Research On Massive Random Access In Machine-to-machine Communications

Machine-to-Machine(M2M)communication aims to realize fully automatic communica-tion between machines without human intervention.In the past few decades,this communi-cation method has experienced rapid development.With the widespread application of M2 M communication in smart grids,Internet of Vehicles,smart healthcare,smart agriculture,smart factories and other fields,the number of machine type nodes(MTDs)is bound to increase on a large scale.At present,considering factors such as coverage rate and reliability,the Long Term Evolution(LTE)system is still the best choice for undertaking M2 M communications.When MTDs access the network,random access(RA)is first required.However,when a large num-ber of MTDs communicate through the LTE system,it will inevitably cause serious radio access network(RAN)overload,leading to problems such as reduced access efficiency,increased ac-cess delay,and increased energy consumption.In addition,considering the characteristics of diverse M2 M communication types and short average data packet length,the existed random access mechanism cannot serve M2 M communication well.This dissertation studies how to achieve the goals of increasing network throughput,meeting the communication requirements of different nodes,and reducing signaling consumption by adjusting the backoff parameter set-tings of each node,which has important reference significance for the design of random access protocol for M2 M communication.First of all,in view of the current situation of severe congestion and diverse application types in massive M2 M random access,two multi-priority scenarios are considered,namely the high-priority MTDs delay absolute limitation scenario and the high- and low-priority MTDs throughput relative limitation scenario.In the scenario where the absolute value of the delay of high-priority MTDs is limited,the MTDs are divided into delay-sensitive MTDs and delay-insensitive MTDs.All delay-insensitive nodes are divided into a group.These MTDs have no requirements of delay but their total access efficiency should be as high as possible.For delay-sensitive MTDs,these MTDs are divided into multiple groups according to the different delay requirements of various applications.The access behavior of all MTDs can be charac-terized by a double-queue model.Based on this,the throughput and average access delay of each group of MTDs can be expressed as expressions about the number of MTDs,the total data packet arrival rate,and backoff parameters of each MTD.In order to meet the delay require-ments of delay-sensitive MTDs,the average delay of delay-sensitive nodes is required to be lower than a specific threshold.The analysis shows that in order to maximize the throughput of delay-insensitive MTDs under the delay requirements of delay-sensitive nodes,the access backoff parameters of each group of MTDs need to be adjusted accordingly.In particular,the optimal backoff parameters of delay-sensitive MTDs only depend on the delay requirements of these MTDs,while the optimal backoff parameters of delay-insensitive MTDs depend on the number of MTDs in the network and the number of MTDs in each group.In scenarios where the throughput of high and low priority is relatively limited,the throughput of different priority MTDs should meet a certain percentage requirement.The analysis shows that in order to maxi-mize the network throughput and meet the throughput requirements of different priority nodes,the optimal backoff parameters of each group of MTDs not only depend on the parameters of the group of MTDs,but also depend on the throughput ratio limit among the groups of MTDs.Secondly,for the problem of excessive signaling consumption in random access of mas-sive M2 M communication,this dissertation proposes a distributed optimal backoff parameters determination algorithm.Using this algorithm,each MTD can determine the optimal backoff parameters by itself,and the network throughput can be maximized.In particular,each MTD only needs to observe its own number of successful access requests and the total number of access request attempts within a certain estimation period,and the estimated optimal backoff parameters can be obtained according to the observed information.For the proposed method for determining the optimal backoff parameters in distributed manner,analysis shows that the estimation period has an important influence on the performance.In particular,the selection of the estimation period should consider both the accuracy of the estimation and the time re-quired for the estimation.When the estimation period lasts long,each node can obtain a more accurate estimation result,so it is more likely to obtain the optimal throughput,but if the num-ber of MTDs in the network changes quickly,this setting will cause the MTDs cannot update the backoff parameters timely.By comparing with two well-known centralized algorithms,it can be seen that the centralized algorithm can update the backoff parameters in each time slot to adapt to network changes,but they also incur a lot of signaling consumption and cannot achieve the optimal throughput.Finally,in view of the low efficiency of random access in large-scale M2 M communica-tions,a deep neural networks based analyzable Double Contention Random Access(DCRA)mechanism is proposed.In particular,the base station(BS)first divides the received random access preamble(RAP)into 4 categories,namely,not be selected,be selected by one MTD,be selected by two MTDs and be selected by three or more MTDs.In the DCRA mechanism proposed in this dissertation,based on the classification results of RAPs,the BS will allocate two uplink resources for the third type of RAP,that is,the RAP selected by two nodes,so the two conflicting MTDs can randomly select an uplink resource to continue the random access process,which means that these conflicting MTDs can obtain an extra chance to successfully access the network instead of suffering doomed access failure.For the performance analysis of this mechanism,a double-queue model is used to describe the random access process of each node and the performance of the random access process is analyzed.In addition,this thesis also provides a method to determine the optimal backoff parameters.When this method is applied,the BS only needs to collect statistic information such as the arrival rate of data packets and the service requirements of each MTD to determine the optimal backoff parameters without track-ing the dynamic information of the network,so when the network scale is large,the complexity of the algorithm is greatly reduced compared to the algorithms proposed in other documents.A large number of simulation results show that the throughput of the DCRA mechanism has a huge improvement compared with the traditional random methods.