Dear Poul-Henning,

I have been suspecting this very mechanism to exist in the HP5065 among
others. I have not been overly impressed by the stability by which the
current is produced.

It would be interesting to see to what degree the surrounding
temperature as well as the mains supply voltage does. The mains supply
voltage will creep in two ways, one way for the raw rectified voltage
and second on the burning off the difference voltage causing the main
voltage to power to temperature variation.

It would be interesting to see how good damping the main voltage to
various internal volages there really is.

Cheers,
Magnus

On 08/18/2015 10:05 PM, Poul-Henning Kamp wrote:

I'm impressed with your great documentation ... but I wish to differ with some of your circuit analysis. I don't have an HP5065A and can only see the portion of the schematic you copied. Here is what I can surmise from your measurements (assuming you don't have any ground loops or measuring instrument or test lead thermoelectric issues):

The change of the 20 V power supply when you lock is roughly 400
uV (20 ppm).

The change of the C-coil current isn't really repeatable. I see a big
500 nA (111 ppm) drop at 1,000 seconds, but no such clear change earlier
in the test. But the actual voltage changes you are measuring seem to be
roughly correlated between the 20 V power supply and C-coil current
sense resistor changes. In the first 900 seconds I would guess that the
C-coil current sense voltage is rising at about 40% of the rate of the
20 V supply, then during the 1,000 second event the 20 V supply dropped
maybe 30 seconds before the C-coil current sense voltage dropped, and
the current sense voltage dropped about 80% of the power supply drop.
I'm guessing about these values, based on converting your current
numbers into voltage based on a perfect R10 || R11 parallel combination
of 691.6 ohms. I see that R10 has an * asterisk, and I wonder what is
shown for that note. R10 might be a selected value at manufacturer, or
it might have a specific temperature coefficient.

The temperature coefficient of the resistors may be much more important
than anything else, especially for an old product. I also wonder if
those old electrolytic capacitors (C4, C6, and C7 for example) are still
OK, or whether they are showing any changing leakage currents. You might
want to change them with new capacitors just in case.

So I'm not convinced that the time curve is showing a correlation based
on the 20 V power supply affecting the C-coil current. It looks to me
like the noisy jumps in these are not related to each other. It's
possible that measuring system errors (such as where you connected the
measurement system ground) might cause some of these changes. So you
might want to check your setup and the instrument and test lead
computed accuracy.

I also disagree with your estimate of the CR5 zener current. By your
report of the C-coil current and R10 || R11 sense resistance (691.6 ohms
if those stated resistance values were perfect), the current sense
voltage is 3.1266 V. So the voltage at the base of Q6A is also 3.1266 V.
I assume that Q6A/Q6B are a matched dual transistor in the same case, so
they are at very close to the same temperature. This means that the
current through R8 is (8.786 - 3.1266 V)/1333 = 4.2456 mA. The current
through R7 is (20.089 - 8.786 V)/925 = 12.2195 mA. So if there is
nothing drawing significant current in the wire extending left from C4,
the CR5 diode current must be (12.2195 - 4.2456 mA) = 7.9739 mA. That's
pretty close to the zero temperature coefficient current for the 1N938
you describe. So I do not recommend changing the diode current.

So I recommend changing all of the old electrolytic capacitors in that
area of the schematic for good measure (C4, C5, C6, and C7). Then
check the warmup voltage as you did before from terminal 1 of the C-
coil to the ground end of the current sense resistors R10 || R11. Be
sure to not use a different ground, since there may be significant
current and ground drop through the ground trace or plane. Be sure to
have the covers in place so the airflow and thermal characteristics
are as HP designed.

--
Bill Byrom N5BB

On Wed, Aug 19, 2015, at 04:45 PM, Poul-Henning Kamp wrote:

In message 1440039768.738221.360877705.34D2F4EC@webmail.messagingengine.com,
Bill Byrom writes:

The manual is on K04BB if you want the full monty.

The measurement setup isn't optimal, I can only do one measurement
every five seconds on each point and the HP34972A is only a 6½ digit
instrument, but otherwise the setup is solid.

It didn't actually lock until 1000 seconds, so one difference is that
the logic "continous operation" logic didn't reach final state and
its lamp didn't turn on in the previous attempts.

As they should be, because both are derived from the same A15CR5 zener.

The asterix means "selected".  I have not been able to figure out
what criteria it is selected for.

All the important ones are wire-wound, probably for exactly that reason.

Good point.

No, the main correlation is via the common zener, but the step at 1000s
is not present in the zener voltage, which means that the current
generator has really bad supply sensitivity.

I think the way I've done it is OK.  The 34972A has floating inputs
and I measure from a local GND for all six points.  I'm not seing
any noise-artifacts.

Good point.

The 7.5 mA optimum is from a much later data-sheet, and may be for
a particular "high performance" variant of the 1N938, so there is
no guarantee that there even is a zero-tempco current for the one
in my HP5065.

Either way, fixing the zeners tempco is only half of the solution,
it looks to me like the "real" solution involves an entirely new
C-field current driver.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Poul-Henning,

I abuse ADEV too! It really works well in some cases, doesn't it. Very clever of you to try it on a 5065A with voltage, current. etc. I assume you use the y-form (frequency) not the x-form (phase) in this case?

Notes:

1. The scale is more intuitive, I find, when sqrt is used -- so I'm curious why you chose MVAR instead of MDEV?

2. When plots mostly head down with a -1 slope, consider TDEV instead of MDEV, which effectively rotates by 45 degrees turning -1 slopes into 0 slopes. For some kinds of data a rise above a normal (zero) slope is more informative than a bend of a steep -1 line. Psychologically too, it removes the "things are working better and better as time goes on" impression that happens with a -1 slope, e.g., when ADEV is used on data from a locked loop.

3. Before you settle on MDEV, also try ADEV. There are cases where the massive averaging inside of MDEV ruins interesting noise periodics that would show up in ADEV. Then if you want to combine (2) and (3) you can compute ADEV(tau) * tau. Look at http://leapsecond.com/tools/adev5.c for the undocumented z flag. Like the real TDEV is to MDEV, this is a sort of TDEV for ADEV, but without the sqrt(3). It plots a zero slope where ADEV would plot a -1 slope.

/tvb

----- Original Message -----
From: "Poul-Henning Kamp" phk@phk.freebsd.dk
To: "Discussion of precise time and frequency measurement" time-nuts@febo.com
Sent: Sunday, August 23, 2015 11:16 AM
Subject: [time-nuts] Running voltages through MVAR (In re: HP5065)

In message 85E3D5A82F314700BCA6493EEF245CCF@pc52, "Tom Van Baak" writes:

Actually the plot is MDEV now that I think of it...

I've added a footnote.

Good point.

I generally hunt periodics with FFTs, but yes, ADEV is useful for the
sort of "almost has a stable frequency" like HVAC's turning on/off etc.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Poul-Henning,

Using ADEV and MDEV, and indeed TDEV as TVB points out, for analyzing
other physical signals than phase/frequency is maybe not very common,
but if Kocher if you know your field well enough.

David Allan and I discuss this every once in a while. ADEV/MDEV and
friends is really the statistical tools to analyze and separate noises
of different slopes. Together with the bias functions you can use it to
verify you have the white noise you assume, and only then you can
rightfully make use of that property to meet the intention of GUM.

Flicker noise is indeed one such noise-form that does require better
tooling.

Don't forget to use FFT as well as analyzing non-repetitive systematic
trends. I'd love to see the raw data more closely.

So keep up the good work.

Cheers,
Magnus

On 08/23/2015 08:16 PM, Poul-Henning Kamp wrote: