Multi-objective genetic array pattern optimization via dynamic subarray partitioning and sidelobe clustering

Abstract To address the insufficient flexibility and limited sidelobe/grating lobe suppression of traditional weighting methods in array signal processing, a novel array pattern optimization approach is proposed. This method integrates dynamic subarray partitioning and sidelobe clustering with genetic algorithm (GA), hereafter designated as dynamic clustering with K-means-based genetic algorithm (DyCK-GA). First, a greedy algorithm is employed to implement dynamic subarray partitioning, ensuring subarray spatial diversity. Second, dynamic weights are designed based on K-Means sidelobe clustering, and a multi-objective fitness function is constructed that incorporates mainlobe constraints in terms of position and width, sidelobe and grating lobe suppression, phase smoothness, and null preservation. Finally, a real-coded GA is utilized for global optimization via tournament selection, blend crossover, and Gaussian mutation operators. Comparative experiments are performed against fixed arrays and particle swarm optimization methods. The results show that the proposed DyCK-GA achieves a grating lobe peak of -3. 73 - 3. 73 dB and an average sidelobe level of -18. 30 - 18. 30 dB. For comparison, the improved PSO obtains a grating lobe peak of -3. 61 - 3. 61 dB and an average sidelobe level of -15. 94 - 15. 94 dB. It delivers superior pattern performance under multiple constraints, providing an effective solution for array optimization in radar, communication, and various relevant domains.