AIXhaul: 深度强化学习赋能的星地 Xhaul 流量工程与神经波束成形调度

AIXhaul Deep Reinforcement Learning Cross-Domain Routing Pipeline

Continuous closed-loop telemetry feedback tuning physical RF beams and optical crosslinks

Central System: Network Telemetry Collector - Samples SNR, packet loss, and buffer occupancy at 1 kHz

Network Nodes

• Network Telemetry Collector
• Deep Q-Network / PPO Agent
• Neural Beamforming Controller
• P4 Programmable Hardware Switch

Architectural Protocol Layers

• Spaceborne PHY/MAC Layer (3GPP Rel-17 TS 38.300): Doppler compensation, HARQ disabling, and timing advance extension.
• Orbital Mesh Routing (CCSDS / Delay-Tolerant Networking): Dynamic topology routing over intermittent laser crosslinks.
• Regenerative Core Processing (OpenRAN / 3GPP CU-DU Split): Onboard user plane function (UPF) local breakout.

Key Metrics: < 18ms (LEO Orbit-to-Ground Roundtrip) | 100 Gbps (Optical Crosslink Throughput) | ±48 kHz (Doppler Shift Compensation) | 99.999% (Orbital Constellation Availability)