20dB 'RF' attenuator - seeking feedback to improve

[rf-messkopf]

Interesting thread, which shows that it is not at all easy to make good attenuators at the board level from standard chip resistors. I became curious and ran a quick impedance measurement on four standard 0805 resistors (27 Ohms, 100 Ohms, 510 Ohms, 1000 Ohms) up to 1.8 GHz.

The lower resistor values tend to increase with frequency, whereas the higher ones tend to drop. The sweet spot seems to be somewhere around 100 Ohms. It becomes a complex optimization problem to manufacture a flat attenuator from these.

For the measurement the resistors were soldered to a PCB test jig. The calibration planes of the measurement lie exactly at the outer edges of the SMD pads. The analyzer was calibrated by TRL with TRM extension at low frequencies. Since the impedance standard in this calibration is the wave impedance of a trace on the test jig (a grounded and fenced coplanar waveguide, manufactured in a cheap and non-impedance controlled pool process), the absolute values of the resistances indicated in the analyzer printout may be off by a few Ohms (the 100 Ohms resistor measures almost exactly 100 Ohms at DC, so that's roughly 5% high at low frequencies). In fact, I measured the coplanar waveguide impedance a few Ohms below 50 Ohms, so the values should read a little high. But the general trend is clearly visible anyway.

See here for a picture of the test jig: https://www.mariohellmich.de/projects/trl-cal/img/trl-fixture.jpg

[ogden]

Interesting thread, which shows that it is not at all easy to make good attenuators at the board level from standard chip resistors.

Your TRL Calibration and cheap SMA cal kit characterization projects are even more interesting read. Well done! 

The lower resistor values tend to increase with frequency, whereas the higher ones tend to drop. The sweet spot seems to be somewhere around 100 Ohms.

We can conclude that best "DIY 50-Ohm load standard" could be made out of 2x100 Ohms, not 1x50. Thanx for valuable input.

[joeqsmith]

Welcome to the discussion Mario and again, thank you for all the help you've provided.     

I wonder how other package sizes would behave and also what happens at higher frequencies.   After making those changes to that low cost attenuator, the world above 2GHz seems to come with a whole new set of problems. 

Thinking about your test board, assuming that this was used as the standard and that the reference seems to be made from 2X100 ohm 0805s, is this biasing the the results?

[joeqsmith]

Looking at 0603s with the Nano sweeping from 50K to 1.5GHz.   Values are 24.9, 100, 499 and 1K ohms.   There does appear to be an upward trend with the lower value and opposite with the higher values.   

[rf-messkopf]

[...] and again, thank you for all the help you've provided.     

You are very welcome, Joe.

Your TRL Calibration and cheap SMA cal kit characterization projects are even more interesting read. Well done! 

We can conclude that best "DIY 50-Ohm load standard" could be made out of 2x100 Ohms, not 1x50. Thanx for valuable input.

Thank you, that's appreciated.
Unless the resistors are dedicated RF resistors, two 100 Ohms in parallel should indeed be more flat.

I wonder how other package sizes would behave and also what happens at higher frequencies.   After making those changes to that low cost attenuator, the world above 2GHz seems to come with a whole new set of problems. 

Thinking about your test board, assuming that this was used as the standard and that the reference seems to be made from 2X100 ohm 0805s, is this biasing the the results?

The board only works up to about 1.8 GHz because then the phase shift of the line approaches 180 degrees and the TRL calibration breaks down. For higher frequencies I would have to make a new test board with another shorter line and then use multiline TRL for calibration.

Concerning your second question: The primary impedance standard on the test board is the wave impedance of the line, not the two 100 Ohms resistors in parallel. They have not even been used for calibration. I did measure a match standard in addition to the TRL calibration standards, but that is not really necessary when measuring from 50 MHz onward. It would only serves to extend the calibration to lower frequencies where the phase shift of the line approaches 0 degrees. It should have almost no influence on the displayed results. Moreover, in an independent test I have measured the wave impedance of the line and found it to be about 6% below the nominal 50 Ohms. That matches the measurement result of the 100 Ohms resistor, which reads about 5% high.

