Op-Amp Accuracy

Different experts have different criteria for what is "good". Either pick just one and do everything he says, or understand the reasoning behind the recommendations.

In the case of your input offset vs. range problem, the output of the amplifier is going to be more or less equal to gain * (offset + input). That should make things clear right there.

If that doesn't help, then consider this: if you have the full offset of

35uV on the amp, then you will read it as a current of 35mA. Can you stand a current measurement error of +/- 35mA? If so, then you're home free. If not, then you need to look at different amplifiers, or more clever ways to use the one you have.

--
My liberal friends think I'm a conservative kook. 
My conservative friends think I'm a liberal kook. 
Why am I not happy that they have found common ground? 

Tim Wescott, Communications, Control, Circuits & Software 
http://www.wescottdesign.com

off of a low side shunt. I need to be able to read +/- 30A, I'm thinking I'll use a 1 mOhm shunt and read +/- 30mV across it, but I'm having a hard time understanding how to figure out the accuracy or precision I'd get from an Op Amp.

of 10mV full-scale"

accuracy you will need 100mV full scale across the shunt.

full scale...

First, keep in mind "specsmanship". When an instrument guy says "1% accuracy", what they generally mean is that +/-1% of the rated full scale input is the maximum error (under some perhaps optimistic conditions such as 25+/-0.1°C and within 45 seconds of last calibration). This number is the smallest that can honestly be quoted, so it's a popular one to quote.

So, +/-1% of a 30A full scale reading is +/-300mA. That means that you could have a +/-10% of reading error in a 3A measurement, and a

300mA-ish measurement would be meaningless.

A typically more useful spec is a percentage of reading (gain error) plus an offset (zero error). Your gain error will be a function of the resistor accuracy (trimmable) and the amplfier gain (not well controlled and temperature-senstive) relative to the gain required in your circuit. For example, if you're turning 30mV into 3V, you have a gain of 100, so your amplifier needs to have a gain of around 100,000 to keep gain errors due to the amplifier around 0.1%. Not much of a problem at DC, but this can quickly become a problem if you need high frequency response- for example to accurately measure a 50kHz switchmode signal with a gain of 100, you'd need a gain-bandwidth product in the GHz range. It's also possible the shunt doesn't look like a pure resistor even at audio frequencies. To get 0.1m ohm of impedance at 50kHz takes only 0.3nH of inductance- which is not very much- about the inductance of 1mm of straight wire.

Also, keep in mind that a stable offset is not inherently a source of error- it can be calibrated out (just measure the voltage on the amplifier output with no current and subtract that number from future measurements). If the offset changes with temperature- which it will- (and is uncompensated) or time, then it becomes an uncorrectable error. Same with gain errors. Temperature changes are both externally and internally provoked.

There are several sources of offset error in making small DC measurements such as you are looking at doing- offset (Vos) and offset drift of the amplifer (TCVos), and thermal EMFs in the shunt connections. I assume you're using a 4-wire shunt resistor so the connections are properly designed. Asymmetry in the heat sinking can lead to temperature differences and thermal "tails" on the response etc.

Generally, at least for relatively high frequency test purposes, it may be easier for you to use a feedback-type sensor with a good (and specified) bandwidth such as LEM products rather than trying to roll your own.