3

I want to connect a 4-wire PT100 sensor to a MAX31865 RTD-to-digital converter.

The MAX31865 has a resolution of 15 bit, however I need a resolution of 17 bit.

Will I be able to use the oversampling and averaging principle (AN118) to gain 2 effective number of bits by averaging 16 sampled values?

Velvel
  • 3,591
  • 3
  • 12
  • 30

2 Answers2

5

If your 15 bit readings have noise then oversampling will improve the resolution.

If the 15 bit readings are noise-free (stuck in a bit), then oversampling will not improve anything.

The analog noise of the conditioned sensor signal must be at least 1/2 LSB of the 15 bit converter. If noise is lower, your 15 bit converter will just report the same value over and over and averaging will not change this outcome.

tobalt
  • 18,646
  • 16
  • 73
  • Since temperature noise has an approximate distribution of white noise, I'd assume the PT100 will never be noise free. Would you agree? – Velvel Nov 21 '22 at 15:10
  • 2
    No, i wouldn't agree. it will surely have analog noise, but maybe not enough, to be larger than 1 LSB on a 15 bit ADC. Depends on your signal conditioning. – tobalt Nov 21 '22 at 15:13
  • The J-N noise voltage of a Pt100 is <15nV/rtHz RMS at room temperature, which is negligible by a couple orders of magnitude at the excessive current that this chip typically uses. – Spehro Pefhany Nov 21 '22 at 17:59
  • 1
    @SpehroPefhany It is even below 1.5 nV/rtHz. – tobalt Nov 21 '22 at 18:12
  • 1
    If you don't have natural noise, [artificial](https://en.wikipedia.org/wiki/Dither) is fine too. – Simon Richter Nov 22 '22 at 05:10
1

As long as the sampling (and other) noise meets certain requirements, such as non-correlation with the signal, and you can guarantee that the input signal holds steady for your repeated samples, averaging \$N\$ samples will result in a noise reduction by a factor of \$\sqrt{N}\$

Scott Seidman
  • 29,274
  • 4
  • 44
  • 109
  • the main requirement is that thr noise is larger than 1/2 LSB of the converter..If noise is lower, it wont work. – tobalt Nov 21 '22 at 15:14
  • @tobalt since the IDEAL quantization noise is \$\pm \frac{1}{2} \text{LSB}\$, isn't that pretty much guaranteed? – Scott Seidman Nov 21 '22 at 15:24
  • 3
    @ScottSeidman quantisation noise is only +/- 1/2 LSB peak IFF the ADC is making several LSBs of transitions, whether from the actual signal, or from an added dither signal. If the ADC is quiet, and the signal is quiet, and no transitions are being made, then 'quantisation noise' doesn't exist. – Neil_UK Nov 21 '22 at 16:28