scott wurcer said:
Hi Scott,
The distortion I was talking about doesn't stem from the 22pF caps, neither from C124 and R125 (another compensating thingy). The point is that the amp (as published in that patent) is highly unstable. To overcome this serious shortcoming, some kind of additional Miller compensation is definitely needed. Regrettably, such kind of compensation makes the EC mechanism completely ineffective.
For more info, you might start here: http://www.diyaudio.com/forums/showthread.php?postid=1518061#post1518061
Cheers,
Edmond.
john curl said:
John,
You don't "NAIL" anyone with your bluster and mostly unsupported assertions. Nor does your "rebuttal" on the TIM thing. That is obvious to anyone who read that and my response when they were made available for anyone to see.
There is nothing wrong with PMA doing what he is trying to do. As I said, I do not claim that PIM does not exist; just that the generalization that feedback and/or low open loop bandwidth makes it worse.
Indeed, I believe that if I was to measure a very low distortion high-feedback amplifier, your JC-1, and a no-feedback amplifier together with my PIM analyzer, I would find that the PIM was lowest in the high-feedback, low-distortion amplifier and highest in the no-feedback amplifier. I would expect that the JC-1 would lie in the middle. Obviously, that is just a guess on my part.
As good as SPICE is, I agree that it may be difficult to find PIM with confidence in SPICE transient simulations. However, it may be possible to infer PIM from phase shift changes in an AC simulation where the DC operating point is moved around on a DC-coupled amplifier.
Bob
john curl said:
This thread has turned boring, so I read it only if I have excess time,
but I would surely have noticed if you had nailed anybody.
(And I don't mean the connotation this had if you translated it word by word into German).
The habitual name dropping may impress hobbyists (maybe that's your target audience)
but nobody who's in the electronics industry for > 25 years and able to develop ideas of his own.
john curl said:
Spice models are not perfect, neither virtually or practically.
They are as good as the effort behind them. When I designed
some analog functions on AEG B1000 arrays in 5um bipolar back when
a VAX was a nice computer, we had _excellent_ Spice models
and excellent parasitics extraction that we could rely upon.
This cannot be said about most models you can download
from the web. Funny, how all transistor models generated by
the PSPICE model extractor share the same 5? Ohms base
spreading resistance!
Douglas Self wrote that the Signetics 5532/4 contained 3 nested
feedback loops that were required for its performance. Can you
see anything about this in the simplistic model published here?
For bipolars, Spice may be OK, but for LDMOS it's hopeless.
That's why there is a market for Agilent ADS , Genesys or
Microwave Office. OK, they have other merits, too.
john curl said:
While I was was away, has this terminal rebuttal surfaced for anybody to read?
I have asked for it repeatedly to no avail.
john curl said:
I think you suppose that this is the Blowtorch Shrine. After ~20 years,
given your excellence, you should be so far away in the future that it should pose
no problem to publish some _hard_ facts about it. Seldom has such a
vapor theme produced such a large thread.
Regards, Gerhard
chascode said:
AFAIK it's 12bit but I don't think it really matters. The TDS3052 does 5 samples/nS which allows to estimate (after averaging) +/-100pS delays. If there is any measurable PIM effect, it's obviously well below that.
I don't care about the Tek FFT module myself, it's very primitive.
Exactly. See attached plots of closed-loop gain vs DC bias at input (0V, +-1V, +-2V, and +-3V -- 3.2V is the clipping limit) of a simple amp with moderate "semi-global" feedback and common source MOSFET outputs. The increasing output capacitance with lower Vds is most probably the root cause here. Differences from bias polarity, emphasizing P-ch vs N-ch differences can be seen, too.Bob Cordell said:
- Klaus
Attachments
Phase Intermodulation Distortion (PIM) is not really that difficult to understand. PIM is simply the dynamic change in amplifier transfer function phase shift as a function of signal. This is analogous to common SMPTE Amplitude Intermodulation distortion (AIM) where the incremental gain of the amplifier is a function of signal excursions. The test for them is the same. A large low-frequency signal, typically 60 Hz, is used to cause the operating points in the amplifier to traverse a large-signal range. A smaller signal, typically 4 times smaller and at a high frequency like 7 kHz then has its amplitude and phase modulations measured. Modulation on the small signal at 60 Hz or its harmonics is either AIM or PIM depending on whether amplitude or phase detection is used.
Amplifiers without negative feedback have PIM. One common source of PIM in an amplifier without negative feedback is nonlinear junction capacitance that changes the bandwidth, and thus the phase shift, of the amplifier as a function of signal swing. Miller effect in the VAS from collector-base junction capacitance is an example here. Another source can be the changing ft of the output transistors affecting the bandwidth as their ft droops at signal extremes. The main point here is that the PIM is pretty much directly related to modulation of the bandwidth of the amplifier by signal excursions. When the signal modulates circuit parameters in such a way as to reduce the bandwidth, then phase shift through the amplifier increases. Note that a given percentage change in bandwidth will cause a smaller amount of in-band phase shift if the 3-dB point being modulated is at a higher frequency.
