F-SMART: Target-Protected Slow-Time Reconstruction for Automotive FMCW Radar Interference Mitigation

Abstract

Noncooperative frequency-modulated continuous -wave (FMCW) interference degrades automotive radar sensing when an asynchronous interference ridge crosses a target beat-frequency line. Masking, excision, or independent per-chirp recovery can then remove target support or perturb slow-time phase consistency, spreading target energy and increasing false alarms in the range–Doppler (RD) map. This article proposes a Fourier-synchrosqueezing slow-time multimodal adaptive reconstruction technique (F-SMART), a detection-oriented predetection mitigation method. Target-prioritized evidence constructed by the Fourier synchrosqueezing transform (FSST) and dynamic programming (DP) identifies target support, interference trajectories, and observation reliability. The resulting evidence is mapped to range–slow-time groups, where weighted multimodal modeling estimates supported Doppler modes and selectively rewrites low-reliability observations within accepted target support. F-SMART is evaluated through a 300-case held-out comparison involving nine processing pipelines, controlled representation and structural ablations, operating-boundary and high-load studies, and publicly available measured mutual-interference data. In the held-out comparison, F-SMART achieved a three-family macro-average $F1$ of 0.9162. F-SMART is therefore best viewed as a detection-oriented RD cleanup method for noncooperative FMCW interference, with explicit tradeoffs among detection performance, false-alarm control, and spectral fidelity.