So you've built a diy amp, it tests ok with a restive dummy load but when you plug it into a test speaker it shows signs of instability, how do you personally go about testing and then making the amp stable?
So far I've only found two references online to the method, the best being this one by Patrick Turner - amplifier-frequency-tests I looked in the Tek Cookbook and in Metzler - nada. There is obviously some info in Radio Designers but for those who aren't big on theory it can be a bit opaque. After having read Patricks article there's a few things I'm unsure of but AFAIK the procedure is this.
With the amp in open loop and closed loop and with various capacitive loads do frequency sweeps looking for where THD is greater than 2% or where the OP of an amp peaks.
So for instance I ran several sweeps yesterday with my recent 807 PPP amp with a 100n and then a 470n as load, noting down where THD exceeded 2%, no peaks were noted or increase in THD.
I then ran a sweep with my big test speaker which is a cobbled together job of a 15" Celestion woofer cab, 10in Celestion in another cab, and two Fane horns in another cab, all through a Kef Xover, this typically is a difficult load. I got more clear results where THD shot up to 80% at 15hz and at 45khz and again at 111khz, this was using a HP3903B to monitor the OP as well as a B&K AC voltmeter, signal source an Advance Type 81A.
I'll run some more but would be very interested in other builders methods, I'm after practical methods rather than maths heavy methods involving bode plots, Nyquist diagrams and imaginary numbers please. I'll probably change the odd coupling cap and fit a few step networks as well as a zobel network to reduce gain at HF but interested in other members wrinkles.
Andy.
So far I've only found two references online to the method, the best being this one by Patrick Turner - amplifier-frequency-tests I looked in the Tek Cookbook and in Metzler - nada. There is obviously some info in Radio Designers but for those who aren't big on theory it can be a bit opaque. After having read Patricks article there's a few things I'm unsure of but AFAIK the procedure is this.
With the amp in open loop and closed loop and with various capacitive loads do frequency sweeps looking for where THD is greater than 2% or where the OP of an amp peaks.
So for instance I ran several sweeps yesterday with my recent 807 PPP amp with a 100n and then a 470n as load, noting down where THD exceeded 2%, no peaks were noted or increase in THD.
I then ran a sweep with my big test speaker which is a cobbled together job of a 15" Celestion woofer cab, 10in Celestion in another cab, and two Fane horns in another cab, all through a Kef Xover, this typically is a difficult load. I got more clear results where THD shot up to 80% at 15hz and at 45khz and again at 111khz, this was using a HP3903B to monitor the OP as well as a B&K AC voltmeter, signal source an Advance Type 81A.
I'll run some more but would be very interested in other builders methods, I'm after practical methods rather than maths heavy methods involving bode plots, Nyquist diagrams and imaginary numbers please. I'll probably change the odd coupling cap and fit a few step networks as well as a zobel network to reduce gain at HF but interested in other members wrinkles.
Andy.
Last edited:
"Are you using the amplifier for HF or VHF?" Neither, AF. If you roll off gain at too low a frequency, like at 20khz unless your using a very sharp cutoff filter it will effect gain at lower frequencies, you can't get a sharp cutoff using a simple RC filter. Also as PCL200 mentions that would affect transient response.
Anyhoo, that's beside the point, I'm after folks testing methods and methods of fettling an amp for unconditional stability.
Andy.
Anyhoo, that's beside the point, I'm after folks testing methods and methods of fettling an amp for unconditional stability.
Andy.
Fully agree here. Firstly you must have a solid basic knowledge of the NFB theory which is difficult without maths and those bode plots or Nyquist diagrams.
Then the subject of stability : how do you want your NFB amplifier is behaving when applying different loads (Pure capacitive load? Pure inductive load? Pure resistive load? A combination of those?) Do you want it unconditionally stable so independenly of the kind of load or conditionally stable (the most practical approach)?
How do you test this is quite another part of NFB
A whole lot of NFB theory/practice is published on that subject which is not at least simple to understand without math.
