Project1 Number: EUJ-01-2016-723171
Project Acronym: 5G-MiEdge
Project title: Millimeter-wave Edge Cloud as an Enabler for 5G Ecosystem
Periodic Technical Report
Part B
Period covered by the report: from 01/07/2017 to 30/06/2018
Periodic report: 2nd
1 The term ‘project’ used in this template equates to an ‘action’ in certain other Horizon 2020 documentation
Explanation of the work carried out by the beneficiaries and Overview of
the progress
1.1 Objectives
The key measurable objectives of the 5G-MiEdge project are:
1.1.1 Objective 1: Research, develop, and prove the 5G based MiEdge concept whose
viability will be evaluated by detailed theoretical and numerical analysis and
prototyped for proof-of-concept.
This Objective is the foundation of the whole 5G-MiEdge project. It manifests in work
package 1 (WP1), to continuously ensure the effective collaboration between the Japanese and
European teams, define scenarios and use cases that are relevant for this project and can take
advantage of the newly proposed technologies.
In the second period, there has been one deliverable:
D1.3 System architecture and requirements
The details are explained in Section 2.1.
1.1.2 Objective 2: Develop transmission schemes and protocols of mmWave
access/backhauling aimed to assist the mobile edge cloud with caching/prefetching
so as to realize ultra-high speed and low latency service delivery, resilient to
network bottlenecks, such as e.g. backhaul congestion, users’ density, mission-
critical service deployment, assuming three target scenarios: stadium, office, and
train/station.
The second objective aims to design new technologies to implement the 5G-MiEdge concepts
and meet up the new requirements. It is settled in work package 2 and in the second period,
there was one deliverable.
D2.3 Design of mmWave antennas for 5G enabled stadium
Section 2.2 contains all the details on work package 2.
1.1.3 Objective 3: Develop novel ultra-lean and inter-operable control signaling over
3GPP LTE to provide liquid ubiquitous coverage in 5G networks based on
acquisition of context information and forecasting of traffic requirements, in order
to enable a proactive orchestration of communication/computation resources of
the mmWave edge cloud.
The actual implementation of the proposed concept requires substantial changes to the
network architecture, taking into account all the benefits and drawbacks of using mmWave
technology. For this objective, work package 3 will develop and design a liquid edge cloud or
user/application centric orchestration.
In the second period, there was one scheduled deliverable.
D3.1 Architecture of mmWave edge cloud and requirement for control signalling
The current status and results are detailed in Section 2.3.
1.1.4 Objective 4: Develop user/application centric orchestration algorithms and
protocols to adapt radio and computation resources of mmWave edge cloud in 5G
networks by utilizing traffic forecast provided by liquid RAN C-plane to enable
self-organized and proactive reservation of the resources and satisfy low-latency
service requirements.
This objective is assigned to task 3.3 of work package 3. There was no deliverable scheduled
for the second period, but a lot of research and development was conducted and is explained
in section 2.3.3.
1.1.5 Objective 5: Develop a joint 5G test-bed integrating mmWave edge cloud, liquid
RAN C-plane, and user/application centric orchestration to foster an effective
impact of 5G-MiEdge in both Europe and Japan, particularly in preparation of
2020 Tokyo Olympic Games. The 5G-MiEdge test-bed will liaise actively with the
other EU/JP consortium focusing on the network side as well as to leverage
synergies between alternative 5G concepts.
Objective 5 is meant as a proof of concept. Taking the developed designs and specifications
from objective 1, hardware from objective 2 and the algorithms from objectives 3 and 4 and
creating a functional test-bed to further study and refine the development. It was assigned to
work package 4 and there was one deliverable scheduled for the second period.
D4.1 Performance evaluation of 5G MiEdge based 5G cellular networks
In section 2.4 the details are explained.
1.1.6 Objective 6: Contribute to the definition of 5G mobile communications standards
in 3GPP and IEEE, as well as in open fora such as NGMN, Small Cell Forum, and
the International Telecommunication Union (ITU) Industry Specification Group
MEC, in terms of mmWave access, liquid RAN C-plane, and protocols for
user/application centric orchestration by coordination across European and
Japanese partners.
Work package 5 is working on standardization, spectrum regulation and dissemination of the
technologies and results from the 5G-MiEdge project. In this period there was one
deliverable.
D5.2 Second report on dissemination, standards, regulation and exploitation plan
The details of this objective are presented in section 2.5.
Explanation of the work carried out per WP
This section presents the six WPs that compose the 5G-MiEdge project. Each work package is
split into tasks. For each task, we provide information about the status, details about past and
upcoming deliverables, a selection of highlights and the relationship to other tasks. The focus
of this report are the activities concluded in the second year of the project.
2.1 Work package 1: Scenario/use cases, business model, and 5G architecture
and ecosystem
Contributors: Intel, FHG, CEA, TI, URom, TTech, KLAB, PANA
The scope and structure of WP1 has not changed during the second year of the project.
This WP runs throughout the lifetime of 5G-MiEdge and is in charge of some key aspects of
the project:
- It fosters and ensures that an effective collaboration between the Japanese and the
European teams takes place, creating a common vision that maximizes the synergies,
reduces the risks and finally avoids all possible deviations from the common targets.
- It analyses the impact of the project on the existing business models in the wireless
communication markets. This activity will be mainly driven worked on during the last
year of the project, when most of the innovations and results of the project will be
available.
Up until project month twenty, it defines the use cases and scenarios relevant for the project
objectives and capable of showing the advantages of the newly proposed technologies. Finally,
it defines an extended 5G architecture and derives related requirements to be worked out in the
other WPs.
2.1.1 Task 1.1: Joint EU/JP vision for global exploitation of 5G technologies, business
models and impact on the eco-system
Contributors: Intel, FHG, CEA, TI, URom, TTech, KLAB, PANA.
Task period: M01 – M36.
Task status: running.
All project partners have been contributing to this important task, the main scope of which is to
keep the alignment, leverage on the complementing strengths and create synergies among the
two different ecosystems of Japan and Europe. This task interacts continuously with the
technical WPs (WP2 and WP3), so to maximize the return-on investment of the consortium
partners and to leverage in the best way on the project results.
Finally, in synergy with WP5, this task not only identifies international events, venues, public
demonstrations and fora relevant for the focused areas of the project, but also takes care of the
synergy with other research projects in the wireless ecosystem.
Output
A detailed list of international impacting events organized and attended by the project in
synergy with other research projects can be found in the project Deliverable 5.2.
Deliverable
Task 1.1 did not have any deliverables planned in the second year of the project.
Del.no. Deliverable name Task no. Due
1.2 Mid-term report on joint EU/JP vision,
business models and eco-system impact
T1.1 M12
1.4 Final report on joint EU/JP vision, business
models and eco-system impact
T1.1 M36
Highlights
Here below a short summary of the most relevant activities of 5G-MiEdge is provided. For a
full list one can refer to D5.2.
- Presentation to the final SPEED-5G EU-funded project international workshop in BT
premises, London March 2018.
- Co-organization with the TWEETHER and ULTRAWAVE EU-funded projects of the
Special Session Workshop “Economics and adoption of millimeter wave technology in
future networks” and of the related panel at the IEEE WCNC conference in Barcelona,
April 2018.
- Co-organization with the EU-funded projects 5GCity, VirtuWind, MoNarch, 5GEx, and
SPEED-5G of the “2nd Workshop on business models and techno-economic analysis
for 5G networks” and related panel, at the EUCNC conference in Ljubljana, June 2018.
Relationship to other tasks
This task feeds WP5.
This task is fed from the work of all WPs, but the tighter link is to WP5.
2.1.2 Task 1.2: Use cases, scenarios and system architecture
Contributors: FHG, CEA, Intel, TI, URom, TTech, KLAB, PANA.
Task period: M01 – M20.
Task status: completed.
This task defines use cases, scenarios, system architecture and related requirements capable of
leveraging on the new enabling technologies developed in the project. Task 2.1 provides the
basis of the work done in all the more technical WPs of 5G-MiEdge.
Task 1.2 successfully ended in project month 20, with the delivery of the project deliverable
D1.3 “System architecture and requirements”.
Output
Based on the work done in year one w.r.t. the finalization of deliverable D1.1 “Use cases and
scenario definition”, in the second year of the project also the deliverable D1.3 “System
architecture and requirements” has been finalized.
D1.3 clarifies the system architecture, which is specified for each of our relevant use cases and
scenarios as defined in D1.1, and the respective requirements needed to be considered in 3GPP
and ETSI MEC standardization process.
Deliverable
Task 1.2 released deliverable D1.3.
Del.no. Deliverable name Task no. Due
1.1 Use cases and scenario definition T1.2 M09
1.3 System architecture and requirements T1.2 M20
Highlights
Finalization and submission of D1.3.
Table 2.1-1 shows the comprehensive table to capture all the most important info related to
the five chosen use cases.
Table 2.1-1 Summary of 5G-MiEdge use cases and scenarios
Use case Scenario Specific application
1.
Omotenashi
service
Ultra-high-speed wireless access in
dense areas, like
- airport,
- train,
- shopping mall
Ultra-high-speed contents download,
Massive video streaming.
2.
Moving
hotspot
High-speed wireless communication
for passengers in vehicles like
- train,
- bus,
- airplane
Pre-cached/pre-fetched contents download
and SNS contents‘ upload from/onto local
MEC server on vehicle
3.
2020 Tokyo
Olympic
Olympic Stadium area and stands
- In the seating area
- At the gates
File download, high definition content
download and sharing,
Immersive reality
4.
Dynamic
crowd
Outdoor hotspot areas like bus stops,
stations and recreation parks, with
dynamic changes of traffic pattern
Public video surveillance and 3D live
video broadcast of Olympic games
5.
Automated
driving
Automotive traffic environments in
urban city
Cooperative perception by exchanging HD
dynamic map information between
vehicles & roadside units
Baseline system architecture
D3.1 defined a high-level system architecture for the 5G-MiEdge project, as described in
Figure 2.1-1.
Figure 2.1-1 High-level baseline system architecture for 5G-MiEdge
This high-level architecture follows a basic architecture of 5G mobile network, as defined in
[TS23.501], and the MEC framework and architecture defined in [MEC003]. The architecture
proposed in [D1.1] is a preliminary version and is subject to refinements and improvement
coming out of the work of the technical work packages, including WP3. Please note that the
N4 reference point is added between SMF and MiEdge RAN to the original figure in [D1.1].
In the following, only some details are provided on the proposed blocks of the architecture,
for a full-fledged description one can read [D1.1] and [D3.1].
Relationship to other tasks
Output of this task feeds to WP2, WP3, WP4, and WP5.
This task is fed from WP2 and WP3.
2.2 Work package 2: Millimeter-wave edge cloud for 5G RAN deployment
paradigm
Contributors: PANA, FHG, CEA, TI, URom, TTech
WP2 aims to provide ultra broadband access with multi-Gbps throughput, while supporting
mmWave backhaul and relay in order to extend coverage area as well as to improve resiliency
against blocking. The key technologies include not only mmWave physical layer design, but
also sophisticated resource allocation (including MIMO and coordinated beamforming),
mmWave high-gain array antenna, mmWave AP deployment design etc.
2.2.1 Task 2.1: mmWave ultra broadband access for highest capacity 5G scenarios
Contributors: PANA, FHG, CEA, TI, URom, TTech
Task period: M04 – M30
Task status: running
This task focuses on mmWave ultra broadband access technologies. In order to achieve multi-
gigabit throughput while dealing with densely populated environment, various techniques such
as spatial multiplexing, multi-link coordination as well as calibration techniques for super high-
speed mmWave access are investigated. Ultra lean signaling/control plane is also developed for
efficient user management in a dense environment.
