[time-nuts] PLL question

John Miles jmiles at pop.net
Mon Aug 10 14:07:44 UTC 2009


That's a big improvement but I wonder if you're still leaving some
low-hanging fruit, or if there could be a measurement error somewhere.

Here's my reasoning: The floor of an HP 8560E-series is normally -90 dBc/Hz
near 10 Hz as you suggest, but it's actually closer to -100 at 1 kHz
and -115 at 10 kHz.  Your blue trace reaches the -90 dBc/Hz floor at 10 Hz,
but instead of falling away quickly to conform to the analyzer's measurement
floor -- as would be expected if a good OCXO were being measured by
itself -- the plot has a plateau at -85 dBc/Hz to -90 dBc/Hz that extends
out to 3 kHz or so.

Load your two plots into PN.EXE and then load the 8563e.pnp baseline file
that comes with the program, and you'll see what I mean.

Working backwards from your blue trace, it seems that the noise on your 20
kHz comparison frequency, within your ~3 kHz loop bandwidth, is about -140
to -145 dBc/Hz.  You're multiplying this back up by 512, presumably, to get
to 10 MHz, adding 20*log(512)=54 dB to reach -85 to -90 as shown.

Again, that's great in itself, but consider the case where you simply use a
mixer and low-pass filter to lock your cleanup OCXO to the CW25 in a
bandwidth less than 1 Hz.  (Remember that the GPSDO's own loop bandwidth is
closer to 0.01 Hz, so any loop that's wider than that should preserve the
GPS disciplining benefits.)  By the time your analyzer's 10 Hz minimum
offset is reached, your final PN would look more like an unlocked OCXO, -120
dBc/Hz or so, and it would remain far below the 8560E's noise floor
everywhere else.  This is the approach taken in Wenzel's cleanup
oscillators, and others'.

That aside, it's worth asking where that ~-140 dBc/Hz figure at your phase
detector came from.  The CW25's raw output trace looks pretty awful, at
about -60 dBc/Hz within the same 3 kHz BW.  Dividing that down to 20 kHz
should have yielded ~-114 dBc/Hz, not ~-140.  Multiplying *that* back up to
10 MHz in a 3 kHz loop should have restored the noise within that bandwidth
as well, putting it back to the -60 dBc/Hz vicinity.

Instead, you're doing about 25-30 dB better.  Despite the fact that you seem
to have a flat loop response out to ~3 kHz that should have restored the
CW25's noise to its original level within that bandwidth, you're still
managing to get a 25-30 dB improvement somehow.  Am I overlooking
something...?

-- john, KE5FX

> As discussed the CW12 10MHz output has short term noise/jitter that need
> removal for many purposes. The GPSDO I built with one (CW25, same
> thing for
> this ...) locked a Temex double oven 10MHz Osc. The CW12 jitter
> drove me to
> a relatively narrow loop, I ended up using divide by 512 so about
> 20kHz ref
> frequency. Much of the design was dominated by what was in the
> junk box, if
> the attachments work you should find the theoretical closed loop transfer
> function with the attempt to suppress the jitter transfer. The other plot
> shows phase noise on the 10MHz signals measured with an HP8560E
> and KE5FX's
> excellent software. The pink trace is PN on the CW25's 10MHz, the
> blue trace
> is the output seen on OCXO 10MHz which is dominated by the noise floor of
> the analyser rather than the signal.
>
> David
> GM4HJQ
>
>




More information about the time-nuts mailing list