Published Paper Details:

This work compares three techniques for reducing noise in MST radar I/Q data: Wiener filtering, Adaptive Bayesian Shrinkage, and a refined CNN that uses multi-head attention. MST radar echoes often weaken as they pass through atmospheric layers, where turbulence, refractive-index changes, and multipath introduce phase distortions, amplitude fluctuations, and broadband noise. These effects lower Doppler accuracy and affect the retrieval of atmospheric parameters. To test each method, real MST radar measurements and synthetic noisy signals were used. The CNN was trained with a combined loss that includes time-domain reconstruction and a spectral-consistency term, helping it retains Doppler patterns and stable instantaneous phase. Evaluation relied on RMSE, MAE, ?SNR, spectral coherence, cross-correlation, and phase error. Across all these measures, the CNN provided stronger denoising and better Doppler preservation than the other methods. Its balance between noise removal and signal fidelity makes it suitable for MST radar analysis.