Project topics and materials
Get Complete Project Material File(s) Now! »
Cellular system structure and evolution
The mobile cellular network used today has gone through several generations [59][47].
The first generation (1G) appeared in the 1980s which used analog technology to transmit the information. The well-known standard for the 1G system are for example, Analog Mobile Phone Service (AMPS) in American and Total Access Communication System (TACS) in Europe.
In the early 1990’s, the Second Generation (2G) starts to emerge using the digital communication and coding technology to guarantee the correctness of data transmission. In terms of services, voice and text message are provided in 2G systems. Although there were several 2G standards, the Global System for Mobile (GSM) is the most successful system and was highly adapted by the operators around the world. This has enabled GSM to be further enhanced and developed so as to support higher data rates. General Packet Radio Services (GPRS) and Enhanced Data Rates for GSM Evolution (EDGE) are the evolutions of GSM with enhanced Adaptive Modulation and Coding (AMC) and some other coding schemes.
The evolution of cellular systems continue to third generation which introduced the packet-switched concept coexisting with the circuit-switched method used in the previous system. The Univeral Mobile Telecommunications System (UMTS) developed within the 3rd Generation Partnership Project (3GPP) is one of the candidates that meet the 3G requirement of International Telecommunication Union (ITU) in terms of performance, service and spectrum efficiency. It provides enhanced radio interface called Universal Terrestrial Radio Access (UTRA) network and a core network evolved from the last generation. The Long Term Evolution (LTE) of UMTS system is the 3.9th Generation of cellular networks.
It was designed with an Evolve- Univeral Terrestrial Radio Access (E-UTRA) network with a full-IP core network called Evolved Packet Core (EPC). Whole information are supposed to be transmitted in packet-switched network and no circuit-switched service will be offered anymore in 4G. The entire architecture is named Evolved Packet System. Then we have the real 4G system, LTE-Advanced that fulfills the requirement of ITU with the advanced features for both radio and core networks. Nowadays, discussions are launched for 5G. The ITU requirement of 5G has not been defined yet. However, the most recognized system characteristics of 5G are Massive system capacity, Very high data rate everywhere, Very low latency, Ultra-high reliability and availability, Very low device cost and energy consumption, Energy-efficient networks. 5G also includes a lot of specific topics and technology like virtualization and it can support various types of services such as machine-type communication, automatic car, intelligent factory, etc.
Radio resource management
As radio resources are limited, how to fairly share the wireless resources and how to increase the spectrum efficiency become important issues. In the followings, we introduce two properties concerning the wireless resource management. Duplexing The duplexing aims at defining a transmission technique between the downlink and uplink link. There are two common techniques, Time-Division Duplex (TDD) and Frequency- Division Duplex (FDD) as shown in Fig. 2.4. In TDD, the downlink and uplink transmissions are partitioned over time and on the same frequency band. Generally speaking, TDD provides more freedom to the resource allocation. On the other hand, in FDD, the downlink and uplink are allocated to the separated frequency bands during the whole time axis.
Queueing theory and traffic modeling
In this section, we introduce the basic knowledge of queueing theory and how to apply queueing theory for traffic modeling. Queueing theory is a popular mathematical tool used to describe a dynamic system with a shared resource. In this section, we only introduce some concepts applied in our thesis. Regarding the details, we refer the readers to [32][22] .
Processor sharing discipline
Different from the classical FCFS queues, processor sharing queues provide an interesting property calledinsensitivity, mentioned in [62] and [24]. Assuming Poisson arrivals, the stationary distribution of the number of customers does not depend on the distribution of service times, which is not the case with the FCFS discipline. In reality, the service time distribution is not always exponential. Therefore, in practice, this property is very useful and has the practical interest in communication networks of allowing the development of engineering rules independently of precise traffic statistics.
