Hi,
the data sheet of the CR1000 states (for analog inputs, at temp. = -25°C to 50°C):
Accuracy: +/- (0.12% of reading + offset)
So there is a max. possible absolute error ("offset") and a max. possible relative error ("of reading").
In case of a single-ended voltage measurement (-5 V to 5 V), the max. absolute error should be 0.004 V.
The max. relative error should be 0.12% * measured value, e.g. 0.0048 V in case of a measured value of 4 V.
If this is correct so far, then the next question is how to interprete these values.
A "max. possible error" should be identical to an "uncertainty limit", which can be converted to a standard uncertainty,
e.g. assuming a rectangular distribution (of the probability density function) which requires a multiplication by (1/sqrt(3)).
Finally, there is the question how to combine these two standard uncertainties (absolute and relative),
in order to calculate the combined uncertainty which results from these two components.
This depends on the assumption of either a full correlation of these two components (-> linear addition)
or a non-correlation (=independence from each other -> quadratic addition).
(Theoretically, a partial correlation might also be assumed, but this is unusual in our practice.)
Which assumption should be applied?
Thanks
Voltmeter manufacturers (Agilent/HP, Keithly, Fluke) traditionally use the term accuracy instead of uncertainty, a tradition we follow on our dataloggers. A voltmeter accuracy specification usually consists of two portions, a % of reading portion, and an offset or “Percent of Range” portion. These two components of accuracy model instrument performance where there is a signal independent portion of the error (offset), along with a signal dependent portion. Resolution limits on the larger input ranges, and thermal EMF’s on the smaller input range, along with non-ideal offset compensation result in non-zero offset (% of Range) errors, hence the offset portion of the CR1000 specification. Errors in the on-board voltage reference used for the analog-to-digital conversion result in measurement errors that are proportional to the signal to be measured, hence the % of reading portion of the error. While these two error components are independent, they are intended to be added together to arrive at a total measurement error. For example total voltmeter error calculations from Agilent/HP and Keithly simply add the % of reading and “% of range” (offset) errors together and then present a result with a confidence level that is probably at least 99%. There is no standard or agreed upon confidence level for specifications that I’m aware of with voltmeters. As stated on pg. 31-4 of “Calibration: Philosophy in Practice” ISBN 0-9638650-0-5, 2nd Edition, 1994, Fluke Corporation, “It should be remembered that specifications do not equal the performance, they are performance parameters. They can be conservative or aggressive. Manufacturers are not bound by any convention as to how they present specification. Some will specify their products conservatively. Such instruments will usually outperform their specifications. Other manufacturers may manipulate the specifications to make an instrument appear more capable than it really is.” Also on pg. 31-5 it states “Fluke uses a 99% confidence level for its specifications for calibrators and standards”.
We adopt a conservative specification approach for our CR1000. For units within the 2 year recommended recalibration interval, we have a 99% confidence level of the logger meeting the voltage measurement accuracy specification arrived at by adding the % of reading and offset errors together. You could root-sum-square the % of reading and the CR1000 offset errors together for a reduction in total error and confidence level.