The low sidelobe level high-gain adaptive beamforming method for uniform subarrays Authors: Yuxi Du, Weijia Cui, Bin Ba, Guanghui Su, Long Zhang Published: 2025 ABSTRACT Aiming at the weak signal enhancement, target detection and anti-jamming, and reduction of energy consumption faced by array radar technology in complex electromagnetic environments, this paper proposes a low sidelobe level high-gain adaptive beamforming method for uniform subarrays. The proposed method utilizes the virtual interference signals iteration method to greatly reduce the computational dimension on the basis of ensuring the output performance. Firstly the proposed algorithm compensates the delay of the array elements within the subarray with respect to the reference array element by element-level weighting and derives the array output after element-level weighting. Next the subarray-level interference plus noise covariance matrix was derived. Then the Lagrange method is applied to solve the weight vector for each iteration. Finally, the algorithm is made to iterate until convergence by adding virtual interference signals. The theoretical derivation and simulation experiments demonstrate the effectiveness and reliability of the proposed method. Introduction Adaptive beamforming is a technique for achieving directional signal enhancement or suppression by controlling the phase and amplitude of signals in an antenna array. With the core advantages of directional signal enhancement and interference suppression, spatial flexibility and dynamic adjustment capability, high resolution and precise detection capability, as well as multi-source data synergy and system integration, beamforming technology is widely used in many fields such as communications, remote sensing, navigation, disaster monitoring1–5. Enhancement of weak signals to achieve target detection and anti-jamming in complex electromagnetic environments as well as to reduce energy consumption are current challenges that need to be solved in the field of array radar6,7. To address the matrix pathology problem caused by covariance matrix estimation errors (e.g., small snap counts, array calibration errors) in practical applications, Cox proposed the diagonal loading (DL) algorithm8,9. The algorithm can suppress the eigenvalue dispersion through the loading amount by introducing artificial white noise thus avoiding the numerical instability in the inverse of the covariance matrix. However, in practical scenarios, due to unknown noise power and other factors, the DL factor needs to be determined by statistical or optimization methods, which is easy to introduce suboptimal solutions10–12. Overloading can lead to distortion of the signal and reduce the main flap gain. At the same time, some of the improved algorithms require multiple matrix reconstructions and decompositions which can lead to sacrificing the real-time performance of the algorithms. The eigenspace-based (ESB) algorithm project the received signals onto the orthogonal subspaces formed by eigenvectors through the eigen-decomposition of the signal covariance matrix, thereby separating the signal subspace from the noise subspace and using the information of the signal subspace to suppress noise and interference13,14. The separation of the signal subspace from the noise subspace can effectively suppress non-coherent interference and maintain the integrity of the target signal even in scenarios with strong noise or dense interference. The accuracy of subspace partitioning is affected by the estimation error of the covariance matrix. When the number of snapshots is small or the signal to noise ratio (SNR) is low, it is prone to subspace confusion15–17. The algorithm assumes that the signal and noise subspaces are strictly orthogonal. In practice, model mismatch (such as direction of arrival errors) can destroy the orthogonality, leading to performance degradation. The worst-case performance optimization algorithm sets the error range of the signal guidance vector (such as direction mismatch, array calibration error, etc.) to find the optimal weight vector in the worst case within the error set and ensure that the algorithm can maintain the distortion-free response of the signal and suppress interference under all possible errors18. It can effectively counteract the mismatch of the steering vector, array calibration errors and the estimation deviation of the covariance matrix caused by a small number of snapshots, and still maintain stable output under non-ideal conditions. The error radius needs to be