2. 2
王蒞君 講座教授
Prof. Li-Chun Wang (ED813、MIRC516)
Research Areas:
5G Wireless and Mobile Communications
UAV Communications and Networking
Intelligent Internet of Things (IIoT)
Big Data Analysis and AI Solutions for complex
systems
Honor
IEEE Fellow
科技部傑出研究獎 (2011 and 2017)
24. 24
Data-Driven 3D Placement of UAV Base Stations
for Arbitrarily Distributed Crowds
July 4th 2019
Primary result:
Chuan-Chi Lai, Li-Chun Wang, and Zhu Han, “Data-Driven 3D Placement of
UAV Base Stations for Arbitrarily Distributed Crowds.” To appear in IEEE
Globecom 2019.
Extended version:
Chuan-Chi Lai, Li-Chun Wang, and Zhu Han, “Drone Base Station as a
Service for Arbitrary Crowds: A Data-Driven Approach.” Plan to submit to
IEEE Transactions on Mobile Computing.
25. 25
DBS in wireless networks has attracted great attention recently
and DBS can be used to assist ground base stations for increasing
capacity and coverage
In this project, we will develop the dynamic 3D placement of DBS
for satisfying the communications demand of arbitrary flash crowds
25
Background and Motivation
Maratho
n
AR Games
Pokémon Go
Concert
26. 26
System Model and Assumptions
26
o Consider how to let a ground base station (GBS)
and 𝑘𝑘 DBS, coordinately serve 𝑁𝑁 ground users
and thereby improve the sum rate of the cellular
system.
o The GBS is equipped with omni-directional
antennas for ground communications and
mmWave directional antennas for the back-haul
connections of DBSs
o If the spatial information of UEs, 𝑁𝑁, 𝑁𝑁𝐺𝐺, and
𝑁𝑁𝑗𝑗 are known, the system sum rate can be
formulated as
o 𝐶𝐶𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 = ∑𝑗𝑗=1
𝑘𝑘
∑𝑖𝑖=1
𝑁𝑁𝑗𝑗
𝑐𝑐𝑖𝑖,𝑗𝑗 + ∑𝑖𝑖=1
𝑁𝑁𝐺𝐺
𝑐𝑐𝑖𝑖,𝐺𝐺
o 𝑐𝑐𝑖𝑖,𝑗𝑗 is the allocated data rate of a user
associated with the 𝑈𝑈𝑗𝑗, where 𝑗𝑗 = 1,2, … , 𝑘𝑘
o 𝑐𝑐𝑖𝑖,𝐺𝐺 is the allocated data rate of a user
associated with the 𝐺𝐺
o 𝑈𝑈 = {𝑈𝑈1, 𝑈𝑈2, … , 𝑈𝑈𝑘𝑘} is the set of DBSs,
o 𝐸𝐸 = {𝑢𝑢1, 𝑢𝑢2, … , 𝑢𝑢𝑁𝑁} is the set of target users
o 𝑁𝑁 is the total number of UEs,
o 𝑁𝑁𝐺𝐺 is the number of served UEs for the GBS, 𝐺𝐺,
o 𝑁𝑁𝑗𝑗 is the number of served UEs for each𝑈𝑈𝑗𝑗,
and then 𝑁𝑁𝐺𝐺 + ∑𝑗𝑗=1
𝑘𝑘
𝑁𝑁𝑗𝑗 = 𝑁𝑁
Down link:
1. GBS to UE
2. UAV to UE
3. GBS to UAV
27. 27
Existing Art & Motivations
Research Directions
• Maximize the number of covered users[1][4]
• Minimize the number of UAVs to serve users[2]
• Minimize the total power consumption[3]
Limitations
• Most of existing methods only consider the PPP based homogeneous distribution
Model[1][2]
• They do not consider the coexistence of the ground base station[1][2][3][4]
• They do not consider the coverage overlapping/co-channel interference issue[1][2][3][4]
• They do not consider the minimum data rate requirement of UE
1. E. Kalantari, H. Yanikomeroglu and A. Yongacoglu, "On the Number and 3D Placement of Drone Base Stations in Wireless Cellular Networks," 2016
IEEE 84th Vehicular Technology Conference (VTC-Fall), Montreal, QC, 2016, pp. 1-6.
2. J. Lyu, Y. Zeng, R. Zhang and T. J. Lim, "Placement Optimization of UAV-Mounted Mobile Base Stations," in IEEE Communications Letters, vol. 21,
no. 3, pp. 604-607, March 2017.
3. Q. Zhang, M. Mozaffari, W. Saad, M. Bennis and M. Debbah, "Machine Learning for Predictive On-Demand Deployment of Uavs for Wireless Communications," 2018 IEEE Global
Communications Conference (GLOBECOM), Abu Dhabi, United Arab Emirates, 2018, pp. 1-6.
