![15
Challenge: The existing transmission mechanism can not reach the strict delay
requirements of IIoT
Network resources are limited
Traffic characteristics change dynamically
Approach: Delay-aware group-based MTC scheme
Adaptively adjust aggregation buffer size based on the change of real-time traffic
Adaptive cluster partition scheme by bin packing based on delay performance
Gain:
[Uplink transmission latency] 303 ms → 113 ms (↓62%)
5G Machine Type Communications for Latency-Aware Industrial5G Machine Type Communications for Latency-Aware Industrial
Internet of Things (IIoT)Internet of Things (IIoT)
303 ms303 ms
113 ms113 ms
62%62%
Move
Reliable Device-
aggregator Associations
Fixed Aggregator
Moving
Aggregator
IIoT TestbedIIoT Testbed](https://image.slidesharecdn.com/wanglabintroduction12042018-190109111125/85/Wang-lab-introduction_1204_2018-15-320.jpg)
![16
Challenge: The existing 802.11ad link setup latency
can not reach the AR/VR requirement.
802.11ad link setup latency: >40 ms
AR/VR latency requirement: ~ 4 ms
Approach:
AI-based light-weight WiFi fingerprint beam search
Gain:
[Beam search latency] 10.95 ms → 1.09 ms (↓90%)
[Probability of availability] 89% → 98% (↑11%)
Low-Latency High-availability 5G Indoor CommunicationsLow-Latency High-availability 5G Indoor Communications
Training Trajectory
Signal Mapping
Ideal Practical](https://image.slidesharecdn.com/wanglabintroduction12042018-190109111125/85/Wang-lab-introduction_1204_2018-16-320.jpg)
![17
UPER scheme
Almost achieve
URLLC requirement
Delay requirement (1 ms)
Challenge: How to avoid resource collision between urgent uplink URLLC
with other services (e.g., eMBB or mMTC)?
Approach:
User-initiated probabilistic elastic reservation (UPER) scheme
Probabilistic resource selection between dedicated and shared resource blocks.
K-repetition transmit in frequency domain.
Gain: (For 0.25 ms grant-free transmission interval)
[Average delay time] 0.56 ms → 0.385 ms (↓31%)
[1 ms delay reliability] 85.59% → 99.994% (↑14.404%)
Resource sharing for 5G URLLC Cellular SystemsResource sharing for 5G URLLC Cellular Systems
UPER
others
URLLC
Unreserved
others
URLLC
URLLC dedicated resources
Shared resources
URLLC
UL traffic
eMBB/mMTC
UL traffic
Resource blocks
Reserved resources
UL UL UL UL UL UL UL
0.25 ms
URLLC:
Ultra Reliable Ultra Low Late
Communications](https://image.slidesharecdn.com/wanglabintroduction12042018-190109111125/85/Wang-lab-introduction_1204_2018-17-320.jpg)
Wang lab introduction_1204_2018
- 1. 1 Mobile Computing Lab Chair Professor, Li-Chun Wang ( 王 君蒞 ) Research Interests: 5G Wireless and Mobile Communications Data-Driven Learning-Assisted Radio Resource Management Big Data Analytics for Industrial Internet of Things (IIoT) AI-enabled UAV Honors IEEE Fellow (2012) MOST Outstanding Research Award (2012, 2018) Best Paper Award: IEEE COMSOC APB (2015); IEEE VT Jack Newbauer
- 2. 2 intelligent Internet of Things (iIoT) Social Mobile Cloud Big-Data Analytics Security Better Healthcare Safer Planet Information
- 3. 3交通大學電機所 IBM Streams Computing for i-IoT ∗ Cooperation with IBM Watson in promoting “Stream Computing and Big Data Analysis” research and education ∗ Summer Training course have been offered by the experts of IBM Watson for 6 years. ModifyModify Filter / Sample Filter / Sample ClassifyClassify FuseFuse AnnotateAnnotate ScoreScore Windowed Aggregates Windowed Aggregates AnalyzeAnalyze
