Synthetic cellular network modeling via public data and AI-enhanced KPI modeling

Official URL: https://dx.doi.org/10.1109/WCNC65185.2026.11555132

Cellular network design, optimization, and performance benchmarking have traditionally depended on proprietary datasets collected by mobile network operators (MNOs), constraining independent research and limiting replicability. This paper introduces an operator-agnostic modeling framework for synthetic cellular network planning and Key Performance Indicator (KPI) generation using openly accessible geospatial and demographic datasets. The framework integrates land use/land cover (LULC) data, population density grids, and publicly available LTE drive-test traces to characterize radio environments without operator intervention. A hybrid statistical-generative modeling pipeline is developed, combining geo-statistical clustering with deep generative AI models to synthesize realistic traffic loads, interference patterns, and KPI distributions across heterogeneous regions. Frequency, bandwidth, and deployment parameters are user-configurable, enabling scenario generalization and transferability across geographic domains. Furthermore, a domain-adaptive learning module enables the trained model to extrapolate KPI behavior in previously unseen environments, achieving operator-level fidelity in coverage, throughput, and quality-of-service metrics. The proposed approach establishes a data-independent methodology for cellular network emulation, promoting reproducibility, cost-efficiency, and open benchmarking. This work highlights how AI-driven surrogate modeling and open geospatial intelligence can democratize access to network performance analytics, accelerating innovation in wireless communication research, digital infrastructure planning, and policymaking.