AI-Driven Framework for 6G Radio Access Network (RAN) Planning and Dimensioning: Theory and System-Level Simulation

This paper proposes a unified Radio Access Network (RAN) planning and dimensioning framework specifically designed for the heterogeneous requirements of sixth-generation (6G) systems. Addressing the limitations of static fifth-generation (5G) planning tools, the proposed methodology mathematically integrates terrestrial, aerial, and non-terrestrial (NTN) domains into a closed-loop optimization engine. The framework couples physically consistent multi-band propagation models (sub-6 GHz to THz) with an Artificial Intelligence (AI)-driven solver that utilizes Pseudo-Spatio-Temporal Residual Networks (PST-ResNet) for traffic prediction and a Self-Coordinated Dynamic Swarm Control System (SC-DSCS) for resource allocation. Unlike generic architectural surveys, this work explicitly formulates the dimensioning problem to jointly optimize site placement, Reconfigurable Intelligent Surface (RIS) phase-shifts, and Fluid Antenna System (FAS) configurations under strict coverage and latency constraints. Validated through extensive system-level simulations across Urban Ultra-Dense, Industrial Internet of Things (IoT), and Smart City scenarios, the results demonstrate that the proposed AI-driven framework achieves a 22% improvement in coverage probability and up to 30% gains in energy efficiency compared to static 5G baselines. The study provides quantitative design guidelines for deploying sustainable, latency-aware 6G infrastructures, bridging the gap between theoretical channel models and practical network dimensioning.