impedance measurement with VNA using series, shunt/series through methods, graph

[JohnG]

For what it is worth, accurately measuring a 1 mohm resistance is challenging enough at DC, not to mention 0.1 mohm. I would be really surprised if the contact resistance was repeatable within 0.1 mohm.

Something else to consider is that the resistance an RF cable is probably several mohm, perhaps higher. Variation in temperature or connector pressure, etc. can give you substantial errors at this level. It may be that the initial accuracy of any calibration combined with the limited precision of the error correction model is insufficient.

Finally, most low-ohm resistors have a fairly large cross section, and skin effect can make the effective resistance higher. It doesn't really help if they are thin-film unless a lot of care is paid to the shape of the resistive region and the current return. A chip resistor will experience current crowding at it's edges, not just the faces. Even worse is that many low-ohm alloys are partially magnetic, which makes skin depth issues even worse.

Cheers,
John

[coppercone2]

for what its worth, if you use deoxit on a strip of copper, and test leads, it does increase repeatability when measured with real gold kelvin leads, for some reason, on a source meter

take 15 measurements and measure variance if you don't believe me. i used IET brand kelvin leads (not cheapo) and a keithley SM that was calibrated from a fluke multifunction calibrator at a modest current like 100mA, the source meter was powered up for 24 hrs before measurement

they are not lying

its also still not very stable

the usual i heard is "omfg its gold"

[joeqsmith]

I would never apply anything to an RF connector that was not what the manufacture recommends.  Now, if you found me an article published by Keysight, Copper Mountain Tech, Gore or manufactures of high quality RF connectors where they call it out as part of the maintenance and care, I would read it.   

[joeqsmith]

For what it is worth, accurately measuring a 1 mohm resistance is challenging enough at DC, not to mention 0.1 mohm. I would be really surprised if the contact resistance was repeatable within 0.1 mohm.

Something else to consider is that the resistance an RF cable is probably several mohm, perhaps higher. Variation in temperature or connector pressure, etc. can give you substantial errors at this level. It may be that the initial accuracy of any calibration combined with the limited precision of the error correction model is insufficient.

Finally, most low-ohm resistors have a fairly large cross section, and skin effect can make the effective resistance higher. It doesn't really help if they are thin-film unless a lot of care is paid to the shape of the resistive region and the current return. A chip resistor will experience current crowding at it's edges, not just the faces. Even worse is that many low-ohm alloys are partially magnetic, which makes skin depth issues even worse.

Cheers,
John

So far, at least for DC using the current source, it has been repeatable well within 1uOhm.   I can't answer for absolute accuracy.   Once I have the DCR sorted, I'll try and sort out the other main source of error.   Again, just an interesting experiment.   

************
Doing a search, I came up with the following thread:

https://www.eevblog.com/forum/testgear/measuring-low-impedances-with-a-vna/

Attached, showing the modification to the test boards.   At DC I measured the following:

1mOhm
Measured 0.9870A @ 1.0120mV or 1.025mOhms.

100uOhm
Measured 1.8929A @ 189.0uV or 99.85uhms.

Also shown is the last transformer.  The test setup was very similar to what virtualparticles shows.   Based on the data they provided, I suspect I am using the same coax, but more of it, more turns, and a mix of core materials.  The downside is I give up the higher frequency.   I plan to built two more to try and match the Picotest's transformers performance.  One using the hybrid, the other using a single type but I suspect more in-line with what they have based on the data we have been provided.     

[joeqsmith]

Showing results using the two smaller loads. 

[coppercone2]

I am guessing that using the 4 terminal SMD resistor in the same way would have the same results and its not worth doing for the price tag because its clear that the problem was resolved good enough, maybe there would be a very slight improvement, but it would restrict your PCB only for the weird resistors making it not very useful outside of a dedicated calibrator

what is the accuracy of the resistor? at 0.15, the error is very low, assuming you used a high accuracy resistor.

we probobly don't hear about splitting the trace under a normal resistor because its prone to manufacturing errors so no one bothered to write about a 'hack' like this despite the giant error reduction, since mass production would be upset, you probobly need alignment during reflow since that will no longer self align.

I am still waiting on some of the brass tubes and hardline to make a probe that has the plunger inside of the coaxial cable sticking out only a little. I have to solder the plunger into a thin brass tube that I got because its far too thin and slightly enlarge the hardline to fit that tube in the hardline, the calculations show I should have like slightly less then 50 ohms in that transmission segment (I get something like 43 ohms for around 1.5 inches). I will make the same fixture as you but with spring loaded pins rather then press fit for the center conductor and see how that works out despite the slight changing of the dielectric shell.

