Super Low Distortion Ultra Pure Audio Oscillators Revisited

Where we discuss new analog design ideas for Pro Audio and modern spins on vintage ones.
User avatar
mediatechnology
Posts: 5437
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by mediatechnology »

I use that site all the time but forgot to look there for a twin-T.

The twin-T notch filter I published earlier in this thread will likely be what I layout on the PCB.
It's from Jansek.

The component sensitivity hasn't been much of an issue largely due to the fact Mouser stock 10 nF/polypropylene in 1%.
With Mouser now having 200 piece minimums on 1% resistors ($4/200) it also becomes practical to hand-select values.

The notch depth varies somewhat with temperature and power on warm-up drift of Fc but it's not significant WRT A/D overload or FFT measurement of HD2-10.

I want two filter networks so I can do simultaneous Left/RIght measurements AND/OR (and that's a big "AND/OR") balanced and floating measurements.

Balanced and floating with a 40 dB INA and CMRR stage isn't that difficult but making it do both balanced and single-ended (dual channel) may be a real challenge.

Switching the tuning is one issue. Not only the number of contacts but the contact resistance, contact resistance effect on notch depth and introduction of distortion above -150 dBc as the contacts age.

If I make it do both balanced and SE the switching may have to be with jumper links or have three filters: Two SE and one balanced.
With three filters I can have two sets of input connectors and switch their outputs.

The notch filter will also need to have a switch for Generator Monitor and Input connector.
User avatar
terkio
Posts: 304
Joined: Thu Feb 07, 2019 3:17 am
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by terkio »

Thanks for all these informations.
I re read the Jansek's paper. Great work indeed.

10nF 1%. Excellent, the accuracy of the caps is an important point.
1% resistors is good, but no need for better, since there are trimmers. However, they better have a good TempCo ( The caps, too. ).

Switching: Jansek uses 4 jumpers. But you need more switching.
What about mercury wet contacts ?
What about a board, one can configure for a SE measurement session or a balanced measurement session, by means of jumper links.
Just, ideas, I do not see clearly all the switching you want.

I found answers to questions of my own. The key is in the two equations for the "Center rejection frequency" f0 given at
Twin T calculator http://sim.okawa-denshi.jp/en/TwinTCRkeisan.htm
Inspecting those two, one can see how these two f0, shift when trimming a resistor.
With one trimmer one can manage to have the two f0s equal. That makes a filter that perfectly rejects f0, but it is not likely the wanted frequency.
With two trimmers, one can manage to have the two f0s equal AND keyed on the oscillator frequency.
Furthemore, from these equations, I think, one can dig the trimming amplitude needed to correct a known component tolerance ( 1% caps ).
I will recheck my thinking, about these equations.
User avatar
mediatechnology
Posts: 5437
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by mediatechnology »

My reasoning behind selecting the 15K resistors to <1% was to reduce the overall error of that arm and therefor reduce the value of the trim.
With larger-value 15T trims the mechanical backlash in the adjustment is apparent as one attempts to adjust to a really deep null.

The depth of the null however is really not that critical as long as it has adjustment range to be tuned to the Fo of the oscillator.
User avatar
terkio
Posts: 304
Joined: Thu Feb 07, 2019 3:17 am
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by terkio »

I don't get it.
I think, what is conditioning the value of the trim is mostly the caps tolerance.

15T trims and higher is no good answer, I think.
Firstly, what's needed is as low a resistor value, that has a range good enough with reasonable margins. This is not easy, it requires good knowledge of the errors to correct.
I have seen too many trims that where way too much too stiff or too lean.

I knew there is a catch with mercury wetted contacts
A mercury-wetted reed relay is a form of reed relay that employs a mercury switch, in which the contacts are wetted with mercury. Mercury reduces the contact resistance and mitigates the associated voltage drop.
Ok, but what resistor value ?
Surface contamination may result in poor conductivity for low-current signals.
Who knows, how bad, that is ?
How long it takes, to contaminate, what material and atmosphere dependence ?
I have seen in the 60s a project of a DAC based on a R 2R ladder with relay contacts, I presume mercury wet contacts; And that was to go in the Concord navigation system.
User avatar
mediatechnology
Posts: 5437
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by mediatechnology »

I don't get it.
I think, what is conditioning the value of the trim is mostly the caps tolerance.
It's both the C and the R.
But the R is easier to match and that makes more trim range available for the delta-C.

