Answers are grounded in this paper and cite their sources.
The number of codebook beams in a millimeter-wave cellular sector directly affects beamforming gain, inter-sector interference, beam-management overhead, and user fairness, so a single fixed beam count is inefficient under non-uniform traffic.Improving one of these quantities alone is simple but usually degrades the others.This paper therefore optimizes five objectives of equal importance: average effective throughput, fifth-percentile SINR, Jain's fairness index, beam-management overhead, and inter-sector interference.In a network of seven base stations and twenty-one sectors, the beam count of each sector is selected independently from a discrete set using NSGA-III and the Multi-Objective Whale Optimization Algorithm (MOWOA), both with a bounded working Pareto archive and an unbounded historical archive; an equal-weight figure of merit (FoM) selects the compromise solution, which is evaluated on 200 Monte Carlo snapshots.To the best of the authors' knowledge, no earlier study has optimized the beam count of every sector under such an equal-importance five-objective criterion.Over 20 independent runs, NSGA-III ranks first among 15 candidates (nine homogeneous beam counts, two load-aware heuristics, NSGA-III, MOWOA, and two Q-learning policies) in 19 runs (mean pooled FoM 0.8467, 95% confidence interval 0.8425-0.8510),and MOWOA ranks in the top three in 19 runs.NSGA-III attains a significantly higher fairness index than MOWOA (0.727 versus 0.710) and a significantly smaller IGD+ (0.014 versus 0.027) with no significant difference in hypervolume, whereas the differences in average throughput are not significant after Holm correction.No fixed beam count is balanced: the fixed fifteen-beam configuration has the highest average throughput (28.91 Mbps/user) but 11.3% overhead, and the fixed eight-beam configuration has 5.6% overhead but lower throughput and fairness.The optimized allocations reach neither single-metric optimum; their advantage lies in the balance among the five objectives.A sensitivity analysis over 27 perturbations of the beam-selection error, overhead, traffic-activity, and propagation parameters shows that NSGA-III ranks first in 25 scenarios without re-optimization, and re-optimization restores first place in the two remaining beam-error scenarios.A re-implementation of a 2024 reinforcement-learning beam-management method ranks below NSGA-III in all 20 runs (median rank sixth with a shared five-objective reward and fourteenth with a throughput-only reward).
No takes yet. Share an insight, caveat, or question.
Hamdy et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: