A novel method using the compressive sensing (CS) technique to accelerate the multilevel characteristic basis function method (MLCBFM) is proposed in this letter for the solution of electrically large scattering problems. According to the MLCBFM, the solution process involves the full filling of the impedance matrix as well as the construction and solution of the reduced matrix equation on each level, which is time-consuming. Unlike this, we use the CS technique on each level, except for the bottom level, to recover the coefficients of CBFs and total current, through which both memory consumption and computation time would enjoy a significant reduction. Furthermore, we optimize the generalized orthogonal matching pursuit recovery algorithm to the least squares method which achieves more stable recovery accuracy and higher efficiency. The results of numerical simulations indicate that the proposed method is much more efficient and less memory-consuming compared to the MLCBFM while obtaining a higher level of accuracy.
18 February 2025 in IEEE Antennas and Wireless Propagation Letters
In this letter, the multiple-measurement-vectors (MMV) compressive sensing (CS) model is presented to accelerate the singular value decomposition (SVD) based characteristic basis function method (CBFM) for the solution of wide-angle scattering problems of three-dimensional objects. On one hand, the MMV-CS model is constructed to accelerate the generation of characteristic basis functions (CBFs), through which the whole CBFs can be simultaneously recovered from only a small percentage of them. Such obtained CBFs are more computationally efficient in the SVD procedure with fewer CBFs retained compared to the traditional SVD-CBFM. On the other hand, the MMV-CS model is also constructed to accelerate the solution of induced currents, through which the mutual impedance matrices only need to be partially filled and the reduced matrix equation is replaced by an overdetermined CS matrix equation, which can be efficiently solved. Numerical simulation results indicate that the computational efficiency is significantly improved by the MMV-CS model with sufficient accuracy compared to the traditional SVD-CBFM.
For scattering problems involving array structures with randomly distributed cells within a periodic grid, a characteristic mode basis function method (CMBFM) based on convolutional neural network is proposed to rapidly predict the characteristic mode expansion coefficients for each cell. A five-channel feature map and a weighted loss function based on modal significance are designed to enhance the training performance and prediction accuracy. The trained network exhibits exceptional generalization capabilities across varying numbers of cells, enabling accurate predictions for arrays with arbitrary numbers and distributions of cells. In the simulations, for an array with 100 cells and approximately 1.9×105 unknowns, the proposed method requires less than 0.03 s, representing a 99.9% reduction in computational time relative to the traditional CMBFM.
1 September 2026 in IEEE Antennas and Wireless Propagation Letters