Bio-inspired and hybrid evolutionary optimization for robust resource allocation in imperfect-CSI wireless networks: a narrative review of algorithms, applications, and real-world implementation challenges

The evolution of fifth-generation advanced (5G-Advanced) and emerging sixth-generation (6G) wireless networks has intensified the need for robust resource allocation techniques capable of operating under imperfect channel state information (CSI). Traditional convex optimization methods, although computationally efficient, often degrade under CSI mismatch and struggle with the non-linearity, uncertainty, and multi-objective nature of modern wireless environments. Bio-inspired evolutionary algorithms (EAs), including swarm intelligence, evolutionary computation, and nature-mimetic metaheuristics, provide strong global search capability, adaptability, and robustness to modeling uncertainty, making them well suited for imperfect-CSI optimization problems. This narrative review presents a systematic synthesis of bio-inspired and hybrid evolutionary optimization frameworks for wireless systems affected by estimation errors, quantization noise, channel aging, mobility, and hardware impairments. A transparent review methodology and a unified taxonomy are introduced to classify algorithmic principles, hybridization strategies, and application domains spanning non-orthogonal multiple access (NOMA), multiple-input multiple-output (MIMO) systems, reconfigurable intelligent surface (RIS)-assisted communications, unmanned aerial vehicle (UAV) networks, millimetre-wave and terahertz links, and edge-enabled architectures. Quantitative benchmarking trends from the literature are consolidated to compare evolutionary, hybrid, and convex optimization approaches in terms of spectral efficiency, fairness, outage probability, convergence behavior, and computational complexity. The review further clarifies the complementary roles of evolutionary algorithms and learning-based methods, positioning EAs for offline or slow-timescale global optimization and machine learning or deep reinforcement learning for fast online decision-making. Comparisons with end-to-end deep learning approaches are provided, highlighting trade-offs in interpretability, training cost, robustness, and deployment feasibility. Finally, environment-aware and sustainable optimization perspectives are discussed, advocating holistic metrics such as Joules per delivered bit including optimization overhead, and identifying open research challenges for scalable, intelligent, and hardware-aware resource allocation in next-generation wireless systems.