Trouble using ahrs with nano ble sense rev2

[nicolasvair]

Hi,

I love this lib but I'm having trouble using it with a nano 33 ble sense rev 2.

I don't manage to have any result better than manually integrating gyroscope. Maybe I'm using it wrong ?

I need something quite reactive so I'm using

ahrs.setFusionAlgorithm(SensorFusion::COMPLEMENTARY);
ahrs.setAlpha(0.98);

I'd be happy with the accelerometer data for the pitch and roll but I'd like the heading to correct itself over time.

But the heading/yaw data is close to random, even changing direction over time (going from left to right increase then decrease the heading value)

Here is my init :
`#include <ReefwingAHRS.h>

ReefwingAHRS ahrs;
SensorData data;
`

Here is my setup :
` ahrs.begin();

ahrs.setFusionAlgorithm(SensorFusion::COMPLEMENTARY);
ahrs.setAlpha(0.95);
ahrs.setDeclination(2.5); // Paris, France

if (IMU.begin() && ahrs.getBoardType() == BoardType::NANO33BLE_SENSE_R2) {
Serial.println("BMI270 & BMM150 IMUs Detected.");
}
else {
Serial.println("BMI270 & BMM150 IMUs Not Detected.");
while(1);
}

Serial.print("Detected Board - ");
Serial.println(ahrs.getBoardTypeString());

`

And my loop :

`
float roll, pitch, yaw, heading;
static int CC_1, CC_2, CC_3;

if (IMU.gyroscopeAvailable()) { IMU.readGyroscope(data.gx, data.gy, data.gz); }
if (IMU.accelerationAvailable()) { IMU.readAcceleration(data.ax, data.ay, data.az); }
if (IMU.magneticFieldAvailable()) { IMU.readMagneticField(data.mx, data.my, data.mz); }

ahrs.setData(data);
ahrs.update();

if ( millis() - lastPrint > 60 ) {

yaw = (ahrs.angles.yaw - 180) / 180 ;
roll = ahrs.angles.roll / 180 ;
pitch = ahrs.angles.pitch / 180;

}

}
`

I'm then sending the data as midi to control a moving light.

Please telle me I'm doing something wrong ?

[nicolasvair]

Maybe worth mentioning I'm using it perpendicular (ground is pointing the same direction but pitch and roll are reversed). So yaw/heading is 90° off.

Is it something that could screw the average ?
Right now it seems to settle after some minutes but if I move it around its completely lost for at 30s to a minute. Maybe it's normal ?

[reefwing]

Hi @nicolasvair - glad you are enjoying the library.

It is tricky to get a correct yaw/heading from a magnetometer. Here are some suggestions:

1. Calibrate the BMM150 magnetometer

• Perform a full hard- and soft-iron calibration. Move the board slowly through all orientations in a figure-8 and tumble pattern, then apply the resulting offsets and scale factors in your sketch.
• Test by logging raw mx, my while the board is level and rotated on a table. The points should form a centered circle. If you see an offset or an ellipse, the yaw will wander until you correct it.
• Keep the sensor away from steel benches, laptop speakers, and wiring during tests.

2. Verify frame alignment and the 90° mounting

You mentioned the unit is perpendicular and yaw is 90° off. That is fine, but the filter must receive data in a consistent body frame:

• Rotate your raw accel, gyro, and mag vectors into the library’s expected frame before calling ahrs.setData(data)
• A quick test is to hold the board level, rotate it in place, and confirm that roll and pitch stay near zero while yaw changes smoothly. If roll or pitch change during a flat rotation, the axes are swapped or signs are flipped.

3. Read a consistent sensor triplet
   Your loop reads whichever sensor is “available” at that moment. That can mix samples from different instants. You may get a better result if you:

1. Wait until you have all three fresh readings.
2. Read gyro, accel, mag back-to-back into data.
3. Call ahrs.setData(data) and ahrs.update() once per cycle.

The downside is that it will be less responsive due to the slower mag sampling rate.

4. Avoid heading wrap artifacts in your MIDI scaling
   You are mapping heading with
   yaw = (ahrs.angles.yaw - 180) / 180;
   If yaw crosses ±180°, this wrap can look like a direction reversal. Either use an unwrapped heading for control, or map to the range you need after first converting to a continuous representation. For example, convert yaw to a unit vector (cos, sin) or use an unwrap routine before scaling.

[nicolasvair]

Hi,

Thank you for your rapid and detailed answer ! I have some more questions.

1 - About the magnetic calibration : you mentioned in a post in this repo that the ble sense rev 2 sensor has automatic sensor calibration. I’m wondering if this could be part of my problem (sensor re-calibrating then screwing the “average” of ahrs algorithm, this is coherent with the fact that data is mostly ok after 2-5mn).
The doc explain how to disable it using registers but then I don’t know how to do a manual calibration.
Is there a function in the lib that works with ble sense rev2 ?

I’ve seen solutions online but I’m afraid to do it in a way that would screw the data. Should I work with the quaternion data or purely sensor data ?

2 - about the frame alignment : I fear to create more problems doing so. In my actual setup I have no pitch / roll change during flat rotation so I suppose it is ok ? Or is there other situations in which the frame should be absolutely aligned ?

3 - this seems like a real issue that I will test asap

4 - yaw = (ahrs.angles.yaw - 180) / 180;
Do you confirm that this doesn’t affect the ahrs algorithm ? I use this to scale values on 0-127 for midi but then it’s mapped to 0-255 and wrapped so it shouldn’t be sensitive to changes in direction. But maybe I’m getting it wrong ?

Thank you for your help !! 🙏

[nicolasvair]

In your sketchs for other boards, you mention that it’s important to calibrate before running ahrs.

Could it be that because ble sense rev 2 sensor “auto calibrate” over time, I should “reset” the ahrs algorithm to account for it ?

Instead of letting the ahrs slowly adapt for a big change in magneto direction.