Page 11 of 17

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Fri Jan 15, 2021 8:06 am
by mediatechnology
I'm still finishing up "Nacho's" output stage but wanted to post this circuit fragment of the balanced/single-ended version.
When J1-J3 are open the filter is a fully-balanced "double-twin-T" followed by an instrumentation amp.
J1-J3, when linked, change Nacho from a fully-balanced to single-ended Left/Right configuration.
If the single-ended configuration isn't needed, C4-C6 and a pair of trims, VR3/4 could be eliminated.

Image
Nacho 1kHz Notch Filter Circuit Fragment

I'm not yet keen on the idea of introducing electrolytic capacitors in the measurement path but have shown locations for a couple in the +40 dB post-filter stage.

What I've done thus far is let DC offset at the input be subject to DC gain and then AC-couple, with a film cap, the outputs.
The reduction in headroom from allowing offsets to develop is tolerable considering that the rejected fundamental and distortion signal levels are very low, <<-40 dBc.
With 40 dB DC gain and assuming a +/-8V output maximum the inputs will accept up to 80 mV offset before output saturation.

In the unlikely event I ever needed unity DC gain I added locations for two caps in the gain leg.

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Fri Jan 15, 2021 3:11 pm
by mediatechnology
I decided to AC-couple the input to the instrumentation amp input.
I stock 0.33µF polypropylene and tested them in this location.
They didn't add any measureable distortion or change the signature in any way and are working at a very,very low level since they're after the notch.

Image
Nacho 1kHz Notch Filter Circuit

I should point out that the cross-coupled THAT1246, when fed into a balanced input downstream, provide very high common mode rejection.

See: "Composite INA Extends CMRR Frequency Range 10X" https://proaudiodesignforum.com/forum/p ... f=12&t=701

I don't really need high CMRR but what I do need is maximum noise rejection from the output of the notch filter into the A/D to find those distortion needles in the haystack.
The easiest way to get the anti-polarity output is cross-connection of a second diff amp.

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Sun Jan 17, 2021 3:53 pm
by mediatechnology
mediatechnology wrote: Fri Jan 15, 2021 3:11 pm I decided to AC-couple the input to the instrumentation amp input.
I stock 0.33µF polypropylene and tested them in this location.
They didn't add any measureable distortion or change the signature in any way and are working at a very,very low level since they're after the notch.

Image
Nacho 1kHz Notch Filter Circuit

I should point out that the cross-coupled THAT1246, when fed into a balanced input downstream, provide very high common mode rejection.

See: "Composite INA Extends CMRR Frequency Range 10X" https://proaudiodesignforum.com/forum/p ... f=12&t=701

I don't really need high CMRR but what I do need is maximum noise rejection from the output of the notch filter into the A/D to find those distortion needles in the haystack.
The easiest way to get the anti-polarity output is cross-connection of a second diff amp.
Before I committed Nacho to a layout I wanted to see if the addition of C9/C10 and C13/C14 had any effect on the measurement.

I used a Wima FKP2 polypropylene for the 0.33 µF C9/C10 because I stock them.
This gives a differential cutoff of about 4.8 Hz with 100KΩ bias resistors.
The reason I want to maintain a low cutoff frequency is that I may want to make a 100 Hz (or lower) notch filter and oscillator.

Then there's C13/C14 to AC-couple and remove offset from differential DC gain or single-ended DC gain when used Left/RIght.
I stock 1 µF polypropylene for the Phono Transfer System with 25.4 mm lead spacing.
They're huge.

C13/C14 at 1 µF provides a 26 Hz Fc.
The THAT1246 diff Zin is 24KΩ. When cross-coupled it's half that at 12KΩ.

I tried a smaller 10 mm 2.2µF Wima polyester and it seemed to not change the oscillator's measured FFT signature.
I may see if I can provide pads for 5, 10 and 25.4 mm lead spacing for C13/C14.
At the end of the day a physically-smaller larger-value polyester may work fine here or the cap could just become a jumper.

