Re: IM distortion
If you can isolate distortion from signal, you can inject it to get some
advantage (read the paper on SuperSymmetry).
Or you can apply feedback, but then you knew that already.
Alternatively, I did play with inversion of coefficients of successive stages,
so that each cascaded stage has the "opposite" distortion of the previous.
It reduces the total distortion, but the increase in complexity remains.
This phenomenon has been well known since before I got involved in audio,
and I seem to recall Len Feldman writing about it in the 60's.
SCD said:
If you can isolate distortion from signal, you can inject it to get some
advantage (read the paper on SuperSymmetry).
Or you can apply feedback, but then you knew that already.
Alternatively, I did play with inversion of coefficients of successive stages,
so that each cascaded stage has the "opposite" distortion of the previous.
It reduces the total distortion, but the increase in complexity remains.
This phenomenon has been well known since before I got involved in audio,
and I seem to recall Len Feldman writing about it in the 60's.
Mr. Pass,
If IMD only can be cured by feedback, will putting local feedback (resistor from C-B or D-G) on every transistor in the circuit helps lower IM before global feedback applied? It will cut total gain alot , but help linearity, I think.
Or is this suggestion inferior to ordinary global feedback method (where we build up OL gain with many stages, then apply global feedback like usage of opamps).
If IMD only can be cured by feedback, will putting local feedback (resistor from C-B or D-G) on every transistor in the circuit helps lower IM before global feedback applied? It will cut total gain alot , but help linearity, I think.
Or is this suggestion inferior to ordinary global feedback method (where we build up OL gain with many stages, then apply global feedback like usage of opamps).
IMD is curable or preventable by all the same mechanisms that
cure or prevent THD, at least as far the article is concerned.
Resistors D-G, etc form local loop feedback, and we clearly can see
that they while they lower total distortion, they contribute to higher
orders of harmonics and sidebands.
cure or prevent THD, at least as far the article is concerned.
Resistors D-G, etc form local loop feedback, and we clearly can see
that they while they lower total distortion, they contribute to higher
orders of harmonics and sidebands.
Baby steps
So If I am getting this corrrect.
Relatively simple circuits with consistent, predicable, quality parts that are operated in their linear ranges should produce fairly low levels of IM distortion. Or at least levels that keep the music in the listenable range. Feed back should be used to adjust rather than correct. IE season to taste
So If I am getting this corrrect.
Relatively simple circuits with consistent, predicable, quality parts that are operated in their linear ranges should produce fairly low levels of IM distortion. Or at least levels that keep the music in the listenable range. Feed back should be used to adjust rather than correct. IE season to taste
leapcat said:Question
Would a fast amp lower the Im distortion of a complex
signal?
There are some articles that dig into amp bandwidth and distortion.
http://waltjung.org/Classic_Articles.html
At intervals I have suggested ideas for testing methods--most recently the idea of using two square waves instead of sine waves for IM testing. Nelson's seven tone test signal is a giant step in that direction (if you do a Fourier breakdown on a square wave, it has a mathematically infinite sequence of components, but the first few are the highest proportionally; they diminish in amplitude as the frequency rises). I, for one, am grateful that he took the time to do so. I don't have enough hardware to do that sort of thing at this time. I can generate three signals, but only two of them are of decent quality.
Nelson,
I may have missed it in the article, but what frequencies/amplitudes did you use?
Grey
Nelson,
I may have missed it in the article, but what frequencies/amplitudes did you use?
Grey
Took me a bit to find time to read the article. Seems the most technical so far.
Nelson, it's plain great that you took the next step and put out some more technical basics about distortion and especially IM!
If you write it, people will also read it.
All the best, Hannes
EDIT: I may have missed it, but what circuit did you use for the distortion spectra for ClassA and B? Looks to me as if it was something simple with gain, certainly no complementary feedback pair.
Nelson, it's plain great that you took the next step and put out some more technical basics about distortion and especially IM!
If you write it, people will also read it.
All the best, Hannes
EDIT: I may have missed it, but what circuit did you use for the distortion spectra for ClassA and B? Looks to me as if it was something simple with gain, certainly no complementary feedback pair.
Some questions,
1. An amplifier can only put out one voltage at a time?
2. Going from 1 volt to 2 volts it still needs to pass through all the voltages in between?
3. The transfer function is one to one?
4. Even with multiple input tones there is still only one voltage coming out of the amplifier at any point in time?
5. What is the highest frequency the amplifier is trying to reproduce?
1. An amplifier can only put out one voltage at a time?
2. Going from 1 volt to 2 volts it still needs to pass through all the voltages in between?
3. The transfer function is one to one?
4. Even with multiple input tones there is still only one voltage coming out of the amplifier at any point in time?
5. What is the highest frequency the amplifier is trying to reproduce?
MikeW said:
Mike,
1--Yes, an amplifier only puts out one voltage at any given instant. Sometimes it's helpful to think of the signal as an electrical analog of the sound pressure variations in the air at the time of the performance. If you measure, there's only one aggregate air pressure at any given instant, not several simultaneously. The disparate frequencies all add together to create one complex waveform. The ear knows.
2--Mathematically, it is possible to transition instantly from one voltage to another instantaneously. In the real world, there's some degradation, some slowness, due to the finite bandwidth of the circuit. A mathematically perfect square wave has an infinite series of harmonics added together, but in a real amplifier the harmonics are truncated above some arbitrary frequency, be it 50kHz, 100kHz, or whatever.
3--Excepting the amplification (assumed to be a constant multiplier), yes, the function should be one-to-one between the input signal and the output signal. Only...it ain't...
4--Yup. See #1.
5--Nelson? Fundamental = 100Hz, 1kHz...?
