Mobile edge computing (MEC) emerges recently as a promising solution to relieve resource-limited mobile devices from computation-intensive tasks, which enables devices to offload workloads to nearby MEC servers and improve the quality of computation experience. In this paper, an MEC enabled multi-user multi-input multi-output (MIMO) system with stochastic wireless channels and task arrivals is considered. In order to minimize long-term average computation cost in terms of power consumption and buffering delay at each user, a deep reinforcement learning (DRL)-based dynamic computation offloading strategy is investigated to build a scalable system with limited feedback. Specifically, a continuous action space-based DRL approach named deep deterministic policy gradient (DDPG) is adopted to learn decentralized computation offloading policies at all users respectively, where local execution and task offloading powers will be adaptively allocated according to each user’s local observation. Numerical results demonstrate that the proposed DDPG-based strategy can help each user learn an efficient dynamic offloading policy and also verify the superiority of its continuous power allocation capability to policies learned by conventional discrete action space-based reinforcement learning approaches like deep Q-network (DQN) as well as some other greedy strategies with reduced computation cost. Besides, power-delay tradeoff for computation offloading is also analyzed for both the DDPG-based and DQN-based strategies.

Mobile edge computing (MEC) is a promising solution to enable resource-limited mobile devices to offload computation-intensive tasks to nearby edge servers. In this paper, dynamic computation offloading in a non-orthogonal multiple access (NOMA) based multi-user network is investigated by considering stochastic task arrivals and wireless channels. Specifically, a cooperative multi-agent deep reinforcement learning (CMADRL) based framework is proposed to learn decentralized offloading policies to allocate task offloading and local execution powers at each user, which minimizes long-term average network computation cost in terms of power consumption and buffering delay. By leveraging a centralized training and distributed execution strategy, the proposed framework can not only learn efficient decentralized policies, but also relieves users’ computational pressure and works well in coordinating the interference of NOMA-based networks. To reduce the training complexity, it is further improved to train one parameter shared policy network that can be exploited by all users with slightly sacrificed performance. Numerical simulations demonstrate that the proposed CMADRL based framework can learn efficient dynamic offloading policies at each user, which significantly outperforms the conventional independent Q-learning based framework as well as some greedy strategies with lower network computation cost.

In this paper, we consider a fully sustainable cooperative communication system which consists of multiple source nodes with radio-frequency (RF) energy harvesting capabilities, a half-duplex relay node with renewable energy supplies, and a destination node. Specifically, the relay node is powered by the green energy harvested from renewable sources such as solar or wind, while the source nodes are wirelessly charged by the RF energy from the relay node’s forwarding signals to the destination node. An optimal joint time scheduling and power allocation problem is formulated to achieve the maximum system sum-throughput of the users over a finite time horizon. To tackle the formulated NP-hard non-convex mixed integer nonlinear programming problem, we first analyze its upper bound by problem reformulation and relaxation, which can be simplified by the directional water filling algorithm and iteratively solved by sequential parametric convex approximation. We then propose an optimal branch-and-bound framework to solve the formulated problem, and develop an efficient sub-optimal offline algorithm and a heuristic online algorithm to reduce the computational complexity. Finally, extensive simulations are conducted to verify the superiority of the proposed solution and demonstrate that the sub-optimal algorithm approaches the performance upper bound with polynomial time complexity.

In this letter, we study the hybrid-field cascaded channel estimation in the extremely large-scale RIS (XL-RIS) assisted multi-user millimeter wave systems, where the cascaded channel is composed of far-field and near-field radiation components, and has spatial non-stationarity caused by visibility regions. We propose a U-shaped network based on the dedicated multilayer perceptron (MLP) architecture, termed as U-MLP, to realize the high-dimensional channel reconstruction with limited pilot overhead. In U-MLP, a basic feature extraction module-Permutator is designed to capture the long-range dependency of non-stationary channel, while the U-shaped backbone is constructed to exploit effective latent representation of the high-dimensional cascaded channel. Numerical results show that the proposed U-MLP outperforms existing channel estimation benchmarks with less pilot overhead.

In this paper, a joint beamforming design for energy-throughput tradeoff is investigated in a sustainable cloud radio access network (Cloud-RAN), where multiple remote radio heads (RRHs) powered by independent renewable energy sources collaboratively transmit wireless information and energy to the data receiver and the energy receiver simultaneously. To explore the optimal joint beamforming design for both time-invariant and time-varying channels, firstly, optimization problems are formulated to maximize the throughput of the data receiver while charging the energy receiver with sufficient RF energy over a finite time horizon. Then, the formulated non-convex optimization problems are relaxed into a convex form and shown to be upper bounded by the optimal value of the relaxed problem. This upper bound can be further proved to be tight as the optimal solution to the relaxed problem is rank one. Moreover, motivated by the derived optimal solution, an efficient online joint beamforming algorithm is proposed for practical implementation. Finally, extensive simulations are performed to verify the superiority of the proposed joint beamforming strategies to other beamforming designs.

Low earth orbit (LEO) satellite networks are promising to carry out edge computing and reduce the service latency in future 6 G networks. Meanwhile, the microservice architecture provides a lightweight and flexible approach for network function deployment in satellite edge computing networks by splitting complex applications into multiple microservices. In satellite edge computing networks, a brand new problem is how to deploy various types of microservices with service function chain (SFC) constraints onto appropriate satellites to efficiently utilize limited on-board resources and reduce the service latency. In this paper, a distributed microservice deployment strategy in satellite edge computing networks is proposed by taking into account time-varying satellite network topology and diverse satellite resources. Specifically, we formulate the deployment problem as a partially observable Markov decision process (POMDP) to minimize the load imbalance as well as the service latency. A distributed multi-agent reinforcement learning (DMARL) scheme is proposed for learning a dynamic deployment strategy. Furthermore, to accommodate more microservices in SFC, a novel attention-based actor-critic strategy is designed to significantly accelerate the convergence of network training. Simulation results demonstrate that the proposed DMARL scheme can achieve a superior performance.