[joeqsmith]

All four resistors used are 1% parts.   From my previous post, measuring the 1.0 and 0.1 mOhm resistors with the two old HP bench meters with unknown accuracy I would guess we are well within 5%.   

Noise floor of the SA vs the VNA.

[joeqsmith]

Looking at S21 for various decoupling test boards.   Notice the three types of capacitors on this particular board.

With the V2 Plus working well into the GHz,  it may be useful to have some sort of power distribution graphs. 

[joeqsmith]

Basic example of PDN design talking about effects on phase noise.


Nice lab, but I don't see the Nano anywhere.. 

[coppercone2]

if you want to put a 3mm into a lmr400 cable with the center conductor removed, its a snug fit, and a non jacketed but still insulated cable makes a very light fit into a 7.2mm tube.

you need to widdle down the center conductor of the lmr400 cable to get it to solder into a brass tube however

[coppercone2]

So the problem is getting a joint between a 3mm brass tube filled with solder and a 2.7mm copper clad aluminum conductor cut, then gluing a peice of sleeve over it

I am going to try LF460 aluminum solder and flux first off. I think you need a lathe to make your own non open brass tube because I see nothing but problems here. If LMR400 was solid copper, there would be 0 problems. Now it looks like I need a miracle to solder this together or I need to add alot of series connectors

I am thinking to get CNT-400 cable, which has solid copper, unlike the modified LMR, supposedly. try making heads and tails out of this mess, looks like the wild west

edit: managed to tin the aluminum copper with the 460 aluminum solder and flux then bonded it with normal solder to the brass tube filled with solder, so its a 5 metal junction, its probobly pretty weak, but it seems strong enough for the application.

[joeqsmith]

Hi JoeQ

Measured from connector ground to connector ground. I used a pair of pigtails and a copper plate as a low inductance common ground as you can see in the picture. I calibrated to the end of the test cables, attached the cheezy pigtails and then used port extensions to get to the end of them. The attached S-Parameter file should be what you want.

Best,

BW

From Picotest:   

"We use a PCB FIXTURE to short the BODE 100 OUTPUT (center pin only) to both pins at the J2102B input.  Then we connect one receiver to the J2102 input and the other receiver to the J2102 output center pin."

Which makes sense but I wouldn't expect that to cause any difference compared with how you collected the data.   If I run a similar configuration on transformer #8 where the shield and center pin on the input side are shorted and then measure thru the center pin,  the results are the same as looking across the shield.  At 100kHz, it looks like your core measures around -24dB compared with transformer 8 at -37dB, and -90dB with the Picotest graph.   At 1MHz, your at -32dB compared with -47dB for #8 and  -110dB for Picotest's graph.    Again, this transformer would be very poor at higher frequencies but the plan is to construct three others that will attempt to get something closer to the data you have provided.   

Also, their response didn't address the gain shown in their plot.  I have written them back so hopefully we can get to the what is the cause of the discrepancy. 

Also shown is the transformer #9 which is also a hybrid.  It measures about 22dB dB at 100kHz and -32dB at 1MHz.   The old HP is limited to 150MHz.  It's obviously fairly flat in this region.   Transformer #8 also works well at these lower frequencies.   For fun, I stuck it across the NanoVNA.  I wouldn't put a lot of stock in the data above 300MHz.   Once I have the other transformers built, I will take the time to measure them and we can overlay the data with what you provided.   

[coppercone2]

I got a copper bar stock and silver plate so I can braze into a U shape and drill holes and add set screws and solder on spring couplers to make my test fixture

I am gonna make a U shape and then solder oversized silver plated berylium copper springs around the drill holes so when you insert the silver plated brass rod into the thing, the end of it contacts a the spring and makes a good coupling to the U shape body so its not going through set screws. I got tellurium copper for that so its easier to work with. I will cut it into 3 sections and silver braze it together.

it should beat the toolmakers vise I am using now.

I am just not sure how to get the drill holes right, the tubing measures differently in different points. I am not sure how much it will wedge so I might make sample segments and try to fit it in an aluminum block to see what snug is.. I wonder if I need to buy a rotary lap and a precision reamer.

[joeqsmith]

I was thinking something like this.   Copper clamps are soldered together at the spring end.  They would provide more than enough force to clamp an SMD in place. 

[JohnG]

First, a thanks to those pursuing this and showing the results. I did not think that it would make sense to do these measurements down to the single digit milliohms, but given the discussion in this thread, I am motivated to give it a shot.

Second, those are beautiful fixtures. I just got a used Sherline lathe with milling attachment, so they would be something to learn on. There is one gotcha, though. They can overestimate the inductance of a part that would normally be mounted over a ground plane. This will become significant as the width of the chip component becomes larger than the spacing between the surface layer of the PCB and the ground plane, at least for thin chip components.

Third, the solution to getting a wide frequency range CM-blocking inductor should be as simple as putting ones with smaller ranges in series. If the frequency range overlap is small, you should increase the total frequency range of CM rejection without affecting the differential mode of the 50 ohm line. The penalty will be an increase in total size and cable length. You could stack different cores if you wanted to lower the total winding length, but that will most likely hurt the high end rejection.

Cheers,
John

[coppercone2]

the pogo pins have replaceable contacts (they are a sleeve and a pin) so thats the only reason why i gravitate to them, plus it probably slightly hardened and gold plated.

If you do use pins, be aware my half finished one in the vise is a pain in the ass to load, since I do not have a external big spring, I need to position the part in tweezers then use another pair of tweezers to separate the pogo pin and slowly push it back into place. It is annoying and a few parts were shot away from me. I thought about making a little pedestal out of some material to rest the part on on top of a screw to adjust the height  but I am not sure what the screw should be made of, and it seems unnecessary unless i widen it alot to fit a big inductor or something.

A big external spring mechanism is superior so long the RF parameters are OK. I don't have the tools to make a nice joint (I do have a brass surface grinder sharpener fixture that I will likely never have a surface grinder for,which is almost perfect to modify because it has a smooth travel in a long piece of brass, but i also thought to just make an attachment to use that to dress my bench grinding wheels so its probobly better to leave that tool alone). Since I got some beryllium spring copper i will experiment with making springs, I don't have a knack for figuring out what kind of spring I need to get the mechanics right or the tools to make it operate smoothly. Maybe I will try to make a little one built like a watchmakers vise on my proxxon mini mill or modify a watchmakers vise (the one for holding the bowl part).


My current fixture has a discontinuity in the dielectric, I cut it as flush as I could with a razor to expose a piece of center conductor so I can solder the pin holder in the brass tube to it, then slid another piece of dielectric on and soldered the brass tube on top. However if you have the N connectors you can probobly solder the center conductor to the coaxial connector pin and the pogo extension and slide a single piece of dielectric back over that and solder the N connector over that single piece. I am not sure what that thin discontinuity does from razor cut, a TDR would tell me nicely, but I have a feeling it does not matter much at only 300MHz max. I have materials to try to make non interrupted ones if I get access to higher speeds but its already pretty interesting.

I got the aluminum solder to work alot better by using it alone for the joint instead of pretinning, to use my regular soldering iron with the special acid flux and clean it afterwards with mild abrasives and solder it like a normal solder joint, and to really scrub the aluminum solder down with stainless steel wool in soap water prior to soldering and cleaning with alcohol. It is actually more difficult to join the pretinned surfaces then just flowing a normal joint, I think the problems I was having was just really nasty solder, from my brazing drawer in the tool box, in the uncontrolled garage.. I had to throw out all my RG45 rods from corrosion and spent like 2 hours fixing the brass rods filler on the buffing wheel lol. I stopped trying to get the joint the right size and instead overfill it so there is a big bead and then use a dremel sanding drum to get it back into a cylinder shape then polish it from 120 to 2000 grit with the plastic radial disks along the entire splice before sliding the sleeve on top to reduce insertion force. The only problem my current fixture has is that one side needs to be rotated to get the pogo pins to make good contact (it is slightly off center because when I was soldering the pogo pin sleeve into the brass tube I did not have a nice centering fixture for it, I got lucky with one being dead center and the other one is sensitive to rotation or its misaligned. I also don't have whatever N connectors were used with the factory cable I modified, so one of my connectors is a right angle one, but it ends up working because I run the fixture ontop of the VNA when its inside of 19'' rack, next to the other rack, so I will probobly leave it alone for now. Anyway I am excited to see it worked and eventually I will do the refinements.

 I am also going to solder some of those micro alligator clips to the end of some pogo pins so I can put alligator clips inside of the fixture to test small leaded parts, those small copper alligator clips can be bent easily to conform around different diameter leads. Since they are replaceable its cool because you can solder stuff directly to the pogo pins and insert them into the fixture, and even seize the pins with solder to get rid of the spring action in such situations, so it makes a micro pin connector. You can even solder them to empty PCBs without coaxial connectors to make boards that slide into the gold sleeves, so long its acceptable the board is floating from ground, which still may be useful for testing series components chains.