Output
Spatial multiplexing (MIMO, massive MIMO) for mmWave (Subtask 2.1.1):
o Evaluated the performance of the mmWave MU-MIMO systems using hybrid
beamforming with a line of sight (LOS)-component-only channel model over
stadium scenario with high density user stations (STAs) employing user scheduling
algorithm
Multi-link coordination of mmWave access to control interference and blocking (Subtask
2.1.2):
o Proposed and analyzed techniques for multi-link connectivity, in scenarios without
blocking and in blocking scenarios (due to obstacles) to reduce power consumption
at the mobile side, finding the optimal number of links to be used
Channel bonding and higher order modulation for super high speed mmWave access
including phase noise compensation and pre-distortion (Subtask 2.1.3):
o Proposed and evaluated the phase-noise-robust (PN-robust) channel estimation for
the mmWave MU-MIMO OFDM systems, and clarified the relation between the
mean squared error (MSE) performance and the K factor of Rician channel model
Ultra lean signaling/control plane for mmWave (Subtask 2.1.4):
o Proposed the cooperative WiGig/Wi-Fi multi-user connection management which
has been implemented in the WiGig signage prototype, and successfully verified its
effectiveness through measurement
The work performed in this task has led to the following publications.
o M. Kobayashi, T. Urushihara, N. Shirakata, K. Takinami, "Cooperative WiGig/Wi-
Fi Multi-User Content Delivery System with Application-Centric Connection
Management", SmartCom 2017, Rome, Oct. 2017.
o Y. Chang and K. Fukawa, "Phase Noise Compensation for mm-Wave MU-MIMO
OFDM Systems", SmartCom 2017, Rome, Oct. 2017.
Deliverable
Task 2.1 did not have any deliverables planned in the second year.
Del.no. Deliverable name Task no. Due
2.1 Requirement and scenario definition for
mmWave access, antenna and area planning
for mmWave edge cloud
T2.1,
T2.2,
T2.3
M12
2.2 Design of mmWave ultra broadband access for
5G
T2.1 M30
Highlights
Performance evaluation of the mmWave MU-MIMO for the stadium scenario
Hybrid beamforming (HBF) is considered in the spatial multiplexing for the mmWave to reduce
the hardware complexity of the multiple-antenna systems and to enlarge the coverage of the
access point (AP). In the previous works, the evaluations almost focus on more general cases,
or focus on complex optimization methods [AKS18]. However, we have a very clear use case
and scenario; thus, in this work we focus on the sport stadium use cases, which have very higher
density user stations (STAs) in the band-like stands area. We try to provide quantitative
evaluations for this specific scenario with the specific purposes.
Figure 2.2.1-1 shows an example of the mmWave multi-user MIMO (MU-MIMO) systems,
where there are a lot of antennas elements equipped in the AP side, and there are also several
STAs in this MU-MIMO system. It is not economic to employ a radio frequency (RF) chain for
each antenna element due to the huge number of the elements of the massive MIMO systems.
In the transmitter (Tx) of a HBF system the signal output from each RF chain is power divided
into several routs, and phase shifted by the phase shifters, then transmitted by the antenna
elements (antenna array), which is called the analogue beamforming (ABF) that can
compensate the limited coverage due to the large path loss of the mmWave system with lower
hardware complexity. The same concept also can be used in the receiver (Rx), where the power
dividers are changed by the power combiners. The power combiner/divider can be implemented
with the directional couplers. With ABF the AP can control the directivity of each antenna array
at the AP toward each STA, and the STAs also can control the directivities of their arrays toward
the AP. By properly selecting the simultaneously transmitted STAs from the active STAs, the
ABF can enlarge the power of the diagonal entries of the effective channel matrix including the
ABFs and suppress that of the non-diagonal entries. In down-link (DL) systems, the simple
precoding (or post-coding in up-link systems) likes zero forcing (ZF) can be employed to pre-
code the signal inputted into each RF train as the digital beamforming (DBF) to cancel the
remained inter-user interference (IUI) between the STAs.
When the number of STAs is larger than that of the RF trains in the AP, as mentioned earlier,
it is necessary to select numbers of STAs from the active STAs with the user selection algorithm.
Two kinds of user scheduling algorithms at the AP side are considered in the evaluation:
1) Round Robin (RR) algorithm: that selects the STAs in turn, which did not consider the level
of IUI between the selected STAs. 2) Beam RR algorithm: that first selects the ABF beams in
turn, then selects one STA in turn during the group of STAs in the coverage of each selected
ABF beam. The detail of Beam RR is described as follow: 1) In the initialization phase of Beam
RR, the AP has to groups the STAs into the group of each ABF beam pattern. In each Tx timing,
2) the AP selects the ABF beam patterns in turn for each RF chain, which has a non-null STA
group, and then 3) for each ABF beam, the AP selects one STA in turn from the group of this
ABF beam.
Computer simulation is conducted to show the performance of HBF MU-MIMO system with
the user scheduling algorithms over the line of sight (LOS)-components-only channel model.
The simulation condition is shown in Table 2.2-1 that is a DL scenario with one AP and several
(10-50) STAs. Each STA has one antenna array, and the AP has 4 antenna arrays. Each array
in the AP has 16 antenna elements that are constructed on a square uniformly. The array of each
STA has 4 antenna elements that are linearly and uniformly constructed (as the example shown
in Figure 2.2.1-1). The coverage area is shown in Figure 2.2.1-2, where there is an AP with
height ℎAP and a square coverage area of the STAs with side length 𝐿STA, and the minimum
horizontal distance between the AP and the area is 𝑑AP. The STAs are uniformly distributed in
this area. Because the number of RF chains of the AP is 4, the user scheduler has to always
select 4 STAs (Or the number smaller than 4; however, here we only consider the case of 4 for
simplicity) to transmit. There are 10 (shown in Figure 2.2.1-3) and 5 predesigned ABF beam
patterns for the AP and the STAs, respectively, where the beam patterns of the STAs is the
same with the azimuth pattern of the AP (shown in Figure 2.2.1-3 (a)). The ABF beams in the
AP are designed to cover the aforesaid coverage area. Figure 2.2.1-4 shows the simulation
Figure 2.2.1-1 An example of spatial multiplexing for mmWave MU-MIMO systems
results, where “w/ ZF” denotes the results with not only ABF but also DBF (here ZF is
employed), while “ABF only” denotes the results without DBF thus the IUI is remained. We
can see, with ZF the system capacities are improved to larger than twice. And the results of
Beam RR are better than the normal RR; especially, when ZF is employed the improvement is
larger than 20%. With Beam RR and ZF the system capacity of more than 17 bps/Hz can be
achieved when the number of STAs is fewer than 50. Note that the system capacities do not
increase with the total number of STAs, for the scheduling algorithms are all RR-based
algorithms that cannot obtain the benefit of multiuser diversity. When the number of STAs is
smaller, the system capacity becomes larger, because of the higher probability to select the
STAs with lower IUI levels.
Table 2.2-1 Parameters of the evaluated hybrid beamforming MU-MIMO
Height of AP, ℎAP 3 m
Min. distance from AP, 𝑑AP 2 m
Length of the side of coverage area, 𝐿STA 10 m
No. of RF chains of AP, 𝐾 4
Tx power of each RF chain, 𝑃t 10 dBm
No. of elements of each ABF array in the AP, 𝑁AP 16 (4×4 planar array; distance between
elements: 0.5 wavelength)
No. of STAs, 𝐾 10-50
No. of RF chains of each STA 1
No. of elements of ABF array in each STA, 𝑁STA 4 (linear array; distance between
elements: 0.5 wavelength)
Rx noise power of each element −74 dBm (noise figure of LNA: 6 dB)
No. of ABF beam patterns of AP 10 (5 [azimuth] × 2 [elevation])
No. of ABF beam patterns of STA 5
User scheduling methods Round Robin (RR), Beam RR
Figure 2.2.1-2 The evaluated coverage area of the mmWave MU-MIMO system
It is expected that there is fewer multiuser diversity in the mm-Wave MU-MIMO systems with
ABF that will harden the wireless channels, and reduces the fluctuation level of fading, which
also reduces the user selection diversity (multiuser diversity) gain. RR-based user scheduling
algorithms have the best fairness between the STAs, and Beam RR algorithm can guarantee the
reduction of the IUI level between the selected STAs. In the previous works, again, the
evaluations almost focus on more general cases, or focus on complex optimization
methods [AKS18]. The contribution of this work is that we focus on the more practical scenario
of sport stadium, in which there are higher density STAs in the band-like stands area. We
provided the quantitative evaluation for this specific scenario with the specific purposes. In
future works, we are going to evaluate the scenarios of mm-Wave MU-MIMO with multi-AP,
in which we also have to consider the inter-cell interference (ICI) between the APs.
Multi-link coordination of mmWave access to control interference and blocking
During the second year, we investigated more in detail how the exploitation of multi-link
communications can help in scenarios with blocking and without blocking. We suppose the
mobile terminal is endowed with an antenna array, so that it can perform beamforming toward
multiple APs. Multi-link communications, in the context of mmWave, can be useful for two
different reasons:
(a) (b)
Figure 2.2.1-3 ABF beam patterns of AP: (a) azimuth patterns, (b) elevation
patterns
Figure 2.2.1-4 Average system capacity vs. number of users
1) They can improve the spectral efficiency using the same transmit power, or reduce the
transmit power while guaranteeing the same spectral efficiency
2) They can reduce the effect of blocking events due to obstacles or beam collision
Without blocking, we formulated the problem as the minimization of the power consumption
with the constraint of maintaining a minimum spectral efficiency 𝑅min, finding a condition for
the convenience of using 𝑁 links out of Nmax available AP’s. Without going into the details, in
Figure 2.2.1-5 we show the numerical results in terms of power consumption of the mobile
device vs. the number of available AP’s Nmax around the user. For every Nmax, we derived a
strategy to find out the optimal number of links Nopt out of Nmax (the details will be in D3.2).
The results are shown for different values of Rmin , and are averaged over several channel
realizations (i.e. AP’s positions). The numerical results show that the largest gain is obtained,
in every case, by having 2 available AP’s instead of 1. Moreover, increasing the number of
AP’s above a certain threshold, does not yield significant improvements, considering also the
complexity of transmitting over more links. Furthermore, the higher gain in increasing Nmax is
obtained for higher values of Rmin.
Figure 2.2.1-5 Average Transmit power vs. N_max
In scenarios with blocking, the channel is considered to have an ON/OFF behavior, so that the
aim is to minimize the power consumption while maintaining a minimum average spectral
efficiency R̅ greater than Rmin, similar to previous works such as [BAR17] and [BCM17]. In
particular, we assume that the user, before transmitting, checks which channels are available,
and transmits over the Nopt best available links. For this investigation, we used the blocking
model of [BAI14], which takes into account the density of the obstacles, their average
dimension and the distance between the user and the AP. In Figure 2.2.1-6, we show a numerical
example in a scenario with a user surrounded by 4 AP’s. The results are shown in terms of
power consumption vs. the density of obstacles. As in the non-blocking scenario, we can note
that the higher gain is achieved using 2 links instead of 1, since the overall probability of not
finding available AP’s is decreased significantly. Passing from 2 to 3 or more links still yields
some power gain but this improvement does not justify the additional complexity.
Cooperative WiGig/Wi-Fi multi-user connection management
In order to evaluate the effectiveness of the ultra-lean signaling/control scheme, the cooperative
WiGig/Wi-Fi multi-user content delivery system was developed. The system employs the
heterogeneous architecture where WiGig is used for high speed content download and Wi-Fi is
responsible for authentication and connection management. This reduces overhead due to
control signaling at the mmWave frequency band.
