My ultra-low distortion measuring setup

[Helmut Sell]

Hello Pet rushi,

I always respect the performance of people who build, simulate and test circuits themselves, so logically also your approach. With this oscillator design, I don't see any reference voltage that ensures an exact output voltage of the oscillator. There is also no information about frequency stability.
When building Twin-T or Hall-Notch filters with OPAMPS, this is usually done to increase the Q of the filter. Then you have to keep the frequency of the oscillator and the filter very constant, otherwise there is no longer any damping of the fundamental oscillation and all further measurements are useless. Furthermore, the maximum attenuation of the filter usually decreases at resonance frequency.
I have set up all of the filters described here and tested them intensively (with an R&S UPL audio analyzer). The best attenuation at resonance frequency was achieved by a passive Twin-T Notch filter: -110dB. This design also has the advantage that frequency variations of the oscillator of +- 0.5% relative to the resonance frequency (995 Hz .. 1005 Hz) have no influence on the further measurement accuracy. However, you then have to correct the harmonics K2 and K3 etc., as they are damped by the factors 0.3511 and 0.5477.

 

[IVX]

IMO an R divider will not change the harmonics structure. You can check with R divider and LPF which attenuate at the same ratio (but rotate phases harmonics relatively). That ratio can be fine-tuned by changing the playback frequency.

Basically an LPF at about 1kHz will make H2 vs. H1 phase shift different compared to the R divider, then chosing the R values so that LPF attenuates approx same around 1kHz, then fine-tuning the frequency to make the LPF attenuation as close as possible to the (frequency-independent) R divider (because distortion depends on level too). Then you can check what the LPF does to the distortions of the combo DAC + ADC - only DAC distortions will be rotated by the LPF compared to the R-divider path.

Honestly, I don't get what about you talking, you can try 9039S silver sample by yourself, and I'll give you a 19% discount. 19 = Member Joined 2005, each year means 1% 😉

 

[IVX]

right, and 2+2 = 4, I know 😉

 

[jan.didden]

OK, so you agree that the mentioned 1.1uV wideband noise level is not the lower limit to measure harmonic levels.
Glad we cleared that up.

Jan

 

[phofman]

Honestly, I don't get what about you talking,

Let's say DAC outputs H2 and H3 distortions at -140dB phase 90°. That goes to an R-attenuator (divider) of -1.5dB. The ADC chain (the notch) distorts H2 and H3 at -140dB phase -90°. The measurement (vector sum of DAC + ADC H2, H3) will show plain zero distortions, eventhough both DAC and ADC distort H2, H3 at -140dB.

Now replace the R-attenuator with an RC LPF 1k + 100nF which also attenuates the 1kHz H1 at -1.5dB (just like the R-divider).

But now the LPF transfer function changes the DAC H2 from -140dB 90° to -142.6dB 57°, H3 from -140dB 90° to -145dB 28°. The ADC side distortions do not change, because H1 entering the ADC chain has the same frequency and level as in the measurements with the R-divider. However, now the measured vector sum or the DAC + ADC distortions is not zero, but -145dB -43° for H2 and -141dB -56° for H3. Quite a difference compared to zeros for the R-divider path.

If the DAC H2/H3 distortions were really zero, no change in the measured distortions would occur as all the distortions were attributed to the ADC.

I use this simple method for determining DAC/ADC distortion components of the measured vector sum in https://www.diyaudio.com/community/...on-compensation-for-measurement-setup.328871/ . The calculated values can be used for pre-distorting (distortion-compensating) the DAC and ADC by their respective share of the vector sum.

Fine-tuning the generator frequency is for setting the H1 attenuation of LPF (frequency-variable) exactly equal to the attenuation of the R-divider (frequency-fixed). The resultant frequency is used for the measurement (which does not have to be exactly 1kHz and is not anyway since your DAC and ADC devices have separate clocks).

you can try 9039S silver sample by yourself, and I'll give you a 19% discount. 19 = Member Joined 2005, each year means 1% 😉

I do appreciate your offer.

 

[Mark Tillotson]

DON”T use LM4562 unless in the buffer part of the oscillator. I found out that it picks up very easily mobile phone /wireless router signals nevertheless placed in a metal chassis.

