[time-nuts] Re: Allan Deviation of Z3801A

Tom Van Baak tvb at leapsecond.com
Wed Jul 27 12:40:46 EDT 2005

> Dear Tom Van Baak,
> Sorry to bother you in this way but I cannot get registered on time-nuts
mailing list.

Thanks for the mail. Very pleased to meet you.
I'll cc time-nuts and John can add you to our
list manually.

> My name is Lymex Zhang (or Zhang Limin, if direct translate from Chinese),
I am one of a few time-nuts in China.
> I have read you home page many times with great respect - many of your
equipment and the way you are dealing with them are simply my dreams.
> I own a SRS SR620 frequency counter plus SR625 Rb time base, an Agilent
53131A, a HP58540A GPS, a Trimble Thunderbolt GPS, and many OCXOs. Recently,
I acquired two HP Z3801As and became confused in measuring the stability of
their OCXOs.

That's a very nice set of equipment. Yes, you are
a time nut; acquiring and confusion are two key
elements ;-)

> My primary goal is to find a purist, most stable frequency source as I can
afford. One good criterion seems to be the Allan deviation (of many tau

This is a great goal, one that most of us are
trying to achieve one way or another.

> Z3801A seems to use HP 10811-60158 OCXO
(http://www.realhamradio.com/GPS-oven-journey.htm). From the spec of
10811-60158(http://www.febo.com/time-freq/hardware/HP10811-Specs.pdf, page
21), it says <5E-12 for 0.1s and 10s, <9.8E-13 for 1s. This is the best time
domain stability I have ever seen for frequency source. However, by your
article (http://www.leapsecond.com/pages/z3801a-osc/), many of the Z3801A
are not satisfying this. Is it might be that not all Z3801A use the same
crystal? Or because the crystal degraded as many years have passed by?

All Z3801A use the same model crystal. The 1 s
noise should not change much over time.

That's an interesting question. It seems all 12
of those Z3801A that I tested meet spec for
every tau except for the 9.8e-13 spec at
1 second. It seems most of them are slightly
above this at 1 second.

One theory is that the OCXO itself meets the
spec but we are not really measuring the bare
OCXO - we are measuring the 10 MHz BNC
output of the Z3801A and there could be added
noise in the output buffering circuitry or something.

This is something I could check. I'll let you know
the result.

Also the "9.8e-13" is one of the stranger specs
that I have seen. Sometimes you see 5e-13, or
1e-12, or 2e-12, but "9.8"? It seems a very odd
number and there must be a story behind it.

> The other concern for me is to how actually measure the Allan deviation. I
use some of my standards(mentioned above) in turn, and make calculation
afterwards to cancel the instability caused by the frequency source of its
own( reading^2 = f1^2 + f2^2, where f1 is the Allan dev of the measured
freq. and f2 is the reference freq.). Even though, the noise of the
frequency counter is a big problem. I use the SR620 to measure SR625 (also
Tee the output of SR625 as a reference), here are the results for Jitter:
> Interval, Allan dev., Std. dev., N
>   0.1sec,   3.9E-10,   3.9E-10, 200
>   1.0sec,   3.7E-11,   3.8E-11, 100
>  10.0sec,   3.9E-12,   4.0E-12, 100
> It seems that:
> 1, the floor noise is very high, higher than many OCXOs.
> 2, the noise is inverse proportional with interval
> 3, Allan deviation and standard deviation are almost the same in this

1. Good observations. You should not use the
SR625. It is sold as a low drift reference, and
yes, rubidium has very low drift. But it has poor
short-term noise. Almost any OCXO will beat
a compact rubidium, short-term. If you want to
calculate ADEV for tau 1000 to 1 day then use Rb.

If you want to calculate tau much less than that,
then use a good quartz standard as a reference.
The SR625 spec is 2e-11 at 1 s. This is 10x to
50x worse than a really good OCXO.

2. When you see noise inversely proportional to
tau it's the -1 slope on the log-log ADEV chart.
It usually means you are seeing measurement
noise. This makes sense. A SR620 is sensitive
to about 20 ps at one second, or 20e-12, 2e-11
so it cannot measure stabilities much less than
2e-11 at one second.

3. Yes, ADEV and stddev will be pretty close to
each other for many measurements, especially
those containing noise. Where ADEV begins to
be less than stddev is when there are slow trends
in the data, trends (like frequency offset or drift)
that rise above the noise.

> I also did the similar measurement with my 53131A. It shows that Std. dev.
is about the half of the SR620. However, no Allan dev. function is available
for 53131A.

For the purpose of comparing all your standards
against each other using the HP53131A stddev is
perfectly fine.

> I understand that TSC-5110A is the best for this purpose (the floor noise
is 2.5E-14 for 10MHz). But, is there another way to measure the low Allan

Yes, the typical way is to beat the two sources
against each other and measure the stability of
the beat note. This technique gives you orders
of magnitude improvement in short-term resolution.

If you can manually tune any of your sources
0.1 Hz or 1.0 Hz off frequency then mix that and
your UUT in a DBM (double balanced mixer) and
low-pass filter the output (a simple RC filter will

You will see a beautiful 10 Hz or 1 Hz sine wave.
If you then measure and record each period of
the sinewave to, say 1 us, then you will have a
measure of the relative stability of your two
sources down to parts in 10^13th.

I can go into this in more detail if you wish.

> Thank you for your patient in advance.
> Best regards,
> Lymex Zhang
> p.s. I'm also a HAM(call sign is BG2VO)


More information about the time-nuts mailing list