Figure 2.2.1-7 illustrates a block diagram of the prototype system. It consists of WiGig modules,
a Wi-Fi access point (AP), a local storage and an AP controller (APC). While the user terminal
is connected to Wi-Fi, the APC periodically collects link quality via Wi-Fi for connection
management [Figure 2.2.1-7 (1)]. When the user terminal requests WiGig connection, the APC
connects the terminal to one of the WiGig modules which provides the best link quality [Figure
2.2.1-7 (2)]. When the user requests the content download from the application, the user
terminal starts downloading from the local storage via WiGig [Figure 2.2.1-7 (3)].
Figure 2.2.1-6 Transmit power vs. density of obstacles
To minimize latency due to WiGig connection establishment, the application-centric
WiGig/Wi-Fi cooperative connection management is newly introduced. As illustrated in Figure
2.2.1-8, the users’ behavior is predicted by monitoring the state transition of application, which
enables to establish WiGig connection before the download request from user terminals (i.e.
WiGig connection is established when the application transits (a) content catalogue to (b)
content details).
Figure 2.2.1-9 shows the measured effective throughput. In this measurement, three users
downloaded a 2 GB content per user with 2 second delay interval. Throughput via Wi-Fi (20
MHz BW, 2x2 MIMO) was also measured for comparison. The maximum throughput via Wi-
Fi was about 100 Mbps, which is deteriorated as the number of users increases. On the other
hand, the maximum throughput of WiGig reached about 1.8 Gbps, achieving the 2 GB content
download in 10 sec for all users. Figure 2.2.1-10 shows the latency between the user’s download
request and download start. In the conventional method, the user’s download request triggers
WiGig connection establishment which introduces more than 0.2 sec latency before starting the
download. The proposed method, on the other hand, establishes the WiGig connection before
the download request with about 90% probability, which validates the effectiveness of the
proposed method.
Figure 2.2.1-7 Cooperative WiGig/Wi-Fi multi-user content
delivery system
(a) ContentCatalogue
(b) ContentDetails
(c) DownloadingStatus
(d) DownloadCompleted
Start
Connect Wi-Fi
Select Content /Start Connect WiGig
Back Button /Disconnect WiGig
Download Button
Finish Download /Disconnect WiGig
Cancel Button/Disconnect WiGig
Back Button
Play Button
(e) ContentPlayback
Back Button
Figure 2.2.1-8 Proposed connection management using state transition of application
(a) Wi-Fi
0
0.1
0 10 20 30470 480 490 500
Th
rou
gh
pu
t [G
bp
s]
Tim e [ s]
User1
User2
User3
(b) WiGig
0
1
2
0 10 20 30470 480 490 500
Th
rou
gh
pu
t [G
bp
s]
Tim e [ s]
User1
User2
User3
Figure 2.2.1-9 Measured throughput
Relationship to other tasks
T2.1 takes inputs from Task 1.2 on the use cases.
The task will provide outputs for proof of concept in WP4 as well as standardization
and dissemination activities in WP5.
2.2.2 Task 2.2: Design of mmWave antenna for specific scenarios toward Tokyo
Olympic 2020
Contributors: TTech, FHG, TI
Task period: M01 – M24
Task status: completed
This task focuses on the development of mmWave antennas for various use case and selected
scenarios, together with WP1, toward Tokyo Olympic 2020. The antenna prototypes will
include a planar array antenna for gate system (mmWave shower) where large contents such as
videos, application software etc. are downloaded instantaneously at the entrance gate. High gain
antennas will also be investigated for mmWave backhaul.
Output
Designed mmWave antennas for gate system (mmWave shower)
Designed mmWave antennas for backhaul link
Presented the CATR system in D2.3, which will be used for experimental evaluation
of the mmWave antennas in the third year
Deliverable
0
10
20
30
40
50
60
70
80
90
100
0 0.2 0.4 0.6 0.8 1
CD
F [%
]
Latency [s]
Proposed
Conventional
90
Figure 2.2.1-10 Latency of download start (The horizontal axis indicates the latency
between the user’s download request and download start)
Task 2.2 released deliverable D2.3.
Del.no. Deliverable name Task no. Due
2.1 Requirement and scenario definition for
mmWave access, antenna and area planning
for mmWave edge cloud
T2.1,
T2.2,
T2.3
M12
2.3 Design of mmWave antennas for 5G enabled
stadium
T2.2 M24
Highlights
mmWave antennas for gate system (mmWave shower)
Figure 2.2.2-1 illustrates the 60 GHz-band GATE (Gigabit Access Transponder Equipment) to
be equipped as the fixed terminal in public areas such as in corridors and escalators located in
stations and departments stores, and at other locations. When a user holding a mobile terminal
passes through a GATE, gigabit access is available in its clearly defined coverage area named
here as the “compact-range”.
In order to realize GATE system, large array antennas with high efficiency, wide bandwidth
and high gain are required. 32×32-element array was fabricated by diffusion bonding of thin
aluminum plates as shown in Figure 2.2.2-2. Electric field distribution is calculated by
measured near field distribution on the antenna aperture. The field distribution is shown in
Figure 2.2.2-3. Desired electric field intensity of more than -1.3 dBV/m is obtained up to 14 m
away from the antenna.
Reception Zone
Large array antenna
for the access point
Multi-Gb/s
Figure 2.2.2-1 Concept of gate system (mmWave shower)
mmWave antennas for backhaul link
To utilize the frequency resources more efficiently, full duplex, also known as Directional
Division Duplex (DDD) is adopted to double the transmission capacity compared to the
conventional Time Division Duplex (TDD) or Frequency Division Duplex (FDD). That is, a
wireless terminal in DDD will use two independent antennas operated in same frequency and
with same polarization to realize simultaneous bidirectional communication as shown in Figure
2.2.2-4.
In this study including the past study [IEEETAP16], a double-layer waveguide slot array is to
be designed in the 40 GHz band. Figure 2.2.2-5 shows the prototype antennas with V-
polarization and horizontal arrangement. Isolation between Tx and Rx antennas were measured
using a signal generator and a spectrum analyzer. Measured isolation is shown in Figure 2.2.2-6.
Isolation higher than -76 dB was confirmed.
Figure 2.2.2-2 Fabricated 60-GHz band 32x32-element slot array antenna made of aluminium
2 4 6 8 10 12 14
Position z (m)
-0.5
0
0.5
Posi
tion
x (
m)
0
5
10
15
20
25
Ele
ctr
ic f
ield
in
tensi
ty (d
BV
/m)
Figure 2.2.2-3 Near field electric field distribution
Measurements of antennas by TIM in CATR
The antenna measurements is an essential activity required to verify if the actual performance
match the requirements. In particular, the radiation pattern has to be measured and verified
because it represents the fundamental input of any design tool used in the cellular coverage
prediction.
CATR (Compact Antenna Test Range) enables large antenna measurement where obtaining
far-field distance is infeasible using far-field antenna measurement. Figure 2.2.2-7 shows
Same Polarization
Same Polarization
Gain: 32dBi
Gain: 32dBi
Space Propagation loss 1km: -125dB@40GHz
Figure 2.2.2-4 Full duplex, directional division duplex
Figure 2.2.2-5 Prototype of the aluminium antennas
39 39.5 40 40.5 41 41.5
70
80
90
100
Frequency [GHz]
Isola
tion [
dB
]
ANT only + wall + wall + radome
Figure 2.2.2-6 Frequency characteristics of isolation
sketched images of the CATR installed in TIM and its main characteristics is shown in Table
2.2-1.
Table 2.2-2 Main characteristics CATR
Frequency band GHz 3 – 110
Quiet zone shape Circular
Quiet zone dimensions (Ø x depth) m 1 x 1
Maximum Cross Polarization dB -30
Amplitude total variation dB 2.0 @ 10 GHz Worst Case
Amplitude Taper dB 0.5 (20 – 40 GHz)
Amplitude ripple dB
± 0.75 (< 15 GHz)
± 0.4 (15 - 50 GHz)
± 0.5 (50 – 110 GHz)
Phase ripple degrees
± 8° (< 15 GHz)
± 4° (15 - 50 GHz)
± 5° (50 – 110 GHz)
Reflector dimensions cm 218x218
Focal distance cm 350
Surface accuracy (QZ area) µm <80 peak to peak
Surface accuracy (QZ outside) µm <125 peak to peak
Max AUT Weight kg 160
Relationship to other tasks
Figure 2.2.2-7 Sketched images of the CATR installed in TIM
The specification of the antennas are derived by the use cases shown in T1.2.
The designed antennas are used as a part of a testbed of Task 4.2.
2.2.3 Task 2.3: Site specific deployment of mmWave edge cloud with
caching/prefetching and relay
Contributors: FHG, URom, CEA, TTech
Task period: M01 – M26
Task status: running
This task is tightly connected to WP1 and focuses on deployment of mmWave edge cloud
systems in specific scenarios. It also builds the foundation for hardware development and
outdoor deployment in WP4.
At first the deployment method is defined to match the scenario requirements regarding
throughput and user density and to provide caching, prefetching and relay capabilities. After
the successful deployment, the radio resources are dynamically optimized. Based on the current
and predicted traffic load, the backhaul links and allocated channels are adapted.
Output
Area planning of mmWave edge cloud with traffic forecast
Investigated the deployment and optimization of a multi-RAT heterogeneous network.
Inteference management in mesh backhaul
Dynamic resource allocation and prediction aware load matching
Proposed and analyzed techniques for dynamic resource allocation for computation
offloading, based on stochastic optimization, to find the best trade-off between computation
queues stability and power consumption
Mesh Backhaul Network
The network consists of multiple nodes that are interconnected via mmWave links, forming a
dynamic meshed backhaul network. Some of the nodes offer mobile edge cloud functionality
like caching and computation, while others act as network relays. To enable decentralized
management, a SDN controller orchestrates the network.
Antenna and System Prototyping
In order to achieve the desired link range, the frontends are enhanced with a passive reflect
array. This array offers a gain of 26 dBi and narrows the beam to 6°. When mounted directly
before the internal antenna there is a high increase in transmit power, allowing link distances
of over 200 meters.
Figure 2.2.3-1 shows the prototype design of a mesh backhaul antenna, it has three separate
compartments, two for mmWave antennas, the third for a computation device and can be
mounted on a variety of surfaces, e.g. on a wall like shown here.
Figure 2.2.3-1 Mesh backhaul antenna design prototype
Our developed system consists of nodes, each offering the following features.
Node Features
Each node offers multiple mmWave links equipped with a 60 GHz WiGig interface, multiple
Gpbs throughput, a range of over 200 m and dynamic alignment capabilities. In addition there
are MEC features available, like storage/caching and computation.
Orchestration Features
The controller is capable to adapt the routes and also dynamically change the alignment of each
link, counteracting events like blockage and optimizing the overall network performance. In
addition it is possible to power off unused nodes and power them on again when needed.
MEC Features
With the MEC capabilities of the nodes, it is possible to cache and store requested content,
reducing the load on the core network and increasing responsiveness. The computation
allows for running tasks with high computation power, low latency and reduced power
usage of mobile devices.
The work performed in this task has led to the following publications:
G. Ghatak, A. De Domenico, and M. Coupechoux, ”Coverage Analysis and Load Balancing
in HetNets with mmWave Multi-RAT Small Cells, IEEE Trans. on Wir. Com., May 2018.
G. Ghatak, A. De Domenico, and M. Coupechoux,” Accurate Characterization of Dynamic
Cell Load in Noise-Limited Random Cellular Networks” IEEE VTC Fall 2018.
Deliverable
Task 2.3 did not have any deliverables planned.
Del.no. Deliverable name Task no. Due
2.1 Requirement and scenario definition for
mmWave access, antenna and area planning
for mmWave edge cloud
T2.1,
T2.2,
T2.3
M12
2.4 Method of site specific deployment of
mmWave edge cloud
T2.2 M26
Highlights
Modeling and optimization of multi-RAT HetNet
In the first year, we have focused only on coverage aspects (i.e., the SINR distribution and
related metrics). In the second year, we have modelled the cell loads to optimize the small cell
deployment in order to satisfy the specific load requirements and show how tier and RAT
selection biases can improve the user average throughput.
In this study, we have considered a multi-user system where the users share the available radio
resources according to a round robin policy.
Past researches have characterized the load by using the average number of associated users in
a cell; however, we have provided a more realistic characterization by considering a dynamic
traffic model. In addition, in contrast with the literature, we have optimized the user throughput
while considering SINR outage constraints as well as overloading constraints.
First, we study the system performance from a network perspective. In Figure 2.2.3-2, we show
the minimum deployment density (number of small cell per macro cell) required for a given
traffic density such that outage and overloading probabilities constraints are meet. The more
stringent the constraints are, the more small cells the operator should deploy. When the traffic
density is low, the outage probability is the limiting constraint and accordingly, the minimum
deployment density is the one required to maintain coverage. However, as traffic density
increases, overloading probability is determining.
Next, we analyze how cell and RAT selection biases (QT and QR) affects the user performance.
Figure 2.2.3-2 Minimum required deployment density for a given traffic density
Figure 2.2.3-3 Effective user throughput for λ_S/λ_M =200
In very dense small cell deployments (see Figure 2.2.3-3), the users do not suffer excessive
outage probability even in the case of high tier biases. In this case, QR should be high enough
to maximize the mmWave association probability. In case of λS
λM = 200, this results in a
maximum throughput of around 30 Gbps for QT = 10 dB and QR = 6 dB.
In moderate small cell deployments (Figure 2.2.3-4), high values of QT are desirable to offload
traffic from overloaded MBSs. However, as the small cell ranges increase, mmWave becomes
unattractive for users at the small cell edges. We can observe that increasing QR beyond a certain
limit pushes these users in outage thereby decreasing the effective throughput. The maximum
average throughput in this scenario, considering the regime of biases where the macro cell tier
is not overloaded, is 10 Mbps at QT = 6 dB and QR = 3 dB.
Interference management in mesh backhaul networks
Optimization for interference management of mmWave mesh backhauling including node
placement, channel selection and beam steering is NP-hard due to the large number of involved
unknown parameters. Instead, we are proposing a heuristic approach which can achieve
reasonable performance via three steps i.e. mesh node deployment, channel assignment and
best beam selection. Unfortunately, there are still interference bottlenecks which prevent
highest performance of mmWave backhaul links. To solve that issue, the transmission power
of each node is optimized in a centralized manner to maximize the SINR (Signal-to-
Interference-and-Noise Ratio) of the bottleneck link. In summary, we proposed a dynamic
interference management scheme for MmWave Mesh Backhaul Networks (MMBN) via
suitable deployment of mmWave APs, channel assignment, beam selection and power
allocation. In MMBN, we assume a mesh topology such that each mmWave AP has several
interfaces for connecting with the GW via multi-hop relay. Owing to this topology, the system
becomes robust against propagation loss via route diversity, and attains higher relaying capacity
Figure 2.2.3-4 Effective user throughput for λ_S/λ_M =50
via route multiplexing, at the expense of slightly higher complexity. The overall architecture of
MMBN is depicted in the below figure. The LTE macro cell basically is responsible for the C-
plane of the network via which control signaling to dynamically construct MMBN is undergone.
MmWave GW is the only small cell base station (AP) with high-speed wired backhaul working
as the gateway for MMBN to the core network. MmWave APs are the other small cell base
stations (APs) without wired backhaul. They play both roles as the sink receiving information
from the GW and the regenerative relay nodes (except the end node on a route). They have both
access interfaces to connect with UEs and backhaul interfaces to connect with other adjacent
mmWave APs. Regenerative relaying is conducted by the latter interfaces. Interference
management for MMBN can be realized by the steps summarized in Figure 2.2.3-6.
Figure 2.2.3-5 Architecture of MMBN
Figure 2.2.3-6 Four steps to reduce interference
1) Zigzag deployment both horizontally and vertically
2) Switch of channel per every two hops
3) Beam steering to exploit multipath route e.g. reflection from wall
4) Optimal power allocation to maximize the worst bottleneck link
Numerical analysis was conducted to evaluate the effectiveness of the proposed algorithm via
ray-tracing simulation. Simulation environment and parameters are summarized in the below
figure and Table 2.2-3. Terrain model is made in reference to Shibuya, a crowded station in
Tokyo and mmWave APs are deployed approximately every 50m based on lampposts
placement. The conventional approach is defined as linear deployment of mmWave APs (green
points), while the proposed approach applied zigzag deployment (red points). Simulation
parameters are based on IEEE 802.11ad (WiGig) standard. The number of paths and APs are
set as 25 and 28 respectively. Dielectric constant and conductivity of concrete.
Table 2.2-3 Simulation parameters
We evaluate by the simulation assuming routes are extended radially by the arrows in the above
figure. We compute throughput of backhaul links by mapping MCS and the achieved SINR.
Due to multi-hop relay, the performance of the network is evaluated at the bottleneck link. We
compare with the conventional approach with linear node deployment and therefore channel
switching per every one hop. Numerical results are summarized in Figure 2.2.3-7, where the x-
axis shows the AP distance from GW and the y-axis shows the bottleneck capacity up to that
distance. The maximum achievable throughput of each link is upper-bound by 6.76 Gbps
according to IEEE 802.11ad standard. The figure shows that the conventional approach cannot
reduce interference effectively because mmWave APs were deployed linearly. On the other
hand, the proposed approach can alleviate interference up to 250 m from mmWave GW.
We conducted another simulation with route multiplexing, where certain APs are assumed to
be sink nodes. For each sink node, we assumed source information from the GW are relayed to
each sink node via two-route multiplexing, thus the maximum throughput becomes 13.5 Gbps.
As results, throughput achieved only about 10 Gbps in conventional approach compared to 13.5
Gbps in the proposed approach. The result again confirms the effectiveness of the proposed
mechanism for interference management in even multi-route multiplexing scenario. It is due to
intra-route and inter-route interference that could not be tackled by the conventional scheme.
Parameter Value
Carrier freq. 60 GHz
Bandwidth 2 × 2.16 GHz
Link Down
Tx power 10 dBm
Antenna gain 32 dBi
Halfwidth of azimuth 5 deg
Halfwidth of elevation 5 deg
Propagation loss 20 log10(4𝜋𝑑 𝜆⁄ )
Dielectric constant 3.3
Conductivity 0.015 S m⁄
Noise power density −174 dBm/Hz
Noise factor 10 dB
Dynamic resource allocation, considering the best trade-off between queues and power
consumption
While in the first year, we concentrated on a static optimization, in the second year we focused
on dynamic resource allocation in MEC environment with mmWave communications, building
on stochastic optimization tools [NEE10] for the offloading of computationally heavy
applications. In particular, we extended the approach of [MAO17] to a scenario with multiple
AP’s and MEH’s, where tasks continuously arrive with unknown probabilities, and can be split
into a local execution at the mobile side and a remote execution at the MEH side. Each user has
a local computation queue and a remote computation queue. The aim of the optimization
algorithm is to assign users to MiEdge resources (AP and MEH), choose the split between local
and remote execution and allocate resources to find a trade-off between the power consumption
for the offloading and the stability of the computation queues.
The optimization is performed in a per-slot fashion, solving a deterministic problem in each
time slot. Building on the methodologies of [NEE10], it can be shown that the mean rate
stability of the computation queues is guaranteed, while minimizing the long-term average
power consumption. The power consumption we consider is the one necessary to offload part
of the application to the MEH, plus the one used to perform local computations at the mobile
side. Moreover, due to the binary nature of the assignment problem, we devised a low
complexity algorithm based on matching theory with transfers to associate users to MiEdge
resource.
The association process should be ideally performed in each time slot to achieve optimal results,
but, due to its complexity, we show how the algorithm performs when the assignment is
performed every 𝑁 time slots, with 𝑁 = 10, 50, 100. Results are compared with an SNR-based
association, in which every user is assigned to its best AP in terms of signal to noise ratio.
Using stochastic optimization tools in this context, we can guarantee the stability of the queues
and the long-term average power consumption, finding an optimal trade-off between average
Figure 2.2.3-7 (a) w/o route multiplexing
(b) w/ route multiplexing
power consumption and average delay. Indeed, in general, stabilizing the computation queues
can lead to an unnecessary power consumption. Instead, taking into account both terms can
achieve optimal results in terms of long-term average power consumption and queue stability.
The variables of the optimization problem are the association variables (binary), the bandwidth,
the transmit power, the scheduling policy at the MEH side, and the local CPU cycle frequency
used for local execution. Once the association is found, closed forms solution are used for
resource allocation by each AP-MEH pair in each time slot. The exemplary scenario used for
numerical results is shown in Figure 2.2.3-8.
The scenario is composed by 5 UE, 3 AP’s and 3 MEH’s. In the figure, the blue arrows show
the SNR-based association. The results are shown in Figure 2.2.3-9 and Figure 2.2.3-10, in
terms of average power consumption and average queue length, as a functions of 𝑉, a parameter
that weights the power consumption and represents the trade-off between computation queue
length (or, equivalently average delay) and long-term average power consumption. However,
changing the user association at every time slot means to transfer the state of the application
running in one MEH to another MEH, so it is a very complex operation in a real context. To
reduce the complexity, we show the performance when the matching algorithm is used every
slot or every 100, and 500 slots. Moreover, we compare these results with the SNR-based
association. Note that, as 𝑉 increases, the average queue length increases as well, as illustrated
in Figure 2.2.3-10. The long-term average power consumption vs. 𝑉 is plotted in Figure 2.2.3-9,
showing its optimal values for the highest values of 𝑉. Note that, for such values, the long-term
average power consumption reaches the optimal value while stabilizing the queues. We can
Figure 2.2.3-8 Scenario
also note that using the strategy of optimizing the association every 100 time slots achieve
results close to the ones in which the association is changed in each time slot. This suggest that
this lower complexity strategy can be used without degrading performance. In general, we
observed that our strategies incorporating both SNR and computational state outperform
conventional user association methods based only on SNR.
Figure 2.2.3-9 Long-term average power consumption vs. 𝑽
Figure 2.2.3-10 Average queue length vs. 𝑽
Relationship to other tasks
There is a close connectivity between Task 2.3: Site specific deployment of mmWave
edge cloud with caching/prefetching and relay and Task 1.2: Use cases, scenarios and
system architecture.
Task 2.3 is based on the specifications of scenarios and use cases that are developed in
Task 1.2, yet those specifications are shaped and presented to be usable for the
deployment of an mmWave edge cloud system.
2.3 Work package 3: Design of 5G liquid edge cloud for user/application centric
orchestration
Contributors: KLAB, CEA, URom, TTech, Intel
This WP focuses on designing control signalling for joint radio and computation resource
orchestration algorithm for distributed mmWave edge cloud of 5G wireless heterogeneous
networks. It is composed of mmWave edge cloud integration into 5G mobile network, context
information management for traffic map prediction and user/application centric orchestration
to realize 5G liquid edge cloud.
2.3.1 Task T3.1: Integration of mmWave edge cloud into 5G cellular networks with inter
operable control plane
Contributors: KLAB, Intel.
Task period: M04 – M32.
Task status: running.
This task focuses on integration mmWave edge cloud into the 5G mobile network. Control
signalling is analyzed so to better serve contents and services, which are deployed by utilizing
edge cloud in the network in a distributed manner. Signalling is enhanced for accessing
advanced contents and services, and for reaching an optimal radio/computation/storage
resource utilization. The following phased approach has been planned for this task:
First year: Define architecture baseline for mmWave edge cloud integration into 5G
mobile network.
Second year: Detail the signalling design for access to contents and services, and an
optimal resource utilization and performance evaluations by the use of
simulations.
Third year: Total evaluations.
Output
The system architecture and control signalling were designed for access to contents
and services. As an example, some of the control signaling procedures are reported
here below:
1) Initial access to the service on MEH (Mobile Edge Host),
2) MEH resource compensation at resource shortage event, and
3) Path switching to the optimal MEH based on user movement
Related talks on 1), 2) and 3) were delivered at the panel sessions in IEICE SmartCom
2017 and the general conference 2018.
Simulations were performed on 2) and 3), and the results showed a very good efficiency for
all of them.
Submissions of patent applications and papers are planned on the related topics in the
third year.
Deliverable
Task 3.1 contributed to deliverable D3.1.
Del.no. Deliverable name Task no. Due
3.1 Architecture of mmWave edge cloud and
requirement for control signalling
T3.1,
T3.2,
T3.3
M18
3.2 Integration of mmWave edge cloud into 5G
cellular networks
T3.1 M32
Highlights
The deliverable D3.1 has been finalized.
In this second year, we studied architecture and signalling for access to services on Mobile
Edge Host (MEH).
System architecture definition
The system architecture was defined with integration of 3GPP and non-3GPP access based on
the high level architecture defined in D1.3 in cooperation with WP1 as shown in Figure 2.3.1-1.
This is the configuration of which the access point (AP) is not connected with the 5G core
directly in order to deploy MEH also on the non-3GPP access side and just MEHs have
interworking each other between both sides. In addition, it is assumed that the facilities on the
non-3GPP access side are deployed as the 5G operator’s own or entirely/partially leased to the
5G operator to be operated under the 5G operator’s control.
Figure 2.3.1-1 5G MiEdge System Architecture
with interworking between 3GPP and non-3GPP networks
Signalling for the initial access to the service on MEH
UE AP GW MEH
RAN UPF MEH
MiEdge RAN for 5G
MiEdge RAN for non-3GPP access
UPF
AS(DN)
AMF SMF NEF
MSF
NL2 NL2
NL1 NL3NL4
NL5N9
Mp3
N3
N2
N1
N4
N4
N11 NNEF
N6
The initial access procedure for the service is shown in
Figure 2.3.1-2 based on the basic signalling defined in the first year, along definitions in
3GPP [TS23.502] and ETSI MEC [MEC003] on the edge computing. The MEC network is
deployed overlaying the 5G mobile network. Then each is controlled by its management
entity(s) independently from management of the 5G network basically. Therefore, service
deployments to MEHs are performed inside the MEC network. Routing rules are applied to
the 5G mobile network, provided by the management entity(s) of the MEC network, e.g. the
orchestrator, to the Policy Control Function (PCF) to access services on MEH. When a UE
(User Equipment) requests session establishment for a service, the 5G mobile network is not
aware of details of the request from the UE because the application layer of that packet is not
inspected. Instead of inspecting details of the session establishment request message,
slice/service type (SST) and/or data network name (DNN) are indicated by the UE for the
service. Those information will help to set up routing and QoS policy in the 5G network for
that service. After the session establishment for the service, the UE performs DNS (Domain
Name System) procedure to resolve IP address of host server (MEH) for the service by using
the service name, for example. That packet for DNS procedure is sent for DNS server in the
global (cloud) network. A MEH (the default MEH for the session) responds IP address of the
serving MEH to the UE by intercepting the DNS packet when the requested service is
matched with the DNS record of services on the MEH. Then the UE is able to receive the
service from the serving MEH.
UE RAN AMF UPF SMF PCF UDMDefau lt
MEH
6 . UPF selection
1 . PDU Session Estab lishm en t Request
2 . Nam f_ PDUSession _ CreateSMCon text Request
3 . Sub scrip tion retrieval
4 . Nam f_ PDUSession _ CreateSMCon text Response
5 . Session Managem en t Policy Estab lishm en t
7 . N4 Session Estab lishm en t Request/Response
8 . Nam f_ Com m un ication _ N1N2MessageTran sfer
9 . N2 PDU Session Request
10 . AN-specific resou rce setu p (PDU Session Estab lishm en t Accep t)
1 1 . N2 PDU Session Request Ack
Up lin k Data (DNS resolver request)
Dow n lin k Data (DNS resolver response)
SST/DNN& cell I D
Retrieve rou tin g ru le from PCF
UPF selection and rou tin g ru le ap p lication
Notification of th e selected UPF to RAN
12 . Nsm f_ PDUSession _ Upd ateSMCon text Request/Response
13 . N4 Session Mod ification Request/Response
ServingMEH
DNS record check
MEC service access ( service p rovision in g)
5 G m ob i le n e tw or k M EC n etw or k
Orchestra tor
The servin g HEH is d eterm ined for each service in sep arated p rocedu re from session estab lishm en t b y th e UE in ad vance. Then DNS record is
u p d ated on the d efau lt MEH.
NEF
Rou tin g p olicy
DNS record u p d ate
I P add ress is resolved by the requested serv ice nam e basica lly .
Figure 2.3.1-2 Initial access procedure for MEC service
Signalling for MEC resource compensation in the event of resource shortage
The computation and storage resources should be prepared sufficiently on MEHs based on long
term traffic/load estimation, but not too much. So, unexpected spike of service requests may
exceed the prepared resources on the MEH for the service. Two types of compensation methods
are considered for that computation/storage resource shortage. Signalling examples for those
compensations are shown in Figure 2.3.1-3.
Assumed compensation methods of service resources on a specific MEH
a) Perform the next task on an alternate MEH
b) Perform the next task by shifting spare resources from a lower prioritized service on
the same MEH
UE RAN AMF UPF SMF PCF UDMDefau lt
MEH
6 . UPF selection
1 . PDU Session Estab lishm en t Request
2 . Nam f_ PDUSession _ CreateSMCon text Request
3 . Sub scrip tion retrieval
4 . Nam f_ PDUSession _ CreateSMCon text Response
5 . Session Managem en t Policy Estab lishm en t
7 . N4 Session Estab lishm en t Request/Response
8 . Nam f_ Com m un ication _ N1N2MessageTran sfer
9 . N2 PDU Session Request
10 . AN-specific resou rce setu p (PDU Session Estab lishm en t Accep t)
1 1 . N2 PDU Session Request Ack
Up lin k Data (DNS resolver request)
Dow n lin k Data (DNS resolver response)
SST/DNN& cell I D
Retrieve rou tin g ru le from PCF
UPF selection and rou tin g ru le ap p lication
Notification of th e selected UPF to RAN
12 . Nsm f_ PDUSession _ Upd ateSMCon text Request/Response
13 . N4 Session Mod ification Request/Response
ServingMEH
DNS record check
MEC service access ( service p rovision in g)
5 G m ob i le n e tw or k M EC n etw or k
Orchestra tor
The servin g HEH is d eterm ined for each service in sep arated p rocedu re from session estab lishm en t b y th e UE in ad vance. Then DNS record is
u p d ated on the d efau lt MEH.
NEF
Rou tin g p olicy
DNS record u p d ate
I P add ress is resolved by the requested serv ice nam e basica lly .
Figure 2.3.1-3 Signalling for MEH resource compensation
Simulations were performed for evaluation of these two compensation methods. Figure 2.3.1-4
shows system topologies for simulations of the method-a. In Topology-1, the service is
provided by the cloud server when the MEH1 does not have enough resources for a request. In
Topology-2, the service is provided by a MEH in upper level when a MEH does not have
enough resources for a service request. When the service is provided from the internet,
communication latency of +100ms is considered. Dual connectivity is also considered by
deploying small cells over macro cell. Data plane communications are provided always via
small cell in this simulation scenario.
Figure 2.3.1-4 Topologies for simulation (method-a)
Figure 2.3.1-5 shows the traffic model and resource allocation for each MEH and service for
simulation. It is assumed that relevant resources are occupied on the MEH until completion of
data delivery to the served UE. Occupied resources are equally 5 for each service this time.
The compensation method-b was also verified in Topology-1 and Topology-2.
UEMEH-1
( defau lt)MEH-2
( serving )
Status updateStatus update
Detection of resource shortage
Serving MEH change request
DL/UL data for the service
MECm anag er
Decision of the serv ing MEH
for task offload
DNS record change
UE Serv-1 Serv-2
Status updateResource shortage
Localm anag er
Resource shift
decision
Resource shift com m and
Resourceshift
MEH
a) Procedure for serving MEH change b) Procedure for resource shift inside the MEH
Detection of the resource shortage
for Serv-1
DL/UL data for the service
DNS resolver for the serviceDNS resolver for the service
Serving MEH change response
MEH1
cloud
Latency+100m s
MEH1
MEH2
MEH3
cloud
●Topology-1 ●Topology-2 ●Cell deploym ent
Latency+100m s
10Gbps
10Gbps
: Macro cell(2GHz)
: Sm all cell(30GHz)
Radius of macro cell= 200m
Figure 2.3.1-5 Traffic model and resource allocation on MEHs for simulation
Results are shown in Figure 2.3.1-6.Figure 2.3.1-6. We verified that communication latency
distributions in CDF, actually response time completing reception of requested contents at the
UE, were improved by compensation methods a and b as shown in the result for the all
services,“All services”. Without compensation, around 6% of task requests were responded in
more than 10 milliseconds, and the latency became more than 100ms in that case. With a
compensation method, almost all task requests were treated at one of MEHs within 10msec
latency. (Some were responded in more than 10 msec because of their bigger data size. They
were Serv_3.) The right hand side shows the response time for each service. Compensation by
resource shift from lower prioritized services inside a MEH, that is “1MEC, Shift_on” (red
solid lines), decreased the response time, on Serv_1 and Serv_2. However, lower prioritized
services, especially Serv_4, were degraded. It is because serv_1 and serv_2 were more
prioritized. In the topology-2, it was compensated very well by MEHs in multiple levels.
There were sufficient resources (3 x 25) for each service. Of course, it is possible to use
combination of these 2 methods for the compensation.
Figure 2.3.1-6 Simulation results on compensations for MEH resource shortage
Path switching to the optimal MEH based on user movement
In the mobile network, the inter-cell handover of wireless connectivity is performed with
respect to the UE mobility. When the edge computing is applied to the mobile network, it is
required to consider switching the serving MEH to a closer one to satisfy the service
ServicesData size Priority
Traffic generation
QoS characteristics (3GPP)
Resourcetype
Prioritylevel
Delaybudget
Serv1 V2X messages,Real time gaming
1 kbits
High Very frequent GBR3
(high)50ms
Serv2 Augmented Reality,Low latency
10 kbits
FrequentNon-GBR
6(middle)
10ms
Serv3 TCP-based www, SNS, chat
100kbits
MiddleNon-GBR
9(low)
300ms
Serv4 TCP-based FTP, Buffered streaming
1 Mbits
Low Less frequentNon-GBR
8(low)
300ms
Serv1 Serv2 Serv3 Serv4 Total
MEH1 25 25 25 25 100
Serv1 Serv2 Serv3 Serv4 Total
MEH3 25 25 25 25 100
MEH2 25 25 25 25 100
MEH1 25 25 25 25 100
●Traffic m odel ●Resou rce a llocation for Topolog y-1
●Resou rce a llocation for Topolog y-2
90%
92%
94%
96%
98%
100%
0 10 20 30 40 50 60 70 80 90 100 110 120
CD
F
Latency (msec)
No compensation
1MEC, Shift_on
3MEC, Shift_off
All services
90%
92%
94%
96%
98%
100%
0
10
20
30
40
50
60
70
80
90
10
0
11
0
12
0
CD
F
Latency (msec)
Serv_1
90%
92%
94%
96%
98%
100%
01
02
03
04
05
06
07
08
09
01
00
11
01
20
CD
F
Latency (msec)
Serv_2
90%
92%
94%
96%
98%
100%
01
02
03
04
05
06
07
08
09
01
00
11
01
20
CD
F
Latency (msec)
Serv_3
90%
92%
94%
96%
98%
100%
01
02
03
04
05
06
07
08
09
01
00
11
01
20
13
01
40
15
0
CD
F
Latency (msec)
Serv_4
Improved
Degraded
requirements such as communication latency. There are many literatures on it and [MB17]
provides the summary of them. They generally provide the optimal threshold policies to
determine whether to initiate the service migration to another MEH or not.
Figure 2.3.1-7 depicts images of service continuation from MEH when the inter-UPF handover
is performed. In the figure-A, the original MEH continues to provide the service also even after
the cell handover. But the data transfer latency becomes bigger because the distance from the
UE to the MEH becomes farther. In the figure-B, the data transfer latency is smaller by the
service provisioning from the closer new MEH. However, it is necessary to have the service
migration between involved MEHs to provide service continuity and cooperative path
switching between the mobile and MEC network. The above mentioned existing literatures
provide algorithms to minimize aggregated overhead for service migrations in the system by
minimizing service migrations.
Figure 2.3.1-7 Images of service continuation for the inter-UPF handover
In the second year, we have devised a procedure for path switching to the optimal MEH
according to the inter-UPF handover of the UE. Figure 2.3.1-8 is its simplified procedure. As
seen in the step 7, the new base station which the UE connects with after the inter-cell handover
notifies the MEH change with respect to the service migration between MEHs. In order to
realize this procedure, the 5G core requires to know the existence of the service migration
between MEHs. That information should be notified before or in the procedure.
Analysis and simulations are under going to evaluate efficiencies comparing with the existing
researches with respect to this topic.
UPF1 UPF2MEH1 MEH2
RAN1 RAN2
UE UE
UE m ovem ent
UPF1 UPF2MEH1 MEH2
RAN1 RAN2
UE UE
UE m ovem ent
A) The orig inal MEH continues to servefor the UE
B) Mig rate the rest o f the task to the op tim al new MEH
Service m ig ration
Figure 2.3.1-8 Procedure for MEH change notification while the inter-UPF handover
Toward the final report
Task 3.1 will deliver the final report through the deliverable D3.2 in Feb. 2019 (M32). It will
verify the goodness of the proposed innovations and will include the following topics:
Combined control signalling,
Performance evaluations considering selected project use cases
Signaling studies for resource optimizations derived from Task 3.2 and Task 3.3.
Relationship to other tasks
This task is based on the use cases defined by WP1.
This task provided information to T1.2 for the definition of the whole 5G-MiEdge
architecture.
The results will be input to Task 4.2 for PoC definition.
The results will be input to Task 5.2 for standardization activities.
2.3.2 Task T3.2: Context information management for traffic map prediction
Contributors: CEA, URom, KLAB, Intel
Task period: M04 – M32
Task status: running
UE RAN-2RAN-1 MEH-2MEH-1 UPF-2UPF-1
1 ) I n ter-cell h andover
2 ) Path sw itch ing request
3 ) New UPFselection
6 ) Path sw itch ing response(w ith the MEH chang e notification )7 ) MEH chang e notification
8 ) MEH chang e
UL data
DL data
UL data
DL data
5 ) Path m od ification request/response
4 ) Path estab lishm en t request/response
5G Core
This task focuses on defining new procedures to transport and share measured context (e.g.,
location, traffic, action, etc.) of users with the edge, on understanding the impact and roles of
terminals, (e.g. identify and define measurements on the terminal side to fulfil liquid resource
management, or new procedures and functional splits between terminals and the edge cloud),
and finally on the management of liquid resource, which requires adapted signaling for joint
communication and computing cluster formation. Cluster formation can either require a
centralized intelligence for orchestrating its formation and update or can be accomplished by
distributed intelligence.
Output
Analysis of computation caching policies for the exploitation of context information in
uplink traffic reduction schemes. These results appear in all detail in: N. di Pietro and
E. Calvanese Strinati, “Proactive computation caching policies for 5G-and-beyond
mobile edge cloud networks,” to be presented at EUSPICO 2018, Rome, 3-7 Sept.
2018.
Patent under preparation on the topic of computation caching policies.
Analysis of the trade-off between transport and caching energy costs associated to
delivery of contents in information networks, exploiting context information. These
results appear in all detail in: S. Sardellitti, F. Costanzo, M. Merluzzi, “Joint
optimization of caching and transport in proactive edge cloud”, to be presented at
EUSIPCO 2018, Rome, 3-7 Sept. 2018.
Deliverable
Task 3.2 contributed to deliverable D3.1.
Del.no. Deliverable name Task no. Due
3.1 Architecture of mmWave edge cloud and
requirement for control signaling
T3.1,
T3.2,
T3.3
M18
3.3 Context information management to create
traffic map for mmWave edge cloud
T3.2 M32
Highlights
In the following subsection, we propose an analysis on computation caching policies for the
reduction of uplink traffic and related costs. This analysis integrates the results of the first year
of the project, when computation caching was already introduced. In summary, the
computation caching policies that we propose are based on several kinds of context information:
a catalogue of offloadable computation tasks, their popularity, their computational requirements,
the size of their inputs, and the size of their results. Through learning techniques, this context
information can be learnt and updated by Serving Small Cells (SSCs) or Mobile Edge Hosts
(MEHs) and then exploited to optimize the information exchange management and the
minimizing the communication costs. Context information can be collected by SSCs during a
“training phase” that precedes the regular networking phase. During this phase, SSCs increase
and refine their knowledge of the traffic parameters involved in computation offloading and fill
their cache memories in preparation of the regular networking phase. Based on the learnt
context information and upon application of the described computation caching algorithm, the
network can predict and estimate the reduction of uplink traffic volume and associated costs
(both communication and computational). This allows the SSC to reduce the amount of
resources involved in the serving process, still guaranteeing the same Quality of Service (QoS).
Finally, notice that in many scenarios, the context information used for our computation caching
algorithm is UE-independent and can be measured and registered anonymously, without
violating the privacy of users.
Management and exploitation of context information: computation caching policies for
uplink traffic reduction
In MEC networks, Serving Small Cells (SSCs) endowed with radio access technology,
computing units, and local cache memories can be charged by User Equipments (UEs) to run
computing tasks on their behalf. The procedure of entrusting these computational assignments
to small cells is called task or computation offloading. It allows UEs to save both time and
energy and revolutionizes the classical interaction between UEs and mobile terminals.
The role of content caching in MEC networks is critically important and deeply investigated
[AFK17], [BBD14], [IW16], [WZZ+17]. In the context of task offloading, a new form of
caching was recently introduced, after noticing the pointlessness of repeating many times the
same computation for the same reiterated offloading request. This paradigm is called
computation caching [OC15], [Oue16], [OC17] and suggests to exploit small cells’ memory to
store the results of offloadable computations [EBS17], so that they can be simply retrieved from
the cache instead of being recomputed each time they are requested. The goal is to decimate
redundant and repetitive processing and has several
advantages, e.g., drastically reducing computation time and saving energy for both UEs and
SSCs, preventing uplink bottlenecks, freeing network resources and decreasing the SSCs’
workload, diminishing the number of virtual machine instantiations. Although computation
caching can be applied to several aspects of 5G-and-beyond networking, this part of 5G-
MiEdge’s work focuses on the interaction between a single UE and its SSC. The goal of our
analysis is to introduce, evaluate, and compare three different computation caching policies that
depend on task popularity. These policies are enablers for proactive computation caching
[EBS17], intended as the strategy of dynamically adapting the content of cache memories based
on the continuous learning of task popularity and other statistics. The leading concept is that
future offloading traffic can be predicted and computation caching can be proactively adjusted
to smoothly react to traffic fluctuations. Here we will show how some uplink traffic parameters
can be exploited to design efficient and effective computation caching policies: in particular,
we will focus on three quantities that characterize an offloadable task: its popularity, the size of
its input, and the size of its result. In particular, this last parameter plays a crucial role: in
computation caching, the size of the data to cache and to download (the task result) can be
significantly different from the size of the data to upload from the UE to the SSC (the task
input). This adds a novelty to the work of the first year of the project and marks a sharp
difference with classical content caching, in which cached data essentially have the same size
of the corresponding data travelling through the network. Our main contribution is to show that
this difference can be exploited to significantly improve the effectiveness of computation
caching.
In our setting, a UE offloads computational tasks to the MEC via its SSC. The communication
rates are denoted 𝑅𝑈𝐿 in uplink and 𝑅𝐷𝐿 in downlink and are measured in bits per second. We
suppose that the computational capacity of the SSC is 𝑓 CPU cycles per second and that the
SSC can store up to 𝑚 bits to perform computation caching on a local memory.
Offloadable tasks belong to a finite set 𝒞 = {𝑐1, … , 𝑐𝐾}, that we call the computation catalogue.
A task 𝑐𝑘 ∈ 𝒞 represented by a triplet: 𝑐𝑘 = (𝑊𝑘, 𝑒𝑘 , 𝑊𝑘′), where 𝑊𝑘 the input data (a sequence
of bits) to be processed, 𝑒𝑘 is the number of CPU cycles per bit needed to elaborate the data,
and 𝑊𝑘 is the computation result (another sequence of bits). We denote |𝑊𝑘| and |𝑊𝑘′| the sizes
in bits of 𝑊𝑘 and 𝑊𝑘′.
In order to represent the content of the SSC’s cache, we define the cache indicator as the vector
𝜎 = (𝜎1, … , 𝜎𝐾) ∈ {0,1}𝐾 such that 𝜎𝑘 = 1 if and only if the result 𝑊𝑘′ of 𝑐𝑘 ∈ 𝒞 is stored in
the SSC’s cache. Thus, a cache indicator fully identifies the cache content. Since the cache size
is limited to 𝑚 bits, in general not all vectors in {0,1}𝐾 correspond to a feasible cache
configuration. Therefore, we define the set of feasible cache indicators as follows:
ℱ = {𝜎 ∈ {0,1}𝐾 ∶ ∑ 𝜎𝑘
𝐾
𝑘=1
|𝑊𝑘′| ≤ 𝑚}.
Task offloading starts with a request from the UE to the SSC specifying the task to run and a
time delay within which the UE needs to retrieve its result. Such an offloading request is
denoted 𝑟 = (𝑘, 𝑡), meaning that the UE asks for the execution of the 𝑘-th task and to receive
its result within 𝑡 seconds. If the SSC has enough available resources to elaborate the task, the
request is accepted, otherwise it is denied.
