T.C. measurements on precision resistors

[Andreas]

Hello,

After the Z201 120R resistors as comparison some PWW 120R resistors from Ultron plus (UP805_120)
Setup has changed:
I now use a PWW resistor with 1K (UP805#3) as reference resistor.

Edit:
And I had to move to the basement with the whole setup since
otherwise it gets too warm for the reference resistor during summer.
(The 27.5 deg setpoint was close to run out of regulation during the last measurements).

UP805_120#3 delivery date 1510
T.C. 3ppm/K from datasheet

17.05.2015: first measurement normal polarity
18.05.2015: 2nd measurement reversed polarity (check for thermal EMF)
19.05.2015: 3rd measurement normal polarity

no significant hysteresis

LMS interpolation of 19.05.2015

A 0 =  1.06481425389144E+0000
A 1 = -1.60431994557356E-0001
A 2 = -2.25165649662830E-0002
A 3 =  1.44281949166390E-0004

So T.C. from LMS at 25 deg C is -0.16 ppm/K 
-> not bad compared to datasheet and even better than most of my Z201_120 samples

The "box" T.C. is around -0.45 ppm/K including noise
and around 0.30 ppm/K from LMS interpolation (without noise)

No significant drift (1.15 ppm = within noise level) during the 3 days

With best regards

Andreas

[Andreas]

Hello,

Next PWW 120R resistors from Ultron plus (UP805_120)
Reference resistor is again UP805 1K #3

UP805_120#4 delivery date 1510
T.C. 3ppm/K from datasheet

21.05.2015: first measurement normal polarity
22.05.2015: 2nd measurement reversed polarity (check for thermal EMF)
23.05.2015: 3rd measurement normal polarity

no significant hysteresis

LMS interpolation of 23.05.2015

A 0 =  3.80606747298013E-0001
A 1 = -4.11315154136265E-0001
A 2 = -2.05365763017324E-0002
A 3 =  3.33819468006340E-0004

So T.C. from LMS at 25 deg C is -0.41 ppm/K 

The "box" T.C. is around -0.58 ppm/K including noise
and around 0.43 ppm/K from LMS interpolation (without noise)

No significant drift (1.09 ppm = within noise level) during the 3 days

With best regards

Andreas

[Andreas]

Hello,

Next PWW 120R resistors from Ultron plus (UP805_120)
Reference resistor is again UP805 1K #3

UP805_120#5 delivery date 1510
T.C. 3ppm/K from datasheet

25.05.2015: first measurement normal polarity
26.05.2015: 2nd measurement reversed polarity (check for thermal EMF)
27.05.2015: 3rd measurement normal polarity

no significant hysteresis

LMS interpolation of 27.05.2015

A 0 =  3.84755349289595E-0002
A 1 = -3.24576744031759E-0001
A 2 = -2.12792039565395E-0002
A 3 =  1.94117540677847E-0004

So T.C. from LMS at 25 deg C is -0.32 ppm/K

The "box" T.C. is around -0.53 ppm/K including noise
and around 0.39 ppm/K from LMS interpolation (without noise)

No significant drift (0.03 ppm = within noise level) during the 3 days

With best regards

Andreas

[Andreas]

Hello,

and again a overview.
This time for the up to now measured 120 Ohms resistors.

So Z201 metal foil against UP805 PWW resistors.

With best regards

Andreas

[Andreas]

Hello,

I have built a new toy to measure the resistors.
Similar to AN96 figure 8.
http://www.linear.com/docs/6637

To eliminate the thermocouple voltages I make a AC excitation of the resistor divider.
All is supplied from the voltage reference.
As the H-bridge does only switch between the measurements the switching noise does not affect measurement.

By adding the normal and the inverse difference voltages over the resistors
the thermocouple voltages are canceled out without losing signal to noise ratio.
(this would be the fact if I simply would switch off the reference voltage
each 2nd measurement to read the thermocouple voltage).

