[#3300] Wavetek 7000, the hidden gemstone.

[TiN]

Yay, post 3300... 
As some already know, MM Labs got greedy hands on (not cheap at all) Wavetek 7000 cells, sold recently on ebay.

While I'm working on the article and study of design, here's glimpse on data over last few days:



Two W7000 cells are measured.
Cell 1 TW7000 measured by calibrated 3458A in differential to calibrated Fluke 732B (9.99992738 V).
Cell 2 MM7000 measured by calibrated K2002 in differential to same Fluke 732B.
Environment suboptimal due to ambient temperature oscillation from aircon, but even then output is within +/-0.2 ppm window  .

More to come..

[Pipelie]

cool!
I think the resolution of the y-axis (voltage)  is too small to read.
HP3458A hates air condition... for sure. at least mine does.

[kj7e]

When I was a kid, I would read the old HP catalogs cover to cover.  I remember seeing in the 1969 catalog how they kept the Cal Lab room held to + or - .03deg C, IIRC they used water pipes in the walls and floor to regulate temp.  Think I was 9 when I first saw that, it amazed me then and still does now.

[quarks]

looking forward for more to come

[Pipelie]

When I was a kid, I would read the old HP catalogs cover to cover.  I remember seeing in the 1969 catalog how they kept the Cal Lab room held to + or - .03deg C, IIRC they used water pipes in the walls and floor to regulate temp.  Think I was 9 when I first saw that, it amazed me then and still does now.

I think that's how we do in the north of China in winter.

[TiN]

More ppm hunting commited.



Allan deviation plot on data also shows that at short observation time τ, the ADEV is higher due to reference and meter noise. At longer τ, instability slightly decrease because the noise averages out. At still longer τ, the ADEV creep up again, suggesting that the meter stability is gradually drifting over time and impact from other factors.



Here was result with ADEV of same 3458-002 measuring reference directly on 10V range.

[TiN]

Completing expensive inner-circle USA cal club round for 2018, box with my Wavetek 7000 is on the way to me, yay 
Before thing arrives, I can share the following photos taken during the calibration in May 2018.

Since 7000 has no history or calibration certificate, decision to adjust it was made.
Prior to adjustment value was measured on direct DCV range, with in-cal 3458A/002 (KS Loveland annual service).



After this, Wavetek 7000 was connected opposite to in-cal Fluke 732B, which has assigned value with uncertainty 0.33ppm and accumulated few calibrations to establish predicted value at the time of W7000 cal.
732B value is expected to be 9.99992738 VDC.



Wavetek 7000 was adjusted 78.3 uV higher than 732B, to result new calibrated value +10.00000568 VDC. (ignore incorrect multiplier on VFD).

Resulting direct measurement: +10.0000056 VDC, which is spot on.



I plan to calibrate some of my LTZ references , one 3458 and one 2002 to this Wavetek, and then ship same 2002 to calibration, to increase confidence that DC Volt import this year for my lab is good.
Once all is confirmed, this will be used as my primary reference, together with my zombie 792X battery-backed LTZ FX 10V-reference box.

[TiN]

Another day, another Doug.

W7000 finally delivered to me today. Now it's going thru it's conditioning thing.
Once it finishes conditioning (that takes about 8.5 hours) logging will start in both direct measurement and differential to my 792X LTZ-box (large scary box on top of 3458As). 



You can see K2400 powering the unit on backstage right. Keep heatsources like SMU far from precision stuff

[doktor pyta]

TiN,
can we expect photos of the PCBs of this rare device?
Existing ones don't satisfy my curiosity...

[TiN]

Yes, you can expect photos and schematics. It's just too much distractions here, sorry... Silly equipment from ebay keep coming and coming... 

[branadic]

I pretty much appreciate tear down pictures. This unit is really interesting and with the upgraded battery pack with 1700mAh instead of 850mAh and thus 30h instead of previously 16h together with it's low power consumption <1W this is the perfect candidate by now for hot shipping. So Illya, you might want to put the teardown in the upper place of the priority list?
I'd also like to know more about the recovery methode, especially the details not covered by the patent or the paper. Would be great. A live stream on that is also welcome

-branadic-

[TiN]

Okay, I'll try to move it up, sorry.

I'm arranging my 732 calibration versus new SI Volt (Happy metrology day everyone!), so I want to close up 7000 as add it into my current 24/7 DC ref bank as 4th reference. So this has to be done earlier than later.

Some of pictures are already taken: here

[branadic]

So I was looking at the pictures and do wonder a bit.
Around LTZ they use TDP16031000 and TDP16031002 resistor network as well as a 30k foil resistor. Can someone explain how they are arranged to set temperature? Is 100R network to form precision 120R resistor?
Furthermore they have 3x TDP16031002 to form 7V/10V boost. Fluke shows, that they need 16 resistors to to do that, so third network must be wired in a different fashion, but I can't figure out how.
Has someone a schematic of Wavetek/Fluke 7000 or an idea?

-branadic-

[branadic]

Reverse engineered front panel pcb by the pictures available on https://doc.xdevs.com/docs/Wavetek/7000/ and simplified the resistor network arrangement.

