🤖 AI Summary
This work proposes a dual-process architecture integrating large language models (LLMs) and numerical optimization to enable policy-driven power reconfiguration in communication systems under safety constraints. The LLM acts as a high-level policy interpreter, translating natural language instructions into adjustments of channel weights and power budgets, while a fast optimizer performs constrained power allocation via projected gradient ascent with a weighted mutual information objective. Inspired by System 1/System 2 cognitive mechanisms, the framework incorporates multiple reliability safeguards—including exponential smoothing, normalization, and fallback strategies—to ensure robust operation. Experimental results in an 8-channel setting demonstrate the system’s ability to flexibly satisfy diverse policy requirements and autonomously reconfigure upon abrupt channel changes, reducing the dispersion of mutual information distribution by 60%.
📝 Abstract
Large language models (LLMs) are increasingly being explored as high-level decision modules in closed-loop systems, but their stochastic nature makes safe integration challenging. In this paper, we propose LLM-Steered Power Allocation, a dual-process architecture for parallel QPSK channels inspired by Kahneman's System 1/System 2 framework. A fast numerical optimizer (System 1) continuously performs projected gradient ascent on a weighted mutual-information objective, while an LLM navigator (System 2) periodically interprets natural-language policies and updates only the channel weights and the operational power budget. The LLM never manipulates the power-allocation variables directly, and constraint satisfaction is enforced structurally by the optimizer. To mitigate LLM unreliability, we further incorporate multi-layer guardrails including normalization, exponential moving-average smoothing, and fallback mechanisms. Numerical experiments on an 8-channel system show that, with a fixed optimization core and unchanged system prompt, different natural-language policies induce qualitatively different operating points, including throughput-oriented allocation, channel prioritization, power-aware operation, and channel shutdown. In addition, under an abrupt channel-gain reversal, the proposed system autonomously reconfigures its steering signals and reduces the final mutual-information spread by 60% compared with the optimizer alone. These results suggest that LLMs can serve as policy interpreters for safe, flexible reconfiguration of communication-system optimizers without controller reimplementation.