MikeB said:
Mike,
Thanks for data, very educational.
The comment I like to make is firstly that 20% sag is unrealistic, even under prolonged full power you will probably see not more than 5%, 10% tops in a bad case.
Secondly, did you model the sag on cycle-by-cycle basis or what? I think you will NOT see a 'sag-unsag' on a 10kHz or even a 1kHz ritmh. What is realistic is to have a certain tone(s) playing for a short period, and the supply sagging say 5% from the beginning of the tone period until the end. Is this how you modeled it? Because that is very different (and much more benign) than assuming that there is a supply modulation on a cycle-by-cycle basis.
Jan Didden
Power supply ripple has two components. One is resistive and happens in each cycle, but it makes things much like if the collector or drain swing of all devices was slighly increased, except when sensitive things are referenced between rails and ground (great mistake). Power ground also suffers from ripple relative to signal ground, and that's a good reason to avoid to use it as a reference.
The other component is due to PSU capacitors discharging, it's made of sub-audio frequencies and follows the "density" of the music program (only for unregulated PSUs)
The other component is due to PSU capacitors discharging, it's made of sub-audio frequencies and follows the "density" of the music program (only for unregulated PSUs)
darkfenriz said:
Yes - circletron source follower.
Balanced topology, all n-p fet, class A, cascoded buffer before follower.
Tested 20kHz, 25w, 8hOm load.
Eva said:
Indeed, but that is NOT the same as a cycle-to-cycle supply modulation. So it is not at all clear to me how that modulation mechanism works. One can argue that if the supply slowly sags, say with a period 10 or 50 times the signal period, what are the modulation effects? Just would like some more info on exactly how MikeB modelled this.
Jan Didden
janneman said:
Jan, I would respectfully disagree - most amps don't have a regulated PSU, so for them, 'sag' is the lowest rail voltage at any given time. Looking at the average or RMS will give you the wrong result as far as amplifier maximum output is concerned. Looking at the unloaded PSU (or very lightly loaded, at average power) and the fully loaded PSU (transient or steady state), a 20% variation in lowest rail voltage is not overly pessimistic. In fact, soft power supplies that can show even worse results, are often used as a form of overload protection.
If the rails are shared (without any filtering) by the front end, what they see a waveform that has a rising edge somewhat like a sawtooth, as the rectifiers charge the rail PSU capacitor, with small steps as halfwave rectified output current needs to be shunted from transformer to the output. Then, you have a part where the capacitor is discharged by halfwave rectified output current (looks kind of like a series of downward steps). One look at it is enough to see this waveform is nasty.
Essentially, we have the halfwave rectified output current, the rectification process makes plenty of harmonics. Even with a perfect DC supply, any amount of internal resistance of this supply feeds back these harmonics back into the front end, subject to PSRR.
Then, we have the rectification sawtooth, which is comprised of 100/120Hz fundamental and plenty of higher harmonics. The fundamental here is not a great concern for most amps as PSRR for low frequencies is usually not a problem. But, depending on the characteristic of the rectifiers (reverse recovery effects, amongst others), the harmonic spread can extend quite high. Again, PSRR may become a problem, though lesss evere compared to the above.
Finally, and perhaps of most concern - both of the above are combined through the nonlinear impedance of the PSU - the impedance changes depending in what phase the rectifiers are. So, you get intermodulation. PSRR issues will recirculate this into the output, ad infinitum. So, when one mentions PSU audibility, what really needs to be looked at is the PSRR.
When modeling this, two key points should be observed:
1) You must model the ESR of the filter caps, and the impedance of the transformer.
2) Be sure to check how this works for tle lowest decade of the audio spectrum - results are often eye-opening, you just need to look at the power rail waveform 🙂
Once more, let me reiterate that all of this is at close to maximum output, in real world settings, this is relevant with transients and clipping.
