Signal processing for space surveillance radar

Space Situational Awareness (SSA) by ground-based Digital-Array Radar (DAR) systems
has been attracting global attention in recent years. One of its fundamental
tasks is to provide accurate collision forecast between space debris and active satellites,
leading to increased avoidance probability. In a DAR system, this is achieved
by a rigorous Signal Processing (SP) stage that successfully detects stray space debris
and accurately estimates its parameters: range, Direction-of-Arrival (DOA), radial velocity,
and more. However, classical SP methods, which are not tailored to the SSA
environment, ultimately lead to sub-optimal performance.
The objective of this thesis is to improve the DAR performance through the adaptation
of the SP scheme to the SSA scenario. We present the shortcomings of several
key traditional SP aspects and propose new methods for improved target detection and
parameter estimation, supported by numerical demonstrations of a real SSA DAR.
One such aspect is the SP of large DAR with a high number of receiving antenna
elements, which contributes to a better target detection and DOA estimation. Commonly
used data reduction transformations harness existing resources to serve only the
target detection aspect, which is not necessarily optimal for SSA. We present a novel
parameter-controlled design method to construct a lossless (or optimal) transformation,
with respect to the available resources and an acceptable trade-off between detection
and DOA estimation performance. Moreover, a new tool is provided to analyze the
potential performance of a given array without the need for simulations. The abovementioned
concept is also demonstrated in a multi-static radar network configuration,
showing significant performance gain.
A common topic in most radar systems is the DOA Maximum Likelihood Estimator
(MLE). In SSA, we deal with targets in the Low Earth Orbit (LEO) region, moving at
great orbital velocities. Pulsed radar systems therefore experience the so-called ‘DOA
migration’ effect from pulse to pulse, where each pulse echo returns with changing DOA
and unequal amplitude. With a classical MLE, these effects result in a large target
localization estimation bias (in the order of kilometers). For that purpose, the orbital
mechanics of LEO targets are implemented in the DOA motion model, rendering the
estimation bias removed.
Another issue of great impact is the target masking phenomenon. Classical SP gives
rise to a miss-detection in two specific cases related to SSA. A far range target will
go unnoticed in the presence of a short range target. In addition, a small target will
remain undetected due to a nearby larger target. With this challenge in mind, two new
waveform design concepts are successfully demonstrated.
The above-mentioned techniques ultimately lead to superior estimation accuracy,
higher resource efficiency, and robust detection capabilities, contributing to the SSA
goal. Based on this research, new estimation methods and operational modes could be
explored in the context of a single station and radar networks.