4. A. V. Savkin and H. Huang, "Deployment of Unmanned Aerial Vehicle Base Stations for Optimal Quality of Coverage," in IEEE Wireless
Communications Letters, vol. 8, no. 1, pp. 321-324, Feb. 2019.
5. C.-C. Lai, C.-T. Chen and L.-C. Wang, "On-Demand Density-Aware UAV Base Station 3D Placement for Arbitrarily Distributed Users with Guaranteed Data Rates," in IEEE
Wireless Communications Letters, vol. 8, no. 3, pp. 913-916, June 2019. doi: 10.1109/LWC.2019.2899599
28. 28
28
Goal
• Maximize the system sumrate using a minimum number of DBS while
satisfying the minimum data rate requirement of UE
Novelty (Contribution)
• To support arbitrarily distributed UEs[4] for dynamic events/flash crowds
• Coordinate multiple UAVs with the ground base station
• Effectively reduce the overlapping coverage/co-channel interference
• The proposed method can automatically determine the appropriate
number, locations, altitudes of DBSs
Goal & Novelty of Our Work
29. 29
Problem Formulation
29
o Thus, we need to search the optimal 𝑁𝑁𝐺𝐺 and 𝑁𝑁𝑗𝑗 to optimize the considered System
Sum rate Optimization (SSO) problem and it can be defined as
o SSO is a NP-complete 0-1
Multiple Knapsack-like problem
30. 30
Step 3 : (Association) Associate each UE with a DBS using SINR constraint and update SINR of each UE
Step 3-A : (Constraints Check) If some UEs does not meet the communications constraints, try to
re-associate these UEs with the other nearby DBSs
Step 3-B : (Refinement) Solve the Minimum Covering Circle Problem of each cluster to minimize the
overlapping coverage and thus reduce the effect of co-channel interference
Step 3-C : Repeat Steps 3, 3-A, and 3-C until the communications constraints are satisfied; If not,
after repeat 𝑘𝑘 − 1 times, set 𝑘𝑘 = 𝑘𝑘 + 1 and jump to Step 2
Step 3-D : Compute the part B
Step 1 : (Initialization) Maximize the part A by associating neighboring UEs with
the GBS and then let 𝑘𝑘 = (𝑁𝑁 − 𝑁𝑁𝐺𝐺) 𝑐𝑐min/𝐶𝐶max
Proposed scheme
Data-Driven 3D Placement of DBSs
A B
Step 2 : (Clustering) Cluster unassociated UEs into to 𝑘𝑘 clusters using Balanced k-Means
1. Simple and effective
2. Do not need to know
the exact value of k
in advance
31. 31
31
Simulation Settings
o Settings
o 100 different artificial datasets as
the input spatial information
o Each dataset contains 600 to 1, 300
UE locations which are arbitrarily
distributed
o The size of target area is a 1, 200 ×
1, 200 m2 area
o We consider the urban scenario and
its environmental parameters
o The maximum allowable path-loss
of the DBS to the UE link is
34. 34
34
Conclusion
• The proposed approach can effectively maximize the system
sumrate and satisfy the traffic demands caused by arbitrary flash
crowds
• Unlike the conventional k-means++ which cannot determine the
value of 𝑘𝑘 by the algorithm itself, the proposed approach will
automatically determine an appropriate value of 𝑘𝑘 according to the
input spatial information, environmental parameters, and
communications constraints
• In this way, the system can save a lot of computational costs (time and
energy) on checking the impossible value of 𝑘𝑘.
35. 35
Enabling Mobile Service Continuity across
Orchestrated Edge Networks
Student: Osamah Ibrahiem Abdullaziz
Advisor: Prof. Li-Chun Wang
Department of Electrical and Computer Engineering, National
Chiao Tung University, Taiwan
July 4th 2019
36. 36
36
Service interruption issue in edge-enabled WiFi networks
o Handoff decision is locally made by users and the re-association process takes 1~
2s[1]. During this process, clients experience connectivity interruption.
o Hop count increases as users move away from the initial serving edge. This may
effect user experience especially for latency-sensitive applications.
Why service continuity is important?
o Wireless communication systems are driven by the quality of services they provide to
the end users. Industrial verticals such as smart transportation, e-health, cloud
robotics require very low-latency.
o Service interruption due to user mobility creates challenges to sustain the reliability and
latency requirements of these services.
1. A. Mishra, M. Shin, and W. A. Arbaugh, “An empirical analysis of the IEEE 802.11 MAC layer handoff process,” Comput. Commun. Rev.,
vol. 33, no. 2, pp. 93–102, Apr. 2003.
