Low-distortion Audio-range Oscillator

Disabled Account
Joined 2012
Here is one mode of analyzer..... 1v from a generator to the AD725D and to a VM and QA401 via USB and display.

All indicate 1v or FS or 0 dBv.


DSC02821.JPG



THx-RNMarsh
 
Last edited:
Yes in the analysis mode the amplitude of the harmonics is the only information, it's like an FFT where the resolution is set by the fundamental frequency i.e. the bins are at fo, 2Xfo, 3Xfo, etc. This is from when FFT systems were rare and/or expensive. With an FFT you only need the spurs to be 10db > than the noise level to be within 1dB accuracy, at 20dB for 0.1dB accuracy.

I ordered a couple of Victor's oscillators, as Bruce Hofer showed me 30yr ago a passive notch is unambiguous and with care can approach the -140/150 THD level and at single frequencies the easiest way to verify performance.

If Richard's plot is normal mode from the block diagram IMO there is an error in computing the reference level.

Where about is the sighted error? What stage?
 
Disabled Account
Joined 2012
I get a lot more noise when trying to measure Victors osc/gen.

Gain error in level mode is very accurate to -110dBv.

[No need to measure below that for the analyzer mode monitor output.... noise limited ]

Anyway..... is there an inverting amplifier used in the QA401's input?



-Richard
 
Last edited:
Some practical comparisons regarding averaging.
Before measurement the boards were about two hours in running for to prevent frequency drift in the warm up process and smoothing as result. The measurement without averaging was done after the averaged measurement.
"Typical" 1kHz board without averaging:
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHz22nF.jpg
The same board with averaging 30 times:
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHz22nFAver30.jpg
My own 1kHz oscillator example without averaging (measurement was done with the double shielding):
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHzMy.jpg
With averaging:
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHzMyAver30.jpg

Victor.
 
Member
Joined 2004
Paid Member
Some practical comparisons regarding averaging.
Before measurement the boards were about two hours in running for to prevent frequency drift in the warm up process and smoothing as result. The measurement without averaging was done after the averaged measurement.
"Typical" 1kHz board without averaging:
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHz22nF.jpg
The same board with averaging 30 times:
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHz22nFAver30.jpg
My own 1kHz oscillator example without averaging (measurement was done with the double shielding):
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHzMy.jpg
With averaging:
http://content32-foto.inbox.lv/albums/e/elterra/1kHzAver/1kHzMyAver30.jpg

Victor.

Victor- Your oscillator has HD3 at -168dB if I read that correctly? Now we are going to hound you on how to get more of those. Because we can't measure -150 accurately we need something even more out of reach (and beyond attainability).
 
Victor- Your oscillator has HD3 at -168dB if I read that correctly? Now we are going to hound you on how to get more of those. Because we can't measure -150 accurately we need something even more out of reach (and beyond attainability).

Again, details please to put this in context. If need be use Samuel Groner's thesis as a guide. Trustworthy repeatable measurements at -168dB in a home lab environment are not believable.
 


Noise floor is ~165dB.


Noise floor is now ~170dB. I hate sounding like a broken record, but frequency domain averaging cannot lower the noise floor, while synchronous averaging would lower the noise floor by about 10*LOG(N)~14dB.

I assume the number of frequency bins are the same. Then this 5dB difference must be explained.
 
Again, details please to put this in context. If need be use Samuel Groner's thesis as a guide. Trustworthy repeatable measurements at -168dB in a home lab environment are not believable.

I am surprised too :). My target was only to show, how averaging works in my system. My own very old 1kHz prototype board mounted in the aluminum enclosure shows this unbelievable spectrum, when averaging is used. I can't measure harmonics from this board since I replaced one old Wima FKP2 capacitor with one selected Xicon polystyrene. This old board is fully hand made without mask. I had suspicions regarding that the PCB material affect to the oscillator performance. Some time ago, when I replaced 1,5mm FR4 to 1,2mm, I got better performance in the serial boards. Maybe chemicals traces from the manufacturing process can affect... Also AC voltage across the AGC FET is lower then in my serial boards - around 30mV p-p instead of usually 40mV (at maximum output), and this board has small shied over the AGC region - similar as on my 10kHz boards. Now I am in the investigation, what I can measure with the averaging. Seems, that the my new twin T built from the LCR polystyrenes is little better then the old with Wima FKP2 caps...
 