Grey
The analysis to which you refer is without reference to circuit
bandwidth or delay.
You may assume the bandwidth to be infinite, and toward that end
you may also assume a continuum of voltages.
The seven test tones are simply ratios, as given previously. You
may assign any arbitrary frequency to the lowest of the test tones.
Any discussion you want to get into with reference to time lags or
bandwidth would be on top of the results presented.
bandwidth or delay.
You may assume the bandwidth to be infinite, and toward that end
you may also assume a continuum of voltages.
The seven test tones are simply ratios, as given previously. You
may assign any arbitrary frequency to the lowest of the test tones.
Any discussion you want to get into with reference to time lags or
bandwidth would be on top of the results presented.
MikeW said:2- Could be but I doubt it. I do have a degree in Mathematics.
Consider something like digital. In the abstract, theoretical sense, the zeros and ones are assumed to be perfect, mathematical representations, with no unpleasantness resulting from finite rise times and such. On paper, a perfect computer need not worry about capacitance and inductance hampering bandwidth, hence rise times, and the zeros and ones are just that...zeroes and ones...not 0.01957s and 0.99372s when the clock pulse says they should equal 0.00000 and 1.00000. The math works perfectly...on paper. In the real world, it doesn't.
SY said:Bye, Fred.
There are any number of comments that come to mind, but I'll restrict myself to noting that things must be really, really dull over at the little home away from home that the rejects built if Fred has to come back over here to vent his spleen.
He hasn't changed a bit.
Nelson Pass said:The analysis to which you refer is without reference to circuit
bandwidth or delay.
You may assume the bandwidth to be infinite, and toward that end
you may also assume a continuum of voltages.
The seven test tones are simply ratios, as given previously. You
may assign any arbitrary frequency to the lowest of the test tones.
Any discussion you want to get into with reference to time lags or
bandwidth would be on top of the results presented.
I'm not sure whether you're posting to me or Mike. For my part, yes, I was assuming that there were differences between the mathematical ideal and the real world implementation, which is what I was trying to say--perhaps not as clearly as I'd hoped. On paper it's possible to go from 0 to 1 instantly if you sum an infinite series of harmonics in the right ratios. In reality it's not because the harmonic series isn't infinite and the rising/falling edges begin to lean a bit. Perfection assumes infinite bandwidth, which implies no reactance in the circuit. Clearly not possible. In a real circuit, the answer to Mike's question is that the voltage does indeed travel through all intermediate voltages; if frozen on a fast scope, you could see an intermediate discrete voltage, V, at any arbitrary time, T. Increasing the slew rate would asymptotically approach the ideal zero-to-one instantaneous transition, but with fast enough test equipment you could still see a less than infinite slope for the rise time.
This train of thought is exactly what lead me to wide bandwidth circuits, in the hopes of getting cymbals and other instruments with substantial high frequency content to sound right. In some ways I've exceeded even my wildest dreams, in others there's still something missing in sounds that I would assume would be improved by better high frequency content. It's better, but not all I'd hoped for. Still pondering. That's what makes all this so interesting. If the answers were easy, everything would have been settled long ago.
I had understood that the ratios were arbitrary and that the choice of fundamental could be as well. I was just curious as to what fundamental frequency you had chosen. My reasoning being that if the fundamental was 1kHz--the 28 ratio then being 28kHz--then the bandwidth limitations of, say, a 50kHz amplifier would lead to slightly different results than if the fundamental was 100Hz and the 28 was 2.8kHz, which would leave more room for upper harmonics (i.e. distortion products).
This is the point at which someone will draw the conclusion that limiting bandwidth reduces distortion. Yes, it does. My HP distortion analyzer has buttons to limit the bandwidth and pushing them does indeed lower the distortion readings. You can accomplish exactly the same thing by building a low bandwidth circuit. The problem is that it's one of those "throwing the baby out with the bathwater" situations. I choose wide bandwidth (I shoot for at least 250kHz) and live with slightly higher distortion than if I were to aim for 50kHz bandwidth. You could argue that the HP can 'hear' distortion ca. 100kHz that I can't hear, but that spins off in a completely different direction. I'm trying to get closer to the perfectly reproduced square wave school of thought, the idea being that the ear can detect rise time--hence waveform--even if it can't hear a discrete frequency that you might derive by separating the waveform into its Fourier components. Needless to say, others reach other conclusions.
You can view bandwidth limitations as leading directly to distortion, in that the signal that you get out is not the same as the one you put in. This is a wonderful indictment of the sine wave as test signal strategy. A single frequency will go in and come out with only the expected changes due to THD. But if you put in a square/triangle/sawtooth, it comes out visibly changed...distorted, in other words.
Conceptually, it's not that difficult to imagine a distortion meter based on a square wave instead of a sine wave, but since the result is--as far as I know--going to more or less reflect the bandwidth, why bother? Just increase the bandwidth and move on to the next thing.
Of course, my blind assumption that all such a test would show is the bandwidth limitation is exactly the same foolishness that I criticize in others when they make unfounded assumptions, so excuse me while I go cuss myself out...
Grey
That was a very informative article. It was stated "if you can make a single stage operate at .01% 2nd harmonic with a single tone without feedback, you could also achieve the .1% peak in the complex IM test"
I understand that this has been done - although the device really is only good at powering full range, ultra sensitive speakers. I wonder if anyone will ever develop the device that can do the same for larger amounts of current - say 10-12 amps? Then that would be an amp to build!
I understand that this has been done - although the device really is only good at powering full range, ultra sensitive speakers. I wonder if anyone will ever develop the device that can do the same for larger amounts of current - say 10-12 amps? Then that would be an amp to build!