Upload
others
View
2
Download
1
Embed Size (px)
Citation preview
Emerging Technologies in 5G
ENGINEERING
ELECTRICAL AND COMPUTER ENGINEERING
Broadband Communication Research Group (BBCR)
Dept. of ECE, University of Waterloo200 University Avenue West
Waterloo, Ontario, Canada N2L 3G1
Nan Cheng
Guest Lecture
ECE 710 (Lectured by Professor Xuemin Shen)
March 20, 2018
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
2
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
3
Evolution of Mobile Networks
4
Mobile 1G: Foundation of Mobile
5
Mobile 2G: Enabled More Users
6
Mobile 2G: Enabled More Users
7
Mobile 3G: Mobile for Data
8
Mobile 3G: Mobile for Data
9
Mobile 4G: Mobile Broadband
10
What will 5G be?
• 5G mobile networks need
• Support rich applications and services
• Satisfy diversified requirements
• Evolutionary network architecture
• Orchestration of new and existing technologies11
5G Provides A Platform for New
Connected Services
12
VR/AR: Ultra High Data Rates
13
Autonomous Car: Reliable/Latency
14
Autonomous Car: Reliable/Latency
Communicating intent and sensor data even in
challenging real world conditions
15
Internet of Everything (IOE):
Trillions Connectivity
• Very dense & huge no. of devices
• Low power consumption
• Long battery life16
5G: Key Performance Indicators
17
5G Use Cases/Scenarios
Use case development completed at
3GPP SA (TR 22.891):
Enhanced Mobile BroadBand (eMBB):
10-20Gbps network-level data rate.
Ultra-Reliable Low latency
Communications (URLLC):
at most 1ms delay experience
Massive Machine-Type Communication
(mMTC):
106 links/km2 connection density.
18
5G Distinguished Features
5G
Features
Scalable, Flexible,
Sustainable 03.
Diversified Capabilities 02.
Improved KPI 01 06 New Technologies
05Enhanced Security
04 Reduced Costs
19
5G New Opportunities
New Ecosystem
Software-based
Vertical Markets Big Data Analytics
IoT
HetNetsIaaS
New Research Issues
SecurityCloud
20
21
5G Key Technologies
RSU
RSU
3G/4G 3G/4G
3G/4G
802.11p
RSURSU
RSU
RSU
3G/4G 3G/4G
3G/4G
802.11p
RSURSU
Seamless Handover
D2D
VANET scenarios
Cloud
Fog nodes Fog nodes
SDN/NFV Controller
SDN/NFV Controller
Vehicles Vehicles
Core Network:
SDN
NFV
Cloud computing
Data center
Radio Access Tech:
Polar Coding
Massive MIMO
Full Duplex
Millimeter Wave
Access Network:
Vehicular Network
IoT
Fog computing
Big
Data
Analytics/A
rtificial In
tellig
ent
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
22
23
5G Requirements on Networking
• Global control
• Data forwarding
• Network management
• Intelligent control with big data and AI
• Flexible
• Scalable
• Larger network
• Hardware/software update
• New application/service
*Some slides are from Prof. Raouf Boutaba slides introducing SDN at University of Waterloo
24
Network Planes
• Data Plane
• responsible for forwarding packets
• Control Plane
• responsible for defining the behavior of the network
• Service Plane
• provides services beyond packet forwarding
• Management Plane
• configures routing protocol, monitors network devices
and handle anomalies
25
Data Plane
• Responsible for moving packets
• local, per-router function
• Input port->output port in a router
• Other data plane functions
• Packet scheduling and buffering
• Packet fragmentation
• Bandwidth metering
• Access control
• Traffic monitoring
26
Control Plane
• Responsible for moving packets
• Network-wide logic
• Determines how packet is routed among
routers along end-end path from source
host to destination host
• Route computation function
• Other control plane functions
• Topology discovery
• Access control list generation
• QoS enforcement
27
Service Plane
• Provides services beyond packet forwarding
• Load balancers, web proxy, web cache
• Firewall
• Intrusion detection system
• Mobility control
28
Management Plane
• Configures routing protocol, monitors network devices
and reacts to anomalies
• Responsible for managing FCAPS
• Fault
• Configuration
• Accounting
• Performance
• Security
• Provisions resources for network services
29
Traditional Networking
• Control plane is vertically integrated with data plane
• Each router runs its own control plane
30
Traditional Routing
• Traditional routing protocols are distributed
• Routers exchange connectivity info. with each other
• Each router builds its own routing table
• Forwarding tables (FIB) are populated based on
information available in the routing table
31
Distributed Control Plane
• Distributed per-router control plane
• Vertically integrated router contains switching hardware,
runs proprietary implementation of Internet standard
protocols (IP, RIP, ISIS, OSPF, BGP) in proprietary router
OS (e.g., Cisco IOS)
32
Traditional Network Element
33
Why Not Traditional Networking in 5G?
