Measuring passive parts with VNA

[tszaboo]

Continuing the discussion from here:
https://www.eevblog.com/forum/testgear/12-ghz-active-probe-project/75/
So I ordered a test PCB to measure passive parts with a VNA. It's for 0402 components because that's what I use for antenna matching.

The reference plane for the measurements is the SMD pad of the 0402 component.
The measurement method is a though measurement, based on this article:
https://coppermountaintech.com/measurement-of-electronic-component-impedance-using-

The measurement method should yield the smallest errors. Reflection based measurements are possible on a second PCB.
I'm interested how well passive parts behave. I have some experience in antenna matching, mostly for 2.4 GHz, but I'm really not an RF guru. I had the idea a while back because some RF inductors were acting atrociously, doing all kinds of crazy things when I tried to match an antenna. And I don't have an expensive simulation tool, it's more a trial and error journey for me.
The tool to measure it is a Megiq 440e, which is a VNA that goes between 400MHz and 4 GHz.

[joeqsmith]

Have a look at Mario's board.  Note the short, open and thru.  Also note the reference plane for the match. 

https://www.mariohellmich.de/projects/trl-cal/trl-cal.html

[joeqsmith]

Some time ago I attempted to replicated an experiment where the VNA was used to measure the parasitic shunt capacitance of some resistor.    Not sure you will find anything of use there.

https://www.eevblog.com/forum/rf-microwave/shunt-capacitance-of-1206-smd-resistors-jeroen-belleman-december-2010/25/

[tszaboo]

Have a look at Mario's board.  Note the short, open and thru.  Also note the reference plane for the match. 

https://www.mariohellmich.de/projects/trl-cal/trl-cal.html

I'm not sure which difference are you trying to point out. The VNA I have at work only supports OSLT calibration, so I wouldn't be able to use that fixture. I've used an impedance controlled pool to make the board, and calculated the waveguide. I see that he is not using a clearance on the RF connector. A Würth appnote was specifically mentioning that there should be more space around the pin of these end launch connectors. Not a concrete guide, and I have no way to simulate it. I also don't recognize the connector he is using so maybe that has a different footprint.
The other thing I see is the via fencing. As I understand this is more for EMC compliance, and a waveguide will work just as good without it, at least for the frequencies in question. But I'm ready to be proven wrong.

[joeqsmith]

Added inductance and discontinuity. 

[rf-messkopf]

I'm not sure which difference are you trying to point out. The VNA I have at work only supports OSLT calibration, so I wouldn't be able to use that fixture.

I ran a quick test on a 22pF 0805 capacitor in the series fixture using TRL, TOSM and unknown thru (UOSM in R&S speak) calibration on that very board you are discussing. All calibrations only use the cal standards on that board and assume them to be ideal. As you can see, there is not a huge difference when you calculate the impedance magnitude from S21 data. TRL generally yields the smoothest traces with least ripple on this board, especially when you look at S21 data, and it surely is the most precise way to get the reference planes of the measurement directly at a particular point in a planar geometry.

Further, the advantage of TRL is that it does not need fully known standards. For example, the reflect only needs to have some not too small reflection coefficient, which is assumed to be the same at both ports. Also, the line only needs to have a precisely defined impedance (this determines the system impedance of 50 ohms), but it can be lossy and have an arbitrary propagation constant. Therefore you can make very precise measurements that way.

There will be some systematic error in the TRL cal because the line standard does not have exactly 50 ohms since the board was manufactured in a cheap pool process, without proper impedance control.

I've used an impedance controlled pool to make the board, and calculated the waveguide. I see that he is not using a clearance on the RF connector. A Würth appnote was specifically mentioning that there should be more space around the pin of these end launch connectors. Not a concrete guide, and I have no way to simulate it. I also don't recognize the connector he is using so maybe that has a different footprint.

The part No. of that connector is mentioned on my homepage. It is designed specifically for coplanar waveguides.

The other thing I see is the via fencing. As I understand this is more for EMC compliance, and a waveguide will work just as good without it, at least for the frequencies in question. But I'm ready to be proven wrong.

This is a grounded coplanar waveguide, hence the front side ground planes must be tied in some way to the rear ground plane. That's what the via fences are for.

Another option you could try is making a board with a footprint and two connectors like the one pictured in the attachment, then calibrate to the connectors (for example with TOSM/SOLT), and do a port extension to the SMD pads on the board (i.e., add the required offset length at each port).  You can check if your offset lengths are correct by temporarily soldering a short to ground at each SMD pad. This might already be precise enough below a couple of GHz.

