Should ADC reference pin also be a kelvin connection?

Question

Reasking this question with a proper schematic:

I am doing a ratiometric measurement of a Thermistor. The thermistor is located over 5 metres away from the ADC and all other components. This is going to be a resistance of the cables of upwards of several Ohms.

The opamp driving the ADC will be doing a Kelvin 4-point measurement.

The Thermistor is supplied by a 2.5V Voltage source, which will be the reference for the ADC.

However, where should I connect the reference pin of the ADC - on the PCB, or 5 metres away right on the thermistor voltage divider?

If I do it on the PCB, the voltage the ADC sees as the reference will be different to the actual voltage 5 metres away will it not? Will this defeat the point of a ratiometric measurement?

Then if I do it 5 metres away, there will be a lot of inductance in this wire, so I will need a larger buffer capacitor next to the Vref pin to ensure the voltage doesnt have large spikes during each conversion

Which option is better?

Answer 1

You can still do it with just 4 wires !

1. Recognize that it is only important to know the current in the thermistor and the V across it.

2. Move the upper resistor back into your board (i.e. in series with the output of the 2.5V. Connect a channel of the ADC to each end of the resistor (say VREF and VOUT). Now the current is (VREF-VOUT)/R.

3. With kelvin connections to the thermistor, measure the V across the thermistor itself -- the thermistor resistance is that V divided by the current measured above.

4. Any 'extra' wiring resistance between the output V and the thermistor (or in series with the thermistor's GND will just reduce the actual current flowing, but the calculation above will negate that effect, and still will be correct.

Answer 2

You'll want a Kelvin connection if the gain error caused by it will be less than the gain error you'd get without it. So you'll have to evaluate the designs and calculate errors for each of them. The three designs to evaluate, at maximum wire length, are:

1. Direct reference connection.
2. Kelvin reference connection with a series resistor acting as an LC spike isolator.
3. Kelvin reference connection with a transient voltage suppressor acting as the spike isolator.

When designing the reference input decoupling capacitor against dynamic current spikes, you should probably assume an open circuit anyway. The capacitor's parasitics will dominate the AC response. The layuout, wiring and connector series resistances will determine the gain error. The resistive spikes must be suppressed by series isolating resistance or parallel isolating shunts.

In most cases a 10uF tantalum parallel with 100nF NPO ceramic will work well, with an isolation resistor so that any source inductance won't produce spikes that destroy tantalums. You should be able to determine the effect of the series resistance, including wire and connector resistance, on the gain of the ADC. In most cases the currents are in the tens of microamps or less, and with up to 10Ohms of series resistance cause gain errors of a couple LSB at most at 16 bits.

Tantalums do not tolerate LC voltage spikes beyond their rated voltage. Each such spike will eat up the life of the capacitor and lead to it failing short, either setting itself on fire, or melting the wire harness. Thus the need for series resistors, or fast-acting transient voltage suppressors. TVS is a good alternative to isolating the inductive spikes from the decoupling capacitor.

Do not use X or Z ceramics: they are microphonic and vibrating the circuit board will modulate the reference voltage. That's not usually desired.

Answer 3

^{simulate this circuit – Schematic created using CircuitLab}

I would prefer you maximize your CMRR in the cables and sensor + 3 matched resistors. Shield is terminated at sensor end only to prevent ground loops.