Sixth-generation (6G) wireless networks have the potential to offer several emerging technologies that require efficient resource allocation. However, traditional resource allocation methods have limited performance in mitigating interference, which hinders scalability and compromises joint optimization of spectral, energy, and computational efficiency. This motivates the implementation of deep reinforcement learning (DRL) techniques for resource management in device-to-device (D2D)-enabled 6G systems towards self-sustainable networks (SSNs). Hence, this paper presents a review of DRL algorithms for resource management in D2D networks. The state-of-the-art DRL algorithms, including value-based, policy-based, and hybrid methods, are reviewed. It highlights their strengths and limitations for resource allocation and power control. Moreover, DRL-based solutions are discussed for mode selection, spectrum allocation, and power allocation, as well as the joint optimization of resource management to enhance network performance in terms of both energy and spectral efficiency. It also presents advanced learning techniques, such as multi-agent DRL and federated DRL, as potential solutions to improve network scalability and preserve users’ privacy. Finally, the paper summarizes the open issues and future research directions for adaptive and explainable resource management techniques in D2D-enabled Industrial Internet of Things (IIoT) and Integrated Sensing and Communication (ISAC) systems in the upcoming 6G networks.

Device-to-device (D2D)-assisted sixth-generation$(6 \mathrm{G})$wireless networks necessitate intelligent resource management and power control to improve the overall system performance. Additionally, D2D underlay communication, a key enabler for improving spectral efficiency, is often hindered by interference, which adversely impacts the quality of service (QoS) for cellular users (CUs) and also degrades the performance of D2D users (DUs). Considering this, we propose a deep reinforcement learning-based method for resource management and power allocation in D2D-assisted heterogeneous networks (HetNets). A joint optimization problem integrating power control and channel selection is formulated for maximizing the system's spectral energy efficiency (SEE). Specifically, a multi-agent, double-deep Qnetwork (DDQN) algorithm is proposed in which each user equipment (UE) is trained as an action selection and target estimation agent. These agents dynamically optimize transmit power levels and channel selection to meet the QoS requirements of both CUs and DUs. Simulation results indicate that the proposed approach achieves significant improvement in average SEE by 19.84%, 35.27%, and 69.37% compared to the advantage actor-critic (A2C), deep Q network (DQN), and random allocation (RA) methods, respectively. This highlights the robustness of the proposed method in enabling efficient resource management for D2D communication, contributing to the development of self-sustainable networks.

Device-to-device (D2D)-assisted heterogeneous wireless networks are a key enabler for ultra-massive connectivity in sixth-generation (6G) networks. A critical challenge in realizing self-sustainable networks is achieving high energy efficiency (EE) while ensuring ubiquitous connectivity, meeting quality-of-service requirements, and preserving user privacy. This paper proposes a federated multi-agent double deep Q-network (F-MADDQN) framework for joint mode selection and resource allocation in D2D-assisted heterogeneous cellular networks. The proposed approach aims to maximize system EE under minimum signal-to-interference-plus-noise ratio and throughput constraints for both cellular users (CUs) and D2D users (DUs). In the proposed framework, agents exploit the local observations for channel selection, power allocation, and transmission mode decisions, while federated learning aggregates local DDQN models to preserve privacy and address partial observability. Simulation results demonstrate that the F-MADDQN method significantly enhances EE, achieving 17.55%, 35.24%, 66.72%, and 110.26% higher EE compared to multi-agent advantage actor-critic (MAA2C), multi-agent deep Q-network (MADQN), Hungarian algorithm (HA), and the random selection algorithm (RSA)-based resource allocation schemes, respectively. Moreover, the proposed scheme reduces the outage probabilities of CUs by 21.70%, 39.76%, 51.25%, and 55.43% compared to MAA2C, MADQN, HA, and RSA, respectively. These results show that the proposed framework offers a privacy-preserving solution for improving EE and reliability in D2D-assisted 6G networks.

The traditional layered-protocol architecture seems to be inefficient in wireless network environment as the wireless channel is time-varying in nature. The IEEE 802.16 standard which the WiMAX system is based upon is also designed according to such layered-protocol stacks. To obtain overall performance gain however, cross-layer optimization is essential. In this paper, we propose a cross-layer optimization architecture for WiMAX system. It consists of a cross-layer optimizer which acts as an interface between MAC and PHY layers. It gathers and optimized the parameters from both layers to achieve optimum performance gain. Preliminary simulations were conducted to justify the need for cross-layer optimization.