Desiging a budget high-speed RF Power Sensor...

[wilhe_jo]

Hi Folks!

It's time to present my progress on designing a high-speed RF power sensor.

So first, let me quickly explain why the hell you would want to have something like this:
I run a consulting business where I do pre-compliance EMC and Radio in my own lab.
Regularly, I get some IoT device that (of course) needs some fancy antenna that is not covered in some test-report.
For now, I mostly did measurements relative to known samples prior sending them to some lab.
However, I wanted to improve on this.... at least in the 2.4G band.

You may know ETSI EN 300 328.
To measure the TX power, you should use a power-sensor that samples with at least 1MSPS, has a sensitivity of at least -40dBm and measures RMS values.
So I put a sensor with those specs on my "shopping list" and waited for some deals... and none came along...

Last year, I had an enlightening talk with "the guy" from LUMILOOP at a symposium here in Austria.
I learned, that these fast linear-in-db rms detectors are quite nice besides their need for calibration.
This was when I decided to take "some" hours (if I knew back then how many hours...) to give them a try.

My choice was a AD8363 detector IC that should provide 60dB of dynamic-range and up to 6GHz.
The main reason for this was it was available at JLC, and the sig-gen with the highest frequency gets up to 6.4GHz.
If you carefully study the datasheet, you'll quickly notice, that the performance of the RMS power detection quite nicely matches EN 300 328.

However, there's a huge elephant in the room: Thermal stability!
As a measurement instrument, +-0.5dB (the IC's spec on the first page) may be a little on the high-side, but digging deep into the datasheet, you quickly learn, that temperature compensation is not broad-band. It is pretty much limited to maybe 1GHz of bandwidth (there are resistor values for several bands in the datasheet) where the characterization was done.

Last part of the puzzle was the interface chip that magically gets the 1MSPS of data to the PC where the bursts get processed.
Some of you may have seen the threads on the CH32V307 here on the forum. This very interesting chip has integrated Ethernet PHY AND an integrated high-speed USB PHY!
Together with the integrated 1MSPS ADC, this means, a single chip solution to do everything in the digital domain

The design got to my lab in November last year and got shelved until mid of December.
Some small mistakes happened, but I still work on Rev1 hardware - so nothing really terrible

As you can see, I fitted everything in one of these cheap 2-part aluminium extrusions... maybe not the best idea for sensitive RF equipment, but I thought it should be fine for a very first test.

The most demanding task was to get the CH32V307 to work.
This device is very, very interesting, but the supplied libs and datasheets are lacking in the few spots that were important to me.
It took quite some time to figure out how to transmit the sampled data with tiny-usb (mostly) reliable to the PC.

I mentioned in several spots, that I'm working on a quite versatile test automation software. In fact, my whole lab's already running on that.
So it was pretty obvious, that everything needs to be included into this framework... that'll get its own post in the mid-future as it gets open-sourced eventually.
Anyhow, until recently, USBTMC was my way of communicating with the device.

While calibrating CW signals (of cource, that's automated), I got quite some discrepancies.
Most of them were solvable by just doing 3 measurements and using the median.
For a real test-instrument, this isn't really the way it should be.
Luckily, meanwhile, my troubles with tiny-usb were resolved. This means, my next step was to implement some kind of "scope" interface... to measure Wi-Fi signals, some streaming-data-interface was needed anyways.

What you see is my LTE router sitting next to my development setup happily causing EMI... 2.5dB spikes on a -30dBm signal...so around 2µW
Using the streaming interface, I'm able to do some magic... sampling 1 second, sorting the values, killing a good portion of them and taking the average helps quite a lot.
Long story short, I can get the standard variation down to 20mdBm by this very simple trick

With this encouraging developments, I thought it would be interesting how far we could go with some shielding...

As I mentioned above, I was a worried about the thermal properties (I must have been wrong by some decades with my napkin-calculations!).
This fear drove me into the insane thinking, that I could get away without ground-pour in the RF section besides what's needed for the microstrip.

The 10dB attenuator you see in the photos does the trick with matching and shifts the signal a convenient level. I figured -40 - 5dBm would nice.
Considering the characteristics in the datasheet, you'd try to stick to levels below -5dBm. So the 10dB attenuator nicely solves 2 problems.

So here we are, I have some working prototype of a power sensor that's able to do >1MSPS, gets down to -40dBm with ease and some software framework that just waits to get the EN 300 328 procedure for TX power measurements implemented

Next steps are improving the calibration procedure, ordering Rev2 and thinking about how to publish the whole design.


73

[tatel]

(...) I decided to take "some" hours (if I knew back then how many hours...) to give them a try.

(...) as it gets open-sourced eventually.

Release soon, release often