QPPG: Quantum-Preconditioned Policy Gradient for Link Adaptation in Rayleigh Fading Channels

Reliable link adaptation is critical for 6G wireless communications in dynamic fading environments. However, classical reinforcement learning (RL) often suffers from unstable convergence due to the Euclidean geometry of the standard gradient updates, which ignores the curvature of the policy parameter space. We propose the quantum-preconditioned policy gradient (QPPG), a geometric optimization framework that approximates the quantum natural gradient (QNG) for classical link adaptation. We stress that the prefixquantumrefers solely to the Fubini-Study metric and quantum-tomography-inspired regularization from which the preconditioner is derived, not to quantum hardware or exponential speedups. Evaluations across five network scenarios demonstrate that QPPG bridges the gap between quantum geometry and classical engineering. It achieves a 28.6% gain in throughput and a 43.8% decrease in transmit power compared to classical baselines, while exhibiting superior robustness to channel estimation errors and noise uncertainty.