An enhanced genetic algorithm with ant colony optimization for phase shift design in RIS-assisted MIMO networks

Abstract Reconfigurable intelligent surfaces (RISs) have emerged as a promising technology for seamless integration with future sixth-generation (6G) wireless systems. However, their performance critically depends on the optimization of numerous phase shifts across RIS elements, making efficient phase shift design crucial. Existing optimization methods often face limitations, including unbalanced exploration and exploitation, slow convergence, suboptimal channel gain, high computational overhead, poor scalability, and reduced robustness. Among them, the Genetic Algorithm (GA) is a widely used meta-heuristic for RIS phase shift optimization, valued for its strong exploration capabilities in complex, constrained, non-convex problems. Nevertheless, GA suffers from limited exploitation and high sensitivity to control parameters, leading to performance degradation. To address these limitations, this paper proposes an enhanced GA that integrates pheromone-guided selection and heuristic-driven recombination—adapted from ant colony optimization (ACO)—into the GA’s crossover operation only. This novel lightweight hybridization strengthens exploitation without compromising exploration. A new entropy-based mathematical framework is developed to rigorously analyze the exploration–exploitation trade-off. Simulations are performed to assess the enhanced GA while comparing it with the standard GA and another popular optimizer, the particle swarm optimizer (PSO). The results demonstrate that the enhanced GA consistently outperforms both the standard GA and PSO. In small-scale system, its exploration–exploitation balance enables it to reach the 95% balance point in only 6 iterations, compared to 66 and 43 for the original GA and PSO, respectively. It also converges faster with 7 against 62 and 43, achieving up to 1.6 dB higher channel gain. With a minimum runtime of 55 ms, it is more computationally efficient than the original GA (140 ms) and PSO (95 ms), reducing runtime by up to 2.6 times. Its scalability is evident, as its channel gain advantage grows from 1.6 dB to 2.4 dB with increasing network size. Furthermore, the enhanced GA demonstrates superior robustness under algorithmic parameter variations and channel estimation errors.