Accelerometer Viability Test: Axis-Calibrated Thingy ADXL vs Lord Reference

Corrected workflow: align by raw magnitude peaks, calibrate Thingy raw axes into the Lord g-frame, then compute magnitude and compare time-domain and FFT agreement. No smoothing is applied to the time-domain signals.

Magnitude Pearson r

0.9547

Magnitude RMSE

0.0401 g

Mean Bias

-0.0089 g

NRMSE / Range

3.81%

Axis RMSE ch1

0.0192 g

Axis RMSE ch2

0.0425 g

Axis RMSE ch3

0.1430 g

FFT Corr. 0-12.5 Hz

0.9516

1. Three-axis source data

Lord channels are measured in g. Thingy channels are raw interpolated counts on the aligned SensorConnect time base.

2. Three-axis calibrated data

Thingy raw axes are mapped to the Lord frame using a 3-axis affine calibration before computing magnitude.

3. Combined acceleration magnitude after axis calibration

Magnitude is computed after calibrating each axis into the Lord g-frame.

4. Residual error

Residual is axis-calibrated Thingy magnitude minus Lord magnitude.

5. FFT viability comparison

FFT uses mean-removed calibrated magnitude with a Hann window. The display is limited to 0-12.5 Hz, matching the original 25 Hz Thingy Nyquist limit.