20dB 'RF' attenuator - seeking feedback to improve

[joeqsmith]

The standards provided with the Nano are obviously going to have a different reference plane than the attenuator.   At least with the version of firmware I am using, I am not aware of it having a way to compensate for this.   

The physical length from the SMA connector to the resistor on that test fixture I show is obviously MUCH longer than with the PCB.  If I use the same value resistor on both and measure them using my homemade PCB cal standards, the one mounted on the PCB would be reasonable.  The other I would expect to be off by a fair amount.  If it were a perfect 50 ohm resistor and everything else was perfect, there would be no return loss and the length would not matter. 

For fun, let's test a 1K ohm resistor.  I'll stay with the 1MHz to 300MHz to avoid nanoproblems. 

PCB1&2 are looking at the resistor mounted to the PCB.  1) looking at the Smith chart, there's not a lot of rotation and 2) we can see the resistance is fairly flat over the entire frequency range.   If we were to put say a meter of coax on the nano, we would expect to see a lot of rotation, starting at the far right and working around the outer circumference.  We would expect the same with the test jig as that's what we have done. 

JIG1&2 are looking at the same 1K mounted in my test jig.  No soldering.  1) notice that rather than a dot at 1K, we have this rotation.   2) if we calculate the resistance for this data, you can see even at 300MHz we are not even close. 

There's about a 40mm difference between these two setups.   Using my 166ps/inch rule of thumb (obviously this is GREATLY oversimplified) we would expect about 280ps of delay.   

JIG3&4 notice that I have added a 340ps port extension.  3) notice that we now basically have our dot at 1K again and  4)  the part appears to be somewhat flat.

Keep in mind these are very back of the napkin sort of measurements, but it should provide you with some idea how important the mechanics are.   I suspect after building these attenuators,  you're starting to see that things are much less critical at lower frequencies.     

As I mentioned earlier, Mario is helping me sort out some home made standards.   I made up a set that I thought would be good, but these were not even close.  I machined up a new set of parts and knew Mario would not approve of them.     After the third attempt, I came up with something that appears reasonable.  These standards will be used at much higher frequencies and again, are just for ballparkish measurements. 

[joeqsmith]

http://zone.ni.com/reference/en-XX/help/373153D-01/vnahelp/reference_plane/

[enut11]

Thanks @joeqsmith.
It is now obvious that HF circuits are a mixture of art and science.
This has been an interesting journey for me. Starting off at audio frequencies with a nominal cutoff frequency of 100KHz, I now have great respect for radio frequency circuit designers.
enut11

PS
Was the Smith chart named after you

[rf-messkopf]

As I mentioned earlier, Mario is helping me sort out some home made standards.   I made up a set that I thought would be good, but these were not even close.  I machined up a new set of parts and knew Mario would not approve of them.     After the third attempt, I came up with something that appears reasonable.  These standards will be used at much higher frequencies and again, are just for ballparkish measurements.

The point is that with SOLT, for doing the vector error correction, the S₁₁ responses of the one-port cal standards as well as the S₂₁ response of the thru have to be fully known with respect to the reference plane for the connector family used.

On the positive side, this means that the standards can be quite imperfect or imprecisely machined as long as these imperfections are known and accounted for, either by a model and a few parameters from which the analyzer can calculate the S-parameters of the standards, or directly by their full S-parameters (e.g. from a Touchstone file). When you are not able to determine these responses by theoretical means (e.g. an EM simulation), you can always measure them against a known cal kit.

And you do not get away without corrections, at least for the open, because there will always be a fringing capacitance. Moreover, for many connector families the physical open and short will inevitably be offset from the connector reference plane. This is especially true in N connectors (look at the location of the reference plane -- in the female N connector the center pin protrudes beyond the reference plane, whereas in the male N the center pin tip is recessed).

Was the Smith chart named after you

Wouldn't that mean he is 115 years old? 

[joeqsmith]

I feel old some days but not that old!  lol

On the positive side, this means that the standards can be quite imperfect or imprecisely machined as long as these imperfections are known and accounted for, either by a model and a few parameters from which the analyzer can calculate the S-parameters of the standards, or directly by their full S-parameters (e.g. from a Touchstone file). When you are not able to determine these responses by theoretical means (e.g. an EM simulation), you can always measure them against a known cal kit.

And you do not get away without corrections, at least for the open, because there will always be a fringing capacitance.

Of course, you can't expect to use just anything for standards at 10GHz.  Using an open in place of a load may not work so well no matter how you characterize it.  There's a limit to how poor these can be.  I believe I am pushing these limits. lol.  I had read a paper that was talking about common mistakes people make when taking measurements.   One of the first problems talked about how some engineers believe SOLT can compensate for bad cables.  If I come across it again, I will post a link.   

