Poor performance of DIY SMA-M Precision Load

[W3AXL]

Hi all!

Decided I'd try and make myself a nice OSL test kit since I'm starting to accumulate more and more higher-frequency gear. I based my design off of several different articles showing various ways of making a precision 50Ω load that would work well into the multi-GHz range.

The design I came up with, while crude as a first iteration, seemed promising. I found a bar of hex stock that would fit the 4-hole flange SMA connectors nicely. So I built one:



The load is very simple - two 0.1% Panasonic 0805 100Ω resistors in parallel. This was a common design I found in many others' homebuilt loads, so I figured it would work well.

Next step was to measure them. Fortunately at work we have a myriad of RF test gear, so I fired up the E8362 and set up a 10 MHz - 10 GHz sweep using our 85033C 3.5mm cal kit.

I had on hand a set of LibreVNA and NanoVNA cal standards as well (the cheap ones they send with the units) so I figured I'd characterize those as well.

And here are the results (note that the S-parameter tool I was using plots return loss as a positive number, so higher on the chart here is better):



Sadly, and unexpectedly, my homespun 50ohm load was actually the worst performer out of the group! I'm curious what the forum thinks could be the culprit for the terrible response. It doesn't event hit 1 GHz before the return loss drops below 40 dB. I was hoping to at least have a usable load up to 3 GHz (where my home VNA caps out) but I may have to rethink the design significantly.

[tszaboo]

75 Ohm connector by accident?

[W3AXL]

I wasn't aware they made 75Ω SMA connectors - but as far as I can tell, these look identical to other SMA loads I've got at my desk.

Here's a smith chart of the DIY load - I left the S1P files on my work PC so enjoy a terrible screenshot instead:



Looks like I've got some capacitance on the lower end and some inductance above 5 GHz.

[RoGeorge]

The load is very simple - two 0.1% Panasonic 0805 100Ω resistors in parallel.

What part number?  Are they specified as "High Frequency/RF" smd resistors?
Common smd resistors might not behave well at high frequency.

[shabaz]

I think you're expecting too much from it. I wouldn't expect such a DIY load to be used beyond a few hundred MHz, perhaps 1 GHz at most. Also, I think most of the DIY loads use two 0603 resistors (0603 do fit in the gap between the SMA center pin and the outer conductor, measured from a reel of resistors to be as close to combined 50 ohm as possible, since sometimes just 1% thick film is used. Plus, it's a bit risky not securing the resistors. I don't know a good solution, but I have tried using epoxy glue in the past, covering all that area including the dielectric (not ideal again).

However, reasonable 50-ohm loads are available ready-made at a few tens of $. Example: cheapest Huber & Suhner one should look like a dot on the VNA across your frequency range, and could be glued into the alu housing.

[tooki]

The load is very simple - two 0.1% Panasonic 0805 100Ω resistors in parallel.

What part number?  Are they specified as "High Frequency/RF" smd resistors?
Common smd resistors might not behave well at high frequency.

The most beautiful SMD resistors I ever soldered were Vishay Sfernice CH series flip-chip 0603 resistors that my customer had specified by mistake, not realizing they weren’t in the signal path and thus didn’t need to be high-performance ones like those. They’re beautiful little white cubes that cost about $6 each. But they’re good to 50GHz, which is something an ordinary 0603 resistor can’t claim!

[EggertEnjoyer123]

You can get a bit better frequency response by flipping the resistor over.

See this article for more information on resistor frequency response at microwave frequencies: https://www.vishay.com/docs/60107/freqresp.pdf

You could also just buy these for a few bucks and select the best ones: https://www.mouser.com/ProductDetail/Amphenol-RF/132360?qs=V8V9vCxO5JtDrelx3RqrVA%3D%3D

[W3AXL]

Thanks for the great suggestions everyone!

To those saying just buy a pre-made load, I absolutely could do that (and I've got a small pile of precision loads). But part of this adventure is manufacturing my own, and understanding all the complexities that go with it.

I might throw a small quantity of some RF 0603s in my Digikey cart, to see how they compare. Good suggestion on mounting them upside down, I'll give that a shot as well.

[tszaboo]

For antenna matching, that I make in-situ PCB OSL calibration standards, I've switched to the Vishay FC series resistors.
Previously, I had a lot of trouble with antenna matching, because the simulated matching network just didn't work as expected, and I was spending easily an afternoon soldering 0402 parts and scratching my head why it doesn't work. The FC series is like 2-3 EUR each.
I had matchings that worked on first try since.

[szoftveres]

I'm assuming that the SMA coax inner conductor - SMD resistor junction is a big capacitor and it just looks inductive at higher frequencies  - at 10GHz λ/4 is ~7mm in vacuum (even shorter in teflon), so the capacitance could look inductive on the other side of the SMA connector - the clue is that the smith chart is orbiting around the 50Ω point.

The bottom end looks more of a concern - the clue is 50.9Ω at 1GHz.
I would suggest to characterize your DIY load between 100MHz and 500MHz - at these frequencies the Smith plot is supposed to show somewhat consistent resistance and reactance readings - e.g. 50.9Ω and some capacitance across the band. From this you can deduct an RC model, then calculate a delay, based on the frequency where the DIY load reactance flips sign (becomes inductive) and apply all this data when using your cal standard; or just use this data to reduce the capacitance and further improve your DIY cal kit.

At very high frequencies mode discontinuity could be a problem, e.g. the homogenous TEM mode coming out from the coax all suddenly has to split into two, inductive SMD resistors. Usually a good load is a homogenous, lossy dielectric all around the center conductor, tuned for 50Ω. In the DIY world this only can be approximated by e.g. 4x 200Ω resistors in a cross pattern.

[szoftveres]

I'm able to approximate your measurement results with the below model - the only odd thing is the 18mm long SMA connector (QucsStudio could be using vacuum propagation speed by default, not sure what to change it to to match the SMA connector's dielectric)

[G0HZU]

I think you're expecting too much from it. I wouldn't expect such a DIY load to be used beyond a few hundred MHz, perhaps 1 GHz at most.

It is possible to get fairly respectable performance using 100R 0805 resistors. See below for an old plot of my original SMA 50R load that I made for my old HP 8714B VNA. This was made using a really good quality SMA end launcher connector and two 0805 100R resistors.

I made this the best part of 20 years ago and I spent some time selecting 100R resistors from various SMD manufacturers. Once I found the best ones, I selected two that gave the closest to 50R on a decent ohmmeter.

You can see it managed 40dB return loss at 3GHz when using a decent Agilent cal kit as the reference. Sadly, it isn't quite this good today. It has some wear and ageing effects. I really should make another one.

[shabaz]

That's a good idea to use a more suitable connector and 0805 (0603 probably wouldn't reach for those decent end-launch connectors with their tiny center-pins).

[joeqsmith]

... It doesn't event hit 1 GHz before the return loss drops below 40 dB. I was hoping to at least have a usable load up to 3 GHz (where my home VNA caps out) but I may have to rethink the design significantly.

Was the data collected with the unit assembled?  Was the photo taken with the load off the hex stock to show how it was soldered up?  With it having so much capacitive reactance from the start, I question the parasitics due to the hex stock. Sadly, you don't provide any details about how that was machined.  Nor did you show the inside of the hex stock where it mounts.   

If you took the data with the load detached as you show in the photo, then something else is amuck. 

[DaJMasta]

While I don't disagree with the recommendations already made, there could be a quick one to check: resolder the ground ends of those resistors.  They don't look that well wetted onto the resistor terminals themselves (or, at least, there's a noticeable pinch point between the terminal and the connector body), and you may not be getting the ground connection you're aiming for even if there is some contact/adhesion being made.

[W3AXL]

... It doesn't event hit 1 GHz before the return loss drops below 40 dB. I was hoping to at least have a usable load up to 3 GHz (where my home VNA caps out) but I may have to rethink the design significantly.

Was the data collected with the unit assembled?  Was the photo taken with the load off the hex stock to show how it was soldered up?  With it having so much capacitive reactance from the start, I question the parasitics due to the hex stock. Sadly, you don't provide any details about how that was machined.  Nor did you show the inside of the hex stock where it mounts.   

If you took the data with the load detached as you show in the photo, then something else is amuck.

I did have this suspicion, and I took measurements both with the load removed from the hex stock and without, and they were essentially identical. Maybe 1 or 2 dB different in spots.

The hex stock just has a hole drilled about 1/4 deep and wide enough to accommodate the resistors and solder without pinching anything.

