EEVblog #1104 - Omicron Labs Bode 100 Teardown

[precaud]

Well this is interesting. Last week I had requested data from Picotest, specifying the response of their J2100A and J2101A 1:1 injection transformers with 50 Ohm source into 1 Ohm load.

They just sent me the response plot (S21) for the J2101A. I've attached it. It's quite impressive. -3dB at about 8MHz into 1 Ohm. That's an order of magnitude better than what I'm getting. Hmmm....

[T3sl4co1l]

To get leakage lower, keep doing more of it.  More pairs in parallel!

This reduces Zo overall, however, and with magnet wire, it's probably not a good match to the 50 ohm source anymore.

And to do it at 1 ohm, or so on, use a lower ratio, etc.

Tim

[Wolfgang]

Dont forget that cable impedance will be the limiting factor.
At 10MHz, 1cm of cable with an inductance of 10nH has an impedance of 0,628Ohms, a gross measurement error if you count with 1Ohm.
Current injection directly at the point of measurement or a VNA method (series thru) is probably smarter than an extreme injection transformer that cannot be brought close enough to the DUT because its big and bulky.

Some ideas:
https://electronicprojectsforfun.wordpress.com/power-supply-impedance-measurements-using-the-bode100-lf-vna/
https://electronicprojectsforfun.wordpress.com/power-supply-impedance-measurements-for-various-power-supplies/

[precaud]

Dont forget that cable impedance will be the limiting factor.
At 10MHz, 1cm of cable with an inductance of 10nH has an impedance of 0,628Ohms, a gross measurement error if you count with 1Ohm.

Yes, but:
 a) the upper-frequency limit of this setup is 1MHz. If I need to look higher than that, I can power the device down and use any number of other tools/techniques, without having to worry about the DC.
 b) the test leads only extend about 3 inches out from the transformer, and
 c) the current sense resistor is mounted right on the test clip.
 All of which is well within the ability of the OSL math to compensate for, well beyond the 1MHz target.

Current injection directly at the point of measurement or a VNA method (series thru) is probably smarter than an extreme injection transformer that cannot be brought close enough to the DUT because its big and bulky.

Less bulky than the boxes you have for your Shunt-Thru setup... 

Also, there are real advantages to xfmr-coupling the signal source when making low-level measurements... the infamous "braid error" problem is eliminated.

[Wolfgang]

Dont forget that cable impedance will be the limiting factor.
At 10MHz, 1cm of cable with an inductance of 10nH has an impedance of 0,628Ohms, a gross measurement error if you count with 1Ohm.

Yes, but:
 a) the upper-frequency limit of this setup is 1MHz. If I need to look higher than that, I can power the device down and use any number of other tools/techniques, without having to worry about the DC.
 b) the test leads only extend about 3 inches out from the transformer, and
 c) the current sense resistor is mounted right on the test clip.
 All of which is well within the ability of the OSL math to compensate for, well beyond the 1MHz target.

Current injection directly at the point of measurement or a VNA method (series thru) is probably smarter than an extreme injection transformer that cannot be brought close enough to the DUT because its big and bulky.

Less bulky than the boxes you have for your Shunt-Thru setup... 

A few *inches* of cable is even long for 1MHz. Anyway, it always depends on the accurracy you want.

... The bulky ones are for higher voltage / lower frequency. I have made really small solder-in parts for the fast jobs, but they are not yet on the webpage. Thanks for reminding me

[precaud]

A few *inches* of cable is even long for 1MHz. Anyway, it always depends on the accurracy you want.

But I am having no problem with accuracy... if I was, and was unable to solve it, I'd be posting questions about it !

I have experimented with Series-Thru (not good for low impedances), Series-R, Shunt-Thru, and Xfmr-V/I methods extensively for quite some time, and for this application (active in-circuit measurement), I prefer the latter overall.

... The bulky ones are for higher voltage / lower frequency. I have made really small solder-in parts for the fast jobs, but they are not yet on the webpage. Thanks for reminding me

May I suggest, you might want to make the 'lytics in your boxes nonpolar so you don't have to worry about supply polarity...

