Dale CMF55-series Resistor Distortion

[mediatechnology]

Cross-posted from the Build thread for ULDO-Nacho:

I've assembled quite a few ULDO_Nacho boards using value-matched resistors for the 15KΩ in the Wien bridge and the 10KΩ in the inverter.

I've been installing Xicon or Yageo which appear to be identical and consistently get THD levels around -143 dBc or -139 dBc with a 600Ω load.

I decided to try the 0.1% Dale CMF55-series for the 909Ω, 10KΩ and 15KΩ.
Checking them with the 4-wire Ohmeter they were super close to 15K and very consistent within 2-3Ω.

Distortion with the CMF55 was at least 15 dB worse.
Don't use them. You'll regret it.
I replaced all 8 resistors with Yageo/Xicon and THD levels returned to normal.

It broke my heart to clip out about $10 worth of resistors but the 3 cent resistors working far better made up for it.

Lesson learned...

[flyboy71]

Good info! I have seen this same phenomenon in RF designs as well based on one brand over another affecting the noise floor when measuring phase noise in oscillators.

Many years ago I was tasked with evaluating SMD inductors when everything went lead free. Of all the brands Vishay-Dale were the only ones that we couldn't get solder to properly reflow. Many just fell off the boards or the end terminations broke. We actually talked to engineers at Vishay about it and they denied anyone else ever had the issue. They didn't seem to care about losing a lot of business over it.

[mediatechnology]

I posted this over at DIY Audio.

I now have some interesting comparative data.

I built a bridge from 4X 10KΩ resistors and drove the upper and lower arms of the bridge with ULDO at +26 dBu.
The center of the bridge was connected to the balanced input of the notch filter.
The post-filter gain is 40 dB with a CM rejection stage following the post-filter amp.
The CM voltage was +20 dBu; 7.75V RMS.
The test Yageo was value-matched within about 4 Ohms to the near-perfect value of the CMF55.

The lower two arms were 1984-era Dale RN55s that I had previously matched to 0.01% with a 4 wire Kelvin bridge.

In the following FFTS one or both of the upper bridge arms contained a test resistor.
Remember that the drive level is +26 dBu, attenuated by 6 dB in each arm, and that there is 40 dB gain.
A -60 dB indicated 1 kHz FFT level is actually about -120 dBu.
A 1Mpt flat-top FFT was averaged 10 times.

The first FFT is with 2X Yageo MFR-25 in the upper arms and 2X Dale RN55 in the lower:



As you can see in the above it appears "distortionless." If the distortion signatures of the upper arms' resistors match each other - and the lower two arms' signatures match there is cancellation. This doesn't mean any of the resistors are "perfect" it just means that the upper pair distort similarly to each other and the lower pair distort similarly to each other.

This FFT is with 4X 1984-era RN55s:



Again there is cancellation of distortion or the possibility perfection. It's a coin-toss as to which.

The following FFT is with 3X Dale RN-55 and a single Yageo MFR-25. One of the RN-55s is in the upper arm and we can now see a relative difference in the signature.



OK, so the Yageo distorts differently than the Dale RN-55s.

The final FFT may be the smoking gun. It has 3X Dale RN-55 and a single CMF55:



Relatively speaking the CMF55 has about 5 dB more HD3 and introduces HD2 compared to the Yageo. The insertion loss of the notch at HD2 is about 9 dB so the HD2 introduced is far worse than it appears.

This is a comparative test but may well explain the increase in HD3 when I use the CMF55 in the oscillator.

UPS arrives in a few minutes and I'll test the Holsworthy and Vishay MBB0207.

Later that day...

UPS came bearing resistors from Holsworthy and Vishay.

The FFT results for 3X RN55 and 1X Holsworthy Y1RB:



A near-perfect match to the Dale RN55s and perhaps all are near-perfect or again maybe are equally bad LOL.
Note that the Holsworthy isn't introducing the HD3 compared to the RN55s like the Yageo MFR-25 does.