[joeqsmith]

John,  the following is an article Picotest published on sub 100uOhm measurements: 
https://www.picotest.com/images/download/Ultra-low.pdf

A bit odd that while they have presented a picture of what is marked a 20uOhm test device, the picture they choose to include in the report for their DCR is showing 33uOhm.   Note how they used the J2102A rather than the J2102B.  The article was dated April 2018.  If you look at this undated paper, the graph from the VNA is from March of 2019.  So I assume they just didn't have the J2102B yet.

https://www.picotest.com/downloads/INJECTORS/Picotest%20Releases%20Update%20to%20Popular%20Power%20Integrity%20Test%20Tool%20J2102B.pdf


Picotest responded about the discrepancy but sadly I still can't make sense of it.  They talked about a possible reason for the gain:

It can have slight gain, because of the loading impedance of the probe (high Z) vs port (50 Ohms).

I could calibrate that out, but since this was only a relative measurement it wasn't important.

The point is that at low frequency, the common mode transformer can't help, the solid state isolator can. Above a kHz or so the transformer wins.

I won't pursue it further.  It seems like a lot of small, and what I would consider important details are left out of their documents. 

I've been pretty clear about my use of more than one core material.  From the S-parameter data, I suspect the Picotest design uses a single material.   It's possible that you could improve on their design using multiple materials with the same cable type and length.   If you decide to make your own, I am interested in seeing what you come up with.

[joeqsmith]

The P2130A DC Blocker lets through frequencies between 500 Hz and 8 GHz while simultaneously providing low insertion loss and excellent VSWR. Its maximum differential (input – output) voltage is 50V.

From the attached data, the typical insertion loss at 6GHz is around 0.8dB. 

I spent some time making a couple DC blocks to use with the transformer.  Still waiting on parts. 

[JohnG]

Thanks for the links. I have been taking a more careful look at the method.

I do intend to build a few CM transformers and see what I can do. If there are any interesting results, I will post them. It won't be for a few weeks, as I have other priorities right now. Unfortunately, I also only have a NanoVNA 2+4 at my locale. Might be fine to measure the inductance, but not sure how low an impedance can be measured. If travel starts again, I may be able to visit my company HQ and do some measurements on our Keysight VNA.

Cheers,
John

[joeqsmith]

I haven't looked to see if anyone has attempted to run the Plus 4 below 50KHz.  The software that our local member had modified for the original Nano supports down to 10KHz.   The one I have isn't too bad at 20KHz.   For your own transformer,  you may be better off waiting until you can get some time with some better equipment so you can see how they perform at lower frequencies. 

Look forward to seeing what you come up with. 

[joeqsmith]

The new cores arrived.  I tried making up a couple of binocular transformers.  The first with a single material which was a total bust.  The second using a mixture which worked much better.   

Shown are the -3dB and insertion loss at 6GHz, along with the common mode attenuation for the two best transformers I came up compared against transformer #8  and the Picotest.   
******

My last attempt used three different core materials which gains about 4dB in that 100kHz region.  Not too shabby.   

[coppercone2]

I tried my silver plate on tellurium copper, seems to work

working on the BeCu springs now

[joeqsmith]

You should post some pictures of this jig you are working on.   

From what I learned in that last round of transformer testing, I'm making one last attempt on material selection.  The goal is to get the common mode below what was achieved with #8, but get the insertion loss at the high end closer to the performance of the Picotest transformer.    I think it's doable and the cost would still be well below the price of the Picotest transformer. 

[coppercone2]

when its done, i have focus problems getting the small parts to focus and i might remake them so its a waste of time because i am an impatient worker, so it looks sloppy and half assed

The spring is a big unknown to me and I need to make the parts out of a piece of aluminum junk first to see if the tools I had in mind will work because I need to do the machining and then fixture/braze then solder

[JohnG]

If you are looking for best attenuation per turn at low frequency, it's tough to beat a toroidal tape-wound core, especially an amorphous or nanocrystalline one, e.g. Metglas or Finemet. For the most inductance per turn, avoid cut cores or "distributed gap" cores. Also avoid "square-loop" cores, and look for so-called flat-top annealed cores.

Here's an example: https://www.digikey.com/en/products/detail/T60006L2040W422/2258-T60006L2040W422-ND/12531985?itemSeq=363262270

It's not very tolerant of any DC, but CM DC should be negligible. The permeability starts rolling off below 10 kHz, but it's still pretty good at 100 kHz.

It will start to get quite lossy above that, but it's still CM loss so it helps. Combining with another ferrite core may help at the high end if needed.

Cheers,
John