john curl said:
Hi John,
Let me try and explain PIM so that you can understand that it is not dependent on open-loop bandwidth. PIM is phase modulation of the signal induced by the amplitude variations of the program signal, and it is dependent on dynamic movement of the closed-loop pole of the power amplifier.
Consider an amplifier with 20 dB of NFB and an open loop bandwidth of 20 kHz with a first-order rolloff. It will have a closed-loop bandwidth of about 200 kHz. The pole at 200 kHz will result in a phase shift of about 4.4 degrees at 20 kHz. Assume that the amplitude of the program material has caused a compression of 10% in the incremental open loop gain. This will cause the closed loop bandwidth to decrease by 10% to about 180 kHz. The closed-loop pole has thus moved inward by 10%, increasing the phase shift at 20 kHz to about 4.8 degrees. There is thus about 0.4 degrees of PIM at 20 kHz, corresponding to about 56 ns of effective time modulation. Note also that there will be about 1% of ordinary distortion at 20 kHz, the 10% open loop distortion having been reduced by the 20 dB of NFB. This amplifier has quite a bit of ordinary distortion.
Now consider an amplifier with 40 dB of NFB and an open loop bandwidth of 2 kHz. It will also have a closed loop bandwidth of about 200 kHz. Again assuming that the program material causes a 10% compression of the open loop gain, the closed loop pole will move in by about 10% to 180 kHz. We can see immediately that the situation is essentially identical insofar as phase modulation goes. It is numerically about the same as before. This shows that NFB-induced PIM is not dependent on the open loop bandwidth. Notice in this case there will also still be considerable ordinary distortion at 20 kHz, on the order of 1%, since the open loop distortion has been reduced by the 20 dB of NFB available at 20 kHz.
PIM is dependent on the closed-loop bandwidth of the amplifier. Imagine increasing the open-loop gain of both of the amplifiers above by 20 dB while keeping the open loop pole at the same frequency. This would push the closed-loop pole out to about 2 MHz. It is now easy to see that, with the closed-loop pole 10 times further removed from the audio band, its movement inward by 10% will have far less effect on the phase at 20 kHz. Thus, we see a situation where increasing the amount of negative feedback (to 40 dB at 20 kHz in this case) decreases the amount of PIM, all else remaining equal.
A deeper discussion of PIM can be found on my web site in the paper “Phase Intermodulation Distortion – Instrumentation and measurements” under the Published Papers section. The next-to-last paragraph on page 19 is especially illuminating, as it discusses the measured PIM in an experimental amplifier with and without negative feedback. In that case, introducing NFB REDUCED PIM by a factor of 3-6 times.
Cheers,
Bob
Maybe this can explain why no feedback amp or DartZeel considered good
http://hyperphysics.phy-astr.gsu.edu/hbase/sound/hearcon.html#c1
interesting articles : organ of corti, hair cells, place theory, sharpening
http://hyperphysics.phy-astr.gsu.edu/hbase/sound/hearcon.html#c1
interesting articles : organ of corti, hair cells, place theory, sharpening
lumanauw said:
I did not find that particularly enlightening. I recommend
The Psychology of Music, Second Edition, Diana Deutsch.
It is quite clear that hearing is far more complex and subtle than
the approaches used to measure amplifiers.
The Dartzeel seems to sound different. Sometimes that alone
is perceived as better or the short term - I remember when the
Phase Linear 700 ushered in a new era.
Nelson, wasn't it Diana Deutsch who wrote the article about perception of hearing from a right brain-left brain aspect. It was stated that orchestra positions were laid out so that most people (right handers/left brains) heard the violins more diffusely. In other words, "imaging" was purposely nipped in the bud!
Bob Cordell said:
I disagree with this. I'm thinking only about feedback-generated PIM here, and not the intrinsic PIM of the amplifier. These are my reasons for disagreement.
At the time you wrote your article on PIM, I don't think Gilbert's article had been written yet, or at least not widely discussed. It's my view that Barrie's article puts into clear mathematical form what Otala was originally thinking of. It's complex enough to show the problem, yet simple enough to still be mathematically tractable. He's assuming a nonlinear input stage, with the rest of the amplifier assumed to be a distortionless integrator. I'll try to describe his approach as concisely as possible.
1) Assume an undistorted sinusoidal output.
2) Compute the voltage fed back to the input (easy, it's just B*vout)
3) Work backwards from the output to the differential input. First get the undistorted diff amp output current. Then, get the distorted difference mode input voltage with the help of the Taylor series expansion for the inverse tanh(). Retain only the linear and cubed terms of this expansion.
4) Using a KVL equation around the input loop, compute what the amp's input voltage must have been to get that output. This is just the phasor sum of the voltages found in (2) and (3) above.
