Actually what it does is magnify the nonlinearities by reducing the noise gain that reduces the nonlinearities. its a pretty valid way to check an opamp except there are probably some things it won't show if the opamp is run with full gain.
The -117 dB comes from the input noise of the analyzer (usually around 3-5 nV/rtHz) and the residual distortion of the analyzer in the 30KHz or 80 KHz bandwidth of the measurement. Getting better as a traditional analyzer its exceptionally difficult.
A high resolution FFT after a passive notch where the noise is reasonably low can look much deeper for harmonics. However the linearity of the passive components becomes an issue as well.
The -117 dB comes from the input noise of the analyzer (usually around 3-5 nV/rtHz) and the residual distortion of the analyzer in the 30KHz or 80 KHz bandwidth of the measurement. Getting better as a traditional analyzer its exceptionally difficult.
A high resolution FFT after a passive notch where the noise is reasonably low can look much deeper for harmonics. However the linearity of the passive components becomes an issue as well.
I don't want to keep on about the THD too much,
but the important thing is that THD is level dependent and needs a
reference to be meaningful.
Ohtherwise you can not compare different amps and measurements.
And THD goes very fast to zero if the level is reduced and if noise is ignored,
this follows from Taylor series expansion of the nonlinear transfer function.
The High-End is in Munich this weekend, is anybody there?
Best wishes,
Udo
but the important thing is that THD is level dependent and needs a
reference to be meaningful.
Ohtherwise you can not compare different amps and measurements.
And THD goes very fast to zero if the level is reduced and if noise is ignored,
this follows from Taylor series expansion of the nonlinear transfer function.
The High-End is in Munich this weekend, is anybody there?
Best wishes,
Udo
Yes and that is exactly where you misinterpreted. -90dB THD on a -60dBV signal is not -140dB, it is -90dB THD. The reference is at -60dBV.
Jan
Jan, i wanted to know if *you* are at the High-End 
The nonlinear harmonics go much faster to zero than the fundamental.
If you reduce the input amplitude by 10, the first harmonic is reduced by 100,
the 2nd harmonic is reduced by 1000, therefore the THD improves by 10.
Here is a 15 minute reference which nicely explains the thing:
http://www.ittc.ku.edu/~jstiles/622/handouts/Intermodulation Distortion.pdf
If you still don't belive, try a spice simulation of a differential pair with
different input levels...
Udo
The nonlinear harmonics go much faster to zero than the fundamental.
If you reduce the input amplitude by 10, the first harmonic is reduced by 100,
the 2nd harmonic is reduced by 1000, therefore the THD improves by 10.
Here is a 15 minute reference which nicely explains the thing:
http://www.ittc.ku.edu/~jstiles/622/handouts/Intermodulation Distortion.pdf
If you still don't belive, try a spice simulation of a differential pair with
different input levels...
Udo
Jan, i wanted to know if *you* are at the High-End
The nonlinear harmonics go much faster to zero than the fundamental.
If you reduce the input amplitude by 10, the first harmonic is reduced by 100,
the 2nd harmonic is reduced by 1000, therefore the THD improves by 10.
Here is a 15 minute reference which nicely explains the thing:
http://www.ittc.ku.edu/~jstiles/622/handouts/Intermodulation Distortion.pdf
If you still don't belive, try a spice simulation of a differential pair with
different input levels...
Udo
Hmm ... if you look at this the other way round ... if you increase the level by a factor of say 10 the high order harmonics would explode, e.g. the 42th harmonic would be by a factor 10^42 bigger ... terrifying ;-)
Last edited:
Jan, i wanted to know if *you* are at the High-End
The nonlinear harmonics go much faster to zero than the fundamental.
If you reduce the input amplitude by 10, the first harmonic is reduced by 100,
the 2nd harmonic is reduced by 1000, therefore the THD improves by 10.