With best regards

Andreas

[ltz2000]

To eliminate the thermocouple voltages I make a AC excitation of the resistor divider.

Good idea. Many metrology grade platinum thermometers use the same technique.

[Andreas]

Many metrology grade platinum thermometers use the same technique.

Hello,

thanks for the hint:
do you have a schematics of (at least the analog frontend including ADC) of such a thermometer?

With best regards

Andreas

[Andreas]

Hello,

what does the AC-excitation bring in the end?
A measurement of the Z201_1K#2 with UP805_1K#3 as reference gives following results.

30.05.2015 normal multiplexer reverse polarity
31.05.2015 normal multiplexer normal polarity
06.06.2015 AC excitation (both polarities)

If there were significant thermocouple voltages I would see differences in hysteresis openings
(by different temperature direction up/down) between the diagrams.

Each +/- 1uV thermocouple voltage would give +/-0.4ppm difference.
Obviously I have no significant thermocouple voltages since I use the thermal
coupling of the wires with aluminium sheets (isolated with thermal silicone stripes)
according to the suggestion of Emanuel.

With best regards

Andreas

[Andreas]

Hello Emanuel,

you are right the cirquit in the paper seems to be really not the best.
3 different amplifiers used with 3 different drifts and offsets.
And also the LTC2442 has 10 fold temperature drift compared to the LTC2400 that I use.

The CN0155 cirquit is really the best (lowest component count) for ratiometric measurements.

Only the gain drift of the AD7195 is rather high with 1 ppm/K typ.
The LTC2400 is specced with 0.02ppm/K typ.
In my cirquit I also sense the resistors directly (4 wire) so that the RDSon does not influence accuracy.
In this case the RDSon only slightly reduces signal to noise ratio by some ppm.
Unfortunately the LTC2400 has rather large noise. (so I need to average many samples).
I am still looking for a ADC which has low gain drift and low noise.

With best regards

Andreas

[branadic]

Hi Andreas,

what about using something like picostrain instead of an ADC? They are used for weightscales measurements and have high resolution, ENOB up to 18.9bit. I was thinking of PS081. Maybe you want to check this way to measrure your resistors?

branadic

[Andreas]

Do you use a simple voltage divider or a wheatstone bridge?

Its a voltage divider with (pseudo differential) 4-wire measurement.
Schematics of the resistors is on the first page of thread:
With the AC-Multiplexer GND and VRef are interchanged after each measurement cycle.



With best regards

Andreas

[Andreas]

what about using something like picostrain instead of an ADC? They are used for weightscales measurements and have high resolution, ENOB up to 18.9bit. I was thinking of PS081. Maybe you want to check this way to measrure your resistors?

Hello Branadic,

perhaps it would be worth a try.

But I do not see how I can do a 4 wire measurement with such a device.
In my case the wiring resistance (and T.C. of wiring) cannot be neglected.

Do you have soldered break out boards for these devices?
(my hand soldering will not create reliable connections for the QFN package)

With best regards

Andreas

[Vgkid]

If you want to read up an an AC resistance bridge here is a manual of a LR400
http://www.lakeshore.com/ObsoleteAndResearchDocs/LR400.pdf
no schematic is given, but a rather detailed description is there.
I was debating on bidding on one earlier in the year, it was a parts unit that appears to have been in a barn.
I did not bid, it was sold.

[Andreas]

4 channels? Don't you use a LTC2400? (Several ones maybe?)

Of course I use a LTC2400 with a external multiplexer like MAX4052A or MAX4051A
and a buffer (ADA4538-1) between multiplexer and ADC.

With different ADCs you would have different gains and offsets within the signal chain.

With best regards

Andreas

[branadic]

Hello Branadic,

perhaps it would be worth a try.

But I do not see how I can do a 4 wire measurement with such a device.
In my case the wiring resistance (and T.C. of wiring) cannot be neglected.

