In this paper, multi-UAVs serve as mobile aerial ISAC platforms to sense and communicate with on-ground target users. To optimize the communication and sensing performance, we formulate a joint user association, UAV trajectory planning and power allocation problem to maximize the minimum weighted spectral efficiency among UAVs. This paper exploits the centralized and the decentralized deep reinforcement learning (DRL) solutions to solve the sequential decision-making problem. On one hand, we first introduce the centralized soft actor-critic (SAC) algorithm. Then, we explore the equivalent transformation of the optimization objective based on symmetric group, propose the random and the adaptive data augmentation schemes to design the replay memory buffer of SAC, and accordingly propose SAC algorithms assisted by data augmentation to tackle the transformed problem. On the other hand, the multi-agent soft actor-critic (MASAC), a decentralized solution, is also introduced to solve this sequential decision-making problem. The experiment results reveal the effectiveness of the centralized and the decentralized solutions in considered scenarios. Specifically, the SAC assisted by the adaptive scheme significantly outperforms other centralized solutions in the training speed and the weighted spectral efficiency. Meanwhile, the decentralized MASAC algorithm behaves best in the early training speed.

With the development towards the next generation cellular networks, i.e., 5G, the focus has shifted towards meeting the higher data rate requirements, potential of micro cells and millimeter wave spectrum. The goals for next generation networks are very high data rates, low latency and handling of big data. The achievement of these goals definitely require newer architecture designs, upgraded technologies with possible backward support, better security algorithms and intelligent decision making capability. In this survey, we identify the opportunities which can be provided by 5G networks and discuss the underlying challenges towards implementation and realization of the goals of 5G. This survey also provides a discussion on the recent developments made towards standardization, the architectures which may be potential candidates for deployment and the energy concerns in 5G networks. Finally, the paper presents a big data perspective and the potential of machine learning for optimization and decision making in 5G networks.

As a novel paradigm, Vehicular Edge Computing (VEC) can effectively support computation-intensive or delay-sensitive applications in the Internet of Vehicles era. Computation offloading and resource management strategies are key technologies that directly determine the system cost in VEC networks. However, due to vehicle mobility and stochastic arrival computation tasks, designing an optimal offloading and resource allocation policy is extremely challenging. To solve this issue, a deep reinforcement learning-based intelligent offloading and power allocation scheme is proposed for minimizing the total delay cost and energy consumption in dynamic heterogeneous VEC networks. Specifically, we first construct an end-edge-cloud offloading model in a bidirectional road scenario, taking into account stochastic task arrival, time-varying channel conditions, and vehicle mobility. With the objective of minimizing the long-term total cost composed of the energy consumption and task delay, the Markov Decision Process (MDP) can be employed to solve such optimization problems. Moreover, considering the high-dimensional continuity of the action space and the dynamics of task generation, we propose a deep deterministic policy gradient-based adaptive computation offloading and power allocation (DDPG-ACOPA) algorithm to solve the formulated MDP problem. Extensive simulation results demonstrate that the proposed DDPG-ACOPA algorithm performs better in the dynamic heterogeneous VEC environment, significantly outperforming the other four baseline schemes.

The device-to-device (D2D) communications mode has been regarded as an effective technique for solving/relieving the contradiction between the exponentially increased data traffic requirements of mobile customers and the essence of scarcity of radio resources in wireless communication networks. In the presence of unfavorable direct communication links between D2D peers, cooperative relays may play an important role in enhancing both reliability and flexibility of D2D communications. Aiming at maximizing the sum throughout of the network with a low computation complexity, a new scheme, which jointly considers a number of critical aspects, such as power control, interference limit based on the location information, optimal relay selection based on delineated area, and optimal link selection based range division, etc., is proposed in this paper. Numerical results reveal that the proposed scheme is capable of substantially improving the system's performance compared with either the existing brute-force technique or the area-division scheme. Furthermore, the proposed scheme also exhibits its advantages over the existing techniques in terms of computational complexity.

Both Device-to-device (D2D) and full-duplex (FD) have been widely recognized as spectral efficient techniques in the fifth-generation wireless communications systems. By enabling FD mode in D2D communications, the attainable spectrum efficiency can be (in theory) up to twice as high as that of conventional half-duplex (HD) mode D2D technology, provided that the self-interference signal in the former can be sufficiently suppressed. Considering the fact that FD technology does not always outperform the HD technology in arbitrary channel condition, we will delve into the FD gain as well as its preconditions for acquisition in this paper. In particular, the performance of D2D- aided underlaying cellular networks is investigated by assuming that the D2D users are capable of operating at both HD and FD modes. Under a given user workload, the sum throughput in each DU is shown to be always improved by activating D2D links despite an extra DU-induced interference could be imposed on its co-spectrum cellular users. Numerical results show that the FD-D2D mode exhibits its superiority in terms of sum throughput than the HD-D2D mode in a light-workload scenario even if a non-ideal SI cancellation has been implemented in the former, but the latter pulls back a game in a heavy-workload scenario. Furthermore, to maximize the sum throughput, an appropriate mode-selection scheme (i.e., by choosing either FD or HD mode in the current time-slot for each individual D2D user) should be implemented for sufficiently exploiting the FD gain according to the instantaneous radio frequency environment.