Abstract:
Ultrasonic backscatter provides a non-destructive
approach for characterizing microplastic particles; however, the
spectral features extracted from an echo can depend on the signal
segment selected for analysis. This study investigates the sensitivity
of polyethylene (PE) and poly(methyl methacrylate) (PMMA)
spectral-feature differences to echo-analysis window length using
a publicly available high-frequency ultrasonic backscatter dataset.
Signals sharing a common 50-μm size label were analyzed using
peak-centered windows of 64, 128, 256, and 384 samples. After
mean removal and Hamming tapering, spectral centroid, spectral
spread, and normalized spectral entropy were extracted from the
Fourier-domain representation. Spatially neighboring
measurements were grouped before statistical comparison to
reduce the influence of repeated local observations. The results
show clear feature-dependent window sensitivity. The PE–PMMA
centroid difference remained comparatively stable, changing by
only about 3.2% between the shortest and longest windows,
whereas the spectral-spread difference increased by approximately
115%. Spectral entropy also showed substantial but non-
monotonic variation with window length. These findings
demonstrate that echo-analysis window selection is an important
digital signal processing parameter and should be considered
when evaluating the robustness and reproducibility of ultrasonic
spectral features for microplastic characterization.
