AD587LW - 10V precision travel standard

[Andreas]

Overview:

Hello,

I will describe a 10V travel standard based on the AD587JQ.
Inspired from the US calibration club from Jason (cellularmitosis).
The cirquit is based on the proposal from Lars Walenius a enhanced SVR-T cirquit.
So the project name had to be AD587LW.

Goal is to have a temperature compensated cirquit (with NTCs)
and the good long term stability of a AD587JQ.
I intentionally wanted to have a travel standard which gets shipped "hot".
So no room for a heater and only minimal battery consumption. (around 5 mA)
Further targets: keep it below 5 cm heighth so that it can be shipped as letter within germany.
And weight should be below 500 g.

As Lars told me that the former discontinued AD587LQ had around 2 ppm/K
I will have to select among the JQ devices the best with < 2 ppm/K.
First measurements turned out that most of the currently sold devices have around 7-8 ppm/k and
some (3-4 within 10 devices) have near 2 ppm/K or even below.

So with temperature compensation (factor 10 enhancement over a small +/- deg C range)
a device with 2 ppm over my lab temperature range and 2 ppm ageing/year after a run-in time of 5-10 kHrs
seems to be feasable so with 1 ppm reserve for noise it could have around 5 ppm over a year.

- The device will have a switching power supply from a USB-charger
  (so do not measure during charging)
  with 12xNiMH AAA batteries for about 150 hours. (when new).

- The temperature compensation cirquit.

- a buffered output (highly recommended on a review by Lars)
  so mis-treating the output will only have minor changes on
  the output voltage after exchange of the LTC2057.

- and a controller which is able to measure temperature
  (for logging by RS232) and battery voltage
  and switch off the power supply before the batterie goes dead.
  Of course the connection to the PC is galvanically isolated.

work in progress: posts will be updated soon

[Andreas]

schematic: (first version)

the whole schematic is attached as .pdf.

It consists of several parts:

- the battery charger unit.
  Intention is to use every USB charger with USB-mini connector.
  Including a mobile power bank.

- the LT1763 based voltage regulator to keep the AD587 and
  the output buffer under constant conditions.
  Can be switched off in case of low battery.

- The AD587 part which is essentially the schematics that Lars published here:
  https://www.eevblog.com/forum/metrology/best-out-of-the-box-10v-reference/msg1197526/#msg1197526
  except that I replaced the passive NTC for temperature read out by a active read out.

- A LTC2057 based output buffer (to avoid damages to the AD587 in case of short cirquit etc.).

- And a battery monitor with a own power supply
  which samples the temperature and the battery voltage
  and sends the info to a PC by serial connection.

with best regards

Andreas

[Andreas]

power supply:

a) USB-charger.
Design goal was to have a charger that can be used with every (0.5A) USB Port/charger with a Mini-USB connector.
I first wanted to have one of those tiny SOT-23-5 1.6 - 2.5 MHz step up converters with tiny 4.7-10uH inductor.
The problem was that I did not find any device which can reliably limit the input current to below 500mA under all conditions.
Overloading a USB-port with more than 500mA can lead to a tripping of the polymer fuse.
After once having tripped it will have around double the initial on state resistance which may lead to a unreliable port for higher loads.

So I finally ended in a good old MC34063A design strictly layed out according to data sheet for a minimum input voltage of 4.2V (5V nominal) and
A maximum output voltage of 18.1 V to charge 12 NiMH cells via a diode to 1.45V/cell at room temperature.
The output voltage can be adjusted by P21 to the exact value of 18.1V measured @ room temperature between TP 4 + TP5.
NTC R32 gives a negative temperature compensation of around -3mV/cell/deg C which compensates the cell voltage.
So at high battery temperature the charging voltage (and thus current) is reduced preventing that the cells are cooked on hot summer days.

The current limit is done by resistors R34+R35 to a value below 0.5A.
I first wanted to use 3*1 Ohms in parallel for around 0.9A peak and 0.5A average current.
But in certain load situations the MC34063A goes nearly in a continuous on-state, so you have
Effectively to limit the peak current to 0.5A-0.6 max to get maximum 0.5A input current.
This ends that only R34 (1 Ohms) + R35 (1.5 Ohms) are populated for the 3 resistors.
Remaining charging current is around 80-90mA max for the 12 AAA cells (850-1000 mAH) so around C/10.
So when charging empty cells it will need around 14 hours until the full cell voltage is reached.

