OICC Press Majlesi Journal of Electrical Engineering 2345-3796 2026 11 06 Optimizing Network Performance in Mobile Data Offloading Using Cooperative IEEE 802.11 MAC Protocol and Deep Reinforcement Learning 10.57647/j.mjee.2025.17785 EN Nabeel Abdolrazagh Yaseen Alrashedi Institute of Artificial Intelligence and Social and Advanced Technologies, Isf.C., Islamic Azad University, Isfahan, Iran https://orcid.org/0000-0002-9872-9304 Rasool Sadeghi Department of Electrical Engineering, DOL.C., Islamic Azad University, Isfahan, Iran https://orcid.org/0000-0003-1353-2789 Wael Hussein Zayer Al-Lamy Department of Electronic Engineering, Southern Technical University, Amara, Iraq https://orcid.org/0000-0002-3796-0393 Mehdi Hamidkhani Institute of Artificial Intelligence and Social and Advanced Technologies, Isf.C., Islamic Azad University, Isfahan, Iran https://orcid.org/0000-0002-8041-7337 Reihaneh Khorsand Institute of Artificial Intelligence and Social and Advanced Technologies, Isf.C., Islamic Azad University, Isfahan, Iran Journal Article 2026 11 06 With the exponential growth of mobile data traffic and the increasing demand for latency-sensitive applications, wireless networks face critical resource management challenges. In particular, heterogeneous architectures such as LTE and Wi-Fi necessitate intelligent and adaptive offloading mechanisms to ensure efficiency and Quality of Service (QoS). This paper exploits Deep Reinforcement Learning (DRL) and cooperative communication within the MAC layer of the IEEE 802.11n standard to design an adaptive and scalable framework for data offloading in wireless networks. The proposed model employs the Partially Observable Markov Decision Process (POMDP) structure to optimally select the communication path and the type of communication (direct or multi-hop) in real-time. The proposed learning algorithm, which leverages the Deep Q-Learning structure and the Policy Improvement mechanism, demonstrates significant performance gains compared to conventional methods. Simulation results indicate that the average cumulative reward achieved by the DRL algorithm is 3.4, a substantial improvement in decision-making effectiveness compared to 2.2 for Q-Learning and 0.7 for Heuristic methods. Moreover, the energy efficiency of the proposed method improved by 87% compared to CoopMAC, and the throughput increased by 18%. These results establish the proposed framework as an effective and low-power solution for implementation in Fifth-Generation (5G) network architectures.