Space-Variant Autofocus for Forward-Looking Multichannel SAR via Nonconvex Optimization

Abstract

Motion-error-induced phase errors in forward-looking multichannel synthetic aperture radar (FLMC-SAR) exhibit 2-D space-variant characteristics, which present a critical challenge for high-resolution imaging. Traditional space-variant autofocus (SVAF) methods, which rely on separable echoes, are ineffective for FLMC-SAR due to the identical illumination time and inherent azimuthal aliasing of targets within the same range gate. To address this fundamental limitation, this article introduces a novel method for the estimation of space-variant phase errors based on an image entropy minimization criterion. However, the associated optimization problem is nonconvex with local minima and is complicated by strong coupling between error parameters and significant disparities in their partial derivatives. These factors make the global optimum difficult to locate. To overcome these challenges, we employ a joint damped limited-memory Broyden–Fletcher–Goldfarb–Shanno (damped L-BFGS) optimization scheme, supported by a heuristic initialization strategy. Comprehensive simulations and experimental results confirm that the proposed method effectively compensates for space-variant motion errors, demonstrating its correctness and reliability.