If 6-axis sensor is enabled, the IMU data in an 0x30, 0x31, 0x32 and 0x33 input report is packaged like this (assuming the packet ID is located at byte 0):
| Byte | Remarks |
|---|---|
| 13-14 | accel_x (Int16LE) |
| 15-16 | accel_y (Int16LE) |
| 17-18 | accel_z (Int16LE) |
| 19-20 | gyro_1 (Int16LE) |
| 21-22 | gyro_2 (Int16LE) |
| 23-24 | gyro_3 (Int16LE) |
| 25-48 | The data is repeated 2 more times. Each with 5ms Δt sampling. |
The 6-Axis data is repeated 3 times. On Joy-con with a 15ms packet push, this is translated to 5ms difference sampling. E.g. 1st sample 0ms, 2nd 5ms, 3rd 10ms. Using all 3 samples let you have a 5ms precision instead of 15ms.
The accelerator axes are defined by the output in packet. Because of the placement of the IMU chip, the 2 Joy-Con have an axis reversed each.
The following images show exactly the 6 axes:
The following equation should scale an int16 IMU value into an acceleration vector component (measured in Gs):
acc_vector_component = acc_raw_component * G_RANGE / SENSOR_RES / 1000 [Value in G]
where G_RANGE is the sensitivity range setting of the accelerometer, as explained here.
The Joy-Con are ranged to ±8000 MilliGs (G_RANGE = 16000 MilliGs), the sensitivity calibration is always ±8192 MilliGs and the SENSOR_RES is 16bit, so the above equation can be simplified to:
acc_vector_component = acc_raw_component * 0.000244f. (=16000/65535/1000)
For Gyro the equation to convert the values into angular velocity (measured in degrees per second):
gyro_vector_component = gyro_raw_component * G_GAIN / SENSOR_RES [Value in dps]
where G_GAIN is the degrees per second sensitivity range.
The Joy-Con, based on their configuration, have a gyro sensitivity of ±2000dps (G_GAIN = 4000dps), so the above equation can be simplified to:
gyro_vector_component = gyro_raw_component * 0.06103f (=4000/65535)
To express the full-scale rates in both directions without saturating you can add a 15% (STMicroelectronics LSM6DS3 datasheet) and it's the recommended way to do it.
The new conversion will be:
gyro_vector_component = gyro_raw_component * 0.070f (=4588/65535)
Internally Switch uses revolutions per second instead of degrees. That's why the gyro calibration is always 13371.
Normally to get revolutions/s you need to follow the below equation:
gyro_revolutions = gyro_raw_component * G_GAIN / SENSOR_RES / 360.0f [Value in revolutions/s]
So the equation for the above 3 sensitivities will become:
gyro_revolutions = gyro_raw_component * 0.0001694f (=4000/65535/360)
gyro_revolutions = gyro_raw_component * 0.0001944f (=4588/65535/360)
The SparkFun library for the LSM6DS3 will likely be a valuable resource for future IMU hacking.
The SPI accelerometer calibration, includes 3 important values for each axis:
The cal_acc_origin is the origin scale value when the Joy-Con is held completely horizontally. It's not an origin position but how the raw values are affected by noise and by the unleveled body of the controller (triggers protruding).
The cal_acc_horizontal_offset is the offest (difference) that the Joy-Con/Pro-con has when it's on a flat surface than being completely horizontal. (The Trigger bumps change its position and you can use this offset to calibrate the position when it is on a flat surface)
The cal_acc_horizontal_offset is always the same. Advise here for the values each model (JC Left, JC Right, Pro).
The cal_acc_coeff is used for calibrating sensitivity. The default uncalibrated value for ±8G sensitivity is x4000 (16384). Advise here.
Based on these we can conclude on the following equation to find the final coefficient:
acc_coeff = (float)(1.0 / (float)(cal_acc_coeff - uint16_to_int16(cal_acc_origin))) * 4.0f;
Then we use the coefficient to convert the value into G (SI: 9.8m/s²):
acc_vector_component = acc_raw_component * acc_coeff
*This will level the accelerometer to 0 values when resting on flat surface. It's useful when we perform a manual calibration.
We subtract cal_acc_horizontal_offset and then we use the coefficient to convert the value into G (SI: 9.8m/s²):
acc_vector_component = (acc_raw_component - uint16_to_int16(cal_acc_horizontal_offset)) * acc_coeff
The SPI gyro calibration, includes 2 important values for each axis:
The cal_gyro_offset is the offset when the Joy-Con is stable (held still).
The cal_gyro_coeff is used for calibrating sensitivity. The default uncalibrated value for ±2000dps sensitivity is x343B (13371). Advise here.
Based on these we can conclude on the final equation:
gyro_cal_coeff = (float)(936.0 / (float)(cal_gyro_coeff - uint16_to_int16(cal_gyro_offset)));
gyro_cal_coeff = (float)(816.0 / (float)(cal_gyro_coeff - uint16_to_int16(cal_gyro_offset)));
Then we use the coefficient to convert the value into degrees°/s (SI: 0.01745 rad/s):
gyro_vector_component = (gyro_raw_component - uint16_to_int16(cal_gyro_offset)) * gyro_cal_coeff
Here, unlike acceleration, the origin position is indeed the still position and not a scaling value. So it must be also used in the unit calculation, in addition to coeffition.
The equation will become:
gyro_vector_component = (gyro_raw_component - uint16_to_int16(cal_gyro_offset)) * gyro_cal_coeff * 0.0027777778
This is useful to set a noise cancellation deadzone for acc and gyro.
Based on official values. Also it's a range value (not a ± value).
Accelerometer:
| Sensitivity | Noise level range |
|---|---|
| ±2G | 328 * 2.5 = 2050 |
| ±4G | 164 * 2.5 = 410 |
| ±8G (default) | 82 * 2.5 = 205 |
| ±16G | No official value |
Gyroscope:
| Sensitivity | Noise level range |
|---|---|
| ±250dps | 236 * 2.5 = 590 |
| ±500dps | 118 * 2.5 = 295 |
| ±1000dps | 59 * 2.5 = 147 |
| ±2000dps (default) | 30 * 2.5 = 75 |