Here is a 15 minute reference which nicely explains the thing:
http://www.ittc.ku.edu/~jstiles/622/handouts/Intermodulation Distortion.pdf
If you still don't belive, try a spice simulation of a differential pair with
different input levels...
Udo
It is not a matter of believing. There is indeed a (limited) range of amplitude in a diff pair where this holds.
But that was not the discussion, and your own measurement contradicts you.
Jan
back to where we started
Hello,
“…magnify the nonlinearities by reducing the noise gain that reduces the nonlinearities…” is close to a direct quote from the data sheet attached in post 1140, I read that too. That is an abbreviated or simplified version of the longer winded more complete feedback / feedforward explanation.
"What it actually does" is attenuate the input by a value equal to the feedback factor; R1/(R1+R2). The attenuated input plus the input-error-signal are amplified by the circuit gain; (1/R1/(R1+R2)).
The net result is that the input sees unity gain and the error-signal sees a gain equal to (R1+R2)/R1. Another way of looking at it is that R1 is bootstrapped to selectively amplify the error signal. This is back to where we started.
DT
Hello,
“…magnify the nonlinearities by reducing the noise gain that reduces the nonlinearities…” is close to a direct quote from the data sheet attached in post 1140, I read that too. That is an abbreviated or simplified version of the longer winded more complete feedback / feedforward explanation.
"What it actually does" is attenuate the input by a value equal to the feedback factor; R1/(R1+R2). The attenuated input plus the input-error-signal are amplified by the circuit gain; (1/R1/(R1+R2)).
The net result is that the input sees unity gain and the error-signal sees a gain equal to (R1+R2)/R1. Another way of looking at it is that R1 is bootstrapped to selectively amplify the error signal. This is back to where we started.
DT
Udo, no offense taken, and I do not criticise your calculation or simulation. I only pointed out that in the graph showing a signal at -60dBV with a THD 90dB below that, that is -90dB THD, not -150dB. That's all.
Jan
peak to peak noise displayed on the oscilloscope
Hello Chris,
The point:
We do not want to limit ourselves.
Sometimes a simple model (KISS) works, sometimes we need to look a little deeper.
The TI op-amp data sheet is an example. An AP analyzer with a specification distortion sensitivity of -117dB is used to measure distortion levels in the range of -150dB.
Another example is Ed Simon’s use of a bridge to extend an AP analyzer to -170dB for resistor distortion measurements.
Also keep in mind the limits to this approach. FFT’s are never real time. FFT’s average and smooth nonperiodic data to the point that periodic data becomes visible above the FFT “noise floor”. We get the impression that we can pick the fly poop out of the pepper.
Look at the same data on an oscilloscope; the same Harmonics that are plainly visible in a 20 times averaged FFT plot will be completely lost in peak to peak noise displayed on the oscilloscope.
DT
Hello Chris,
The point:
We do not want to limit ourselves.
Sometimes a simple model (KISS) works, sometimes we need to look a little deeper.
The TI op-amp data sheet is an example. An AP analyzer with a specification distortion sensitivity of -117dB is used to measure distortion levels in the range of -150dB.
Another example is Ed Simon’s use of a bridge to extend an AP analyzer to -170dB for resistor distortion measurements.
Also keep in mind the limits to this approach. FFT’s are never real time. FFT’s average and smooth nonperiodic data to the point that periodic data becomes visible above the FFT “noise floor”. We get the impression that we can pick the fly poop out of the pepper.
Look at the same data on an oscilloscope; the same Harmonics that are plainly visible in a 20 times averaged FFT plot will be completely lost in peak to peak noise displayed on the oscilloscope.
DT
If you have a bank of narrow band analog filters those harmonics will show up. The scope has a very wide band so the energy in the whole spectrum is visible all at once. Those harmonics are still there, along with lots of other stuff.
Depending on your goal one view or the other may be relevant. The averaging used with an FFT is similar to the waveform averaging that happens with a scope. Both remove signals that are not repeating like noise. A single shot in both cases has lots of extra stuff. A single shot FFT is certainly possible and for larger signals you can see what you need. A tuned filter can also show a lot.
