9-10GHz and FR4?

[daqq]

Hi guys,

I need a cheap way to get a frequency adjustable (digitally, so PLL) 9 - 10 GHz microwave signal - I don't need any particular signal strength, just a signal at anything above around -10dBm, just so a peak of a known frequency will show up on a spectrum and I can tune a different oscillator against it.

Now, I've been looking, there's a few PLLs that should be able do this, with pretty mundane components around, such as:
http://www.st.com/resource/en/datasheet/stuw81300.pdf
http://www.analog.com/media/en/technical-documentation/data-sheets/ADF5356.pdf
Or maybe something with a lower frequency and then pass it through a doubler...

The signal from this would then go directly onto a small antenna, keeping the microwave stuff confined to a very small area of the PCB.

Now, onto the iffy bit - would it be OK to use an FR4 PCB? 4+ layer, 0.1mm between top and internal.
The ADF5356 datasheet specifies:

Do not consider using FR4 material because it is too lossy above 3 GHz

But seeing as I'm not very interested in the actual output power, and the microwave stuff would be confined to say, 20mm of trace, this should not present much of a problem? Or is there some extra problem I'm not seeing? Since the problematic parts (VCO, buffers, PLLs) are already on chip, they shouldn't mind the material of the PCB at all.

Thanks,

David

[daveshah]

I reckon you'll get away with it. Plenty of digital systems around 10Gbps seem to be FR4 (including FPGA dev kits), so I reckon it will do for this application too - the losses should be acceptable.

[darrell]

It's fine and something I have done when the loss is acceptable. Expect to lose around 1 dB to that 20 mm of trace. Given your thin dielectric, metal loss may dominate over dielectric loss.

FYI: Oshpark 4 layer is on Isola FR408 which has about half the loss of plain FR4.

Here are some 60 mm long lines from OSHPark I measured. It doesn't go to 10 GHz, but you can see where it's headed. The mask reduces loss since the finish is ENIG.
https://pbs.twimg.com/media/DYXKuF-VoAAO6Sb.jpg

[hagster]

My guess would be that you will be ok so long as the tracks are short.

The oshpark PCB is probably a good option or Eurocircuits do a very low loss RFpool isola itera PCB.

[ejeffrey]

Yeah, I have used FR4 at 4-8 GHz, it works fine.  It is a little lossy, but if you can tolerate the loss and don't worry too much about flat frequency response it is fine.

A friend of mine claims that the weave of the fiberglass makes the impedance slightly different in the horizontal and vertical directions, and that you can get around that by routing everything at +/- 45 degrees.  I don't know if that is true, used to be true but more modern FR4 is better controlled, or simply superstition that doesn't really hurt.

That said, eurocircuits offer relatively affordable 4 and 6 layer RF pool service that has low loss cores (i-tera or rogers) on the outside and standard prepreg in the center (reversed buildup).  This is fine if you use microstrip on the top and bottom as they only see the good dielectric.  I have used this service quite a bit, and it is really nice.  It is still way more expensive than off the shelf cheap FR4 protos, so for your application I probably wouldn't worry about it.

[Neganur]

looks about right, TXLINE and CST are at about 0.5 dB/cm at 10 GHz  (0.1 mm substrate, 35um copper, 0.16 mm trace width)

[jungle vegetable]

In addition to all said above, consider that FR4 has Er that depends on the frequency (up to say 5% difference between LF and 10-15GHz)

Also, if your traces are narrow enough (they generally will be for 50 ohm target impedance on multi-layer boards) effective Er will depend on where the trace is located relative to the woven pattern of the glass (both linear offsets and relative angles matter). Different prepregs will have different glass bundle widths/periods, so YMMV depending on the exact PCB stackup. It's even possible to get resonant/filtering behavior due to periodic Er variation under your trace where the resonant frequency would depend on the trace angle relative to the woven material

See, for example, here: http://www.microwavejournal.com/ext/resources/Webinars/2014/SLIDES_Isola_26feb14.pdf

That said, if your design is tolerant to some (few dB perhaps) frequency-dependent extra loss and/or reflections and the traces are not too long, it's probably gonna be ok

[daqq]

Awesome, thanks guys for the info!

[geggi1]

This is a very late reply.
Have a look at w1ghz.org
Paul made a transverter for 10Ghz on FR4.
He used a 4 layer FR4 with the top layer as signal and the next as GND.
By doing it that way the FR4 is wery thin and losses go down.