1 GHz synthesizer

I'm developing a 1 GHz fractional-N synthesizer. The reference input is a
50 MHz sine wave which I'm converting to LVDS as follows:

1. 50 MHz sine wave arrives at board mounted SMA in 50-ohm system
2. Converted to 100 ohms differential using Mini-Circuits ADT2-1T
transformer
3. Carried 10mm along pair of closely-spaced traces to 100-ohm 0402 resistor
adjacent to differential input of LMH7324 high-speed comparator.

I think some 1 GHz from my VCO (3 inches away) is getting into the (3.84
Gbps rated) LMH7324 comparator input and modulating the zero-crossings.
Result: integer boundary spurs when the VCO frequency is set very close to
an integer multiple of the reference.

Reducing reference drive level increases integer boundary spurs.
Increasing reference drive level reduces integer boundary spurs (quite
encouraging reduction is possible).
Touching one side of the transformer secondary with the metal blade of a
plastic-handled scalpel increases integer boundary spurs.

Presumably, the 1 GHz enters the comparator as a common-mode signal. I
would like to try two small caps from the comparator inputs to ground.
Unfortunately, there are no convenient grounds to be had in that area. The
transformer secondary centre-tap is grounded and I could easily put small
(10pF?) caps across the secondary windings; but that's 10mm away. I will
just have to try a few different things and maybe drill some holes through
to the continuous copper ground plane on the bottom. Annoyingly, there are
some decouplers on the bottom just under the 0402 100-ohm.

How/why is the 1 GHz leaking all over the board like this?

Should I have split my ground plane between reference frequency and VCO
frequency areas of the board?

Any other suggestions?

TIA

 

Thanks for the suggestions. Using a tuned circuit sounds good. I was
planning to support 10 or 50 MHz reference inputs; but I could drop that
requirement and stick to one frequency.

I wanted it to take the -10 dBm 50 MHz STD output from my spectrum analyzer.
I'm currently needing to boost this by 20dB to push the spurs down.

Can you elaborate on why a slower comparator might result in less phase
noise? I can see how it might be less susceptible to 1 GHz and therefore
lower the spurs; but wouldn't it have more random jitter?

 

<ouieeeeeeeeee ...> (sound of alarm siren)

If there is no ground plane in that area that's you first problem, needs
to be fixed. If there is one and you just don't have a via to it place
copper tape to the next via so you get a part of a plane there. If too
crowed put that on the other side and drill a hole. Say a prayer before
drilling

If the comparator is well bypassed, which I hope it is, then you cold
also solder caps from inputs to its VDD and VEE. If not well bypassed
then that needs to be accomplished first.

As John said transformers have plenty of leakage inductance. This picks
up RF and you may have to shield the transformer.

Add a little snippet of copper tape to bridge the 10mm with a fairly
wide swath of copper.

In the RF world that is usually a recipe for desaster. A ground split is
like roulette. You win on one number only, on black nothing happens, on
red you lose ;-)

Yes, can you add make-shift shield cans? Shallow shield tops of a few
millimeters height can be bought but when in a rush at a client I
sometimes cut up a butter cookie can. The contents are dutifully eaten
by the crew

Another more nasty trick that borders on cheating is to solder a 2-3cm
snippet of wire to one side of the transformer and bend it around. Try
both sides. Then try to find a minimum in the spurs by bending. It acts
as an antenna and the goal is to make it feed 180 degrees against the
1GHz that leaks in through other paths. Typically one would use a longer
than anticipated snippet and when the reaction upon bending is too gross
cut a chunk off and see whether it affords more neutralization control.
Bending needs to be done by plastic tweezers. Or in a pinch wooden chop
sticks, always a good excuse to have lunch at an Asian place.

Make sure a black cat crosses the lawn and is seen by others when you do
this so when it works you will be awarded guru status

 

Joerg a écrit :

Did you really use that one? Might be a bear to find the right position...

Better to bring one with you. I'll add one to the box next time I'll go
to a customer...

