Electronic – ADC eliminating noise for improved accuracy

adcisolation

I am working on a 24 bit ADC and would like to eliminate noise as much as possible in order to maximize the performance of the 24 bit ADC.

There are two main areas where I am concerned where noise can come from:

  • Internal noise where the noise from the circuit itself.
  • External noises where noise from the environment can affect the reading.

I have a general idea of how to protect from these noises but I lack specifics if my counter measures are enough.

Internal Noise

Internal noise from traces and power supply can affect the reading so I have used digital isolator to isolate the ADCs both power, ground, and data lines from the rest of the system. The power supply can either come from a power over rthernet module or a simple 5V USB cable.

Taking an advice from an answer to a previous question, the power supply is filtered by a series of low – pass RC filters. Should I place low pass filters on the AVDD of the ADC, too?

Parts used that is involved in Power Rails: ISOW7481 ISO7731

enter image description here
enter image description here

Another solution I found is using a multi-layered PCB with the signal traces be like this. According to an article I read, it helps reduce noise. With the top and bottom layers are either GND or Vcc.

enter image description here

External Noise

This can be a bit tricky since the device will be in close proximity (less than 1 meter) of 50Hz AC mains lines. Those lines carry a maximum 50-75A of AC current (beyond this master breaker would trip.) My first idea is to use some of this shielding material which can be tediously cut and soldered to the board. Just like one of those wifi/bluetooth modules that has a metal covering the IC. Would that be sufficient? Also does shielding need to be connected to the ground of the circuit? If so which ground (digital or analog?)

enter image description here

If that is not enough I might go as far as placing the whole module into a solid metal container with holes just enough for the wires to go through. This should block those low frequency noises. Does it block the high frequency noise too? If so then the shielding above may be redundunt, but how thick would that need to be for it to be effective? What metal should it be?

enter image description here

Since its metal near power lines, I assumed this solid metal should be connected to earth. Should this metal also be connected to the circuit's ground? Would that affect the reading?

Conclusion

Although I may not need this many bits, I may as well go all out. Even modern oscilloscopes do not need this much of bit accuracy. Noise can still be induced by the probing. But since this project allows me to go a bit further I might as well go ahead and learn as much as I can.

Best Answer

24 bits is hard. luckily most 24 bit ADC's have an ENOB (effective number of bits) of around 19, which means your measurements "only" need to be good to 2ppm of the input voltage range,

Supply noise: This is where you pull up the PSRR (power supply ripple rejection) of your ADC and work out how much of an error source a noisy supply will cause, in these cases I tend to break these charts up in to frequency bins so I can add up all the noise sources for different bands to know exactly how I need to fight it, independent noise sources add as the sum of square roots,

You can also pull up the expected noise output graph of your power supply and see how much you need to fight it down

Then there is thermal noise to add in to the mix, as the ADC will have its own internal noise in the measurement

Next up external noise sources, for this level of resolution, you want differential measurements, this makes it easier to remove most of the impact of common mode noise, and when differential the ground in your PCB is acting more like a shield than a signal wire, on top of this you will want some capacitance between the differential lines to soak up the differential mode noise.

For any capacitors in this circuit, I would also recommend NPO or COG, you do not need the capacitor being part of the noise source.

Shielding is ideally to your circuits ground, you want that enclosure your chip is in to always remain the same voltage as it, otherwise it will start coupling in to sensitive nodes.