Problems with accuracy of LSM6DSL pedometer functionality

[KReng.1]

Hi all,

I am using LSM6DSL chip as a pedometer and sensing other stuff like tap detection, tilt detection etc on a wearable device. The pedometer counts are pretty close to accurate when I compare it with other standard devices like the apple watch/ fit bits. But the issue is it is also counting the hand movements as steps for eg: when I am not walking and just moving my the sensor or just picking up something etc. I was wondering if there is a filter of some sort to remove false step counts?

Thanks,

Kesav

[TBomb.1 · Accepted solution]

Hi, a pedometer output can sometime be biased by movements that are similar to the step pattern. The accuracy of the device also depend upon the position (wrist or waist, for example). You could vary the main parameters of a pedometer, such as the debouncing in register ****_DEB_REG (14h), the threshold in register CONFIG_****_THS_MIN (0Fh), and the delta STEP_COUNT_DELTA (15h) to tune it for your application and to make it less sensible to other-than-steps movements.

Debouncing parameter

threshold parameter

https://www.st.com/resource/en/application_note/dm00402563-lsm6dsl-alwayson-3d-accelerometer-and-3d-gyroscope-stmicroelectronics.pdf

Tom

[KReng.1]

Thanks for directing me to the right place, Tom.

I increased the threshold all the way to the maximum value (0xFF), the accuracy was reduced by a small factor (8% less steps - I don't mind that) when actually walking. But it didn't make much difference with respect to false steps though. Changing the pedometer de-bounce did not making any difference either.

I am using the sensor on the wrist and ankle to count the steps. I wonder if there is any pre-set values for these register for using it wrist or ankle or hip etc.

Also, I am trying to leverage the significant feature detection functionality to turn on the pedometer. But I am not sure what they consider it as "significant movement".

Anyways, thanks Tom.

Regards,

Kesav.