Depending on your goal one view or the other may be relevant. The averaging used with an FFT is similar to the waveform averaging that happens with a scope. Both remove signals that are not repeating like noise. A single shot in both cases has lots of extra stuff. A single shot FFT is certainly possible and for larger signals you can see what you need. A tuned filter can also show a lot.
easily identified offenders
Hello,
Demian, you are right on target, the oscilloscope is a much wider BW tool than the FFT. As I was writing the above post I was scratching my head, simple or more complex. I opted not to bring up BW limits.
I use an AP to measure noise in tube designs. With the BW set to 10hZ - 500K the noise number is huge. With the BW set to 400hZ – 22K the measured noise is much reduced. I prefer a C-weighted filter as being more representative for humans listening to audio.
For harmonics 5th and higher they are most likely to be lost in the noise, also to have huge variation from tube to tube. For harmonics 2nd and 3rd they are the most easily identified offenders.
For looking at power supplies I prefer the BW limited scope first and then the FFT to identify the offenders. In this case, the FFT is a scope feature.
Yes, both oscilloscope and FFT are useful tools.
DT
Hello,
Demian, you are right on target, the oscilloscope is a much wider BW tool than the FFT. As I was writing the above post I was scratching my head, simple or more complex. I opted not to bring up BW limits.
I use an AP to measure noise in tube designs. With the BW set to 10hZ - 500K the noise number is huge. With the BW set to 400hZ – 22K the measured noise is much reduced. I prefer a C-weighted filter as being more representative for humans listening to audio.
For harmonics 5th and higher they are most likely to be lost in the noise, also to have huge variation from tube to tube. For harmonics 2nd and 3rd they are the most easily identified offenders.
For looking at power supplies I prefer the BW limited scope first and then the FFT to identify the offenders. In this case, the FFT is a scope feature.
Yes, both oscilloscope and FFT are useful tools.
DT
@udo
Just a remark on your findings on THD for low level. Apparently there are amplifiers where THD increases for low amplitudes, in particular B-class push-pull with these cross-over distortions.
I read the paper by Jim Stiles. His findings are based on a Taylor expansion of the transfer function. However, for some functions the Taylor expansion does not even exist, and for many functions this is not always a good approximation. I think the transfer function for a class-B amplifier is one of those cases where a Taylor or a more general polynomial expansion is problematic.
The paper by Stiles addresses distortions of (RF) amplifiers close to clipping, where a Taylor expansion is apparently more suitable.
Wolfgang
Just a remark on your findings on THD for low level. Apparently there are amplifiers where THD increases for low amplitudes, in particular B-class push-pull with these cross-over distortions.
I read the paper by Jim Stiles. His findings are based on a Taylor expansion of the transfer function. However, for some functions the Taylor expansion does not even exist, and for many functions this is not always a good approximation. I think the transfer function for a class-B amplifier is one of those cases where a Taylor or a more general polynomial expansion is problematic.
The paper by Stiles addresses distortions of (RF) amplifiers close to clipping, where a Taylor expansion is apparently more suitable.
Wolfgang
Any updates?
Trying to be patient, know it will be some time.
+1
Even just a schedule update would be great.
I'm going to let him relax and work on getting them out. Any pressure will not be to our benefit. We have a rough time frame and that's good enough for me.
Hi DT,
I agree with that. I am very sceptical of extremely low noise measurements that you are quoting though. At some point, even thermal noise in a wire will limit our ability to see these low levels unless we use some super cool cooling method. But then, we lose relevance to real world circuit behaviour.
-Chris
Hi DT,
I agree with that. I am very sceptical of extremely low noise measurements that you are quoting though. At some point, even thermal noise in a wire will limit our ability to see these low levels unless we use some super cool cooling method. But then, we lose relevance to real world circuit behaviour.
-Chris