The Focusrite 2i2 has 30KΩ/leg input impedance so when connected to the Left/Right outputs the THAT1246 inputs are in parallel with the A/D if they're populated on the PCB.
The THAT1246 CM Zin is 18KΩ. With a second parallel-connected THAT1246 the CM Zin is half that at 9KΩ.
The total load with the A/D connected is just under 7KΩ putting Fc, for single-ended outputs, at 23 Hz with 1 µF caps.

I wanted to avoid introducing any electrolytics in the path and was reluctant to add film blocking caps.
But AC-coupling has so many measurement advantages its worth doing.
Now Nacho can measure in the presence of Phantom power.

I think I'm ready to stick a fork in Nacho too...

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Wed Jan 20, 2021 1:46 pm
by terkio
I have done more work about the two ways to build a balanced notch filter. This is about J1 and J2. What is best two jumpers or one jumper.
I had thought a twin H was better rather than a double Twin T. It looks good and can be built with all equal capacitors. Then I realized that this was not a rational choice. So, I worked more on this puzzling question.
Building with ideal components, equal value capacitors and resistors C, C, 2C, R, R, R/2 the two ways make no difference at all. To reject 1KHz one needs C=10nF, R=15.915494K.
The proof is from the symmetries of the twin H circuit, the two virtual grounds must have the same voltage.
With real components, the symmetries do
not exist anymore, after trimming they get close but not perfect. So, I tackled this issue with simulations where I deliberately skew one capacitor 10%, to see what happens on the notch depth.
On six simulations, I kept the upper twin T ideal. I skewed the lower twin T using 11nF then 9nF at the left cap, then 22nF then 18nF at the central cap, then 11nF then 9nF at the right cap.
All six simulations give a deeper notch with the double twin T. The double H notch is not so good.
The notch depth are typically -95dB versus -80dB when skewing a 10nF and -95dB versus -75dB when skewing a 20nF.
I can give all the details, the values of trimmed resistors for instance.
I was not expecting such a consistent result over a set of cases that I think well covers the not so ideal circuits.
May be this is splitting hair, with capacitors 1% may be there is no significant difference.
I think there is no need for J1 and J2, only one is fine, a tad better assuming notch depth is the feature that matters.

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Wed Jan 20, 2021 3:04 pm
by mediatechnology
Thanks for doing the sims.
My instincts told me that that J1/J2 could be combined but then I started thinking about it and talked myself out of it. :roll:

Another consideration is calibration.

Nacho needs to be tweaked for the single-ended configuration (links closed) and then, when opened for balanced, the depth is what it is.
The chips fall where they may: As drawn there's no optimizing it for both.

As a practical matter the notch depth isn't that critical since its' whole point is preventing converter overload.
For a +26 dBu input a 30 dB notch has a -4 dBu output.
That's 12 dB of headroom for most converters.
I'm thinking one nulled optimally for unbalanced will null at least 40 dB when switched to balanced.

If I were to combine J1/J2 I could put Balanced/Unbalanced on a DPST switch.
If I were doing it with a relay I think I'd do it a little differently to reduce crosstalk.

To keep the input/output switching down I think two sets of jacks (3 In/3 Out) is simplest and avoids a lot of contacts.
I may use 4 RCA for the unbalanced and two TRS for the balanced connections.
Since its a test instrument I don't mind re-patching the input and output.

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Wed Jan 20, 2021 4:47 pm
by terkio
I was amazed to see very good notch depth despite a +/- 10% capacitor.
Thanks to the two trimmers adjusted in SE mode to accurately get Fc 1KHz, I get -96dB notch depth on the 6 cases I simulated.
The trimmed 15.9k range is: 14.5k to 17.7k.
The trimmed 15.9k/2 range is: 7.4k to 8.6k.
So, I think you do not need accurate costly capacitors, trimming will fix this like wonder. Very good TempCo seems more important than accuracy.

