When your board becomes an antenna

[jmw]

I've been working on an active probe project for a while, and now almost everything is looking good except for an unexplained dip in the transfer function around 1.7 GHz  . I can't account for it using a lumped or transmission line model of the tip, so I tried simulating the board in openEMS, modeling the input network but considering the rest of the board as ground plane. At the tip, there's no coax or waveguide to contain the fields, so they just spread out and it looks like it is radiating and also reflecting off the far end of the PCB. The s-parameter model of the input shows some ripples from the reflection and a dip that could be what I'm seeing on the prototype. Are there ways to test this theory, or design features that would reduce the common mode coupling and reflections? Would an RF can around the amplifier and input do anything? The dip/ripple frequency in the simulation changes with the board length (76 mm) but I don't think I can make the actual board much shorter.

[RoGeorge]

I've read resistors (instead of vias stitching) at the edge of the PCB should extinguish edge reflections, but I didn't try:
https://www.edn.com/keeping-rf-from-leaking-out-the-edges-of-your-board/

[selcuk]

Did you try placing vias on the upper edge? That side is empty in the image. Can you simulate again by decreasing the pitch of vias on the bottom edge? They are far from each other.

How many layers are there in the board? Is there a power plane in the inner layers?

[coromonadalix]

have you tried with a shielding cage

i dont know if you have impedance issue, capacitance, ...    vias stitching is normally better ...  and yes   how many layers, 

ground in the inner layers ?? or it's a 2 layer pcb ?

[jmw]

It is four layers:
top signal/ground
L2 ground
L3 power/ground
L4 ground

In the simulation I stitched all layers together at essentially every grid intersection 

[MarkT]

Model the shield of the cable running off the back of the PCB - the RF currents will run down that and perhaps reflect a lot less.

Is the probe grounded at the tip to the excitation source?

[jmw]

The left wall in the picture is metal/PEC, and the excitation is a lumped resistance between this plane and the tip. The probe ground connection touches all the way to the wall.

[tszaboo]

BUF802 with a capacitive voltage inverter with integrated LDO?
It's such a fascinating project, I also did the schematic for a design just like this.
You can test the radiated emission issues with a near field probe.

[jmw]

Yes, good read of the PCB
The schematic (and related stuff) is here: https://github.com/jmwilson/oshw-active-probe/blob/master/board/Mfg/Schematic%20Prints.PDF

[tszaboo]

I was thinking of using the LM27762 for the power. But this works as well. Good luck with the project.

[T3sl4co1l]

Try a band of flexible rubber ferrite-loaded shield stuff wrapped around the board?

Could also try a CMC at the tip, but it needs to be good enough characteristic impedance to handle the diff pair properly.  Which IIRC you want as short as possible, so, that's a thing.

You might also add some intentional CM impedance, just with a very short transformer/CMC, and resistor, to make a compromise between mode conversion and signal gain.

Tim

[jmw]

After some accidental experimentation, I think I can now say the source of the strange dip is ... my hand  As long as my fingers are away from the area forward of the bulk caps, the response looks like this. There's a point where the signal level rolls off quickly near the tip resonant frequency, but it doesn't look like a notch anymore, and the overall bandwidth is still very good. It wasn't in the simulation because I wasn't in the simulation... and so far I've learned getting a repeatable and reliable measurement setup for probes is a bear.