AD587LW - 10V precision travel standard

[Andreas]

Hello,

some days dreams come true.
See also:
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg3737494/#msg3737494

On the way of my ADR1001 reference with temperature controlled PWM correction of the output voltage I thought it was a good idea to test the new cirquit parts on a cheaper AD587JQ device. And I think it was not wasted time...
The design is called AD587TC (AD587 Temperature Corrected).

AD587TC hardware description: (see attached .PDF)

The upper part of the schematic is the battery charger and the 14V supply for the AD587 reference.
It is completely overtaken from AD587LW cirquit.

The reference is AD587JQ (Cerdip package). I have selected a device which has ~2.4 ppm/K.

The rough trimming is done either by R25 or R26 (some ppm below 10.00000V).
The fine trimming (only upwards) is done via T3 per PWM with the trim range determined by R4+R5 together with the trim pin resistance.
PWM is filtered by a 2nd order low pass filter (C22 + C23).
The output is buffered by U5 (ADA4522-1) and protected by a EMI filter and a transient zener.

5V supply for the processor is done via XC6216 which is also the reference voltage for battery voltage sensing.
NTC R22 senses the AD587 temperature. The 12 bit ADC of U1 gives a resolution of ~0.025 deg C.
A 3rd order temperature correction can be calculated to give a 24 bit sigma delta PWM to the PWM output.
From the 10 Bit Hardware PWM only the upper 8 bits are used together with a dither to create 24 bit resolution.
PWM frequency is chosen as multiple of 300 Hz (1200 Hz with a 2.4576 MHz X-TAL) to give a good PWM suppression within  usual NPLC sampling of DMMs (50/60 Hz).

The total cirquit needs ~4.7 mA. The processor takes ~1.3 mA @ 2.45 MHz since the PWM has to be always active. The Interrupt all 833 us with the PWM calculation takes around 50% of the processor load. ~1 mA is for the output buffer and the rest for the reference and the power supply. With low battery there is also a low power off state which switches off the reference and most of the processor internal peripherals resulting in ~30-40 uA average current for sampling the ADC from time to time.

First changes during initial operation.

During power on I recognized a overshoot on the 10V output. So I had to reconnect C22 from GND-Pin to Pin6 from reference U2 to avoid this.

Since I did not want to use a costly ADG419 Multiplexer, I tried a simple BSS138 FET pulling the Trim-Pin down to increase the voltage. But the leakage current introduced a large non-linearity (and reduction of the trim range). As a work around I used 47K pull up and a 110K pull down at the Drain of the FET. This way I meet the idle voltage on the trim pin so that the trim range is not changed.

I had forgotten to spend a Transient Zener on the output. Fortunately it fits on the solder side of the pins of C14.
Initially I wanted to have a smaller trim range by PWM (22Meg for the PWM-Filter). But then I decided that with 24 Bit PWM I could use a larger trim range which also could allow some ageing compensation (later). With the 10 Meg in PWM filter I get +/-100 ppm trim range.

with best regards

Andreas

[Andreas]

Hello,

for further calculations I need the exact sensitivity of the PWM output.
so I did a scan from 4.000 to 252.000 (integer part) in 4.000 steps.
Note that due to the dithering 0% and 100% PWM are not possible without overflow.
The output is measured with K2000 and HP34401A (which is reading ~21 ppm too high).

I did a LMS fit simultaneously for PWM and temperature since temperature is not calibrated so there is a ~24 uV/K temperature drift.

The result of the fit is ~8.1 uV per each integer PWM step (and of course the ~24uV/K temperature sensitivity).
The total adjustment range is +/-1 mV around center.

with best regards

Andreas

[Andreas]

Hello,

first temperature run with fixed PWM = 128.000.

Ideally I want to have a device from 10 deg C to 40 deg C.
So with ~2 deg C self heating of the AD587 reference I will need to compensate from 15-42 deg C.

The output voltage is shown in red.

The other colours are different temperature sensors.

NTC1 is on the heat spreader below the AD587TC housing which controls the setpoint temperature.

Then in pink the internal sensor of near the AD587 reference. (which is also X-axis)

The remaining sensors are above the AD587TC housing.

There is quite some spread between the sensors which are at different positions.

Unfortunately the internal sensor of the first run ranges only to ~39.5 deg C.
But since the output sensitivity may not be exact (and probably temperature dependant). I will need in every case a 2nd run for fine trimming.

The 3rd order LMS evaluation for the first set of temperature coefficients shows a (theoretical) possible adjustment within a ~1 ppm window over the temperature range. This is to be proved.

There is also some kind of hysteresis (temperature lagging?) between rising and falling temperature. So perhaps I will need some kind of temperature prediction.

with best regards

Andreas

[iMo]

..since today, I’ve found myself no longer quite as fond of the Wima red bricks as I once was.. 

PS: btw, afaik, Wima stopped production of most of the 5mm pitch MKS2 red bricks (like 1u-4u7 50-100V and perhaps other MKS2 too) this year..

[MiDi]

MKS2 up to 4.7uF are back at least in the DS.

[Andreas]

No worries there is a obsolence notification due to the Mylar manufacturer.

https://www.tti.com/content/dam/ttiinc/products/PCN/WIMA/Wima-OBSL-NDC_20250422_001.pdf

I still have some in my drawer. The new ones may be larger in size. (the last that I bought where higher in size).

with best regards

Andreas

[iMo]

..anyhow, re 1u-4u7/50V red bricks - better test them on leakage at 40-55C if used at that temperatures..
PS: I would do it with any red brick..

[Andreas]

Do you have any numbers?
What do you use else?
I have seen no excessive non-linearities during my T.C. tests.

with best regards

Andreas

[iMo]

Do you have any numbers?

https://www.eevblog.com/forum/metrology/hp-34401a-hacks-and-upgrades/msg6255547/#msg6255547

[Andreas]

Hello,

first temperature run with activated temperature compensation.

Compensation coefficients where taken from first T.C. raw measurement. (coarse calibration)
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg6255709/#msg6255709

The result is that with the coarse calibration I get already values within a 12 uV window (1.2 ppm window).

I calculated some offset coefficients for a fine calibration. (LMS diagram).
But there is still a hysteresis between rising and falling temperatures which has to be investigated.

with best regards

Andreas

[Andreas]

Hello,

I did some fine calibration with the delta calibration offsets of previous post.

Of course this does not improve the hysteresis (still ~10 uV with some noise) but it will ease further evaluations.

with best regards

Andreas

[Andreas]

Hello,

first I thought that the hysteresis is due to the temperature sensor lagging against the reference so I would need some temperature prediction.
There is a PWM-Filter which has effective ~30 seconds time constant and additionally the NTC ADC value for the temperature is digitally filtered with 16 seconds time constant used to smooth the PWM output when the ADC is toggling between 2 values.

So the first idea was to set the temperature sensor filter time from 16 seconds to zero.
But surprise: there was no improvement (rather a slight higher hysteresis).
Then I tried 128 seconds as filter time: the hysteresis got worse. Additionally the upwards ramp and downwards ramp had the opposite direction for the hysteresis loop.
So the optimum should be somewhere in between 16 and 128 seconds for the PT1 filter of the NTC.
Actually the reference temperature is lagging against the temperature sensor!

After several runs a PT1 filter of ~52 seconds gave the best result.

I had to do a further small fine calibration for the 3rd order correction.

The result is shown in the diagrams.
Virtually no hysteresis. Peak-Peak noise of ~6 uV. Less than 0.5uV deviation for LMS evaluation. (42 deg to 10 deg to 42 deg C)
Practically only the the noise of my ADC#27 which I had to use since the K2000 was needed for other tasks.

There is some run-in deviation at the beginning of the measurement. I have to check wether it is related to the change in ramp slope or if it is due to the fact that I charge the batteries immediately before start of the measurement. Perhaps I have to wait for 1-2 hours after charging.

Further steps to improve are the thermal coupling between NTC and AD587 chip.

with best regards

Andreas

[Andreas]

Hello,

made a repeatability test run with 2 hrs waiting time after charging.
Result: no run in deviation.

So charging (with the battery pack on the other side of the AD587 as the NTC) definitively influences precision.

with best regards

Andreas

[Andreas]

Further Tests:

doubling and halving the ramp speed setpoint from nominal 0.1K/min to 0.2K/min and 0.05K/min.

Result:
this has no influence on hysteresis.
Note that due to my cooler the -0.2K/min are not reached at low temperatures and also the 10 deg C level is not reached.

with best regards

Andreas

[Andreas]

Hello,

Noise test 0.1-10Hz 1/f noise with 10000 fold LNA amplifier:

Result is 3-4 uVpp across 32 measurements with 100 sec.
Corresponding to datasheet value 4uVpp typical.
Note that there is a additional low pass 4K7 + 3.3uF in front of the output buffer which limits total band width.

with best regards

Andreas

[Andreas]

Wide band noise test.

