Testing Global feedback stability and question about practical componsation for outputstage

[trobbins]

I'd recommend disconnecting the previous stage, as your test signal can then be applied in the normal manner (ie. to ground), and not applied as a voltage across a resistor within the feedback loop.

Have you checked the frequency response for your DMMs, as a cross-check for measuring sine and square wave rms levels (many DMMs roll off above mains ac or 1kHz type frequencies) against your scope (which probably calculates rms quite well).

You need a remote turn-on feature for a few valve amps in your garage - as a preheat

Measuring the total output stage gain with resistive load and no feedback is the easier way to check for unsuspecting gain peaks at very low and very high frequencies, as they may not be controlled by the feedback.

Setting up an X-Y plot of input and output voltages is also a very easy way to see when phase shift starts to kick in, and when phase shift exceeds 90deg and gets close to or exceeds 180deg, and align that with gain response. If you can get that data, then you can identify how stable the response is in a formal phase/gain margin manner.

[SoulFetish]

You need a remote turn-on feature for a few valve amps in your garage - as a preheat

Ha! ain't that the truth

I'd recommend disconnecting the previous stage, as your test signal can then be applied in the normal manner (ie. to ground), and not applied as a voltage across a resistor within the feedback loop.

okay, sounds good. This is the approach I'll take.

Have you checked the frequency response for your DMMs, as a cross-check for measuring sine and square wave rms levels (many DMMs roll off above mains ac or 1kHz type frequencies) against your scope (which probably calculates rms quite well).

I just tracked down the specs for both DMMs I'm using. I'll post them here, if anyone is interested in them.
Extech EX205T:


Aneng AN8001:

Interestingly enough – I think the Aneng uses a AD536A chip. I took a quick look through the datasheet and it seems like they're barely tapping into that chips capability.
https://www.analog.com/media/en/tech...ets/AD536A.pdf
https://www.analog.com/en/products/ad536a.html

Setting up an X-Y plot of input and output voltages is also a very easy way to see when phase shift starts to kick in, and when phase shift exceeds 90deg and gets close to or exceeds 180deg, and align that with gain response. If you can get that data, then you can identify how stable the response is in a formal phase/gain margin manner.

I...uh.. only have one probe at the moment.
I bought some surplus Pamona insulated panel mount BNC connectors to install in my load box so I can run a direct BNC cable and monitor output.
I should probably wire a 10:1 voltage divider to attenuate the signal at the BNC jack then... 1MΩ:100k fine?

[trobbins]

Yeh I have two AN8009 - great up to 1kHz sine.

Your scope will be interesting to see what rms accuracy can be achieved, but it should happily get to above the range you need to be aware of (eg. a few hundred kHz). It appears you need to use the FFT screen, and probably have to play with averaging and windows - there may be some relevant forums/posts (eg. over at eevblog and testgear) or you could start one if you were keen, as it is obviously a tool worth using to its best advantage.

I'd suggest just getting a few clone 10:1 standard scope probes off ebay - they may indicate the BNC input capacitance range they can cope with, to compare with your scope spec. They may not get you 50MHz bandwidth, but that's not the point, as they are cheap and would be fine to a MHz. I have a few, and a 100:1 for use with soundcard inputs. Trying to make your own probe may not easily yield acceptable bandwidth or flatness, and cause some frustration, and make it difficult for others to believe your plots unless you can show how you calibrated it.

[SoulFetish]

Status update:

I was able to get back to this project tonight. So I set up my bench to measure the open loop and closed loop gain, run some other tests, and share the results. Sounds simple enough, right?

