mandat said:
Regarding the switching PSU unit, this is a question of engineering design, shielding, filtering. It is possible to achiveve the results provided the work is done by competent designer (I love these words that I learnt yesterday from John Curl). The only objection I would have was the maximum peak current delivered by this kind of PSU, probably that's why Halcro speak about that current limitation at 10A - they explain that they want to prevent speaker damage, which I do not suppose to be the case.
Doing a Google search, I found one article that gives an overview of the problem: It was written for 'Electronic Design' back in 1998. The link is:
www.elecdesign.com/Articles/ArticleID/7207/7207.html
I hope I got it transcribed right, however this short article references the earlier work by Otala and Barrie Gilbert.
Otala's main contribution is: "Feedback Generated Phase Modulation in Audio Amplifiers" 65th Convention AES 1980, London. Preprint #1576
This is a tough read, and not for the math challenged.
www.elecdesign.com/Articles/ArticleID/7207/7207.html
I hope I got it transcribed right, however this short article references the earlier work by Otala and Barrie Gilbert.
Otala's main contribution is: "Feedback Generated Phase Modulation in Audio Amplifiers" 65th Convention AES 1980, London. Preprint #1576
This is a tough read, and not for the math challenged.
Yes, they seem to have removed access to that Barrie Gilbert article. That specific article is important, because it shows that Matti Otala was on the right track in 1980. Actually, most of you here can get the necessary concept of FM distortion from Walt Jung's article. The 'proof' is fairly math intensive. If you cannot appreciate what is said by Walt's article, then there is little to further say in order to explain why we use as little negative feedback as possible, in many modern designs.
On the other hand, the linked article deals only with primitive structures, so the distortion is horrible. In case of local linearization one closes feedback loop across very linear circuitry and the distortion compounds can be of order -120dB and better. The question is whether it is audible. Every of those simple amps emphasized here has lines in distortion spectrum far higher than -120dB even for high order harmonics, that are considered to be most harmful.
Nobody argues that feedback brings new distortion components, this is apparent that yes. But the magnitude should be always mentioned.
Nobody argues that feedback brings new distortion components, this is apparent that yes. But the magnitude should be always mentioned.
john curl said:
john curl said:
Actually i read Gilbert and Otala's work refered to here quite sometime ago.....The later merely produced a quantitative analysis of baxandal's findings...
viz:feedback does infact increase distortion....but he neglected to add that such an increase is observed if insufficient feedback is applied....in other words, with enough global feedback, and an inherently linear amplifier, distortion of any complexion will be negligible....
I find it rather amusing that Walt's article is aluded to in this context.....This gives prominence to high open-loop bandwidth as a desirable quantity.... Infact what is desirable is a high foward path gain across as much of the audio band as possible, without compromising stabilty.......viz: maximise feedback factor across as much of the audio band as possible....
Amplifier design is vastly more straightforward than some would have us believe.....
Of greater significance perhaps is the fact that there is no 'new' distortion out there that cannot be detected by THD+N......
Thanks Pavel for your great common sense and insight....
Phase distortion
The Walt Jung and Barrie Gilbert (I presume, since I can't access the article) articles speak of the limitations of feedback bandwidth to prevent phase distortion due to nonlinearities within the feedback loop. And most Op-amps would be used in inverting mode to process audio, so as to include the input diff. amp. stage within the loop.
But most audio power amplifiers do NOT operate in inverting mode. Any nonlinearity between the input transistors including the long tail, or current source, are not corrected. Variable capacitance in the current source collector junction especially would cause phase distortion.
A high level low frequency could cause dynamic phase modulation of a higher frequency. Unless, really gross, this would NOT be detectable by normal THD or even IMD tests. For example, looking at the spectrum generated, the higher frequency would only appear to have a slightly widened peak. The usual distortion tests use notch filters to separate the distortion products from the intended signals, and cannot detect this nearby distortion. The ear might well, however, detect this phase or FM modulation. Possibly explaining some listener preferences for a resistive long tail rather than an active current source. By the way, I tried modest phase modulating a synthetic instrument waveform from a digital waveform storage sound card and several people were able to hear this.
