03-Dec-2008

Introduction

The M12 (Motorola M12+T or iLotus M12M) has a UTC-synchronized pulse output which can software configured to 1 Hz (1PPS) or 100 Hz. An interesting question is how the stability compares between the two modes.

The M12 has provision for a sawtooth correction (message @@Hn) which can be used by either hardware or software to reduce 1PPS jitter. The CNS-II clock implements hardware sawtooth correction. Another interesting question is how the stability compares between a normal and a sawtooth corrected 1PPS.

In this experiment 1,000 seconds of data was collected for each of three runs: 100 Hz, 1PPS raw, 1PPS sawtooth corrected. This is enough for stability computations out to about tau 100 seconds. The time interval counter was a Pendulum CNT-91 (50 ps resolution). The phase/frequency pulse reference was a Stanford Research SR625 rubidium.

Comparative Allan Deviations Plots

The first three plots below show 100 Hz, 1PPS, and 1PPS with sawtooth correction.

100 Hz output (ADEV and MDEV)

1PPS output (ADEV and MDEV)

1PPS sawtooth corrected output (ADEV and MDEV)

The next three plots below show combinations of each pair of runs.

1PPS and 1PPS sawtooth corrected output (ADEV and MDEV)

100 Hz and 1PPS output (ADEV and MDEV)

100 Hz and 1PPS sawtooth corrected output (ADEV and MDEV)

And the final busy plot shows all runs together.

100 Hz and 1PPS and 1PPS sawtooth corrected output (ADEV and MDEV)

Phase Strip Plots - 1000 seconds

Below are three phase charts from the raw data at the same scale: 1000 seconds in duration and ± 50 ns in scale. The frequency offset was removed and the data normalized before plotting. Note these are not simutaneous runs and are not meant to line up with each other.

100 Hz output (9.6 ns sdev)

1PPS raw output (10.2 ns sdev)

1PPS sawtooth correction output (5.4 ns sdev)

Phase Strip Plots - 60 seconds

Below are three phase charts from the raw data at the same scale: Just one minute in duration and ± 25 ns in scale. The frequency offset was removed and the data normalized before plotting. Note these are not simutaneous runs and are not meant to line up with each other.

100 Hz output

1PPS raw output

1PPS sawtooth correction output

Conclusions

• The MDEV plots are always lower than the corresponding ADEV plots. This is normal, and especially true for GPS signals due to the large white noise content. The more averaging the better at almost every tau. Depending on the implementation of a GPSDO an MDEV plot might be more predictive of final performance than ADEV.
• The obvious wiggles at short tau in the 100Hz output confirm that the 100 Hz is generated with some sort of modulo arithmetic based on the internal (and unsynchronized) clock.
• At tau 1 second, the raw 1PPS has an ADEV of just over 15 ns compared to just under 5 ns for the sawtooth corrected 1PPS.
• At tau 100 seconds, the raw 1PPS has an ADEV of 1.7×10^-10 compared to 1.0×10^-10 for the sawtooth corrected 1PPS.
• However the MDEV at tau 100 seconds is about the same for each: just over 2×10^-11.
• For all tau beyond 1 second, the ADEV of the 100 Hz output exactly matches the ADEV of the 1PPS output.
• Until about tau 100 seconds, the MDEV of the 100 Hz output is 3x to 5x better than the ADEV of either the 100 Hz or the 1PPS output.
• At tau 100 seconds, the MDEV of 100 Hz and 1PPS are nearly equal.
• For all tau, the ADEV of the sawtooth corrected 1PPS is better than the ADEV of either 100 Hz or raw 1PPS outputs.
• Beyond about tau 100 seconds, all the MDEV start to merge.
• It would be fun to extend these plots another decade or two.
• A sub-optimal (i.e., indoor) antenna was in use during these measurements. Next time I'll be more careful.
• Perhaps the same tests could be run comparing a M12+T and M12M; just to be sure they have the same performance.