It should be noted that even when the front end power supply is 'perfect' regulated DC, there are mechanisms that produce intermodulation between the PSU waveform and the output - one example would be the nonlinear Cgd of output MOSFETs, where one end of this capacitance is connected to the driver output, and the other to the power rail. At high outputs, where Vg ~~ Vd, the output will contain intermodulation products of the power rail waveform and the output waveform, which the NFB must compensate for. Still, this will be a far smaller problem than all of the above where the intermodulation products enter the amp via power rails.
ilimzn,
Rather than repeating your post, let me say the folowing. I agree with almost all you said; the mechanisms are not new and well understood.
But a back-of-the-envelope calculation shows that in a 50Hz, 40V supply with 20.000uF cap, it takes a 16A load puls to sag from 40 t0 32 V between recharging of the cap. Not impossible, but certainly few and far between in normal listeneing events.
I fully agree to your conclusion that it can almost all be translated to PSRR. Taking it from there, I would think that even a lowly FFT would show these as a wide spectrum noise.
For instance, I cannot imagine that you would have a predicted high supply output noise from a bad PSRR and not see it in an FFT or THD measurement!
So, again, although in theory the mechanism will lead to FM modulation, it will be seen in the regular measuments. And even when it IS seen in for instance an FFT, there is still the question whether it is audible.
Showing that something exists doesn't mean it is audible. We always seem to come back to show technically the existing of a phenomenon, but we are totally speechless on a controlled, repeatable way to find out whether we can hear it!
Jan Didden
Rather than repeating your post, let me say the folowing. I agree with almost all you said; the mechanisms are not new and well understood.
But a back-of-the-envelope calculation shows that in a 50Hz, 40V supply with 20.000uF cap, it takes a 16A load puls to sag from 40 t0 32 V between recharging of the cap. Not impossible, but certainly few and far between in normal listeneing events.
I fully agree to your conclusion that it can almost all be translated to PSRR. Taking it from there, I would think that even a lowly FFT would show these as a wide spectrum noise.
For instance, I cannot imagine that you would have a predicted high supply output noise from a bad PSRR and not see it in an FFT or THD measurement!
So, again, although in theory the mechanism will lead to FM modulation, it will be seen in the regular measuments. And even when it IS seen in for instance an FFT, there is still the question whether it is audible.
Showing that something exists doesn't mean it is audible. We always seem to come back to show technically the existing of a phenomenon, but we are totally speechless on a controlled, repeatable way to find out whether we can hear it!
Jan Didden
But this is nothing new, supply ripple forces the Vce or Vds of components to change in the exact same way as the own amplifier output or input voltage swing does. And of course, this Vce and Vds change unavoidably affects the parameters of active devices.
Cascoding and proper regulation of reference currents and voltages are two common practices to reduce the impact of these effects.
By the way, there are interesting things that can be done to investigate into these effects with real components. For example, I've been just watching in my oscilloscope the VAS current waveform (amp isolated from earth, probe grounded to one rail and connected to the other end of the VAS emitter resistor) when playing full power 10Khz tone bursts into 2.5 ohm with in my current amplifier project. The current swing is very small, only 300 uA peak to peak just before clipping, but the waveform appears clearly distorted and shows different flavours of zero-current crossing spikes depending on bias adjustment. All the open-loop defects of the amplifier appear in that kind of measurements 😀
Note that simulation is very bad at predicting the non-linearities of real components.
Cascoding and proper regulation of reference currents and voltages are two common practices to reduce the impact of these effects.
By the way, there are interesting things that can be done to investigate into these effects with real components. For example, I've been just watching in my oscilloscope the VAS current waveform (amp isolated from earth, probe grounded to one rail and connected to the other end of the VAS emitter resistor) when playing full power 10Khz tone bursts into 2.5 ohm with in my current amplifier project. The current swing is very small, only 300 uA peak to peak just before clipping, but the waveform appears clearly distorted and shows different flavours of zero-current crossing spikes depending on bias adjustment. All the open-loop defects of the amplifier appear in that kind of measurements 😀
Note that simulation is very bad at predicting the non-linearities of real components.
janneman said:
Agreed - though I just realized I was talking about the total DC and AC as sag (here one has to calculate the transformer regulation, fuses if present, nonideal rectification, etc) - taking this into account shows that again the 20% is not overly pessimistic (just sort of a worst case - like a 8 ohm nominal amp with 4 ohm load, and the ubiquitous beancounter influenced undersized transformer. Still, for this discussion, it is the AC component that is relevant and for that your back of envelope calculation is the right one. Certainly this will not be seen often, it would take sustained bass notes - anything else would fry the speakers, which we assume are conencted 🙂
Yes. Although a transient driven test with waterfall spectrum diagram would probably reveal even more.
Not much I can say to that except AMEN 🙂
Hi, MikeB,
I may be wrong, but I think in 3 stages power amp, PSRR is mainly determined by VAS transistor (where its base and emitors are connected to rails).
Is there any recipe to make power amp with good PSRR, in particular how to make a good PSRR VAS stage?
I may be wrong, but I think in 3 stages power amp, PSRR is mainly determined by VAS transistor (where its base and emitors are connected to rails).
Is there any recipe to make power amp with good PSRR, in particular how to make a good PSRR VAS stage?
Hi, EVA,
I might be wrong (one more time 😀), but as Janneman said in few post earlier, isn't that weird looking at the base/gate is supposed to be like that? It should looks like the error signal, and shouldn't looks good like the input signal?
Hi, Janneman,
I might be wrong (one more time 😀), but as Janneman said in few post earlier, isn't that weird looking at the base/gate is supposed to be like that? It should looks like the error signal, and shouldn't looks good like the input signal?
Hi, Janneman,
Is there an opposite situation? 😀 Something like we can hear it, but we can't measure it?
lumanauw said:
David,
I don't know. Maybe, maybe not. I doubt it though, I mean, after many 10-s of years of research by 1000's of engineers and listeners, we still don't know for sure how much HD we can hear, how much phaseshift we can hear, how much freq response deviation we can hear.
There are many studies that show that IN VERY SPECIFIC circumstances we can hear a certain phase shift, for example. Or that we can hear 0.01% 3rd harmonic. But as I said, all under very specific circumstances like a single tone, or with headphones in A/B mode.
What can we hear with real music? I think nobody really knows.
What I DO know is that all these "non-linearities" in all these studies were readily measureable. You can easily measure 1 degree phase shift (well, reasonably easily any way), you can easily measure 0.1dB freq response non-flatness, you can easily measure 0.001% 3rd harmonic.
See what I mean? But no, I cannot PROVE that there don't exist audible things we cannot measure.
Jan Didden
lumanauw said:
I don't understand exactly what you mean.
What you can see in the VAS current waveform is the signal that has to be actually fed to the output stage in order for it to mimic the input waveform. That "drive" signal is usually ugly, and a good looking one would be a proof of open-loop linearity. Furthermore, that signal contains all the distortion products inverted (assuming a linear enough input differential stage).
I'm just tempted to amplify it, record it, and listen to it later 😀😀😀
Hi, Janneman,
I read in "John Curl Interview" pdf, he mentioned that he developed a specific measurement device when he can hear something, but measurement devices he has cannot detect it. He modded capacitor measurement device until he can see various distortions from various capacitor types.
One very simple thing. Listening Fatique. What actually causes it? Why listening to tube amps (SET) seldom fatigue like a wrong designed high feedback transistor amp? If all the mechanism is well understood, I think some amplifier company has already release a SET sounding feedback amp (with all SET's distortions) 😀
Hi, EVA,
I've seen those ugly signals at base/gate, but sometimes it is difficult to see. In some points, when I touches the scope probe, the whole thing just oscilates 😀
If the test tone is sinusoidal. Do you think that a more sinusoidal looking at VAS gate indicates that the openloop is having good linearity (while with worse OL linearity, the look will be more ugly?)
I read in "John Curl Interview" pdf, he mentioned that he developed a specific measurement device when he can hear something, but measurement devices he has cannot detect it. He modded capacitor measurement device until he can see various distortions from various capacitor types.
