Abstract: The rapid deployment of 5G networks has introduced unprecedented complexity in Radio Access Network (RAN) optimization, particularly in antenna tilt management, where the choice of algorithm significantly impacts performance. Traditional fixed tilt settings (e.g., 9◦) applied uniformly across cells fail to adapt to spatiotemporal variations in user density and traffic, resulting in 30–50% performance deficits compared to optimal configurations, with edge-heavy scenarios suffering over 35% capacity degradation. While recent AI-driven methods improve adaptability, they are limited to a single optimization paradigm, typically deep reinforcement learning or heuristic bargaining, missing scenario-specific algorithmic strengths.

This seminar presents an agentic AI-driven meta-selector that dynamically chooses the optimal algorithm from a heterogeneous portfolio: Fixed Tilt, Nash Bargaining Solution (NBS), Weighted NBS (WNBS), Kalai-Smorodinsky (KSBS), Shannon Entropy (SEBS), and Deep Q-Network (DQN). In multi-cell environments, it maintains 31.80% gain over Fixed and near-perfect fairness (JFI = 1.000). Under extreme loads (10–300 users/cell), the AI agent sustains 2–8% better performance than individual algorithms. The end-to-end pipeline executes in low latency, integrates with O-RAN RIC interfaces, and reduces manual configuration effort significantly, enabling fully autonomous, scenario-aware optimization.

Event Details
Title: Agentic AI-Driven Dynamic Algorithm Selection for 5G Antenna Tilt Optimization
Date: March 27, 2026 at 3:00 PM
Venue: ESB 244
Speaker: Mr. Uma Kishore (EE15D013)
Guide: Dr. Giridhar K
Type: PHD seminar

Updated: