用于物联网设备中可靠性受限的节能物理下行控制信道监测的专家混合深度强化学习

中文摘要 AI 辅助

在第五代（5G）物联网设备（IoT-D）中，持续监测物理下行控制信道（PDCCH）是能源消耗的主要来源。即使没有有效的调度授权，用户设备（UE）也必须盲目检测下行控制信息。预测性动态功率管理虽能减少不必要的接收活动，但激进睡眠可能导致授权丢失并降低接收可靠性。本文将UE端的PDCCH监测问题表述为可靠性受限的长期能量最小化问题。具体而言，IoT-D在观察实际调度结果之前，决定是否监测PDCCH或切换接收器链到低功耗状态。目标是最小化长期平均能耗，包括接收器运行能耗、组件切换能耗和预测相关计算能耗，同时确保调度授权检测的误报率保持在规定阈值以下。由于授权到达的突发性和时间相关性以及长期可靠性约束耦合的二元监测决策，该问题是非凸的。为解决此问题，我们提出了一种基于专家混合输入输出隐马尔可夫模型（MoE-IOHMM）的预测监测方案，其中多个IO-HMM专家捕获异构授权到达模式，门控网络自适应地组合它们的预测。仿真结果表明，与始终开启的PDCCH监测和传统预测基线相比，该方案有效降低了IoT-D端的能耗，同时将误报率保持在规定的可靠性阈值以下。

英文摘要

The continuous monitoring of the physical downlink control channel (PDCCH) is a major source of energy consumption in fifth-generation (5G) Internet of thing device (IoT-D), since the UE has to blindly detect downlink control information even when no valid scheduling grant is present. Although predictive dynamic power management can reduce unnecessary receiver activity by skipping PDCCH monitoring in grant-free slots, aggressive sleeping may lead to missed grants and degrade reception reliability. To address this tradeoff, this paper formulates UE-side PDCCH monitoring as a reliability-constrained long-term energy minimization problem. Specifically, the IoT-D determines, before observing the actual scheduling outcome, whether to monitor the PDCCH or switch the receiver chain into a low-power state. The objective is to minimize the long-term average energy consumption, including receiver operating energy, component switching energy, and prediction-related computational energy, while ensuring that the false negative rate of scheduling-grant detection remains below a prescribed threshold. The resulting problem is non-convex due to the bursty and temporally correlated nature of grant arrivals, and the binary monitoring decisions coupled by a long-term reliability constraint. To solve this problem, we propose a mixture-of-experts input-output hidden Markov model (MoE-IOHMM)-based predictive monitoring scheme, where multiple IO-HMM experts capture heterogeneous grant-arrival patterns and a gating network adaptively combines their predictions. Simulation results show that the proposed scheme effectively reduces IoT-D-side energy consumption compared with always-on PDCCH monitoring and conventional predictive baselines, while maintaining the false negative rate below the prescribed reliability threshold.

发表机构

• National Key Laboratory of Science and Technology on Communications, University of Electronic Science and Technology of China（电子科技大学通信抗干扰技术国家级重点实验室）
• China Mobile Internet of Things Co., Ltd.（中国移动物联网有限公司）
• Xinyi Information Technology (Shanghai) Co., Ltd.（新易信息技术（上海）有限公司）

机构由 AI 辅助整理，请以论文原文为准。