One of its known failings (although TI don't want to know!), the other being some units have signidicantly higher noise than the datasheet. Doug Self recommended using a 100 ohm resistor in a unity gain follower using the LM4562 to reduce EMI pickup, IIRC.

 

[Pet rushi]

I wish I knew that when I got 25 LM4562 from TI a year ago.

 

[Pet rushi]

Hello Pet rushi,

I always respect the performance of people who build, simulate and test circuits themselves, so logically also your approach. With this oscillator design, I don't see any reference voltage that ensures an exact output voltage of the oscillator. There is also no information about frequency stability.

The oscillator that I built oscillates at 1001.2Hz. It takes around a couple of minutes to get there. It starts at 1002Hz or so and gradually goes down to 1001.2Hz. If the chassis is open that time may be different and the oscillation frequency is not very stable (+/-0.2Hz), most probably due to temperature variations. If the chassis is closed the oscillation frequency is very stable.

I had difficulties to get stable notch below -75dB - for both Twin-T (more stable) and and Hall-network. I suspect that the POTs that I use (not cheap ones) to adjust the notch frequency fluctuate. It is possible to adjust the notch to -90dB but next time I turn it on I need to readjust. So I preferred to use a couple of notch filters in series and not readjusting all the time.

If there is a trick to adjust the notch to -100dB and not readjusting I would be very thankful if someone let me know.

I can't completely agree that an active notch has the same depth as a passive one. I expected that this would be the case but my experiments showed different outcome. It is possible to get deeper notch by 6-12dB adjusting the Q of an active notch (especially Hall network notch).

 

[Pet rushi]

On the subject of the "deeper notch" when active configuration is used (continuation from the post #35).

The specs of the op amps used are crucial. For example, when using NE5534 the notch depth can be barely influenced. On the other hand, using OPA2211 or OPA1656 the notch depth can go down by 12dB. So the open loop gain is very important factor.

 

[Helmut Sell]

Hello Pet rushi,

The frequency stability of your oscillator is really very good. My oscillators drift by about 0.2% after an hour, i.e. from 1000Hz to 1002Hz. This is due to the negative temperature coefficient of the film capacitors I use. (MKP 1837 Vishay or Wima FKP2).
I used Bourns 3296W trimmers with 200R and 100R on my passive Twin-T Notch filters.All fixed resistors 0.1% tolerance and 25ppmTK. Capacitors Wima FKP2 or Vishay Roederstein MKP1837 or MKP1830 1% tolerance.The resistance values in the upper part must be exactly the same. Measure with a precise DMM. Then adjust the notch depth to the maximum value by varying the trimmer in the lower arm.

 

[Pet rushi]

For me Panasonic MKP worked out very well. I was very surprised to find out that TDK C0G are equally stable.
Contrary to the logic it looks like that smaller size caps don't drift so much.

 

[torgeirs]

Ultra-low distortion 1kHz signal generator (-130dB or better) - https://www.kaltecs.com/ultra-low-distortion-oscillator/

I have to agree with poster that reacted to the 140 db THD + N claim at kaltecs
Maybe 126 dB?
Then it can be used to measure Thd about 20 db lower with 1000 bin fft
But stating THD + N without bandwith is not serious. And leads the reader to asume the normal 20 to 20k

 

[torgeirs]

But of course the whole point is the super low THD of the signal.
But maybe only a analog sharp peaking filter is needed after a normal audio dac to achieve the same?

 

[torgeirs]

Dac with 1k tone -> sharp bandpass filter -> buffer -> DUT

 

[torgeirs]

Or maybe a difference amp for LR channel to remove second harmonics and analog 24 db LP filter is enough (if the dac is not hampered with bad output caps)

But don’t get me wrong. A oscillator has always it’s own worth😍

 

[jan.didden]

It is not a simple task to keep the oscillator and filter exactly aligned in the face of time and temperature drift of both the oscillator and the filter suppression peak. And the buffer may distort more than the source.
Things always seem so easy if you don't have to do it yourself 😎

Jan

 

[Moderation Team]

The thread starter has been banned. Several forum rules were broken. Some posts have been deleted, apologies if some innocent posts needed to be removed as collateral due to their quotes.

 

[Moderation Team]

This thread has been closed, however if a member feels it appropriate to take posts from this thread to begin another on the same topic, an exception will be made if you report this post.