While I don't disagree with the recommendations already made, there could be a quick one to check: resolder the ground ends of those resistors.  They don't look that well wetted onto the resistor terminals themselves (or, at least, there's a noticeable pinch point between the terminal and the connector body), and you may not be getting the ground connection you're aiming for even if there is some contact/adhesion being made.

Not a bad idea. I'm also planning on making a second load using an SMA-female flange connector to compare (less moving parts, etc). Plus, I'm looking at getting some better RF resistors to play with.

[joeqsmith]

Just an FYI,  PNA was calibrated using an 85033E.  Black is return loss for the cal standard I use for low frequency.  It is 2X1206 100 ohm mounted to FR4.  The load supplied with my original NanoVNA (yellow) works very well for the 300MHz the VNA is rated for. 

The benefit of the home made standard was to help compensate for some of the errors due to the PCB and connectors.

[joeqsmith]

And here are the results (note that the S-parameter tool I was using plots return loss as a positive number, so higher on the chart here is better):

Both METAS and AppCad are available for free and plot the data as expected.  Showing three different loads made with metal film axial parts and plotted with AppCad.   

One thing to keep in mind, if you use the low cost VNAs and the open source software with them, I ran into a problem with my software trying to import Touchstone files from QT where they write 0s for S21, which if you know your math, causes problems.   I have not played with other software enough to know if there are similar problems but if they copy one another's work there is certainly that potential.  These viewers can also have problems when presented with bad Touchstone files as well. 

[tomnut]

I've made SMA kits like this for years. I would have expected slightly better than this, maybe -30 dB up to 6 GHz (I didn't go much beyond that).

A couple of thoughts for getting higher in frequency though:
- try four 200 Ohm resistors?
- test it removed from the steel cavity, or enlarge the cavity behind the connector
- don't worry about the resistor tolerance, 1% is like -50 dB reflection coefficient.
- use Suhner / Telegaertner connectors instead of eBay cheapies
- look very carefully at the Smith chart, and then start fiddling with the solder around the centre pin, and possibly the ground, to adjust it.

For the last point, before you try to read the Smith chart, you need to port extend. To my eye your chart looks like it has a half-wave-at-10ghz of uncorrected cable.
- First calibrate with a better cal kit than yours
- then install your short, and adjust the port extension so that you get a clean short circuit on the screen. Cal kits have various reference plane positions, not necessarily at the back of the basic 4-hole SMA. I assume your short is made from copper tape flat on the back of the connector.
- now try the load, you should see just one tail of L or C, not a spiral.
- a spiral means you need a better connector.
- it's probably inductive. Add a blob of solder, or a short wire perpendicular to the Rs, or bring some ground closer to the pin.
- or add a small disk of something, to enlarge the pin to twice its diameter, before connecting the resistors (and switch to 0603, etc). We would do this with the centre pin of a PCB-mount connector, tweaking the size of the pad and the ground clearance for best VSWR.


Thank you for fighting the good fight, plotting Return Loss as a positive number.

[Co6aka]

FWIW... Connector construction matters, and each resistor adds a tiny bit of capacitance (and inductance) so you might be better off with one quality 49.9-Ohm resistor.

[eb4fbz]

Try a better quality SMA connector and 3x 150ohm 0603 resistors mounted upside down.

[virtualparticles]

You'll see some lead inductance from the resistors. You also have a short post that sticks up from the middle. This has an inductance of about 20 nH per inch. You can still use this, but you would have to database it using a golden VNA and apply the database in the cal kit table of whichever VNA you use.

[Randy222]

nanoVNA-H can do GHz ?

I would try to get LCR numbers of the small SMD resistors. Spec sheet might show it, but measuring it is a must.
What does the terminator as-is show on an LCR meter?
Then do the same on the bulkhead ground and pin from one end to the other to make sure the continuity is a good zero.

[Randy222]

The most beautiful SMD resistors I ever soldered were Vishay Sfernice CH series flip-chip 0603 resistors that my customer had specified by mistake, not realizing they weren’t in the signal path and thus didn’t need to be high-performance ones like those. They’re beautiful little white cubes that cost about $6 each. But they’re good to 50GHz, which is something an ordinary 0603 resistor can’t claim!

0603, rated 400mW. Wow, it's a tiny package for 400mW (68 x 33 thou.in.), I wonder how hot it gets at 400mW in open air.

[Randy222]

Just a side question.
With the level of  wanted "precision" in this DIY config, I wonder if some form of DIY laser diode spot welder, or resistance welding, would help?