I hooked the board to the analyzer again with the 100 Ohms 0805 resistor in place. The attached plot now also shows the real and imaginary components of the impedance. The resistor becomes slightly inductive at higher frequencies.

I also soldered a 100 Ohms 0603 resistor to the 0805 footprint of the test board. There is virtually no difference to the 0805.

[joeqsmith]

Sorry, I completely missed the TRL cal.   With the 100 ohm being the most stable of the four, I wouldn't expect to see as dramatic effect using the 0603 compared with the 1K.   With the Nano, it may be difficult to see any sort of trend. 

[joeqsmith]

Using the Nano to log 20 sweeps from 50KHz to 1.5GHz using the 1K ohm 0603 compared with a 1K 1206 package.  At least with the Nano, there appears to be little difference.  Your lab grade VNA may show otherwise.

[rf-messkopf]

With the 100 ohm being the most stable of the four, I wouldn't expect to see as dramatic effect using the 0603 compared with the 1K.   With the Nano, it may be difficult to see any sort of trend.

Yes, one would have to run more extensive tests with several values for each size. But that becomes tedious since the resistor needs to be soldered down each time, though that is the most faithful way of measuring.

Another question is how such measurements could be used for attenuator design. Since the TRL cal using the test board puts the reference plane right at the edges of the SMD pads, one could import the measured S-matrices into a simulation program, add traces (e.g. microstrip) and vias, and calculate or optimize the overall response. Again a tedious task. And it will never become as good as a high performance coaxial attenuator: https://www.mariohellmich.de/projects/sma-cal-kit/img/20db-att-meas.png.

[joeqsmith]

With the 100 ohm being the most stable of the four, I wouldn't expect to see as dramatic effect using the 0603 compared with the 1K.   With the Nano, it may be difficult to see any sort of trend.

Yes, one would have to run more extensive tests with several values for each size. But that becomes tedious since the resistor needs to be soldered down each time, though that is the most faithful way of measuring.

Another question is how such measurements could be used for attenuator design. Since the TRL cal using the test board puts the reference plane right at the edges of the SMD pads, one could import the measured S-matrices into a simulation program, add traces (e.g. microstrip) and vias, and calculate or optimize the overall response. Again a tedious task. And it will never become as good as a high performance coaxial attenuator: https://www.mariohellmich.de/projects/sma-cal-kit/img/20db-att-meas.png.

What's the Mfg and PN of the attenuator in your plot?  That's a very nice bit of hardware!!   

To answer your question, there could be a few reasons to roll your own.  The requirements may warrant the cost or be such that its easily achieved.  Perhaps it needs to be integrated as part of a larger design and on the same PCB.  In the case of this thread, I assumed OP was wanting to learn more about it, which is certainly a valid reason as well.   

Just for some sort of sanity check, the attached plot is using the same setup used to cal the Nano. This is not the newer standards I am attempting to put together.  I'm still waiting on my friend to send over those parts they characterized.   We can now see some difference between the two resistor packages. I would imagine if I ran it higher the difference would become more apparent.

[enut11]

No, you did not place connectors as I told - when no parts of connector touch bottom side of PCB. Is it so that you use double sided PCB with top copper layer still in place, very, very close to central pins? - Not good.

@ogden
Since I have only 2 SMA connectors left, I want to get it right. I will be using 0805 type SMD resistors.
I am building the 20dB 50 ohm Pi attenuator using four 120 ohm to ground and two parallel 510 ohm in series with the SMA center pins.

Now, the center pin to ground plane distance is ~3mm, ie too far for the 2mm 0805 devices.
The distance of the center pin to (grounded) side pins is ~2mm.

I seems the only way I can build this attenuator is off the ground plane. Is this what you intended?
enut11

[ogden]

I am afraid that connector do not lay flat, central pin is tilted away from PCB. My "cheap" SMA edge mount connector for 1.6mm PCB measure 2.5mm which is "good enough" to put 0805 in-between. You may want to desolder this connector, then clamp or tie it down to new PCB in proper position. Double check that it stands properly & still, ONLY THEN solder. BTW you got 1.6mm or 1.8mm PCB connector? Latter could be "too big" indeed.

