Sensing Accuracy Optimization for Multi-UAV SAR Interferometry with Data Offloading

TL;DR

This paper presents an optimization framework for UAV swarms performing multi-baseline InSAR to enhance DEM accuracy, integrating data offloading and joint optimization of formation, velocity, and power for real-time high-precision Earth surface monitoring.

Contribution

It introduces a novel joint optimization approach for UAV formation, velocity, and communication power to improve InSAR sensing accuracy and real-time data transmission.

Findings

Outperforms baseline optimization methods like GAs, SA, and DRL.

Achieves sub-decimeter vertical accuracy in DEM reconstruction.

Demonstrates the effectiveness of coordinated UAV swarms for high-precision Earth observation.

Abstract

The integration of unmanned aerial vehicles (UAVs) with radar imaging sensors has revolutionized the monitoring of dynamic and local Earth surface processes by enabling high-resolution and cost-effective remote sensing. This paper investigates the optimization of the sensing accuracy of a UAV swarm deployed to perform multi-baseline interferometric synthetic aperture radar (InSAR) sensing. In conventional single-baseline InSAR systems, only one synthetic aperture radar (SAR) antenna pair acquires two SAR images from two distinct angles to generate a digital elevation model (DEM) of the target area. However, multi-baseline InSAR extends this concept by aggregating multiple acquisitions from different angles, thus, significantly enhancing the vertical accuracy of the DEM. The heavy computations required for this process are performed on the ground and, therefore, the radar data is transmitted in real time to a ground station (GS) via a frequency-division multiple access (FDMA) air-to-ground backhaul link. This work focuses on improving the sensing precision by minimizing the height error of the averaged DEM while simultaneously ensuring sensing and communication quality-of-service (QoS). To this end, the UAV formation, velocity, and communication power allocation are jointly optimized using evolutionary algorithms (EAs). Our approach is benchmarked against established optimization methods, including genetic algorithms (GAs), simulated annealing (SA), and deep reinforcement learning (DRL) techniques. Numerical results show that the proposed solution outperforms these baseline schemes and achieves sub-decimeter vertical accuracy in several scenarios. These findings underline the potential of coordinated UAV swarms for delivering high-precision and real-time Earth observations through radar interferometry.