非正交多址接入（NOMA）由于能够显著提升频谱效率和设备连接数而成为物联网中支持海量连接的关键技术之一。而许多物联网通信场景，如工业自动化、车联网等，对通信的时延性能有较高的需求。在NOMA技术应用于时变信道中时，由于存在用户间干扰，导致分析统计时延性能十分困难，为保障统计QoS的NOMA传输带来了极大挑战。为此，本文深入分析了上下行NOMA的服务过程的统计特征，给出了NOMA系统中包括时延超标概率上界和有效容量的统计QoS指标的表达式，设计了保障统计QoS的功率分配方案。本文的创新点主要有：（1）针对用户已配对的上行NOMA系统，在无色散的块衰落信道中，推导静态功率分配下服务过程Mellin变换的表达式，并对Mellin变换应用随机网络演算（SNC），给出了排队时延超标概率上界；基于Mellin变换首次给出了上行NOMA用户有效容量的闭式表达式；分别基于排队时延超标概率上界和有效容量，设计了满足给定统计时延QoS需求的最小化发射功率的准静态功率分配方案和在$\alpha$公平性下最大化有效吞吐量的准静态功率分配方案；仿真表明，与正交多址接入（OMA）相比，所提方案能有效提升系统的统计QoS性能。（2）为了进一步提升上行NOMA的统计QoS性能，设计了最大化有效容量之和以及最大化有效能量效率的动态功率分配方案；仿真表明，相比于现有NOMA传输方案以及OMA，所提方案能显著提升系统的有效容量与有效能量效率。（3）针对下行NOMA，分别在无色散的Nakagami-$m$和Rician块衰落信道中推导了静态功率分配下服务过程Mellin变换的闭式表达式，并基于SNC给出了排队时延超标概率上界；基于该Mellin变换导出了有效容量的闭式表达式，并分别在高低信噪比条件下推导了有效容量的渐近表达式；在上述分析和性能表达式的基础上，分别对排队时延超标概率和有效容量应用最小最大和最大最小公平性准则，提出了两种新的保障NOMA用户对的统计QoS和公平性的功率分配方案；推导了所提功率分配方案在高低信噪比条件下的闭式解；仿真表明，所提方案能在保障用户公平性的同时显著提升统计QoS性能。此外，本文还将上述基于排队时延超标概率上界的工作扩展到了基于信号对齐的下行MIMO-NOMA系统中。（4）在NOMA中考虑统计QoS需求，在保障弱用户的最小有效容量的前提下，设计了最大化强用户的有效容量的动态功率分配方案，仿真表明，与现有的NOMA传输方案以及OMA相比，所提方案能显著提升系统的有效容量。
Non-orthogonal multiple access (NOMA) has become one of the key technologies supporting massive connections in the internet of things (IoT), due to its potential in significantly improving the spectrum efficiency and the number of connected devices in wireless communication systems. Many IoT communication scenarios, such as factory automation and vehicle-to-vehicle communications, have stringent requirements on both the number of connected devices and delay quality of servicce (QoS). However, it is intractable to analyze the statistical delay performance of NOMA in time-varying channels, due to the inter-user interference incumbent to NOMA. This brings a great challenge in designing NOMA transmission to guarantee the statistical delay QoS. To surmount this challenge, in this thesis we investigate the statistical characteristics of the service processes in uplink and downlink NOMA, and derive closed-form expressions of the statistical QoS metrics including delay violation probability and effective capacity in NOMA systems. Based on the analysis, power allocation schemes are designed to guarantee statistical QoS for NOMA. The major contributions of this thesis are listed as follows:(1) For an uplink NOMA system where users are already paired, the Mellin transforms of the service processes under static power allocation are derived in non-dispersive block fading channels. Then, stochastic network calculus (SNC) is applied to translate the Mellin transforms to the upper bounds of the queueing delay violation probability. Moreover, we provide closed-form expressions of the effective capacity (EC) for uplink NOMA for the first time, based on the derived Mellin transforms. By taking the upper bounds of the queueing delay violation probability and EC as the delay QoS metrics, we propose two new quasi-static power allocation schemes that minimize the transmitting power while satisfying given statistical QoS requirements and maximize the effective throughput of uplink NOMA under $\alpha$-fairness, respectively. Simulations demonstrate that the proposed schemes outperform OMA in terms of delay violation probability and EC.(2) In order to further improve the statistical QoS performance of uplink NOMA, we design two dynamic power allocation schemes that maximize the sum EC and the effective energy efficiency respectively. Simulation results demonstrate that the proposed dynamic schemes can remarkably enhance the EC and the effective energy efficiency when compared to either existing NOMA transmission schemes or OMA.(3) For downlink NOMA, we derive the closed-form expressions and their convergent truncated approximations for the Mellin transforms of the service processes of NOMA users in non-dispersive Nakagami-$m$ and Rician fading channels under static power allocation. The upper bound of the delay violation probability is established by applying SNC to the Mellin transforms. The closed-form expressions for the ECs of NOMA users are derived based on the Mellin transforms and analyzed in both low and high signal-to-noise ratio (SNR) regimes. By applying the min-max and max-min rules to the delay violation probability and EC, respectively, we propose two new power allocation schemes, which provide fairness to inherently unfair NOMA users in terms of delay guarantee and EC. Closed-form solutions are derived in low and high SNR regimes. Corroborated by extensive simulations, we show that the proposed power allocation schemes can significantly promote the statistical QoS performance of downlink NOMA while guaranteeing user fairness. Furthermore, we extend the work based on the upper bound of the delay violation proability to signal aligmanet based downlink MIMO-NOMA.(4) We bring statistical QoS requirements into downlink NOMA, and design a novel dynamic power allocation scheme which maximizes the EC of the strong user while ensuring the minimal EC reqruirement of the weak user. Simulations show that the proposed dynamic scheme can remarkably improve the EC comparing to either existing NOMA transmission schemes or OMA that dynamically allocates time slots.