preset. Meanwhile, if it does not match the actual error distribution, it may lead to the degradation of the algorithm performance. In practical applications, due to the limitation of numerical solution accuracy, the algorithm may not converge to the global optimal solution, affecting the interference suppression effect19–21. The performance of the above-mentioned algorithms is all much lower than that of the minimum variance distortionless response (MVDR) method due to the self-cancellation phenomenon at high SNR. The weighting method achieves precise control of the sidelobe level (SLL) while ensuring the width of the main lobe by taking advantage of the isogamy characteristics of the Chebyshev polynomial22. This method can obtain the lowest SLL under the constraints of the same number of array elements and the width of the main lobe. Meanwhile, the closed-form solution based on Chebyshev polynomials can directly calculate the weights without iterative optimization. However, the width of the main lobe and the SLL cannot be adjusted independently. Therefore, when designin g, the performance of one side needs to be sacrificed to meet the requirements of the other side23–25. It is easily affected by manufacturing tolerances, which leads to the actual SLL performance deviating from the theoretical value due to the large excitation amplitude of the array edge element. The Olen beam graph algorithm minimizes the energy of the side lobe region by setting a virtual interference source and adjusting the array weights based on iterative optimization, while maintaining the directivity gain of the main lobe26. However, the iterative optimization process involving large-scale matrix operations will lead to a sharp increase in the computational load when the number of array elements increases. The quantity and location of virtual interference sources and the iteration termination conditions need to be finely tuned. Improper Settings can easily lead to distortion of the main lobe or residual side lobes27,28. The beam graph optimization algorithm based on second-order cone programming (SOCP) transforms the beam design problem into a convex optimization model, uses second-order cone constraints to describe performance indicators such as side lobe suppression and main lobe shape, and efficiently solves the global optimal weights through the interior point method29–32. The SOCP can ensure the uniqueness and global optimality of the weight solution to avoid the local convergence defect of traditional gradient-based algorithms because it is a convex optimization problem. When optimizing large-scale arrays, the SOCP model involves high-dimensional matrix operations, and the solution time increases exponentially with the number of array elements, which is not friendly to scenarios with high real-time requirements. However, the above-mentioned beam optimization design algorithm only considers one of the SLL of the beamforming device or the array gain. In fact, these indicators are often interrelated. Unconsidered indicators often lead to a significant decline due to complex environments and errors and other factors. This paper proposes a beam-forming method with low SLL and high array gains for uniform subarrays. The desired signal can be superimposed because the element-level weighting is used to compensate for the delay of the elements within the subarray. The subarray-level weighting is used to suppress interference signals and increase array gain. The side lobe stage is controlled by adding virtual interference signals. The contribution of this paper lies in the combination of reducing the SLL and increasing the array gain and their application to subarrays, providing a new idea for the development of beamforming. Capital letters, bold lowercase letters and bold capital letters are used to represent variables, vectors and matrices respectively in this article. Given a matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{D}}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{D}}^{{\mathrm{T}}}$$\end{document} represents the transpose of the matrix and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{D}}^{{\mathrm{H}}}$$\end{document} represents the conjugate transpose of the matrix. Signal model Suppose a desired signal with an incoming wave direction of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta_{0}$$\end{document} is incident on a uniform linear array (ULA) composed of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P$$\end{document} array elements. The wavelength of the signal is represented