17 January 2020 in IEEE Internet of Things Journal
With the unprecedented development of smart mobile devices, e.g., Internet of Things devices and smart phones, various computation-intensive applications are explosively increasing in ultra-dense networks (UDNs). Mobile edge computing (MEC) has emerged as a key technology to alleviate the computation workloads of smart mobile devices and decrease service latency for computation-intensive applications. With the benefits of network function virtualization, MEC can be integrated with cloud radio access network (C-RAN) in UDNs for computation and communication cooperation. However, with stochastic computation task arrivals and time-varying channel states, it is challenging to offload computation tasks online with energy-efficient computation and radio resource management. In this paper, we investigate the task offloading and resource allocation problem in MEC-enable dense C-RAN, aiming at optimizing network energy efficiency. A stochastic mixed-integer nonlinear programming problem is formulated to jointly optimize the task offloading decision, elastic computation resource scheduling, and radio resource allocation. To tackle the problem, Lyapunov optimization theory is introduced to decompose the original problem into four individual subproblems which are solved by convex decomposition methods and matching game. We theoretically analyze the tradeoff between energy efficiency and service delay. Extensive simulations evaluate the impacts of system parameters on both energy efficiency and service delay. Simulation results also validate the superiority of the proposed task offloading and resource allocation scheme in dense C-RAN.
5 January 2021 in IEEE Transactions on Vehicular Technology
Technological evolutions in unmanned aerial vehicle (UAV) industry have granted UAVs more computing and storage resources, leading to the vision of UAVs-assisted edge computing, in which the computing missions can be offloaded from a cellular network to a UAV cloudlet. In this paper, we propose a UAVs-assisted computation offloading paradigm, where a group of UAVs fly around, while providing value-added edge computing services. The complex computing missions are decomposed as some typical task-flows with inter-dependencies. By taking into consideration the inter-dependencies of the tasks, dynamic network states, and energy constraints of the UAVs, we formulate the average mission response time minimization problem and then model it as a Markov decision process. Specifically, each time a mission arrives or a task execution finishes, we should decide the target helper for the next task execution and the fraction of the bandwidth allocated to the communication. To separate the evaluation of the integrated decision, we propose multi-agent reinforcement learning (MARL) algorithms, where the target helper and the bandwidth allocation are determined by two agents. We design respective advantage evaluation functions for the agents to solve the multi-agent credit assignment challenge, and further extend the on-policy algorithm to off-policy. Simulation results show that the proposed MARL-based approaches have desirable convergence property, and can adapt to the dynamic environment. The proposed approaches can significantly reduce the average mission response time compared with other benchmark approaches.
16 October 2019 in IEEE Internet of Things Journal
Unmanned aerial vehicle (UAV)-assisted communication is envisioned as a potential solution to the data traffic explosion in the massive machine type communications (mMTC) scenario. In this paper, we investigate the UAV-assisted cellular networks, where a UAV acts as a flying relay to offload part of the data traffic from the overloaded cell to another. We utilize the practical spatial distribution of data traffic and a convincing air-to-ground channel model. The quality of service (QoS) is defined as a UAV utility function which is designed based on a packet loss ratio (PLR)-related users cost function to represent the performance improvements brought by the UAV. We formulate a joint optimization problem to maximize the UAV utility function and then decompose it into the sub-problems about the access point selection and the UAV path planning, which influence the PLR by influencing the packet collision rate and channel state. Since the access point selection sub-problem is NP-hard, a game theory-based distributed algorithm is proposed, instructing the users to select the base station (BS) or the UAV as the access point autonomously. To achieve the most superior channel state, we solve the UAV path planning sub-problem by a deep reinforcement learning (DRL)-based approach, instructing the UAV to take the optimal action in each position. Simulation results show that the proposed access point selection scheme can significantly reduce the average cost of users and the proposed UAV path planning method can achieve a path with smaller average channel pathloss compared with other approaches.
11 June 2021 in IEEE Access
Computation-intensive mobile applications are explosively increasing and cause computation overload for smart mobile devices (SMDs). With the assistance of mobile edge computing and mobile cloud computing, SMDs can rent computation resources and offload the computation-intensive applications to edge clouds and remote clouds, which reduces the application completion delay and energy consumption of SMDs. In this paper, we consider the mobile applications with task call graphs and investigate the task offloading and resource scheduling problem in hybrid edge-cloud networks. Due to the interdependency of tasks, time-varying wireless channels, and stochastic available computation resources in the hybrid edge-cloud networks, it is challenging to make task offloading decisions and schedule computation frequencies to minimize the weighted sum of energy, time, and rent cost (ETRC). To address this issue, we propose two efficient algorithms under different conditions of system information. Specifically, with full system information, the task offloading and resource scheduling decisions are determined based on semidefinite relaxation and dual decomposition methods. With partial system information, we propose a deep reinforcement learning framework, where the future system information is inferred by long short-term memory networks. The discrete offloading decisions and continuous computation frequencies are learned by a modified deep deterministic policy gradient algorithm. Extensive simulations evaluate the convergence performance of ETRC with various system parameters. Simulation results also validate the superiority of the proposed task offloading and resource scheduling algorithms over baseline schemes.