Scenario-Aware Multi-Objective Bayesian Optimization of Antenna Placement in Dual-Band LTE/5G Distributed Antenna Systems

Distributed antenna layouts optimized for nominal demand may deteriorate under changing hotspots and correlated shadowing. We present SA-MOBO-DAS, a scenario-aware multi-objective Bayesian optimization framework for placing four shared LTE/5G remote antenna units in a synthetic 500 × 500 m urban area. Its contribution is a placement and validation protocol combining worst-band coverage and throughput objectives, frozen operating scenarios, a fixed evaluation budget, and independent post-search risk assessment. A reproducible simulator incorporates 3GPP TR 38.901-inspired path loss, obstruction footprints, correlated shadowing, strongest power association, full-frequency reuse, and load-aware throughput at 2.1 GHz LTE and 3.5 GHz NR. Independent Matérn-5/2 Gaussian processes and a finite-pool approximation of q-noisy expected hypervolume improvement guide an 80-evaluation search. Thirty paired runs used 64 development and 160 held-out scenarios. SA-MOBO-DAS achieved held-out lower-tail hypervolume of 0.006619 (SD 0.000822), with no statistically detected difference from ParEGO (+0.7%; adjusted p = 0.968). Hypervolume exceeded NSGA-II and Sobol by 27.2% and 24.4%, respectively (both adjusted p < 0.001). Direct CVaR optimization yielded 7.6% and 0.9% higher point estimates of stress tail coverage and throughput, respectively, but reduced overall hypervolume by 10.6%. Paired 95% confidence intervals for both stress tail differences included zero.