Other option for you would be to copy "low cost attenuator" PCB by etching/cutting/dremelling it.

How did your attenuator shown in this post measure with NanoVNA?

[rf-messkopf]

What's the Mfg and PN of the attenuator in your plot?  That's a very nice bit of hardware!!   

That is a HP part no. 85053-60001 attenuator with 3.5mm air dielectric connectors, rated to 26.5 GHz, see the attached picture (not easy to take a photograph due to the glare). It it part of the HP 85053A verification kit. I don't know exactly but I think it will set you back by a four digit amount when you buy it separately.

It would surely be possible to get an even more flat response by a better test setup. It was connected by Sucoflex 104A cables with precision SMA, which are good but not intended for precision VNA measurements, and are not designed to be phase stable. At 0.01 dB/division cable flexure has a significant influence on flatness, and connector torque becomes a real concern (and at 0.001 dB/div you can sometimes see the effect of thermal expansion when you touch your hardware). The mating of 3.5mm connectors with precision SMA is also a compromise.

Should I become bored during the shutdown I will get the 3.5mm test port cables out and see if I can get a better result.

To answer your question, there could be a few reasons to roll your own.  The requirements may warrant the cost or be such that its easily achieved.  Perhaps it needs to be integrated as part of a larger design and on the same PCB.  In the case of this thread, I assumed OP was wanting to learn more about it, which is certainly a valid reason as well.   

I wonder at which point you would run into repeatability issues with different resistors when you design with measured S-matrix data, and different PCB batches. FR4 is not the most stable material when it comes to RF, and you may want to use a Rogers material. But it is certainly true that there are applications where attenuators from standard components make sense.

[joeqsmith]

Thanks.  I figured it had to be some sort of metrology grade component.

I came across this YT channel and started watching a few of his videos.  This one talks about a study that was done using several 1206 resistors.  It may be of some interest.   
https://youtu.be/D5mIJzKAMSI?t=664

His most recent video adds a bit more. 


Getting things somewhat true when soldering or using glue,  I will normally block up the parts.  The M/F was made by soldering all four prongs, then inserting a bit of teflon where they PCB would normally be.  A shield was then soldered to the lower half.  It could be covered with copper foil tape to seal it. 

[enut11]

How did your attenuator shown in this post measure with NanoVNA?

Thankfully my latest attenuator (from post #147) seems to measure up well against the silver 6GHz commercial 20dB attenuator (post #131 and #135)
NanoVNA plots look very similar in the 300MHz and 900MHz sweeps. Again, beyond 300MHz, the Nano seems to be out of its depth.

Currently working on an improved SMD version suggested by @ogden. Waiting for more SMD resistors to arrive.
enut11

[joeqsmith]

I've been thinking about your TRL fixture but am VERY ignorant about it and have been doing a little reading.   The following link has a spreadsheet to calculate the lengths and number of them:     

https://www.microwaves101.com/encyclopedias/trl-calibration

They also offer some practical guidelines, like using precision connectors.   Most articles talk about the TRL obtains the most accurate results at the higher frequencies.  I find it odd they offer standard SOLT kits rated for 10+GHz.     

I've been working on setting up to be able to run experiments in the 2GHz+ which is where these new SOLT standards come in.   While it's been a bit of a struggle even with your help,  things look promising.  Now based on what I have read, I would like to try a few experiments using TRL.   It seems like one way to do this may be to use hardline coax to make the test fixture and cal standards.   

Before I go down this rabbit hole, could you have a look at that spreadsheet and give your thoughts.

[rf-messkopf]

Most articles talk about the TRL obtains the most accurate results at the higher frequencies.  I find it odd they offer standard SOLT kits rated for 10+GHz.     

