Font Size: a A A

Research On Random Access Scheme Of Massive Machine Type Communications For Differentiated QoS Requirements

Posted on:2022-12-11Degree:MasterType:Thesis
Country:ChinaCandidate:M T XieFull Text:PDF
GTID:2518306761960259Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
With the rapid growth of the number of devices connected to networks,the Internet of Things(Io T)technology is booming.As one of the main forms of the Internet of Things,machine type communication has become the focus of people's attention.In order to provide wireless connection for a large number of Io T devices,the fifthgeneration mobile communication system(5G)presents the scenario of massive machine type communications(m MTC).In the m MTC scenario,a large number of machine type communications devices(MTCD)initiate wireless access requests simultaneously,which poses a great challenge to the wireless access network.Random access collision based on cellular network architecture is the key problem for network access.Therefore,it is particularly important to investigate the random access mechanism to meet the differentiated quality of service(QoS)requirements and access demands of a large number of devices.When the MTCD accesses the wireless cellular network,a preamble is sent through the physical random access channel to perform a contention-based random access process,and after success,the available resources in the physical uplink shared channel are used to transmit data.In order to solve the two problems of preamble collision on PRACH and the shortage of data transmission resources on PUSCH in the random access process of MTCD in m MTC,this thesis constructs a model for dynamically dividing virtual small cells that can flexibly respond to the increase and decrease of equipment,and classifies MTCD to meet the differentiated QoS requirements of different applications,and then,this thesis proposes two schemes of backoff access decision and resource allocation based on machine learning:(1)In order to solve the backoff access problem after contention-based random access collision of MTCD which have differentiated QoS requirements,this thesis proposes a backoff access decision algorithm combining analytic hierarchy process and Actor-Critic for each group of MTCD competing for the same preamble: The multiattribute decision-making method is used to solve the QoS objective function that meets the diverse needs.The access process of MTCD is set as double queue,and the state transition process of data queue is modeled as Markov decision process,and the reinforcement learning model is further established.In order to maximize the sum function of QoS of the successful access MTCD while meeting the individual requirements,the AC algorithm is used to learn autonomously to obtain the optimal backoff access strategy.(2)In order to solve the problem of the shortage of uplink data transmission resources in random access process of MTCD with diversified QoS requirements,this thesis proposes a resource allocation scheme based on grouping and deep reinforcement learning on the basis of virtual small cells system model: By summarizing historical data to divide the activity level and active time period of MTCD,and a dynamic grouping(DG)algorithm based on geographical location and activity characteristics is proposed.For the MTCD whose events are randomly driven and active periods overlap each other in each group,with the goal of maximizing the number of MTCD that can successfully complete data transmission on the premise that the diversified QoS requirements are met,a resource allocation algorithm based on deep Q-network is proposed to dynamically allocate uplink data transmission resource blocks.Finally,it is verified by simulation that these two schemes proposed in this thesis can effectively improve the access success rate of MTCD and the performance of system,and reduces the average access delay while meeting the personalized QoS demands.
Keywords/Search Tags:Massive machine type communications, Quality of service, Random access, Reinforcement learning, Resource allocation
PDF Full Text Request
Related items