This paper addresses the problem of traffic splitting in the multi-access system for real-time applications in the uplink. We first understand the Long Term Evolution (LTE)-uplink-scheduling mechanism using Network Simulator ($\mathrm{N}\mathrm{S}-3$) and then analyse the LTE-Packet Data Convergence Protocol queue length process at the end of User Equipment (UE). We provide an end-to-end delay jitter analysis on a single LTE network for the Constant Bit Rate (CBR) Traffic. We observe that the delay jitter only depends on the allocation (Transport Block Size) provided by eNodeB but not on the allocation rate. The main finding is the structural dependency of the delay jitter on one of the CBR parameters: inter-burst time. We then utilize this to develop an online-adaptive traffic splitting algorithm in the multi-access system to improve the overall delay jitter performance of the system. We formulate the traffic splitting problem as an optimization problem applicable to different objective functions. We validate and share our real-world experimental results for the video/Real-time Transport Protocol traffic.

本文解决了多路访问系统中的流量拆分问题，用于上行链路的实时应用。我们首先使用网络模拟器了解长期演进（LTE）-上行链路调度机制（$\mathrm{否}\mathrm{小号}-\mathrm{3个}$)然后分析用户设备(UE)端的LTE-Packet Data Convergence Protocol队列长度过程。我们针对恒定比特率 (CBR) 流量在单个 LTE 网络上提供端到端延迟抖动分析。我们观察到延迟抖动仅取决于 eNodeB 提供的分配（传输块大小），而不取决于分配率。主要发现是延迟抖动对 CBR 参数之一的结构依赖性：突发间时间。然后，我们利用这一点在多路访问系统中开发了一种在线自适应流量拆分算法，以改善系统的整体延迟抖动性能。我们将流量拆分问题制定为适用于不同目标函数的优化问题。