[time-nuts] Si5351B

[Achim Gratz]

[typo in subject corrected]

John Ackermann N8UR writes:
> On 2/3/19 11:48 AM, Achim Gratz wrote:
> I've been playing with the SiLabs 53xx series chips.  There are two
> that are quite interesting for time-nuts.

I am really trying to go for the octal output parts.  :-)

I'd rather use the Si5351C since that one has a designated reference
input, but the boards I'm looking at only come with the B part.  Judging
from the photos, these are factory pre-programmed to god knows what
function, so my guess is that these are left-overs from some production
run.

Anyway, no later than I figured out that the "C" part would be the ideal
one for me I found this:

https://www.jackenhack.com/ntp-server-extreme-accuracy-for-under-200/

Actually I had looked at the article before, but didn't remember it.
Anyway, this is almost exactly what I'm planning to do, except I plan to
run off a OCXO at least initially and I don't want to remove the
crystal.  Feeding from a GPS timing module is in the cards, of course.

For my application I don't really need outstanding noise performance, as
the resulting clocks are used to run an SBC from, which in turn will run
an NTP server.  The SBC will create a lot more timing uncertainty than
the difference between a good and a very good reference, I think.
Anyway, I plan to use even divisors, so that should get me relatively
good performance all the way to the output.  The part has been used
quite a lot for amateur radio, so I've also found a good deal of
real-world data to look through.

> The 5340/41 are pretty good, though there's a tradeoff in using them -- 
> with a good reference you can get decent close-in noise, but the noise
> floor sucks.  Using a cheap 48 MHz crystal you can get a pretty good
> noise floor but the close in noise and ADEV suck.

That's the first of your plots, right?  That looks … interesting.  Do
you know were this behaviour originates?

> The 5342/44/45 chips add an additional PLL using a cheap 48 MHz
> crystal as a clean-up stage.  It gives performance that's sort of like
> the best of both 5340 reference options.  These chips also allow you
> to adjust the PLL bandwidth, though you can't get as narrow as our
> applications might want.
>
> Attached are some PN plots at 144 MHz with both the 5340 and 5342, and
> a few others for good measure.
>
> To decode the abbreviations for the reference oscillators, "XO" is the
> cheap 48 MHz crystal on the evaluation board, "PP" is a PulsePuppy
> board with a $20 eBay Isotemp 10 MHz OCXO, and "ULN" is an expensive
> Wenzel Ultra Low Noise oscillator.  All the external oscillators gave
> similar noise performance, establishing that you don't need a
> super-clean reference to drive these chips.
>
> I don't have plots to show it, but a higher frequency reference seems
> to have a significant impact on the noise floor -- if you can provide
> a 50 MHz source with decent PN, the 5342 will be quite a bit quieter
> than with a 10 MHz reference.

The Si5351 needs something between 25MHz and 27MHz as reference.  I'm
currently mulling if it's worth the risk going slightly lower, as there
are nice and cheap OCXO at 24.576MHz I could actually buy, whereas my
options at the actual range boil down to just a single unit (I've only
looked at two distributors, though).


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf rackAttack:
http://Synth.Stromeko.net/Downloads.html#WaldorfSounds