The point of using TRL with the test board is not precision, but to place the reference planes on the PCB and to calibrate the coaxial-to-coplanar waveguide transition out. This is the reason why TRL is usually the method of choice when a calibration in a planar geometry is desired. For precision measurements at higher frequencies one would, however, not use connectors, but a probe. This also works on wafers, and is used for on-wafer VNA measurements in MMICs.

You can also do TRL in coaxial geometries using airlines. This is, for example, done in metrological contexts since airlines can be manufactured to extremely tight tolerances. Moreover, in this way scattering parameters can be traced to dimensional standards, and that's what national metrology labs use TRL for. However, airlines are delicate and tedious to handle, and very easy to break. So you don't want to do TRL when you only need to calibrate to a coaxial connector unless extreme accuracy is required at high frequencies. In almost all cases you simply want a good quality SOLT/TOSM kit.

Another feature of TRL: Cal standards do not have to be fully known (as opposed to OSM). For example, the line can have an arbitrary transmission coefficient (which includes loss); only the wave impedance matters. Similarly, the reflect standard can have (in theory) an arbitrary nonzero reflection coefficient (but which is assumed to be equal at each port). In fact, the TRL mathematics allows to obtain the transmission coefficient as well as the reflection coefficient from the underlying system of equations. But normally VNA firmware doesn't output them, but this could be done offline with your own software written in e.g., MATLAB or Octave. This is sometimes done in a two-tier calibration to measure wave impedance in planar geometries.

There is a frequency restriction due to line length: The phase shift of the line must be kept away from 0° and 180°; for wider bandwidths several lines have to be used. Extension to low frequencies (in fact, to DC) require an additional match (this is then called TRM). Modern VNAs automatically take care of that.

See the writeup here for some more details: https://www.mariohellmich.de/projects/trl-cal/trl-cal.html, and also the quoted literature.

They also offer some practical guidelines, like using precision connectors.

Yes, the test board shown above depends on the equality and repeatability of the connectors. Since the reference planes are on the PCB, the assumption is that the path from the VNA port to the reference plane is equal for each cal standard, as well as for the DUT connection. And this includes the connectors and the connector-to-coplanar waveguide transition.

I don't think that with my test board I have too much of a repeatability issue (more a feeling than a fact) and results are fairly reproducible, but that's only up to 1.8 GHz with half decent connectors (about €8 each). For more precision and/or higher frequencies you will have to invest in 3.5mm screw-on launchers (e.g. from Huber-Suhner), be prepared to pay at least €150 each. The board dimensions must also be sufficiently accurate.

Before I go down this rabbit hole, could you have a look at that spreadsheet and give your thoughts.

Sorry, I have a ban on Microsoft Office here and can't take a look at it. But you only have to observe the phase shift restrictions with the line standards, and when using two lines allow for sufficient overlap. You can also do the math with paper and pencil. Or start with just a single line when you don't need a wide frequency band.

[joeqsmith]

At the lower frequencies, I wonder if you really gained anything over just using the SOLT setup on that same board.   I assume to made the board to run this test.  Do you remember if there was any advantage? 

Sorry, I have a ban on Microsoft Office here and can't take a look at it. But you only have to observe the phase shift restrictions with the line standards, and when using two lines allow for sufficient overlap. You can also do the math with paper and pencil. Or start with just a single line when you don't need a wide frequency band.

No problem.  If I use the numbers screened on your PCB, the spreadsheet looks like it wants two lines.  One at 115.9 the other 28.4mm.  It seems like a pretty big discrepancy. 

I was thinking with using coax, I would get better results than with FR4.   I was also thinking that in the case of the attenuator, it could be built inside the coax easy enough.       

[rf-messkopf]

At the lower frequencies, I wonder if you really gained anything over just using the SOLT setup on that same board.   I assume to made the board to run this test.  Do you remember if there was any advantage? 

I suppose you mean using SOLT elements on the PCB right in the fixture where the 0805 component under test goes. Because calibrating to the connector by a standard coaxial SOLT cal kit will not yield accurate results since the DUT is connected by pieces of transmission lines to the connectors, and these will transform.

