Discrete Opamp Open Design

[abraxalito]

Three tones would certainly be a step in the right direction, but why stop there? Here's a test signal I composed recently but haven't tried out on any real circuits yet as I'm not done playing with it. I plan to add some LF components - this signal has just over 100 discrete tones about -40dBFS each. The second image shows the time domain - the crest factor is quite unlike a sine and I've not done the 'Schroeder Phase' thing to minimize it

[RNMarsh]

Or -- A random wide frequency band (flat) noise generator as source used with a high precision subtractive circuit to get I/O difference level. -RNM

[SY]

Originally Posted by RNMarsh

After 20,000 tones have been applied, that .01% THD of each of thousands of tones adds up to enough to reach audible levels.

Does this account for masking (i.e., audibility in the presence of 20,000 large signals)? And how do you fit 20,000 different tones into the audible bandwidth and still be able to see any harmonics? Any references on this?

[SY]

Originally Posted by abraxalito

Of course it does - my point was that the AP made it more accessible to the average punter. Just as computers went back prior to the founding of Microsoft yet it was Bill Gates who made them widely accessible.

Gee, I had a Heathkit IM analyzer when I was a young lad back in the '60s. Lots of them were around (also Eico, Knight). A hell of a lot more accessible to THIS average punter than a $10k AP. Now, I'd certainly rather have the AP, it does a lot more and does it better, but the price is well out of the range of any non-rich non-professional.

[abraxalito]

Originally Posted by SY

Gee, I had a Heathkit IM analyzer when I was a young lad back in the '60s. Lots of them were around (also Eico, Knight).

How did that allow you to create your own multitone test waveforms?

[RNMarsh]

Originally Posted by SY

Does this account for masking (i.e., audibility in the presence of 20,000 large signals)? And how do you fit 20,000 different tones into the audible bandwidth and still be able to see any harmonics? Any references on this?

Charting new territoty here -- See #1052 for yet another way to do it besides using AP equipment.

It isnt a masking test. But the affect has been discribed and it seems to corrallate with this total tones harmonics higher number's affect.

Do the rt1/2 sum of the sq of all the harmonic levels and tell me what it comes out to be with 20K tones each with harmonic of .01% Any approach you choose will give a large total distortion number. Or even 1000 tones and thier harmonics at .01% each (say 2,3, 4 and 5th only). Would it be audible if the detection threshold is .05% or less? Should be. -RNM

[fas42]

Originally Posted by RNMarsh

Do the rt1/2 sum of the sq of all the harmonic levels and tell me what it comes out to be with 20K tones each with harmonic of .01% Any approach you choose will give a large total distortion number. Or even 1000 tones and thier harmonics at .01% each (say 2,3, 4 and 5th only). Would it be audible if the detection threshold is .05% or less? Should be. -RNM

Wait a minute ... is that a valid way of looking at how a circuit "adds" further distortion when the signal is more complex? As far as the circuit and the parts within are concerned, all they ever see is an input wiggle, a varying voltage signal, they don't "know" that they're supposed to break everything down into sine waves, amplify each individually with at attendant measure of distortion, and then add it all up in a nice, mathematical way. Of course, if they do, hats off to such smart componentry!!

At any point in time the output signal will be a certain value incorrect, the instantaneous distortion component so to speak. And to me that will never vary substantially from what the circuit state would be if it happened to be amplifying a pure sine wave with the waveform exactly corresponding to a part of that curve. That is, unless you want to take a lot of "memory" and similar effects into account ...

Frank

[dchisholm]

Originally Posted by RNMarsh

Or -- A random wide frequency band (flat) noise generator as source used with a high precision subtractive circuit to get I/O difference level. -RNM

A somewhat related idea would be the "Noise Power Ratio (NPR)" test used to evaluate the FDM telephony systems from about the 1940's through the 80's.

The idea was to apply a random, broadband signal (band-limited noise) to the whole information bandwidth of the system. Then a very sharp, narrow, notch filter removed the noise from a small segment of the input spectrum (roughly 1% or less of the total information bandwidth). As that clean notch in the input spectrum passed through the system it would collect corruption, crud, and grunge from a wide range of sources - basic thermal noise; THD products from spectral components at lower frequencies; IMD products from spectral components at a variety of frequencies; power-supply ripple modulation of spectral components near the notch frequency; etc.

At the system output, an equally sharp, very narrow band, bandpass filter would measure the power in the notch. Comparing this to the power of the broadband input signal gave the "Noise Power Ratio". Usually there were three measurement notches - one at the lower end of the system passband, one at midband, and one in the upper end. This was primarily an evaluation test to gauge how good a system was overall, not an analytical test to localize a problem. (Sort of like a physician using temperature, pulse, and blood pressure to evaluate your general health - by themselves, those 3 measures don't diagnose any particular illness.)

While this technique was developed for use in FDM telephony systems where the information bandwidth extended from about 50 KHz to several MHz (a frequency ratio around 100:1), I suspect the principles could be applied to an audio bandwidth of 20 Hz - 20 KHz (1000:1 ratio). As with the FDM test sets, the challenge would be to create the notch and bandpass filters. Forty years ago they did it with passive analog filters; today it may require a combination of analog and digital filtering.

More information:
"Noise Power Ratio (NPR)" (Walt Kester) at < http://www.analog.com/static/importe...als/MT-005.pdf >
"Noise Power Ratio Measurement Tutorial" (Allen Katz and Robert Gray) at < http://lintech.com/PDF/npr_wp.pdf >
Thirty years ago the common reference book for NPR testing was published by Marconi Instruments and called (as I recall) "The White Noise Handbook", but it's probably long out of print.

Dale

[RNMarsh]

Originally Posted by fas42

Wait a minute ... is that a valid way of looking at how a circuit "adds" further distortion when the signal is more complex? As far as the circuit and the parts within are concerned, all they ever see is an input wiggle, a varying voltage signal, they don't "know" that they're supposed to break everything down into sine waves, amplify each individually with at attendant measure of distortion, and then add it all up in a nice, mathematical way. Of course, if they do, hats off to such smart componentry!!

At any point in time the output signal will be a certain value incorrect, the instantaneous distortion component so to speak. And to me that will never vary substantially from what the circuit state would be if it happened to be amplifying a pure sine wave with the waveform exactly corresponding to a part of that curve. That is, unless you want to take a lot of "memory" and similar effects into account ...

Frank

Its as valid as a single tone or two tones or three tones or multiple tones (AP) we use now... just more tones. Its a way to gauge the total level of all the harmonics generated... similar to when music frequencies are the input being worked upon.

[RNMarsh]

Originally Posted by dchisholm

A somewhat related idea would be the "Noise Power Ratio (NPR)" test used to evaluate the FDM telephony systems from about the 1940's through the 80's.

The idea was to apply a random, broadband signal (band-limited noise) to the whole information bandwidth of the system. Then a very sharp, narrow, notch filter removed the noise from a small segment of the input spectrum (roughly 1% or less of the total information bandwidth). As that clean notch in the input spectrum passed through the system it would collect corruption, crud, and grunge from a wide range of sources - basic thermal noise; THD products from spectral components at lower frequencies; IMD products from spectral components at a variety of frequencies; power-supply ripple modulation of spectral components near the notch frequency; etc.

Dale

This or something like it that gets at the total noise/harmonics of a system rather than a single tone's harmonics would be more realistic of what we are being exposed to when using the audio gear for listening to music signals. Thx for the idea. -RNM