A method of detecting this phase or FM distortion would be to apply a large level, low frequency signal plus a low level, high frequency signal to an amplifier. The amplifier output would then be high pass filtered to remove the low frequency component. The remaining high frequency component would then be mixed, via a multiplier chip, with the original high frequency signal source. The output of the mixer/multiplier would then be low pass filtered to see if any phase distortion product remained (the difference frequency). (Alternatively, one could just sum the two high frequency signals, with matched amplitude, and observe the result on a scope to look for low freq. enveloping due to phase shifting, but would not be as sensitive a detection method for slight phase shifts. )
Don
The Walt Jung and Barrie Gilbert (I presume, since I can't access the article) articles speak of the limitations of feedback bandwidth to prevent phase distortion due to nonlinearities within the feedback loop. And most Op-amps would be used in inverting mode to process audio, so as to include the input diff. amp. stage within the loop.
But most audio power amplifiers do NOT operate in inverting mode. Any nonlinearity between the input transistors including the long tail, or current source, are not corrected. Variable capacitance in the current source collector junction especially would cause phase distortion.
A high level low frequency could cause dynamic phase modulation of a higher frequency. Unless, really gross, this would NOT be detectable by normal THD or even IMD tests. For example, looking at the spectrum generated, the higher frequency would only appear to have a slightly widened peak. The usual distortion tests use notch filters to separate the distortion products from the intended signals, and cannot detect this nearby distortion. The ear might well, however, detect this phase or FM modulation. Possibly explaining some listener preferences for a resistive long tail rather than an active current source. By the way, I tried modest phase modulating a synthetic instrument waveform from a digital waveform storage sound card and several people were able to hear this.
A method of detecting this phase or FM distortion would be to apply a large level, low frequency signal plus a low level, high frequency signal to an amplifier. The amplifier output would then be high pass filtered to remove the low frequency component. The remaining high frequency component would then be mixed, via a multiplier chip, with the original high frequency signal source. The output of the mixer/multiplier would then be low pass filtered to see if any phase distortion product remained (the difference frequency). (Alternatively, one could just sum the two high frequency signals, with matched amplitude, and observe the result on a scope to look for low freq. enveloping due to phase shifting, but would not be as sensitive a detection method for slight phase shifts. )
Don
Parenthetical to this discussion, the late Oscar Heil was a
firm believer that human ears are extremely sensitive to
these very small modulations, and he attributed much of
the difference in the finest stringed instruments to this.
We should also note that the distinction between phase
modulation and amplitude modulation might be more subtle
than you think. They have some interesting similarities in a
spectral plot.
firm believer that human ears are extremely sensitive to
these very small modulations, and he attributed much of
the difference in the finest stringed instruments to this.
We should also note that the distinction between phase
modulation and amplitude modulation might be more subtle
than you think. They have some interesting similarities in a
spectral plot.
Hi Nelson,
I can well imagine that a stringed instrument might allow a large amplitude string vibration to modulate the tension of the string and hence the frequency of other notes, and wood case stiffness would affect this too. If someone can hear this effect, then it might be reasonable to look for a similar sound degradation in a power amplifier. Of course, it all depends on the magnitude of the effect, so some measurements would be illuminating as to whether this is a real concern in an amplifier. As you point out, for large phase modulation or FM, the spectra spreads out like for AM, and would be detectable by THD tests. But I think the likely magnitude of phase modulation in a power amplifier is fairly small, so would only show up as spectral line widening, and so would avoid detection by normal THD/IMD tests. One would also like to have some measurements as to what level is detectable by the listener and subjective opinions about sound quality.
One last note on the alternate detection scheme mentioned above, using the sum of the two high frequencies and observing enveloping on a scope. The sensitivity of this scheme could be improved considerably by using a 90 degree phase shifted signal from the HF signal source to sum with the high pass filtered amplifier output (at equal amplitudes). This way, any small phase shift immediately causes envelope level changes, much as the old FM phase discriminators worked. ( I think this 90 degree mod. would also improve the mixer/multiplier method sensitivity too.)
Don
I can well imagine that a stringed instrument might allow a large amplitude string vibration to modulate the tension of the string and hence the frequency of other notes, and wood case stiffness would affect this too. If someone can hear this effect, then it might be reasonable to look for a similar sound degradation in a power amplifier. Of course, it all depends on the magnitude of the effect, so some measurements would be illuminating as to whether this is a real concern in an amplifier. As you point out, for large phase modulation or FM, the spectra spreads out like for AM, and would be detectable by THD tests. But I think the likely magnitude of phase modulation in a power amplifier is fairly small, so would only show up as spectral line widening, and so would avoid detection by normal THD/IMD tests. One would also like to have some measurements as to what level is detectable by the listener and subjective opinions about sound quality.
