2.5GHz resonance problem

My service monitor gives me a couple of issues on some frequencies, to verify this, I placed a current meter inline with the circuit to see if there was a noticeable change at the same place as on the spectrum sweep. Of course, the change most likely is minimum but enough that it should tell you if you actually do have a resonator on board somewhere loading down your signal.

Jamie

20 years ago, I had a computer motherboard that had resonances in the 400MHz range with harmonics going up as high as I could measure. Turned out to be the ground and power planes resonating. I tried adding/deleting caps. I could change the Q a little, but couldn't change the frequency much. The impedance was so low that adding discrete parts didn't do much.

I got curious and TDR'd and frequency swept some other production boards. They had the same problem.

I decided I was trying to fix something that wasn't broke.

You don't have that luxury. You probably don't need a power plane.

I'd try a TDR on the output looking back...or a directional coupler and sweeper. Another thing to try is to sweep the power and ground planes (powered off) and look at the output connector to see if anything is coupling in. Or stuff the resonant frequency in the output and look for signal on the board. But that's gonna be a difficult probing problem.

Got a pic of the PC board?

--

John Larkin         Highland Technology, Inc

jlarkin at highlandtechnology dot com
http://www.highlandtechnology.com

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom laser controllers
Photonics and fiberoptic TTL data links
VME thermocouple, LVDT, synchro   acquisition and simulation

I'll guess layout issues. Lambda/2 for 2.487GHz is ~2.373 inches (61 mm) and for 2.287GHz is 2.6 " (65.6 mm). Look at your board traces around those lengths & relevant multiples thereof.

Whomp up a 2-3 GHz RF sniffer with a half-wave antenna. Wave it slowly over a board in operation. Look for dips in the output amplitude when you align the sniffer antenna with the guilty trace(s) so as to detune them. Picking them out shouldn't be hard.

Fix? Dielectric constant of FR4 is what, four point not much? Maybe Dremel out some of the board material next to suspicious traces to spoil the resonator Q.

Mark L. Fergerson

(see

2

I use 4.6. Some people claim 4.8. It's not well controlled.

--

John Larkin                  Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

This is the HF part:

The next layers are ground and power. I already tried to make the 'S' shorter with a wire but no difference at all.

--
Failure does not prove something is impossible, failure simply
indicates you are not using the right tools...
nico@nctdevpuntnl (punt=.)
--------------------------------------------------------------

Have you contacted Doug Smith? He may have even addressed something that relates to your problem in one of his monthly news letters.His articles are full of practical tips and DIY test equipment ...into the GHz ranges.

Assume not any test equipment artifact,

just looked at the PCB. some nitpicking: 4-layer?! I'd go to 6 or even 8 layer to get the layers closer together. FR4?? that stuff has horrible dielectric above 1GHz, especially if 20 mils thick . I know, less than an inch losses are not 'too' bad. But, there may be quite a difference in Zo, which just happens to go the wrong way for your circuit - Even Vswr of 2:1 can play havoc when all line up against you. The cutouts around GND vias look huge. Your output trace does NOT look like it would have uniform Zo along it. Was the PCB built with controlled Zo spec? Have you TDR'd looking back into a PCB with no drive chip, just to see what your SMA transition to matching network looks like? What is the trace running parallel to your output strip? How much coupling can occur there?

From that dip in the output it does look like the chip's power is starved, and/or the output drive power is starved. which means, nothing comes out the output trace, so no matching will improve that. Fun circuit, thanks for sharing!

I can't tell from picture whether this is sitting on a large area or truncated in the pic. What are the 'natural' resonance frequencies for your board? What happens if you extend the PCB BND planes outware, making larger with a 1 inch boundary around your active board area. [same layout] NOTE:GND planes are only GND planes *IF* they extend to infinitiy.

Is that 'S' shape even 50 ohm trace?

I use femm 4.2 to calculate trace widths [very accurate], or an old DOS tool called DIMSTRIP.exe that Engineers claimed better than 1% accuracy to 1GHz. If you want a copy of the 50k program let me know.

Given a box lid 1 inch above surface of PCB, FR-4 dielectric 4.2 and loss tangent of 0.06, 1 mil copper metallization layer [balance between 0.7 and 1.4 mil], dielectric thickness of 20mils; Dimstrip calculates the S curve trace width should be 38.9 mils to get Zo=3D50 ohms.

What did the PCB Fab house suggest for the trace width? They have their own programs.

