Combined Multi-Agent and Centralized Resource Allocation in Cloud Radio Access Networks (C-RANs)

Abstract

The rapid growth of fifth-generation (5G) mobile networks demands architectures capable of supporting massive connectivity, ultra-low latency, and high data throughput. The Cloud Radio Access Network (C-RAN) addresses these demands by centralizing Baseband Units (BBUs) in a virtualized cloud pool while distributing Remote Radio Heads (RRHs) across the network edge. Despite its advantages, C-RAN faces a persistent challenge in allocating limited computational and network resources—CPU, memory, and storage—to dynamically fluctuating user demand. Conventional fixed and random allocation strategies cannot adapt to real-time traffic variation, resulting in high unmet demand and inefficient resource utilization. This paper proposes a hybrid resource-allocation framework that combines centralized decision-making with a Multi-Agent System (MAS), in which Virtual Base Stations (VBSs) act as autonomous agents reporting real-time demand to a centralized BBU controller that retains historical usage patterns. The base hybrid model is further extended using a Multi-Objective Genetic Algorithm (GA) that simultaneously optimizes resource utilization, fairness, and unmet-demand minimization. The proposed system is implemented as a simulation prototype and evaluated against the conventional Fixed Priority-Based Resource Allocation technique. Experimental results demonstrate that the hybrid and genetically-optimized models substantially reduce unmet demand, improve resource-utilization efficiency, and enhance fairness among competing VBS agents compared with the baseline, confirming the suitability of the proposed method for nextgeneration cloud-based mobile network deployments.