| With the continuous emergence of the new wireless applications,the contradiction between the limited spectrum resources of wireless networks and the exponentially increasing data transmission requirements has become increasingly prominent.In addition,different wireless applications have differentiated service requirements such as data transmission reliability,delay,and energy efficiency.Although the current wireless networks can satisfy some refined network requirements,it is difficult to guarantee the high speed,low latency and high reliability communication for dense communication scenarios and ultimate virtual reality(VR)applications.In addition,future network must provide intelligence services so as to support the leap from "Internet of Things"to "Intelligent of Everything".Therefore,this work proposed a smart cellular network architecture that includes two solutions to meet the stringent requirements of future wireless network applications,which are expanding spectrum resources and improving transmission efficiency.For the first solution,this work used Terahertz(THz)and visible light bands with broad spectrum resources for data transmission meet the high reliability and low-latency data transmission requirements.We also deployed unmanned aerial vehicles(UAVs)in wireless networks to provide ubiquitous network accesses.For the second solution,we studied semantic communication driven wireless networks that uses limited spectrum resources to transmit the meaning of original data so as to provide imprecise communication.For the proposed two solutions,the key challenge is jointly analyzing wireless service requirements and wireless resource characteristics and proposing targeted and intelligent network resource management solutions.Due to the complex relationship between the differentiated requirements of different applications and the wireless resource allocation strategies under the dynamic network states,it is difficult for traditional optimization algorithms to provide flexible and efficient technical support.Therefore,the resource management of wireless network urgently requires innovation and breakthrough in theory and paradigm.The resource management based on artificial intelligence technologies such as machine learning algorithms is a promising solution.However,in future wireless networks,the machine learning based resource management technology faces many challenges,such as joint analysis of spatiotemporal characteristics of network environment,adaptability in dynamic network environment,and multi-agent collaborative optimization.In order to satisfy the requirements of four specific network applications and manage wireless resources in different networks efficiently and intelligently,the following innovative work is carried out in this thesis:(1)In the VLC-enabled UAV networks,we considered the impact of ambient illumination on the data rate and illumination requirements of users.A deep learning model that can jointly analyze the temporal and spatial characteristics of ambient illumination is proposed to predict the illumination distribution in advance.Based on ambient illumination distribution prediction,the UAV deployment and the user association can be optimized to minimize the total transmit power of UAVs.(2)A THz/VLC-enabled wireless VR network is proposed to provide high data rate VR image transmission and user indoor positioning services.A reinforcement learning algorithm based on meta-learning framework is proposed to optimize LED selection and THz small base station-user association.The network resource management strategy trained by the proposed algorithm can quickly adapt to dynamic user movement patterns,thus maximizing the network reliability in dynamic environments.(3)A semantic communication network for textual data transmission is developed.The proposed network uses knowledge graph to model the semantic information of textual data.A reinforcement learning algorithm based on attention network is proposed.The proposed algorithm can evaluate the importance of the extracted semantic information as well as optimize the resource block allocation and semantic information selection based on importance distribution,thus maximizing the semantic communication system metrics that jointly consider semantic accuracy and semantic completeness.(4)A covert semantic communication network for image data transmission is developed.A multi-agent reinforcement learning algorithm is proposed,which enables legitimate devices and jammers to cooperatively optimize the transmission power and determine the semantic information required to be transmitted,thus jointly improving the performance and security of the semantic communication. |