• Lack of programmability
• Changes in the network requires manual reconfiguration
• Error-prone and time consuming
• Network becomes hard to scale
• Vendor dependency
• Handful of vendors have monopoly in the market
• Lack of open interface
• Software vertically integrated with hardware
• Over specified
• Slow protocol standardization
34
Why Not Traditional Networking in 5G?
• Less opportunities for innovation
• Only vendors can write the code
• Slow time-to-market for new services/features
• High operational cost
• More than 50%
• Human error still causes most outages
• Buggy software
• Hardware with 20+ million lines of code
• Cascading failures, vulnerabilities, etc.
35
Software-Defined Network (SDN)
• Software Defined Networking (SDN) emerged as a
new networking architecture that
• Separates control plane from data plane
• Software running on a logically centralized server
remotely controls network hardware
• Open standards (vendor neutral)
• Directly programmable
• Facilitates automation & programmability
36
SDN Architecture
37
Network Evolution with SDN
38
Why SDN in 5G?
• Separation of control and data planes
• Enables independent development of new features
without hardware modification
• Global network view at control plane
• Network-wide view allows easier application of policies
• Management Automation
• Minimize human intervention
• Lower operational cost
• Monitoring
• On-demand monitoring of resources
• Dynamically change what to monitor
• Application performance can be improved by advanced
monitoring features
39
Why SDN in 5G?
• Elasticity
• Resources can be scaled up or down on-demand
• More Flexible/Easier Performance Management
• Traffic engineering
• Capacity optimization
• Load balancing
• Faster failure recovery
• Network Function Integration
• On-demand provisioning of resources
• Firewalls, IDS, IPS, etc.
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
40
41
Big Data Analytics in 5G
• Proliferation of network traffic (volume and type)
• Highly real-time streaming data (Youtube, Twitch, Douyu)
• Mobile/IoT devices everywhere (Peaceful city, mobile phone,
AR/VR, e-health…)
• Increased complexity of network and traffic monitoring
and analysis
• Difficulty in predicting and generalizing application
behavior
• Too many sources of knowledge to process by human
• Too many black boxes
• Need for aggregated solutions: get the value out of data
42
Data Usage in Mobile Networks
• Mobile data traffic will grow at a compound annual
growth rate (CAGR) of 47%
• Mobile devices will grow at a CAGR of 8%
• Of all IP traffic in 2021, 50% WiFi, 30% wired, 20%
mobile
• By 2021, there will be 8.3B mobile devices, and 3.3B
M2M connections
43
Traffic in Data Center Networks
• IoT/analytics/database workloads are growing fast
• Video and social networking will lead the increase
• Cloud (virtualized data center) traffic dominates
growth
44
Big Data Generations
Data collected:Smart phone
Movement
Traffic
Environment
Electricity
Monitor
Light
Etc.
45
How is Data Generated in 5G?
