I've been doing a lot of tests of the u-blox modules with their
TIMEPULSE outputs set to RF-ish frequencies. Attached are phase noise
plots of the NEO-M8T in several groups of related frequencies.
The frequencies that are integer MHz values seem to be the worst.
4, 8 and 12 MHz have worst-case spurs in the -10 to -20 dBc range at
about 30 Hz offset. At 10 MHz the worst case spur is -34 dBc at 150 Hz
offset; at 5 MHz it's about -42 dBc at about 850 Hz offset. So based on
biggest spur, the 5/10 MHz outputs are significantly better than the
4/8/12 MHz. Otherwise, the spectra of 4/8/12 MHz look quite different
than 5/10 MHz, particularly beyond 1 kHz, but I don't know that they are
better overall.
Some power-of-two frequencies seem to be quite a bit better.
It looks like 4.096 is the winner, with a maximum spur of -52 dBc at 1
kHz (that value is from the spur table; on the plot it looks to be < -65
dBc). At 8.192 MHz, the peak spur is -35 dBc also right at 1 kHz. By
eyeball, the overall spectra of both look much better than those of the
integer frequencies.
At 2.56 and 5.12 MHz the maximum spurs are also around -50 dBc, but
there are more and bigger spurs in the range of 1 kHz to 10 kHz that
make the overall spectra appear worse than at 4.096 MHz.
So contrary to the u-blox paper, it doesn't seem that sub-multiples of
48 MHz are a "sweet spot", and in fact they generate the worst spurs.
From this set of an admittedly small number of test frequencies, 4.096
MHz seems the best. [ The u-blox paper tested the M6T while I am
working with the M8T. I don't know if there are changes between the two
series that would change these results. ]
I'm scheduling similar tests of the ZED-F9T and NEO-M9N and should have
that data in another week or two. Preliminary tests didn't show any
surprises compared to the M8T results.
John
I've been doing a lot of tests of the u-blox modules with their
TIMEPULSE outputs set to RF-ish frequencies. Attached are phase noise
plots of the NEO-M8T in several groups of related frequencies.
The frequencies that are integer MHz values seem to be the worst.
4, 8 and 12 MHz have worst-case spurs in the -10 to -20 dBc range at
about 30 Hz offset. At 10 MHz the worst case spur is -34 dBc at 150 Hz
offset; at 5 MHz it's about -42 dBc at about 850 Hz offset. So based on
biggest spur, the 5/10 MHz outputs are significantly better than the
4/8/12 MHz. Otherwise, the spectra of 4/8/12 MHz look quite *different*
than 5/10 MHz, particularly beyond 1 kHz, but I don't know that they are
better overall.
Some power-of-two frequencies seem to be quite a bit better.
It looks like 4.096 is the winner, with a maximum spur of -52 dBc at 1
kHz (that value is from the spur table; on the plot it looks to be < -65
dBc). At 8.192 MHz, the peak spur is -35 dBc also right at 1 kHz. By
eyeball, the overall spectra of both look much better than those of the
integer frequencies.
At 2.56 and 5.12 MHz the maximum spurs are also around -50 dBc, but
there are more and bigger spurs in the range of 1 kHz to 10 kHz that
make the overall spectra appear worse than at 4.096 MHz.
So contrary to the u-blox paper, it doesn't seem that sub-multiples of
48 MHz are a "sweet spot", and in fact they generate the worst spurs.
From this set of an admittedly small number of test frequencies, 4.096
MHz seems the best. [ The u-blox paper tested the M6T while I am
working with the M8T. I don't know if there are changes between the two
series that would change these results. ]
I'm scheduling similar tests of the ZED-F9T and NEO-M9N and should have
that data in another week or two. Preliminary tests didn't show any
surprises compared to the M8T results.
John
