STARMAP: S paceborne T arget A cquisition R adar With M eta-RL A ssisted P lacement

The urgent requirement to monitor and identify unmanned aerial systems (UASs) within restricted airspace has become increasingly critical. Traditional methods fail to detect low-observable (LO) UASs effectively, thus presenting considerable threats in defense and civil sectors. While radar systems are traditionally favored for their robust detection capabilities, the standard active radar configurations—where transmitters are collocated with receivers—face numerous operational challenges. A more promising solution is adopting passive radar technology, which leverages ambient environmental signals, thereby obviating the need for proprietary transmitters. In this vein, we introduce STARMAP framework, a cutting-edge methodology employing spaceborne illuminators. This approach is particularly effective for extensive range operations such as the surveillance of missiles and fighter jets at high altitudes. Despite the benefits, the major challenges include measuring distance and pinpointing targets, exacerbated by the unknown positions of transmitters. STARMAP overcomes these limitations by integrating time difference of arrival (TDOA) methods with bistatic Doppler shift evaluations. A crucial yet often neglected aspect in passive radar systems is the influence of receiver spatial configuration on localization precision. STARMAP underscores the necessity to optimize the arrangement of receivers to minimize errors induced by unfavorable geometries, a task complicated by its NP-hard nature. To address this, we have developed an advanced meta-reinforcement learning (meta-RL) algorithm, enhancing a double deep Q-network (DDQN) to optimize receiver placement. Through rigorous testing across various scenarios and dimensions, our findings demonstrate that STARMAP substantially improves localization accuracy by reducing geometry-induced errors compared to traditional placement strategies.