Good idea, ask Viktor. Or at least a position for a connector.
I am also modifying my Viktor oscillators for balanced out; would be nice if there was a provision so I don't have to add two resistors haphazardly ;-)
Jan
But the RTX6001 has an internal xtal clock for the DAC generator, no? That should be more than stable enough for this purpose!
I will try that out this week when I get time.
Jan
Hi Jan,
Almost. With everything on the same reference clock, 1001 Hz from the RTX measures 1001.000 Hz on the test equipment. So if I am modulating something, I know exactly what the offset is and any broadening of the peak is noise or another signal for certain. If I then decode the signal and read 1001 Hz with jitter, I know the LO is unstable, and by how much.
-Chris
Almost. With everything on the same reference clock, 1001 Hz from the RTX measures 1001.000 Hz on the test equipment. So if I am modulating something, I know exactly what the offset is and any broadening of the peak is noise or another signal for certain. If I then decode the signal and read 1001 Hz with jitter, I know the LO is unstable, and by how much.
-Chris
Conceptually, the high quality Victor oscillator is a *really* high Q filter. That's why the harmonics are so low. So, the reference clock may not have to be super clean because the oscillator will filter it.
This makes me wonder if the very same idea could be applied to the QA401. The locking is the key.
Is the AP input circuit published for that version? I have found thatlowering noise requires looking at all the input dividers and protection circuits. Often you cannot improve one aspect without compromising some other aspect.
The easy way to lower the noise would be a bigger fft. However I don't think the AP supports that. However the RTX does and you can use Diana for a better analysis.
The easy way to lower the noise would be a bigger fft. However I don't think the AP supports that. However the RTX does and you can use Diana for a better analysis.
If you can get long sequence of samples to PC, any FFT length is possible.
Jan, congrats to your important result. What I just do not understand - analog technology is being developed for almost 100 years, countless research teams, academic, commercial. Yet no such elegant and relatively simple solution has been uncovered. I see it in other areas too. Advanced simulation techniques have moved the research to further levels. One would expect all of this being explored to every tiniest detail by now.
Jan, congrats to your important result. What I just do not understand - analog technology is being developed for almost 100 years, countless research teams, academic, commercial. Yet no such elegant and relatively simple solution has been uncovered. I see it in other areas too. Advanced simulation techniques have moved the research to further levels. One would expect all of this being explored to every tiniest detail by now.
Yes, those AP circuits have been very much optimized. You can't improve them simply by changing opamps or resistor types. The circuits are not published, but we know that they are very similar in concept to the System 1 (S1) circuits, and these are publicly available.
The AP does support very long FFTs, like 4M points at 32k SR. But I don't know how this is processed with the 'no window - move to bin center' setting.
I understand that you can transfer such very long records to Octave or Matlab to FFT but I have no experience with that.
Jan
I have been thinking about that, but replacing the AP input board is a big undertaking.
With my external twin-tee at the AP input, the fundamental is attenuated say 40-50dB. The AP autoranger automatically applies the appropriate gain to bring that signal back up to the sweet spot of its analyzer, around 3V. I guess that's where most of the noise occurs, although the limit might also be at the ADC circuitry.
So one idea could be to build this make-up amplifier externally with extremely low noise and manually set the AP input ranger to 0dB straight through, which probably is it's lowest noise setting. That external amplifier need not be very low distortion because the DUT distortion is 40-50dB increased due to the twin-tee action.
So, lots of things to try! ;-)
Edit: ARailsback we posted simultaneously ...
Jan
Jan, please I do not understand the FFT image - the twin tee attenuates by 50dB, yet the fundamental is measured at -170dBr.
Longer FFTs on PC running on exact digitally-generated signals in some softwares (probably running at float precision) have artefact peaks reaching to -170dBFs - that is exact 24bit signal, no analog conversion along the way. The chart shows measurements at -180dB, hair-thin line. Please what were the FFT settings?
Longer FFTs on PC running on exact digitally-generated signals in some softwares (probably running at float precision) have artefact peaks reaching to -170dBFs - that is exact 24bit signal, no analog conversion along the way. The chart shows measurements at -180dB, hair-thin line. Please what were the FFT settings?
Yes, the Y-scale in in dBr, meaning it is referenced to whatever I set as reference. The test oscillator output is 2V or +6dBV. The twin tee attenuates the 2nd by 8dB. So I set the reference for the dBr scale to -2dBV.
