Author Topic: [Newbie] On-PCB VNA calibration  (Read 3946 times)

0 Members and 1 Guest are viewing this topic.

Offline KoenTopic starter

  • Frequent Contributor
  • **
  • Posts: 502
[Newbie] On-PCB VNA calibration
« on: January 19, 2018, 07:11:47 pm »
Hello,

      I still have no clue where to ideally do the on-PCB calibration of a VNA with regards to the optimisation of an antenna matching network. I had luck until now but a recent error had me re-read some documentation on antenna optimisation from TI, Cypress and Agilent but I still couldn't figure the ideal.

      Assuming the PCB below with the chip's RF input pin expecting 50ohm, an u.FL connector for the VNA behind the pin, a ~50ohm coplanar wave guide, should I do the on-board open/short/load at point A, B or C ? Ground plane fill is hatched for readability.

Thank you very much,
Koen



 

Offline ThomasDK

  • Regular Contributor
  • *
  • Posts: 139
  • Country: dk
  • B.Eng. EE
Re: [Newbie] On-PCB VNA calibration
« Reply #1 on: January 19, 2018, 11:14:13 pm »
Right at the tuning network (B).
 

Offline KoenTopic starter

  • Frequent Contributor
  • **
  • Posts: 502
Re: [Newbie] On-PCB VNA calibration
« Reply #2 on: January 20, 2018, 01:14:25 pm »
Thank you for your answer !

Ideally, should the calibration be "at D" :

   Open
     D : NP (not populated)
     E : NP
     F : NP
   Short
     D : capacitor
     E : NP
     F : NP
   Load
     D : 50ohm
     E : NP
     F : NP

or "at F" :

   Open
     D : NP
     E : capacitor
     F : NP
   Short
     D : NP
     E : capacitor
     F : capacitor
   Load
     D : NP
     E : capacitor
     F : 50ohm

Thank you very much,
Koen

 

Online dmills

  • Super Contributor
  • ***
  • Posts: 2093
  • Country: gb
Re: [Newbie] On-PCB VNA calibration
« Reply #3 on: January 20, 2018, 01:27:35 pm »
I think I find myself in slight disagreement here as I would go for calibrating at A, as we do not want to de embed the length of feedline as it may or may not be 50 really ohms.

What matters is the impedance at the chip pin so we should have our reference plane defined at that point?

If the feedline is an accurate 50 ohms then setting the reference plane at the input to the matching network is useful for tuning the matching network as it means the effect of the feedline can be more easily ignored, but for checking the match to the chip you want the reference plane at the chip, surely?

Incidentally, that RF connector may or may not be sufficiently small for the footprint as a stub to be ignorable, I would have left it off and just placed a solder mask opening behind the RF pin of the chip so that I could tack on a bit of small hardline or such for measurement (Also makes the open and short cases easy).

Thoughts from the other RF folks?

Regards, Dan.
 

Offline ThomasDK

  • Regular Contributor
  • *
  • Posts: 139
  • Country: dk
  • B.Eng. EE
Re: [Newbie] On-PCB VNA calibration
« Reply #4 on: January 20, 2018, 05:15:55 pm »
I think I find myself in slight disagreement here as I would go for calibrating at A, as we do not want to de embed the length of feedline as it may or may not be 50 really ohms.
As long as the line is reasonable close to 50R it will be fine. If you move the reference plane away from the point where you do the actual tuning, things get "fun", as the line shifts the phase by an unknown amount. (Depending on velocity factor and length of the line).

What matters is the impedance at the chip pin so we should have our reference plane defined at that point?

If the feedline is an accurate 50 ohms then setting the reference plane at the input to the matching network is useful for tuning the matching network as it means the effect of the feedline can be more easily ignored, but for checking the match to the chip you want the reference plane at the chip, surely?

Ideally you should add a second matching filter at the chip, if the chip is not internally matched to 50R.
Incidentally, that RF connector may or may not be sufficiently small for the footprint as a stub to be ignorable, I would have left it off and just placed a solder mask opening behind the RF pin of the chip so that I could tack on a bit of small hardline or such for measurement (Also makes the open and short cases easy).
Agreed. An alternative is drilling a hole in the feedline and adding a SMA connector on the back of the board.

In any case, I would move the rf ic to the filter, and avoid the long feed line completely.
 

Offline ThomasDK

  • Regular Contributor
  • *
  • Posts: 139
  • Country: dk
  • B.Eng. EE
Re: [Newbie] On-PCB VNA calibration
« Reply #5 on: January 20, 2018, 05:21:15 pm »
Thank you for your answer !

Ideally, should the calibration be "at D" :

   Open
     D : NP (not populated)
     E : NP
     F : NP
   Short
     D : capacitor
     E : NP
     F : NP
   Load
     D : 50ohm
     E : NP
     F : NP

or "at F" :

   Open
     D : NP
     E : capacitor
     F : NP
   Short
     D : NP
     E : capacitor
     F : capacitor
   Load
     D : NP
     E : capacitor
     F : 50ohm

Thank you very much,
Koen



At "D", but a piece of copper tape as the "short" instead of a capacitor. Remove the soldermask at the opposite side of the line and short to both sides.
 

Offline TheUnnamedNewbie

  • Super Contributor
  • ***
  • Posts: 1208
  • Country: 00
  • mmwave RFIC/antenna designer
Re: [Newbie] On-PCB VNA calibration
« Reply #6 on: January 20, 2018, 05:51:45 pm »
Where to put the calibration depends on what level or precision you want. It also depends on the frequency, because at higher frequencies small imperfections can have a more significant effect. What frequency is this design operating at?