One last note on the alternate detection scheme mentioned above, using the sum of the two high frequencies and observing enveloping on a scope. The sensitivity of this scheme could be improved considerably by using a 90 degree phase shifted signal from the HF signal source to sum with the high pass filtered amplifier output (at equal amplitudes). This way, any small phase shift immediately causes envelope level changes, much as the old FM phase discriminators worked. ( I think this 90 degree mod. would also improve the mixer/multiplier method sensitivity too.)
Don
Apologies in advance for the verbose post that follows here 
. I finally took some time to read through the article www.elecdesign.com/Articles/ArticleID/7207/7207.html, and I disagree with some of the points made. His points include the following:
1) The gain-bandwidth product of a conventional voltage feedback op-amp or power amp in rad/sec is gm/C.
2) When the differential input voltage to the input diff amp is not negligible, the instantaneous differential input voltage swing, combined with the diff amp nonlinearity causes gm to be modulated.
3) This modulation of gm causes the gain-bandwidth product to be modulated.
4) Modulation of the gain-bandwidth product causes instantaneous phase modulation, that is, AM-to-PM conversion in RF speak.
5) The performance can be improved for a fixed gain-bandwidth product by decreasing the open-loop gain and increasing the open-loop bandwidth. This is best done by loading the VAS with resistors to ground.
By "modulaton", I mean "varies with the instantaneous value of the difference-mode input signal". I'm describing his points here, not necessarily mine. I agree with 1-4 and strongly disagree with 5.
Let's set aside for the moment whether these effects are audible and just look at the mechanism. Let me try to demonstrate why I disagree with 5. To do this, let's assume we have two hypothetical amplifiers that we can change, subject to the following restrictions:
A) The input diff amp can't be changed
B) The compensation cap can't be changed
C) All we can do is change the open-loop gain and bandwidth while keeping the gain-bandwith product constant.
At a fixed output frequency and amplitude, it's clear that the amount of phase modulation is determined in a direct way by the amount of modulation of gm. Since the phase is being modulated at the same rate (frequency) as the signal itself, the modulation sidebands just show up as normal harmonic distortion. But let's forget about harmonic distortion for the moment and just compare how much gm is being modulated in the two hypothetical amplifiers. Further, let's assume a sinusoidal output voltage whose amplitude is independent of frequency.
Let's take an amplifier with a 20 MHz gain-bandwidth product. Assume its open-loop bandwidth is 2 kHz and its DC open-loop gain is 10000 (80 dB). Suppose we decide to modify this amp so that its open-loop bandwidth is wider, say 20 kHz and its gain-bandwidth product is the same. The DC open-loop gain must be 1000 now, or 60 dB. Further, assume the modifications meet the requirements of A, B and C above, maybe by putting collector resistors to ground at the VAS output as the author suggests.
Let's assume a large 50 kHz signal at the output of both amplifiers, and assume the slew rate is high enough so there's no gross distortion. Since the gain-bandwidth products are the same, the Bode plots for both amplifiers fall right on top of each other at frequencies above 20 kHz. So the two amplifiers will have almost exactly the same open-loop gain at 50 kHz. If we take the output signal amplitude and divide it by the open-loop gain, we get a difference-mode input signal that's the same for both amplifiers. Since the input stages are the same for both amplifiers, and the differential input voltage swings are the same, the instantaneous modulation of gm is the same for the two and therefore the AM-to-PM conversion is also the same.
Now take the same amplitude output signal at 1 kHz. We now have 20 dB more open-loop gain in the amplifier with the narrower open-loop bandwidth than the one with the wider open-loop bandwidth. Dividing the output signal amplitude by the open-loop gain, we get a difference-mode input signal that's 20 dB lower in the amp with the lower open-loop bandwidth. So at lower frequencies, there's actually much less modulation of gm in the amplifier with the lower open-loop bandwidth due to the decreased differential input voltage swing.
So the feedback is not the villain here! At frequencies above about 20 kHz, the performance with regard to the phase modulation is the same for both amplifiers. But at frequencies below 20 kHz, the phase modulation is less in the amplifier with the lower open-loop bandwidth and higher open-loop gain (because we're assuming a fixed level output signal, and the open-loop gain is larger, and therefore the difference-mode input is smaller). While it's true that this phase modulation will no longer be independent of frequency from low frequencies to 20 kHz, it is always better in the amp with the lower open-loop bandwidth than the one with the higher open-loop bandwidth (remember we're assuming the gain-bandwidth products are the same between the two). Not only is the feedback not the villain, but it's improving matters at the lower frequencies.
