Checking Accelerometer Filtering Before Reducing Output Data Rate

A lower accelerometer output data rate reduces the number of samples delivered to the host, but it does not by itself describe the sensor’s complete frequency response. Unwanted vibration can enter before a decimation stage or be aliased during later processing. Review the entire sampling chain before changing the rate.

Identify each filter and sampling boundary

The Analog Devices ADXL354 and ADXL355 data sheet distinguishes the analog antialiasing response from the digital decimation filters. It also explains that the final passband depends on their combined response, with delay changing across settings.

Draw mechanical response, analog filtering, internal conversion, digital decimation, output registers, and host processing. Label the rates and available filter settings at each boundary. A host read interval is not necessarily the sensor’s conversion interval.

Define wanted and unwanted motion

Specify the highest wanted vibration frequency and its permitted amplitude error. Separately identify strong out-of-band sources such as motors, gears, impacts, or structural resonances. Their amplitudes matter because finite filter rejection may leave a meaningful residual.

For a hypothetical host stream sampled at 200 samples per second, a remaining 170 Hz component can appear at 30 Hz after sampling. This arithmetic identifies a possible alias location; its measured amplitude depends on all filtering before that sampling boundary.

Avoid unfiltered host decimation

If the host keeps only every tenth sensor sample, provide a suitable digital low pass filter before discarding the others. Merely reducing the polling rate can produce irregular sampling and repeated or missed values, depending on the device’s register behavior.

Check the data ready signal or supported timestamps to establish the original sample spacing. Include filter group delay when the acceleration feeds a control loop or is compared with another sensor.

Verify with controlled vibration

Apply known motion within the fixture and sensor limits and compare several rate and filter settings. Sweep the wanted band, then examine selected out-of-band frequencies that could fold into it. Keep mounting and stimulus amplitude consistent.

Compare spectral peaks with predicted alias locations. A peak that moves when the output rate changes is a useful clue, but inspect timing and stimulus stability before assigning its cause. Mechanical resonances can also change the observed response.

Save the complete rate map, filter configuration, stimulus frequencies, measured attenuation, and delay. A successful configuration preserves wanted motion while bounding the unwanted response. A low output rate, a visually smooth trace, or a software average alone does not establish that aliased vibration has been removed from the measurement.