Wireless Power Transfer (WPT) enables sustainable voice services in Non-Terrestrial Networks (NTN), yet it introduces a critical interdependence between information and energy. High-power charging could degrade communication quality, while high-quality semantic codecs increase energy consumption. To navigate this trade-off, we propose an online joint adaptation mechanism for NTN voice services that adjusts WPT power levels and semantic codec bitrates dynamically. We formulate an optimization objective to maximize long-term perceptual voice quality under queueing stability, energy sustainability, and receiver-side protection constraints. Our approach uses the Lyapunov optimization to decompose this stochastic problem into per-slot decision-making. These decisions remain nonlinear because the voice-quality response is unknown and safety constrained. To address this, our safe low-rank bilinear upper confidence bound algorithm, termed SR-BiUCB, integrates a conservative safe-action filter and frequent-directions sketching, decoupling the information-energy interdependence in a computationally efficient manner. We implement a multi-rate semantic codec on a live Tiantong narrow-band voice link and emulate the WPT-induced impairment. Prototype traces and trace-driven simulations show that SR-BiUCB achieves over 22% higher lower-tail perceptual voice quality than representative state-of-the-art benchmarks, while maintaining bounded reported backlogs and near-zero receiver-side protection exceedances. These results support safe, high-quality, energy-sustainable voice services in future NTN.
原文链接
分享
引用