So the idea of putting collector resistors to ground at the VAS output only degrades the theoretical performance with respect to phase modulation below 20 kHz, and keeps it about the same above 20 kHz. The original Leach amplifier in the Audio magazine article in February 1976 http://users.ece.gatech.edu/~mleach/papers/lowtim/feb76feb77articles.pdf had these resistors (R20 and R21 in Figure 2). However, later versions http://users.ece.gatech.edu/~mleach/lowtim/graphics/ckt.pdf omit them. This issue was the subject of debate back in the late '70s and Dr. Leach changed the design. He specifically mentioned at that time that what was really important is not the open-loop bandwidth, but the gain-bandwidth product. Removal of the resistors brought about improved conventional distortion performance without sacrificing transient capability.
While the phase modulation concept has a valid theoretical basis, for two otherwise identical amplifiers with the same gain-bandwidth product, the one with the larger low-frequency open-loop gain and smaller open-loop bandwidth actually has an advantage with respect to theoretical performance in this area. It's a mistake to blame feedback for this.
. I finally took some time to read through the article www.elecdesign.com/Articles/ArticleID/7207/7207.html, and I disagree with some of the points made. His points include the following:1) The gain-bandwidth product of a conventional voltage feedback op-amp or power amp in rad/sec is gm/C.
2) When the differential input voltage to the input diff amp is not negligible, the instantaneous differential input voltage swing, combined with the diff amp nonlinearity causes gm to be modulated.
3) This modulation of gm causes the gain-bandwidth product to be modulated.
4) Modulation of the gain-bandwidth product causes instantaneous phase modulation, that is, AM-to-PM conversion in RF speak.
5) The performance can be improved for a fixed gain-bandwidth product by decreasing the open-loop gain and increasing the open-loop bandwidth. This is best done by loading the VAS with resistors to ground.
By "modulaton", I mean "varies with the instantaneous value of the difference-mode input signal". I'm describing his points here, not necessarily mine. I agree with 1-4 and strongly disagree with 5.
Let's set aside for the moment whether these effects are audible and just look at the mechanism. Let me try to demonstrate why I disagree with 5. To do this, let's assume we have two hypothetical amplifiers that we can change, subject to the following restrictions:
A) The input diff amp can't be changed
B) The compensation cap can't be changed
C) All we can do is change the open-loop gain and bandwidth while keeping the gain-bandwith product constant.
At a fixed output frequency and amplitude, it's clear that the amount of phase modulation is determined in a direct way by the amount of modulation of gm. Since the phase is being modulated at the same rate (frequency) as the signal itself, the modulation sidebands just show up as normal harmonic distortion. But let's forget about harmonic distortion for the moment and just compare how much gm is being modulated in the two hypothetical amplifiers. Further, let's assume a sinusoidal output voltage whose amplitude is independent of frequency.
Let's take an amplifier with a 20 MHz gain-bandwidth product. Assume its open-loop bandwidth is 2 kHz and its DC open-loop gain is 10000 (80 dB). Suppose we decide to modify this amp so that its open-loop bandwidth is wider, say 20 kHz and its gain-bandwidth product is the same. The DC open-loop gain must be 1000 now, or 60 dB. Further, assume the modifications meet the requirements of A, B and C above, maybe by putting collector resistors to ground at the VAS output as the author suggests.
Let's assume a large 50 kHz signal at the output of both amplifiers, and assume the slew rate is high enough so there's no gross distortion. Since the gain-bandwidth products are the same, the Bode plots for both amplifiers fall right on top of each other at frequencies above 20 kHz. So the two amplifiers will have almost exactly the same open-loop gain at 50 kHz. If we take the output signal amplitude and divide it by the open-loop gain, we get a difference-mode input signal that's the same for both amplifiers. Since the input stages are the same for both amplifiers, and the differential input voltage swings are the same, the instantaneous modulation of gm is the same for the two and therefore the AM-to-PM conversion is also the same.
Now take the same amplitude output signal at 1 kHz. We now have 20 dB more open-loop gain in the amplifier with the narrower open-loop bandwidth than the one with the wider open-loop bandwidth. Dividing the output signal amplitude by the open-loop gain, we get a difference-mode input signal that's 20 dB lower in the amp with the lower open-loop bandwidth. So at lower frequencies, there's actually much less modulation of gm in the amplifier with the lower open-loop bandwidth due to the decreased differential input voltage swing.