 

I have a continuous ground plane on the bottom but no ground vias near the
comparator because it only has VEE -5V, VCCI 3.3V and VCCO 2.5V. There are
vias to 100nF decouplers on the bottom and I could easily put caps from
these vias to the inputs. I didn't do this earlier because I wasn't sure if
the SRF of the 100nF would be effective against 1 GHz; but I could maybe
replace them with 10nF.

Thanks for putting me straight on that one!

 

If it's regular X7R ceramics in 0402, 0603 or even 0805 it'll be ok at
1GHz. No need to swap those out.

 

I did. It's not that bad. Basically you bend it downwards a bit and then
rotate it slowly until you hit close to zero. If the noise shoots up way
more than 2x after soldering it on snip pieces off. If you can't get
anywhere near zero it needs to go on the other phase.

Another trick I learned way back in my early ham radio days is the poor
man's variable capacitor. If you need just a smidgen of signal injection
and the source is really close by (less than 1cm) use two wires and
twist tightly several times. Then snip off a millimeter at a time until
"right".

Trick #2 did not come without pain. Had to twist some more on my first
one, reached in, forgot about the fact that 300V plate voltage was
riding on one side, snipped end touched index finger ... *BZZZT*

[...]

 

Joerg a écrit :

Yep, knew that one.
One related I used/invented? where I needed really super fine cap
adjustment (to GND) was:
Have two surface pads located on top side, located half an inch to one
inch apart. One unconnected, the other being your cap terminal.
Then solder a small bare wire between the pads, so that it makes a flat
inverted U, something like 5mm above the GND plane.
And voilà, you have your really fine trim cap. Just bent the wire up and
down with a small plastic tool while using the PCB as fulcrum and you
can easily adjust your cap to the few fF.

Thought of this while thinking about the way some chineese tuners coils
are hand adjusted. It made wonders once in prod at tweaking HF CMRR.

 

This is actually how some RF mixer modules are aligned in production. By
pushing and bending those wire stubs.

On coils that's been done since a long time. Pushing the windings
together or if you want to go in the other direction insert the little
flat wooden stick and twist it so the turns spread again. But a really
experienced tuning tech never has to go backwards

[...]

 

Or build a filter with two peaks, one at 10MHz and the other at 50MHz.
They are far enough apart, should work.

 

Joerg a écrit :

Or maybe, if the reference input is a square signal, let it pick the
10MHz fifth harmonic...
With a high enough Q you'll probably have enough signal to almost not
notice.

 

The Zeitgeist pub would have some soothing medicine for that

 

In the end it's like what I have seen at an auto shop in Italy. Alfa
Romeo, hot edition with four independent carburtors, needed adjusting.
"O mama mia, only Giuseppe can do those, you'll have to wait until he
gets back".

 

Oh yeah, and the design is really slick, too. This one was a Julietta
but with four carbs, each having as "air filter" nothing but a stretched
piece of cloth above the throat.

They made a lot of very sporty cars down there. Except they rusted like
crazy, at least in the 80's.

 

Jim Thompson a écrit :

You could have a look at spice opus. It's free and has quite a rich
optimization algorithms set.
http://www.spiceopus.si/optimize.html

It's been a while since I used it and I don't know whether they've
included a GUI or not, so it might bring you back to the good
editor/netlist old days, but for a few RLCs filters this is pb OK.
(my spice being a 3F5-Xspice based one I could use it straight with some
hand written added commands for the spice opus unsupported specific
commands)

 

The Dishal procedure is described in
"Randall W. Rhea: HF filter design and computer simulation"
Mc Graw-Hill ISBN 0-07-052055-0
pages 392-394
Everyone doing practical filter design should have this book.

Rhea has also a good oscillator design book and is
the author of that what evolved to Eagleware / Agilent Genesys.
(has nothing to do with the Eagle layout program)
Genesys also has a nice filter design plugin.

regards, Gerhard

 

Here is a deal, do it for 1 kHz and 5 kHz instead and let others scale it.