In the first plot the THD is a measure of the RTX. The second gives -146 THD The third is more challenging with its different scaling factors but still around -130 dB THD.
With the Shibasoku you can take its monitor output into an FFT and scale appropriately and should get reasonable accurate harmonic levels. However these all may be close to cancellation with internal harmonics, something that can be difficult to check.
With the Shibasoku you can take its monitor output into an FFT and scale appropriately and should get reasonable accurate harmonic levels. However these all may be close to cancellation with internal harmonics, something that can be difficult to check.
Hi Spike,
To see any difference between the supplies (or anything for that matter), you're going to need a shield for starters. Put it into the case intended for the supplies and oscillator. Then, as Demian pointed out, you're going to have to look at the monitor output.
The Mosfet is operating as a CCS into a zener diode, giving you a stable pre-regulated input to the 317 regulator. That's going to be less noisy and have fewer amounts of any HF noise on the line. The worse your raw power is, the more of a difference the CCS - zener is going to make over a C multiplier.
-Chris
To see any difference between the supplies (or anything for that matter), you're going to need a shield for starters. Put it into the case intended for the supplies and oscillator. Then, as Demian pointed out, you're going to have to look at the monitor output.
The Mosfet is operating as a CCS into a zener diode, giving you a stable pre-regulated input to the 317 regulator. That's going to be less noisy and have fewer amounts of any HF noise on the line. The worse your raw power is, the more of a difference the CCS - zener is going to make over a C multiplier.
-Chris
Demian, that's a good data point. I am currently also trying this on the RTX, running under Virtins MI. So far I am mostly trying to figure out Virtins ;-)
How do you do this with DiAna, just DiAna and the RTX as sound card?
Jan
Maybe RTX with compensated distortion would be able to measure the oscillator THD directly. An ordinary soundcard with original THD of -98dB loopback measures opamp THDs of -130dB quite reliably. I would expect RTX to perform significantly better when compensating from -120dB of its original THD.
Last edited:
Actually, ultra sharp Japanese knifes is another of my interests, so, yes, send me some hair and I´ll split them for you ;-)
Notch filter compensation
Above figures suggest that this B&K notch filter is the passive and classic version (and also similar to Jan Didden's balanced T filter), right?
In that case DiAna can adjust the harmonic amplitudes 'automatically' by means of simulation of such a filter. Just go to Option -> Apply notch compensation. If no notch response file was loaded, DiAna will ask "Use simulated values instead?" Press Yes and the harmonics (including the fundamental) will be restored to the levels just before the notch filter (provided that the frequency is exactly tuned to the notch dip, of course)
Cheers,
E.
Hi Demian,
Above figures suggest that this B&K notch filter is the passive and classic version (and also similar to Jan Didden's balanced T filter), right?
In that case DiAna can adjust the harmonic amplitudes 'automatically' by means of simulation of such a filter. Just go to Option -> Apply notch compensation. If no notch response file was loaded, DiAna will ask "Use simulated values instead?" Press Yes and the harmonics (including the fundamental) will be restored to the levels just before the notch filter (provided that the frequency is exactly tuned to the notch dip, of course)
Thank you for support.
Cheers,
E.
I can't leave well enough alone ...
My previous-best was with the synced Vic oscillator (3kHz), no window, move-to-bin-center.
What I did now was to slightly change the nominal 3kHz Vic oscillator I have to make it synchronous to the FFT to avoid the whole move to bin center processing and still not needing to use a window. I can do that now that I can sync Vic (not Vic, his oscillator ;-) to the AP digital generator.
Calculated what the nearest frequency is to 3kHz that will be synchronous when I sample at 48kHz, 32,768 FFT samples record length. Each sample has 1/48,000 = 20.83uS, and each 32,768k samples record thus takes 0.68267 seconds. Which freq closest to 3kHz will fit an integer # of cycles in this period?
204 cycles @ 2988.3Hz
205 cycles @ 3002.9Hz
206 cycles @ 3017.6Hz
... so pulling the oscillator to 3002.9Hz seemed the way to go.
This gave the attached curve, even lower than before, and still no trace of harmonics. Using the 'Sync' setting on the FFT processor means I cannot do averaging so the noise is more ragged than before, but clearly lower.
Vic where are your harmonics??
Jan
My previous-best was with the synced Vic oscillator (3kHz), no window, move-to-bin-center.
What I did now was to slightly change the nominal 3kHz Vic oscillator I have to make it synchronous to the FFT to avoid the whole move to bin center processing and still not needing to use a window. I can do that now that I can sync Vic (not Vic, his oscillator ;-) to the AP digital generator.
Calculated what the nearest frequency is to 3kHz that will be synchronous when I sample at 48kHz, 32,768 FFT samples record length. Each sample has 1/48,000 = 20.83uS, and each 32,768k samples record thus takes 0.68267 seconds. Which freq closest to 3kHz will fit an integer # of cycles in this period?
204 cycles @ 2988.3Hz
205 cycles @ 3002.9Hz
206 cycles @ 3017.6Hz
... so pulling the oscillator to 3002.9Hz seemed the way to go.
This gave the attached curve, even lower than before, and still no trace of harmonics. Using the 'Sync' setting on the FFT processor means I cannot do averaging so the noise is more ragged than before, but clearly lower.
Vic where are your harmonics??
Jan
Attachments
What it does:
When you are performing an FFT on an acquisition record that is longer
than the transform length, only the number of samples specified in the FFT
field will be transformed.
That is why there is an acquisition length choice 'track FFT'. So if you specify a 32k FFT length and select 'track FFT', 32k samples are acquired and transformed. If you select a longer acquisition record, you can select separate frames in that long record for transform, but the transform only happens on a frame with the # of samples specified in the FFT length.
If you want to analyze a longer record than the FFT length you need to do that offline.
Jan
When you are performing an FFT on an acquisition record that is longer
than the transform length, only the number of samples specified in the FFT
field will be transformed.
That is why there is an acquisition length choice 'track FFT'. So if you specify a 32k FFT length and select 'track FFT', 32k samples are acquired and transformed. If you select a longer acquisition record, you can select separate frames in that long record for transform, but the transform only happens on a frame with the # of samples specified in the FFT length.
If you want to analyze a longer record than the FFT length you need to do that offline.
Jan
Yes, the new 555B has a much more capable software (and a few dB less distortion) but starts at $ 30k. And in its wisdom AP now decided that with the new machines you need to pay a yearly software license between $ 1k and $ 2k depending on options ...
Unobtanium for us mere mortals.
@Dimitri: yes, I agree.
Jan
Unobtanium for us mere mortals.
@Dimitri: yes, I agree.
Jan
Not that I know of. Acquisitions are stored in an acquisition buffer before processing. Max acq. buffer is 4MS.
And not the whole buffer can be processed all at once. For instance, the max FFT record length is 32k samples. If you do 4MS, you can 'move' through the 4M to select a 32k set for processing, but not the whole 4M record.
Jan
