Radio Access Network Intelligent Sensing-based Multi-granularity Resource Allocation

With the continuous emergence of unmanned services,holographic communication and other new services,the network has become increasingly complex,and the traditional mobile communication network can not adapt to the development of these emerging services.The adaptive ability of the network will be improved while introducing the sensing function,and ultimately achieve the intelligent,on-demand and flexible management of the network,thus providing reliable network support for these emerging services.However,the method of network management based on intelligent sensing will inevitably lead to cost increase due to frequent resource adjustment,which limits the adaptive dynamic and intelligent development of the network.Therefore,the multi-granularity resource allocation algorithm based on radio access network intelligent sensing is studied in this work.Firstly,the sensing algorithms,such as the traffic prediction algorithm of users and base stations and service types estimation algorithm of users are researched by deep learning in the radio access network(RAN),and then support dynamic and efficient resource allocation algorithms;Based on this,the joint optimization of communication,sensing and computing resources under different time granularity is explored.This algorithm improve the network spectrum efficiency,and reduce the service delay,including transmission delay and processing delay,and reduce the resource adjustment cost.The contributions of this work are as follows:(1)Traffic prediction and traffic estimation based on deep learning are studied in this work.Firstly,the Seq2 Seq algorithm with attention composed of gate recurrent unit(GRU)is used to predict the traffic of the base station and users.Specifically,the mathematical model of traffic prediction problem,the process of traffic data collection and preprocessing are introduced,and the network structure and parameter setting of Seq2 Seq model with attention are given.Then,the convolutional neural networks(CNN)algorithm is used to estimate the user’s service type.Specifically,the mathematical model of the service estimation problem and the process of service sample data collection and preprocessing are introduced,and the network structure and parameter setting of the CNN model are given.Finally,an experimental platform of communication,sensing and computing intelligent endogenous integration system is built,and traffic prediction at different sensing time granularity and service estimation are realized on this platform.By comparing the traffic prediction error and the accuracy of service type estimation of different algorithms under different parameter settings,it is verified that the traffic prediction model has smaller prediction error and service estimation model has higher accuracy used in this paper.(2)A multi-granularity resource allocation algorithm based on network intelligence sensing is proposed.Multi-time granularity is reflected in the effective time of the strategy generated by the algorithm.First,a joint optimization of communication,sensing and computing resources problem is modeled as a problem of maximizing the utility function with the constraints guarantee at multiple time granularity.Wherein,the utility is defined as the weighted sum of delay of all users,spectrum efficiency and resource adjusting cost,and the constraints include the minimum data transmission rate threshold and the resources occupation constraints.Secondly,the dueling deep Q network(Dueling DQN)algorithm is used to solve the problem,and the sensing information,including traffic prediction and service type estimation results,communication information,including signal-to-noise ratio,transmission power and the number of resource blocks,etc.,and computing information,i.e.computing resources of the base station are taken as the state sets;The effective action time of the resource allocation policy,the number of resource blocks,the transmission power,and the computing resources of the base station are taken as the action set.The weighted sum of delay of all users,spectrum efficiency and resource adjusting cost is taken as the reward;Finally,in the experimental platform relying on the open source software,a variety of different resource allocation algorithms are compared,and it is verifies that the algorithm proposed in this paper can improve the network spectrum efficiency,and reduce the service delay,including transmission delay and processing delay,and reduce the resource adjustment cost.