[Wolfgang]

A few *inches* of cable is even long for 1MHz. Anyway, it always depends on the accurracy you want.

But I am having no problem with accuracy... if I was, and was unable to solve it, I'd be posting questions about it !

I have experimented with Series-Thru (not good for low impedances), Series-R, Shunt-Thru, and Xfmr-V/I methods extensively for quite some time, and for this application (active in-circuit measurement), I prefer the latter overall.

... The bulky ones are for higher voltage / lower frequency. I have made really small solder-in parts for the fast jobs, but they are not yet on the webpage. Thanks for reminding me

May I suggest, you might want to make the 'lytics in your boxes nonpolar so you don't have to worry about supply polarity...

Good point, but I almost never use negative polarities, so I cut corners here. For one app where I needed it I made a special negative only variety. Drawback of unpolarized electrolytics is half the capacitance per volume.

[precaud]

Good point, but I almost never use negative polarities, so I cut corners here. For one app where I needed it I made a special negative only variety. Drawback of unpolarized electrolytics is half the capacitance per volume.

Alas, I do not have that luxury... almost everything I work on has bipolar supplies.

[Wolfgang]

What kind of supplies are you working on ?

[precaud]

Varies. But high percentage of linear.

Referring back to reply #250. I received confirmation yesterday from Picotest that the plot they sent me of their J2101A xfmr is indeed the mag portion of the transfer function (S21).

8MHz bandwidth into 1 Ohm is outrageously good performance for a 1:1 xfmr. It appears that, unlike the Omicron "NVT", this transformer has some real engineering in it. To get that BW, L_Leak and C_Winding would have to be an order of magnitude less than what we're getting on these VAC toroids.

Anyone care to speculate how they're pulling this off?

[Wolfgang]

Maybe a split between two transformers, one for HF and one for LF inside

[Jay_Diddy_B]

Hi,

Have you looked at the Picotest J2111A Current Injector?

Link: https://www.picotest.com/products_J2111A.html

It has a bandwidth of DC to 40 MHz.


I am just going to tease with a few photos:





These are current waveforms measured with a Tektronix TCP202 probe 50mA/div


Here is an application:




Measuring the impedance of ceramic capacitor to 40 MHz.

Jay_Diddy_B

[precaud]

Very cool. But way overkill for my needs. And pricey to boot...

[Jay_Diddy_B]

The pictures I posted are from my DIY version not the $1500.00 Picotest box 

Jay_Diddy_B

[precaud]

You don't really need the extra current to resolve the minima on those caps, do you?

I've been surprised by the number of times I've had to dial back the drive current to get a good Z reading when testing regulators on-board. Especially 78Lxx's in circuits pulling only a few mA from 'em. The drive current can modulate the output impedance!

[precaud]

Another data point regarding using these transformers into 1 Ohm loads. This shows the level response of various transformers, with 1VRMS input from 50 Ohms source. The good news is, any of them are able to drive 1 Ohm relatively flat from 10Hz to 1MHz with a useful amount of current for the measurement. So other criteria will (or can) determine which one to use.

Would still be interested if someone who made an "NVT" would post its S21 into 1 Ohm, for comparison.

[precaud]

Another interesting aspect of this is: the interaction between the transformer's nonlinear response vs load and freq, and the workings of the Open-Short-Load math.

In a nutshell, the OSL math works by assigning, at each measured frequency point, real & imag values that define the short and open conditions, plus one "known" impedance value. All measured impedances are then mapped or scaled linearly to those values.

For this to work, the transformer has to have a transfer response that is linear with load and frequency within the ranges you're wanting to use it. This is a tall order into low Z loads.

I decided to evaluate some xfmrs over my desired range of use: from 0 to 10 Ohms, and 10Hz to 1MHz. Other than xfmr saturation, there were no linearity problems at low freqs, so I evaluated them at four freqs (10kHz, 100kHz, 500kHz, and 1MHz) and seven resistances from 1 Ohm to 10 Ohms. (With a 1 Ohm current sense R, this corresponds to a measurement range from 0 to 9 Ohms). I converted the S21 to real/imag values and plotted them on a linear scale. This will give immediate visual indication of what is the best OSL "Load" resistor to use with that xfmr, and over what impedance and freq ranges the xfmr can be used with OSL compensation to give accurate results.

