Research On Quantum Secure B5G Fronthaul Optical Network Resource Allocation Based On Machine Learning

Fronthaul optical network has become one of the essential means to improve mobile communication network’s comprehensive ability.In the future,it will face challenges in security,capacity,management and other aspects.Quantum Key Distribution(QKD)technology is based on the three laws of quantum mechanics.It can provide quantum keys for both sides of remote communication parts and achieve theoretical information security by one-time-pad.In addition,the space division multiplexing technology based on multi-core fiber can multiply the overall capacity of the network.This thesis investigates how to build a B5G fronthaul optical network with high performance for the future from the aspects of network architecture,delay control and resource allocation.A quantum secured B5G fronthaul optical network architecture based on multi-core optical fiber is presented.A noise prediction model based on Boosting algorithm is designed to evaluate the delay introduced by channel noise when evaluating the dynamic allocation of resources in the proposed architecture.Finally,according to the different service needs in the proposed architecture,a resource allocation scheme based on deep reinforcement learning is designed,which achieves efficient allocation of network resources according to service needs.The main research contents are as follows:(1)To overcome the shortcomings of capacity,security and flexibility of the existing fronthaul optical network architecture in the face of future network requirements,a quantum secured B5G fronthaul optical network architecture based on multi-core fiber is proposed.Previously,fronthaul optical networks were generally based on a single-core fiber and WDM architecture,making it difficult to cope with the challenges of future surges in data traffic,and existing encryption methods are based on computational complexity,making it difficult to resist the threat of quantum computers.In this thesis,a quantum secured B5G fronthaul optical network architecture based on multicore fibers is proposed.QKD technology and space division multiplexing technology based on multicore fiber are applied to future fronthaul optical networks,which multiplies the network capacity and gives it theoretical information security.In addition,in order to meet the requirements of highly flexible services management in the future network,the proposed network architecture uses wavelength selection switches to adjust the upstream and downstream wavelengths of the services in order to dynamically distribute classic-quantum services over suitable channels and improve network performance.Through simulation,the architecture can meet the requirements of QKD transmission,and has the advantages of large capacity and flexible wavelength add-drop.(2)To evaluate the network delay introduced by real-time channel noise in the fronthaul optical network,a noise prediction scheme based on Boosting algorithm is proposed,and noise prediction models based on XGBoost and LightGBM algorithms are designed respectively.In dynamic QKD optical network,in order to reduce noise caused by classical optical signals and improve the performance of QKD system,resources of classical and quantum signals should be allocated reasonably according to link noise,resource utilization,etc.However,evaluating the noise on the quantum channel will increase the overall network delay.Based on Boosting algorithm in machine learning,two noise prediction models are designed,and their generalization ability is improved by feature extraction,and the training and prediction speed are improved by using multipleoutput regression method.The simulation results show that the proposed noise prediction model can reduce the noise evaluation time by 98.8%and maintain the noise prediction accuracy of more than 96%.(3)A service requirements-based core-wavelength resource allocation scheme(SR-based CWA)is proposed to solve the resource competition in the fronthaul optical network.The arrival of services in dynamic networks is random.Traditional fixed allocation schemes cannot guarantee optimal performance of quantum channels and waste channel resources.However,real-time allocation schemes based on optimal channel prediction cannot guarantee optimal overall network performance in the long run.In addition,different quantum optical networks have different service requirements.Based on this,a resource allocation scheme based on DQN algorithm in deep reinforcement learning is designed to assist dynamic quantum and classical services resource allocation.The simulation results show that different resource allocation models can adapt to different services requirements and achieve good network performance by adjusting the value function in the deep reinforcement learning model.