No, I have not tried to use SOLT within the fixture. That would be difficult because, e.g., the open would be undefined due to its fringing capacitance, and the load would be a resistor with unknown RF properties. In contrast, TRL does away with these problems, and allows for very precisely located reference planes right at the outer edges of the SMD pads. You will not have that level of control over the reference planes with SOLT in this situation.

If you wanted to use SOLT for some reason I would rather calibrate to the connector in the usual way, and then do a port extension to the DUT in the fixture. The proper delay can be found by adding a temporary short to ground at the pads, and measuring the phase delay. A modern VNA can measure and compensate that delay automatically, and even include losses. But that will certainly have a noticeable influence on accuracy (especially phases), unless you are only interested in the S21 magnitude.

As already mentioned, here the essential problem remaining with TRL is that the wave impedance of the line on the PCB is not exactly 50 Ohms, and that will affect the accuracy; you are effectively not referencing your S-matrix to a system impedance of 50 Ohms, but to 50 Ohms +/- x Ohms. But the line impedance could be measured independently and added into the cal kit definition in the analyzer for correction.

No problem.  If I use the numbers screened on your PCB, the spreadsheet looks like it wants two lines.  One at 115.9 the other 28.4mm.  It seems like a pretty big discrepancy. 

Over which frequency band do you want to calibrate? And in which medium (microstrip, CPW, coax), because that will determine the relation between phase shift and frequency.

I was thinking with using coax, I would get better results than with FR4.   I was also thinking that in the case of the attenuator, it could be built inside the coax easy enough.       

That will depend on your machining skills and your ability to manufacture resistive elements on dielectric substrates, e.g. by thin film technology, that can be put in a coaxial structure.

[joeqsmith]

I just used the numbers that were on your PCB and blog and plugged them in.   

At these lower frequencies, I have never characterized my circuit board SOLT.   Running various tests over the years has never been a concern.   But then again, I am only looking for ballpark numbers.   

I assumed with your board, you had the Short, Open, Match and Mismatch all built up on the board with the same reference.  I assumed you could use the SOL to then measure the Mismatch, then use the TRL to measure the mismatch as well.   

The problem with the Normal SOLT standards is in my case, I would normally mount the DUT on FR4.  The delay is only part of it.  That's the benefit of building it all up on FR4 (or whatever medium used). 

I did make up a few sections of coax to try running the TRL.  Even with my homemade sections, it works well enough to make some basic measurements.  ***Just to add, a MWM 18GHz 50 ohm terminator measures +/-5 ohms up to 4GHz.   So, I'm not suggesting this is great but for some hand made up, unmeasured sections of coax that were not even torqued, I'm amazed it worked this good.   

Building the attenuator into the coax would be no different than building it onto FR4.  It could still all be done with surface mount parts.     

[rf-messkopf]

I assumed with your board, you had the Short, Open, Match and Mismatch all built up on the board with the same reference.  I assumed you could use the SOL to then measure the Mismatch, then use the TRL to measure the mismatch as well.   

Okay, I tried that, see the attachment. I used the open, load, short and thru on the board for a standard SOLT cal, assuming that all standards are ideal, including the thru (i.e., zero delay and zero loss). Then I measured the 100 Ohms resistor soldered to the series fixture and displayed the impedance magnitude like shown in previous posts. For comparison, the same measurement is shown with the same TRL calibration as previously.

You can see that the SOLT (or TOSM) trace shows quite some ripple. This is clearly an artifact and probably due to phase errors incurred by the imperfect open. But I expected it to be much worse (it will certainly get worse at higher frequencies). The phases are essential for the calculation of the impedance Z from S21, even though only the magnitude of Z is plotted, since
\[Z=Z_0\frac{2(1-S_{21})}{S_{21}}.\]
And, as discussed previously, you can see the line impedance being about 5% low, which means measured impedances read about 5% high. You can also see that the DUT impedance obtained from SOLT increases towards 1.8 GHz, where in reality it falls off. This is because the two parallel 100 Ohms resistors on the board that were used as the load standard for SOLT, and their resistance also falls off towards 1.8 GHz, which means that a calibration using them will show impedances too high towards high frequencies.