And the runner-up is the Vishay MBB0207:



Again we have a near-perfect match to the Dale RN55s.

It's starting to look like the CMF55 is the culprit. It's added HD2 and additional HD3 compared to the Yageo is elevating HD3 in the oscillator.

To bust -140 dBc I think I need to install either the Holsworthy Y1RB or the Vishay MBB0207.

[emrr]

Wow, seriously into the dark valley we go. Very interesting.

[AnalogJoe]

Very nice work.

[SandyTodorov]

Give the exact scheme by which you measured the THD of the resistances.
There are many pitfalls.
Where is the evidence that everything else, the generator, the filter and the amplifier after the filter, have at least 10 times less THD than the resistors.
The THD of a non-inverting amplifier or buffer at high impedance of the signal source (filter) is too high.
OPA1612 at 100 ohms on the generator and gain +1 has THD -150dB but at 10k generator output already has THD 122dB
And at gain -1 THD is already -177dB

The sum of the THD of two 0.001% THD devices is not 0.002%, but can be anything from 0.00000% to 0.002% depending on the phases of the harmonics.

[mediatechnology]

Give the exact scheme by which you measured the THD of the resistances.
There are many pitfalls.
Where is the evidence that everything else, the generator, the filter and the amplifier after the filter, have at least 10 times less THD than the resistors.
The THD of a non-inverting amplifier or buffer at high impedance of the signal source (filter) is too high.
OPA1612 at 100 ohms on the generator and gain +1 has THD -150dB but at 10k generator output already has THD 122dB
And at gain -1 THD is already -177dB

The sum of the THD of two 0.001% THD devices is not 0.002%, but can be anything from 0.00000% to 0.002% depending on the phases of the harmonics.

I believe I already explained that here in post #3.

Also follow along here:
https://www.diyaudio.com/community/thre ... st-7665489
https://www.diyaudio.com/community/thre ... st-7665544
https://www.diyaudio.com/community/thre ... st-7667295

Assume four perfectly-matched "distortionless" resistors connected in a bridge.
The top and bottom of the bridge are driven by a balanced ground-referred generator having a low -140 dBc typical THD.
The center of the bridge is sampled by an instrumentation amp with +40 dB gain followed by a cross-coupled pair of THAT1246 to realize common mode rejection.
The output of the CM stage is fed to a balanced input A/D to perform an FFT.

The generator fundamental of 1 kHz - along with its distortion - appear in common mode and are rejected.
My FFTs show about -60 dB rejection which relative to the +20 dBu common mode signal and 40 dB gain is about -120 dB.
About half of that CMRR is due to the bridge itself.

The generator does not have to have 10X lower distortion than the resistors using a bridge because this is a null test.

The notch filter is a floating double twin T in this setup and is actually not needed because it filters the differential 1 kHz which is actually quite low due to R match. It has no effect on the common mode signal other than slightly reducing it due to bias resistor tolerance.
In a future test jig I will eliminate it and just use a THAT1510 with 40 dB gain.

The balanced INA/CMRR stage does introduce distortion but what we're looking at are relative changes when different resistors are introduced.
As it turns out the INA/CMR stage is actually quite good relative to the bridge imbalance and distortion introduced by one "bad" resistor.

By substituting different manufacturers resistors (all 0.1%) in the bridge one can easily determine who the bad actor is by voting.
If 5 different brands/types resistors all measure good and one doesn't then its quite likely the one that measures bad is bad.

Now one could argue that 4 bad resistors in the bridge would measure great. It didn't.

This test confirmed why distortion in the oscillator rose 15 dB when the Dale CMF-55s were used and why it fell slightly when Holsworthy Y1RBs were used instead of Yageo MFRs.

FWIW the 1984-era CMF-55/RN55 resistors I have are super clean and can serve as reference resistors.
1audio at DIYAudio has a CLT-1 component tester with resolution down to -170 dBc.
I've sent him the resistors I tested and we'll see how well his measurements correllate.