Table of contents :
Acknowledgement
Abstract
Résumé
Synthèse en français
List of Acronyms
1 Introduction
1.1 Context
1.1.1 Video categories
1.1.2 Quality of Experience (QoE)
1.2 Objectives
1.3 Contributions
1.4 Publications
2 Background
2.1 Wireless systems
2.1.1 Wireless channel characteristics
2.1.2 Channel capacity
2.1.3 Network deployment
2.1.4 Cellular system structure and evolution
2.1.5 Radio resource management
2.2 Queueing theory and traffic modeling
2.2.1 Queue model
2.2.2 M/M/1 queue
2.2.3 State dependent queue
2.2.4 Processor sharing discipline
2.2.5 Whittle networks
2.2.6 Packet-level modeling v.s. Flow-level modeling
2.2.7 Flow-level modeling
2.3 Machine learning
2.3.1 Supervised learning
2.3.2 Cost function and probabilistic interpretation
2.3.3 Generalized Linear Model (GLM)
2.3.4 Gradient descent algorithm
2.3.5 Support Vector Machine (SVM)
2.3.6 Overviews of machine learning techniques
3 Model of Real-time Streaming Traffic
3.1 Problem statement and the state of the art
3.2 Flow-level and packet-level model
3.2.1 Flow-level dynamics
3.2.2 Packet-level dynamics
3.2.3 Fluid model approximation
3.3 Extension to heterogeneous radio conditions
3.3.1 Flow-level dynamics
3.3.2 Packet-level dynamics
3.4 Simulation results
3.4.1 Quasi-stationary regime
3.4.2 Single class model validation
3.4.3 Multiple class model validation
3.5 Validation with fading effect
3.6 Summary
4 Model of Adaptive Streaming Traffic
4.1 Problem statement and state of the art
4.2 System description
4.2.1 Video content configuration
4.2.2 Wireless access network
4.3 System model with flow-level dynamics
4.3.1 Small chunk duration model
4.3.2 Large chunk duration model
4.3.3 Stability condition
4.4 KPIs definition
4.4.1 Video bit rate
4.4.2 Service time
4.4.3 Deficit rate
4.4.4 Buffer surplus
4.4.5 Performance of Different KPIs v.s. Starvation Probability
4.5 Scheduling schemes
4.5.1 Heterogeneous radio conditions
4.5.2 Round-robin scheme
4.5.3 Max C/I scheme
4.5.4 Max-min scheme
4.5.5 Opportunistic scheduling scheme
4.5.6 KPIs definition for heterogeneous radio condition
4.6 Integration of elastic services
4.6.1 Stability condition
4.6.2 KPIs definition for integrating elastic traffic
4.7 Mobility model
4.7.1 Stability condition
4.7.2 Performance in light traffic
4.8 Summary
5 Simulation of Adaptive Streaming and Approximation Model
5.1 Impacts of chunk duration
5.2 Chunk duration design
5.2.1 One chunk per HTTP request
5.2.2 Multiple chunks per HTTP request following same size of requests
5.2.3 Performance comparison
5.3 Impacts of the number of video bit rates
5.4 Impacts of scheduling schemes
5.5 Impacts of intra-cell mobility
5.6 Discriminatory scheduling scheme
5.7 Impacts of largest video bit rate
5.8 Approximation model
5.8.1 Approximation model for significantly small chunk duration
5.8.2 Approximation model for significantly large chunk duration
5.8.3 Performance of approximation model
5.9 System dimensioning
5.10 Summary
6 Predicting QoE of Video Streaming with Machine Learning Technique
6.1 Problem statement and state-of-the-art
6.2 System Description
6.2.1 Video streaming
6.2.2 Radio access network and traffic characteristics
6.2.3 Flow-level model and maximum arrival rate
6.2.4 Event-driven simulator
6.2.5 Recorded features
6.3 Machine Learning Tool
6.3.1 Generalized Linear Model (GLM)
6.3.2 Support Vector Machine (SVM)
6.3.3 Performance metrics
6.3.4 Libraries
6.4 Simulation Analysis
6.4.1 Simulation configuration
6.4.2 Prediction performance of different HTTP streaming
6.4.3 Prediction performance of different features
6.5 Summary
7 Conclusions and Future Works