Just giving my opinion.
I did find a good WEB link on this written in way that is easy to understand - it takes a simple valve amp and takes Lissajous figures to get the phase response. Cannot remember where I found it however sorry. My experience is even when its stable for small signals you can still get oscillation on larger signals where the output stage becomes non linear.
Here is what I do:
HF stability tests are performed using a 1 watt 10 kHz square wave (as measured when normally loaded) with scope hung on the output, and then the output shorted, opened, or operated in a capacitance only load ranging from .005 uF to 1.0 uF.
At some value, the cap only load will certainly produce a maximized ring on the tops and bottoms of the square waveform, but should never cause it to break into full blown oscillation.
LF stability tests are most easily performed by again hanging a scope on the output of the amp, and lightly pulsing the input of the amplifier when operating first into a normal resistive load, and then into no load.
In both cases, the baseline of the scope should settle rapidly, with little after bounce.
A baseline that is slow to settle or actually is stimulated into full blown oscillation obviously indicates trouble. You can pulse the amplifier with any convenient DC source, including the use of a simple battery. Just be sure to use a pulse voltage appropriate for the sensitivity of the amplifier, and an output level that would otherwise produce an output of about 1 watt.
HF stability tests are performed using a 1 watt 10 kHz square wave (as measured when normally loaded) with scope hung on the output, and then the output shorted, opened, or operated in a capacitance only load ranging from .005 uF to 1.0 uF.
At some value, the cap only load will certainly produce a maximized ring on the tops and bottoms of the square waveform, but should never cause it to break into full blown oscillation.
LF stability tests are most easily performed by again hanging a scope on the output of the amp, and lightly pulsing the input of the amplifier when operating first into a normal resistive load, and then into no load.
In both cases, the baseline of the scope should settle rapidly, with little after bounce.
A baseline that is slow to settle or actually is stimulated into full blown oscillation obviously indicates trouble. You can pulse the amplifier with any convenient DC source, including the use of a simple battery. Just be sure to use a pulse voltage appropriate for the sensitivity of the amplifier, and an output level that would otherwise produce an output of about 1 watt.
Assuming this is an amp with global feedback (where the output transformer is playing a key role in the response when feedback is applied), and you don't know its stability performance (because its diy and the OPT hasn't been tested by others), then I'd start with a baseline gain-phase plot with no feedback.
Practically, the high frequency end can be measured using a soundcard type system (unless you have more exotic means) with a normalised response out to at least 90kHz (ie. 192kHz sampling rate and with loop back response normalised for flat gain-phase). Typically that type of measurement system won't extend down below a few Hz, and so some OPT response may be at or below your measuring system, and so you could resort back to X-Y plot on scope (good perveance helps) - which you could also use for the HF end as well. The key datapoints to gather are at step phase shift angles like 45, 90, 135, 180, and to look for HF resonance dips and picks and identify those frequencies and gains.
That then gives you a good starting point for determining how much feedback can be applied at the low and high frequency ends, and hence guide you as to whether you just want a flat feedback response that doesn't encroach on poor margins at either end, or whether you want to shelf the feedback level to have more feedback mid-range but still adequate margins at low and high ends.
If you just had a wide-band voltmeter, then you could at least measure gain response - but few voltmeters go really low or really high in frequency response (I had to use a HP3400A for a Williamson as the DMM I had at that time rolled off at 2-300kHz).
You can avoid any maths up until when you want to work out what shelving may be useful, although even then you should really retest to confirm the actual forward path response.
Total system (with feedback applied) response is then the next practical measurement activity imho. And although squarewave response for high frequency stability is the typical testing used, the response at low frequency usually forces you back to X-Y or gain-phase plots. I'd then aim to do severe load type testing afterwards, like capacitor only, or no load.