[switchabl]

Another relatively simple approach that works well even at much higher frequencies is 2X-Thru de-embedding. It involves a standard coaxial calibration at the connectors and a single planar thru standard. The T-matrices of the left and right halfs of the fixture can be extracted from the thru measurement using time gating (under some reasonable assumptions) and they allow you to move the reference plane to the middle of the line. The connectors need to be far enough apart so that you can separate them properly in the time domain (this depends on the bandwidth of the VNA/fixture).

Most modern VNAs should have this as an option (Keysight calls this "automatic fixture removal", R&S has something similar). But you could always do it offline:
https://scikit-rf.readthedocs.io/en/latest/examples/networktheory/IEEEP370%20Deembedding.html
https://www.amphenol-cs.com/software

[G0HZU]

Here's a quick measurement of a cheapo 22pF 0805 capacitor vs a decent 22pF 0805 cap made by PPI. The PPI cap should have much lower ESR and a lower SRF and the VNA plot below shows this.

A good way to test your fixture quality would be to see if you can get fairly sensible data for the ESR of the cheapo 0805 22pF cap up to a few GHz. Then compare against a decent 22pF cap to see if you can measure the difference in ESR between the two caps.

You can see that there is a clear improvement with the PPI part. The VNA predicts the ESR is only about 0.11R at 2GHz.  In reality, it's probably a bit higher than 0.11R at 2GHz. The VNA is really being stretched here and there will be a lot of uncertainty in the measurement, but there is a clear difference in ESR between the two 22pF caps.

Note that this was done with a really old homebrew SMA cal kit. it's probably about 20 years old. I think the cal kit is getting tired now, I keep meaning to make a new 50R load as the VSWR above 1.5GHz has definitely degraded slightly over the years.

[joeqsmith]

Comparing various caps with my LRC meter, datasheets and low cost VNA.   Used a few fixed resistors as a sanity check.  The test runs from about 4min in to about 20min.    Data was also taken with homemade standards and the ideal model.   



Of course for these really low number, the ground loop is a problem.   Even with the common mode transformer, the NanoVNA's noise floor limits the measurement.

[tszaboo]

The part No. of that connector is mentioned on my homepage. It is designed specifically for coplanar waveguides.

Thought so. People knowing RF usually don't make simple mistakes like that.

Added inductance and discontinuity. 

The RF signal in theory goes between the ground and the trace. So it doesn't in theory make that much difference between via fenced waveguide or regular waveguide.

Here's a quick measurement of a cheapo 22pF 0805 capacitor vs a decent 22pF 0805 cap made by PPI. The PPI cap should have much lower ESR and a lower SRF and the VNA plot below shows this.

A good way to test your fixture quality would be to see if you can get fairly sensible data for the ESR of the cheapo 0805 22pF cap up to a few GHz. Then compare against a decent 22pF cap to see if you can measure the difference in ESR between the two caps.
You can see that there is a clear improvement with the PPI part. The VNA predicts the ESR is only about 0.11R at 2GHz.  In reality, it's probably a bit higher than 0.11R at 2GHz. The VNA is really being stretched here and there will be a lot of uncertainty in the measurement, but there is a clear difference in ESR between the two 22pF caps.

Note that this was done with a really old homebrew SMA cal kit. it's probably about 20 years old. I think the cal kit is getting tired now, I keep meaning to make a new 50R load as the VSWR above 1.5GHz has definitely degraded slightly over the years.

I'm interested in measuring resistors. So I plan to get some regular 50 Ohm and some ultra high end RF resistors, and we will see.

[nctnico]

The part No. of that connector is mentioned on my homepage. It is designed specifically for coplanar waveguides.

Thought so. People knowing RF usually don't make simple mistakes like that.

Added inductance and discontinuity. 

The RF signal in theory goes between the ground and the trace. So it doesn't in theory make that much difference between via fenced waveguide or regular waveguide.

But the problem is that without good connection between the grounds, the inductance starts to play a role. If you are simulating this in a program like Sonnet, you'll see the HF current to start flowing around the edges of the 'ground' on the top (signal) layer if you don't have enough vias to stitch all the grounds together to form a single ground (at least to a few GHz).

[rf-messkopf]

Thought so. People knowing RF usually don't make simple mistakes like that.

Just a hobbyist here.