Do you have soldered break out boards for these devices?
(my hand soldering will not create reliable connections for the QFN package)

With best regards

Andreas

Hi Andreas,

there is no possibility for 4 wire measurement, because the measurement principle is that one after another resistor is used to form a RC circuit with the same capacitor and discharge time is measured with a Time-to-Digital Converter with ps resolution.
I have the evalkit available at work with serveral measurement modules (high resolution module, standard module, wheatstone module), no seperate breakout board.
If you are able to fabricate a pcb I could solder the QFN package for you.

Regards, branadic

[Andreas]

Aaaaaaah... Ok! Now I understand why you said your measurements take long. I'm convinced this is a great idea to measure on the 4 channels to compensate for various wires losses and drifts, but you miss the possibility of true ratiometric operation (With excitation/output voltage being measured synchronously and not sequentially) that can be brought by a single differential input + reference ADC.

Hello,

I call the measurement "pseudo differential" because I do not measure the high and low voltage over the resistor at the same time.
Since I use the reference voltage as supply for the resistors directly. Each measurement is ratiometric.
I do not measure absolute voltages but only "voltage ratios" between the point of interest and the reference voltage.
So even reference voltage changes between the different measurements will not affect the ratio. (only the signal to noise ratio).
The long time is needed to average several measurement cycles. A single measurement has around 10uVpp noise.
So a single measurement would have around 4ppm uncertainity.
By averaging 85 measurement cycles the noise is reduced to about 1uVpp or 0.4ppm for a "1 minute measurement point".

With best regards

Andreas

[Andreas]

Hi Andreas,

there is no possibility for 4 wire measurement, because the measurement principle is that one after another resistor is used to form a RC circuit with the same capacitor and discharge time is measured with a Time-to-Digital Converter with ps resolution.
I have the evalkit available at work with serveral measurement modules (high resolution module, standard module, wheatstone module), no seperate breakout board.
If you are able to fabricate a pcb I could solder the QFN package for you.

Regards, branadic

Hello branadic,

so this measument system is only usable if all resistors are nearby and
 do not have different (exactly controllable) temperatures (which will need longer wires).

With best regards

Andreas

[Kleinstein]

The LTC2410 might be a slightly better ADC. At such a slow temperature cycle, I don't thing the rate of measurement is already critical - for a faster data rate and thus less final noise, one could in principle use 2 ADCs to measure both voltages (differential) simultaneously with periodic swapping of the ADCs - that is having two max4052.

[Andreas]

The LTC2410 might be a slightly better ADC. At such a slow temperature cycle, I don't thing the rate of measurement is already critical - for a faster data rate and thus less final noise, one could in principle use 2 ADCs to measure both voltages (differential) simultaneously with periodic swapping of the ADCs - that is having two max4052.

After looking at the data sheet:

On the first view the noise is halved at same speed. And by differential measuring the resistors you can double the measurement rate.

But the disadvantage is that the measurement range is limited to -Vref/2 ... +Vref/2 without over range.
So you have carefully to select the reference resistor so that both voltages over the resistors are below half of the Vref supply. So it will not be possible to use a 12K5 reference resistor to measure a 12K DUT.

With best regards

Andreas

[Andreas]

If you want to read up an an AC resistance bridge here is a manual of a LR400
http://www.lakeshore.com/ObsoleteAndResearchDocs/LR400.pdf
no schematic is given, but a rather detailed description is there.
I was debating on bidding on one earlier in the year, it was a parts unit that appears to have been in a barn.
I did not bid, it was sold.

Hello,

AC excitation: yes. But not sinusoidal. I want to have a ratiometric measurement. So during measurement the voltage has to be stable. Sinusoidal AC would complicate the evaluation with lesser accuracy. -> I need rectangular AC with stable amplitudes.

With best regards

Andreas

[Andreas]

Hello,

Next resistors: will be all 12K5 resistors measured with AC excitation.

First measurement:
Z201_12K5#1 as DUT with UP805_12K5#1 as reference resistor.

