A model-driven deep reinforcement learning heuristic algorithm for resource allocation in ultra-dense cellular networks. IEEE Transactions on Vehicular Technology, 69(1), pp. 983-997.There may be differences between this version and the published version. You are advised to consult the publisher's version if you wish to cite from it.Abstract-Resource allocation in ultra dense network (UDN) is an multi-objective optimization problem since it has to consider the tradeoff among spectrum efficiency (SE), energy efficiency (EE) and fairness. The existing methods can not effectively solve this NP-hard nonconvex problem, especially in the presence of limited channel state information (CSI). In this paper, we investigate a novel model-driven deep reinforcement learning assisted resource allocation method. We first design a novel deep neural network (DNN)-based optimization framework consisting of a series of Alternating Direction Method of Multipliers (ADMM) iterative procedures, which makes the CSI as the learned weights. Then a novel channel information absent Q-learning resource allocation (CIAQ) algorithm is proposed to train the DNN-based optimization framework without massive labeling data, where the SE, the EE, and the fairness can be jointly optimized by adjusting discount factor. Our simulation results show that, the proposed CIAQ with rapid convergence speed not only well characterizes the extent of optimization objective with partial CSI, but also significantly outperforms the current random initialization method of neural network and the other existing resource allocation algorithms in term of the tradeoff among the SE, EE and fairness.
This paper investigates the covert performance for an unmanned aerial vehicle (UAV) jammer assisted cognitive radio network. In particular, the covert transmission of secondary users can be effectively protected by UAV jamming against the eavesdropping. For practical consideration, the UAV is assumed to only know certain partial channel distribution information (CDI), whereas not to know the detection threshold of eavesdropper. For this sake, we propose a model-driven generative adversarial network (MD-GAN) assisted optimization framework, consisting of a generator and a discriminator, where the unknown channel information and the detection threshold are learned weights. Then a GAN based joint trajectory and power optimization (GAN-JTP) algorithm is developed to train the MD-GAN optimization framework for covert communication, which results in the joint solution of UAV's trajectory and transmit power to maximize the covert rate and the probability of detection errors. Our simulation results show that, the proposed GAN-JTP with a rapid convergence speed can attain near-optimal solutions of UAV's trajectory and transmit power for the covert communication.
BP neural network is introduced to the fault location field of DWDM optical network in this paper. The alarm characteristics of the optical network equipments are discussed, and alarm vector and fault vector diagrams are generated by analyzing some typical instances. A 17 BP neural network structure is constructed and trained by using MATLAB. By comparing the training performances, the best training algorithm of fault location among the three training algorithms is chosen. Numerical simulation results indicate that the sum squared error (SSE) of fault location is less than 0.01, and the processing time is less than 100 ms. This method not only well deals with the missing alarms or false alarms, but also improves the fault location accuracy and real-time ability.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.