I'm getting to know the Freescale Kinetis L microcontroller. It's ADC has a lot of options, but long story short, it seems like I should be able to get 16-bit resolution in either single ended or differential mode.

The sensors I want to read vary their resistance, so I was going to just hook them up in a voltage divider to determine the resistance:

simulate this circuit – Schematic created using CircuitLab

Is there any benefit to using the differential inputs in this scenario? I was thinking maybe it would have better common mode noise rejection in this mode. Am I over-thinking it? The sub-family datasheet certainly makes me think I get better resolution (effective number of bits) in Figure 7 and Figure 8... So when is it appropriate to use the Differential Inputs?

There is virtual zero benefit in using differential inputs for your application. Assuming the sensor is close to the uC, you'll be able to get almost identical results from either single ended or differential, +/- a couple LSB.

The real benefit of differential signals is in long cable runs and noisy environments, for increased common mode rejection. They're also very useful in instrumentation, when dealing with very small signals. A common way to deal with sensors that vary their resistance in small amounts, such as strain gauges and pressure cells, is to use a Wheatstone Bridge.

simulate this circuit – Schematic created using CircuitLab

This configuration by itself centers the measurement in the ADC span. $V_+-V_-=V_{differential}$ It's important to match the default resistance as close as possible. Any deviation from the default resistance will show up as offset in the measurement.

Again, this is done because:

1. the changes in resistance are very small
2. more often than not, the sensors are far away from the data acquisition and processing equipment
• very nice answer... can you quantify "close" and "long" with rules-of-thumb? – vicatcu Mar 18 '13 at 13:59
• The only rules of thumb I really know of when it comes to single ended and differential are in terms of audio. Generally, single ended is good for about 20 feet before the cable starts acting like a low pass filter. Differential signals can travel very long distances. When I was doing live audio, we sent line level music 1000+ feet down a river bank for a fireworks show. Every 200 feet was an amp rack, with the input impedances of the amps in parallel. Granted, line level is up to +24dBu, but that's a long distance to cover, and it wouldn't have been possible without the differential signal. – Matt Young Mar 18 '13 at 19:44

To REALLY answer the question, we'd need to know what sort of resistance changes you expect to see. If they're small -- a few percent of baseline, the correct answer is the Wheatstone bridge configuration and the differential input seen above.

We'd also need to know what sort of noise you can tolerate. If The signal is big and you don't care about noise, it makes no difference what you do.

In the medium situations, things get a bit grayer. If the signal is going through a long cable in an electrically noisy environment, using twisted pair cable and differential configuration will get you a cleaner signal.

Lastly, if your tolerance for noise is low, whether you use differential or single sided configuration may not be as important as preamplifying your signal so as to cover nearly the full range of the ADC.

The best approach is to figure out how many bits of effective resolution you need, and then calculate how much resolution you're going to lose to noise (might need to measure to see this), and then plan what you need to do to get those numbers to match.

Generally, if an ADC provides a differential input it does so to maximize its performance. I just had a quick look at the data sheet and didn't find too much info on how to connect the inputs differentially and I have a concern that the way you are anticipating connecting them may be incorrect and/or non-beneficial.

Most (maybe all of them?) differential A2Ds require a true differential input to achieve the best performance. I would have thought you need to input diff signals that are anti-phase and centred around half Vref.

On another point, the input resistance of the analogue inputs is quite small and you would get possible errors directly connecting your sensor. I'd urge you to get more information on the ADC before committing to a design that you may be disappointed with.