| Abstract [eng] |
This thesis addresses a key limitation in air-coupled resonant ultrasound spectroscopy: the severe degradation of measurement accuracy caused by overlapping reflections when using long-duration spread-spectrum signals for improved signal-to-noise ratio. The main contribution is a comprehensive solution consisting of three interconnected innovations: 1. SNR-optimized APWP signals — a generalized spectral shaping technique that compensates for frequency-dependent transduction losses and maximizes usable bandwidth (from ~480 kHz to 670 kHz at –20 dB). 2. Inverse filter-based pulse compression with gating and decompression to resolve overlapping reflections in long chirps (up to 350 µs), reducing bias errors. 3. ResidualUNet1D deep learning model for automated reflection separation, achieving up to 45 dB reminder energy for short signals and ~32 dB for 300 µs signals. The proposed methods were validated through simulations on Vitis vinifera leaves and experiments on polycarbonate plates. Results show consistent and significant reduction in bias errors for velocity, attenuation, density, and thickness compared to conventional LFM, NLFM, and pulse excitations. The work demonstrates that spectral loss compensation combined with overlapping signal resolution enables high-accuracy air-coupled TRUS using only a single transducer pair. |