as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document} and the distance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document} between adjacent array elements is half of the signal wavelength. All array elements are isotropic and there are no situations such as channel inconsistency or mutual coupling. Taking the first array element on the far left of the array as the reference array element, the data of the array at time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document} can be described as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{gathered} {\mathbf{x}}(t) = {\mathbf{a}}(\theta_{0} )d(t) + {\mathbf{n}}(t) \\ = {\mathbf{x}}_{s} (t) + {\mathbf{n}}(t) \\ \end{gathered}$$\end{document}among them, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d(t)$$\end{document} represents the expected signal and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{a}}(\theta_{0} )$$\end{document} is defined as the direction vector of the desired signal:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{a}}(\theta_{0} ) = [1,e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }\sin \theta_{0} }} , \cdot \cdot \cdot ,e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(P - 1)\sin \theta_{0} }} ]^{{\mathrm{T}}}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{n}}(t)$$\end{document} is the Gaussian white noise with a mean value of 0 and a power of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma_{n}^{2}$$\end{document}:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{n}}(t) = [n_{0} ,n_{1} , \cdots ,n_{P - 1} ]^{{\mathrm{T}}}$$\end{document} If there are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L$$\end{document} interference signals of the same frequency as the desired signal incident on the array and the desired signal and the interference signals are independent of each other, then the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$l{\mathrm{th}}$$\end{document} interference signal received by the array can be expressed as:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{x}}_{l} (t) = {\mathbf{a}}(\theta_{l} )f_{l} (t){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} l = 1,2, \cdots ,L$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{l} (t)$$\end{document} represents the data of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$l{\mathrm{th}}$$\end{document} interference signal at time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document}. The array receiving data can be rewritten as:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{x}}_{a} (t) = {\mathbf{x}}_{s} (t) + \sum\limits_{l = 1}^{L} {{\mathbf{x}}_{l} (t)} + {\mathbf{n}}(t)$$\end{document} The beamforming output of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N$$\end{document} snapshots can be expressed as:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} y(t) &= {\mathbf{w}}^{{\mathrm{H}}} {\mathbf{x}}_{a} (t){\kern 1pt} {\kern 1pt} \\ &= {\mathbf{w}}^{{\mathrm{H}}} {\mathbf{x}}_{s} (t) + {\mathbf{w}}^{{\mathrm{H}}} (\sum\limits_{l = 1}^{L} {{\mathbf{x}}_{l} (t)} + {\mathbf{n}}(t)){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} t = 1,2, \cdots ,N \\ \end{aligned}$$\end{document} The output signal to interference plus noise ratio (SINR) of the array can be defined as:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SINR}} = \frac{{\left| {{\mathbf{w}}^{{\mathrm{H}}} {\mathbf{x}}_{s} (t)} \right|^{2} }}{{\left| {{\mathbf{w}}^{{\mathrm{H}}} (\sum\limits_{l = 1}^{L} {{\mathbf{x}}_{l} (t)} + {\mathbf{n}}(t))} \right|^{2} }}$$\end{document} The beam response of the array can be written as:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b(\alpha ) = {\mathbf{w}}^{{\mathrm{H}}} {\mathbf{a}}(\alpha ){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \alpha \in \Theta$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Theta$$\end{document} represents the observation area. The proposed method Suppose a ULA is divided into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M$$\end{document} subarrays, and the number of elements in each subarray is equal and both are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B$$\end{document}. The array model of the proposed method is shown in Fig. 1. We achieve a lower SLL of the beam pattern and a higher output array gain through element-level weighting and subarray-level weighting. Taking the position of the first element on the left side of each subarray as the subarray position, the subarray position can be expressed in the order from left to right as:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{k}} = [k_{1} ,k_{2} , \cdots ,k_{M} ]^{{\mathrm{T}}}$$\end{document}and exist10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_{i + 1} - k_{i} = B{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} i = 1,2, \cdots ,M - 1$$\end{document}Fig. 1The array model of the proposed method. The array model of the proposed method. The element-level weighting The element-level weighting enables the desired signal to be superimposed by compensating for the delay of the internal elements of the subarray relative to the reference elements of the subarray. The incoming wave direction of the expected signal is taken as prior knowledge. By observing formula (5), the weighting within each subarray can be expressed as:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{gathered} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} [w_{{ek_{m} }} ,w_{{e(k_{m} + 1)}} , \cdots ,w_{{e(k_{m} + B - 1)}} ] \hfill \\ = [1,e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }\sin \theta_{0} }} , \cdot \cdot \cdot ,e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(B - 1)\sin \theta_{0} }} ]{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} 1 \le m \le M \hfill \\ \end{gathered}$$\end{document} The element-level weighting can be written as:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathbf{w}}_{e} &= [w_{{ek_{1} }} ,w_{{e(k_{1} + 1)}} , \cdots ,w_{{ek_{M} }} ]^{{\mathrm{T}}} \\ &= [1, \cdot \cdot \cdot ,e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(B - 1)\sin \theta_{0} }} , \cdots ,1, \cdot \cdot \cdot ,e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(B - 1)\sin \theta_{0} }} ]^{{\mathrm{T}}} \\ \end{aligned}$$\end{document} The output after element-level weighting can be expressed as:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathbf{x}}_{e} (t) & = {\mathbf{w}}_{e}^{ * } \circ {\mathbf{x}}_{a} (t) \\ & = {\mathbf{x}}_{se} (t) + \sum\limits_{l = 1}^{L} {{\mathbf{x}}_{le} (t)} + {\mathbf{w}}_{e}^{ * } \circ {\mathbf{n}}(t) \\ \end{aligned}$$\end{document}where1415\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{gathered} \hfill {\mathbf{x}}_{le} (t) = [1, \cdots ,e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }((k_{2} - 1)\sin \theta_{l} - (B - 1)\sin \theta_{0} )}} , \cdot \cdot \cdot ,e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{M} \sin \theta_{l} }} , \\ \hfill \cdot \cdot \cdot ,e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(P\sin \theta_{l} - (B - 1)\sin \theta_{0} )}} ]^{{\mathrm{T}}} f_{l} (t){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} 1 \le l \le L \\ \end{gathered}$$\end{document} After superimposing the array element outputs within each subarray, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{x}}_{e} (t)$$\end{document} can be rewritten as:16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \widetilde{{{\mathbf{x}}_{e} }}(t) &= {\mathbf{a}}_{e0} d(t) + \sum\limits_{l = 1}^{L} {{\mathbf{a}}_{el} f_{l} (t)} + {\mathbf{n}}_{e} (t) \\ &= [Be^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{1} \sin \theta_{0} }} , \cdots ,Be^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{M} \sin \theta_{0} }} ]^{{\mathrm{T}}} d(t) + \sum\limits_{l = 1}^{L} {[(1 + \cdots + } \\ &\quad e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }((k_{2} - 1)\sin \theta_{l} - (B - 1)\sin \theta_{0} )}} ), \cdot \cdot \cdot ,(e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{M} \sin \theta_{l} }} + \cdot \cdot \cdot + e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(P\sin \theta_{l} - (B - 1)\sin \theta_{0} )}} )]^{{\mathrm{T}}} f_{l} (t) \\ &\quad+ [(n_{1} + \cdot \cdot \cdot + n_{{k_{2} - 1}} e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(B - 1)\sin \theta_{0} }} ), \cdot \cdot \cdot ,(n_{{k_{M} }} + \cdot \cdot \cdot + n_{P} e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(B - 1)\sin \theta_{0} }} )]^{{\mathrm{T}}} \\ \end{aligned}$$\end{document} Optimization problems The beam response of the array can be rewritten as:17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} b(\alpha ) &= ({\mathbf{w}}_{u} \otimes J_{B \times 1} \circ {\mathbf{w}}_{e} )^{{\mathrm{H}}} {\mathbf{a}}(\alpha ) \\ &= ({\mathbf{w}}_{u}^{{\mathrm{H}}} \otimes J_{B \times 1} )({\mathbf{w}}_{e}^{{\mathrm{H}}} \circ {\mathbf{a}}(\alpha )) \\ &= {\mathbf{w}}_{u}^{{\mathrm{H}}} {\mathbf{v}}^{\alpha } {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \alpha \in \Theta \\ \end{aligned}$$\end{document}among them, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{w}}_{u}$$\end{document} represents