All was tested on a perf-board before making the final layout.

b) batteries
As batteries 12*AAA NiMH cells are choosen.
Giving between 13.8 - 17.2V usable output voltage range.
Main restriction for the size was the aluminium case height.

c) linear low noise regulator.
Here I am using my standard LT1763 cirquit with nominal 14V output voltage for 12 NiMH cells
To stabilize the output voltage for the AD587 which needs minimum 13.5V as input voltage.
This measure is mainly to keep the self-heating of the AD587 constant and improve PSRR.
The LT1763 can be switched off in case of too low battery voltage to protect the batteries.

with best regards

Andreas

Update: some scope measurements on the perf-board cirquit with full load conditions
(either full load current at 18V or near shorted output at around 6V).
URSC = voltage over current limiting resistor.

[Andreas]

AD587 section and selection criterias

The AD587 section corresponds to the enhanced SVR-T cirquit of Lars.
  https://www.eevblog.com/forum/metrology/best-out-of-the-box-10v-reference/msg1197526/#msg1197526

The input voltage is 14V (due to battery supply)
 so above the minimum 13.5 V spec in the AD587 data sheet.
C13 filters the wideband noise.
C12 + C17 are against EMI. (partly missing in the original SVR-T cirquit).
P3 is used to adjust the output voltage: with a reduced trimming range
so that R8 or R7 is needed to do the raw adjustment.
P2 is used to adjust the T.C. together with either R5+R6
 (SMD NTCs on the bottom side of the AD587) or alternatively with R1-R3 as through hole NTCs.
TP1-TP3 are testpoints for adjustment.
If the voltage difference betwee TP1+TP2 is zero then the T.C. compensation is also zero.
T.C. can be adjusted in both directions +/- by around 4 ppm/K/V according to the voltage difference of TP1 + TP2.
Of course this correction works only in a small (around +/- 5-10 deg C range).

R22 is for sensing the temperature of the reference. (will be displayed or logged on a PC).

For the selection of the AD587 references I bought some samples  and measured the T.C. and 0.1-10 Hz noise.
Main goal is to select a device with less than 2 ppm/K.
And a normal noise behaviour: i.e. below 6uVpp for the 1/f noise and low "popcorn" noise (can be seen only partial with the 0.1-10 Hz LNA).
Since the popcorn noise hurts more than the T.C. (which can be adjusted in a small temperature range) I decided the following for the "uncertainity score".
The T.C. in ppm/K is improved by a factor 10 by the adjustment.
 So every ppm/K (10uV/K) counts only 1 uV/K in the final score.
The measured maximum popcorn noise (you never get the maximum) counts * 2 in the final score.
score (uV) = T.C./10 (uV/K) * 10 deg C + 2 * max measured popcorn noise (uV).

To get a final spec of 5ppm/year with 1-2 ppm drift/year I have to select devices with a uncertainity score of less than 30uV which is reached when having 2 ppm/K and 5uV popcorn noise. A good device should stay below 20uV in the score.

So from the listed AD587 I decided to use AD587JQ#03 for my first cirquit and AD587JQ#14 for the 2nd travel standard.
#22 which is a excellent device that needs no T.C. adjustment will be used for a special purpose.

with best regards

Andreas

[Andreas]

output buffer:

the output buffer is my standard cirquit around U5 which I also use for my buffered LTZ references.
R15/C20 stabilizes the cirquit against capacitive loads e.g. the EMI-filter (or long cables).
R17 also compensates voltage drops of R15.
R13 is equal to R17 to compensate errors by input bias currents.
C18 gives a additional filtering for broadband noise above 340 Hz.
The LTC2057 can easily deliver up to 10 mA (near 30 mA in short cirquit condition).

At the output there is also a EMI filter which can be populated differently.
Either a common mode choke L6 (51uH Würth WE-SLM 744242510) together with C14+C15.
(usefull more for the lower frequency range)
Instead of the common mode choke L6 also 2 ferrrites L4+L5 (BLM31PG601) can be populated
which are useful around 100 MHz. (FM-frequency range).

I have built 2 samples one with the ferrites #02
and one with the common mode chokes #01 to make different tests.

The common mode choke has 2*0.32 Ohms max DC resistance so it adds
around 0.1 ppm max error when measuring with the 10 Meg input impedance of the meter.
The ferrites have 2*0.09 Ohms max so adding less than 0.02 ppm error.
https://www.conrad.de/de/line-filter-smd-51-h-032-2500-04-a-wuerth-elektronik-we-slm-744242510-1-st-1087399.html
https://www.conrad.de/de/smd-ferrit-600-l-x-b-32-mm-x-16-mm-murata-blm31pg601sn1l-1-st-442583.html