Last edited:
Member
Joined 2004
Paid Member
Again, details please to put this in context. If need be use Samuel Groner's thesis as a guide. Trustworthy repeatable measurements at -168dB in a home lab environment are not believable.

I hope you didn't think I was completely serious. However both of my Radiometer CLT-1's have a no load residual of approx -170 on HD3 (10 KHz, the only frequency it operates at) in my home lab. Its impressive but not too relevant except for its specific purpose.
 

Attachments

  • CLT-1 residual.png
    CLT-1 residual.png
    804.2 KB · Views: 252
Member
Joined 2004
Paid Member
[/COLOR]

Noise floor is ~165dB.



Noise floor is now ~170dB. I hate sounding like a broken record, but frequency domain averaging cannot lower the noise floor, while synchronous averaging would lower the noise floor by about 10*LOG(N)~14dB.

I assume the number of frequency bins are the same. Then this 5dB difference must be explained.

Two things about the plots. First the -165 or -170 without a known bandwidth for the bins is only a sort of relative number and not the actual noise density. Second, noise being random means that a single sample could have noise at a number of levels and some averaging would be needed to get a real value of the average or RMS noise level. We don't know how that software does an FFT. The different software packages I use usually have a fat line with as much as +/-10 dB on the first plot that gradually converges to a line at a lower level than the initial peak values. I find I need at least 5 averages to get a meaningful plot. Its possible that Victor's software initially shows just the peak values before averaging.
 
AX tech editor
Joined 2002
Paid Member
When I think about how it looks on my dScope, when I do averaging the peak-to-peak noise does get lower, but the *average* noise does not. In other words, if I draw a line through the average noise with no averaging, then activate averaging, the ragged peaky noise settles down to a much more flat line at the original average level. No change in level, only change in pk-pk values.
I think that is what Scott is referring to?

Jan
 
Disabled Account
Joined 2012
here is the 725D/QA401 analyzer test for accuracy at very low levels. At the moment, I have tested down to -90 FS range:

The fund freq is 1KHz and 1v or 0dbv. The injected freq is 2.4KHz at -90. The 725D is on its -90dbv scale. Seems OK.

[note: Using VP-7722Aas 1KHz source and ShibaSoku AG16B for 2.4KHz because of its 0-99.99dB atten, built-in]


2017-03-03 12.38.59.jpg



THx-RNMarsh
 
Last edited:
I hope you didn't think I was completely serious. However both of my Radiometer CLT-1's have a no load residual of approx -170 on HD3 (10 KHz, the only frequency it operates at) in my home lab. Its impressive but not too relevant except for its specific purpose.

The CLT-1 is not home lab equipment, I was referring to the notch filters and oscillator boards folks here are making. I don't recall ever seeing a single IC or audio equipment maker making even a remotely similar measurement claim with any commercially available test equipment. I'm preparing a simple post with pictures to make a few points it will be ready some time today.
 
When I think about how it looks on my dScope, when I do averaging the peak-to-peak noise does get lower, but the *average* noise does not. In other words, if I draw a line through the average noise with no averaging, then activate averaging, the ragged peaky noise settles down to a much more flat line at the original average level. No change in level, only change in pk-pk values.
I think that is what Scott is referring to?

Jan

Yes, you reduce the variance while the mean is unchanged. Statistics 101, roll a die, roll two dice, then three...
 
I am surprised too :). My target was only to show, how averaging works in my system. My own very old 1kHz prototype board mounted in the aluminum enclosure shows this unbelievable spectrum, when averaging is used. I can't measure harmonics from this board since I replaced one old Wima FKP2 capacitor with one selected Xicon polystyrene. This old board is fully hand made without mask. I had suspicions regarding that the PCB material affect to the oscillator performance. Some time ago, when I replaced 1,5mm FR4 to 1,2mm, I got better performance in the serial boards. Maybe chemicals traces from the manufacturing process can affect... Also AC voltage across the AGC FET is lower then in my serial boards - around 30mV p-p instead of usually 40mV (at maximum output), and this board has small shied over the AGC region - similar as on my 10kHz boards. Now I am in the investigation, what I can measure with the averaging. Seems, that the my new twin T built from the LCR polystyrenes is little better then the old with Wima FKP2 caps...

Not trying to be critical at all, just trying to rationalize all the different results. I feel in general folks make too little out of the noise floor when in fact it has great power to keep you honest. I will demonstrate soon.

I'm back so please PM me your Paypal info and an amount and I will just transfer the money, this is the easiest way for me.