Self‐Healing Swarm Beamforming for LEO Satellite Constellations

ABSTRACT Low Earth Orbit (LEO) satellite constellations must be able to communicate reliably, strongly, and with little energy use in order to meet 5G/6G performance goals. This paper introduces Self‐Healing Swarm Beamforming (SHSB), a new way to do distributed beamforming. SHSB lets adaptive digital beamforming happen across the whole constellation by using federated deep reinforcement learning (FDRL) and topology‐aware dynamic antenna arrays. SHSB creates a virtual massive multiple‐input multiple‐output (MIMO) array by treating satellites as a cooperative swarm. This increases signal‐to‐interference‐plus‐noise ratio (SINR), lowers energy use, and improves spatial coverage. Reconfigurable intelligent surfaces (RIS) make beam directivity even better at 28 GHz. In case of a satellite failure, a predictive self‐healing mechanism reallocates beams within five seconds. It uses graph neural networks (GNNs) to predict topology, which fills in gaps in the previous FDRL‐RIS methods that focused on energy but did not have resilient beamforming. SHSB reduces energy consumption by 25%, delivers SINR  18 dB at 16‐dB SNR, and achieves a spectral efficiency of 120 bits/s/Hz at 20‐dB SNR, according to MATLAB simulations. The feasibility is confirmed by theoretical limits on FDRL convergence and system stability. For next‐generation satellite networks, SHSB provides a scalable, autonomous, and high‐capacity solution with direct applications in disaster recovery, Internet of Things (IoT) connectivity, and international broadband services. This study proposes Self‐Healing Swarm Beamforming (SHSB), a distributed framework integrating federated deep reinforcement learning (FDRL), reconfigurable intelligent surfaces (RIS), and graph neural network (GNNs) for resilient Low Earth Orbit (LEO) satellite communications. This work enables predictive self‐healing with beam reallocation in less than 5 s upon satellite failure, ensuring constellation‐wide topology adaptation. This research forms virtual massive multiple‐input multiple‐output (MIMO) arrays via swarm coordination, achieving SINR >18 dB at 16‐dB SNR and spectral efficiency of 120 bits/s/Hz at 20‐dB SNR. This work demonstrates 25% energy reduction compared to centralized DBF baselines through MATLAB simulations with 100 Monte Carlo runs under Ricean fading. This study provides theoretical convergence bounds for FDRL and stability proofs, affirming scalability for 5G/6G applications in disaster recovery and Internet of Things (IoT).