Evolutionary–Neural Hybrids for Interference-Aware Channel Assignment in Ultra-Dense 6G Networks: A Survey, Taxonomy, and Research Roadmap

Abstract

Ultra-dense networks (UDNs) are a defining feature of 6G: thousands of small cells and devices share a limited spectrum, so co-channel interference rather than noise limits performance. Assigning channels to cells or users in such networks is a combinatorial, NP-hard problem whose search space grows exponentially with network size. Genetic algorithms (GAs) offer powerful global search, but they repeatedly evaluate an expensive interference-based fitness function and react slowly to changing traffic. Neural networks (NNs) offer millisecond-scale inference and can learn traffic and interference patterns, but they need large training sets and give no guarantee that constraints are satisfied. Combining the two is a natural but still fast-moving research direction. This survey (i) formalises the interference-aware channel assignment problem, (ii) proposes a taxonomy of five GA–NN hybridisation patterns (surrogate-assisted, predict-then-evolve, neuroevolution, GA-labelled or seeded NN, and NN-guided operators), (iii) reviews more than 40 works spanning classical and evolutionary channel/spectrum assignment, neural and graphbased channel allocation, deep reinforcement learning, and evolutionary-computation foundations, drawing its evidence base specifically from work published between 2015 and 2025 so that the survey reflects the current, AI-native direction of 6G research, and (iv) synthesises nine research gaps, a research problem statement, and a reference framework for future work. We conclude that scalable, constraintaware, multi-objective hybrid GA–NN methods evaluated on common benchmarks remain largely open for 6G UDNs, even as the first dedicated GA–DRL hybrids for 6G spectrum sharing and resource allocation.