subarray-level weighting, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J_{B \times 1}$$\end{document} represents a full 1-vector of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B \times 1$$\end{document} dimension and18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{v}}^{\alpha } = [v_{1}^{\alpha } ,v_{2}^{\alpha } , \cdots v_{M}^{\alpha } ]^{{\mathrm{T}}}$$\end{document}where19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} v_{i}^{\alpha } &= e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{i} \sin \theta_{l} }} + \cdot \cdot \cdot + e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }((k_{i} + B - 1)\sin \theta_{l} - (B - 1)\sin \theta_{0} )}} \\ & = e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{i} \sin \theta_{l} }} (1 + \cdot \cdot \cdot + e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(B - 1)(\sin \theta_{l} - \sin \theta_{0} )}} ) \\ & = e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{i} \sin \theta_{l} }} \frac{{\sin (s\frac{{{2}\pi }}{\lambda }B(\sin \theta_{l} - \sin \theta_{1} )/2)}}{{\sin (s\frac{{{2}\pi }}{\lambda }(\sin \theta_{l} - \sin \theta_{1} )/2)}}e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(B - 1)(\sin \theta_{l} - \sin \theta_{1} )/2}} \\ \end{aligned}$$\end{document} Therefore, the subarray-level weighted values generated by the virtual signal have the ability to suppress the incoming wave signal with direction \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta_{l}$$\end{document}. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta_{s}$$\end{document} represents the angle corresponding to the highest side lobe level, and the corresponding interference plus noise covariance matrix (INCM) can be expressed as:20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{R}}_{{\theta_{s} }} = {\mathbf{v}}^{{\theta_{s} }} ({\mathbf{v}}^{{\theta_{s} }} )^{{\mathrm{H}}}$$\end{document} The subarray-level weighting enables the attainment of the lowest possible SLL and the highest possible array gain while keeping the width of the main lobe unchanged. We suppress the side lobes by increasing the virtual interference signals at the angle corresponding to the highest side lobe of each beam pattern to obtain the lowest SLL. The optimal subarray-level weighting is obtained through multiple iterations. Then the subarray-level weight design problem of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z + 1{\mathrm{st}}$$\end{document} iteration can be expressed as:21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{gathered} \mathop {\min }\limits_{{{\mathbf{w}}_{u}^{z + 1} }} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} ({\mathbf{w}}_{u}^{z + 1} )^{{\mathrm{H}}} {\mathbf{R}}_{i + n} {\mathbf{w}}_{u}^{z + 1} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\mathbf{w}}_{u}^{z + 1} {\mathbf{a}}_{u} (\theta_{0} ) = 1 \hfill \\ \mathop {\min }\limits_{{{\mathbf{w}}_{u}^{z + 1} }} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} ({\mathbf{w}}_{u}^{z + 1} )^{{\mathrm{H}}} {\mathbf{R}}_{{\theta_{s} }}^{z + 1} {\mathbf{w}}_{u}^{z + 1} \hfill \\ \end{gathered}$$\end{document}among them, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{w}}_{u}^{z + 1}$$\end{document} represents the weighted value of the subarray in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z + 1$$\end{document} iteration, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{R}}_{i + n}$$\end{document} represents the INCM at the subarray level, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{R}}_{{\theta_{s} }}^{z + 1}$$\end{document} represents the INCM composed of the virtual interference signals added in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z + 1$$\end{document} iteration. The INCM at the subarray level can be defined as:22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathbf{R}}_{i + n} &= (\sum\limits_{l = 1}^{L} {{\mathbf{a}}_{el} f_{l} (t)} + {\mathbf{n}}_{e} (t))(\sum\limits_{l = 1}^{L} {{\mathbf{a}}_{el} f_{l} (t)} + {\mathbf{n}}_{e} (t))^{{\mathrm{H}}} \\ &= \left[ {\begin{array}{*{20}c} {r_{11} } & {r_{12} } & \cdots & {r_{1M} } \\ {r_{21} } & {r_{22} } & \cdots & {r_{2M} } \\ \vdots & \vdots & \ddots & \vdots \\ {r_{M1} } & {r_{M2} } & \cdots & {r_{MM} } \\ \end{array} } \right] \\ \end{aligned}$$\end{document}we define:23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{li} = e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }k_{i} \sin \theta_{l} }} + \cdot \cdot \cdot + e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }((k_{i} + B - 1)\sin \theta_{l} - (B - 1)\sin \theta_{0} )}}$$\end{document}24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_{i} = n_{{k_{i} }} + \cdot \cdot \cdot + n_{{k_{i} + B - 1}} e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(B - 1)\sin \theta_{0} }}$$\end{document} The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}-th row and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j$$\end{document}-th column of formula (23) can be written as:25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{ij} = \left(\sum\limits_{l = 1}^{L} {C_{li} f_{l} (t)} + D_{i} \right) \cdot \left(\sum\limits_{l = 1}^{L} {C_{lj} f_{l} (t)} + D_{j} \right )^{{\mathrm{H}}}$$\end{document} Because26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C_{li} f_{l} (t) \cdot C_{lj} f_{l} (t)& = \sigma_{l}^{2} (B + (B - 1)(e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(\sin \theta_{l} - \sin \theta_{0} )}} + e^{{{\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(\sin \theta_{l} - \sin \theta_{0} )}} ) + \cdot \cdot \cdot \\ & \quad + (e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(B - 1)(\sin \theta_{l} - \sin \theta_{0} )}} + e^{{{\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(B - 1)(\sin \theta_{l} - \sin \theta_{0} )}} ))e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(k_{i} - k_{j} )\sin \theta_{l} }} \\ &= E_{l} e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(k_{i} - k_{j} )\sin \theta_{l} }} \\ \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma_{l}^{2}$$\end{document} represents the power of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$l$$\end{document}-th interference signal. Therefore, formula (26) can be rewritten as:27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{ij} = \left\{ {\begin{array}{*{20}c} {\sum\limits_{l = 1}^{L} {E_{l} e^{{ - {\mathrm{j}}s\frac{{{2}\pi }}{\lambda }(k_{i} - k_{j} )\sin \theta_{l} }} } {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} i \ne j} \\ {\sum\limits_{l = 1}^{L} {E_{l} } + B\sigma_{n}^{2} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} i = j{\kern 1pt} } \\ \end{array} } \right.$$\end{document} We can rewrite formula (23) as:28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{R}}_{i + n} = \sum\limits_{l = 1}^{L} {E_{l} {\mathbf{a}}_{u} (\theta_{l} ){\mathbf{a}}_{u}^{{\mathrm{H}}} (\theta_{l} )} + B\sigma_{n}^{2} {\kern 1pt} {\mathbf{I}}_{M}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{I}}_{M}$$\end{document} is the identity matrix with dimension \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M$$\end{document}, and the direction vector of the submatrix is written as29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{a}}_{u} (\theta_{l} ) = [1,e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }B\sin \theta_{0} }} , \cdot \cdot \cdot ,e^{{{\text{ - j}}s\frac{{{2}\pi }}{\lambda }(M - 1)B\sin \theta_{0} }} ]{\kern 1pt}^{{\mathrm{T}}}$$\end{document} The INCM of the virtual interference signals in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z + 1{\mathrm{st}}$$\end{document} iteration can be written as:30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{R}}_{{\theta_{s} }}^{z + 1} = {\mathbf{R}}_{{\theta_{s} }}^{z} + {\mathbf{v}}^{{\theta_{s} }} ({\mathbf{v}}^{{\theta_{s} }} )^{{\mathrm{H}}}$$\end{document} Problem solving The Lagrange function is defined by first assuming that the number of virtual interfering signals is zero:31\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{1} ({\mathbf{w}}_{u}^{z + 1} ) = ({\mathbf{w}}_{u}^{z + 1} )^{{\mathrm{H}}} {\mathbf{R}}_{i + n} {\mathbf{w}}_{u}^{z + 1} + \mu ({\mathbf{w}}_{u}^{z + 1} {\mathbf{a}}_{u} (\theta_{0} ) - 1){\kern 1pt}$$\end{document} This is obtained by taking the derivative of formula (32) with respect to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{w}}_{u}^{ * }$$\end{document} and making the derivative zero:32\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{R}}_{i + n} {\mathbf{w}}_{u}^{z + 1} + \mu {\mathbf{a}}_{u} (\theta_{0} ) = 0$$\end{document}which gives us33\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{w}}_{u}^{z + 1} = \frac{{{\mathbf{R}}_{i + n}^{ - 1} {\mathbf{a}}_{u} (\theta_{0} )}}{{{\mathbf{a}}_{u}^{{\mathrm{H}}} (\theta_{0} ){\mathbf{R}}_{i + n}^{ - 1} {\mathbf{a}}_{u} (\theta_{0} )}}$$\end{document} The beam pattern is obtained from formula (17) and searched to obtain the angle corresponding to the SLL. The Lagrange function is constructed again:34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{2} ({\mathbf{w}}_{u}^{z + 1} ) = ({\mathbf{w}}_{u}^{z + 