- 4. 4 Zion Server Extreme OpenFlow Switch OpenFlow Controller NCTU OpenFlow-based Datacenter Testbed Zion Servers 70 Nodes 560 Cores (4Core/CPU;2CPUs/Node) 3.36TB Memory 140TB HD Emulate 490~2000+ VMs Extreme OpenFlow Switch x 8 NOX OpenFlow Controller Machine Machine Machine…. OpenFlow Management Network OpenFlow Controller Production NetworkSoftware-Defined Network 英業達捐贈價 ：值 727 萬
- 5. 5 A Book for 5G 5
- 6. 6 5G Vertical Markets MVNO DMVNO D MVNO: Mobile Virtual Network Operator MVNE: Mobile Virtual Network Enabler MVNO CMVNO C MVNO BMVNO B MVNO AMVNO A Carrier2 Carrier1 Public Network Factory of the Future VR/AR Stadium Tele-Drone Fleet SI(B) Enterprise Private Network VerticalServiceProvider MEC ( vEPC + vAPP ) MVNE Licensed Spectrum Licensed Spectrum Shared Spectrum Smart Grids Courtesy of Taiwan 5G Office
- 7. 7 Research Projects 1. Intelligent Communications for Drone-Cruiser Cellular Networks 2. Latency-Aware Industrial IoT 3. 5G Ultra-reliable and Low Latency Communications 4. Industrial Big Data Application Research
- 8. 8 Intelligent Communications for Drone-Cruiser Cellular Networks Project Goal Develop a new UAV, Drone Cruiser, as the new flying base stations and apply AI techniques for solve the complex issues on dynamic factors in 5G beyond cellular systems Develop a 3D communications QoS prediction platform to help the dynamic placement of UAV-BSs Create more innovative services and applications
- 9. 9 Intelligent Communications for Drone-Cruiser Cellular Networks Issues: Limited hover time for current drone Learning-based Air-to-Ground Wireless Channel Modeling 3D Placement of Drone Base Stations Network Integration and Handover Techniques for the joins and departures of Drone Base Stations Development of Innovative Services The developed techniques can be applied to: •Aviation •5G communications industries •Industry 4.0, •3D Internet of Things, •cloud big data services, •intelligent control, and other fields ApplicationsApplications Solution: Data-driven optimization and AI-based prediction for the management of UAV base stations with real 3D wireless channel data
- 10. 10 Project Architecture Intelligent Communications and Networking Technologies for Drone-Cells Data Science & Analysis Prof. Wang Prof. Lin, Prof. Deng, Prof. Chang Prof. Wang, Prof. Lin, Prof. Shuai Prof. Shuai, Prof. Huang PI: Prof. Wang Industrial Partners GEOSAT Chunghwa Telecom AI UAV-BS Communications System Integration & Innovation Application Service Development AI UAV-BS Network and 3D Placement Innovative Carrier Design of UAV-BS Prof. Cheng, Prof. Li
- 11. 11 Drone Cruiser Designed by Prof. Teng-Hu Cheng, Mech. Engr. of NCTU
- 12. 12 Learning-Based Radio Maps 3D Channel Radio Map Combine with some open 3D maps QoS prediction tool/platform for UAV-assisted cellular networks LoS: Line of Sight NLoS: Non Line of Sight
- 13. 13 Application Scenarios Edge Computing • 3D Placement Service • Precomputed Recommendation • Offline Learning • Reinforcement Learning Backhaul • mmWave • Beamforming • Trajectory planning Environmental Sensing • Location & Mobility Pattern Recognition • Environmental Feature Extraction • 3D Channel Modeling Resource Management • Displacement of UAV-BS • Online learning QoS Analysis • Real-time monitoring