The case for feedback amplifiers is very similar. In this case, dynamic changes in the closed-loop bandwidth of the amplifier cause the modulation of the phase transfer function of the amplifier. That leads to PIM. There are a couple of interesting twists, however. First, the application of negative feedback in general tends to reduce the effects of open-loop variations; it tends to stabilize gain and phase and bandwidth in going from the open-loop condition to the closed-loop condition. To this extent, NFB tends to decrease the PIM that would have been naturally present in the open-loop amplifier without feedback.
On the other hand, the application of NFB causes a conversion from open-loop AIM to closed-loop PIM to occur. As the open-loop gain of the amplifier varies with signal excursions, we know that the closed loop bandwidth of a conventionally compensated negative feedback amplifier will change. So at signal extremes, if the open-loop gain decreases, for what ever reason, the closed loop bandwidth will decrease, and more in-band phase shift will result, and this is PIM. This is no different than saying that the closed-loop bandwidth of the amplifier is changing with signal, just as in the case with a no-feedback amplifier.
In either case, it is just a matter of how much the closed-loop bandwidth changes with signal excursions and how far out that pole is. At this point it is important to recognize that, to first order, a 10% reduction in open loop gain due to signal excursions will cause an approximate 10% reduction in closed loop bandwidth whether the amplifier has wide open loop bandwidth or narrow open-loop bandwidth; the rolloff slope of the NFB at high frequencies is approximately the same in either case if both amplifiers have the same nominal gain-crossover frequency. But the higher this gain crossover frequency, the less PIM will be created by a given percentage modulation of the open-loop gan.
Another twist reveals itself when we examine one of the oft-cited sources of open-loop gain modulation that can lead to closed loop phase variation and thus PIM in negative feedback amplifiers. Consider the input differential pair. When it is stressed by large signals, its gain decreases. The larger the stress on the input stage, the greater is the potential source of PIM. Consider what happens when we have two NFB amplifiers with the same gain crossover frequency and nominal closed-loop bandwidth of 1 MHz. The first amplifier has a wide open-loop bandwidth of 20 kHz. It has an open-loop gain of 34 dB from 20 kHz on down to DC. The second amplifier has a narrow open-loop bandwidth of 2 kHz. It has an open-loop gain of 34 dB at 20 kHz that increases to 54 dB at 2 kHz and is constant from there on down to DC. Assume that both amplifiers employ the identical input differential pair stage. The amplifier with higher negative feedback and lower open loop bandwidth will actually stress the input stage far less than the amplifier with wide open loop bandwidth. The error signal applied to the input stage at frequencies below 2 kHz is ten times smaller than in the case of the amplifier with wide open-loop bandwidth. This means that this source of PIM will actually be smaller in the NFB amplifier with higher open-loop gain and lower open loop bandwidth.
Cheers,
Bob
Amplifiers without negative feedback have PIM. One common source of PIM in an amplifier without negative feedback is nonlinear junction capacitance that changes the bandwidth, and thus the phase shift, of the amplifier as a function of signal swing. Miller effect in the VAS from collector-base junction capacitance is an example here. Another source can be the changing ft of the output transistors affecting the bandwidth as their ft droops at signal extremes. The main point here is that the PIM is pretty much directly related to modulation of the bandwidth of the amplifier by signal excursions. When the signal modulates circuit parameters in such a way as to reduce the bandwidth, then phase shift through the amplifier increases. Note that a given percentage change in bandwidth will cause a smaller amount of in-band phase shift if the 3-dB point being modulated is at a higher frequency.
The case for feedback amplifiers is very similar. In this case, dynamic changes in the closed-loop bandwidth of the amplifier cause the modulation of the phase transfer function of the amplifier. That leads to PIM. There are a couple of interesting twists, however. First, the application of negative feedback in general tends to reduce the effects of open-loop variations; it tends to stabilize gain and phase and bandwidth in going from the open-loop condition to the closed-loop condition. To this extent, NFB tends to decrease the PIM that would have been naturally present in the open-loop amplifier without feedback.
On the other hand, the application of NFB causes a conversion from open-loop AIM to closed-loop PIM to occur. As the open-loop gain of the amplifier varies with signal excursions, we know that the closed loop bandwidth of a conventionally compensated negative feedback amplifier will change. So at signal extremes, if the open-loop gain decreases, for what ever reason, the closed loop bandwidth will decrease, and more in-band phase shift will result, and this is PIM. This is no different than saying that the closed-loop bandwidth of the amplifier is changing with signal, just as in the case with a no-feedback amplifier.
