[time-nuts] Newbie with questions, SR620 ADEV

Tom Van Baak tvb at leapsecond.com
Thu Jul 28 19:37:38 EDT 2005

> Hi everyone,
> I'm very glad to be here for the first time.
> Thank you John for manually registered me in.
> Thank you Tom Van Baak for putting my e-mails here and also
> for your great answer which clear up most of my confusions.
> Sorry for my poor English because its not my mother tongue.
> I come from a place where there are few people who are 
> interested in time and frequency so I learn most of the
> subject by myself mainly. Until one day I came across an
> article by Tom and then his leapsecond.com which I read
> with great interest, then I found this mailing list. After
> read most of the archives I understand I belongs to here.
> I have a lot of questions now (and probably more in the future).
> Here is my first set of questions. 
> We all know that Allan Variance/Deviation (ADEV) is one of the 
> common ways to measure time domain noise of frequency sources. 
> It seems to me that the noise of many time-interval equipment 
> can also be described in this way.  
> Q1A: Is ADEV could be used for all time-interval measuring 
> equipments?
> An example is at: http://www.timing.com/products/testsets.php
> it says ADEV <5x10-14 at 1 second for TSC 5110A

Yes, that is the noise floor specification for the
TSC 5110A - the minimum resolution that it can
measure. Every oscillator has a certain specified,
or measured, ADEV plot; the same is true for any
test equipment that does timing measurements.

The resolution specification is not always listed
as ADEV. Some counters advertise the "number
of digits per second", or the single-shot resolution,
or the LSD (least significant digit). But it's all
essentially the same thing.

> If so, or partly so,
> Q1B: Could it be easily measured/confirmed in general?
> I even have difficulties measuring the ADEV of SR620 which has
> the ADEV function build in. I made wrong assumption in my last
> mail to Tom, so I re-measure it below. 

You can indirectly measure the ADEV of the
test equipment if you know the ADEV of each
of the two oscillators you are comparing with
that piece of test equipment. If, for example,
you have two nice OCXO that you know have
an ADEV of, say, 1e-12 at 10 seconds and
your counter tells you their relative ADEV is
1e-11, then you can be sure it's the counter
that is the limiting factor, not either of the

In general, the ADEV you measure is the RMS
of the stability of each source and the TI counter
itself. And so in the special case that the OCXO
perform identically and the TI counter is much
better than the OCXO the true ADEV of each
of the OCXO is sqrt(2) of the measured ADEV.
Does this make sense?

> Q1C: I actually measured the ADEV of my frequency counter 
> SRS SR620 below (for tau = 1 sec), anything wrong?
> SR620 is able to take many sample frequencies and calculate the
> ADEV and standard deviation (SDEV). What happened if it measure 
> the reference frequency?  

This is a sort of self-test and it gives you a
rough measure of the jitter in the system.

> Using a good 10MHz as a external reference for SR620, also
> Tee this frequency and let the counter measure it at the same
> time. Ideally it will read 10MHz exact. However, because of 
> the jitter (and some other reason) of the counter, the reading 
> is not perfect. Set the gate time to 1 second, number of 
> samples to 100, the reading for Jitter:Allan is as follows:
> the counter at tau=1 second.
> 1.5E-11 for HP Z3801A (no GPS)
> 3.7E-11 for SR625 (Rubidium time base)
> 1.6E-11 for HP 58540A (GPS locked)
> 1.3E-11 for HP 58540A (Holdover)
> 1.1E-11 for Trimble Thunderbolt (GPS locked, time const=400s)
> 1.1E-11 for Trimble Thunderbolt (Holdover)
> 1.4E-11 for SR620 its own internal reference (TCXO)
> (during above measurement, any reading are taken for at least
> three times, I keep steady, and no cables are touched)

This is a nice set of data.

> It is easy to say that:
> 1) the lower the reading, the better the frequency source for
> tau = 1sec


> 2) compact Rubidium is worse than OCXO in short time stability.

