How Can a Passive Radar Identify an Aircraft Without Transmitting Signals?

Passive radars listen to radio waves already present in the atmosphere, without emitting their own signals. This makes them cheap and hard to detect, ideal for monitoring low‑altitude airspace around airports. The challenge is that the received waves come from various, often suboptimal sources, complicating the extraction of aircraft features. To solve this, researchers designed a “recognition‑before‑tracking” method. Before following a target’s path, the system checks if the signal originates from the same antenna beam, using motion continuity. When the target appears on the range‑Doppler map, a feature vector based on the reflected radar cross‑section is built, corrected for bias. Additionally, a “normalized bistatic velocity” parameter describes the aircraft’s orientation relative to the receiving station. With these data, the radar compares the unknown target’s vector with reference vectors, measuring similarity to identify the aircraft type. Tests, both simulated and on real data, show the method is lightweight, fast, and able to distinguish different aircraft models even with weak signals. In practice, passive radars can now recognize an intruder before even tracking its trajectory, boosting low‑altitude flight safety.

The proposed method relies on two fundamental axioms of passive radar theory: (i) the spatial continuity of the antenna beam and (ii) the temporal coherence of the return signal. In a sparse‑target scenario, the probability that two distinct returns originate from the same beam is negligible, allowing association of energy peaks in the range‑Doppler map to a single object. The feature vector is built from the equivalent radar cross‑section (RCS) estimate, corrected for the bias introduced by bistatic geometry: RCS_biased = RCS_measured × (4π / λ²) × (G_tx G_rx)⁻¹, where G_tx and G_rx are the transmitter and receiver antenna gains. The normalized bistatic velocity (V_nb) is defined as V_nb = (v_target · u_b) / |u_b|, with u_b the unit vector linking transmitter and receiver. This parameter captures the target’s angle of incidence relative to the baseline, providing insight into the aircraft’s attitude (e.g., turning versus straight flight). For recognition, a Mahalanobis distance is used between the unknown target’s feature vector and those in a reference database, weighting the RCS and V_nb components according to their estimated variance. The decision threshold is calibrated via ROC analysis on simulated datasets, ensuring a false‑alarm rate below 5 %. Experimental results, obtained with a 3 GHz passive radar prototype, show a 12 % classification improvement over RCS‑only methods, confirming the effectiveness of the multimodal approach. Moreover, computational complexity is O(N·M), with N targets and M features, making the method suitable for real‑time implementation on low‑power hardware.