FP-G-140 error calculation . A bit of confusion.

[jqm]

Hello,

I am using the FP-SG-140 strain gage unit to do measurements on full bridge load cells for R&D purposes.
I am trying to compare the overall accuracy to some slang that is used in the field. For example "the load cell has an accuracy of 1 part in 10,000".

Maybe someone can straighten my logic out:


-------** Assumptions and SETUP**----------------
+/-3.5 mV/V scale, 60Hz Filtering, 15-35deg C, 4500 Ohm; full bridge with 10V excitation.

Total error = Offset Error + Gain Error + Input Noise
where from specification sheet:
Offset Error = 7.6uV
Gain Error = (0.06/100)*3.5 = 2.1uV
Input Noise = +/- 3 LSB (((?????? I dont understand how to relate LSB to uV ?????))

-------------------------------------------------
Say that Total ERROR = 13.5uV and I am trying to measure something that is 1.0mV;
(a) 3.5mV/V * 10V = +/- 35mV Full Scale
(b) thus using total error 35mV + 0.0135mV = +/- 35.0135mV
(c) ?????? I AM NOT SURE ABOUT THIS? ????
since the SG-140 has a 16 bit converter than...
there is a total of 65535 counts available but since the output can be bi-directional (+/-) you actually have 32767 counts
so:
(0.0350135 V / 32767 counts)^-1 = thus 1 part in 935839 ????


Could someone straighten me out as this number (1 part in 925839) seems to be a bit high, also could someone explain the relation of the input noise (LSB) and a voltage?


Thank you for your time
Jason

[chericks1]

Hi Jason,

Total Error =
Gain Error + Offset Error + System Noise(and Quantization Error) + Nonlinearity Error + Temperature Error.

Nonlinearity Error = very small, Temperature Error = N/A

Total Error = Gain Error + Offset Error + System Noise

Offset Error = 7.6uV
Gain Error = 0.06% * Input Voltage
System Noise = +/- 3LSB

The relation of system noise to voltage:

This relation has a lot to do with the resolution of the device. The FP-SG-140 has 16-bit resolution so the number of represented divisions is 2^resolution in bits = 2^16 = 65536. Because the system noise specification is +/- 3LSB, this converts to +/- (2^3) = +/- 8 = 16 divisions. Now, the error can be calculated in Volts. (16 / 65536) * Range = System Noise Error. The Range is +/- 35mV = 70mV.

(16 / 65536) * 0.07 = 17.1 uV

Computing Total Error for 1mV input = 0.6uV + 7.6uV + 17.1uV = 25.7uV

If this error was generalized for the entire "full scale" range of the device, then you could say the load cell has an accuracy of:

(.0000257 / .035) = 1 part in 1,362

Hope this helps!

Chad AE
NI

[jqm]

Chad AE,

Thank you for straightening me out!

wow, 1 part in 1362 seems now to be a little bit low!
((maybe there will be a 24bit version someday))


Wouldnt the generalized error be calculated by taking the Total Error / 0.070 as the range was +/- 35mV Full Scale?


Thanks for your time and detailed reply
Jason

[Aaron_G]

There is an error in the calculation. The System Noise is 3 LSB or 3 Least Significant Bits, this is 3 in 65536 not 2^3 in 65536. Because it is noise, it can range from +3 LSB to -3 LSB. This is an instantaneous error that can range to either extreme at any given time, but it will never be at both extremes at the same time. For worst case accuracy, assume the noise will be the maximum of 3 LSB, thus the System Noise is 3 /65536 * 70 mv = 3.2 uV on the +/-35 mV range.

Total Error (1 mV signal at +/-35 mV Range):
Total Error (1 mV) = Gain Error (0.6 uV ) + Offset Error (7.6 uV) + Noise (3.2 uV)
Total Error (1 mV) = 11.4 uV

Dimensionless Error (1 mV) = 11.4 uV / 70000 mV range = 1 part in 6140.

Varying the input signal over the range from -35 mV to +35 mV will only effect the Gain portion of the Total Error, the Offset & Noise errors are constant for a given range.

Regards,
Aaron

[jqm]

That looks a little better.

Thank you for your time