Yes, this is often true. Many OCXO have much
better short-term stability than compact Rb.

> 3) the ADEV of the SR620 (if it applies) is better
> than any of the above readings

I'm not sure you can say this. It might be
equal to some of the readings.

I would guess that one or more of your sources
is much better than 1e-11. The fact that all the
readings are above 1e-11 suggests that the
SR620 itself is the limiting factor here. And,
based on my earlier email, this is to be
expected -- 1e-11 is 10 ps per second. I can't
see the SR620 getting much better than this

> 4) the ADEV of the SR620 is perhaps not too much
> better than the lowest reading of above, 1.0E-11 is my estimate.


> 5) this 1E-11 is equivalent to 10ps jitter which is good
> because SR620 has 100ps uncertainty for frequency measurement
> as specified.

> 6) SR620 cannot measure ADEV of much lower than 1E-11 for 
> tau = 1 sec. because of its own noise floor. This makes it 
> useless to measure ADEV of good OCXOs.

Yes and no. It is useless to measure ADEV of
OCXO for short tau. It should still be very useful
for longer tau. The resolution is approx 1e-11 / tau
so at 100 seconds the SR620 can produce data
that gives you ADEV down to 1e-13.

It is not uncommon to use multiple methods
to achieve a full ADEV plot, from sub-second
to multiple days or weeks. A single reference
or a single counter is just not good enough to
cover the entire spectrum. Some people, for
example, use their best OCXO for tau 0.1 s
to tau 500 s, and use GPS for tau greater than
about 1000 s.

The SR620 is one of the best, but it does have
its limits, as does every instrument.

Measuring ADEV of good OCXO (1e-12 to 1e-13)
is not that hard; but you have to use methods
other than simple frequency or TI counters. That's
why I mentioned the heterodyne (DBM) method.

> Q1D: Is the ADEV and the SDEV always the same for frequency 
> counter noise?

Mostly this is true. When dealing with white
noise the two will be similar. Where ADEV
and stddev diverge is when there are drift
effects; this is usually something that happens
in the mid- to long-term, not short term.

If you ran a TI counter for a day you would
definitely see periodic, drift, or thermal effects
in the data. stddev sweeps all this data into
one calculation while ADEV(tau) can resolve
different parts of the time spectrum.

> I ask this because on my SR620, it almost the same reading
> for Jitter:Allan and Jitter:STD DEV regardless of the gate
> time and the frequency source. Tom has answered this question
> by saying that it will be the same in his previous mail.

For white noise this will be the case.

With the SR620, you will start to see Jitter:Allan
and Jitter:stddev differ for longer runs.

> Q1E: which is better as for noise, Agilent 53131A or SR620?
> For Agilent 53131A, only the SDEV function is available. 
> I made the similar measurement as above (gate time=1s, N=100 
> samples). The SDEV is worse than those obtained by SR620. 
> For instance:
> 2.3E-11 for HP Z3801A (no GPS)
> 2.1E-11 for HP 58540A (GPS locked)
> 2.3E-11 for Trimble Thunderbolt (GPS locked, time const=400s)
> 2.4E-11 for SR625 (Rubidium time base)
> 3.1E-11 for 53131A its own internal reference (TCXO)

The SR620 should be much better than a
53131A, or even a 53132A. It will be about
the same as a 5370.

> Yes, these readings are quite the same for external references. 
> It seem to me that the 53131A uses external reference directly 
> while SR620 uses the external frequency indirectly by phase 
> locking an internal VCXO to it. This 2.3E-11 deviation is 
> equivalent to 23ps of jitter which is very good comparing to 
> the 500ps error spec. for frequency measurement.

Not sure on this one. It could be that the time
constant of each PLL is different, or the TCXO
in each counter is of different quality.

> correct me if I am wrong.
> Thank you for your attention.
> Lymex, BG2VO


More information about the time-nuts mailing list