Using a Median Filter to Handle Outliers in Load Cell Readings

Overview

This is a story about how I struggled with an unusually large value appearing for just a single frame while trying to capture the peak pressure using a load cell and HX711 with Arduino.

Original Implementation (Moving Average Filter)

I was processing the values from the load cell using a low-pass filter with a simple moving average without thinking much about it.

Revised Implementation (Median Filter)

I referred to the following site:
https://ehbtj.com/electronics/filter-library-useful-for-arduino-signal-processing/

Moving averages seem good for removing fine fluctuations (high frequency), but they don't seem suitable for rejecting outliers.

Based on the following description, I decided to try using a median filter:

The Hampel filter does not smooth the data much.
It feels like it only removes spikes.
There is a similar filter called the median filter, but this one smooths the data slightly more.

I found a library, so I used it:
https://github.com/luisllamasbinaburo/Arduino-MedianFilter/tree/master

It is written in French, but

you can load the library with #include "MedianFilterLib.h",
set the number of samples with MedianFilter<int> medianFilter(5);,
and the result is returned by calling int median = medianFilter.AddValue(rawMeasure);.

It seems to work the same way for decimal values as well.

When I actually used it, even when an outlier occurred in the Raw data, it didn't seem to affect the Med value.