One thing that happens here is that amplitude compression of the open-loop amp affects the results. Mathematically, the trig identity
cos3(x) = 3/4 cos(x) + 1/4 cos(3x)
gives the gain compression term, therefore a nonlinear relationship between the amplitude of the difference mode input voltage and that of the output.
Because the open-loop amplifier is subject to gain compression, the ratio of the amplitudes of the signals computed in steps (2) and (3) above is not constant as the amplitude of the output signal varies. Because of this, and the fact that the signals computed in (2) and (3) above are in quadrature phase, the phase angle of the phasor sum of these (the overall input voltage) is not constant as the output signal amplitude varies. From my POV, this is the root cause of feedback-generated PIM.
Now what happens to this phasor sum if we assume the difference mode input voltage is in phase with the output, as it would be if the amp had a very wide open-loop bandwidth? The phase angle of this phasor sum will be zero, independent of the output voltage amplitude. Therefore, the feedback-generated PIM will be zero. I've attempted to generalize Barrie's results to arbitrary input-output phase shifts on my web pages here.
Of course, there is a heavy cost of doing this. To get the wide open-loop bandwidth requires big reductions in open-loop gain. This will in turn increase AIM and handicap the amp's ability to reduce the intrinsic PIM of the amplifier (not taken into account in Barrie's analysis). So it's like straining out the gnats and letting the elephants through IMO.
The other thing I want to mention is potential interfering variables. TIM and PIM are both analyzed purely as difference-mode phenomena. What about common-mode effects?
It's pretty simple to do a SPICE analysis that makes similar assumptions to what Barrie did. Only instead of an abstract nonlinear diff amp, use transistor models and a current mirror. Make a distortionless VAS with Miller compensation so the correct loading of the input diff amp is established, and use a unity-gain VCVS as the output stage. Then look at the output THD. A second simulation can be created, the same as the first, except using a bootstrapped cascode in the input stage to cancel out common-mode effects. What I've seen, depending on the gain-bandwidth product and the output amplitude, is reductions in simulated THD20k by 10x to over 100x. This strongly suggests that the contribution of the input stage distortion to that of the overall amp is dominated by common-mode effects - not TIM or PIM - if no bootstrapping is used in the input stage.
When I first noticed this, I kind of smacked myself on the forehead. I reached the conclusion that the whole TIM/PIM brouhaha was, again, straining the gnats out and letting the elephants through. Since the common-mode effects are an interfering variable, it signals the need for caution when trying to infer the theoretical effects of feedback-generated PIM from measured data. An example here is when Walt did his TAA articles which established the correlation between TIM and THD at high frequencies and large output amplitudes. One thing of note is that he used inverting mode to eliminate the effects of common-mode input distortion on his results.
I can't seem to find the SPICE sims I have of this at the moment. I know I posted them in a thread where Glen was talking about a unity-gain buffer he was designing - but I can't remember what thread that was.
Fellow engineers, I think that we should look at possible distortions in a REAL op amp, rather than the exotic products put forth here. How about the RC4558? It is the audio 'standard' for our industry. It is made by several manufacturs, including TI, and it costs only 27 cents per/100. This is about 20 times cheaper than the AD797, that I must admit, looks a bit better on the spec sheet, even with the same input stage. But then, think of the price savings, if you don't need low noise, especially.
I'm sure that Scott's 797 will sound and measure somewhat better, but in what way, and why should it be important?
I'm sure that Scott's 797 will sound and measure somewhat better, but in what way, and why should it be important?
Not purposely. What sort of imaging does she have in mind? Right ear is usually dominant but left ear works as well.
It's laid out so quieter instruments are in the front, louder in back.
1st violins are placed so the loudest directional component of its sound go out to audience. Violas and 2nd violins have to play louder to balance.
Generally orchestra layout is done so conductor gets good sound balance bearing in mind needs of the audience.
Orchestral seating arrangements vary depending on what sort of music is performed.
andy_c said:
Hi Andy,
You've said quite a bit here, and at a higher technical level than I was using to explain PIM. I haven't yet fully absorbed what all that you've said, either. Let me just point out a couple of things to make sure there is no confusion.
The explanation I used in my earlier post to demonstrate that PIM is not affected by open loop bandwidth was aimed solely at feedback generated PIM. I'm not sure you thought I was aiming at that.
Although not limited to that, the scenario I discussed was directly relevant to a situation where the nonlinearity is coming from the input differential pair (and no other nonlinearity need be assumed).
There are numerous ways to explain PIM and how it is affected by the application of negative feedback, and the way I described it was deliberately one of the simpler ones, albeit correct in its conclusions. In my paper on my web site, I actually start with Otala's equations and reach the same conclusions.