One very simple thing. Listening Fatique. What actually causes it? Why listening to tube amps (SET) seldom fatigue like a wrong designed high feedback transistor amp? If all the mechanism is well understood, I think some amplifier company has already release a SET sounding feedback amp (with all SET's distortions) 😀
Hi, EVA,
I've seen those ugly signals at base/gate, but sometimes it is difficult to see. In some points, when I touches the scope probe, the whole thing just oscilates 😀
If the test tone is sinusoidal. Do you think that a more sinusoidal looking at VAS gate indicates that the openloop is having good linearity (while with worse OL linearity, the look will be more ugly?)
I have been loking at the voltage drop across the current-mirror resistors of my current project with oscilloscope while listening at full volume before clipping. The resistors are 220ohm and the highest deviation from the steady state value that I have observed is 2mV (triangular spike shape). That's .002/220=9uA of drive plus distortion current.
Input LTP has 68 ohm emitter resistors and it's a cross-quad derivative, so its transconductance is almost 1/68.
Those 9uA across 68 ohms make 618uV. This is the maximum peak worst case deviation between the input and the output. Since the closed-loop gain is 31.3, 618uV in the input represent 20mV at the output.
Peak output voltage swing during the test was +-20V approx. so the worst case "peak THD" figure would be 0.1%. However, those 20mV of deviation were mostly high frequency crossover spikes due to the nature of my circuit. This means that the "average THD" measured or simulated is going to be much smaller.
lumanauw:
Indeed, but a low open-loop gain will mask the uglyness because the fundamental signals may dominate in these circumstances. PSpice tells that my circuit has 125dB open-loop gain up to 1Khz, that's why I can see all the distortion products, I think.
A tip to avoid the thing to oscillate when touched with the probe is to use a 10x probe that will have smaller input capacitance. Anoter tip is to use a common-mode filter in the probe wire.
Input LTP has 68 ohm emitter resistors and it's a cross-quad derivative, so its transconductance is almost 1/68.
Those 9uA across 68 ohms make 618uV. This is the maximum peak worst case deviation between the input and the output. Since the closed-loop gain is 31.3, 618uV in the input represent 20mV at the output.
Peak output voltage swing during the test was +-20V approx. so the worst case "peak THD" figure would be 0.1%. However, those 20mV of deviation were mostly high frequency crossover spikes due to the nature of my circuit. This means that the "average THD" measured or simulated is going to be much smaller.
lumanauw:
Indeed, but a low open-loop gain will mask the uglyness because the fundamental signals may dominate in these circumstances. PSpice tells that my circuit has 125dB open-loop gain up to 1Khz, that's why I can see all the distortion products, I think.
A tip to avoid the thing to oscillate when touched with the probe is to use a 10x probe that will have smaller input capacitance. Anoter tip is to use a common-mode filter in the probe wire.
Eva said:
But Eva, in fact those 618uV represent a full scale input signal for the output because of the open loop gain. That 618uV is amplified by the full open loop gain (125dB @ 10kHz!).
Jan Didden
lumanauw said:
Well, I have a great respect for Mr Curl as designer, but when I read posts like "Barrie Gilbert said so and he wouldn't if it weren't true", I really don't know what to do with it, so I gave up on that.
Listening fatique? I don't know, I've experienced it very seldom. Sorry, no data.
Jan Didden
janneman said:
Typically by excessive high order harmonics AFAIK.
Tubes tend to create low order harmonics only, bjts the full range of an exp-function.
Mike
Greetings from Norfolk
I have seldom experienced it, but when I have it tends to be from amplifiers with serious cross-over distortion, which is accentuated at low listening levels - commercial spec of distortion very good, but measured and specified at full output, thus masking poor performance at low signal levels, even with nfb - poor design.
Richard
janneman said:
I have seldom experienced it, but when I have it tends to be from amplifiers with serious cross-over distortion, which is accentuated at low listening levels - commercial spec of distortion very good, but measured and specified at full output, thus masking poor performance at low signal levels, even with nfb - poor design.