- 14. 14 UAV-Assisted Cellular System 14Macro-cell FBS cell #1 FBS cell #2 FBS cell #3 FBS cell #4 Resource for FBS cell #1 (UTLLC) Resource for FBS cell #2-#4 (eMBB)Resource for Marco-cell Resource dedicated to FBS URLLC eMBB D2D/FANET Communication FBS backhaul Link (mmWave) Traditional Cellular Link a. Learning-based 3D Drone-BS Placement for Bandwidth and Power Allocation (Trent) b. Data-driven 3D Placement of UAV-BSs for Mitigating Inter-drone Cell Interference ( 俊廷 ) c. Learning-based Trajectory Planning for Always Connected FBS ( 郁芳 ) d. Modeling and Analysis of UAV-Assisted Heterogeneous Cellular Systems for Ultra-reliable and Low Latency Communications (Allan) e. Power Control for Uplink Multiplexing of URLLC and eMBB in 5G Networks ( 晨曜 ) f. Modeling and Analysis the Delay and Reliability of D2D Communication in a UAV Swarm with URLLC Requirement (Irene) g. Unified Offloading Analysis Model of Multi-access Edge Computing and Communications for UAV Network (Jim) ① ② ③ ④ FBS cell #5 ⑤ ⑥ Edge Service ⑦
- 15. 15 Challenge: The existing transmission mechanism can not reach the strict delay requirements of IIoT Network resources are limited Traffic characteristics change dynamically Approach: Delay-aware group-based MTC scheme Adaptively adjust aggregation buffer size based on the change of real-time traffic Adaptive cluster partition scheme by bin packing based on delay performance Gain: [Uplink transmission latency] 303 ms → 113 ms (↓62%) 5G Machine Type Communications for Latency-Aware Industrial5G Machine Type Communications for Latency-Aware Industrial Internet of Things (IIoT)Internet of Things (IIoT) 303 ms303 ms 113 ms113 ms 62%62% Move Reliable Device- aggregator Associations Fixed Aggregator Moving Aggregator IIoT TestbedIIoT Testbed
- 16. 16 Challenge: The existing 802.11ad link setup latency can not reach the AR/VR requirement. 802.11ad link setup latency: >40 ms AR/VR latency requirement: ~ 4 ms Approach: AI-based light-weight WiFi fingerprint beam search Gain: [Beam search latency] 10.95 ms → 1.09 ms (↓90%) [Probability of availability] 89% → 98% (↑11%) Low-Latency High-availability 5G Indoor CommunicationsLow-Latency High-availability 5G Indoor Communications Training Trajectory Signal Mapping Ideal Practical
- 17. 17 UPER scheme Almost achieve URLLC requirement Delay requirement (1 ms) Challenge: How to avoid resource collision between urgent uplink URLLC with other services (e.g., eMBB or mMTC)? Approach: User-initiated probabilistic elastic reservation (UPER) scheme Probabilistic resource selection between dedicated and shared resource blocks. K-repetition transmit in frequency domain. Gain: (For 0.25 ms grant-free transmission interval) [Average delay time] 0.56 ms → 0.385 ms (↓31%) [1 ms delay reliability] 85.59% → 99.994% (↑14.404%) Resource sharing for 5G URLLC Cellular SystemsResource sharing for 5G URLLC Cellular Systems UPER others URLLC Unreserved others URLLC URLLC dedicated resources Shared resources URLLC UL traffic eMBB/mMTC UL traffic Resource blocks Reserved resources UL UL UL UL UL UL UL 0.25 ms URLLC: Ultra Reliable Ultra Low Late Communications
- 18. 18 Motivation: The cost of shutting machines down for one month to repair is about NT 6.8 million. The products produced by failed machines are also big cost. A predictive maintenance model to notify maintenance before machine fail is necessary. Challenge: The error of existing anomaly detection model by RNN increases a lot and costs much computation time since the sensor data of numerical control machine is regression data. Approach: An end-to-end non-linear value mapping model for anomaly detection. Gain: Map regression data to classification data. Increase accuracy. Training time speed up 16 times Real-Time Data Analytics and Predictive Maintenance Platform for Numerical Control Machine
Editor's Notes
- #16: 實驗圖