I indeed agree that in real amplifiers the input stage common mode distortion is a factor that must be considered in addition to all the other sources. I believe that input common mode distortion in my original MOSFET power amplifier with error correction was one of the remaining distortions, as I did not drive the cascode bases with a replica of the signal.
If there is specifically something wrong with the simpler explanation of PIM re open-loop bandwidth that I gave above, please point it out so we can discuss.
In reading your post, it sounds like you are assering that PIM is dependent upon open loop bandwidth when the closed loop bandwidth is held constant. If that is the case, and we are both comparing the same apples to apples, we do have a disagreement that we need to sort out.
Cheers,
Bob
pooge said:
I don't know exactly Diana's assertions (I have been trying to buy her book but it has been out of print for some time) but I disagree to this. The left / right brain thing concerns the general focus of the brain on specific aspects of perception. It does NOT concern the fact that the left ear feeds the left brain and the right ear feeds the right brain, because that's not the case.
The brain produces a 'perceptive sound landscape' that integrates whatever the ears feed into it, plus a lot of other inputs like sight, feeling, body state (how do I feel) and memory inputs. That complex set of inputs is 'weighted' with different weighting factors, and these weighting factors are partly determined by the brain's focus in the left and right hemispheres.
For instance, the body state, which represents how I feel: elated, depressed, happy, sad, whatever, is weighted mainly by the right hemispere and added to the perception equation. As a result, the same objective sound vibration input to your ears can lead to quite different perceptions depending on your body state. A similar process happens to other perceptions, like sight and memories that are activated because they somehow are connected to what happens at this moment.
In the same vein, visual perceptions are tainted by other events, meaning that the objectively same visual input can give rise to different perceptions depending on the other factors.
We humans have an exquisitely fine honed perception apparatus that generates a world view for us that integrates a multitude of events and inputs, and any attempt to explain it in simple terms focussing on just a single sense is doomed to be incomplete at best, way off at worst.
Jan Didden
Bob Cordell said:
Yes, we agree on this.
Okay, we're on the same page here. Since Gilbert also assumes that only an input stage nonlinearity is present, and further that it is purely of the AIM type, this should ideally be an apples-to-apples comparison.
If you think about this some more, consider again what I said in my previous post. If the difference-mode input voltage fundamental is in phase with the output at all signal levels, then the PIM must be zero. This is exactly what happens as the open-loop bandwidth becomes wide. Note that I am not advocating this approach however. I'm assuming purely resistive feedback also.
Yes, this is what I was assuming, although it's not a necessary part of my argument.
I'll re-read your article, as it's been quite a long time since I read it last.
john curl said:
Interesting.
Regarding the 4558, I guess a lot of the recordings in our collections have had the music pass through a lot of these - or worse. Seems kind of silly at times to seek perfection at reproducing that, doesn't it?
Then again, who needs audiophile recordings of lousy music?
>Regarding the 4558, I guess a lot of the recordings in our >collections have had the music pass through a lot of these - or >worse. Seems kind of silly at times to seek perfection at >reproducing that, doesn't it?
No, because with this type of distortion the signal may go through 100 4558s, but the last one in the chain screws with the percieved fidelity as much as the first. I.E. if you literally had the 100 opamps in series and went from output to output with a aural probe, you'd find as big as degradation between #99 and #100 as between #1 and #2. The previous distortion does not mask that ahead of it.
No, because with this type of distortion the signal may go through 100 4558s, but the last one in the chain screws with the percieved fidelity as much as the first. I.E. if you literally had the 100 opamps in series and went from output to output with a aural probe, you'd find as big as degradation between #99 and #100 as between #1 and #2. The previous distortion does not mask that ahead of it.
john curl said:
Hi John,
I don't think anyone is disputing that.
That the 4558 has all of these problems does not make Otala's conclusions about negative feedback and open loop bandwidth correct.
No one is disputing that TIM, PIM and IIM exist; I've measured plenty of each in my papers. You should read them sometime.
I'll say it once again, Otala deserves credit for proposing ways to measure each of the distortions he invented. I've implemented and used his distortion measurement approaches in my papers. It was those measurements, based on Otala's own measurement proposals, that were most effective in proving that his conclusions about the relationship between NFB/open-loop bandwidth and those distortions were wrong. His own measurement proposals sowed the seeds of his own un-doing.
Cheers,
Bob
andy_c said:
Hi Andy,
I think that you may be inadvertantly assuming that closed loop bandwidth goes to infinity as you allow open loop bandwidth large in the limit. I suspect that if the difference mode fundamental is exactly in phase with the output signal, that is tantamount to a closed loop bandwidth of infinity. PIM will indeed be zero if closed loop bandwidth is infinity. Bear in mind that zero PIM does not mean zero lagging phase shift from input to output, but rather the absence of change in that phase when open-loop gain changes.
Cheers,
Bob