Ok. That looks good and much better than the AK5572 evaluation board (below).
Attachments
Correct, but look at the orders of magnitude! That little 8kHz spikey is more than 40dB below the 20Hz..20kHz rms noise floor (and some 20dB below 24bit resolution, mind you, only captured by the self-dithering from the random noise). It's not even visible in a 4M FFT without averaging.
Note also that at 768Khz the packet size is large, increasing packet noise. At 44.1Khz there is not even a hint of it even when maxing out all possible resolution (there are some other "birdies" now, equally utterly irrelevant).
The interface is hooked up as isolated and floating as it ever gets, practically. USB feed is via Intona USB2 Isolator which reduces 8kHz packet noise and does full galvanic isolation, and the supply of the RME is ultra low-leakage medical DC/DC. I actually do "hot wire" measurement with this setup.
Attachments
We don't know the measurement conditions (FFT size, notably), so no way to really compare.
Hi Bob,Bob Cordell said:
Thank you for the detailed explanation.
I've been aware of the method ever since I assembled your Distortion Magnifier, but never thought about it in the context of the notch filter measurement. It certainly makes sense knowing that a linear system will preserve the harmonics' profile as the signal level is reduced.
I've got a couple of oscillators laying around, the distortion profile of which can be tweaked, so I'll only need to build an attenuator.
However, since the notch filter input is at the same time the input of my measurement chain, I can currently use the pilot tone for the calibration only. Mixing it with a DUT's signal would probably call for an active summing circuit.
Regards,
Braca
4M.
Btw, REW takes FFT gain into account when selecting rtHz Y-axis scale so once you set the 0dBFS value correctly for the input, noise measurement is correct regardless of FFT size (+- a few dB).
Last edited:
Below 10nV/rtHz is a remarkable noise floor. RME ADI-2 Pro is certainly a very good performer. Just a bit pricey
Hi Braca,
No need for an active summing circuit. Just inject a tiny current from the pilot oscillator into the output lead of the oscillator you are testing. For example, if you are testing a Victor oscillator with 100-ohm output impedance and you connect the test oscillator to it through a 1M resistor, you will have a 10,000:1 attenuator. If Victor's oscillator was operating at 1V and your pilot oscillator were operating at 1V, then you would see the pilot tone on the spectrum analyzer down 80 dB from the Victor oscillator fundamental, and see effectively 0.01% "distortion". If you set the pilot oscillator to 100 mV, you will see the pilot tone marker at a level of 0.001%. You can verify the above by using a signal chain that has no notch filter at first.
In my setup, I built a frequency-compensated passive twin-T filter with an associated LNA. A special filter after the twin T makes the twin T response pretty flat at the 2nd harmonic on up, and actually enhances the effective notch depth.
This is the case because the first stage of the compensation filter is a second-order state variable HPF tuned to the vicinity of the 2nd harmonic frequency with high enough Q to provide peaking that compensates for the passive twin T's loss at the second harmonic. As a result of its being an HPF at 2 kHz, it actually adds further loss at the 1 kHz fundamental frequency.
A second state variable filter is used as part of an arrangement to further flatten the distortion frequency response. Its action is primarily to make the distortion frequency response at the 2nd and 3rd harmonics the same.
The use of the accurately compensated twin T filter greatly simplifies measurements because I no longer have to worry about taking into account the attenuation of the passive twin T filter at the 2nd, 3rd, etc harmonic frequencies. It also helps to avoid cockpit errors.
I then take the output of the LNA and feed it into my THD analyzer set to 1 kHz. There is a further trick I use to provide the analyzer a fundamental signal with enough amplitude for the THD analyzer to properly auto-tune.
I then take the residual output of the THD analyzer and send it for FFT analysis to my QA401. I use a very large FFT with many averages to get the noise floor down below the distortion lines.
Cheers,
Bob
I'll try to demonstrate this pilot tone method later tonight. With the Viktors, it very easy to add the injection to the output as it's a dual mono wired 2xRCA output. Just need to cut the trace to one RCA and insert the 1Meg there, and feed it from the aux oscillator (soundcard output).
Here we go....
I wasted half an hour or so because the pilot level was way off at first (too high, also varying a bit with frequency, wtf?) but after looking closer on the board I saw it has 600Ohms output impedance (correct to the schematic that I should have checked first, meh), so I had to reduce the pilot send level accordingly. It also means the notch correction is off at higher frequencies, for the harmonic levels and the 5.5kHz pilot (by ~1.5dB).
Test level was 1Vrms, at this level this first generation Viktors still does pretty well. At 1V the notch distortion will be somewhat lower, too. Still enough uncertainity here, of course.
Looking at the osc with a faster real-time FFT (32k) showed a lot of fluctuation of H2 level and general noise floor jumping up and down (LM4562 popcorning)
I wasted half an hour or so because the pilot level was way off at first (too high, also varying a bit with frequency, wtf?) but after looking closer on the board I saw it has 600Ohms output impedance (correct to the schematic that I should have checked first, meh), so I had to reduce the pilot send level accordingly. It also means the notch correction is off at higher frequencies, for the harmonic levels and the 5.5kHz pilot (by ~1.5dB).
Test level was 1Vrms, at this level this first generation Viktors still does pretty well. At 1V the notch distortion will be somewhat lower, too. Still enough uncertainity here, of course.
Looking at the osc with a faster real-time FFT (32k) showed a lot of fluctuation of H2 level and general noise floor jumping up and down (LM4562 popcorning)
Attachments
Attachments
It certainly seems so. I've known of an issue with 1kHz in connection with USB2 and have seen it with another laptop.
Thank you for a useful information.
Regards,
Braca
With RME this is a non-issue. With other interfaces, less so, notably when they are not asynchronous isochronous (== not clock master), stuff like the old USB 1.0 PCM2702, or bus-powered devices (supply modulation).
In modern DACs packet noise most often is a layout issue (polluted GND)
EDIT:and even more often it's a cabling error in the test setup (unbalanced, "ground loop"), increasing the error at the receiving end (the analyzer) even when the analog output right at the DAC outputs itself is clean.
In modern DACs packet noise most often is a layout issue (polluted GND)
EDIT:and even more often it's a cabling error in the test setup (unbalanced, "ground loop"), increasing the error at the receiving end (the analyzer) even when the analog output right at the DAC outputs itself is clean.
Last edited:
