A unified spatiotemporal-geometry framework for target classification and localisation in dual-static passive radar

Area of Science:

• * Electrical Engineering
• * Signal Processing
• * Radar Systems

Background:

• * Passive radar offers covert surveillance capabilities by utilizing ambient broadcast signals.
• * Existing methods struggle with joint classification and localization, especially at low SNRs or with slow targets.
• * Independent processing steps in conventional passive radar can lead to inconsistent results.

Purpose of the Study:

• * To develop a joint spatiotemporal-geometry framework for dual-static passive radar.
• * To improve target classification and position estimation accuracy in passive radar systems.
• * To address limitations of independent processing in existing passive radar techniques.

Main Methods:

• * A novel framework combining a spatiotemporal encoder (dilated convolutions, cross-attention) and a bistatic solver (Cramér-Rao-weighted Levenberg-Marquardt).
• * An iterative optimization loop coupling the encoder and solver, enforcing bistatic delay and Doppler equations as hard constraints.
• * Physics-consistent velocity penalty guided by encoder class probability within the solver.

Main Results:

• * Achieved 93.7% classification accuracy and a 0.937 F1-score.
• * Reduced localization error by 28.1% compared to a geometry-only baseline at -6 dB SNR (1.15 km error).
• * Demonstrated consistent outperformance over seven baseline methods in both classification and localization.

Conclusions:

• * The proposed joint framework effectively integrates spatiotemporal information and geometric constraints for passive radar.
• * Iterative optimization enforces physical laws and improves performance at low SNRs.
• * The method offers a robust and accurate solution for dual-static passive radar applications.