I would like to ask if someone can explain the negative sign of the GAIN TEMP DRIFT (Gain vs. Temperature) which can be found in the AD8227 Datasheet.
Should one make a valuation of the total Error (in other words an Error Budget) as in Application Note AN-539, should the value of the error be signed, that is NEGATIVE? Which means that it should be subtracted from the Error Budget, fact which seems to be wrong...
Drift is the value assigned to the change of a given parameter over temperature. Like any change, only has two ways to go: either an increase or a decrease. When no sign is given, you can assume that the value in question will increase with temperature or that the direction of the change is unknown. When a negative sign is given, it means that an increase in temperature will result in a decrease in the value of the parameter in question. It does not mean that you should subtract it from the error budget. The uncertainty should be treated as an absolute value.
Let me explain:
The negative sign on the gain drift of AD8227 (for gains greater than 5, which require an external gain resistor) reflects the behavior of the absolute drift of the internal gain resistors assuming a perfect external gain resistor (with no drift) is used. In other words, we have observed that the gain decreases as temperature increases. While the slope of the change is always negative, the exact drift amount for any particular part is uncertain. However, we guarantee -100ppm/C or less (meaning it could be somewhere from 0ppm/C to -100ppm/C), since this is provided as a maximum.
For example, if you have a gain of 10 at a given temperature, you should expect to have a gain of no less than 9.999 if the temperature goes up by one degree and no more than 10.001 if the temperature goes down by 1 degree from the original temperature. Likewise, if your total temperature uncertainty is +/-5°C you should expect to have a gain that will vary as much as from 9.995 to 10.005 or a total uncertainty of +/-0.05%.
Some people are compelled to ask: can I use an external gain resistor with a temperature coefficient of -100ppm/C and this way track and eliminate the gain drift? The answer is no. Like I said before, we guarantee that the drift will be less than -100ppm/C but we do not guarantee that it will be exactly that much. Likewise, most resistor manufacturers will follow the same logic, so you don't really know what you've got. Moreover, there will be an inherent temperature mismatch between the internal gain resistors and the external gain resistor, which will make matters worse.
The easiest way to obtain the best gain accuracy over temperature, is by using an instrumentation amplifier that will provide the desired gain value with internal gain resistors, such as AD8228, AD8428, AD8290, AD8293G80 or AD8293G160. Of course, AD8227 will have excelent gain drift performance if used in gain of 5 (like most of our inamps when used in gain of 1 since all gain resistors are internal). Programmable-gain instrumentation amplifiers (pin-strappable gains, no digital interface required) also have internal gain resistors which perform very well, such as AD8231, AD8250, AD8251 and AD8253. Alternatively, you can choose an instrumentation amplifier that requires two external resistors to set the gain. In this case, you can use a resistor network of your choice, so that the gain drift will be set by the relative temperature coefficient (typically 5ppm/C for thinfilm resistor networks). Examples of such include AD8420, AD8237, AD8553, and AD8230.
I hope this helps. Please let me know if you have additional questions.