• What is monitored in the network?
• System and services
• Available, reachable, energy consumption
• Resources
• Expansion planning, maintain availability
• Performance
• Round-trip-time, throughput
• Changes and configurations
• Documentation, revision control, logging
• NOC: Network Operations Center
• Coordinate tasks
• Report status of network and services
• Process network-related incident and complaints
46
Hierarchy of Network Data Analytics
• Multiple layers of network data analytics
• Historical analytics:
• Service plan creation, network planning, subscriber profiling
• Near real-time analytics:
• Network optimization, location-based services
• Real-time analytics:
• Dynamic policy, self-optimizing networks
• Predictive analytics:
• Traffic demand forecasting, fault avoidance
• Data is richer when associated to context: layer,
location, time
47
Big Data Analytics in Smart City
48
Big Data Analytics in Cloud
• Leverage high performance of Cloud techniques
• Help delivery the real-time results
• Flexible and easy-to-use visualization
49
Big Data Based Intelligence
• Real-time situational awareness
• Continuous monitoring
• Data collection & historical record
• Learning and intelligence
• Visualization
• Smart application enablement
• Traffic control, planning, optimization
• Emergency response
• Event planning & management
50
Ex1. Dynamic Resizing in a Data Center
• If we know/can predict the workload distribution, we
can control the active servers
51
Ex2. Traffic Differentiation
52
Ex3. Bandwidth Defragmentation
Bandwidth defragmentation based on real-time monitoring and forecasting
53
Ex4. Network Data Visualization
54
Artificial Intelligence (AI) in Big Data
• AI is required in all layers of
next-generation networks
• Ultimate goal is to automate
the management of the
network and make it more
efficient
• AI models are improved with
big data collected over time
55
Fundamental AI Algorithms
• Supervised learning
• Classification
• Regression
• Unsupervised learning
• Clustering
• Anomaly detection
• Semi-supervised learning
• Reinforcement learning
• Transfer learning
• Deep learning
56
Supervised Learning
• Classification
• Identifying to which of a set of categories (sub-populations) a
new observation belongs
• Regression
• Predicting variable is continuous, such as price
57
Unsupervised Learning
• Clustering
• Dimensionality reduction
• Anomaly detection
58
Evaluation Metrics in AI Algorithms
• Accuracy
• A = (TP+TN)/all
• Precision
• p = TP/(TP+FP)
• Recall
• R = TP/(TP+FN)
• F-score
• F = 2/(1/p+1/r)
Classified Positive Classified Negative
Actual Positive True Positive (TP) False Negative (FN)
Actual Negative False Positive (FP) True Negative (TN)
59
An Example: Network Anomaly Detection
• Detect network anomaly (prob. = 1%)
• Data: positive=10 (anomaly); negative=9990 (normal)
• Alg. 1: Predict all: normal
• Alg. 2: Some machine learning algorithm
60
An Example: Network Anomaly Detection
Classified
Positive
Classified
Negative
Actual Positive TP: 0 FN: 10
Actual Negative FP: 0 TN: 9990
Classified
Positive
Classified
Negative
Actual Positive TP: 9 FN: 1
Actual Negative FP: 20 TN: 9970
Alg. 2: Some machine learning algorithmAlg. 1: Predict all: normal
Accuracy
A = (TP+TN)/all
Precision
p = TP/(TP+FP)
Recall
R = TP/(TP+FN)
F-score
F = 2/(1/p+1/r)
• Accuracy
• A = 99.9%
• Precision
• p = 0
• Recall
• R = 0
• F-score
• F = 0
• Accuracy
• A = 99.79%
• Precision
• p = 0.31
• Recall
• R = 0.9
• F-score
• F = 0.46
61
Evaluation Metrics in AI Algorithms
• True positive rate (TPR)
• TPR = TP/(TP+FN)
• False positive rate (FPR)
• FPR = FP/(TN+FP)
• Receive operating characteristics (ROC) curve
• FPR vs TPR
62
Evaluation Metrics in AI Algorithms
• Confusion Matrix
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
63
64
5G Requirements on Channel Coding
• Low decoding error
• Low coding/decoding delay
• Low decoding complexity
• Easy to implement
• Low-cost chips
• Advanced coding techniques
• Multiple code lengths
• HARQ
• Rate matching
65
Polar Codes
• Novel error correction coding mechanisms that have resulted in achieving performance very close to the limits
• Polar codes is the first coding scheme proved to achieve Shannon capacity
• Easy to implement
• 5G channel coding standard for control channel
*Some slides are from Prof. Arikan’s slides introducing polar codes at UC Berkeley,
http://www2.egr.uh.edu/~zhan2/ECE6332/slidesarikan.pdf
66
Channel Model
67
Binary Symmetric Channel Model
68
Channel Capacity
69
Main Idea of Polar Codes
70
Aggregate and Redistribute Capacity
71
Combining
72
Splitting
73
Polarization Is Common
74
Practical Polarization
Kernel
75
Two Bit-Channels
76
Binary Erasure Channel (BEC)
Erasure probability: ε
Capacity: 1- ε
77
Calculate Capacity of BEC(0.5)
0.5
0.5
0.25
0.75
78
Size 8 Construction
79
Size 8 Bit Channel Capacity
0.50.250.06250.000039
0.1211
0.1914
0.6836
0.3164
0.8086
0.8789
0.9961
80
Polar Code Example, N=8, Rate=1/2
81
Polar Code Example, N=8, Rate=1/2
82
Polar Code Example, N=8, Rate=1/2
83
Polarization
N=128 N=256
N=512 N=1024
84
Polarization Theorem
85
Decoding: Successive Cancellation
86
Performance of Polar Code (1)
87
Performance of Polar Code (2)
88
Performance of Polar Code (3)
Polar vs LDPC (Low-density parity check) – short code
89
Performance of Polar Code (4)
Polar vs LDPC (Low-density parity check) – long code
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
90
91
Vehicular Networks (VANETs)
Vehicle-to-Vehicle
(V2V) communications
Vehicle-to-Infrastructure
(V2I) communications
Safety applications
ITS applications
Entertainment
applicationsVehicle-to-Everything (V2X)
communications
92
VANETs Application
Traffic Notification
Remote Control
Mobile Office
Speed Management
Software Update
Collision Avoidance
Visibility Enhancing
Cooperative Driving
Road
Safety
Passenger
Infotainment
Vehicle Traffic
Optimization
Car
Manufacturer
Services
Internet access
93
Application – Safe Driving
Intersection Collision Warning Co-operative Collision Warning Lane Change Warning
Approaching Emergency vehicle Rollover Warning Work Zone Warning
Coupling/Decoupling Road Information Sharing Electronic Toll Collection
94
Application – Safe Driving
95
Application – Traffic Management
• Transportation efficiency
• Traffic jam costs 160 billion
in US in 2017
• $1,200 per driver
• 20% more fuel wasted
96
Application – Infotainment
97
Application - Self-Driving (1)
98
Application - Self-Driving (2)
Sensing error Scope limitation No coordination Prohibitive
99
Application - Self-Driving (3)
• Improve sensing scope• V2V: 300 meters
• V2I: even longer
• 3D sensing
• Add redundancy to sensing• Assist decision making
• Information accuracy
• AI can better utilize V2X information
• Coordination• Coordination among vehicles
• Edge-based decisions for a set of vehicles
• Improve efficiency, reduce cost
100
5G Requirements on V2X Communication
19/03/2018
101
5G Requirements on V2X Communication
102
5G Requirements on V2X Communication
103
Device-to-Device Communication
• Reliable: BS controlled
• Efficiency: spectrum reuse
• High data rate: proximate
• Low delay: proximate
• Offload the cellular
• Context-aware: discover the
surroundings and communicate
with nearby devices
W. Sun, E. G. Ström, F. Brännström, K. C. Sou, and Y. Sui, “Radio resource management for D2D-
based V2V communication,” IEEE Trans. Veh. Technol., vol. 65, no. 8, pp. 6636–6650, Aug. 2016.
