How to remove noise in ADC voltage measurement?

Question

I want to measure the open circuit voltage of a battery cell (differential reading, that is A0-A1). In order to do that, I am using an industrial PLC based in Arduino Mega and a cheap but accurate ADC (ADS1115). I need to get an accuracy of 100uV (which I should be able to get with ADS1115 by far). However, due to some unexpected noise/currents, my accuracy is only 0.01V (really bad). This means that, if I want to measure for example 1.3005V, I get one second 1.3005 volts, the next second I get 1.3030 and the next second 1.2980 and so on. I know that the real voltage value has not changed at all, as I am checking its value with a BTS system (big and expensive separated equipment). I cannot have so much change in so little time, specially changing from increase to decrease in value. I need the measurement to be completely stable along time. Averaging the value alone is not an option.

This is my set up. Note that I am using a breadboard to connect everything.

Some comments which can help to the diagnosis:

1. Could the problem be caused by the grounds? As I understand, with this set up I am having common grounds, and this could add current paths and create noise. I have read that it is recommended to isolate the analog ground (ADC) from the digital ground (PLC) by adding an isolator. Could this help? Anyways, I have read that an ADC has one digital ground but also an analog ground. However, in the ADC datasheet I find only one ground (the analog), which comes from the PLC.

2. Could the problem be related with common Vdd? I mean, the same power supply is used first by the PLC, and then, the PLC gives the Vdd to the ADC. In the past, there was even no external power supply, so the ADC was powered by the proper Arduino Mega by the USB cable. This generated some big variations in the ADC reading that improved after adding the power supply. Please, note that the USB cable is always connected and with traffic (for debugging use only).

3. I have read on this thread that it is a good practice to add an small capacitor to the ADC input to hold the signal. Could this help? I have not added one in my set up.

4. ADS1115 datasheet suggests to add an small cap (0.1 uF) to decouple Vdd. Do I still need to to this, even though that Vdd comes from the PLC?

5. Also ADS1115 datasheet suggests: Analog inputs with differential connections must have a capacitor placed differentially across the inputs. As I am doing a differential measure, could this improve the reading?

6. I have already tried to add a low pass filter to one and both ADC inputs, but the measurement variation got worse!

7. IMPORTANT UPDATE: As i briefly mentioned in the beginning (but I didn't put it in the drawing) I also have a Battery Test System (BTS) connected to the battery cell in order to charge/discharge it. This BTS operation is the following: it provides positive current (around 300 mA) during charge cycle, and then for the discharge cycle it provides -300 mA, then charging again, discharging, and the loop goes one like this. I say this right now because, when I tried to connect the (-) side of the battery (=A1) to the analog ground as some people have pointed out, it removed the noise. HOWEVER, during discharge cycles, a voltage drops appears and the battery loses a lot of its capacity (which is not feasible). As one person has suggested, maybe this is because the current is able to have a path and then it creates a voltage drop? Is there an easy way to avoid this current?. This is the set up with the BTS included:

Answer 1

A differential measurement still needs a ground reference point, due to how most ADC front ends/signal conditioners are designed.

You need to connect the (-) side of the battery to both one analog input and to GND of the ADC module.

The decoupling capacitor on the supply rail to the ADC is always needed. Sometimes you need two of them for good performance. The capacitors need to be on the PCB where the ADC is. They won't do much good if they are connected via the impedances present in typical breadboard construction.

Answer 2

Question 1:

I do not expect ground currents in your analog path that could mess up your readings. The ADC is at an end point of the supply path.

Question 2 and 4:

Yes, absolut possible. A series inductor of 10-100 µH and a capacitor of at least 100 nF at the ADC supply side are good style. This improves the performance of the voltage reference and the amplifier inside the ADC.

Question 3 and 5:

This is true for ADCs, that do not contain buffer amplifiers, like most ADCs integrated in microcontrollers. And it may help on noisy or high impedance signal sources. Your system has a buffer amp in the ADC and the differential noise from a battery should be very low. Summary: This will probably not help.

BUT we can expect a large common mode voltage noise. The battery is not connected to a reference point of your circuit. It has a random floating charge and receives all sort of noise from the environment. The common mode input resistance of the ADS1115 is 6 Mohm in this mode and can not deal with this noise, which can reach several volts in both directions.

The voltages at the ADC inputs may temporary leave the allowed range and the difference seen by the ADC is lower in this case.

The quick solution is to connect battery minus (=A1) to ADC AGND. This way the circuit is no longer a bipolar instrument.

If you want the ADC to operate more in the center of the input range, provide a voltage divider with 2 equal resistors between VDD and AGND and connect this to A1. A capacitor of some µF between A1 and AGND will cancel the common mode noise. This solution allows bipolar differential measurements.

Answer 3

Could the problem be caused by the grounds?

Yes and the cabling could also do this. Current going back through the cable between the ADS and the PLC can cause voltage differences on ground. Know what the current is and use a decently sized wire (the resistance in the wire will create voltage noise V=IR)

Could the problem be related with common Vdd? Yes, measure the ripple and then look at the PSRR and calculate the noise. PSRR is a figure that tells you how much the power supply will affect the measurement.

I have read on this thread that it is a good practice to add an small capacitor to the ADC input to hold the signal. Could this help?

Adding filtering (not a capacitor) helps in rejecting noise, you can't filter everything though because you will start filtering signal. This must be calculated

ADS1115 datasheet suggests to add an small cap (0.1uF) to decouple Vdd. Do I still need to to this, even though that Vdd comes from the PLC?.

Yes you do, it will help with noise from power and ground and common mode noise from cabling. Ideally you should have a voltage regulator next to the ADS (and your reference should be on the same board if it needs one)

Also ADS1115 datasheet suggests: Analog inputs with differential connections must have a capacitor placed differentially across the inputs. As I am doing a differential measure, could this improve the reading?

I would do what the datasheet says.

I have already tried to add a low pass filter to one and both ADC inputs, but the measurement variation got worse!

If you are doing a differential measurement, you need a differential LPF, if you only filter one side then the noise will not be subtracted differentially and will look worse in most cases.

Answer 4

You need an anti-alias filter in hardware

You say

Filtering/averaging is not an option

However if you don't low-pass-filter in hardware at the Nyquist rate or lower, aliasing is going to seriously mess up your measurements. As VoltageSpike says, for a differential signal you'll need filters on both sides of the signal.

According to this link the ADC can sample at up to 860Hz. So all the noise above 430Hz (or whatever speed you measure at) is getting "reflected" back down. This isn't going to be pretty. Just by the way, this is one of the reasons that early CDs in the 1980s had a bad reputation, because they hadn't properly worked out anti-aliasing on the ADCs back then.

You would benefit from filtering in software

Battery voltage doesn't change that quickly, so consider how fast you genuinely need this to respond. If you're sampling at 5ms intervals but you only need the data every 100ms, a digital low-pass filter in software will definitely improve your signal.