ETD

• Banach
• beamforming
• cell-free
• contraction-factor
• convergence
• fiex-point
• interference
• massivemimo
• neural-network
• optimization
• power
• transformer

The thesis investigates joint beamforming and power-control optimization in user-centric cell-free massive MIMO systems, focusing on the convergence behavior of iterative algorithms and on accelerating it through a neural network–based initialization. After implementing the Team-MMSE iterative method, its convergence speed is analyzed across different scenarios, showing how interference, user density, and the ratio between antennas and terminals critically affect the contraction factor and the required number of iterations. To reduce computational cost, a Transformer-based neural network is designed to predict an initial power vector from the spatial configuration of users and access points. Although the prediction is not fully accurate, especially for high-power values, it still leads to a moderate decrease in iterations, confirming the validity of the approach and suggesting that further improvements could be achieved with more balanced datasets and more advanced learning architectures.