There are also 2 options for output connectors.
I use the CLIFF FCR7350 connectors. (gold plated)
https://www.conrad.de/de/sicherheits-laborbuchse-buchse-gewinkelt-stift-o-4-mm-schwarz-cliff-fcr7350b-1-st-419667.html
but as cheaper option also the Hirschmann connectors PB-4 can be used:
https://www.reichelt.de/Buchsen/PB-4-RT/3/index.html?ACTION=3&GROUPID=5920&ARTICLE=76865&SEARCH=pr%C3%BCfbuchse&START=0&OFFSET=16&

with best regards

Andreas

[Andreas]

battery monitor + temperature measurement

the battery monitor is built around a PIC12F1840 which has all the ingredients that are needed for a low power voltage comparator with hysteresis and a temperature measurement.
- 10 Bit ADC used ratiometrically with VDD as reference for measuring the temperature sensor on GP1
- the ADC is also used together with the 2.048V reference to measure the divided (R23/R24) battery voltage
- the hysteresis is then done in software
- a low power watchdog timer (used all 250ms to wake up the processor and make measurements all 1 or 2 seconds)
- A UART on GP0 which outputs the measured values to a terminal or a dedicated sampling software.
- a internal precision RC-oscillator (1%) for the UART

Firmware description:

The watchdog wakes the processor all 250 ms (nominal with factor 0.6/1.6 as tolerance)
In on state every fourth wakeup (in off state every 8th to save battery) a measurement is done.
GP5 is switched on to supply the temperature sensor and to switch on the voltage divider for the battery measurement.
Also the ADC and internal 2.048V (nominal) reference are switched on.
First the 33K NTC which is linearized by a 27K resistor for the 10-40 deg C temperature range
 (max 0.3 deg linearity error) is measured ratiometrically to VDD.
Then the battery voltage is measured.

The NTC is corrected (linear) by a slope factor (-0.93) and a offset (75,9 deg C for ADC zero)
to get a value in tenths of degree C which is converted to decimal and output as ASCII value on the RS232.
The battery voltage (divider+internal ADC reference) is also corrected by a factor from EEPROM.
Since I first wanted to use the 1.024V internal reference the calibration factors and measurement range 20.6V
are calculated from this value. A overall correction by factor 2 is done before the value is converted to decimal
and output as ASCII value on the LED.

; Sample #1+#2 EEPROM constants
;
    de  0xD2,0x49   ; Correction 20/19.6*20.6/25.6V*65536
    de  0xC2,0x73   ; Temperature offset 75.956 deg*655.36
    de  0xED,0x05   ; Temperature slope 0.925857

;  nominal values
;    de  0xCE,0x14   ; Correction 20.6/25.6V*65536
;    de  0xC2,0x73   ; Temperature offset 75.956 deg*655.36
;    de  0xED,0x05   ; Temperature slope 0.925857
;

the corrected values are also used for the calculation of the on/off state (hysteresis) of the reference.
when the voltage is above 15V (1.25V/cell) GP2 is set high to switch on the supply of the reference via LT1763.
Below 13.8V (1.15V/cell) GP2 is set low to switch off the reference and protect the battery.
Between 13.8V and 15V there is also a voltage drop detection (0.6V per minute)
to detect if a single cell in the stack is flat and needs a quick switch off.


Cirquit details:

C9 compensates for the parasitic capacitance of the LED and speeds up the edges.
T2 is used to switch of the voltage divider of the battery measurement. (and save current)
T2 is a logic level FET with < 1.5V threshold voltage. So at 3V minimum supply voltage a voltage of up to 1.5V can be measured at the source of T2.
Power supply of the battery monitor can either be done by a 5V voltage regulator which needs around 4uA idle current or by a BF545C FET which is switched as voltage regulator with his gate pinch off voltage (around 3-5V depending on load). I usually use the FET because it needs zero idle current.

J2 can be used to program the PIC in cirquit or adjust the EEPROM constants. I use a PICKIT3 for this.
If T1 is used I activate the power supply from PICKIT. 

with best regards

Andreas

Edit 30.03.2018: attached the firmware for the PIC and the PC-software with the necessary DLLs if you do not have VC6 installed.
Of course also a terminal program (e.g. HTERM) can be used for the first steps on the PC.

Edit 05.05.2018: Attached bug-fixed PC-Software (Rev 1: no readings in off-state of the AD587LW reference)

[Andreas]

PCB and construction

Edit: 30.03.2018 attached documentation with BOM, population plans and corrected front plate drill plan of Rev A PCB.

Edit: 01.04.2018 attached reworked BOM after hints from Lars

Edit: 07.04.2018 Description of PCB

The core of the PCB is the additional shielded area around the AD587.
Shielding is done against air drafts and also somewhat against magnetical influence by a TEKO 4020 tinned steel housing.
To make it completely shielded one would have also to put a additional cover from the bottom side.
(will see if it is necessary). Fixing of the bottom cover to the PCB can be done by soldering or by a additional screw.
The core area also gets some foam to further reduce air drafts. And also the bottom side of this area is filled with some foam to keep thermal gradients away from the COVAR pins of the CERDIP package of the reference.

On the lower side the output amplifier outside the shielded area is placed. In Rev B of the PCB there will also be a slot between the buffer and the core area to reduce thermal gradients in case of larger loading (or short cirquits) of the output buffer.
Already there a slot between voltage regulator LT1763 and core area for the same reason (different battery voltages).

The microcontroller (battery monitor) and charger are placed as far as possible from the reference to reduce cross talk.
Only the temperature sensor for sensing reference temperature is placed near the AD587.
Another temperature sensor for the battery temperature (to control the charge end voltage) is in the upper right corner.

[Andreas]

Erratas of PCB revision A of 15.01.2018 revised 30.03.2018
==========================================================
mayors:
- J1 drill positions do not fit together
  (connector pins are 0.2 mm too far against mounting holes)
- J1 not connected to case (via G2)
- L6 too close to capacitors C14,C15
- J3,J4 position in front plate drill plan 1.2mm too low
- Thermal isolation of reference against output buffer missing
- C18 connected to power ground instead of output ground

minors:
- C18 should be upgradable to 1 or 3.3 uF (10 Hz corner frequency)
  (7x7 mm outline footprint)
- testpoint for AD587 GND missing
- testpoint for 14V missing
- testpoint for 10V missing
- Pin1 marker LED missing
- D1 pin 1 marker missing
- D2 pin 1 marker missing
- component values: X7R marking missing on several capacitors.

changes for Rev B:
- reduce peak current in case of longer RS232 transmissions (47uF/25V)

[Andreas]

measurement results

1. power consumption from battery in "off state" measured over a 1K shunt -> 1 mV == 1uA.
    Average consumption in "off state" = 11uA (crude first software, could be further optimized if necessary).

2. power consumption from battery in "on state" measured over a 1K shunt
    Average consumption in "on state" = 3.5mA ~2mA for the AD587LW and ~1mA for the buffer.

[Andreas]

Revision B data of PCB. (30.03.2018)

AD587LWB_DOC_REVB.ZIP:
- Documentation with changed schematics, BOM, placing plan, drill plan for the front plate

AD587LWP_B.zip:
- Gerber data (RS-274x)
  note: there is no silk screen in the data (as it makes no sense to print over the SMD pads).

Notes 17.04.2018:
- I have not built a Rev B PCB by now. (So use data with care until I built the first devices).
- Against BOM I will try some different J-FETS (MMBF4416ACT-ND from Digi-Key) for the power supply of the processor.

Edit 22.04.2018:
- The MMBF4416A give only about 1.5/1.8V with a 1K load instead of the processor which is below the 2.3V needed.
   so I went back to my good old BF545C having more headroom. (2.5-2.7V)
- Charger, Processor supply, and 14V supply are running on 2 PCBs of Rev B. So only the AD587 part and the output buffer need to be verified.

with best regards

Andreas

Edit: if you use the gerber data it would be fine if you give me a note so that I know how many PCBs are in the wild.
        Free for non-commercial private use.

[Andreas]

reserve 2

[cellularmitosis]

       

[e61_phil]

Due to the internal diodes between the inputs of the LTC2057 a short of the output buffer will draw around 1mA out of the reference. Doesn't sound like a real problem. But one have to keep in mind, that the reference is only decoupled by the two 4k7 resistors.

[chuckb]

The new (Nov 2017) chopper OPA189 can handle full differential voltage at the input with no extra current draw on the zener. There are no input differential diodes.

[Andreas]

Due to the internal diodes between the inputs of the LTC2057 a short of the output buffer will draw around 1mA out of the reference.

I am in good companionship in this case.
even the venerable DATRON 4910 has this problem (when looking at the OP27 together with the 10K and 3K resistor) shorting even a LTZ1000 zener. (R120+R124+U111)
https://doc.xdevs.com/doc/Datron/4910_4911/4910%20c20090120%20%5B8%5D.pdf

The new (Nov 2017) chopper OPA189 can handle full differential voltage at the input with no extra current draw on the zener. There are no input differential diodes.

Good find as this is rare with precision OP-Amps. Slightly higher input bias current than 2057. But acceptable for this cirquit.

With best regards

Andreas

[Harfner]

When you log the temperature while in off state, perhaps you could also log humidity?