Our goal is to describe strategies to reduce the costs of tasks offloading. The total cost of the
offloading procedure is made of several independent contributions, among which we identify
two main components: i) the cost of uploading the computation inputs from the UE to the SSC;
ii) the cost of running the computation at the SSC. Depending on the application, the word
“cost” can indicate energy consumptions, time delays, or any other metric that measures an
expense or the quality of service. Nonetheless, in all scenarios, there are evident benefits in
keeping available in the cache memory a computation result before it is requested to the SSC:
indeed, whenever a result 𝑊𝑘 is stored, the task 𝑐𝑘 does not need to be run, its input data do not
need to be uploaded, and 𝑊𝑘′ can be straightforwardly sent to the UE. The most important
consequence is that the two cost components mentioned above are zeroed. Hence, the total cost
of offloading a cacheable task 𝑐𝑘 ∈ 𝒞 is:
Γ𝑡𝑜𝑡(𝑐𝑘) = Γ𝑟𝑒𝑞(𝑐𝑘) + (1 − 𝜎𝑘) (Γ𝑈𝐿(𝑐𝑘) + Γ𝑐𝑜𝑚𝑝(𝑐𝑘)) + Γ𝐷𝐿(𝑐𝑘) + 𝛾(𝑐𝑘),
where Γ𝑟𝑒𝑞(𝑐𝑘) is the cost of sending 𝑟 = (𝑘, 𝑡), the offloading request; 𝜎𝑘 is the 𝑘-th entry of
the cache indicator; Γ𝑈𝐿(𝑐𝑘) is the cost of uploading the input data; Γ𝑐𝑜𝑚𝑝(𝑐𝑘) is the cost of
computing the task result (assuming, for simplicity, that the CPU state does not vary in time
and the computation cost only depends on the task parameters); Γ𝐷𝐿(𝑐𝑘) is the cost of sending
the result back to the UE; and 𝛾(𝑐𝑘) includes any other fixed cost that does not directly depend
on c, e.g., any fixed processing cost at the MEC level or the cost of maintaining active the SSC’s
hardware, including the cache memory. The cost of reading a task result from the cache is
considered negligible.
The previous considerations lead to an important question: given a cache of finite size, how to
choose which 𝑊𝑘′’s to store, with the goal of minimizing the overall costs? To answer, let us
consider 𝑅 offloading requests 𝑟1, … , 𝑟𝑅, sent from the UE to the SSC during its service period.
By definition, every request uniquely corresponds to a task: for every 𝑖 = 1, … , 𝑅, we have 𝑟𝑖 =(𝑘, 𝑡𝑖), for some 𝑘 ∈ {1, … , 𝐾} identifying a task in the catalogue and some latency constraint
𝑡𝑖. Thus, Γ𝑡𝑜𝑡(𝑟𝑖) = Γ𝑡𝑜𝑡(𝑐𝑘) for some 𝑘 and we define the cost over the whole serving period
as:
Γ(𝜎) = ∑ Γ𝑡𝑜𝑡(𝑟𝑖)
𝑅
𝑖=1
.
Our goal is to find the cache indicator that minimizes Γ(𝜎):
𝜎𝑜𝑝𝑡 = arg min𝜎∈ℱ
Γ(𝜎) = arg min𝜎∈ℱ
(∑ Γ𝑡𝑜𝑡(𝑟𝑖)
𝑅
𝑖=1
).
Ideally, if 𝜎𝑜𝑝𝑡 is known, the SSC guarantees an optimal cost minimization by storing the 𝑊𝑘′’s
for which (𝜎𝑜𝑝𝑡)𝑘
= 1.
Since the number of cache indicators grows exponentially with 𝐾 , it is not always
algorithmically possible to run through all of them to exhaustively determine 𝜎𝑜𝑝𝑡. The scope
of our analysis is to propose and evaluate strategies to choose cache indicators with close-to-
optimal associated performance. A very natural choice is to assign a hierarchy among tasks and
to fill the cache with the results of the highest-priority ones. A caching metric 𝜆 ∶ 𝒞 → 𝐑+
assigns to each task a “cacheability value”. The caching policy based on 𝜆 is the application of
the following cache filling algorithm that prioritizes the tasks with the highest cacheability
value:
1: let 𝜋 ∶ {1, … , 𝐾} → {1, … , 𝐾} be a permutation such that 𝜆(𝑐𝜋(1)) ≥ ⋯ ≥ 𝜆(𝑐𝜋(𝐾)).
2: set 𝜎 = (0, … , 0) and 𝑠 = 0.
3: for 𝑘 = 1, … , 𝐾, do
4: if 𝑠 + |𝑊𝜋(𝑘)′ | ≤ 𝑚, then
5: set 𝜎𝜋(𝑘) = 1 and 𝑠 = 𝑠 + |𝑊𝜋(𝑘)′ | .
6: end if
7: end for
8: fill the SSC’s cache according to 𝜎.
We call 𝜎(𝜆) the indicator yielded by the previous algorithm. Clearly, a caching policy is based
on a well-designed metric if Γ(𝜎(𝜆)) is close to Γ(𝜎𝑜𝑝𝑡). A first observation, very spontaneous
and common to the context of content caching [IW16], is that a good caching policy needs to
depend on the popularity of tasks. Indeed, to reduce costs, we want to avoid to repeatedly
process frequently requested tasks. In this perspective, we define the popularity 𝑝𝑘 of a task 𝑐𝑘
to be the probability that 𝑐𝑘 is offloaded to the SSC. In our setting (and, in general, whenever
the offloading requests are pairwise independent and if their total number 𝑅 is big enough to be
statistically representative), we can write:
𝑝𝑘 = |{𝑖 ∶ 𝑟𝑖 = (𝑘, 𝑡𝑖), ∃ 𝑡𝑖}| ⋅ 𝑅−1. In general, the task popularity is a typical example of context information that the SSC can learn
and update during its serving period.
First policy: simply based on task popularity, we define:
𝜆1(𝑐𝑘) = 𝑝𝑘, ∀ 𝑘 ∈ {1, … , 𝐾}. 𝜆1 is essentially the metric used in [EBS17]. A better choice comes from the observation that
caching the result of a very popular task with low input uploading and computation costs, can
be less advantageous than caching the result of a less popular task with higher costs. The latter
directly depend on the size (in bits) of 𝑊𝑘, denoted |𝑊𝑘|, which justifies the next policy.
Second policy: based on popularity and input data size, let
𝜆2(𝑐𝑘) = 𝑝𝑘|𝑊𝑘|, ∀ 𝑘 ∈ {1, … , 𝐾}. In this context, 𝜆2 corresponds to the metric used in [OC17], where it was already highlighted
the need for a policy that mixes popularity and input data size.
Third policy: the third policy is based on the observation that caching task results whose size
is small allows to store more of them. Hence, it may be more convenient to cache a high number
of small-size results, even if their popularity and input size do not maximize 𝜆2. To increase the
caching priority of tasks with small |𝑊𝑘′|, we define:
𝜆3(𝑐𝑘) = 𝑝𝑘|𝑊𝑘|
|𝑊𝑘′|
, ∀ 𝑘 ∈ {1, … , 𝐾}.
The introduction of 𝜆3 is one of the novelties of our work. It turns out to be the most
advantageous metric of the three, from several points of view, as we will show from until the
end of this subsection.
In our numerical simulations, |𝑊𝑘| and |𝑊𝑘′| are chosen independently at random for every 𝑘 as
follows: let 𝑦, 𝑌 ∈ 𝐍 satisfy 𝑦 ≤ 𝑌 and let 𝑥, 𝑋 ∈ 𝐑 be two real numbers in [1, 10] (with 𝑥 ≤ 𝑋
if 𝑦 = 𝑌). When we say that |𝑊𝑘| belongs to [𝑥𝑒𝑦 ∶ 𝑋𝑒𝑌[, we mean that 𝑥 ⋅ 10𝑦 ≤ |𝑊𝑘| < 𝑋 ⋅10𝑌 and |𝑊𝑘| = 𝑢 ⋅ 10𝑣, with 𝑢 and 𝑣 randomly fixed as follows: first, 𝑣 is chosen uniformly
in {𝑦, 𝑦 + 1, … , 𝑌}; then, 𝑢 is chosen uniformly either in [𝑥, 10[ (if 𝑣 = 𝑦) or in [1, 10[ (if 𝑦 <𝑣 < 𝑌 ) or in [1, 𝑋[ (if 𝑣 = 𝑌 ). The same rule is used for |𝑊𝑘
′| , independently from the
corresponding |𝑊𝑘|. With this strategy, there is no privileged order of magnitude among the
values taken by |𝑊𝑘| and |𝑊𝑘′|, even when the maximum possible value is much bigger than
the minimum.
In all figures below, the abscissae represent the SSC’s cache size. 0% means that 𝑚 = 0 bits
and 100 % that 𝑚 = ∑ |𝑊𝑘′|𝐾
𝑘=1 bits. Figure 2.3.2-1, Figure 2.3.2-2, Figure 2.3.2-3 and Figure
2.3.2-4 were obtained with the simulation parameters specified in Table 2.3-1.
Table 2.3-1 Parameters for numerical simulations.
Parameter Value Parameter Value
𝑲 in Figure 2.3.2-1 25 𝐾 in Figure 2.3.2-2,
Figure 2.3.2-3,
Figure 2.3.2-4
50000
𝑹𝑼𝑳 125 Mb/s |𝑊𝑘| in Figure 2.3.2-1 [1𝑒6 ∶ 1𝑒9[ bits
𝑹𝑫𝑳 500 Mb/s |𝑊𝑘′| in Figure 2.3.2-1 [1𝑒6 ∶ 1𝑒7[ bits
𝜶 0.6 |𝑊𝑘| in Figure
2.3.2-2, Figure
2.3.2-3,
Figure 2.3.2-4
[1𝑒6 ∶ 1𝑒9[ bits
𝒆𝒌/𝒇 10−8 s/bit |𝑊𝑘′| in Figure
2.3.2-2, Figure
2.3.2-3,
Figure 2.3.2-4
[1𝑒3 ∶ 1𝑒9[ bits
In particular, we considered stable radio channel conditions and constant uplink and downlink
communication rates. First, the cache was filled applying one of the three policies defined
above, then a high number of offloading requests were simulated. We supposed the popularity
of offloading requests to obey the Zipf law [BBD14]: 𝑝𝑘 = (𝐴𝑘𝛼)−1, for constant 𝛼 and 𝐴 = ∑ 𝑘−𝛼𝐾
𝑘=1 . Notice that, without loss of generality, tasks can be assumed to be sorted in the
catalogue by descending popularity. The simulated offloading operation consisted of four main
serial steps: offloading request, input data uploading, task computation, results downloading. If
the results of the computation were found in the SSC’s cache, data uploading and task
computation were skipped and the results directly sent to the UE. In all simulations, we assumed
that a new offloading request was sent instantaneously after the results of the previous one were
downloaded.
Figure 2.3.2-1 shows, as a function of the cache size, the percentage of task input data that did
not need to be uploaded nor elaborated because the corresponding results were cached and
available for downloading. For brevity, we call this the “spared input data”. Measuring the
spared input data is an effective approach to evaluate the goodness of the caching policies: the
more it is, the higher the corresponding saving in energy, time, or any other metric, both for the
UE and for the SSC. Differently from the other figures, in Figure 2.3.2-1 we could compare the
performance of the policies also with respect to the optimal cache configuration found by
exhaustively looking for 𝜎𝑜𝑝𝑡 among all feasible cache indicators. This was practicable because
we kept the number of tasks in the catalogue small enough (𝐾 = 25). Figure 2.3.2-1 suggests
two main considerations: first, that the second and third policy clearly outperform the policy
exclusively based on popularity; second, that the performance of the third policy is always very
close or superimposed to the optimal and beats the policy based on 𝜆2, especially for cache
sizes between 0 and 20 %. These are the sizes which interest us the most, because an effective
caching strategy needs to achieve a good performance with a cache as small as possible. The
curves in Figure 2.3.2-1 are not extremely smooth. The “jumps” reflect the fact that a small
increase in the cache size may suddenly allow to fit in the cache task results with good metric,
but “relatively too big” to be cached before. This behavior is more visible when 𝐾 is small.