See next post.

[mediatechnology]

My first series of tests on 15KΩ resistors used the 1984 CMF-55 10K matched to 0.01% in the bottom leg. I saw a little HD2 increase with all DUT so I changed strategy.

The bridge is driven by a ground-referred balanced output and and is recovered by a floating-input balanced "double-twin-T" (quad-T?) notch filter followed by a 5532 INA and cross-coupled THA1246 for CM rejection. The 1 kHz - and generator distortion - appear in common mode. In this situation the notch filter doesn't do much to reject the 1 Khz from the generator as most of the heavy-lifting is done by the CM stage.

When the bridge was all 10K the CM voltage pivoted around 0V. When the bridge was used with 15K in the top arms and 10K in the bottom the CM range shifted which gave rise to small amounts of HD2 in the INA. So I switched to an all 15K bridge and used the MBB0202 and YR1B as the reference resistors.

I couldn't find a combination of MBB/Y1RB that produced distortion. In this test there are 2X MBB and 2X Y1RB with the Vishay pair in diagonal legs.



No matter how I mixed and matched MBB0207 and Y1RB the results were consistent.

When a modern CMF-55 is introduced into the bridge things turn south.



Not good.

I also decided to test the theory that 4X "bad" resistors in the bridge would test great. This is 4X modern CMF-55.



They don't. The FFT looks like a pine forest.

If I were building a resistor tester using this method I think I'd make the bridge amplifier a THAT1510 or INA217 with the bridge connected directly to the bases and program the gain to be about 40 dB. With the THAT1510 input-referred CMRR would be about 90 dB. The INA217, which seems to have trimmed CM resistors it might be about 20 dB better. The bridge would supply bias current from the generator output or extra Rs at the tippy-top and bottom of the bridge pulled down to ground. That way CM rejection isn't affected by added bias Rs.

[mediatechnology]

My initial tests didn't use a 1646 or 1510 but conceptually it is similar to this:



As I mentioned earlier when RA=RB=RC=RD with any degree of precision the 1 kHz excitation and it's harmonics will be attenuated when sampled differentially across the bridge arms because they appear in common mode.
So the distortion in the source, if reasonably low, doesn't contaminate the measurement because it's a null test.

Distortion in the 1510 and the output coupling cap are another matter but we're looking at relative changes in FFT signature and don't seek to actually quantify the absolute distortion level of the DUT resistor(s). We're just looking for bad actors that stand out among a larger number of good actors.

When RA=RB=RC=RD the CM voltage is low but when 15K was installed in the upper arms and 10K "reference" resistors in the lower the CM voltage increased and an HD2 spur popped up.
This indeed confirms that distortion produced in the differential amp can, and did, contaminate the measurement.
Making all four test resistors 15K not only reduced the CM voltage but balances the source impedances which also contributes to lowered diff amp distortion.
HD2 from the diff amp disappeared back into the noise floor.

In the above circuit DC offset is allowed to accrue all the way to the output coupling cap.
Although it might be 2-3 VDC it does not restrict dynamic range since the output signal level is super low.

As long as CM distortion doesn't contaminate the measurement (when RA=RB but doesn't equal RC=RD) I'm confident the 1510 and it's output cap can amplify sub-millivolt signals with pretty good linearity.
If a 5532 can do it I'm confident a 1510 can as well.

[SandyTodorov]

The idea of ​​using a balanced output for the bridge is a good one, so the CMRR requirement of the amplifier after the bridge is lower.
But its THD requirements are high It is better to have at least 3 amplifiers one buffer and after it two with +20dB gain (OPA1612)
Checking with the bridge shorted is a must and should show at least 5-10 times lower harmonic levels than without it, just be sure that the THD measurement is caused by the resistors.
If we do not have reference 3 resistors with 10 times less THD from measurements, we simply put 10 pieces in series of the same type 10 times smaller actu value, they are guaranteed to distort 10 times less.