研究目的
To propose a deep reinforcement learning based baseband unit aggregation policy that guarantees users’ quality of service while keeping BBU pool energy-efficient.
研究成果
The proposed DRL-based BBU aggregation policy achieves 34% to 80% lower migration traffic with only a small cost of increased power consumption, demonstrating its effectiveness in realizing low RRH migration and energy-efficient CRAN systems.
研究不足
The study assumes traffic changes in a larger time scale (e.g., order of hours), which may not capture real-time traffic variations. Additionally, the simulation is based on synthetically generated RRH requests, which may not fully represent real-world scenarios.
1：Experimental Design and Method Selection:
The study employs a deep reinforcement learning (DRL) method to solve the BBU aggregation problem, focusing on minimizing the number of active BBUs and maintaining user’s QoS by keeping RRHs unchanged with its associated BBU.
2：Sample Selection and Data Sources:
RRHs are deployed in residential and business areas with different traffic patterns. Traffic loads change across time periods, and BBU-RRH re-association is done once per time period.
3：List of Experimental Equipment and Materials:
The simulation assumes RRHs are deployed in macro, micro, and pico cells with different modulation formats. BBUs have specific IQ-BW and BP properties.
4：Experimental Procedures and Operational Workflow:
The policy network is trained in an episodic manner, simulating L episodes for all M RRHs in T time periods to explore the probabilistic space of possible selection of BBUs.
5：Data Analysis Methods:
The performance of the RL policy is compared with two benchmark heuristics, Load Reallocation (LR) and Load Adjustment (LA) policies, in terms of power consumption and migrated traffic load.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