trobbins, I followed your instructions for the testing proceedure – I lifted the tone/presence connection; disconnected the previous stage; using an isolated(battery powered) Velleman function generator, I connected the leads to the 4n7 capacitor and to the grounded side of the 5k6 resistor in the tail.
I set the test signal for a 1kHz sine wave, at an amplitude of 200mV RMS and confirmed the voltage with both DMMs.
I disconnected the global feedback so I could measure the open loop gain. I connected my scope at the 8Ω output to observe the signal, and did the same with my multimeter to measure the RMS voltage. This was pretty straight forward. I got a nice clean looking sine wave at the load measuring 4.61VRMS for a gain of ≈23 (27.25dB). Easy peasy.
Then I connected the negative feedback to measure the closed loop gain and here's where it all went to hell....
At first I didn't suspect anything. I powered down between tests. When I was ready to power up, I set my scope on auto and was checking my meter for a voltage reading and got a measurement of around 447mV at the output. I was like 'What the hell is that.. down something like -20dB??'
That didn't seem right, so I looked at the sine wave on my scope and could see what looked like quite a bit of noise on the signal and noticed that my scope was set to 10µs/div. I set my meter to Hz to double check and it was measuring around 40kHz! When I set my scope to 500µs/div, there was way to much high frequency content to make out anything resembling a 1kHz sine wave.
So,... uh. what do I do now?
did I misunderstand your setup instructions?

[trobbins]

At least you made a good start

I'd suggest you dumb down the feedback level by say inserting 1-2Megohm in series with the 120k. Switching that feedback in/out should then hopefully only show a small change in clean output voltage - if the voltage goes down a bit then good - if it goes up then that's positive feedback.

Perhaps without any feeback connected, it is a good idea to confirm the amp shows relatively clean, lowish distortion sinewave as input voltage varies, and then output stage clipping starts to show at some nominal output voltage and load resistance, and then the clipping gets more towards a square wave with higher input.

If you can vary frequency, then sweeping across the spectrum (well below clipping level) to find when output voltage halves at each end of the spectrum is a good initial measurement to. If you had a spectrum analyser that would then identify if the noise floor was flat, and not showing any spurious frequencies starting up unexpectedly as you raised signal level to clipping.

[SoulFetish]

At least you made a good start

I'd suggest you dumb down the feedback level by say inserting 1-2Megohm in series with the 120k. Switching that feedback in/out should then hopefully only show a small change in clean output voltage - if the voltage goes down a bit then good - if it goes up then that's positive feedback.

Perhaps without any feeback connected, it is a good idea to confirm the amp shows relatively clean, lowish distortion sinewave as input voltage varies, and then output stage clipping starts to show at some nominal output voltage and load resistance, and then the clipping gets more towards a square wave with higher input.

If you can vary frequency, then sweeping across the spectrum (well below clipping level) to find when output voltage halves at each end of the spectrum is a good initial measurement to. If you had a spectrum analyser that would then identify if the noise floor was flat, and not showing any spurious frequencies starting up unexpectedly as you raised signal level to clipping.

A couple of things I've been thinking about over the last hour or so:
I'm wondering if I left the feedback vulnerable to oscillations and instability in carelessly making physical changes to accommodate the test setup. I had jumpers and shorting leads, coupled with meter probes spilling out of the small area where this circuit is located.
The other thing is now that I have a pretty good idea what the open loop gain is, can I determine the feedback ratio from the 5k6 /120k resistors? Will this give me a "in a perfect world" idea of what to expect for a closed loop gain?

[trobbins]

There is a lot of text-book circuit analysis to identify and wade through to preempt what feedback dB would occur. I recommend doing a few steps in feedback resistance to get some measured dB drop levels for starters.

Unless your layout and wiring is very tidy and adequately avoiding signals from parasitically coupling to circuitry it shouldn't, then there is some doubt as to how clean your signal is being amplified (before even feedback is considered).

Perhaps an easy path is to use square wave response to identify the onset of instability when choosing a certain dB level of feedback and a bad load like resistance and parallel capacitance - there are a few good magazine articles and websites describing how to confirm unconditional stability - well worth finding and reading, as they will give you confidence if you want to make the effort of testing.

In the long run, the feedback level may end up determining the amp gain structure for your input level and clipping/cranking.

