Wavelet-Fused Deep Learning for Computational Phase Correction in Dual-Comb Ranging

Dual-comb ranging enables rapid, high-precision absolute distance measurements, but its performance is constrained by intrinsic phase noise, which induces temporal jitter and degrades pulse-to-pulse mutual coherence. Here, we propose a deep learning network Wavelet-Fused DenseNet (WFDNet) for hardware-free computational phase correction in dual-comb ranging. Through integration of complex wavelet decomposition and physics-guided feature encoding, the network, trained on model-generated data, can directly extract multi-scale time–frequency features to correct phase distortions and recover temporal coherence of the signals. Results from both simulated and experimental scenarios reveal that the approach can effectively suppress spectral noise and retrieve robust and unambiguous phases information, achieving high ranging accuracy with a standard deviation of 0.6 μm.