Since Tim's 3:1 xfmr with paralleled secondary gave the best HF performance, it was first. The plots shows S21 at the 4 freqs and 7 impedances for each, with a line from the 1 Ohm point to one of the upper-impedance points to highlight the linearity of the data points.

The first plot shows the 3:1 with linearity to 10 Ohms. Up to 100kHz, it is excellent. So used with a 9 Ohm Load R, this xfmr would deliver accurate results from 0 to 9 Ohms up to 100kHz.
But at 500kHz and 1MHz the OSL math would not correct the measurement accurately. The measured points are highly nonlinear. The 2nd plot shows that its linear range at high freqs is only to 2 Ohms. So this xfmr would give good results up to 1MHz with a 1 Ohm Load R measuring 0 to 1 Ohms, with increasing error at higher Z's.

This is a much narrower operating range than expected or hoped. It wipes out any advantage gained by the unit's better HF response.

[precaud]

Let's look at the xfmr that HP used to recommend for this test setup. A North Hills 0017CC 1:1 50 Ohm xfmr.
It's linearity to 10 Ohms is much better than than the home-rolled one. Not excellent, but probably good enough; at least the vector is pointing in the right direction  Greatest error would be between 100k and 500k Hz. But it is indeed useable over that Z range to 1MHz.
Drop the Z range and reference Load R to 4.7 Ohms and its better. And you can see that at 3.3 Ohms, it would be quite good.

[precaud]

And here's one that is even better yet. A North Hills 1111LA 1:1 75 Ohm xfmr. Who woulda thunk that a 75 Ohm xfmr would give better results at low Z's. (There are two versions of the 1111 currently being offered on eBay, either one would work fine. I have no affiliation with the sellers.)
https://www.ebay.com/itm/North-Hills-Wideband-TRANSFORMER-75-Ohm-20-HZ-NOS-w-Box-r6/302863195770
https://www.ebay.com/itm/North-Hills-1111LB-20Hz-6MHz-Wideband-Transformer-NEW/254149491185?epid=12021216572

The 1111's data linearity is better that the 0017 at 10 Ohms and 4.7 Ohms up to 1MHz. With 3.3 Ohms (the next dot down), it is about as good as you can expect to get from any xfmr.

Comparing these three xfmrs, what appears to make the difference is how gradual the S21 HF rolloff is. Higher-Q responses that maximize the flatness but cut off faster aren't as useful as a well-damped rolloff.

[BFX]

I did some research and here is the result
Any similarities is only by chance:D

[Wolfgang]

I did some research and here is the result
Any similarities is only by chance:D

(Attachment Link)
(Attachment Link)
(Attachment Link)

...looks too close to the original and has no nude virgins. Something for married people. 

[sixtimesseven]

The pictures I posted are from my DIY version not the $1500.00 Picotest box 

Jay_Diddy_B

Do you have a post for your DIY injector? 

[Jay_Diddy_B]

The pictures I posted are from my DIY version not the $1500.00 Picotest box 

Jay_Diddy_B

Do you have a post for your DIY injector? 

No   

At least not at the moment …

Regards,
Jay_Diddy_B

[ch_scr]

We have tested a different option and found that two huge chinese MnZn cores (search for "Cores Green 75mm x 39mm x 13mm") stacked on top of each other and afterwards wound in a similar fashion (not in the number of turns, but in terms of covering the core with windings, style of twisted wire and connection) to the Bode 100 transformer gives the attatched result. Seems like a lot of wideband transformer for about 10€ for the cores. The much increased volume of core material might also help with saturation?

[David]

 Apologies to break up the discussion on the injection transformer for a moment but could someone explain in a little more detail the architecture of the bode 100? Is this essentially a superhet performing a sweep of the frequencies via sweeping the LO to maintain the IF which is fed to the ADC? Is this the same technique used in FFT analysers? (E.g. the Stanford SR series). I didn't see a mixer or LO (assuming one of the DDS?) In the teardown?

Cheers,

Dave