As I mentioned, I've never done anything to correct for my home made standards.  Certainly my low frequency measurements could be improved upon but they are good enough.  Actually, I am not sure if I used a brand new set of top of the line Keysight standards with the Nano and the Nano supported using their compensations,  would we even see any measurable difference.  Part of me doubts it.  I'm not suggesting that my home made standards are all that good, but we are working at fairly low frequencies with a $50 bit of hardware. 

This presentation goes down the VNA calibration rabbit hole pretty deep.  It's a very high level talk, so nothing real complex.  There are a lot of little things to consider.  It's an eye opener. 
http://resource.npl.co.uk/docs/networks/anamet/members_only/meetings/22/juerg_ruefenacht.pdf

[joeqsmith]

Time to break out Mario's scripts again.   

[joeqsmith]

https://www.microwaves101.com/encyclopedias/how-to-not-trash-a-calibration-kit
https://www.microwaves101.com/uploads/Connector-Care-Poster-2007.pdf

I spent some time with the microscope today inspecting the standards.   Outside of a little cleaning, everything looks in good shape.  I was pretty hesitant to borrow these because the finger contacts are fairly delicate.   Shown are the male and female sides.  These are not your 10 cent Chinese connectors we see on these attenuators. 

[joeqsmith]

 

Anyone who has made a set of standards that provide good results above even 3GHz, you have my respect.  I'm not sure which is worse, my first attempt at the N style or these SMA ones.  I had hoped that with the T-check looking as good as it did, these would be better than they are.   Back to the drawing board.

 

[ogden]

Anyone who has made a set of standards that provide good results above even 3GHz, you have my respect.

Well.. Yes indeed. Some time ago you were insultingly confident about "standards" shipped with NanoVNA.

[joeqsmith]

Still playing around with the home made standards but I  thought as long as I have these Agilent standards on loan, we could try and measure the low cost 20dB attenuator with a little higher confidence.   

The following link is the datasheet for the Mini-Circuits VAT10 which is a 10dB attenuator:
https://www.minicircuits.com/pdfs/VAT-10+.pdf

Attached is the data I collected for the VAT 10.   It seems to fall well within the published typical numbers.

Looking at the X13773, which is the green PCB,

Level 1: 10 0.8 dB
Level 2: 20 1.1 dB
Level 3: 30 1.5dB
Standing wave ratio: 1.20

It seems like it meets their flatness but the SWR appears higher than they claim.     

Also shown is a Midwest Microwave 20dB attenuator as a comparison.  I don't have anything near as nice as Mario's metrology grade attenuator.   

[joeqsmith]

Attached is looking at the same low cost 20dB attenuator after calibrating with my home made standards.  Mario's scripts was used to determine the coefficients based off the Agilent standards.   As you can see, there is some difference when compared with the Agilent standards.

Also, as a side note, the T-check is really poor.   It is based on the same hardware used to make the short and open and has roughly the same reference plane.  I am using fixed resistors and suspect they are just not going to be close enough for this application.

[joeqsmith]

Mario was kind enough to update his Octave script to add a plot showing the phase error between the model and measured data.

You can download it here:
https://www.mariohellmich.de/lostfound/files/ck-model

I've attache the plots for my female SMA open and short as a reference of what to expect if you try it.  Basically the steps are: Calibrate the instrument using the Agilent standards. Use adapters rather than cables.  Test the short & open and collect their Touchstone files.   Feed these into Marios script.  You may need to change the script to support your VNAs Touchstone file.  In my case the header length needed to be changed. 

In my particular case, Agilent has a few different stand alone editors for their database.  Their newer software doesn't work with my old VNA and will corrupt the file causing a memory overflow error.   

[joeqsmith]

This is looking at the data for my home made N standards after the third attempt at machining the parts and trimming them.  These were made from an old set of Narda terminators. 

Also shown is the 180 degree phase difference between the open and short while sweeping up to 6GHz.   

For the terminators I knew I could not make anything that was going to work at these higher frequencies, so I left one of the Narda terminators stock.  I have an old HP terminator that I compared it against and it seems reasonable.  For the SMA I am using a part from Midwest Microwave.   

I have also characterized a few other bits I had to use as a sort of check.  It would be interesting to see how Kirkby's would compare with my how made setup.  Considering the time invested, and the advantage that I could loan these two Agilent sets, buying something ready made would be a much better solution assuming they are any good.   

https://www.kirkbymicrowave.co.uk/

[joeqsmith]

Looking at the return loss of the Mini-circuit ANNE-50 and MidWest Microwave terminators. 
(This is after calibrating the system with the Agilent standards using an adapter rather than cables)

https://www.minicircuits.com/pdfs/ANNE-50+.pdf

[ogden]

Looking at the return loss of the Mini-circuit ANNE-50 and MidWest Microwave terminators. 

