shows no ringing or overshoot, there will be no pronounced high frequency peak in the response. (Both square wave and frequency response will be different into a speaker, though).

Also the effect of the presence control as well as the level difference without NFB are rather small (this correlates of course).

If you just need to reduce some shrillness you might consider a "cut circuit" as used in AC30s, either with a trimmer or in place of the presence control.