Pressure water level sensor: different readings

Vitruvius · June 21, 2026, 12:49pm

I have set up a DFRobot KIT0139 pressure water level sensor. The board (Olimex ESP32-PoE2) is delivering 24V to the sensor. The returned voltage is read out through one of the ESP32 GPIOs. I read ADC on the ESP32 can be quite noisy. So I suppose this is common behaviour?

The GPIO I'm using on the ESP32 has a 10k resistor, might this be interfering with the readings?

I'm going to use the exponential moving average filter to even it out a bit.

A few sample readings:

[13:53:51.178][S][sensor]: 'Water tank level' >> 1.54 V
[13:54:21.183][S][sensor]: 'Water tank level' >> 1.47 V
[13:54:51.176][S][sensor]: 'Water tank level' >> 1.41 V
[13:55:21.181][S][sensor]: 'Water tank level' >> 1.53 V
[13:55:51.175][S][sensor]: 'Water tank level' >> 1.48 V
[13:56:21.179][S][sensor]: 'Water tank level' >> 1.51 V
[13:56:51.171][S][sensor]: 'Water tank level' >> 1.48 V
[13:57:21.175][S][sensor]: 'Water tank level' >> 1.49 V
[13:57:51.184][S][sensor]: 'Water tank level' >> 1.45 V
[13:58:21.172][S][sensor]: 'Water tank level' >> 1.48 V
[13:58:51.180][S][sensor]: 'Water tank level' >> 1.46 V
[13:59:21.185][S][sensor]: 'Water tank level' >> 1.50 V
[13:59:51.178][S][sensor]: 'Water tank level' >> 1.51 V
[14:00:21.182][S][sensor]: 'Water tank level' >> 1.43 V
[14:00:51.175][S][sensor]: 'Water tank level' >> 1.53 V
[14:01:21.179][S][sensor]: 'Water tank level' >> 1.43 V

Thanks

neel-m · June 21, 2026, 2:35pm

Not sure if there was a question there.

Yes, the ADC in esp chips are known to be less than great.

Yes, it is possible to make them better with software.

You didn't show the design/layout of your circuit. The sensor you are using has a current output, so you need something to convert it to voltage. What are you using?

Here is a good basic description of one way to do it.

For the few analog sensors I have (ratiometric water pressure sensors) I use them with the ADS1115. It is a decent quality ADC, so it needs much less software help to get good readings.

Vitruvius · June 21, 2026, 3:06pm

Thanks for responding, and my apologies. First time I built something like this, I've been using ESPHome for little more than BT proxies until now. The questions are:

If this is what one can expect when using the ESP32 ADC indeed.
If using a GPIO with a resistor would cause less accurate readings.

I'm using the DFRobot KIT0139 which comes with a SEN0262 analog to current voltage converter that can send up to 3V.

This is the layout:

Karosm · June 21, 2026, 3:52pm

Looks noisier than my experience with ESP32. But I don't know if you use multisampling or not (you should). Also, don't you have any free adc pin without resistor?
And shouldn't you power your converter at 3.3V since esp is 3.3V board

neel-m · June 21, 2026, 4:05pm

Here is what the manufacturer says about it:

There are better ADCs. It depends on what you are doing with the data if you need one or not.

Precision, accuracy, repeatability all matter for some applications and not so much for others.

I use temperature sensors to control some heaters for freeze protection. Absolute accuracy doesn't matter much, since I have the control point well above freezing. Precision and repeatability do matter some, since I want to minimize the energy used.

So, what are you really trying to do that you need a pressure sensor?

Karosm · June 21, 2026, 4:29pm

While not best in class, I find esp32 adc suitable for most of cases. Where it fails is <200mV and >3100mV ranges. C3 instead can't make it above ~2900mV.
Setting samples: 64 kills noise sufficiently. Small capacitor would improve even more.