All in all this demonstrates the advantage of TRL over SOLT in this case.

Also attached is a S11 measurement of the mismatch standard on the board with SOLT and TRL for comparison.

I did make up a few sections of coax to try running the TRL.  Even with my homemade sections, it works well enough to make some basic measurements.  ***Just to add, a MWM 18GHz 50 ohm terminator measures +/-5 ohms up to 4GHz.   So, I'm not suggesting this is great but for some hand made up, unmeasured sections of coax that were not even torqued, I'm amazed it worked this good.   

That's very good and surely gives confidence that TRL works for you in principle. Although using the MWM load with SOLT would most probably yield more accurate results, provided you have well corrected opens and shorts. These cables are clearly not intended to be calibration standards.

Building the attenuator into the coax would be no different than building it onto FR4.  It could still all be done with surface mount parts.     

Okay, I thought you were thinking about machining a "real" coaxial attenuator like the one pictured in this forum thread: https://www.mikrocontroller.net/topic/477931#5917781. The resistive element is on the horizontal disc sitting in the groove within the cylindrical center part, and is contacted at its face side by the two pins in the attenuator and caps, which also contain the connectors. I feel like sacrificing one of my attenuators to take a look inside. 

[joeqsmith]

Thanks for running it.   The results are not what I was expecting.  I thought that the SOLT would show tighter grouping.   

I'm still waiting on the parts to try and characterize my SMA standards.   

I remeasured my coax and suspect this is part of the problem with my TRL.  I'll recalculate the time based on the measure lengths and try it again with things torqued.   You may also note I made both an open and a short for the reflection.   I only tried one so there's a bit to do. 

For some added humor, my quick test jig to keep the coax centered.   Shown with a 1206 laying in there.    There's about 2mm of plastic tube between the coax and thumb screw.     

[enut11]

Just received my green 'Multi-Atten' board that looks similar to the one that @joeqsmith tested in Reply #122.

NanoVNA tests compare my 20dB MicroAtten to the 20dB section on the green Multi-Atten board at both 1-300MHz and 1-900MHz.

I welcome comments on what you see, especially the trace notches on my MicroAtten.
enut11

[ogden]

NanoVNA tests compare my 20dB MicroAtten to the 20dB section on the green Multi-Atten board at both 1-300MHz and 1-900MHz.

Your 'Multi-Atten' results are so much worse compared to @joeqsmith that I suspect that there is some problem with calibration or your NanoVNA. You are advised to watch @joeqsmith youtube videos about NanoVNA, then SOLT-calibrate before each measurement to be sure that it is indeed attenuator you measure.

[joeqsmith]

Using what I am guessing is the same model 20dB attenuator, the cables and standards that were supplied with my NanoVNA, running off battery power and sweeping from 300KHz to 300MHz. 

S11 (5dB/div), S21(5dB/div) and SWR

Also shown is my cobbled up shunt test jig with 1206.  Like the coax TRL references, I am not expecting very good data off of these. 

The Keysight standards have been shipped.  Rather than using components that were characterized with these standards, the plan now is to follow Mario's lead like I did with the N standards.  The Keysight set doesn't have the airlines.   This should at least get me to the point where I can finally use this new boat anchor to get some decent data. 

[enut11]

NanoVNA tests compare my 20dB MicroAtten to the 20dB section on the green Multi-Atten board at both 1-300MHz and 1-900MHz.

Your 'Multi-Atten' results are so much worse compared to @joeqsmith that I suspect that there is some problem with calibration or your NanoVNA. You are advised to watch @joeqsmith youtube videos about NanoVNA, then SOLT-calibrate before each measurement to be sure that it is indeed attenuator you measure.

Thank you @ogden. After Cal, the 1-300MHz plot for my MicroAtten was much improved. Live and learn!
enut11