MIMO Radar Waveform Design for Range-ISL Optimization via Iterative Deep Unfolding Network

Ziwei Zhao,Jinfeng Hu,Kai Zhong,Yongfeng Zuo,Huiyong Li,Bozhou Zhang

DOI: https://doi.org/10.1109/lgrs.2024.3368446

IF: 5.343

2024-03-19

IEEE Geoscience and Remote Sensing Letters

Abstract:Multiple-input multiple-output (MIMO) radar unimodular waveform design with range-integrated sidelobe level (ISL) optimization is a key technology in remote sensing. Due to the nonconvex quartic objective function and constant modulus constraint (CMC), the problem is NP-hard and nonconvex. Existing methods mainly include relaxation methods or nonrelaxation methods with huge computational costs. We notice that complex circle manifold (CCM) naturally satisfies the CMC. By projecting onto the CCM, the problem is transformed into an unconstrained minimization problem that can be addressed using the Riemannian gradient descent (RGD) algorithm. Furthermore, we notice that the RGD algorithm can be unfolded into a deep-learning model. Hence, a computationally efficient method without relaxation, iterative deep unfolding network (IDUN), is proposed. First, this problem is converted into an unconstrained fourth-order polynomial minimization problem on the CCM. Then, by unfolding the RGD algorithm as the network layer, IDUN is developed by adaptively learning the step sizes. Compared with existing methods, the proposed method has superior performance and less computational cost.

imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics

What problem does this paper attempt to address?

The paper primarily addresses the single-mode waveform design problem in Multiple Input Multiple Output (MIMO) radar, specifically focusing on the optimization of the Integrated Sidelobe Level (ISL). Specifically, the research aims to design MIMO radar single-mode waveforms with excellent range-ISL characteristics under the Constant Modulus Constraint (CMC).

Due to the non-convex quartic objective function and the constant modulus constraint, this problem is NP-hard and non-convex. Existing methods are mainly divided into two categories:

1. Relaxation Methods: These simplify the problem by approximating the objective function, such as the Cyclic Algorithm New (CAN) and the Weighted Cyclic Algorithm New (WeCAN). Although these methods can solve the problem, they inevitably introduce relaxation errors.

2. Non-Relaxation Methods: These directly solve the original problem without relaxation, such as the Block Coordinate Descent (BCD) and the Nonlinear Alternating Direction Method of Multipliers (ADMMs). These methods usually have high computational costs.

To address the issues present in the above two types of methods, the paper proposes a new, more computationally efficient non-relaxation method—Iterative Deep Unfolding Network (IDUN). This method first maps the problem onto the Complex Circle Manifold (CCM), transforming it into an unconstrained minimization problem, and solves it using the Riemannian Gradient Descent (RGD) algorithm. Furthermore, the authors note that the RGD algorithm can be unfolded into a deep learning model, thus developing IDUN to accelerate the convergence process by adaptively learning the step size.

Compared to existing methods, IDUN achieves superior performance while maintaining low computational costs. Experimental results show that IDUN reduces the Maximum Autocorrelation Sidelobe (MAS) and Maximum Cross-correlation Sidelobe (MCS) by 27 to 32dB and 33 to 37dB, respectively, compared to other methods. Additionally, IDUN demonstrates faster convergence speed and lower execution time.