Which MidWest Microwave terminator (part.no) used? There are many different. I take shown Minicircuits any day - 10x lower cost yet perform better.

[joeqsmith]

To get some idea how repeatable the new set is, I had removed all of the adapters and left the VNA off for a day.   I then let it warm up for an hour, installed the adapters and loaded the previous calibration file.

You can see the original Touchstone files compared with the set taken today.   About 0.2dB of error.  The adapters I am using are not the cheap garbage like seen on that low cost attenuator, but they are certainly not metrology grade. 

I had a few lower cost MWM terminators.   While I had everything setup, I ran all four of these.  One of these I had marked suspect but it appears that two of them are are not very good. 

When I compare these with my home made PCB standards.  These are the ones I use with the 8754A, not the ones I show with the Nano.  These are using 0805s vs the 1206s.   You can see up to about 500MHz, these are much better than the commercial parts.  Once we get beyond a GHz, it gets pretty bad.   

[joeqsmith]

A section of semi-ridged was sacrificed.  Yellow tape was added to mark the direction for later.  The Nano was set to sweep to 1.5GHz and calibrated using the Agilent standards.  Data was then collected for the ANNE.   Next the Nano was calibrated using my home made PCB standards that I use with the 8754A and again the ANNE was installed and data was collected for it.     These two files were then compared against the two previous files collected with the Agilent. 

The Nano, when calibrated with the Agilent 85033E standards, is right in there with old Agilent VNA up to about 250MHz.   As expected, at 300MHz it starts to have some trouble and at 900 it gets much worse.    But again, for $50 it's pretty impressive.   

The graph was pretty hard to see, so I captured it at a higher res. 

[enut11]

Thanks @joeqsmith. It is reassuring to know that the inexpensive little NanoVNA can be relied on up to about 250MHz.
This will also help me characterise some of my lower frequency equipment, and for only $80AUD. 
enut11

[joeqsmith]

So far I've held off setting up yet another account on Tinde to order a Nano V2.  I'm surprised we are not seeing any reviews for it yet as I would have thought the developers would have sent out a handful to the larger channels some time ago.   It will be interesting to get one with their cal standards and cables to see how it compares.  At 3GHz, the accessories should be much better than what was supplied with the Nano.   The big advantage as I understand it is the V2 does not use harmonics. 

When you start to do more with the Nano, you may find you require additional hardware.  For example, to check an amplifier you may need Bias-Ts, step attenuators, filters (for the harmonics to prevent overdriving).  Bits that would normally be built into older used systems.  We don't need to be concerned with filtering out the fundamental with these older systems (8753...).    Pretty soon, that $80 may turn into $500.     

[ogden]

When you start to do more with the Nano, you may find you require additional hardware.  For example, to check an amplifier you may need Bias-Ts, step attenuators, filters (for the harmonics to prevent overdriving).  Bits that would normally be built into older used systems.  We don't need to be concerned with filtering out the fundamental with these older systems (8753...).    Pretty soon, that $80 may turn into $500.   

Such purchasing strategy does not make any sense unless you know that those 500$ worth of accessories will be used with proper instrument in the not so far future.

[cdev]

I sometimes use edge mount SMAs which terminate right at the edge of the PCB but since they are made for etched boards instead I will solder their feet to the ground plane and instead will have the trace a bit higher and use components on top or Ive also used mini coax. That seems to work really well, best of both worlds. The advantage of that kind of mucking around is that everything is reusable and rearrangeable.

Axial resistors like that in a cavity are good to 1 GHz so hug the ground plane.  Use a connector which terminates closer to the ground plane like edge mount.  Through-hole 1/8 watt parts are smaller and will perform a little better.  Mount the shunt resistors to make a cross along the ground plane instead of a tee.

[joeqsmith]

I picked up a few ANNE terminators to see if I could hand select something a little tighter for my calibration kit.    #1 is the part that was shown in the previous graphs.  I have also included the load from the Agilent 85033E. 

[joeqsmith]

Comparing the measurement results for the original Mini-Circuits ANNE-50+ 50ohm terminator with the last two times it was measured.    This part is designated #1.   

[joeqsmith]

Comparing all 6 pcs and finally #1, #4 and the Agilent standard. 

[ogden]

Comparing all 6 pcs and finally #1, #4 and the Agilent standard.

Seems like you may return four of six terminators as faulty because specs say that upto 1GHz RL is no worse than -43dB. Why didn't you include NanoVNA "standard" in this comparison?