Added inductance and discontinuity. 

The RF signal in theory goes between the ground and the trace. So it doesn't in theory make that much difference between via fenced waveguide or regular waveguide.

But the problem is that without good connection between the grounds, the inductance starts to play a role. If you are simulating this in a program like Sonnet, you'll see the HF current to start flowing around the edges of the 'ground' on the top (signal) layer if you don't have enough vias to stitch all the grounds together to form a single ground (at least to a few GHz).

In fact, on my board, the front side ground plane even would be cut in two halves without a rear ground plane and the via fences since there are traces that run from board edge to board edge. Also, the rear ground plane adds some shielding.

On the other hand, on my TRL board the launcher to trace transition is not super critical since it is behind the calibration plane, and hence eliminated by calibration. The assumption is, however, that all launcher to trace transitions are identical for the cal standards and the DUT. This is also true for TOSM/SOLT. The biggest problem with TOSM the above measurement is that the cal standards were uncharacterized and assumed to be ideal. This is a problem especially for the open since there is always fringing capacitance.

But when you calibrate to the connectors and use port extension to the DUT fixture SMD pads, then you do not eliminate the launcher to trace transition, and you should be careful to avoid added inductances and discontinuities.

[tszaboo]

So I got the boards, got the connectors, and I'm still waiting for the ultra high bandwidth RF resistors to make the calibration possible against the pedestrian cheap 50 Ohm resistors.

[tszaboo]

Added inductance and discontinuity.

A Würth appnote was specifically mentioning that there should be more space around the pin of these end launch connectors.

But yeah, maybe the coax transition is not optimal. Good thing than that it get's calibrated out then.

[tszaboo]

So I finally got some Vishay CH0402-50R resistors as calibration resistor, and the plan is that I'll compare them to some plain old thick film resitors this weekend to start with.

[tszaboo]

Note to self: measurement error happens if you don't tighten the SMA connectors correctly!
Also please modify footprint of the RF connector its PITA to solder it correctly.


DUT: Samplekits.de (no longer active AFAIK) generic resistor, unknown brand. 51Ohm.
Sweep: 360 steps (overkill) between 400MHz and 4000MHz, 0dBm.

[tszaboo]

DUT: Vishay CRCW 49.9 Ohm CRCW040249R9KFED
This is the resistor I ordered, and intended to use for VNA calibration in the future
At 4GHz the S11 measures 70 + 31J
I'm actually doubting this second board is assembled correctly. I'll put it under the microscope next week.
The measurement is repeated on a 3rd assembled board, and it's pretty much the same.

[tszaboo]

DUT: Samplekits 1.1KOhm resistor, because I wanted a ~950 Ohm but grabbed the wrong one.
I've also added the S22 result zoomed in.
400MHz: 1129 + 1.6J
4000MHz 91 +310J 

[coppercone2]

can you try the same dimension short circuit by soldering something like precisely filed copper?

or buy a 0402 short circuit. keysight sells little blocks, not for soldering. i wonder if you can buy em cheaper

probobly easier to buy a larger resistor and try to duplicate it with a short circuit, 0402 is hard to just manufacture. Maybe it can be painfully lapped with tweezers under a microscope to make something generally similar out of a bit of wire or something

[tszaboo]

can you try the same dimension short circuit by soldering something like precisely filed copper?

or buy a 0402 short circuit. keysight sells little blocks, not for soldering. i wonder if you can buy em cheaper

probobly easier to buy a larger resistor and try to duplicate it with a short circuit, 0402 is hard to just manufacture. Maybe it can be painfully lapped with tweezers under a microscope to make something generally similar out of a bit of wire or something

Good idea. A perfect piece of copper wouldn't be much different from the "Trough" calibration, that measures as 50Ohm exact, since it was used for the calibration.
For now I measured a 0 Ohm resistor, Vishay CRCW series. It Shows as much as 50Ohm + 0.4nH on the S22. I need to look at those calculations again, if this means that the 0 Ohm resistor is actually 0.4nH or not.

[tszaboo]

So I tried to solder a piece of copper into the resistor's place, and this is what I'm getting:
Measures as 44+4J Ohm, or 44 Ohm + 0.2nH. A 0Ohm resistor is clearly better choice.
Also note to self: When assembling another test fixture, sanity check it with a 0 Ohm resistor. Also, solder down all pins of the SMA connector properly.

