Low-Complexity Neural Network-Based Processing Algorithm for QCM-D Signals

This article introduces a novel approach to estimate the oscillation frequency and decay time constant of quartz crystal microbalance with dissipation monitoring (QCM-D) signals using low-complexity neural network (NN) models. The proposed method involves a two-step process: frequency-domain preprocessing using fast Fourier transform (FFT) followed by a shallow neural network for frequency estimation, and time domain preprocessing for envelope detection followed by a second shallow neural network for time constant estimation. The networks were trained and tested on datasets from both a dedicated QCM-D testbench and a numerical simulator. The results demonstrate accurate estimation with errors below 10 Hz for the frequency and $1~{\mu }$ s for the time constant, making the approach promising for various QCM-D applications. The simplicity of the neural network models facilitates the implementation on embedded platforms, reducing the system complexity and offering potential for cost-effective QCM-D sensing systems.