Estimation, Tracking, and Bounds in Close-Range Radar Systems

This thesis investigates the use of performance bounds in radar signal processing, with particular emphasis on the Cramér-Rao bound (CRB) and its misspecified and Bayesian variants, the MCRB and the BCRB. These bounds characterize fundamental limits on estimation accuracy and can therefore be used to assess the performance attainable under a given signal model. However, they can also be used prospectively, as design objectives for selecting sensors or planning future measurements.

Simplified models and assumptions are vital when modeling radar systems in the real world, as an exact model may be too complicated to use in practice. Such simplifications introduce a trade-off between computational tractability and estimation accuracy. Paper A studies this trade-off for close-range radar reflections in the presence of time-varying clock offsets. The MCRB is used to quantify the effects of neglecting wavefront curvature and clock jitter, and a computationally efficient estimator is proposed to mitigate these effects.

The remaining papers consider how performance bounds can be used to design the radar system itself. Paper B addresses sensor selection for the identification of deceptive jammers. The sensors are selected to improve the estimation of the deception range, which provides a means of distinguishing a retransmitted jamming signal from a genuine target reflection. Paper C extends the idea of sensor selection to drones by considering path planning for a MIMO radar drone swarm. The drone trajectories are planned using the BCRB to improve target localization while considering constraints on motion and resource usage. Finally, Paper D considers target tracking by a drone swarm in a GPS-denied environment, where the drone locations themselves are uncertain. Direct-path and target-reflected measurements are both used to estimate the sensor geometry and track the target.

Taken together, the papers demonstrate two complementary uses of performance bounds. First, they provide a means of understanding how modeling assumptions, hardware imperfections, and uncertain sensor locations limit estimation accuracy. Second, they provide a framework for selecting sensors and controlling drones so that future measurements are as informative as possible.