13 December 2024 in IEEE Transactions on Geoscience and Remote Sensing
Polarization adds an additional dimension to the radar signals, contributing to waveform diversity. Codesign of unimodular waveforms and filters with polarimetric power allocation for maximizing the signal-to-interference plus noise ratio (SINR) plays a key role in the polarimetric radar system. The problem is challenging to solve due to the non-convex nature of the objective function and constraints, coupled with the interdependence of multiple variables. Existing methods mainly solve this problem by fixing the power allocation or relaxing the objective function, and obtain the receive filters with matrix inversion. We directly address this problem without matrix inversion by using the proposed Adaptive Unified Manifold Optimization (AUMO) framework. Specifically, a unified manifold space is constructed to satisfy the constraints of unimodular waveform, filters and power, transforming the problem to an unconstrained optimization problem over the manifold. To solve this problem, a parallel conjugate gradient algorithm is derived. This algorithm can adaptively change the step size by exploring the local features of the manifold space. The experimental results based on the measured data show that the proposed method outperforms existing methods in terms of SINR gain and execution time.
7 December 2023 in IEEE Signal Processing Letters
Intelligent reflecting surface (IRS) is a key technique for enhancing the performance of wireless communications. In this letter, we focus on the sum-path-gain maximization (SPGM) problem in an IRS-aided MIMO communication system, which is non-convex due to the constant modulus constraint. The existing works mainly include the relaxation method with relaxation error and the non-relaxation methods with high complexity. Different from the existing methods, we notice that constant modulus constraint can naturally satisfy the Riemannian manifold, and the deep learning method has strong non-convex learning ability. By exploiting these characteristics, the Riemannian gradient descent network (RGD-Net) is proposed. In the proposed method, we first project the non-convex SPGM problem to the Riemannian manifold. Then, the Riemannian gradient descent iterations are unfolded as the network layers. Finally, the step sizes of each layer are learned in unsupervised manner to ensure converged performance. Compared with the existing methods, the proposed method achieves higher spectral efficiency with lower computational cost
23 August 2024 in IEEE Antennas and Wireless Propagation Letters
The phase-only null beamforming synthesis plays an important role in suppressing interference within radar systems. This problem poses a challenge as it involves non-convex constraints, including constant modulus and inequality constraints. Existing methods are mainly divided into two types. The first type is the relaxation method, which inevitably brings relaxation errors. The second type is non-relaxed methods, which are very sensitive to parameter selection. We noticed that complex circle manifold naturally satisfies the constant module constraint, and further noticed that in actual optimization, the inequality constraint are't necessarily strictly satisfied, and the penalty function method can punish the parts that don't satisfy the constraint. Based on the above considerations, a Riemannian conjugate gradient method based on exact penalty (RCG-EP) is proposed. Simulation results demonstrate the following advantages of the proposed method: (1) Its performance is insensitive to parameter selection; (2) Null depth is 10.16 dB deeper than [9], [10], 12 dB deeper than [5], [6], and 16.81 dB deeper than [7], [8]. Moreover, the advantages of the proposed method become more pronounced as the number of array elements increases.
11 June 2024 in Signal Processing
13 June 2024 in IEEE Geoscience and Remote Sensing Letters
The joint design of waveforms and filters has key applications in polarimetric radar target detection. This paper studies the joint design of waveforms and filters to maximize the signal-to-interference-to-noise ratio (SINR) of polarimetric radar, which is a non-convex and NP-hard problem. Most existing works solve it based on matrix inversion and problem relaxation, which inevitably introduce high complexity and relaxation errors. We notice that a unified manifold space naturally satisfies the constant modulus constraint (CMC) and the norm constraint. Based on this characteristic, we proposed a parallel manifold joint optimization (PMJO) method to solve it without relaxing the objective function. Specifically, the unified product manifold is constructed to satisfy both waveform and filter constraints. Subsequently, the problem is transformed into an unconstrained one by projecting it onto the product manifold space. Finally, a parallel conjugate gradient method is proposed to simultaneously optimize waveforms and filters, which can adaptively adjust the step size and fully explore the product manifold space. Simulation results show that our method can obtain a 2dB performance advantage compared to existing methods, while having a half order of magnitude advantage in time complexity.
20 January 2026 in IEEE Transactions on Aerospace and Electronic Systems
Polarization introduces an additional dimension to radar signals, providing significant advantages in aerospace applications. Constrained transceiver design for maximizing the Signal-to-Interference-plus-Noise Ratio (SINR) is critical in fully polarimetric radar systems. Existing methods primarily focus on average SINR maximization, which lacks robustness in uncertain regions. To address this, we propose a method for worst-case SINR maximization, aiming to maximize the minimum SINR within uncertain regions, subject to the constant modulus constraint of the waveform and the complex sphere constraint of the receive filter. The problem is non-smooth and non-convex, making existing methods inapplicable. To overcome this, we introduce an auxiliary variable to smooth the problem using the LogSumExp function, which transforms the non-smooth objective into a smooth one by converting the inequality constraints into a continuous, differentiable form. Additionally, we propose the Inequality Constrained Product Manifold Optimization (ICPMO) framework, which guarantees convergence. Experimental results based on measured data show that our method outperforms existing methods in terms of worst-case SINR and target detection probability.