Dual-Frequency Avoidance and Learning-Based Correction for Altitude Error Analysis in Multipath Environments

This study investigates the impact of ground-reflected multipath on low-altitude target height estimation and proposes a novel approach that exploits multipath as an auxiliary information source rather than treating it solely as interference. While conventional dual-frequency avoidance reduces the root-mean-square error, it still suffers from residual bias and instability in the tail region. To overcome these limitations, we introduced Multipath-Assisted Radar Sensing (MARS), which learns spectral features to generate auxiliary corrections and fuses them with avoidance-based estimates through an uncertainty-weighted combination. The simulation results demonstrate that the proposed method consistently outperforms the baseline, avoidance, and auxiliary-only schemes in terms of the RMSE, bias, and CVaR95. In particular, MARS significantly improves estimation reliability under worst-case scenarios, thereby enhancing the operational robustness of low-altitude radar systems.