2. Machen, A., Wang, S., Leung, K.K., Ko, B.J. and Salonidis, T., 2018. Live service migration in mobile edge clouds. IEEE Wireless
Communications.
37. 37
37
A holistic service continuity approach
• ARNAB: Architecture for transparent service continuity via double-tier migration.
• The term ARNAB means rabbit in Arabic. The service with ARNAB behaves like a
rabbit hopping through the WiFi infrastructure.
• Fundamentally, the proposed architecture is divided into two tiers:
Tier 1: User connectivity migration
o Utilizes virtual access point (vAP) to deliver seamless WiFi experience using
commodity hardware.
Tier 2: Edge application migration
o Proposes a pre-copy container migration scheme to deliver seamless applications
experience at the proximity of the users.
38. 38
38
Work Contributions
• Proposed a novel architecture to provide transparent service continuity for mobile
users in edge-enabled WiFi infrastructure.
o Developed a virtual access point solution to enable ARNAB connectivity
migration.
o Developed a pre-copy migration scheme to relocate containerized applications in
edge environment.
• Identified the root-causes of prolonged downtime during container migration.
• Enhanced the container migration scheme by incorporating real-time computing
capabilities and fast storage.
39. 39
39
Tier 2: Edge Application Migration
Tier 1: User Connectivity Migration
o When C2 join the network, it associates with
a dedicated vAPC2.
o The RSSI received by C2 is monitored so that
when C2 moves away from AP1, vAPC2 gets
migrated to the nearest AP (AP2).
o An edge containerized application gets
created to serve C2 (e.g., video streaming
server).
o Edge application is also migrated to the
proximity of AP2.
C2
ARNAB in a Nutshell
40. 40
40
Migration Anatomy – downtime root-causes
n
Checkpoint latency
Memory to be copied Copying latency
Restore latency
Migration
Downtime
41. 41
Tier 1
user connectivity migration
Tier 2
edge application migration
41
Experimental Results – Service downtime
42. 42
42
Collaboration
Contributed to EU-TW H2020 Project (5G-CORAL) http://5g-coral.eu/
Publication
1. O.I. Abdullaziz, Wang LC, S. B. Chundrigar, Huang KL “Enabling Mobile Service Continuity across Orchestrated Edge
Networks", IEEE Transactions on Network Science and Engineering, (Under Review).
2. O.I. Abdullaziz, Wang LC, S. B. Chundrigar, Huang KL ”ARNAB: Transparent Service Continuity across Orchestrated Edge
Networks", In IEEE Global Communications Conference (GLOBECOM), Dec. 2018
Dissemination & Publication
43. 43
Ph.D. Group Meeting
August 1, 2019
Optimizing Offloading Ratio for 5G Multi-access Edge
Computing and Communications
Yung-Sheng Chao
Institute of Electrical and Computer Engineering
, National Chiao Tung University, Taiwan
44. 44
Background
50 billion smart mobile devices in use in the world by 2020 [1].
Due to the growing number of mobile devices, it is imperative
to enhance services using small base stations (BSs). Especially
in high-density hotspot scenarios.
44
[1] D. Evans, “The internet of things: How the next evolution of the internet is changing everything,” CISCO white paper, vol. 1, no. 2011, pp. 1–
11, 2011.
45. 45
System Architecture
Network construction using the Multi-access Edge Computing
(MEC) platform as the core equipment with radio access
network (RAN) is called MEC-RAN.
Improve network coverage and capacity.
The MEC-CRAN provides different protocol services for users
using different Radio Access Technologies (RATs).
45
Base Station Edge Computing User Equipment
Communication Link Computing Offloading Link
46. 46
Problem
In the MEC-RAN system, There is no effective management
mechanism to handle large packet traffic, which can cause
serious delays.
The fairness and effectiveness of resource allocation is a fundamental
issue in distributed computing systems.
How to implement optimal and efficient management of the
MEC -RAN will be a challenge.
46
Base Station Edge Computing User Equipment
Communication Link Computing Offloading Link
47. 47
Existing Art
47
Reference 1 2 3 4 5 6 7 Our
work
Minimize Latency X X X
MEC Resources allocation X X X X
Low Complexity X X X X
Communication Offloading X X X X X X
Computing Offloading X
1. Y. Lin, Y. Lai, J. Huang and H. Chien, "Three-Tier Capacity and Traffic Allocation for Core, Edges, and Devices
for Mobile Edge Computing," in IEEE Transactions on Network and Service Management, vol. 15, no. 3, pp. 923-933,
Sept. 2018..
2. X. Sun and N. Ansari, "Latency Aware Workload Offloading in the Cloudlet Network," in IEEE Communications
Letters, vol. 21, no. 7, pp. 1481-1484, July 2017.