Z201_12K5#1 date code B1142-
T.C. +/-0.2ppm/K typ +/-0.6ppm/K  from datasheet

15.06.2015: first measurement AC
16.06.2015: 2nd measurement AC
17.06.2015: 3rd measurement AC

relative large hysteresis about +/-3ppm

LMS interpolation of 17.06.2015

A 0 =  3.97841696233471E+0000
A 1 =  5.80125106318059E-0001
A 2 = -7.92049996619505E-0003
A 3 = -4.17013633414582E-0005

So T.C. from LMS at 25 deg C is +0.58 ppm/K 

The "box" T.C. is around 0.59 ppm/K including noise
and around 0.55 ppm/K from LMS interpolation (without noise)

Visible drift (4.13 ppm) during the 3 days
So either reference resistor or DUT drifts.

With best regards

Andreas

[Andreas]

Hello,

2nd measurement: (the counterpart: both resistors exchanged)
UP805_12K5#1 as DUT with Z201_12K5#1 as reference resistor.

UP805_12K5#1 delivery date 1510
T.C. +/-3 ppm max. from datasheet (typical +/-1 ppm/K)

18.06.2015: first measurement AC
19.06.2015: 2nd measurement AC
20.06.2015: 3rd measurement AC

hysteresis about +/-1.5ppm

LMS interpolation of 20.06.2015

A 0 =  2.89372285322921E+0000
A 1 =  7.31179052260287E-0001
A 2 = -2.89803039080773E-0002
A 3 =  6.27147475010951E-0005

So T.C. from LMS at 25 deg C is +0.73 ppm/K 

The "box" T.C. is around 0.69 ppm/K including noise
and around 0.65 ppm/K from LMS interpolation (without noise)

So again the T.C. for the PWW resistor seems to be specced very conservative.

Visible drift (2.59 ppm) during the 3 days
So either reference resistor or DUT drifts.
I will exchange the reference resistor (Z201) with UP805_12K5#2 to see if the drift gets better.

With best regards

Andreas

[Andreas]

Who said that?

http://www.linear.com/solutions/1365

Hello,

I do not know what you want to say with that link.
The LTC2402 is more a dual channel LTC2400
(with the same overrange but noise doubled against the later)
and has no real differential measurement like the LTC2410

With best regards

Andreas

[Kleinstein]

This simple circuit will not fully compensate for the wire resistance. So it may work for an RTD but not for very precise measurements.

The way to go would be using a ref. resistor that is smaller (slightly less than 50%) of the DUT and use only the voltage at the reference resistor as ADC reference. So the reference resistor may need to be something like 10 K , 10 K and 470 K in parallel.

The alternative would be having 2 channels and measure the voltage on both resistors one after the other. This could be just one AD and a MUX (e.g. max4052) or 2 ADCs that are exchanges by two multiplexers. For testing drift the limited normal mode rejection should not be a problem, as the average voltage does not change much.

[Andreas]

I don't understand why it won't work with a 12.5k reference resistor and a 10k resistor under test. A LTC2440 (And others ADCs from the same family) accept 0...5V differential on the reference pins. Only the input is limited to -Vref/2 to +Vref/2.

In this case yes. But the other way round or when you have nearly equally resistors (with one resistor changing by +/-10 ppm) it would not work. (The measured value is limited to 2500mV).

This simple circuit will not fully compensate for the wire resistance. So it may work for an RTD but not for very precise measurements.

Thats what I have overlooked on the first view.

The only annoying wire resistance comes from the wire between the reference resistor and the resistor under test. The others are fully compensated if you connect the IN and Ref pins close enough to the resistors.

I have to keep the reference resistor on constant temperature and the DUT on variable temperature.
With 12K5 resistors the voltage drop between reference and resistor under test is 0.12 mV.
(With 1K0 resistors its even more).
0.12 mV against 2500 mV over the resistor is  around 50 ppm.
Copper wire has around 3850 ppm/K. With a temperature difference of 30 deg C
I would have up to 6ppm error by the wiring T.C.
-> 4 wire measurement is absolutely necessary.

With best regards

Andreas