[SoulFetish]

Unless your layout and wiring is very tidy and adequately avoiding signals from parasitically coupling to circuitry it shouldn't, then there is some doubt as to how clean your signal is being amplified (before even feedback is considered).

some considerable thought went into the physical layout, it more or less still looks like this:

[Dave H]

The other thing is now that I have a pretty good idea what the open loop gain is, can I determine the feedback ratio from the 5k6 /120k resistors? Will this give me a "in a perfect world" idea of what to expect for a closed loop gain?

Feedback ratio is 5.6/(120+5.6) = 0.045 (i.e. the 120k to 5k6 potential divider ratio)
Open loop gain is 23
Which would make the closed loop gain 11.3

But you are not going to be able to measure it with all that oscillation going on.

[trobbins]

Dave shorthanded the calculations. Perhaps the easiest reference to look at for your circuit is http://www.aikenamps.com/index.php/d...ative-feedback.

Can you confirm you connected an 8 ohm output load resistor (instead of a speaker) to the 8 ohm tapping to make your open-loop gain measurement - as that wasn't clarified in your earlier post.

[Helmholtz]

Dave shorthanded the calculations. Perhaps the easiest reference to look at for your circuit is http://www.aikenamps.com/index.php/d...ative-feedback.

Can you confirm you connected an 8 ohm output load resistor (instead of a speaker) to the 8 ohm tapping to make your open-loop gain measurement - as that wasn't clarified in your earlier post.

As already mentioned, the open-loop gain depends on load impedance (see Aiken link "The Effect of Changes in Load Impedance"), so using a load resistor is not representative for a speaker load.

In other words: There isn't much sense in measuring a tube amp's open loop frequency response into a fixed load resistor.

[trobbins]

Using a fixed resistance loading is the best way to take benchmark measurements, if you were keen enough to make measurements. The measurement being discussed was a single mid-band measurement for the purpose of identifying what nominal level of feedback was being applied (ie. 6dB if the open-loop gain measurement was with an appropriate resistive load).

Another measurement set is to look at the open-loop frequency response, mainly to identify if there are unexpected gain peaks and phase responses at the ends of the audio bandwidth. That type of measurement can be more onerous to perform if a suitable test setup is not available, but can be very simple and cheap to make nowadays with software tools. Doing that measurement with anything but a fixed resistor load makes it difficult to appreciate why unexpected peaks are occurring, as equivalent circuit models are used to interpret response quirks and that analysis is typically based on fixed resistance loading, especially where an output transformer is included in the signal path.

Although any typical speaker part has a non-linear response, that doesn't mean the amplifier, or the speaker unit doesn't have linearising networks in place (especially for mid-band to high=band frequencies), or they couldn't be added.

One aspect of assessing if an amp is effectively unconditionally stable is to not use a fixed load resistance, but rather replace it with effectively worst-case speaker loading examples (eg. capacitance only, and no load). If stable response is observed, then its highly likely that any speaker load can be connected without concern.

[SoulFetish]

Can you confirm you connected an 8 ohm output load resistor (instead of a speaker) to the 8 ohm tapping to make your open-loop gain measurement - as that wasn't clarified in your earlier post.

Yes i can confirm that

[Helmholtz]

Although any typical speaker part has a non-linear response, that doesn't mean the amplifier, or the speaker unit doesn't have linearising networks in place (especially for mid-band to high=band frequencies), or they couldn't be added.

I wasn't speaking about speakers' frequency responses but of frequency dependent speaker impedance. (Guitar speaker don't usually have "linearising networks" in place.)
An 8 Ohm guitar speaker has an impedance of over 30 Ohm @ 10kHz and over 100 Ohm @ 100kHz. (Above around 400Hz a speaker's phase is inductive.)
Consequently open loop high frequency gain with a speaker load will be significantly higher than at a resistor. Remember that the open loop output impedance of tube amps is in the 100 Ohm range.

Testing tube amps' stability with capacitive or purely resistive loads is unrealistic and will not reveal problems with a real (inductive) speaker load at high frequencies.

[Dave H]

Dave shorthanded the calculations.

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.