I understand the trimming is done SE for the two channels, then you switch to balanced and there is no need for more trimming.
In case you make a version dedicated for balanced measurements with 5 x 10nF capacitors and 5 resistors. How will the balanced filter be trimmed ?
It is clear the central resistor needs one trimmer ( instead of the two trimmers of the two channels SE ). I do not see clearly about the two other trimmers.
Are they redundant, so one of them can be suppressed ? Nulling the filter with 2 trimmers is likely easy.
Are they both needed ? Then nulling the filter with 3 trimmers seems a challenge.

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Wed Jan 20, 2021 7:24 pm
by mediatechnology
Thanks for running the sims.
It is amazing that loose tolerance caps can produce deep notches.

Mouser stock the 10 nF polypropylene in 1% and the price is pretty reasonable. https://www.mouser.com/ProductDetail/505-FKP20.01-63-1/
ULDO and Nacho combined use 11 so the total cost would be modest around $18.

I use the 10 nF 1% in the RIAA EQ for the PTS and will be using them for the oscillator so I might as well use them in Nacho.
Being the same type as in the oscillator their tempcos should match.

I haven't found any other 1% values Mouser stock so its good to know 5% or even 10% can be made to work particularly for 100 nF and 1 nF that would be used for 100 Hz and 10 kHz.

Bateman in his article (linked to in post #1) adds a trim in series with the Wien bridge's capacitor in the feedback loop of the oscillator.
He says he got the idea from the AD797 datasheet to reduce distortion.
What I think his trim does is peak the frequency of the bridge to correct for component tolerances which in turn reduces distortion.

For a balanced-only build using 5 caps that little voice in my head tells me that I probably ought to use 3 trims to benefit common mode rejection since the bias resistors, R13 and R18, are now on the right-hand side in series with the trims and fixed resistors.
With even 4 trims it's not hard to make it converge in the balanced configuration.
Obviously two trims in the middle leg of the T are redundant so one would become a jumper.

Image

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Thu Jan 21, 2021 12:59 pm
by Gold
terkio wrote: Wed Jan 20, 2021 4:47 pm So, I think you do not need accurate costly capacitors, trimming will fix this like wonder. Very good TempCo seems more important than accuracy.
What about a C0G ceramic? They have very stable TempCo and are very linear.

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Thu Jan 21, 2021 8:20 pm
by terkio
Yes.

A link given by mediatechnology in a previous post.
http://www.janascard.cz/PDF/An%20ultra% ... 0%20dB.pdf
Very informative, a key source.
With:
Resistors are preferably 0.1%, 15 ppm/K through-hole types, capacitors are preferably polystyrene foil types, second best choice are NPO (Negative Positive Zero) with their smaller size and lower temperature coefficients, but slightly higher distortion

Re: Super Low Distortion Ultra Pure Audio Oscillators Revisited

Posted: Fri Jan 22, 2021 7:16 am
by mediatechnology
In through-hole larger-values (e.g. 100nF) COG become insanely expensive.
A 581-SR505A104JAR 0.1/50V/5% is $7.28. (Mouser)
At >6800 pF is where they get really expensive.
SMT seems to support larger values in COG but its not worth the hassle and I haven't checked pricing.

The Wima FKP2, which is what I'm using, is the only series I find where Wima even specs a tempco.
The FKP2 is -200 ppm/°C.
Since the oscillator and Nacho use the same value they track together.
I haven't found the FKP2 with it's 200 ppm tempco to really have much effect on frequency stability and only a minor amount on notch depth which is not the be-all end-all.

What I found more interesting is resistor tempco and self-heating.
For the output build-outs it seemed that I was able to lower what tiny distortion was left by using the Vishay PR01 1W metal film power resistors.

The inexpensive Xicon 1/4W 1% are spec'd at 50 ppm/°C and work fine in every low signal level location.
The build-outs see some power.
The PR01 are 200 ppm/°K (200 ppm/°C) but their thermal mass is a lot higher so their delta-R under load are less.
Whether this is necessary or not I'll figure out once I get PC boards.
On the Protoboard the PR01's seemed a tad cleaner.