With 1000 fold LNA according to LT AN83 (10Hz-100kHz)

Here the 1.2 kHz PWM frequency clocks in with ~47 uVpp.

To prove this I removed the PIC processor, resulting in ~14 uVpp.

According to data sheet there should be ~160 uVpp in a 1 MHz bandwidth with the 1uF capacitor attached to the AD587.
So in a 100 kHz bandwidth there should be 50 uVpp without further measures.
But I still have the 10 Hz low pass filter (4K7 + 3.3uF) in front of the output buffer which further reduces wide band noise.

To reduce the influence of the PWM gate charge current I bridged the ferrite bead L3 which connects Analog and Digital Ground.
Schematics is attached here:
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg6254225/#msg6254225

This results in nearly invisible PWM frequency.

Now this may have an influence on T.C. measurement (PWM edges may be different).

With best regards

Andreas

[Andreas]

I have repeated 1/f noise with L3 bridged with 0 Ohms:

No visible change within statistical stray.

with best regards

Andreas

[Andreas]

Hello,

I built a 2nd sample (AD587TC_02) to test a different temperature sensor position.

A 33k SMD NTC (size 0805) is glued directly (with epoxy) to the top of the AD587 chip.
Connection to the PCB is done with thin wires.

The SMD NTC (Murata) has a slightly different B-value than the Vishay sensor.
The difference in the temperature range 10-40 deg C results in less than 0.15 deg C so I did no different temperature calculation for the new sample.

The inner shield is now soldered completely (not only at the 4 pins to the PCB) to give better equalisation of the temperature.

Reference is AD587JQ#08 which has been measured with ~1.6-1.7 ppm/K temperature coefficient some time ago.
So slightly better than previous sample (2.4 ppm/K)

Untrimmed output voltage is slightly higher than previous sample. So this time I had to use R25 (2*30 Meg in series) to trim down the output voltage to center the PWM adjustment range around 10.0000 V.

With best regards

Andreas

[Andreas]

Hello,

further checks on AD587TC#02:

again leakage current of PWM-FET is too high -> 47K+110K added -> linear behaviour.
the slope for one integer step is now higher 8.147 uV/step instead of 8.1 uV/step on previous sample.

Uncalibrated (raw) T.C. measurement with fixed PWM = 128.0 shows 2 things with same temperature sweep as previous sample:
- the NTC is now ~2-3 deg C hotter directly on chip so I am no longer reaching the 10 deg C minimum temperature.
- effective T.C. at 25 deg C is around 1.86 ppm/K

with best regards

Andreas

[Andreas]

Hello,

some days later with AD587LW#02 measurements:

Actually the delay needed to reduce hysteresis has been shorted from 52 secs to ~28-30 secs.
But there is still no prediction (DT1 part) needed to compensate for the 30 secs PWM filtering time constant.
The firmware and PC software have been prepared for a DT1 part but this is still not needed for the new NTC position.

On the other side I have found ~5uV popcorn noise which is visible at the same time on both DMMs.

When looking at the repeatability of the temperature points there is still something odd. (see avg measurement).
The question is: are the DMMs at its limit (linearity, temperature behaviour) or is it some other problem (NTC sensor?)
I fear also the PWM filter capacitors are at their limit together with the high ohmic resistors (dielectric absorption ?)

I think I should try some better capacitors.

with best regards

Andreas

[Andreas]

Before I forget it:

the NTC position on the chip (instead on the PCB near the chip) reduces stabilisation time after charging from ~2 hours to ~30 minutes.

with best regards

Andreas

[Andreas]

Hello,

initially my assumption was that I could correct up to 0.1 uV with a 12 Bit ADC/DAC.
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg3737494/#msg3737494

With the 24 bit PWM I have plenty of resolution. But when recalculating the 12 Bit ADC:
The temperature increment is about 0.025 deg C per ADC step.
So with 2 ppm/K = 20 uV/K gives a voltage step of 0.5 uV per temperature step.

For a 0.1 uV correction I will need a higher resolution for the temperature sensor.

Another thing is the resolution of my 6.5 digit DMMs. Here I have 1 uV at the serial interface.
This also limits the possible repeatability.

So the "hysteresis" at higher temperatures may be already explained.

When doing a long term test at 35 deg C setpoint (on the heat spreader) I get about 11 uV spread or a RMS value (standard deviation) of 1.8 uV. Allan deviation shows also 1 uV for longer measurements.

So I will try better PWM capacitors (MKP2), although they are huge compared to the typical MKS2 foil capacitors, and do not fit in an upright position within the shielded area.

But I fear I am already at the system limit.

with best regards

Andreas