Performance Analysis And Resource Management For Dense Cellular Internet Of Things

As a communication network that considers general connections of people-devices and devices-devices,Internet of Things(Io T)gradually comes into the human production and living,and becomes an indispensable factor to realize urban intelligence.In order to achieve the long-distance transmissions,cellular Io T has been widely studied.There are many applications for cellular Io T,such as: bike sharing,smart metering,wireless alarm,etc..With the development of those applications,people put forward higher requirements for cellular Io T in terms of reliability,service quality,spectrum efficiency and computing efficiency.Recently,the spring up of the Io T devices makes the cellular Io T develop towards intensity.There are many characteristics of dense Io T networks,such as,complicated network interferences,high devices access demands,high uplink data traffic ratio,high data volumes,etc..These features cause the decreasing of the system coverage probability,the declining of the quality-of-service(Qo S)and the increasing of the computing burden.To overcome the challenges above,it is necessary to introduce advanced technologies to explore the limits of the system performance and optimize the resource allocation so as to improve the system performance.Combining the characteristics of the dense cellular Io T network above,this dissertation studies the performance analysis and resource management strategy from both transmission and application levels.The research content involves four parts of the communication networks: the network structure,devices access,signal transmission and data computing.At the transmission level,firstly,in order to obtain the reliable coverage of the network,Chapter 2 analyzes the signal-to-interference ratio(SIR)coverage probability and provides theoretical support for base station deployment.Then,in order to improve the quality-of-service,Chapter 3 and Chapter 4 consider the massive access requirements and the uplink/downlink transmission features of the cellular Io T network,introduce the non-orthogonal multiple access(NOMA)and full-duplex(FD)technology,respectively,and explore the device scheduling and power allocation methods to improve the spectrum efficiency.Finally,at the application level,with the purpose of reducing the computing burden and guaranteeing the data security,Chapter 5 studies the wireless federated learning method based on the prorogation network.The main contributions of the dissertation include:1)Due to the complicated interference feature of the dense cellular Io T network,under the random arrival of the data packets,the data queue length among the base stations appears to have strong spatial correlations.The dissertation first presents the sufficient and necessary conditions to guarantee the stability of the queues.Then,a two-queue-length approximated model is introduced based on the queue lengths of the base stations at the steady state.Using tools from stochastic geometry and queueing theory,the SIR coverage probability in the steady state is derived.An iterative algorithm is proposed to calculate the active probability of the short-queue length BSs.To reduce the complexity,Beta distribution is applied to approximate the distribution of the service rate in each iteration of the algorithm.Finally,numerical results present the similarity between the analysis results and the results obtained by Monte-Carlo simulations,and show the convergence of the algorithm.2)Considering the large number of serving devices feature of the dense cellular Io T network,in order to improve the Qo S,this dissertation introduces the NOMA technique and investigates the expression of the achievable SINR region and the corresponding resource allocation that achieves Pareto optimal.To characterize the achievable SINR region,the dissertation first derives the necessary and sufficient conditions for an SINR vector to be feasible,then formulates an optimization problem to calculate the boundary points of the SINR region of the downlink NOMA system and its' dual uplink system,respectively.Based on matrix analysis,the closed-form solutions of the SINR region boundary of the two systems are derived and the results are strictly proved to equal to each other.To obtain the Pareto optimal resource allocation solutions,the satisfaction conditions of the optimal devices decoding order are analyzed.Based on this,an iterative algorithm related to the decoding order and power allocation is proposed.Finally,simulations verify the convergence and the effectiveness of the algorithm under high load cases.3)Considering the feature that more uplink transmission requirements of the dense Io T network,to improve the Qo S,this dissertation introduces FD transmission technique.Firstly,the dissertation proposes a frequency reuse scheme so as to alleviate the intercell interference.Then,a jointly max-min-fairness rate maximization problem of the transmission direction assignment(TDA),device pairing(DP)and power allocation(PA)are studied.Since the variables are changed over different frequency,the problem is formulated by a two-time-scale manner.Before solving it,the optimization problem is proved to be non-deterministic polynomial(NP)-hard.Then,two different two-stage algorithms,obtaining the suboptimal values of TDA and DP,are proposed under non-full load and full load case,respectively.PA variables are solved by applying successive convex approximation or weighted minimum mean square error.Finally,simulations verify the validity of the algorithms and the impact of traffic loads on system performance.4)Due to the large device number and large data volumes features of the dense cellular Io T network,the data sets are non-independent and identically distributed(non-i.i.d.)at different devices,which causes an additional computing burden to the central server.In order to guarantee the safety of the data and reduce the computing burden,this dissertation introduces the propagation network and studies the devices scheduling and parameters updating strategy for the wireless federated learning.Firstly,the Katz centrality of the propagation network is introduced to characterize the influence of the device(or data).Then,a device scheduling and parameter updating scheme of the wireless federated learning algorithm is proposed based on the device influence.Since the channels are nondeterministic,the parameter signal successful transmission probability is derived by using stochastic geometry.Based on the basic assumptions of loss functions,the convergence of the algorithm is proved by norm inequalities.Simulations verify the validity of the algorithm and compare the effects of the algorithm parameters on the accuracy and convergence speed of the algorithm.