Richard
Further random thoughts
Some years back, while sipping martinis at the beachfront, Ed. Cherry opened the Op Amp hood to expose monsters happily lurking inside (Are Operational Amplifiers really linear?). Monsters like amplitude to phase nonlinearity induced modulation, actual phase relationships and so forth.
For the unwary, this may have come as a sort of shock. How comes? Were we happily designing and working with ideal gain blocks and just now come to realize something was terribly wrong? Specs are lying? What seems to work fine is just a mirage?
Not so. Analogy sometimes can help shed light on confusing things, and with your benevolence I propose this.
Assume you have unlimited financial resources, and you are confronted with only two business choices, no other but these two.
You are offered to invest 1000 in a project, and are 100% guaranteed to recover your 1000 and a net profit of 5000.
Alternatively, you are offered to invest 1:000,000, and also 100% guaranteed to recover the investment, only now the net profit is 6000 instead.
While the first alternative makes sense as good business, the second seems ludicrous and inelegant on all fronts.
Yet we must agree the second is the natural choice. Remember there are no other alternatives, and you have unlimited resources.
In a sense, this is what happens with audio topologies, particularly the no or little feedback to gross GNF choices. It is true as Cherry exposed, our cherished signal is beaten to almost death within the Op Amp sordid inner dungeons. It is true in the process, all sort of spurious byproducts are generated. Yet in the end, given enough gain, everything not intended to be reproduced, i.e. anything not already present in the input signal, is forcefully beaten in turn by an amount equal to the OL to system gain. And this includes not only harmonic, IM, PM, TIM, PIM or whatever you can recall, but also noise sources including power supply disturbances leaking into the internal signal path.
What separates the well understood and well executed GNF designs from the mediocre ones that probably with good reason have made a bad name of it, is the assurance of an adequate distortion attenuation margin, available throughout the entire operating frequency range, and dimensioned so as to crush the OL inherent nonlinearity generated artifacts below the target specification. This must hold under all load conditions, something frequently overlooked yet essential given the sometimes nasty combinations of frequency -reactance to be found in the real world.
Rodolfo
Some years back, while sipping martinis at the beachfront, Ed. Cherry opened the Op Amp hood to expose monsters happily lurking inside (Are Operational Amplifiers really linear?). Monsters like amplitude to phase nonlinearity induced modulation, actual phase relationships and so forth.
For the unwary, this may have come as a sort of shock. How comes? Were we happily designing and working with ideal gain blocks and just now come to realize something was terribly wrong? Specs are lying? What seems to work fine is just a mirage?
Not so. Analogy sometimes can help shed light on confusing things, and with your benevolence I propose this.
Assume you have unlimited financial resources, and you are confronted with only two business choices, no other but these two.
You are offered to invest 1000 in a project, and are 100% guaranteed to recover your 1000 and a net profit of 5000.
Alternatively, you are offered to invest 1:000,000, and also 100% guaranteed to recover the investment, only now the net profit is 6000 instead.
While the first alternative makes sense as good business, the second seems ludicrous and inelegant on all fronts.
Yet we must agree the second is the natural choice. Remember there are no other alternatives, and you have unlimited resources.
In a sense, this is what happens with audio topologies, particularly the no or little feedback to gross GNF choices. It is true as Cherry exposed, our cherished signal is beaten to almost death within the Op Amp sordid inner dungeons. It is true in the process, all sort of spurious byproducts are generated. Yet in the end, given enough gain, everything not intended to be reproduced, i.e. anything not already present in the input signal, is forcefully beaten in turn by an amount equal to the OL to system gain. And this includes not only harmonic, IM, PM, TIM, PIM or whatever you can recall, but also noise sources including power supply disturbances leaking into the internal signal path.
What separates the well understood and well executed GNF designs from the mediocre ones that probably with good reason have made a bad name of it, is the assurance of an adequate distortion attenuation margin, available throughout the entire operating frequency range, and dimensioned so as to crush the OL inherent nonlinearity generated artifacts below the target specification. This must hold under all load conditions, something frequently overlooked yet essential given the sometimes nasty combinations of frequency -reactance to be found in the real world.
Rodolfo
- Status
- Not open for further replies.