severe clock noise on FMCOMMS1

I'm seeing intermittent junk-o-rama on the output from the FMCOMMS1 we just received.  Sometimes it's perfectly clean with negligible residual FM, but presently it's a royal mess, with the test tone output spread 300khz wide.  Probing the clock, I see a signal added in with a fundamental that wanders around 1.5MHz and an amplitude about 25% of the clock's.  It's pretty obvious this explains the spreading I'm seeing at the output.  Yesterday when I looked at the same signals, both were clean, with negligible residual FM at the output and under 100ps n-cycle jitter on the clock (and none of the 1.5MHz junk).

Game over for the board?  Needs replacement?  (Update from end of the day: this is just some sort of intermittent connection problem on the PCB, so yes it needs replacement, which Analog is working on.  The symptoms seem like a DC-DC converter filter cap with a bad joint, but I only see one in the design, U18, an ADP2323, and it maxes out at 1.2MHz, so eh, who knows.  In any case, pressing lightly on the underside of the board around U13-U15 cleans up the signals momentarily.  I looked at that area under a microscope and nothing jumped out, so I'm baggin' it.)

/

Below is what an IQ plot looks like with the noise on the clock.  The input signal is <1hz rms residual FM, set to match the LO frequency (2.3999995GHZ works best).  The sample window is maxed out at 10000 to show the angle slewing.  When I did this yesterday they came up as dots.  Not so much, today!  I should add that on Friday when I first fired up the RF section, it was junked up like it is today.

  • 0
    •  Analog Employees 
    on Feb 22, 2013 10:07 PM

    Daniel:

    Sorry for the delay.

    Are you still having issues? We have seen this on some boards - but it's normally from something not being seated properly.

    -Robin

  • Well now we're waiting for the replacement board, and the glitchy one is at Avnet awaiting shipment back to Analog.  And AD is trying to pick which of the several already-committed boards in stock to reroute to us instead.

    Maybe our returned board will be salvageable after all?  I assume by "not being seated properly" you mean bad solder joints?  Are you just going over the problem area of the board with a judicious jet of hot air?  I was afraid to do that, so the board I returned will be as we received it.  By all means AD should feel free to rework it to your heart's content :-)

  • 0
    •  Analog Employees 
    on Feb 23, 2013 1:00 AM

    Daniel:

    No - by "not seated" I mean we have seen issues on the Zed board, where a card will not work on it well, but work just fine on a ML605 - KC705, or ZC702. We are not exactly sure why. We test on ML605s before they ship.

    What carrier are you using?

    -Robin

  • Ah, yes, I suspected the FMC connection too, and tested that theory early and late -- removing and reinstalling the daughterboard, and trying it with and without the retention screws cinching the FMC interface together.  Didn't discern any effect.  The clock waveform cleaned up when the implicated part of the PCB was pressed from either side, and only when pressed in that one area I identified (vicinity of U13-U15), so torque affecting an intermittent connection on the FMC connector was apparently not part of the phenomenon.

  • 0
    •  Analog Employees 
    on Feb 23, 2013 1:42 AM

    Did you try it on a Xilinx platform or a Digilent one? We have never seen any issues on the Xilinx platforms.

    You still have not said what carrier you are using...

    -Robin