SFPL-UnfoldingNet: A Deep Unfolding Network for Super-Resolution of Martian Radar Echoes With Spectral Fidelity and Peak Localization

Subsurface radar sounders on Mars orbiters inherently face a physical tradeoff between penetration depth and range resolution. Mars Advanced Radar for Subsurface and Ionosphere Sounding (MARSIS) has deep penetration depth but low resolution, while Shallow Radar (SHARAD) has high resolution but shallow penetration depth. To bridge this performance gap, this study proposes SFPL-UnfoldingNet, a deep unfolding network for super-resolution (SR) reconstruction of Martian radar echoes. The network incorporates two key modules: 1) a Spectral Amplitude Learning Module (SALM) and a Spectral Phase Learning Module (SPLM) that enhance the amplitude accuracy of subsurface reflections, and 2) a peak localization loss function that minimizes positional deviations of subsurface reflectors. Comprehensive simulation experiments demonstrate that the SALM and SPLM module ensure amplitude accuracy, while the peak localization loss function effectively mitigates positional drift of reflective interfaces. Applied to real MARSIS observations from diverse geological regions, our method reveals distinct subsurface interfaces that were previously obscured in original data. Crucially, cross-validation with co-located SHARAD data shows that SFPL-UnfoldingNet improves the range resolution of MARSIS by a factor of 9.5, achieving a final resolution of 15.7 m. The super-resolved MARSIS echoes exhibit a main-lobe width of ~0.3 μs (comparable to SHARAD) and a minimum resolvable echo time difference of ~0.31 μs. Thus, our method effectively bridges the resolution gap between the two instruments while maintaining the deep-penetration advantage of MARSIS, providing a powerful new tool for planetary subsurface exploration.