PaulZC
commented
2 years ago Hi @camelator ,
Walking through this:
getAGMT requests and returns the latest accel, gyro, mag and temperature readings from the 20948. It calls ICM_20948_get_agmt to do that. ICM_20948_get_agmt reads 23 bytes from the 20948 register bank 0, starting with ACCEL_XOUT_H. So agmt ends up containing a faithful copy of the contents of the 20948 registers.
There is a note in the datasheet for ACCEL_XOUT_L which says "To convert the output of the accelerometer to acceleration measurement use the formula below: X_acceleration = ACCEL_XOUT/Accel_Sensitivity".
The Accel_Sensitivity is defined by ACCEL_FS (Accel Full Scale) which is set in the ACCEL_CONFIG register:
So the important things to note are that:
- ACCEL_FS sets the full scale in g (+/- 2g to +/- 16g)
- In 2g mode the sensitivity scale factor is 16384 Least Significant Bits / g
- In 16g mode the sensitivity is 2048 LSB/g
So, in 2g mode, we need to divide the accel reading by 16384 to get to g. But to get to milli-g we need to divide by 16.384.
accX calls getAccMG to convert the accel value into milli-g:
When the accel is in 2g mode, agmt.fss.a is 0. To convert to milli-g we need to divide by 16.384 and, looking at the code, we do:
To my eyes, the code looks correct. Are you getting g and milli-g mixed up?
In the Adafruit code, I'm guessing they have previously converted the accel reading into m/s^2 and then need to divide by 9.8 to convert back into g? I don't know. I've never delved into their code...
Please think more about this and let me know if you still think it is an issue.
Best wishes, Paul
camelator
I don't know how to submit a change, but maybe the following function is not correct. I think it should be 16384 instead of 16.384, 8192 instead of 8.192, etc...
float ICM_20948::getAccMG(int16_t axis_val) { switch (agmt.fss.a) { case 0: return (((float)axis_val) / 16.384); break; case 1: return (((float)axis_val) / 8.192); break; case 2: return (((float)axis_val) / 4.096); break; case 3: return (((float)axis_val) / 2.048); break; default: return 0; break; } }
then I not sure where it comes from but there is a gap between the mindset of this code and what we found in the imucal.ino code: Serial.print("Raw:"); Serial.print(int(accel_event.acceleration.x8192/9.8)); Serial.print(","); Serial.print(int(accel_event.acceleration.y8192/9.8)); Serial.print(","); Serial.print(int(accel_event.acceleration.z8192/9.8)); Serial.print(","); Serial.print(int(gyro_event.gyro.xAdafruit_SensorLab::DEGREES_PER_RADIAN16)); Serial.print(","); Serial.print(int(gyro_event.gyro.yAdafruit_SensorLab::DEGREES_PER_RADIAN16)); Serial.print(","); Serial.print(int(gyro_event.gyro.zAdafruit_SensorLab::DEGREES_PER_RADIAN16)); Serial.print(","); Serial.print(int(mag_event.magnetic.x10)); Serial.print(","); Serial.print(int(mag_event.magnetic.y10)); Serial.print(","); Serial.print(int(mag_event.magnetic.z10)); Serial.println("");
Please note in this code to convert imu to raw we have to multiply by scale and divide by gravity
so maybe to convert raw to imu we have to divide by scale and multiply by gravity instead of simply divide by scale ?????