# Chaos Fusion Mutation-Based Weighted Mean of Vectors Algorithm for Linear Antenna Array Optimization

## TL;DR

This paper introduces a new optimization algorithm for linear antenna arrays that achieves better performance in suppressing interference and reducing signal noise.

## Contribution

The novel algorithm combines chaos-based parameterization and cloud mutation for improved antenna array optimization.

## Key findings

- CFMINFO achieves an SLL of −32.30 dB and a −125.1 dB deep null at 104° for interference suppression.
- The algorithm outperforms PSO, GA, IWO, HSA, and FPA in constrained optimization tasks with a Friedman rank of ≈1.36.
- It demonstrates faster and more stable convergence across five simulation scenarios and CEC2020 problems.

## Abstract

What are the main findings?

CFMINFO is a weighted-mean optimizer with good-lattice initialization, STC chaos, and cloud mutation, designed for constrained array synthesis.

It optimizes both element spacings and amplitudes, ensuring prescribed deep-null steering.

What is the implication of the main finding?

CFMINFO achieves SLL ≈ −32.30 dB and a −125.1 dB deep null at 104°, while preserving the main lobe for effective interference suppression.

It outperforms PSO/GA/IWO/HSA/FPA, demonstrating the best Friedman rank ≈ 1.36 on 7 CEC2020 constrained optimization tasks.

This study proposes the Chaos Fusion Mutation-Based Weighted Mean of Vectors Algorithm, an advanced optimization technique within the weighted mean of vectors (INFO) framework for synthesizing unequally spaced linear arrays. The proposed algorithm incorporates three complementary mechanisms: a good-point-set initialization to enhance early population coverage, a sine–tent–cosine (STC) chaos–based adaptive parameterization to balance exploration and exploitation, and a normal-cloud mutation to preserve diversity and prevent premature convergence. Array-factor (AF) optimization is posed as a constrained problem, simultaneously minimizing sidelobe level (SLL) and achieving deep-null steering, with penalties applied to enforce geometric and engineering constraints. Across diverse array-synthesis tasks, the proposed algorithm consistently attains lower peak SLLs and more accurate nulls, with faster and more stable convergence than benchmark metaheuristics. Across five simulation scenarios, it demonstrates robust superiority, notably surpassing an enhanced IWO in the combined objectives of deep-null suppression and maximum SLL reduction. In a representative engineering example, we obtain an SLL and a deep null of approximately −32.30 and −125.1 dB, respectively, at 104°. Evaluation of the CEC2020 real-world constrained problems confirms robust convergence and competitive statistical ranking. For reproducibility, all data and code are publicly accessible, as detailed in the Data Availability section.