3. Q. Li, J. Lei and J. Lin, "Min-Max Latency Optimization for Multiuser Computation Offloading in Fog-Radio Access
Networks," 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Calgary, AB,
2018, pp. 3754-3758.
4. J. Zhang, W. Xia, F. Yan and L. Shen, "Joint Computation Offloading and Resource Allocation Optimization in
Heterogeneous Networks With Mobile Edge Computing," in IEEE Access, vol. 6, pp. 19324-19337, 2018.
5. J. Ren, G. Yu, Y. Cai and Y. He, "Latency Optimization for Resource Allocation in Mobile-Edge Computation
Offloading," in IEEE Transactions on Wireless Communications, vol. 17, no. 8, pp. 5506-5519, Aug. 2018.
6. A. Roy, P. Chaporkar and A. Karandikar, "Optimal Radio Access Technology Selection Algorithm for LTE-WiFi
Network," in IEEE Transactions on Vehicular Technology, vol. 67, no. 7, pp. 6446-6460, July 2018.
7. C. Zhou, C.K. Tham, and M. Motani, “Online Auction for Truthful Stochastic Offloading in Mobile Cloud
Computing”, IEEE GLOBECOM 2017, Singapore, Dec. 2017. C142.
48. 48
System Model
𝐿𝐿𝑖𝑖 denote the average latency for the i-th UE group’s packet.
48
, , , , , , , , ,
1 1 1 1
* * *
N L N L
C C P P C C S
i i j k j k j s j s i j k j k i j
j k s k
L L R L R L R R
= = = =
= + +
∑ ∑ ∑ ∑
49. 49
Problem Statement
Input :
In MEC-CRAN system, there are M UE groups and N BSs, each BS
with L Multi-RAT network interface functions. Each UE with traffic
arrival rate 𝜆𝜆𝑖𝑖, UE – BS selecting table 𝑅𝑅𝑖𝑖,𝑗𝑗
𝑆𝑆
, and the BS with RAT’s
capacity of communication service function 𝜇𝜇𝑗𝑗,𝑘𝑘
𝑐𝑐
and capacity of
service function 𝜇𝜇𝑗𝑗
𝑝𝑝
.
Output :
The created communication offload ratio table parameters {ℛ𝑗𝑗,𝑘𝑘
𝑐𝑐
} and
computing offload ratio table parameters {ℛ𝑖𝑖𝑖𝑖
𝑝𝑝
}.
Objective :
Minimize the total latency ∑ 𝑳𝑳𝒊𝒊 of packets.
49
50. 50
Proposed scheme
We propose flexible offloading ratio decision optimization
(ForDo) to decide the offloading ratio of the communications
and computing for each MEC.
All Services in MEC-RAN system are modeled as M/M/1 queuing
system.
According to the UEs‘ traffic arrival rate (𝜆𝜆) and the MECs’ computing
capacity (𝜇𝜇)
Minimize the latency by optimizing the traffic offloading over
communications and computing.
Under M/M/1 queuing model, the latency minimization problem is
proved to be convex.
Find the analytic solution under linearly time complexity O(n) by
Cauchy–Schwarz Inequality.
50
51. 51
Objective function
51
1
,
1 1 1, ,
, , ,
1 1 1, ,
,
1 , ,
1
min *
2 1
min min
2 1
min min
2
min
M
i i
i
C PN L N
j k s
C C P P
j k sj k j k s s
C C C P P PN L N
j k j k j k s s s
C C P P
j k sj k j k s s
C
j k
C C
k j k j k
L λ
λ
λ λ
λ µ λ λ µ λ
µ µ λ µ µ λ
λ µ λ λ µ λ
µ
λ µ λ
=
= = =
= = =
=
+ − −
− + − +
+ − −
=
−
∑
∑
∑ ∑ ∑
∑ ∑
∑ ∑ ∑
∑ ∑
∑ 1 1
,
1 1 1, ,
1
1 min 1
2 1
* min min
PN L N
s
P P
j s s s
C PN L N
j k s
C C P P
j k sj k j k s s
N L N
µ
λ µ λ
µ µ
λ µ λ λ µ λ
=
= = =
− + − −
= − + + − + − −
∑ ∑ ∑
∑
∑ ∑ ∑
∑ ∑
Communications part Computing part
The latency minimization problem is proved to be convex.
Find the analytic solution under linearly time complexity O(n) by Cauchy–
Schwarz Inequality.
53. 53
Conclusion
Based on the packets arrival rate of UEs and the computing
capacity of the MECs , ForDo provides different offloading
ratios for different UEs packet traffic demands.
Operators use the capacity and latency limits of each MEC to
quickly assess the upper limit of the UE’s traffic arrival rate.
Operators also use the UE’s traffic arrival rate to assess how
much computing capacity needs to be added in MEC-RAN to
provide efficient computing with time constrain in 5G.
53