In either case, it is just a matter of how much the closed-loop bandwidth changes with signal excursions and how far out that pole is. At this point it is important to recognize that, to first order, a 10% reduction in open loop gain due to signal excursions will cause an approximate 10% reduction in closed loop bandwidth whether the amplifier has wide open loop bandwidth or narrow open-loop bandwidth; the rolloff slope of the NFB at high frequencies is approximately the same in either case if both amplifiers have the same nominal gain-crossover frequency. But the higher this gain crossover frequency, the less PIM will be created by a given percentage modulation of the open-loop gan.
Another twist reveals itself when we examine one of the oft-cited sources of open-loop gain modulation that can lead to closed loop phase variation and thus PIM in negative feedback amplifiers. Consider the input differential pair. When it is stressed by large signals, its gain decreases. The larger the stress on the input stage, the greater is the potential source of PIM. Consider what happens when we have two NFB amplifiers with the same gain crossover frequency and nominal closed-loop bandwidth of 1 MHz. The first amplifier has a wide open-loop bandwidth of 20 kHz. It has an open-loop gain of 34 dB from 20 kHz on down to DC. The second amplifier has a narrow open-loop bandwidth of 2 kHz. It has an open-loop gain of 34 dB at 20 kHz that increases to 54 dB at 2 kHz and is constant from there on down to DC. Assume that both amplifiers employ the identical input differential pair stage. The amplifier with higher negative feedback and lower open loop bandwidth will actually stress the input stage far less than the amplifier with wide open loop bandwidth. The error signal applied to the input stage at frequencies below 2 kHz is ten times smaller than in the case of the amplifier with wide open-loop bandwidth. This means that this source of PIM will actually be smaller in the NFB amplifier with higher open-loop gain and lower open loop bandwidth.
Cheers,
Bob
Re: Re: Re: Re: PIM
Hi Glen,
Indeed, one more source of uncertainties. Actually, I know for years (due to my work on the PC based distortion analyzer) that for precise phase measurements, one need more powerful techniques. As mentioned before, based on FFTs, for example. If that leads to too much computational overhead, one can also use a simple convolution with a sine and cosine.
Anyhow, the clue is to use as much data points as you can. Only then the disturbing effects of (quantization) noise are minimized.
Cheers,
Edmond.
G.Kleinschmidt said:
Hi Glen,
Indeed, one more source of uncertainties. Actually, I know for years (due to my work on the PC based distortion analyzer) that for precise phase measurements, one need more powerful techniques. As mentioned before, based on FFTs, for example. If that leads to too much computational overhead, one can also use a simple convolution with a sine and cosine.
Anyhow, the clue is to use as much data points as you can. Only then the disturbing effects of (quantization) noise are minimized.
Cheers,
Edmond.
PIM
Care to explain why you think that an AC simulation may provide more reliable results than a transient simulation? Or is it just a wild guess?
Also, how would you interpret the results? The phase shift of a signal with 1 and 10V amplitude respectively (for example), will certainly be different from a AC analysis with comparable (1...10V?) DC offsets.
Cheers.
Bob Cordell said:
Care to explain why you think that an AC simulation may provide more reliable results than a transient simulation? Or is it just a wild guess?
Also, how would you interpret the results? The phase shift of a signal with 1 and 10V amplitude respectively (for example), will certainly be different from a AC analysis with comparable (1...10V?) DC offsets.
Cheers.
Bob Cordell said:
Yes.
Not all of the tax payers money is wasted:
http://tf.nist.gov/phase/Properties/main.htm
publication survey:
http://tf.nist.gov/cgi-bin/showpubs.pl
Replace "carrier" with "test tone". The NIST people are more
interested in AM/PM conversion at the noise level, but the
same mechanisms apply for large signal excitation.
Even worse.
http://tf.nist.gov/timefreq/general/pdf/1287.pdf
http://tf.nist.gov/timefreq/general/pdf/1139.pdf
http://tf.nist.gov/timefreq/general/pdf/1134.pdf
and another FET low noise amplifier:
http://tf.nist.gov/timefreq/general/pdf/821.pdf
http://tf.nist.gov/timefreq/general/pdf/655.pdf
BTW: Walls cites the formula for the input noise density:
enq = sqrt(2/3 * 4KT/gm) published by van der Ziel in 1962.
Since K is a constant and T is close to constant unless you have
a refrigerator, this leaves maximizing gm if you want low noise density.
I have no doubt that they knew in 1962 how to maximize the
gm of a FET: make it fat.
If you have the money, make a fat chip like TI's P8000/P8002,
CP643-lookalikes from Teledyne-Crystalonics or, more recently,
Interfet IF3602.
If you don't want to spend the money, you must plow with the
power of a thousand chickens and parallel many small FETs to
get a big one. That's how it was done in the BT, but the way
was well known in 1962 from first principles.
regards, Gerhard
- Status
- Not open for further replies.