Gyroscope Calibration
Digital gyroscopes are just like those spinner toys you've seen when you're a kid, where you pull the string and you can balance the whole thing on your fingertip easily. The ones in chips don't have metal rings, instead they use very very small springs that help it measure rotation.
Compared to accelerometers and even magnetometers, digital gyros are surprisingly complex, but thanks to their existence we can match them up with other sensors to create orientation sensors. Magnetometers are too slow and are easily messed up by magnets and accelerometers cannot measure spin along the gravitational axis (they can only measure tilt) so give a hand to the gyroscope!
When gyros are manufactured, they have some zero-offset error, much like magnetometer hard-offset error, that can make measurements difficult. It's easy to detect and remove this offset, we just take many measurements and look for the 'offset' from zero
For example, here's a common MPU-6050. If we put it down flat on a table, and take measurements, we will see that neither X, Y or Z (red, green, blue) are at zero. Even though its not moving! That's the zero offset. In this case, its about -0.43 for X, 0.32 for Y and 0.34 for Z.
Once calibrated, you can see that there's still a little noise (there always is!) but its only 10% as much as the offset, so we'll get much better measurements. We can try to improve the noisy measurements with filtering if necessary.
Page last edited March 08, 2024
Text editor powered by tinymce.
Install SensorLab
Since there's dozens of different sensor manufacturers out there, and we don't want to have a ton of #ifdef's in our code to manage each kind, we'll be using Adafruit SensorLab to manage detecting the various magnetometers, accelerometers, pressure sensors... etc!
Adafruit SensorLab automatically detects a wide range of sensors, over I2C, no matter what I2C address it's on. It will return an Adafruit Unified Sensor object that we can query for events. You can't do advanced stuff like manually setting ranges or internal filters, but for many projects the basics will do just fine!
We'll be assuming you have the sensor on the main I2C port, and of course use the matching Adafruit library to verify the sensor is working and powered right before you continue!
Install SensorLab
Since there are a ton of sensors, and we also use Arcada in a few examples, there's a lot of libraries to install
No really, we have a lot of software involved here - probably 20 or so libraries total!
For that reason we really strongly recommend you use Arduino 1.8.10 or greater which handles automatic library dependency installation. Otherwise you will be frustrated...
Select the Sketch -> Include Library -> Manage Libraries...
Search for Sensor Lab and install the Adafruit library you see
Page last edited March 08, 2024
Text editor powered by tinymce.
Simple Gyro Calibration
If you don't want to set up a graphical interface for calibration a gyroscopic sensor, you can do a simple zero-g offset calibration using just the serial interface. The nice thing about this is it will work for any and all boards, and does not require any additional software installation!
Step 1 - Upload the SensorLab zero rate simplecal Example
We have a simple sketch that will read a few seconds of gyroscope data and calculate zero-rate offset for you
Open up the Adafruit_SensorLab->calibration->gyro_zerorate_simplecal
Open the serial port to launch the SensorLab calibration. You should see your gyroscope detected
You'll see a countdown to warn you that the gyro must be kept perfectly still during calibration
Place the board on a stable flat surface and hold it down with a book or something. The board will take 5 seconds of data and print at each reading
The first three numbers are the current gyro readings
The middle three numbers are the average values (zero rate offsets) in radians/second.
The last numbers are the peak-to-peak values.
At the end, you'll get a summary:
In this case shown above, the screenshot indicates x = 0.0003, y = 0.01, z = 0.0004 rad/s
You may want to take a few measurements to get a couple calculations!
Page last edited March 08, 2024
Text editor powered by tinymce.
Gyro Calibration with Jupyter
Jupyter Notebooks are a powerful cross-platform method for analyzing data using Python
You can definitely use Jupyter to plot, analyze and calibrate your sensor data. This method is the most powerful because you can do plotting and calculations. However, we assume you already have Jupyter installed (either desktop or thru Anaconda) and have some familiarity with running 'notebook' style Python!
Step 1 - Download Calibration Notebook
The gyro/magnetometer notebook lives in the SensorLab Arduino library, in the notebooks folder.
Look for this section in the website:
You must open this notebook within Jupyter - you cannot run it direct from github or from the command line as a text file!
Once open, your browser will look like this:
Step 2 - Upload the SensorLab imucal Example
Next we have to tell the microcontroller board to send the magnetometer (and, if there is one, accelerometer and gyroscope) data out over serial in the right format.
Open up the Adafruit_SensorLab->calibration->imucal
Open up the serial console, you'll see SensorLab initialization and detection of whatever magnetometer is available. In this case is a LSM9DS1, but any gyroscope can be calibrated!
You'll then see a stream of data that looks like:Raw:-58,-815,8362,76,-121,-95,-375,-159,-24Uni:-0.07,-0.98,10.00,0.0832,-0.1327,-0.1046,-37.50,-15.93,-2.50
The first three numbers are accelerometer data - if you don't have an accelerometer, they will be 0
The middle three numbers are gyroscope data, they should definitely not be zeros!
The last three numbers are magnetometer - if you don't have an magnetometer, they will be 0
Close the serial port, and go back to Jupyter. In the first cell, find where we define the PORT and change the port to match your serial/COM port. For windows it'll be something like COM4 for Mac/Linux it'll be like /dev/cu.USBSERIAL or something
Run the first cell so the serial port is set
Then run the second cell, you should see output like this - the serial port is opened and IMU raw data is output as numbers
If you get errors or no numbers, hard-reset the board (click the reset button once) then try re-running the cell again.
Skip the magnetometer cells, and find the Gyroscope offset calibration cell
Place the board down on a flat stable surface, we like to put a book on top of it to keep it flat and steady. You don't want any shifts or movement while you run the calibrator!
You'll see a 3 second countdown, then the serial port is opened and data collected
Once it's done you'll see a plotting of the data output. You should see something like this with red, green and blue lines. There may be some offset as expected, but the lines should wobble around a point without big spikes or increases/decreases.
Above the graph you'll see analysis of the average value (zero-g offset) like so:
In this case, the zero-g calibration is X = 0.044 Y = -0.0148, and Z = -0.0111
The units are in SI units radians / second not degree/s (dps)!
You'll also see the results of removing the offset, this should be 3 noisy but nearly-zero-centered lines. The higher the quality of the gyro, the lower the jitter in the measurements, and the lower the offset.
Page last edited March 08, 2024
Text editor powered by tinymce.
Comparing Gyroscopes
When making an inertial measurement unit (IMU) especially one used for orientation calculations, the quality of the gyroscope has a big impact on the jitter/speed/drift. Accelerometers are pretty good quality these days, we've figured out how to make them decades ago. And magnetometers are not as important - they don't update often and are used for orientation correction. Gyro's have the most error possibilities.
We compare two basic measurements - the zero offset and the zero noise. Zero offset is easy to correct for, and you should calibrate your gyroscope once its mounted in the final project/PCB - simply take a lot of measurements, find the offset that would bring the gyro to zero. The noise that you get when not moving can be minimized with low pass filtering (sometimes handled in-chip).
Uncalibrated offsets are low, the one we grabbed had 0.006 rad/s (0.35 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±1 deg/s.
No-motion observed noise was an incredibly low ±0.002 rad/s (±0.06 deg/sec) when running at 104 Hz and no filters on. Check the datasheet for more details!
LSM6DSOX
This is an high quality 6-DoF IMU with great performance - at an expected higher price than the LSM6DS33. This is the 'commercial usage' version of the ISM330DHC
Uncalibrated offsets are low, the one we grabbed had 0.007 rad/s (0.42 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±1 deg/s.
No-motion observed noise was an incredibly low ±0.003 rad/s (±0.17 deg/sec) when running at 104 Hz and no filters on. Check the datasheet for more details!
LSM6DS33
This basic 6-DoF IMU is inexpensive and works well for basic projects. We have it both in 6-DoF and paried with a magnetometer for 9-DoF,
Uncalibrated offsets are fair, the one we grabbed had 0.034 rad/s (2 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±10 deg/s!
No-motion observed noise was ±0.015 rad/s (±0.85 deg/sec) when running at 104 Hz and no filters on. Check the datasheet for more details!
The ST LSM6DS3TR-C is a great entry-level 6-DoF IMU accelerometer + gyro and is very similar to the now-discontinued LSM6DS33. We have it both in 6-DoF and paired with a magnetometer for 9-DoF
The one used here had 0.03 rad/s (1.7 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±3 deg/s.
No-motion observed noise was ±0.0015 rad/s (±0.085 deg/sec) when running at 104 Hz and no filters on. Check the datasheet for more details!
LSM9DS1
This popular sensor is a 9-DoF all-in-one with a LIS3MDL for the magnetometer, and a 'LSM6DS-like-but-not-the-same-register-map-at-all' accelerometer/gyro IMU
Uncalibrated offsets are pretty good, the one we grabbed had 0.02 rad/s (1.2 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±30 deg/s!
However, we noticed spikes of gyro data well outside the expected range. When those spikes are ignore, the no-motion observed noise was +- 0.007 rad/s (±0.4 deg/sec) at 1 KHz with the 408 Hz bandwidth filter on. Check the datasheet for more details!
MPU-6050
This is a fairly old sensor, but very popular and low cost! The quality of the sensor is pretty good considering how old/low cost it is
Uncalibrated offsets are fair, the one we grabbed had 0.04 rad/s (2.3 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±20 deg/s!
No-motion observed noise was +- 0.05 rad/s (±0.29 deg/sec) with the 260 Hz bandwidth filter on. Check the datasheet for more details!
NXP FXAS21002
This 3-axis gyroscope sensor is no longer available but we use it in our NXP 9-DoF (and will continue selling it until we can't get the chip anymore) At the time of release it was pretty good, since then better gyros have come out, but it's still not too bad
Uncalibrated offsets are pretty good - the one we grabbed had 0.01 rad/s (0.57 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical post-mount ±50 LSB (NOT deg/s) - at 250 deg/s rate, that translates to ±0.4 deg/s
No-motion observed noise was +- 0.01 rad/s (±0.55 deg/sec) at 100 Hz output. Check the datasheet for more details!
Uncalibrated offsets are not bad, the one we grabbed had 0.023 rad/s (1.3 deg/s) max offset. Datasheet's Angular rate zero-rate level is typical ±5 deg/s
No-motion observed noise was +- 0.015 rad/s (±0.86 deg/sec) at 1.1KHz. Check the datasheet for more details!
Page last edited March 08, 2024
Text editor powered by tinymce.