1} )^{{\mathrm{H}}} {\mathbf{R}}_{i + n} {\mathbf{w}}_{u} + ({\mathbf{w}}_{u}^{z + 1} )^{{\mathrm{H}}} {\mathbf{R}}_{{\theta_{s} }} {\mathbf{w}}_{u}^{z + 1} + \mu ({\mathbf{w}}_{u}^{z + 1} {\mathbf{a}}_{u} (\theta_{0} ) - 1)$$\end{document} The same reasoning is available:35\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathbf{w}}_{u}^{z + 1} }} = \frac{{({\mathbf{R}}_{i + n} + {\mathbf{R}}_{{\theta_{s} }}^{z + 1} )^{ - 1} {\mathbf{a}}_{u} (\theta_{0} )}}{{{\mathbf{a}}_{u}^{{\mathrm{H}}} (\theta_{0} )({\mathbf{R}}_{i + n} + {\mathbf{R}}_{{\theta_{s} }}^{z + 1} )^{ - 1} {\mathbf{a}}_{u} (\theta_{0} )}}$$\end{document} Update the subarray-level weighting by adding a virtual signal corresponding to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta_{s}$$\end{document}. This process is looped until the SLL of the beam response converges. Methodological process In summary, the proposed method is realized as follows: Step1: the element-level weighting is performed via formula (12) so that the desired signal of array elements within the same subarray are superimposed; Step2: the INCM at the subarray-level is obtained via formula (28); Step3: the initial subarray-level weighting is obtained through formula (33); Step4: search for the corresponding angle of the virtual signal and update the INCM using formula (30); Step5: update the subarray-level weighting using the INCM obtained in step 4 through formula (35); Step6: the iteration terminates if the loop reaches the preset upper limit or if the SLL between neighboring iterations is less than a self-defined threshold, otherwise return to step4. Simulation results A ULA with 24 array elements is assumed to be divided into 8 subarrays, each containing 3 array elements. The array element noise is modeled as additive Gaussian white noise with power 1. If not specified the input SNR was 10 dB and the number of snapshots was 200. the interference to noise ratio (INR) was set to 5 dB. 100 Monte Carlo experiments were performed at a time to get more accurate results. In the simulation plots we denote the proposed method with the optimal SINR, the Dolph-Chebyshev method, the MVDR method, the DL method, the ESB method and the matrix reconstruction algorithm using ‘The proposed’, ‘Optimal’, ‘Dolph’, ‘MVDR’, ‘DL’ ‘ESB’ and ‘Reconstruction’ respectively. The loading factor in the DL method we set to 10. Beam pattern The beam pattern of the proposed method when the desired signal is incident from the normal direction is shown in Fig. 2. From the figure, it can be seen that the Optimal, DL and ESB SLL are -12.9 dB, − 12.4 dB and − 13.2 dB, respectively, and the SLL of the proposed method and the Dolph method are -15.0 dB and − 17.0 dB, respectively. The results of the beam pattern when the desired signal is incident from the normal direction and there exists an interfering signal with an incidence direction of.Fig. 2Incidence from normal: no interfering signals. Incidence from normal: no interfering signals. 10° are shown in Fig. 3. From the figure, it can be seen that the Optimal, DL, and ESB SLL are − 12.2 dB, − 12.1 dB, and − 12.3 dB, respectively, and the proposed method and the Dolph method have SLL of − 14.7 dB and − 17.0 dB, respectively. The results of the beam pattern when the desired signal is incident from the 10° direction and there exists an interfering signal with an incidence direction of 20° are shown in Fig. 4. From the figure, it can be seen that the Optimal, DL, and ESB sidelobe levels are − 12.9 dB, − 12.7 dB, and − 12.9 dB, respectively, and the SLL of the proposed method and Dolph method are − 14.6 dB and − 17.0 dB, respectively. The main lobe of Reconstruction is widened, although its SLL in the three cases are − 21.5 dB, − 20.8 dB and − 20.5 dB respectively. It can be seen from the simulation results that there is a slight degradation of the performance of the proposed method when the incoming signal is not normal to the direction of incoming beam and when there is an interfering signal. The proposed method has significantly higher sidelobe levels than the Optimal, DL and ESB methods under the above conditions, and is about 2 dB away from the limit of the Dolph method.Fig. 3Incidence from the normal: there is an interfering signal.Fig. 4No incidence from the normal: there is an interfering signal. Incidence from the normal: there is an interfering signal. No incidence from the normal: there is an interfering signal. The output SINR varies with input SNR The variation curves of output SINR with input SNR for the three cases of desired signal