[bitseeker]

This looks pretty cool, Andreas. Thanks for putting it all in a nice thread.

[Andreas]

When you log the temperature while in off state, perhaps you could also log humidity?

Note: Logging is only done to a attached PC not within the board.
Humidity is not planned since I do not expect a large dependency on humidity.

with best regards

Andreas

[The Soulman]

Where will it travel to?

[Andreas]

Hello,

I have not decided finally since I first need a adjusted unit.
And want to do at least some stability measurements ...
But within EU would be a option. (without custom fees).

but be aware that I missed my 500 g target.
So postage will be at least for a 1000 g letter.
(so 7 EUR from germany to netherlands + eventually 2.50 EUR for a tracking number).
So I think for EU-shipments I will have to look for a different housing of the reference.
As below 500 g I have only 3.70 EUR for shipment.

With best regards

Andreas

[quarks]

very nice project

I am also located in Germany and you can count me in to make meassurements with 8.5 digit gear

[Andreas]

Hello,

Thanks, I will come back to your offer.

with best regards

Andreas

[Conrad Hoffman]

Nice! How important do you think continuous power is? I know for some references it matters. OTOH, my old Fluke 731s don't seem to care at all. They power up and are quickly at the same value as always. IMO, if the reference doesn't really need it, you save a lot of weight and complexity- more room for other stuff.

[Andreas]

Interesting question if it will have a influence on a AD587JQ.

for the LT1027DCLS8-5 I have a answer:

In ADC21 the readings (after 2:1 divider) on LTZ#4 (my most stable one)
decreased over 140 days from 3597.85 mV to 3597.81 mV and stabilized there.
I then switched ADC21 off for a longer time.
After that it started from 3597.83 mV and run opposite to 3597.85V (back to initial value).

Of course this behaviour depends on internal construction (mainly on die attach
which seems not to be the best on the newer SMD hermetically devices).
So it could be not necessary for a AD587JQ.

But yes I would expect from a constantly powered device that it stabilises somewhere.

with best regards

Andreas

[lars]

Nice! How important do you think continuous power is? I know for some references it matters. OTOH, my old Fluke 731s don't seem to care at all. They power up and are quickly at the same value as always. IMO, if the reference doesn't really need it, you save a lot of weight and complexity- more room for other stuff.

For the AD587xQ I don't think continuous power is very important. I enclose some long-term tests for AD587xQ, REF102,LT1021 and LT1031 in metal cases. All data is with my SVR ser.no 721 as reference. The first three graphs is with ten small plastic boxes 55x35x20mm with cheap binding posts from eBay with six AD587KQ or JQ from the nineties and REF102CM from end of nineties. The LT1021CMH is from France via eBay without date code. The fourth graph with REF102+LT1031 is an older box. All refs are compensated with NTC's (similar to AD587LW except LT1031).

In January to March 2017 I turned off the power to all 10 refs for three weeks and turned it on for three hours. I repeated this three times. On the AD587 and LT1021 it is difficult to see any effect. AD587-5 seems to have some seasonal effect. AD587-6 also may have a small seasonal effect. Both are AD587JQ with date code 9243 otherwise they share the same circuit as the others.

The last graph is older and you can see that I under two periods turned the power off for about 3 weeks between measurements. The interesting is the LT1031 that for me seems to be just a three pin version of the LT1021 but has a completely different drift.

But back to the battery in the AD587LW. I have not discussed this with Andreas as I still think it is a good idea to have a battery in the box for a travelling standard like this. The reason is to be independent from different Lab PS's and also if you enclose a small PS (adapter) it will still have a path to the mains as the DMM you compare it to, so you get a loop.

I haven't built in any batteries in my own boxes but has an extra box with two 9v alkaline batteries and an LP2951 set for 15V. I have used this when I a few times have travelled to cal labs and had the chance to test against old very stable Fluke 5700 and 8508's. I should say I have not seen any difference with battery or a small adapter during measurement but I think it is nice to have a battery to get good isolation.

During my travels to cal labs I have also had a min-max thermometer (and hygrometer) in the box as I think large temperature excursions are more of  a problem than power for the AD587xQ. Might even be important for the LTZ1000 as Mr. Pickering commented here:
https://www.eevblog.com/forum/metrology/the-calibration-of-hp-3458a-'gold'-version/msg1374690/#msg1374690

By the way it seems that my eight SVR-T's have drifted upwards about 0-4ppm in last six years and also my four main REF102CM's (that has been continuously powered 17 years, and not the same as in the graphs). So in average less than 0.5ppm/year. Probably the uncertainty is in the region of +-0.5 to 1ppm/year.

Lars