Practically, the high frequency end can be measured using a soundcard type system (unless you have more exotic means) with a normalised response out to at least 90kHz (ie. 192kHz sampling rate and with loop back response normalised for flat gain-phase). Typically that type of measurement system won't extend down below a few Hz, and so some OPT response may be at or below your measuring system, and so you could resort back to X-Y plot on scope (good perveance helps) - which you could also use for the HF end as well. The key datapoints to gather are at step phase shift angles like 45, 90, 135, 180, and to look for HF resonance dips and picks and identify those frequencies and gains.
That then gives you a good starting point for determining how much feedback can be applied at the low and high frequency ends, and hence guide you as to whether you just want a flat feedback response that doesn't encroach on poor margins at either end, or whether you want to shelf the feedback level to have more feedback mid-range but still adequate margins at low and high ends.
If you just had a wide-band voltmeter, then you could at least measure gain response - but few voltmeters go really low or really high in frequency response (I had to use a HP3400A for a Williamson as the DMM I had at that time rolled off at 2-300kHz).
You can avoid any maths up until when you want to work out what shelving may be useful, although even then you should really retest to confirm the actual forward path response.
Total system (with feedback applied) response is then the next practical measurement activity imho. And although squarewave response for high frequency stability is the typical testing used, the response at low frequency usually forces you back to X-Y or gain-phase plots. I'd then aim to do severe load type testing afterwards, like capacitor only, or no load.
Last edited:
Rather than talk about a specific amp I'm after generic testing.
"Problem is there's not easy way to understand NFB without the maths and bode plots. Anything else is just magic." Problem is the maths defeats me at present, also one may know the theory but still has to test whether that theory or maths is correct. Lastly there's more than one way of skinning a cat or approaching a problem.
"I'd start with a baseline gain-phase plot with no feedback." I've tried looking at phase/gain on an analogue scope, it's not easy to measure phase accurately. If using the scope in XY mode how would one practically wire it up? Gain is I presume the difference between IP & OP amplitude and I would presume that goes into the vertical amp, but how does one separate phase?
I'll read that article Ray and do some research on this subject. Thanks all for your IP. Andy.
"Problem is there's not easy way to understand NFB without the maths and bode plots. Anything else is just magic." Problem is the maths defeats me at present, also one may know the theory but still has to test whether that theory or maths is correct. Lastly there's more than one way of skinning a cat or approaching a problem.
"I'd start with a baseline gain-phase plot with no feedback." I've tried looking at phase/gain on an analogue scope, it's not easy to measure phase accurately. If using the scope in XY mode how would one practically wire it up? Gain is I presume the difference between IP & OP amplitude and I would presume that goes into the vertical amp, but how does one separate phase?
I'll read that article Ray and do some research on this subject. Thanks all for your IP. Andy.
This is a topic that has interested me, coming from an IT background.
In IT terms, what Diabolical Artificer is looking for is 'black box testing', whereas the consensus is that it has to be 'white box testing', i.e. it is not possible to test an amplifier without exposing the implementation in some way.
The bandwidth of an amplifier is part of the implementation, so should enable it to be tested in a black box way, but from reviewing that article posted above, it is very difficult to test an amplifier in a non-intrusive way, hence there will always be an element of white box testing.
What is the goal of the testing? To identify potentially dangerous oscillation, and make the amplifier safe before it is connected to real speakers? Or to analyse the metrics of performance, e.g. bandwidth over a pre-determined frequency range?
Is it possible to identify a design issue that causes oscillation without having to perform multiple tests?
Reading a bit between the lines, there is a temptation to ask valve amplifiers to perform at more extreme levels than they did historically, and should we be more realistic in the performance expectations?
I have just restored a couple of Quad IIs, and I find it hard to be subjective about the sound. Yes, it is OK, and yes , it is different to the Tubelab SSE, but I am also interesting in getting a better understanding of the philosophy of performance verification, and making it quantifiable, so following this thread with interest!
In IT terms, what Diabolical Artificer is looking for is 'black box testing', whereas the consensus is that it has to be 'white box testing', i.e. it is not possible to test an amplifier without exposing the implementation in some way.
The bandwidth of an amplifier is part of the implementation, so should enable it to be tested in a black box way, but from reviewing that article posted above, it is very difficult to test an amplifier in a non-intrusive way, hence there will always be an element of white box testing.
What is the goal of the testing? To identify potentially dangerous oscillation, and make the amplifier safe before it is connected to real speakers? Or to analyse the metrics of performance, e.g. bandwidth over a pre-determined frequency range?
Is it possible to identify a design issue that causes oscillation without having to perform multiple tests?
Reading a bit between the lines, there is a temptation to ask valve amplifiers to perform at more extreme levels than they did historically, and should we be more realistic in the performance expectations?
I have just restored a couple of Quad IIs, and I find it hard to be subjective about the sound. Yes, it is OK, and yes , it is different to the Tubelab SSE, but I am also interesting in getting a better understanding of the philosophy of performance verification, and making it quantifiable, so following this thread with interest!
diyaudio has an Equipment and Tools section. Many threads in that section go to how to do your testing. Many find it quite easy to connect an amplifier to their test setup and measure gain and phase, and there are a few threads that go down that path. Some test gear is 'intensive' and uses custom hardware and is aimed for very low noise and distortion measurements. Some test gear uses a relatively cheap USB soundcard and free software like REW, and all the testing is basically automated by the software. So effectively black box testing for starters, up until you want to do some changes to the amp.
Why has the world forgotten how to use a sweep oscillator and a gain-set to measure amplifiers??
Pink noise and FFT is pretty, and can tell you something. WHAT it tells you is not always clear.
Sweep the oscillator full-range to the ACVM, note the errors (mostly flutter in subsonic and slump in high and low). Now turn-down and feed through the amplifier. Any *sudden* bends or kinks in the gain-curve (you don't have to plot it except to report to your Boss) are likely trouble. Keep track of dB/octave (slope). 6dB/8ve is benign, 12dB/8ve is too steep.
On some amps the low-low-level response, normal level, and clipping level responses are very different. My last AM transmitter (supposed to be a headphone amp) was fine until it clipped and then it went nuts. Yes, it can be hard to find ALL the corner cases.
Pink noise and FFT is pretty, and can tell you something. WHAT it tells you is not always clear.
Sweep the oscillator full-range to the ACVM, note the errors (mostly flutter in subsonic and slump in high and low). Now turn-down and feed through the amplifier. Any *sudden* bends or kinks in the gain-curve (you don't have to plot it except to report to your Boss) are likely trouble. Keep track of dB/octave (slope). 6dB/8ve is benign, 12dB/8ve is too steep.
On some amps the low-low-level response, normal level, and clipping level responses are very different. My last AM transmitter (supposed to be a headphone amp) was fine until it clipped and then it went nuts. Yes, it can be hard to find ALL the corner cases.
"Why has the world forgotten how to use a sweep oscillator and a gain-set to measure amplifiers??" That is essentially what I've been doing but with different loads. I have a HP3903B Audio Analyser that'll do a frequency sweep from 20-50khz automatically, but I prefer a manual sweep for this job, the auto sweep goes too fast at HF, easy to miss things.
Having read the original Patrick Turner article here's what I did.
Why test?BTW this applys to tranny amps too, say we have a tranny amp, we've just built it and tested it on a test speaker it sounds ok, but we noticed a bit of instabilty on SW off or fuzzyness of the sinewave at HF or built a valve amp using non audio transmitter type valves, it could be singing away at HF and we'd be non the wiser. Example, I had a HH S500D, sounded sweet, used it for years, prior to selling it I hooked it up to the scope and did a sweep, it was oscillating away at 20v RMS @ several hundred khz.
So, to testing as prescribed by P T. Hook your amp up to a sig gen with half decent THD, P T says better than 0.5%, monitor the OP of the amp with a scope and or wideband AC voltmeter, load is a 8r non inductive resistor. We want to sweep from DC to at least 200khz, though 10-20hz is more realistic, check for peaks or increased THD using a distortion meter, FFT or whatever, there's a bit of free software called Soundcard scope, it's THD meter is as accurate as a HP3903B. Anyhoo, we note down any peaks or changes or do a frequency response graph either by hand or with software, though in my experience most free software doesn't go over 20khz, like SCS. All this is done at 0dB, IE at full whack/just before clipping and with no NFB.
