Bridge sensor calibration calcs (2023 Update)

Work out - measure - what the current actually is and scale the correction by "actual current" divided by 1mA. It means multiply by roughly 0.7 in your application, which doesn't sound too bad.

People do bridges ratiometrically, but the calibration shop will do whatever is easiest for them, which usually means sticking with what they've always done..

<snip>

Measuring the current at 2.5V (or measuring the bridge resistance) is something I'd like to avoid if possible. It seems to me I *should* have all the information I need, but I can't quite work it out.

Yes, that seems likely.

It's a resistance sensor, so just turn the offsets into resistance changes using V=IR, then plug that into your bridge calculation.

Cheers

Phil Hobbs (Who is this very minute doing something similar with a PbSe photoconductor, which is the same idea but more complicated.)

Yes, of course! Thanks.

Unfortunately, I can't make it work without knowing the bridge resistance.

Let's say a full bridge with two pairs of resistors

____|____ | | R1 R2 |--(V)--| R2 R1 |_______| _|_

For Vex excitation, V = Vex * (R2-R1)/(R1+R2)

At 1mA constant current excitation Vex = 1E-3 (R1+R2)/2

So V = 1E-3 (R2-R1)/2

I know (or I'm told) the V at 1mA excitation under some defined conditions, and I want to know the V at, say, Vex = 2.5V. But I can't. Bugger.

There could be a modest accuracy penalty for using a constant-voltage excitation, as opposed to cc. It depends on the tempco of the resistors. If they are silicon, the difference could be significant.

On a sunny day (Fri, 6 Aug 2021 17:02:51 +0100) it happened Clive Arthur snipped-for-privacy@nowaytoday.co.uk> wrote in <sejmfd$oiq$ snipped-for-privacy@dont-email.me:

Cannot be that hard to make a curent mirror at 3.3V,

2 transistors and a resistor?

Discrete transistor based mirrors are horrible. Integrated ones are fair.

Post the datasheet. I suspect you're making this way too hard.

Cheers

Phil Hobbs

>

Keller series 7 LI is the part, or close to it. They'll give you calibration stuff with 1mA constant current excitation. Ain't no way to work this into a constant voltage ratiometric thing without device-specific measurements

Cheat a little!

Vref - good resistor - bridge - ground.

Have the resistor make close to 1 mA.

Have the ADC measure the diff drop across the resistor, then the diff bridge output, divide.

We measure 4-wire RTDs that way.

What's the bridge resistance?

<snip>

You can buy dual transistors which are both on the same substrate. Analog Devices make a few.

The gains can match to 0.5%. Stick a pair of 0.1% resistors in series with each emitter, and they do a bit better. Throw in a trimpot, and they can do better still.

People who do (or once did) electronic design know about this kind of stuff. John Larkin doesn't seem to.

0.5% is about right for the sensor, but it would have to be accurate up to 180'C. Something with feedback would seem better on the face of it, but it's well worth a look.

The bridge is 3500R +/-20%. The given calibration figures are for 1mA, but they do say that you can run from between 500uA and 1.5mA and scale accordingly.

So, I could run it at 700uA which is nominally 2.45V, then use 850R (okay 820R) from 3.3V supply, or maybe a ref. As it happens, the ADC I'm using has a spare differential channel and its input can go as high as the supply. It also has programmable gain, so that looks like a result thanks.

Not quite as simple as ratiometric, but hey.

Well, throw in several trimpots and temperature cycle the rig and tweak the pots until the tempco and supply sensitivity are low enough. That should only take a couple of days per unit. That seems to be the professional way to do it.

That sounds better than my single resistor ratiometric suggestion.

It is ratiometric, because you need to divide! A stable ref is good because the ADC won't measure the current and the bridge output simultaneously.

You can use the drop across the curreent sense resistor *as* the ADC ref, but that can have issues.

The diff-input 24-bit delta-sigma ADCs are perfect for ratiometric measurements.

TI's ADS1247 is nice. Some of the Analog Devices parts are weird and flakey on the digital side. TI knows digital!

If I understand you correctly (?) the resistances in the bridge change with temperature so you're going to be getting a different calibration without that information. Can I assume you've already consulted the manufacturer about what magnitude of error to expect?

Seems like constant current excitation is the easiest solution that preserves full usability of that calibration data.

It wouldn't be too hard to make a constant voltage buffer that provided a current signal you could measure, but that seems more complex than just making a constant current source to begin with.