-branadic-

[doktor pyta]

Tin, maybe I missed some thread but have You seen this burned resistor (pic) ?

[TiN]

Yes, already fixed that. Onboard battery pack charger / power supply escaped some smoke.

[TiN]

In case somebody want to dig into reverse-engineering LTZ cell, here's spaghetti-schematics loosely based on physical placement.
Will need to check all nodes again and redraw it all into human readable shape.

[chekhov]

Hi, it's been silent here for quite some time, let's add a bit more info about W7000.
I thought it would be great to have 'complete' schematics of this little reference, at least to better understand what's going on.
TiN and branadic did a great job posting schematics of reference board and front panel board, why don't add main board to the list ?

So here we are. I drew traces how they actually go at the top layer of main PCB, put it all in KiCAD and tried to decrypt it.
Main board image with traces and components


Spagetti-schematics and following draft are as pdfs, too huge to be pictures  :'(

That's still looks like draft and will likely stay in that shape, but may help someone who curious about what's happening here.
KiCAD labels are used absolutely improperly and component index may look weird, sorry for that .

Most of the circuit is doing power-related job, a few places which control battery and input voltage and turn on/off other parts, there is a block that manages connection of the output to the back connector as well as helps control unit to get module's state. I haven't really tried to understand exact communication, anyway I don't have control module, and still there a few question about signals going from reference board.

And of course output trimming circuit, the most valuable thing. Though it was already discussed in publication, here is how it looks in reality.

I also added schematics of input part of power supply (made of separate components, though newer version uses LT1533) and front panel (keeping component names the same as in branadic's image).

Though I tried to avoid errors, they are always may exist, obvious or not  .
Everything is also stored in our invaluable place - https://xdevs.com/doc/Fluke/7000/sch_chekhov/, including high-resolution images.

[chekhov]

Draft schematics of reference PCB was added.

[doktor pyta]

Thank You very much, chekhov !

[branadic]

Thanks for sharing your latest version, chekhov.

Some good additional information on the design solution can be found in Setting New Standards for DC Voltage Maintenance Systems and Setting new standards for DC Voltage Maintenance Systems - A Solid State DC Reference System

-branadic-

[dietert1]

The LTZ1000 circuit shown includes TC compensation derived from the temperature behavior of Ube of Q18. The signal path is by R33. This is an improvement over the datasheet proposal, where TC compensation is derived from the heater voltage. While the first one is almost linear in temperature (at about 2 mV/K), the latter is proportional to sqrt(T) - an unnecessary complication.

In my LTFLU references i am also using an Ube derived TC compensation and found it very convenient to first set TC compensation to near zero and then fine tune the oven set temperature to the min/max of the residual TC curve. Anyway one needs to vary the oven temperature to find the flat min/max of TC.

Regards, Dieter

PS: Of course another point why deriving TC compensation from the heater signal is a mistake: When i wrote sqrt(T) in reality it is sqrt(TDUT - Tambient). Another complication, since Tambient will normally vary. In my opinion a clean design extracts a precise TDUT measurement from the reference together with the voltage.

[branadic]

I agree with you, but think it is much easier than one might think. It follows the approach, that you would do with any other temperature compensated reference. I always wondered where the z.t.c. temperature of the non-ovenized LTZ with R1=120 Ω as per datasheet schematic can be found. Unfortunately I don't have an oven going high enough in temperature to investigate that, but expect it to be around 80 - 90 °C.

I've tested a simplified version of the W7000 schematic myself - doesn't include the conditioning and not all of the trimming - but was able to set the z.t.c. temperature of the non-ovensized zener to about 48 °C, measured with a temperature profile monitoring the LTZ internal temperature sensor. I've then adjusted the LTZ oven to that temperature and trimmed the remaining and already small and completely linear t.c. to -0.0106 ppm/K (see figure1).
From there I've started with the amplifier stage to get the zener voltage to 10 V. This stage again needed t.c. trimming, but also trimming of the output voltage itself, which influence each other.
Unfortunately, the classical resistor divider and mainly the output voltage trimming inside of it added a small portion of non-linear t.c.. The result I've reached so far is an output voltage of what I think is 10.000 005 4 V - average over 100 samples - with a linear t.c. of +0.0138 ppm/K, but containing a non-linear part (see figure2).

-branadic-

[dietert1]

The Wavetek 7000 supports independent fine adjustment of TC (jumpers around RN3) and oven set temperature (jumpers around RN1). In this respect the LTZ1000 datasheet circuit is a mess and it's near impossible to demonstrate TC alignment or misalignment with that circuit.

Regards, Dieter

[branadic]

Doing a lil' bit more of a tweaking and measuring this is what I ended up with, a linear t.c. of 0.0068 ppm/K (figure3). I do hate the higher order portion though. Stability is good, when comparing the reference inside the thermal chamber at 23 °C against my F7000-2 sitting at room temperature (figure4). So the variation that can be seen mostly comes from the meter itself.
Am I'm happy with the result? Actually not, I guess I will remove the gain, turning it into a simple buffer for the zener, while scratching my had how to improve the gain stage without adding any high order t.c. component to it.

There is so much to learn from the W/F7000.

Edit: A full article about that reference is available here

-branadic-