[jonpaul]

See vintage HP app note 77-3

https://www.hpmemoryproject.org/an/pdf/an_77-3.pdf

Modern imp  meas handbook 4th ed

https://www.pearl-hifi.com/06_Lit_Archive/15_Mfrs_Publications/20_HP_Agilent/HP_App_Notes-Handbooks/Z_Measurement_H-book_4th_Ed.pdf

We use HP4195A and test fix 41952

Good 100k..500 MHz.

Jon

[coppercone2]

the proper kits come with copper and also dielectric blanks for every component size that you measure.

I believe but I am unsure that the machine does some kind of regular calibration (say with a APC-7 OSL) and then a special short circuit and open circuit calibration on the fixture, without the 50 ohm load.

I think they have 50 ohm standard fixture calibration resistors for some sizes, but not all. I think they extrapolate some measurements from some components based on open and shorts in the fixture without a exact replica load.

[G0HZU]

Here's a VNA measurement of a SMD 27nH inductor in an 0603 package. The measurement covers up to 6GHz.
The plot below shows Rs, Ls as well as Rp, Lp and inductor Q vs frequency. I used the fixture correction feature of the (Agilent) VNA to set the reference plane right at the pads of the little SMD inductor. The VNA can model (and therefore cancel out) the fixture connections using two port s-parameter data and this gives really good results over a wide frequency range.

The SRF of the inductor is just over 3GHz and the Q peaks at about 75 and this is similar to the manufacturer's data.

[tszaboo]

Here's a VNA measurement of a SMD 27nH inductor in an 0603 package. The measurement covers up to 6GHz.
The plot below shows Rs, Ls as well as Rp, Lp and inductor Q vs frequency. I used the fixture correction feature of the (Agilent) VNA to set the reference plane right at the pads of the little SMD inductor. The VNA can model (and therefore cancel out) the fixture connections using two port s-parameter data and this gives really good results over a wide frequency range.

The SRF of the inductor is just over 3GHz and the Q peaks at about 75 and this is similar to the manufacturer's data.

So the VNA has the model of the test fixture in it? And you don't have to calibrate it, just enter the type number?

[G0HZU]

My old Agilent VNA has FS (fixture simulator) mode that can accept an s2p file based model of the test fixture at each port. This is usually better than using a basic port extension to model the fixture.

Often, it is possible to measure or model the test fixture as a two port network and so this VNA can correct for the test fixture using the s2p file data.

[rfclown]

I often use an SMA connector with the center pin clipped short. A 0402 isn't quite big enough to bridge the gap between center pin and ground, so I blob solder on the ground side to make the connection. I grabbed a few things out of a box to show in the picture. The larger inductor was probably something I wanted to measure. The resistors are actually a cheap homemade mismatch load that I use for testing. I think the resistors are 0603.

Do a single port OSL calibration on VNA. Measure SMA connector with NO component. Dial out electrical length of connector so that you see an open. Solder component on SMA connector and measure.

[tszaboo]

https://www.vishay.com/docs/31006/tnpw.pdf This is an example datasheet of a general purpose resistor where the RF performance is mentioned. So it's aligns with jonpaul's app note, the low value resistors are slightly inductive, the high value ones are slightly capacitive (Figure 1-9). Honestly, I was expecting it to be the opposite, high value resistors have a lot of "wigle-wigle yeah" in them which is inductive. I wonder if each series has a "golden value" that has high bandwidth and good RF parameters.

[LM21]

Are you really telling me that there is a difference between various 22pF capacitor brands. I'll probably use something larger than 0402 but still.

[coppercone2]

propery rf caps are quite expensive! I am not sure what you mean. That you think there is no difference between various brands of the same type of ceramic capacitor?

[tszaboo]

Are you really telling me that there is a difference between various 22pF capacitor brands. I'll probably use something larger than 0402 but still.

Absolutely, even at 1-2GHz. Imagine one capacitor with one pair of plate and one with two pairs at double the distance. They are going to have different inductance. The ceramics used matters as well.

[coppercone2]

the reason why it sometimes works is because the ESR of the capacitor is low. Even if it becomes inductive, the ESR is so short that even with the effect of inductance its still a pretty good short circuit. but the question is for how long and how consistently. Thats why random unspecified capacitors might work in RF even if they are past their SRF.

The ESR represents the lowest possible impedance for the capacitor, which is at SRF.

For instance this graph.


Does not look like the slope before and after the SRF is the same or even really a slope! it shows you how you can possibly get screwed making assumptions about capacitor performance from just ideal equations that relate to SRF

[LM21]

Are you really telling me that there is a difference between various 22pF capacitor brands. I'll probably use something larger than 0402 but still.

Absolutely, even at 1-2GHz. Imagine one capacitor with one pair of plate and one with two pairs at double the distance. They are going to have different inductance. The ceramics used matters as well.

One worry more. I  have known this before kind of, but didn't think of it  much. And my interest is mainly below 1GHz.

[joeqsmith]

Still, even below 1G, materials used can become very important.   High Q circuits for example. 

[ProphetM]

My old Agilent VNA has FS (fixture simulator) mode that can accept an s2p file based model of the test fixture at each port. This is usually better than using a basic port extension to model the fixture.

Often, it is possible to measure or model the test fixture as a two port network and so this VNA can correct for the test fixture using the s2p file data.

G0HZU,
Please return to the uk vintage forum?

[tszaboo]

50 Ohm thin film resistor. The previous ones were thick film.
TE Connectivity RP73PF1E49R9BTDF
It measures as 86Ohm (40+50), + 16 J at 4GHz so you don't want this as your cal resistors either.
Has similar inductive behavior as the thick film resistors.

[tszaboo]

Now, something else.
DUT:Würth Electronic WE-MK Multilayer ceramic inductor 15nH part number 7447840115
Has self resonant frequency of 2.3GHz minimum

I measured the SRF at 3.32GHz. It's very nicely on the 1.0 circle.

[tszaboo]

DUT:Würth Electronic WE-MK Multilayer ceramic inductor 1.5nH part number 7447840015
Has self resonant frequency of 6GHz minimum, which is beyond what I can measure with this VNA.

[tszaboo]

DUT: 2.2pF +/- 0.5pF NP0 Capacitor WE 885012005051
This is their jelly bean 0402 capacitor.

[tszaboo]

DUT: 2.3pF +/- 0.05pF NP0 Capacitor WE 885392005009
This is their RF range 0402 capacitor.
I'm not sure if I can tell the difference with this test setup. The tolerance is of course a big differentiator, otherwise I see very similar curves.
Someone more knowledgeable than me, can maybe tell me what sort of difference should I be looking at?

[coppercone2]

i think you might need to measure 50 of each from different batches  to see the difference

[tszaboo]

I'm taking the VNA for anther spin.
Aliexpress cheap attenuator board.
The phase is out, because the electrical length is different than the calibration (that happened on the connector)

What I noticed is, these don't have their SMA connector soldered on the back side of the board. It's also not a nice SMA, it's a bit "crounchy" to connect.
The electrical length, compared to the cal kit is 233ps, that has been de-embedded.
I changed the graphs. I think it was a mistake to just go with the default settings for the previous measurements, while I could display an impedance vs frequency graph.
"0dB" through:

[tszaboo]

The -10dB attenuator.
Made with 0603 resistors. 2x 63.5 Ohm to GND, 33.9 + 29.5 in series.
It measures between -10.15 and -9.71dB.

[tszaboo]

The -20dB attenuator. It measures between -18.5 and -19.5. The electrical length is different (strange) at 213ps.
0603 resistors, 52 to GND, 46 in series, 32.2 to GND, 49 in series, 54 to GND.
I took these measurements in circuit, so these are not the actual resistor values, but I don't feel like solving a bunch of math right now. I't more interesting that it doesn't seem symmetrical, or the used components have a bunch of variability. Maybe they are 5% resistors?

[tszaboo]

The -30dB attenuator. Electrical length is 195ps. Why does it get shorter?
Gain is between 29.6 dB and -30.75dB

[szoftveres]

I wonder what would happen if this board was designed with symmetrical "mode" termination - i.e. the coplanar wg has TE waves on both sides; on this particular board, TE waves hits a resistor one side but go freely on the other.

Would be great to see a design with double value shunt resistors and having soldered two on each side (essentially in parallel), and do an apple to apple comparison with this board, particularly at high frequencies.

[tszaboo]

For me it's unclear what kind of layer stackup they have on the board. I would rather design one with the option to place resistors on both side and work from that. I'm not sure how it would differ, or if it would be better in practice. The track width is about the same as the pad size for this resistor, so the track would need to change thickness, or a non-symmetrical placement would be necessary. Or some wierd CPW dimensions that are hard to launch from the end connector with a thick trace. So maybe on two layers?