Smart Helper-Aided F-RANs: Improving Delay and Reducing Fronthaul Load

TL;DR

The paper tackles high fronthaul load and latency in dense Fog-RANs by introducing smart helpers (SHs) that overhear nearby eRRH transmissions and cache popular content without fronthaul connections. It develops a two-time-scale framework combining MWIS-based user assignment/power allocation with dual MARL-based cache updates for eRRHs and SHs, aiming to minimize the average content delivery delay. The authors prove convergence properties, analyze computational complexity, and demonstrate via simulations that SHs substantially reduce fronthaul load and delivery delay, with caching policies outperforming several baselines and even allowing SHs to substitute for some eRRHs. The findings highlight SH-aided F-RAN as a practical, scalable approach for improving QoS in high-demand scenarios, including urban events and remote deployments. Overall, the work provides a proactive caching-and-delivery strategy that leverages overheard communications to optimize edge caching and resource allocation in F-RANs.

Abstract

In traditional Fog-Radio Access Networks (F-RANs), enhanced remote radio heads (eRRHs) are connected to a macro base station (MBS) through fronthaul links. Deploying a massive number of eRRHs is not always feasible due to site constraints and the cost of fronthaul links. This paper introduces an innovative concept of using smart helpers (SHs) in F-RANs. These SHs do not require fronthaul links and listen to the nearby eRRHs' communications. Then, they smartly select and cache popular content. This capability enables SHs to serve users with frequent on-demand service requests potentially. As such, network operators have the flexibility to easily deploy SHs in various scenarios, such as dense urban areas and temporary public events, to expand their F-RANs and improve the quality of service (QoS). To study the performance of the proposed SH-aided F-RAN, we formulate an optimization problem of minimizing the average transmission delay that jointly optimizes cache resources and user scheduling. To tackle the formulated problem, we develop an innovative multi-stage algorithm that uses a reinforcement learning (RL) framework. Various performance measures, e.g., the average transmission delay, fronthaul load, and cache hit rate of the proposed SH-aided F-RAN are evaluated numerically and compared with those of traditional F-RANs.

Figures (8)

• Figure 1: SH-aided F-RAN architecture.
• Figure 2: Two-time scale cache optimization and transmission management time frame.
• Figure 3: Content delivery process.
• Figure 4: Average reward of agents of one SH and one CE-Relay at each training iteration for different values of the Zipf's parameter $\gamma$.
• Figure 5: Total cache hit rate using the proposed algorithm for $S = 4$ with various number of SHs.