I am using the LSM6DSL for a high frequency vibration measurement involving the accelerometer sensor only (gyroscope is turned off). Output data rates are variable between 400-3.3kHz. A recent frequency sweep experiment (0-6kHz applied on external speaker) with the 3.3kHz ODR in high performance mode, analog anti-aliasing filter bandwidth set at 1.5kHz and digital filter at ODR/2 as per the datasheet. In the resultant spectrogram, the frequency sweep signal aliases after crossing ODR/2, ODR, and 1.5*ODR multiple times over, with no obvious roll-off as would be expected with analog filter cutoff at 1.5kHz and digital filter at 1.66kHz. Forcing the analog filter to 400Hz and digital filter to ODR/4 does not seem to change anything.
These anti-aliasing filters at minimum ODR/2 should be active at all times, but this seems not to be the case. Is it possible that switching off the gyroscope might be causing issues with the enabling of these necessary filters? Similar aliasing has been noted in experiments with human voice harmonics.
1 ACCEPTED SOLUTION
Accepted Solutions
Hi @HAAA1 ,
Happy New Year to you too!
The LSM6DSL is actually quite an "old" product, and current state-of-the art devices such as the LSM6DSO(X) embeds an improved analog anti-alias filter in the acquisition chain (see p.28 of LSM6DSO datasheet, not so explicit it has been improved with respect to the elder one, actually).
-Eleon
2 - data_constant_sweeps.csv corresponds to a different device and firmware which uses the LSM6DSL, sampling at 1.6kHz. The device was placed on a speaker and 4 linear frequency sweeps were performed. First a sine wave 0-1.5kHz. Then a triangle wave 0-1.5kHz, then a square wave 0-1.5kHz. A fourth sine sweep was performed 0-3kHz. this is followed by a period of various constant tones either satisfying nyquist criterion or not. Spectrogram here:
3 - data_1-001.csv corresponds to a data set taken on a pediatric subject. You can see in the attached spectrogram that the harmonics of the voice are reflecting off of the nyquist frequency and aliasing down to interfere with the fundamental frequency of speech.
We have noticed these concerning aliasing signals across several device configurations and separately developed firmware. Hopefully there is a simple reason for this, but we have not been able to sort it out based on documentation alone.
Hello Eleon,
Do you have any updates for this? We have since tried simultaneously sampling two LSM6DSL accelerometers at 6.6kHz and inputting a 10-15kHz sweep. As far as the documentation shows, the highest analog anti-aliasing filter possible should be attenuating frequencies >1.5kHz, however there is clear signal aliasing all the way up to 15kHz as if there is no anti-aliasing filter at all.
File attached for upper spectrogram. We would appreciate a response soon as this sensor is deployed in several studies.
Hi @HAAA1 ,
Happy New Year to you too!
The LSM6DSL is actually quite an "old" product, and current state-of-the art devices such as the LSM6DSO(X) embeds an improved analog anti-alias filter in the acquisition chain (see p.28 of LSM6DSO datasheet, not so explicit it has been improved with respect to the elder one, actually).
-Eleon