incident from normal, desired signal incident from normal direction with an interfering signal in 10° incident direction, and desired signal incident from 10° with an interfering signal in 20° incident direction are shown in Figs. 5, 6 and 7. It can be seen from the figure that when there is no interference signal, the output SINR of the proposed method is close to that of the Optimal, Reconstruction and the Dolph algorithm, and all change linearly with the input SNR. The output SINR of the DL algorithm almost coincides.Fig. 5Output SINR: normal incident and no interference signal.Fig. 6Output SINR: the normal is incident and there is an interference signal.Fig. 7Output SINR: no incidence from the normal and there is an interfering signal. Output SINR: normal incident and no interference signal. Output SINR: the normal is incident and there is an interference signal. Output SINR: no incidence from the normal and there is an interfering signal. with the proposed method at a low SNR, and gradually converges to 30 dB at a high SNR. The output SINR of the ESB algorithm is much lower than that of the other methods at a low SNR, and basically coincides with the proposed method at a high SNR. When the desired signal incident direction is 10°, only the output curve of the proposed method is close to that of the Optimal and Reconstruction algorithm and much higher than that of the other algorithms. Among them, the DL and ESB algorithms gradually converge to 25 dB, and the performance of the Dolph method is approximately 3 dB lower than that of the proposed method. It can be known from the simulation results that the proposed algorithm has good output SINR performance whether the desired signal is incident normally or not, and at the same time has a strong suppression ability for interference signals. The output SINR varies with the number of snapshots When the desired signal is incident from 10° and there is an interference signal with an incident direction of 20°, the variation curve of the output SINR with the number of snapshots is shown in Fig. 8. It can be seen from the figure that the convergence speeds of Dolph, Reconstruction and The proposed are all very fast. The output performance of the proposed method after convergence is close to the ideal situation.Fig. 8Output SINR varies with the number of snapshots. Output SINR varies with the number of snapshots. There are errors in the direction vector The variation of the output SINR with the desired signal angle error for the case where the desired signal is incident from 10° and there is an interfering signal with an incident direction of 20° is shown in Fig. 9. From the figure, it can be seen that the proposed algorithm has a 1 dB difference with the Optimal method in the presence of an error of about ± 2.5°. The output performance is lower than the Dolph method when the error of the desired signal DOA is greater than 4.3°. The performance of the DL and ESB methods decreases sharply in the presence of angular error in the desired signal. Same as the simulation condition in Fig. 9, the output SINR versus input SNR curve when the desired signal direction vector has a phase error that varies randomly between [0 2π] is shown in Fig. 10. From the figure, it can be seen that the proposed method decreases about 0.1 dB, and the performance of the DL and ESB algorithms is severely degraded at high input SNR. From the simulation results it is clear that the proposed method is robust to angular error and random phase error.Fig. 9Output SINR: there is an angular error.Fig. 10Output SINR: there is a random phase error. Output SINR: there is an angular error. Output SINR: there is a random phase error. Conclusion In this paper, a beam forming method with low SLL and high gain is proposed for uniform subarrays. Firstly, the desired signal of the array elements within the same subarray are superimposed by array element-level weighting and the weighted array output is derived. Secondly, an optimization problem is formulated based on the requirement of low SLL and high gain and the INCM of the optimization problem is derived. Then the Lagrange method is applied twice to solve for the weights of the signal without virtual interference and with virtual interference, respectively. Finally the subarray weights with virtual interference signals are iteratively updated until the method converges. The superiority of the proposed algorithm is verified by simulation experiments of the proposed method and the comparison method on the direction map and output performance.