A Synthetic Data Generation Framework for Model‑Based Deep Learning in LDACS-Based Passive Radar

Abstract

Air traffic surveillance relies on cooperative systems that depend on the congested 1090 MHz band, making them vulnerable to outages and interference. Additionally, the only non-cooperative system available is being decommissioned in multiple countries, motivating the search for non-cooperative alternatives. Previous works have proposed to employ passive radar utilizing the L‑band Digital Aeronautical Communication System (LDACS) communication signals as signals of opportunity to detect aircraft. While feasible, the low LDACS transmit power leads to very weak reflections, requiring long integration times and advanced processing. As a result, a super-resolution Sparse Bayesian Learning (SBL) approach was proposed for enabling multi-target detection while maximizing the signal-to-noise ratio. However, this approach relies exclusively on the analytical modelling of the reflected signal. To overcome the sensitivity of the SBL to model mismatch, we propose a hybrid model‑based deep learning approach where a neural network enhances the model-based computations. In light of the difficulty of collecting a large quantity of real‑world LDACS passive radar measurements, this work focuses on developing a synthetic data generation framework that enables a future proof‑of‑concept demonstration of the proposed model‑based deep learning approach.