The twin-tee attenuates the 3rd by 4dB. So with these settings, the 3rd (if visible) would be shown 4dB higher than it actually is.
Jan
Not sure I understand this, but the FFT settings were 32k SR and same FFT length. The digital generator DAC runs at 48kHz SR, and the analyzer ADC at 65kHz SR.
If you refer to the spike at -170dBr @ 3kHz, that of course is the fundamental, suppressed by the notch filter.
Jan
Last edited:
Thanks. Please where am I making a mistake?
fundamental at output is +6dBV
notch -50dB => fundamental behind notch at -44dBV
your settings 0dBr = -2dBV => fundamental at -44 - (-2) = -42dBr
The chart shows -170dBr
Where are the missing 128dB? Is there some other setup in the AP for notch? Thanks a lot, I really do not understand the extremely low -170dBr for the fundamental. How can a passive filter attenuate the +6dBV fundamental way below the extremely low harmonics? I must be missing something important in the scheme. Thanks.
fundamental at output is +6dBV
notch -50dB => fundamental behind notch at -44dBV
your settings 0dBr = -2dBV => fundamental at -44 - (-2) = -42dBr
The chart shows -170dBr
Where are the missing 128dB? Is there some other setup in the AP for notch? Thanks a lot, I really do not understand the extremely low -170dBr for the fundamental. How can a passive filter attenuate the +6dBV fundamental way below the extremely low harmonics? I must be missing something important in the scheme. Thanks.
My point is -170dBFs is already hitting resolution/precision of some FFT calculations, it is an extremely low number. But my fault, I did not realize those were dBr, not dBFs. What would be that in dBFs?
Samples from ADC are always dBFs, I assume FFTs are in dBFs too, IMO moving the zero (conversion to dBr) is just before presenting the values.
If calculations run in double (float64), the precision will be much better than -170dB. However long FFTs in single precision (float32) are hitting -170dB. Look at the chart of jaaa running 8M FFT of a precise digital sine produced by single-precision libfftw3f https://www.diyaudio.com/forums/equ...ort-samplerates-sw-analyzers.html#post6133979
If calculations run in double (float64), the precision will be much better than -170dB. However long FFTs in single precision (float32) are hitting -170dB. Look at the chart of jaaa running 8M FFT of a precise digital sine produced by single-precision libfftw3f https://www.diyaudio.com/forums/equ...ort-samplerates-sw-analyzers.html#post6133979
Last edited:
Lets do some mental calculations then, I also want to understand it. Start with 1V to keep it simple. Through the twin-tee gets you -50dBV.
Next, the AP autoranger applies say 60dB gain, which it remembers as a correction to be applied later. The fundamental signal is now 3V.
Next comes the AP notch filter. I don't know it's performance, but let us say -110dB; that seems a reasonable assumption looking at many 'normal' AP 2722 spectra. That puts the fundamental at 10uV.
I have set the reference to -2dBV. 10uV is -110dBV or, -108dB below the reference -2dBV. Correct by the 60dB internal gain puts it at -168dBV.
Have I got that right?
Edit: that notch is even better apparently ..
Jan
Next, the AP autoranger applies say 60dB gain, which it remembers as a correction to be applied later. The fundamental signal is now 3V.
Next comes the AP notch filter. I don't know it's performance, but let us say -110dB; that seems a reasonable assumption looking at many 'normal' AP 2722 spectra. That puts the fundamental at 10uV.
I have set the reference to -2dBV. 10uV is -110dBV or, -108dB below the reference -2dBV. Correct by the 60dB internal gain puts it at -168dBV.
Have I got that right?
Edit: that notch is even better apparently ..
Jan
Attachments
Last edited:
Thanks.ARTA, for ex, can optionally use 64bit math when distorsion is low.
Is there a table where one can see needed nr of digits in the math software code vs wanted FFT resolution?
Is there a table where one can see needed nr of digits in the math software code vs wanted FFT resolution?
That -170dBr is referred to a certain number. It does not mean that there has been a -170dB calculation. See my post above.
But if you want you can calculate how much (little) -170dB is; your Windows calculator can do that {10^(-170/20)}. Then you know how many digits are needed.
Jan
But if you want you can calculate how much (little) -170dB is; your Windows calculator can do that {10^(-170/20)}. Then you know how many digits are needed.
Jan
Last edited: