Robust Covariance Estimation under Imperfect Constraints using an Expected Likelihood Approach

TL;DR

This paper introduces a robust covariance matrix estimation method for radar STAP that effectively handles imperfect prior constraints by using an expected likelihood approach, improving accuracy over existing methods.

Contribution

It develops a novel EL-based framework for setting covariance constraints under non-ideal conditions, with analytical proofs of uniqueness and practical algorithms.

Findings

Outperforms existing covariance estimators in simulations

Provides efficient algorithms for constraint determination

Demonstrates effectiveness on KASSPER data set

Abstract

We address the problem of structured covariance matrix estimation for radar space-time adaptive processing (STAP). A priori knowledge of the interference environment has been exploited in many previous works to enable accurate estimators even when training is not generous. Specifically, recent work has shown that employing practical constraints such as the rank of clutter subspace and the condition number of disturbance covariance leads to powerful estimators that have closed form solutions. While rank and the condition number are very effective constraints, often practical non-idealities makes it difficult for them to be known precisely using physical models. Therefore, we propose a robust covariance estimation method for radar STAP via an expected likelihood (EL) approach. We analyze covariance estimation algorithms under three cases of imperfect constraints: 1) a rank constraint, 2) both rank and noise power constraints, and 3) condition number constraint. In each case, we formulate precise constraint determination as an optimization problem using the EL criterion. For each of the three cases, we derive new analytical results which allow for computationally efficient, practical ways of setting these constraints. In particular, we prove formally that both the rank and condition number as determined by the EL criterion are unique. Through experimental results from a simulation model and the KASSPER data set, we show the estimator with optimal constraints obtained by the EL approach outperforms state of the art alternatives.