104
D2D-based V2V Communication
• Outage probability
• Reliability constraint (10-5)
• Latency constraint
105
D2D-based V2V Communication
• Optimization Problem
• Maximize the CUEs’ rate
• Satisfy VUE’s constraints on reliability and delay
• Transmit 12800 bits with Pout<10-5
a. CUE sum rate b. VUE transmitted bits
Intelligent Transmission
106
Measurement Big Data
Machine LearningContext Insight
Context-Aware Design
• V2V communication
measurement
• Measurement data analysis
• Identify the context factors
• Design context-aware
intelligent protocol
• Each vehicle broadcast safety message beacon periodically.
• Position, speed, heading, acceleration, turn signal status, etc.
• For safety applications in ITS, and self-driving.
• Requirements
• Delay (beacon per 100 ms)
• Reliability (packet delivery ratio)
• Fairness (vehicles equally important)
• Connectivity (dense urban scenario)
107
V2V Safety Beaconing in VANETs
[1] N. Cheng, F. Lyu, J. Chen, W. Xu, H. Zhou, S. Zhang, and X. Shen, "Big Data Driven Vehicular Networks," IEEE Network, submitted.
[2] F. Lyu, N. Cheng, H. Zhou, W. Xu, W. Shi, J. Chen, and M. Li, "DBCC: Leveraging Link Perception for Distributed Beacon Congestion Control in VANETs," IEEE
Internet of Things Journal, submitted.
108
A Time-Division Broadcasting Scheme
• Requirements
• Delay
• Frame length 100 ms
• Reliability
• Only one vehicle transmit in a slot
• Fairness
• Dedicated slots for each vehicle
• Connectivity
• V2V performance is significantly affected by environment.
• Vehicle speed, distance
• Obstacles
• Packet may be lost even no collision. Cannot be solved by
simple TDMA MAC!
• Question:
• What factors impact the V2V performance in practice?
• How to consider the real V2V performance to improve vehicular
beaconing?
109
However?
110
V2V Measurement in Shanghai
Data Type Contents
GPS Speed, height, location
V2V DSRC Measurement Packet delay, loss
Video (front camera) Forward video
(For NLoS/LoS labelling)
Data Volume 110 GB
Measurement in Suburban, Highway, and UrbanVideo screenshot (LoS)
Video screenshot (NLoS)
People Square
• Overall PDR is high
throughout urban areas
• NLoS time is much less than
LoS time.
• NLoS significantly degrades
the V2V performance
111
Insights from Measurement Data
• Target: Max Safety Benefit
• Every vehicle assign minimum
beacon rate
• According to LoS/NLoS to
neighbors, calculate link weight
• Increase beacon rate of vehicle with
highest link weight, until max rate
• Increase beacon rate of vehicle with
2nd highest link weigth…
• Until channel capacity
112
Context-Aware MAC Design
?How to get LoS/NLoS conditions?
113
Online NLoS Condition Detection
Feature Target
PDR_1s PDR_5s PDR_10s LoS/NLoS
1 1 0.89 1
0.5 0.72 0.86 0
0.5 0.65 0.55 0
… … … …
114
Performance Evaluation
• Introduction to 5G Networks
• Software-Defined Network
• Big Data Analytics
• Polar Coding
• Vehicular Networks
• Conclusions
Outline
115
• 5G network has much more strict KPIs
• 5G network enables a whole set of potential
applications/services
• 5G uses a series of new technologies:
• SDN
• Big data analytics/AI
• Polar coding
• V2X communication
Conclusion
116
117
THANKS!