[time-nuts] Ettus OctoClock measurements
attila at kinali.ch
Wed Sep 14 16:38:15 EDT 2016
On Mon, 12 Sep 2016 17:54:57 +0300
Anders Wallin <anders.e.e.wallin at gmail.com> wrote:
> The results are decent but not stellar:
> In particular the AM noise on the 10MHz looks high. Any thoughts/comments?
I am not surprised. They used a digital clock distribution chip. The chip
is not designed for lowest phase noise, neither is the surrounding electronics.
And selling that whole thing, which is nothing more than the example circuit
in the datasheet, for 1700 is a bit of a rip-off, IMHO.
IIRC your design beats theirs already in performance, doesn't it?
> For 1PPS a 200ps skew between fastest and slowest channel is not the end of
> the world, but I would be interested in the cause. To produce 200ps skew
> both a 4cm trace-length-difference and/or 200 mV of DC-offset (for a 1V/ns
> slewrate signal) seem large?
Not really. They use some unnamed 74x04 instead of a specialized clock
distribution chip (like e.g. LMK00101). They have split the PPS distribution
into two parts, which are symmetric. You can see that clearly from your
measurements. My guess would be that the track lengths on different layers
are not equal for all paths, which introduces quite a bit of skew (thanks
to \epsilon_r changes). That only one port is so much off probably stems from
the need of going around the chip in the hex-inverter package.
Any simple idea will be worded in the most complicated way.
More information about the time-nuts