Vitruvius · June 21, 2026, 4:46pm

I have set exponential_multi_average now, yes. The values you quoted were single readings still.

    filters:                      
      - exponential_moving_average:
          alpha: 0.1
          send_every: 10

And values are much closer to one another:

[16:21:53.465][S][sensor]: 'Water tank level' >> 1.47 V
[16:31:53.482][S][sensor]: 'Water tank level' >> 1.49 V
[16:41:53.488][S][sensor]: 'Water tank level' >> 1.49 V
[16:51:53.489][S][sensor]: 'Water tank level' >> 1.48 V
[17:01:53.485][S][sensor]: 'Water tank level' >> 1.49 V
[17:11:53.504][S][sensor]: 'Water tank level' >> 1.49 V
[17:21:53.512][S][sensor]: 'Water tank level' >> 1.49 V
[17:31:53.512][S][sensor]: 'Water tank level' >> 1.48 V
[17:41:53.513][S][sensor]: 'Water tank level' >> 1.48 V
[17:51:53.515][S][sensor]: 'Water tank level' >> 1.48 V
[18:01:53.532][S][sensor]: 'Water tank level' >> 1.49 V
[18:11:53.533][S][sensor]: 'Water tank level' >> 1.49 V
[18:21:53.531][S][sensor]: 'Water tank level' >> 1.48 V
[18:31:53.544][S][sensor]: 'Water tank level' >> 1.49 V

As for the 5V, the manufacturer states this:

This module employs 0.1% high precision sense resistor and ultra-low noise rail-to-rail zero-drift op amp, high accuracy, no need to calibration, so it is convenient to use. 3.3V~5.5V wide voltage power supply [...].

So 5V should be OK.

I do have free pins without resistor, I'll see if that makes a difference, thanks.

Measuring the level of our rainwater tank. It does not need to be super accurate; monitoring is mostly to make sure we add tap water once it runs too low (you don't want to find out you're out of rain water when trying to flush the toilet). So I will be setting like a 10% treshold, we'll see how that goes.

Thank you for the insights, everyone

neel-m · June 21, 2026, 5:51pm

That seems about right 1.48V +/-10mV

What does that translate to in terms of tank full/empty?

The fuel gauge in my newest car is digital and seems somewhat non-linear. It only has about 10 levels from full to empty. Surprisingly, that doesn't feel as good as the analog gauges I am used to, which have more precision (but probably not accuracy).

Vitruvius · June 21, 2026, 6:56pm

Almost full, I have this sitting in the YAML as well to convert to a percentage, but it's disabled for now since I wanted to monitor the voltage reported.

unit_of_measurement: "%"
[...]
filters:
  - calibrate_linear:
      datapoints:
      - 0.9 -> 0         
      - 1.55 -> 100

OldSurferDude · June 21, 2026, 7:06pm

I write to share my experience with the ESP ADC and the concern about its "noise"

This is a graph of the cistern of my gravity fed irrigation system over two days.
I am using a 240-33 Ohm Float Sensor. and wire it like this

+5
sensor
---------------------> to ESP8266 ADC
resistor (sized to keep the voltage < 1V)
GND

I also have full and empty level sensors (which are "filtered").

At 6PM a zone turns on. Shortly after, as the water flows, the full sensor indicates not full which turns on the pump from the well (and off when full). Similarly, in the early morning a series of zones turn on and off.

You can see a lot of noise. I ask myself, "What would I do differently if that noise were filtered out?" What the noise does tell me is that the ESP8266 is still working

The level sensor is not triggering the pump so it does not cause my pump to cycle.

Karosm · June 21, 2026, 7:33pm

Much better, but i would still use multisampling. In certain cases the first sample is not accurate Try with:
samples: 64
attenuation: 12db

I don't know the circuit of that converter, but since you use 3v3 microcontroller, in my opinion that's the correct supply voltage. Otherwise converter might scale the output up to higher voltage.

Karosm · June 21, 2026, 7:36pm

That's interferences, not ADC noise. You have some noisy device running at 6AM, probably the pump.

Vitruvius · June 21, 2026, 9:48pm

Where do I set that? I don't find it in the ESPHome sensor documentation. I'll set attenuation to 12 dB to see if that helps.

I'll see about moving to 3V and using another GPIO next weekend .

Karosm · June 22, 2026, 1:41pm

Where you set all ADC parameters like attenuation. And it's documented right below attenuation.

Vitruvius · June 22, 2026, 2:24pm

Ah, I see it now. Was looking through the general sensor documentation for ESPHome.

When I set that, do I still keep the exponential moving average filter?

Karosm · June 22, 2026, 2:53pm

You can keep it since reaction time is not usually factor for tank level sensor.

Vitruvius · June 22, 2026, 6:19pm

Thanks. I have moved the sensor to the 3,3V line and to a resistorless pin. Voltage is now reporting around 1,9V, still fluctuating, but good enough for what I need.

I'll play a bit with the samples and exponential_moving_average settings.