15T trims are fine.
A super-deep null isn't needed.
I wouldn't over think this.
User avatar
terkio
Posts: 304
Joined: Thu Feb 07, 2019 3:17 am
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by terkio »

Selecting 15K resistors < 1% is the right way to go.
I was wrong.
Based on caps 1%, with resistor selecting < 1%, indeed, it does halves the trimming requirement.

Using the twin T calculator, I figured the f0 sensibility to the Resistors and Capacitors.
I finally found:
With +1% on all C and R, f0 goes -2%.
With +1% on the Cs and << 1% on the Rs, f0 goes -1,%

Now, what is the accuracy of the oscillator frequency ?
IRC a Wien network uses two resistors and two capacitors.
User avatar
mediatechnology
Posts: 5437
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by mediatechnology »

The oscillator network is almost identical to the twin-T.
(10 nF, 10+10nF, 15K9, 7K95)
One arm is 20 nF + 7K95. The feedback arm is 15K9||10nF.

I pulled mine to 996 Hz using 3X randomly-selected 15K8.
I think my C's are a tad high in value.

It uses the same values for both C and R.
The tolerance sensitivity should be the same.

I now have two Nacho's on the Protoboard so I can look at both outputs simultaneously.
I'm hitting +18 dBu balanced into 604Ω with HD2/HD3 just under the noise floor.
Haven't plugged it into the spreadsheet but it looks like they are under -140 dBc.

The problem with converting the "stereo" Nacho into a differential one is that Fc of the notch filter shifts up an octave if the values stay the same.
The common mode Fc of the "stereo" filter is 1 kHz; the differential mode Fc is twice that because the two complete twin-Ts are in series differentially.
It's starting to make more sense to just have three filters and switch their inputs/outputs.
I can scale the resistors 2X to use the same C values in the differential filter.
User avatar
mediatechnology
Posts: 5437
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by mediatechnology »

I built two Nacho filters and used them to monitor both output legs of the oscillator.
I then extracted the differential output of the notch filters using a THAT1240 to check the distortion in a balanced load.
(Looking at both legs individually, in common mode, doesn't tell you how the distortion actually sums in the balanced load.)

The notch filters were nulled to nearly-equal depths and were somewhat phase-matched.
Though the common mode rejection is poor because the notch filter is not a truly differential and floating, it's good enough to figure out how the balanced output measures.

This is +18 dBu into 604Ω.
The -100 dBu line is actually -140 dBu due to notch filter gain.
The FFT is 262 kpts with a 10 sample average.
The distortion is below the noise level.

Image


At +20 dBu into 604Ω HD2 and HD3 begin to peek out of the noise floor.

Image

With some distortion to actually measure I plugged the HD2 and HD3 values into the spreadsheet to find the overall THD.

Image

The THD at +20 dBu into 604Ω is 0.0000076%, -142 dBu or about 0.076 ppm or 76 parts per billion.

I think it's time to quit while I'm ahead... :lol:
User avatar
JR.
Posts: 3697
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by JR. »

yup stick a fork in it, and back off to only +18dBu

JR
Cancel the "cancel culture", do not support mob hatred.
User avatar
terkio
Posts: 304
Joined: Thu Feb 07, 2019 3:17 am
Contact:

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Post by terkio »

0.0000076% THD
Great result.


Trimming a Twin T to null its output.
How difficult, it is ?
How many iterations going around at the two trimmers, does it converge smoothly to the final result ?

Obtaining a stable phase looks awkward to me, because for a perfectly nulled filter the phase jumps from 90° to -90° instantly when crossing the fc.
A stable phase can only be obtained for a "near" nulled filter. So, this needs a subtle compromise between rejection depth and phase stability.

You should have noticed the two 1K trimmers have different sensibilities. There is a x2 difference in how hard they pull

I found that, in theory, for a x% tolerance on the capacitors and much less on the resistances ( All in the Twin T filter and the Wein bridge ) one needs 4x% on the trimmed resistors.
For x%=1% tolerance and 16K and 8K, that gives 640 Ohm and 320 Ohm trimmers.

In earlier posts, you mentioned "differential notch filter."
Is it:
Two twin T connected together at a virtual ground node.
Or
A twin H. That would be made of five 16K resistors and five 10nF capacitors.

The latter, seems to me, a true differential circuit, whereas the former is not. I checked with Spice that this differential "Twin H" does behave like the SE Twin T.
Post Reply