A Large-Scale Energy Efficient Time-Domain Pilot Muting Mechanism for Non-Stationary Channels: A Deep Reinforcement Learning Approach

To address rapid fluctuations in non-stationary channel environments, adaptive pilot patterns are commonly used. However, frequent pilot pattern changes require an interactive mechanism for terminal communication. In this letter, we introduce a time-domain pilot muting mechanism (TDPMM) based on the densest pilot pattern, combined with power optimization using deep reinforcement learning. This approach aims to mitigate issues associated with frequent pilot pattern adjustments in non-stationary channel fading environments. We first formulate an energy efficiency (EE) optimization model that balances normalized mean square error (NMSE) and energy consumption (EC) for large-scale continuous resource blocks (RBs) using TDPMM. Then, we propose a deep Q-network (DQN) based learning strategy tailored to optimize TDPMM selection and the power ratio between pilot and data signals. Simulation experiments confirm the superior EE performance of the proposed TDPMM. Furthermore, our DQN-based approach demonstrates lower complexity and only slightly inferior performance compared to the exhaustive search strategy.