It would be easier to match to 50 ohms at the begining of the matching network in terms of understanding what is going on (the relationship between an increase in the value of one of the components vs it's effect on the smith chart). As ThomasDK says, if we have the line there as well it will be a significant phase rotation which means any issues will be "shifted" (ie, rotated on the smith chart) and we will potentially have a harder time understanding what component change we need to do the match.

However, the same applies to the IC: if we deembed the entire feed line we will not be "seeing" what the IC "Sees". After all, the purpose of the matching network is to make sure the IC sees the right impedance - we don't really care what is going on in between really, we just care about the S11 we get at the IC.

If the chip is in fact not 50 ohm you can consider not doing an additional tranformation to 50 ohm at the IC pins, and instead just keep that impedance going up to the matching network. Main annoyance here is that you have a harder time doing any probing since your equipment is all 50 Ohm.

In addition, is this a CPW or GCPW line? Either way, your network is quite a-symetrical and that might result in components performing differently than you would expect. CPW requires signal balance between the ground planes - if it isn't there with a matching network like this one, you aren't going to get a nice CPW mode propagating so your line might not be 50 ohms at all.

Ideally, check with a simulation tool to verify. This would also allow you to look at what will happen as a result of manufacturing tolerances, and if it needs to be considered.

The best part about magic is when it stops being magic and becomes science instead

"There was no road, but the people walked on it, and the road came to be, and the people followed it, for the road took the path of least resistance"
 

Offline E Kafeman

  • Regular Contributor
  • *
  • Posts: 107
  • Country: se
    • AnTune VNA software
Re: [Newbie] On-PCB VNA calibration
« Reply #7 on: January 22, 2018, 07:28:07 pm »
This is a kind of job I do several times each day. In most cases is it wide band or multi band matching.
What procedure and how to calibrate is a bit depending on if 50 Ohm is a requirement.
For embedded antennas is 50 Ohm as impedance seldom an optimal goal.

Assume VNA and its main cable is calibrated reasonably well.
As standard do I then add a thinner coaxial pigtail, ~100mm. One end with SMA and other end is soldered at point F.
Near F, direction antenna, is CU trace cut.
F is shortcut with an exacto to ground
Then do I set auto port forwarding and auto impedance calibration in a special PC software, special designed for this kind of job, AnTune. It reads data from VNA over LAN/USB/GPIB.
It do then compensate for time delay and impedance losses. Time delay is caused by added cable length, 100 mm, and also these few mm CU traces and pad between cable connection point and pad of point F have added.
Impedance losses are due to non ideal 50 Ohm cable in this 100 mm cable and CU-trace. Even size of PAD and internal PCB ground losses can affect tuning result if not proper handled.
Next, in AnTune select type of matching components, so that these non ideal behavior can be auto compensated for. Also slect preferred topology or let AnTune select optimal alternative.
Then can an actual measurement be performed. This is done by shortcut to point D. Trace in antenna direction is since previous cut. activate radio and measure TX and RX impedance. Save these curves in AnTune.
Be careful to not overload VNA if TX power is above what VNA can handle, typical is power up to 20dBm not any problem, for higher power must attenuation be added, before calibration of main cable.
If it exist additional TIS or TRP requirement, add them in AnTune.
Set resulting curve in AnTune as reference impedance. AnTune will use its conjugate for antenna impedance matching.
Now remove shortcut between D and F and repair the cutted trace against antenna with a small wire or solder blob.
Now is antenna impedance measured and in AnTune can it be directly read what component values should be used for the selected topology and type of component family.
Solder these three components in place and then confirm that Tx/Rx impedance now is a perfect wideband conjugate match of antenna impedance by once again remove solder blob against antenna.

At high frequencies, ++2GHz, and if it is measurable losses and delays between each matching component, do this last confirmation fail or is not good enough. In these cases, depending on situation but then is just the component at F soldered in place and measurement cable is moved just behind E and procedure is repeated, measuring antenna impedance and component F, to find optimal value for E, and finally component D.

Above procedure do maybe seem complicated, but it do never take more then a few minutes and final matching result can never be an surprise. Trace reactive loss and delay between D and actual radio is then not a problem. Radio internal impedance becomes optimal loaded even if it not is 50 Ohm, which it never is even if data sheet says so.
A big advantage is that AnTune requires almost no settings except selecting components, topology and frequency, and have a learning curve around two minutes.

AnTune calculates optimized matching network live. Result is presented at PC screen as fast as it can be delivered from VNA. That makes tuning directly at antenna very easy as final matched result can be seen while adjusting antenna shape. No else software is able to do this.

My opinions about AnTune is very biased as I am the developer of the extremely fast matching algorithm.
AnTune was developed by me with the only goal, to make embedded wide band antenna design with highest possible efficiency and optimized matching network design as fast and simple as possible. Both these, best matching and best efficiency are depending on each other so that is why I prefer above described procedure, which includes measuring both radio and antenna impedance and include internal PCB traces in these measurements. Traces are maybe short and with an ideal impedance close to radio impedance. Radio impedance is maybe very close to data sheet numbers . It have just not happen for me yet.

Another important thing is that when a optimized network is decided can AnTune save a document including all relevant curves and BOM with art.no for actual tuning components.
That is important as especially inductors of different types and brands, even if same inductive value, can have very different behavior at higher frequencies.
« Last Edit: January 22, 2018, 07:49:22 pm by E Kafeman »
No signature
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf