Adafruit Feather CO2 Meter

✨📝 Read all about this project in my blog post! 📝✨

A simple, small, handheld carbon dioxide meter!

It's also battery powered!

2021 onwards saw a surge of people using CO2 meters to help gauge indoor air quality. The Aranet4 seemed popular but I also saw people making their own and thought "that's a fun project, I'll do it too!"

It totally didn't take me 8 months after buying parts to start the project...

How to use this repository

Tutorials

The Tutorials folder contains lift and shift code to do the exact same steps I did. They include the correct binaries too so it just works straight away. There's also a readme there to go over what each of the folders contain.

Running the code

Assuming you have all the components below, the src folder has all you need to lift and shift to get a working project.

1. The src\code.py file is the code itself
2. The src\lib folder has all the extra imports

Copy both of these to your root CIRCUITPY drive and you're good to go.

The version of the code in the src folder is the most up to date and includes any of the newest features listed in this readme.

Components

Item	Cost (USD)
Feather RP2040	$11.95
Adafruit 2.9" Grayscale eInk	$22.50
Lithium Ion Polymer Battery 400mAh	$6.95
SCD-40 - True CO2, Temperature and Humidity Sensor	$49.50
STEMMA QT / Qwiic JST SH 4-pin Cable	$0.95
Brass M2.5 Standoffs 16mm tall	$1.25
Total	$93.10

The only soldering needed is to attach the given headers onto the RP2040.

Pictures

Position	Image
Front
Top
Back

That single spacer to hold the sensor to the RP2040 is used as a makeshift kickstand for the back. It works pretty well!

And for a preview of other CO2 ratings with expert photography:

Fun power efficiency improvements

After finishing version 1.0, it was time to begin working on improvements for the power consumption. Having worked a little bit with Raspberry Pi Zero 2 W boards and looking at power usage, I figured this microcontroller without a proper OS must get a lot out of the 400mAh battery... But it didn't.

The 400mAh battery lasts (very) approximately 12 hours with the first stable release and I think it can do far better. But first, let's understand the power usage patterns. I'll be using a Multifunctional USB Digital Tester - USB A and C to get the readings.

Scenario	Description	Reading	Notes
Baseline 1	Blinking LED, nothing attached	0.21W-0.23W	Our absolute baseline
Baseline 2	Blinking LED, display attached (nothing displaying)	0.21W-0.23W
Baseline 3	Blinking LED, display attached (nothing displaying), SCD-40 (nothing reading)	0.21W-0.25W
Baseline 4	Stable release code	0.23W-0.74W	Big spikes (more on that below), otherwise about the same as previous scenarios

Now that we have baseline measurements, it's time to optimise!

Power saving with a deep sleep

Let's deal with the easy stuff: Busy waits. I love optimising things (self plug: Shrinking a Self-Contained .NET 6 Wordle-Clone Executable) and there's a whole heap of ways for programs to sleep across different languages. I suspect that in CircuitPython that the time.sleep() call might be a bit busy in the background. Doing some digging, it turns out we can use alarms instead of sleeps. I picked up on this fantastic tutorial called Deep Sleep with CircuitPython which explained the different types of sleep in CircuitPython.

I ran some tests (which are in the Tutorials/DeepSleep folder) for deep sleeps. The test case was a blink based on the tutorial and here are the results:

Scenario	Description	Reading	Notes
Sleep 1	time.sleep()	0.21W-0.23W	This one is in the Blink folder
Sleep 2	alarm.light_sleep_until_alarms()	0.17W-0.25W	But more often around 0.17W-0.23W
Sleep 3	alarm.exit_and_deep_sleep_until_alarms()	0.11W-0.23W	Both: I did see it hit 0W a couple of times The RGB NeoPixel also fires due to it booting up

Note: Test when connected to a power supply, and not PC as the board will not actively sleep when connected to a host computer.

The deep sleep looks like what we want. So let's apply it to the stable release:

Scenario	Description	Reading	Notes
Efficiency 1	Stable release code with improved power efficiency	0.13W-0.64W	A good improvement but with the same big spikes (see further down)

A ~0.1W drop and the spikes remain.

As at version 1.1 the battery lasts about 21 hours, or 1.75 time longer than version 1.0.

Power saving by turning off the sensor

Spike time. Every 3-5 seconds as it seems the SCD-40 sensor does a reading regardless of whether the values will be read or not. Below is a quick look at the meter, note the red wattage reading on the lower left and how it spikes:

Note: The display on the reader presents averages between updates. It may not show the proper spike on each display update due to this.

Notice that these spikes did not happen in scenario Baseline 3 even with the sensor connected - it only started happening in Baseline 4 when the sensor has been activated. We can prove this by using the same blink code as Baseline 1, but added in the start measurement code for the sensor from Baseline 4:

import time
import alarm
import board
import digitalio
import adafruit_scd4x

led = digitalio.DigitalInOut(board.LED)
led.direction = digitalio.Direction.OUTPUT

i2c = board.I2C()
scd4x = adafruit_scd4x.SCD4X(i2c)
scd4x.start_periodic_measurement()

while True:
    led.value = True
    time.sleep(1)
    led.value = False
    time.sleep(3)

Scenario	Description	Reading	Notes
Baseline 5	Blinking LED, no display attached, SCD-40 activated	0.23W-0.71W

Nailed it. Confirmed that the sensor measurements need to be kicked off before we see the power usage spikes.

In theory if there is a start then there should be a stop. And there is! stop_periodic_measurement() is the exact call we're looking for. So the code will now:

1. Only start a measurement just before needing the value
2. Read and store the result
3. Immediately stop the measurement

Let's see what that looks like:

Scenario	Description	Reading	Notes
Efficiency 2	Efficiency 1 + power spike removal	0.11W-0.15W	Then with the spike to 0.71W at the 5 minute mark to do a single read

After these improvements, the battery now lasts 54 hours, or 4.5 times longer than version 1.0.

Future ideas

Maybe these would be neat to implement, maybe they won't 🔮

1. Use the eInk display buttons to switch to a graph mode
2. Add symbols to help define what each reading is
3. Properly center all the elements
4. Make use of the four greyscale colours
5. Take readings every x amount of time between display refreshes to get an average
6. Have a button push to refresh asap (as soon as it's been 3 mins since the last display refresh)
7. Battery indicator
8. A way to tell whether the battery has run out on the display (currently the green LED on the sensor is the only indicator)