So the feedback is not the villain here! At frequencies above about 20 kHz, the performance with regard to the phase modulation is the same for both amplifiers. But at frequencies below 20 kHz, the phase modulation is less in the amplifier with the lower open-loop bandwidth and higher open-loop gain (because we're assuming a fixed level output signal, and the open-loop gain is larger, and therefore the difference-mode input is smaller). While it's true that this phase modulation will no longer be independent of frequency from low frequencies to 20 kHz, it is always better in the amp with the lower open-loop bandwidth than the one with the higher open-loop bandwidth (remember we're assuming the gain-bandwidth products are the same between the two). Not only is the feedback not the villain, but it's improving matters at the lower frequencies.
So the idea of putting collector resistors to ground at the VAS output only degrades the theoretical performance with respect to phase modulation below 20 kHz, and keeps it about the same above 20 kHz. The original Leach amplifier in the Audio magazine article in February 1976 http://users.ece.gatech.edu/~mleach/papers/lowtim/feb76feb77articles.pdf had these resistors (R20 and R21 in Figure 2). However, later versions http://users.ece.gatech.edu/~mleach/lowtim/graphics/ckt.pdf omit them. This issue was the subject of debate back in the late '70s and Dr. Leach changed the design. He specifically mentioned at that time that what was really important is not the open-loop bandwidth, but the gain-bandwidth product. Removal of the resistors brought about improved conventional distortion performance without sacrificing transient capability.
While the phase modulation concept has a valid theoretical basis, for two otherwise identical amplifiers with the same gain-bandwidth product, the one with the larger low-frequency open-loop gain and smaller open-loop bandwidth actually has an advantage with respect to theoretical performance in this area. It's a mistake to blame feedback for this.
rough estimate of frequency deviation
I was just curious to see how much frequency deviation/distortion one might expect from a non-inverting input amplifier with a bipolar current source on the diff. amp. stage. So here goes a rough calculation. (please correct me if I am wrong!)
First, phase modulation delta f is proportional to freq. of modulation f sub m (will use 10,000 Hz here) times the phase modulation factor Mp ( degrees/360 per volt units). So worst case will occur for a maximum input signal at high frequency.
So assuming 4 volts peak to peak common mode input for a maximum output. Using the Moto. data sheet for a BC550 npn transistor as current source, with 6V to 10V collector voltage variation, gives about 2.1pf to 1.8pf variation of collector output capacitance, or 0.3pf variation. Using 10K Ohms as input impedance in series with about 1650 Ohms emitter resistance (beta/gm) forms an RC network. The 0.3pf cap has about 5*10 to the 7th Ohms reactance at 10,000 Hz. So, arctan of 11650/5*10 to 7th gives about 1.335 * 10 to -2 degrees or 1.335/360 * 10 to -2 of a cycle. for Mp. Then Mp times 10,000 Hz gives delta f of 3.7 Hz.
This 3.7 Hz deviation from the original signal is pretty small, I doubt whether this could be avoided by any notch carrier elimination filter in a THD or IMD tester at 10KHz, so won't show up in THD or IMD testing. Question is, can anyone hear this level of distortion?
Did I miss anything in the calcs?
Don
I was just curious to see how much frequency deviation/distortion one might expect from a non-inverting input amplifier with a bipolar current source on the diff. amp. stage. So here goes a rough calculation. (please correct me if I am wrong!)
First, phase modulation delta f is proportional to freq. of modulation f sub m (will use 10,000 Hz here) times the phase modulation factor Mp ( degrees/360 per volt units). So worst case will occur for a maximum input signal at high frequency.
So assuming 4 volts peak to peak common mode input for a maximum output. Using the Moto. data sheet for a BC550 npn transistor as current source, with 6V to 10V collector voltage variation, gives about 2.1pf to 1.8pf variation of collector output capacitance, or 0.3pf variation. Using 10K Ohms as input impedance in series with about 1650 Ohms emitter resistance (beta/gm) forms an RC network. The 0.3pf cap has about 5*10 to the 7th Ohms reactance at 10,000 Hz. So, arctan of 11650/5*10 to 7th gives about 1.335 * 10 to -2 degrees or 1.335/360 * 10 to -2 of a cycle. for Mp. Then Mp times 10,000 Hz gives delta f of 3.7 Hz.
This 3.7 Hz deviation from the original signal is pretty small, I doubt whether this could be avoided by any notch carrier elimination filter in a THD or IMD tester at 10KHz, so won't show up in THD or IMD testing. Question is, can anyone hear this level of distortion?
Did I miss anything in the calcs?
Don
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.