Hi,
I'm doing Allan variance analysis on OAK's IMU. I recorded 10 hours of rosbag for each OAK camera. Now I read ACCELEROMETER_RAW and GYROSCOPE_RAW at 100Hz. For BMI270, these seem to be the only two available. But since BMI270 has functions such as programmable filter and CRT, I'd like to confirm whether ACCELEROMETER_RAW and GYROSCOPE_RAW are real RAW data. And you have defined other items in depthhai.IMUSensor. I'd like to understand their purpose more clearly. Are these post-processing intended to be implemented by the IC of IMU or OAK's firmware?
Here is the data I measured from one device.
From the graph it looks like the filter is not enabled, or at least the filter has a high bandwidth. This is what we want! We don't want any latency brought by the built-in filter, our own algorithm will handle the noise propagation. But the random walk in the graph of gyroscope looks a bit weird. Not sure if this is related to the CRT.
#Accelerometer
accelerometer_noise_density: 0.0014050369537968298 # m/s^2/√Hz
accelerometer_random_walk: 7.268205240359128e-05 # m/s^3/√Hz
#Gyroscope
gyroscope_noise_density: 0.0001116082909576279 # rad/s/√Hz
gyroscope_random_walk: 1.0601803124823304e-05 # rad/s^2/√Hz
The noise density from datasheet is 160 ug/√Hz and 0.007 deg/s/√Hz (BW=100Hz), and is close to our measurement. Bias instability is not available from datasheet.
Hi,
I'm doing Allan variance analysis on OAK's IMU. I recorded 10 hours of rosbag for each OAK camera. Now I read ACCELEROMETER_RAW and GYROSCOPE_RAW at 100Hz. For BMI270, these seem to be the only two available. But since BMI270 has functions such as programmable filter and CRT, I'd like to confirm whether ACCELEROMETER_RAW and GYROSCOPE_RAW are real RAW data. And you have defined other items in depthhai.IMUSensor. I'd like to understand their purpose more clearly. Are these post-processing intended to be implemented by the IC of IMU or OAK's firmware?
Here is the data I measured from one device.
From the graph it looks like the filter is not enabled, or at least the filter has a high bandwidth. This is what we want! We don't want any latency brought by the built-in filter, our own algorithm will handle the noise propagation. But the random walk in the graph of gyroscope looks a bit weird. Not sure if this is related to the CRT.
The noise density from datasheet is 160 ug/√Hz and 0.007 deg/s/√Hz (BW=100Hz), and is close to our measurement. Bias instability is not available from datasheet.