Cooperative Deep Learning Strategy for Green Traffic Offloading in Heterogeneous Ultra Dense Networks

In the 6G era, energy efficiency and varying network traffic are a crucial concern because the power consumption of macro base stations (MBSs) is directly related to their traffic load and is comparatively high. In this regard, the emergence of ultra dense networks (UDNs) has revolutionized wireless communication by densely deploying small base stations (SBSs) to offload traffic from macro base stations (MBSs) while maintaining quality of service (QoS) for macro user equipments (MUEs). Nevertheless, traffic offloading causes network congestion, interference, and higher aggregate power consumption at the SBS, presenting significant challenges to the offloading process. This paper proposes a flexible traffic offloading scheme that utilizes system load information from both MBSs and SBSs to devise a green traffic offloading method, determine the optimal operational state for SBSs, and ensure the QoS for MUEs. In this direction, a multi-agent cooperative double deep Q-network is proposed, incorporating a policy learning technique based on value decomposition network to enhance long-term system performance. Numerical findings prove that the efficacy of the suggested method, in terms of energy efficiency and QoS for MUEs, is superior to existing methods.