It should be :-)

These are the specs from the board house concerning the layer distances:

The dielectric is 14 mils. The S trace is 26 mils wide.

I've used an online trace impedance calculator.

A bit of the reason for the whole exercise is to be able to take some measurements to see whether there are serious mismatches in circuits. I'm planning on ordering a directional coupler for that purpose.

--
Failure does not prove something is impossible, failure simply
indicates you are not using the right tools...
nico@nctdevpuntnl (punt=.)
--------------------------------------------------------------

How did you measure that? IOW how did you keep the spectrum analyzer and the ADF4350 in sync?

Did you make sure to turn all WLAN gear in the area off? Because that operates right around 2.5GHz. Not sure who uses 2.3GHz over there but that's a miniscule dip.

--
Regards, Joerg

http://www.analogconsultants.com/

The ADF4350 outputs the signal for much longer than the sweep time of the spectrum analyzer. I've wrote some software for the controller which lets the ADF4350 step through a range of frequencies. I have tried various hold and sweep times but all with the same end result.

True but except for the PCB the cabling is shielded.

--
Failure does not prove something is impossible, failure simply
indicates you are not using the right tools...
nico@nctdevpuntnl (punt=.)
--------------------------------------------------------------

Good, that should exclude aliasing.

Then I'd put the PCB into a metal can. Danish butter cookie cans are good for that but some volunteers have to eat the contents first.

The other way to find out is probe before the filter at the ADF4350. If it's clean there then the filter is the culprit. Although it looks like there isn't too much filtering going on. One of the grounds in the middle of the filter looks a bit iffy but hard to see.

--
Regards, Joerg

http://www.analogconsultants.com/

A short trace like that won't have much effect at low frequencies like

2.5 GHz. The board layout looks fine to me.

I assume that the SMA goes to a 50 ohm coax into a 50 ohm spectrum analyzer. Did you try different lengths of coax to your SA? If the SA were a bad 50 ohm match, you could get reflections. Quick check: stick a 6 dB attenuator on one end of the cable.

What value of resistors (inductors?) did you use for the output pullups on the ADF4350?

I wonder if it's falling out of lock in the dip region. I think you need some other instrument to cross-check the SA. A diode detector would be a good start, or a sampling oscilloscope.

That ADF4350 looks nice. I may have a use for it, if I ever get around to designing my pulse generator. But the digital interface is awesome: I've seen entire microprocessors that were simpler.

--

John Larkin                  Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

Appcad puts that at 49 ohms. The trace is basically perfect, and is short enough that it wouldn't matter much anyhow.

--

John Larkin                  Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

My experience wih issues like this is that they are very likely to be some subtle ground related issue. I would say there is greater than

50% chance there is a false ground or lengthy ground somewhere on your board. An example of a false ground might be a capacitor that grounds to the upper layer "ground" but the via to the actual microstrip ground (I am assuming you are using microstrip) might be half an inch away.

I tried a different cable and a different SMA to N converter. No change.

I used a 20dB attenuator but no change.

I use two 100 Ohm 0402 resistor in parallel. The downside of the ADF4350 is that it doesn't output a sine wave. AD cleverly covered that up by specifying different filters for specific frequency ranges.

I tried a diode detector much like this with a BAT54 diode:

That doesn't work very well to say the least. Maybe I should dig up some PIN diodes.

Wait until you get to the math to calculate the frequency :-) The rest of my PCB is a USB gadget BTW.

--
Failure does not prove something is impossible, failure simply
indicates you are not using the right tools...
nico@nctdevpuntnl (punt=.)
--------------------------------------------------------------

diode:

a bit more work to wire up, but maybe something like this:

-Lasse

That did bother me, on the data sheet. The output stage is apparently an npn current-steering diffamp, and I'd expect it to look like a square wave current source with finite edge rates.

Everybody lies on data sheets, or at least tries to hide the truth.

No, PINs are very slow. A low-c, low-barrier schottky, designed to be a detector, would be best.

Or just buy an SMA detector from Mini-circuits.

Do you have access to a sampling (or fast digital) oscilloscope? It wouldn't even need to be triggered, although that would be instructive, to see the actual waveforms.

Yeah, that does look formidable. They should furnish a nice parameterized C source routine, so their customers don't have to do it hundreds of times.

--

John Larkin         Highland Technology, Inc

jlarkin at highlandtechnology dot com
http://www.highlandtechnology.com

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom laser controllers
Photonics and fiberoptic TTL data links
VME thermocouple, LVDT, synchro   acquisition and simulation