LSM6DSV: Z acceleration values too high by 1g

PeterK · ‎2025-03-18

Hi,
I read the LSM6DSV on the STEVAL SMKI239AA using a Raspberry Pi Pico running Circuit Python. The CTRL1 register is set to 0x07 (240 Hz ODR, high performance). My output is converted to m/s^2 and looks like that when I move the sensor board slowly:

The fourth column is the vector sum of the three axes. As can be seen, X and Y look credible; Z, however, runs up to twice the gravitational acceleration all by itself. When the sensor is positioned horizontally, Z is about zero (but should be -1g / -9.81 m/s^2). When the sensor is inverted (upside down), Z grows to about 2g / 19.6 m/s^2 (but should be +1g / 9.81 m/s^2). I would like Z to behave like the two other axes. I can fake it by subtracting the 9.81 m/s^2, but I prefer to get correct values in the first place.

What setting did I screw up? I only set CTRL1 and leave all others with their default setting.

Regards,

Peter

Federica Bossi · ‎2025-03-21

Hi @PeterK ,

That's very strange.

Are you reading the outputs correctly? I mean are you implementing the 2's complement correctly? Are you also using the STEVAL-MKI109V3?

In order to give better visibility on the answered topics, please click on 'Accept as Solution' on the reply which solved your issue or answered your question.

PeterK · ‎2025-03-21

Hi Federica,
thank you for answering! Yes, I agree, that is strange. Since I treat all three channels equally, and only Z is looking wrong, I doubt that have a mistake when reading the values. And no, I do not use the MKI109V3. I connect the pins for GND, VDD, VDDIO, SCL and SDA directly to the corresponding Pico pins and have no problem to initialize the I2C connection, inquire 0x0F (WhoAmI) or with any other I2C-related stuff. All other pins are left floating. Since I also read the rotations in the same way, and they look fine to me, I very much doubt that my implementation is wrong.
Here is the code (in Python, not C!):

dataGyr = read(i2c, devDSV, 0x22, 6) # 6 Bytes, starting with Reg. 22
dataAcc = read(i2c, devDSV, 0x28, 6) # 6 Bytes, starting with Reg. 28

gyr_x = struct.unpack_from("<h", dataGyr, 0)[0] - xrotSum
gyr_y = struct.unpack_from("<h", dataGyr, 2)[0] - yrotSum
gyr_z = struct.unpack_from("<h", dataGyr, 4)[0] - zrotSum

acc_x = struct.unpack_from("<h", dataAcc, 0)[0]
acc_y = struct.unpack_from("<h", dataAcc, 2)[0]
acc_z = struct.unpack_from("<h", dataAcc, 4)[0]

As you can see, I first collect the rotation values at rest and subtract them from the measurements for gyro bias calibration. The acceleration numbers are the raw sensor values.

PeterK · ‎2025-03-24

Hi Federica,

I should add that I now also set the offset registers (X_OFS_USR, Y_OFS_USR, Z_OFS_USR) actively to zero before reading values and still see the offset in the Z channel, but not the X and Y channel. Also the filtering is reduced as much as possible:

write(i2c, devDSV, 0x18, 0x00)
write(i2c, devDSV, 0x73, 0x00)
write(i2c, devDSV, 0x74, 0x00)
write(i2c, devDSV, 0x75, 0x00)

The syntax of the command should be obvious: First parameter is the I2C instance, second the instance of the I2C device with the bus address 107, the third parameter is the register to be written to and the last parameter is the value.

I also should add that the offset is creeping down with time. Here are the first 36 measurements with the sensor in an upright position (where Z should read -9.8):