Next we do the same thing but at -3dB but with a capacitor as load, and with NFB applied, how much? Depends on how stable the amp is, I usually use a 5k 10 tun pot starting with a few dB of NFB, increasing as i go/make changes. We start with 0.1u, it could be that the amp won't like this at -3dB, -12dB might be more realistic. Do the same again but with a 1u or 2u cap as load. Lastly a test with no load is done but at low amplitude to check LF stability. I hope I have that right, read P T's article - amplifier-frequency-tests for more info.
In my case I tested my latest amp and got a dramatic increase of THD at 18hz, 45khz and a slight peak at 111khz. To counter any LF instability I used LF shelving filter which was inserted between the IP stage and LTP PS, EG a 1m resistor and 47n cap in parallel, this cuts gain at 0.33hz from memory. To deal with the 45khz I used a 12r6 resistor and 270n cap across the secondary of the OPT as a zobel network as well as a 630p cap in parallel with the 5k6 NFB resistor. To deal with the 111khz I put a 470p and 3k resistor across the LTP grid leak resistor.
After this I did further testing, these cured the issues described and the amp has good frequency response from 20hz to 20khz.
I'll look into the phase method when i get a mo, thanks for the link and IP.
Andy.
Having read the original Patrick Turner article here's what I did.
Why test?BTW this applys to tranny amps too, say we have a tranny amp, we've just built it and tested it on a test speaker it sounds ok, but we noticed a bit of instabilty on SW off or fuzzyness of the sinewave at HF or built a valve amp using non audio transmitter type valves, it could be singing away at HF and we'd be non the wiser. Example, I had a HH S500D, sounded sweet, used it for years, prior to selling it I hooked it up to the scope and did a sweep, it was oscillating away at 20v RMS @ several hundred khz.
So, to testing as prescribed by P T. Hook your amp up to a sig gen with half decent THD, P T says better than 0.5%, monitor the OP of the amp with a scope and or wideband AC voltmeter, load is a 8r non inductive resistor. We want to sweep from DC to at least 200khz, though 10-20hz is more realistic, check for peaks or increased THD using a distortion meter, FFT or whatever, there's a bit of free software called Soundcard scope, it's THD meter is as accurate as a HP3903B. Anyhoo, we note down any peaks or changes or do a frequency response graph either by hand or with software, though in my experience most free software doesn't go over 20khz, like SCS. All this is done at 0dB, IE at full whack/just before clipping and with no NFB.
Next we do the same thing but at -3dB but with a capacitor as load, and with NFB applied, how much? Depends on how stable the amp is, I usually use a 5k 10 tun pot starting with a few dB of NFB, increasing as i go/make changes. We start with 0.1u, it could be that the amp won't like this at -3dB, -12dB might be more realistic. Do the same again but with a 1u or 2u cap as load. Lastly a test with no load is done but at low amplitude to check LF stability. I hope I have that right, read P T's article - amplifier-frequency-tests for more info.
In my case I tested my latest amp and got a dramatic increase of THD at 18hz, 45khz and a slight peak at 111khz. To counter any LF instability I used LF shelving filter which was inserted between the IP stage and LTP PS, EG a 1m resistor and 47n cap in parallel, this cuts gain at 0.33hz from memory. To deal with the 45khz I used a 12r6 resistor and 270n cap across the secondary of the OPT as a zobel network as well as a 630p cap in parallel with the 5k6 NFB resistor. To deal with the 111khz I put a 470p and 3k resistor across the LTP grid leak resistor.
After this I did further testing, these cured the issues described and the amp has good frequency response from 20hz to 20khz.
I'll look into the phase method when i get a mo, thanks for the link and IP.
Andy.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.