Figure 2.3.2-1 Spared input data for 𝑲 = 𝟐𝟓.
Figure 2.3.2-2 shows the spared input data for 𝐾 = 50000 . In this case, the optimal
performance could not be traced, but the quality of the third policy is clearly confirmed by the
separation among the curves. Remarkably, for a cache only as big as 2 % of the total size, the
third policy allows to spare more than 80 % of the input data, whereas the first and second
policy respectively achieve around 20 and 30 %.
Figure 2.3.2-2 Spared input data for 𝑲 = 𝟓𝟎𝟎𝟎𝟎.
Figure 2.3.2-3 Cache hit ratio for 𝑲 = 𝟓𝟎𝟎𝟎𝟎.
Figure 2.3.2-3 compares the policies in terms of cache hit ratio: by definition, this is the number
of times that the results of the offloading requests were found in the cache, divided by the total
number of requests. The figure suggests two observations: first, that weighing the task
popularity by the cache input size to define 𝜆2 causes a loss in the cache hit ratio performance;
this is quite natural, because 𝜆1 is by design a metric aimed at maximizing the cache hit ratio.
Nonetheless and more importantly, this loss is completely recovered and even outdone by the
third policy, which promotes for being cached the tasks with small-size results. We deduce that
storing in this way more results, even if they do not correspond to the most popular tasks in
absolute, yields a considerable gain: the other two policies are beaten and cache hit ratios of
almost 60 % are obtained with the third policy when the cache size covers only 1 % of the total
cacheable data (a performance from 4 to 7 times better with respect to the other two policies).
Figure 2.3.2-4 Gain in served offloading requests per hour for 𝑲 = 𝟓𝟎𝟎𝟎𝟎.
Figure 2.3.2-4 shows the ratio between the average number of offloaded tasks per hour with
and without computation caching. Measuring this gain involves the computation of the
offloading time for every request. In the notation of the beginning of this subsection, where in
this case Γ𝑡𝑜𝑡(𝑐𝑘) indicates the total offloading time of 𝑐𝑘 = (𝑊𝑘, 𝑒𝑘, 𝑊𝑘′), we have Γ𝑟𝑒𝑞(𝑐𝑘) =
128/𝑅𝑈𝐿 (where we supposed that a request 𝑟𝑖 = (𝑘, 𝑡𝑖) has a standard size of 16 bytes,
Γ𝑈𝐿(𝑐𝑘) = |𝑊𝑘|/𝑅𝑈𝐿 , Γ𝑐𝑜𝑚𝑝(𝑐𝑘) = 𝑒𝑘|𝑊𝑘|/𝑓, and Γ𝐷𝐿(𝑐𝑘) = |𝑊𝑘′|/𝑅𝐷𝐿 . We also added to
the previous terms a latency of 2 ms, corresponding to 𝛾(𝑐𝑘). Figure 2.3.2-4 reasserts the
superiority of the third policy, which allows gains of up to a factor 10 for cache sizes of less
than 20 %, whereas the gain with respect to the other policies does not go beyond a factor 4.
These gains translate into reduced uplink transmissions and facilitate the prevention of uplink
bottlenecks.
Joint proactive caching and transport optimization strategy to favor caching of the most
popular contents while decreasing the transport cost.
In this task we devised a strategy for finding the optimal trade-off between the transport and
caching energy costs associated to the distribution of contents in information networks. The
results of this task led to the publication [BSC18] and [SCM18]. The proposed strategy is
proactive with respect to the users’ requests, as contents are pre-fetched depending on the
distribution of their (estimated) popularity. In particular, we develop a dynamic energy-efficient
strategy that jointly optimizes caching and delivery costs within each cluster of nodes, i.e.
moderate size network in which each node (entities like Mobile Edge Host (MEH)) has storage
capabilities. The approach fits naturally in the network paradigm named Information Centric
Network (ICN), which integrates content delivery as a native network feature and then it is
perfectly suited for an efficient access and distribution of contents throughout the network
[JST09], [ADI12]. One of the main benefits of ICN is to reduce user content access delay and
network bandwidth usage by storing the contents at the network edge close to the end user. ICN
is based on named data objects, for example, web pages, videos, documents, or other pieces of
information. In contrast, current networks are host-centric since communication is based on
named hosts, for example, web servers, PCs, mobile handsets. We proposed to incorporate the
ICN strategy in the edge cloud to have a robust mechanism to handle mobility in the mmWave
scenario. In fact, the current network is based on retrieving contents based on their IP address.
This means that, if there is an abrupt blocking event in some mmWave link for a relatively long
time (blocking intervals of half a second are not uncommon), the session is doomed to drop.
Conversely, in the ICN architecture, since content objects are retrieved by their names, if a link
drops, the request will automatically traverse the network until it will reach another node that
can deliver the requested content with, possibly, no session disruption. The question becomes
then how to distribute contents through the network. We addressed this question by finding the
optimal tradeoff between content replication and delivery time.
In ICN networks, serving small base stations (MEHs) are equipped with storage capabilities to
cache contents as they are requested by the end users. Many nodes contribute actively in content
caching to reduce the network congestion, the access delay and origin servers’ load [ZLZ15],
[WAN14]. Several content caching strategies have been proposed to maximize local hit rate, or
the fraction of requests served by a given cache, optimizing the placement and routing of
information objects in a static way [HMR17], [CGK12], [KSK18]. Clearly, an effective caching
strategy builds significantly on the ability to learn and predict users’ behaviors. This capability
lies at the foundation of proactive caching [BBZ15] and it motivates the need to merge future
networks with big data analytics [ZBB16]. A distributed dynamic content replacement strategy
that refreshes the caches contents as they travel through the network have been proposed in
[LTV15], where the authors considered the problem of finding the time evolution of the optimal
placement and routing of contents which minimizes the sum of the transport and caching
energies.
In our work, we consider a joint proactive caching and routing strategy aimed at minimizing
the sum of the caching and transport energy consumption in edge cloud networks, assuming
that the content objects to be delivered are stored in the edge nodes. Starting from the strategy
proposed in [LTV15], we devise a proactive in-network caching method based on online
popularity learning of the object contents. In particular, we incorporate a cost of caching the
information objects that depends, dynamically on the local and global popularity of the objects,
in order to encourage the edge nodes to host the most popular contents. The content distribution
results as a tradeoff between replication and delivery time, as described below.
Let us consider a transport network represented by a graph 𝒢 = (𝒱, ℰ, 𝒦), composed of a set
of nodes 𝒱, a set of links ℰ, and a set of information objects 𝒦. We assume that contents can
be permanently or temporarily stored over the nodes of this graph or travel through its edges.
Specifically, in some nodes, called repository, the content objects are stored permanently, at
least as far as their popularity does not change, and in the other nodes the contents may appear
and disappear, according to users’ requests and network resource allocation. To simplify our
formulation, we assume that all contents can be split into objects of equal size, identified by an
index 𝑘 ∈ 𝒦. Each node and edge is characterized, respectively, by a storage and transport
capability. We assume that time is divided in slots of fixed duration ∆τ and we consider time
frames, each composed of T time slots. At time slot n, each node 𝑢 ∈ 𝒱 in the network can act
as a repository node of a set of information objects 𝐾𝑢[𝑛] ⊆ 𝒦 , and can request a set of
information objects 𝑄𝑢[𝑛] ⊆ 𝒦 . Let us denote with 𝒒[𝑛] ∈ {0,1}|𝒱||𝒦| the request arrival
process such that 𝑞𝑢[𝑘, 𝑛] = 1 if node u requests object k at time n, and 𝑞𝑢[𝑘, 𝑛] = 0 otherwise.
The random process 𝒒[𝑛] depends on the time evolution of the contents’ popularity that is
modelled as a Poisson process, with an average arrival rate 𝑃𝑢𝑘[𝑛] at node u for object k follows
the Zipf distribution
where 𝛼𝑢[𝑛], 𝛽𝑢[𝑛] and 𝑟𝑢𝑘[𝑛] are, respectively, the Zipf parameter, the request rate and the
rank of object k, at node u at time n. The content objects popularity evolves in time following
a rule based on a local and a global popularity measures, associated to each time frame s
according to a forgetting factor 𝜂 ∈ [0,1], taking into account all the previous probabilities
A similar formula applies for the global popularity 𝑃𝑘𝑔[𝑠].
Considering the graph of our network, we can define a vertex signal over the nodes 𝑠𝑢[𝑘, 𝑛]
and an edge signal 𝑡𝑢𝑣[𝑘, 𝑛] over the its edges, both with binary values, where 1 means the
content k is stored in node u, or delivered form node u to node v, at time n and 0 otherwise.
Typically, each content may be hosted on every node, removed from the actual position and
moved whenever convenient to another location, by meeting the dynamic process of users’
requests. Moreover, each content is stored at least in one repository node, which can be the one
at the edge of the network, i.e. may directly access a content delivery network, or a node chosen
in a proactive way. The choice of the node where pre-fetching a content object is made
following a probabilistic measure depending on the centrality of each node u, respect to the
other nodes requesting the object k
choosing the node where 𝑤𝑢[𝑘] takes its minimum value, taking into account the length of the
shortest path (in number of hops) 𝐵𝑢𝑣, between node u and node v, and the average popularity
of object k at node v. The goal of our work is to devise a proactive caching strategy that
minimizes the sum of caching and transportation costs by taking into account that the cost of
caching strictly depends on the time evolution of the popularity of the requested contents, so as
to make this strategy proactive and context-aware. We define the energy cost for storing a
content k on node u during T consecutive time slots, in the time window [𝑛′ − 𝑇 + 1, 𝑛′], where
n’ is the time frame, as
where 𝑐𝑢[𝑘, 𝑛] is the time-varying energy cost for keeping content k on node u at time n, defined
as
where 𝛾 ∈ [0,1] is the trade-off coefficient between the local and global popularity costs.
Note that the cost 𝑐𝑢[𝑘, 𝑛] is low for contents with the highest popularity to encourage the
storage of these frequently requested contents. Then, we can define the cost associated to the
content transport as
.
The proposed proactive caching optimization problem is then defined as
where 𝜆 is a positive parameter controlling the ratio between transport and storage energy costs.
The constraint set 𝜒 is defined in order to satisfy all constraints associated to moving the
contents throughout the network. More specifically, the set incorporated the following rules:
o if object k is requested by node u at time slot n, then k either is already present in the
cache of node u at time n or it has to be transported to node u from a neighbor node
𝑣 ∈ 𝑁𝑢 within a maximum delivery time slots;
o if k is being cached at node u at time n, then k either was in the cache of u at time n−1
or was received by node u from a neighbor node 𝑣 ∈ 𝑁𝑢 at time n−1;
o if object k is delivered to node u from a neighbor node 𝑣 ∈ 𝑁𝑢 at time n, then this
object either was in the cache of v at time n−1 or was transferred to u from a neighbor
node 𝑤 ∈ 𝑁𝑢 at time n−1;
o an initial condition constraints that assure a proactive selection of the repository nodes
that always store the objects in 𝒦𝑢𝑝, and at n = 0 nothing else;
o a constraint that assures that the total amount of contents stored and delivered meets the
storage and capacity constraints of the nodes and the edges of the network;
o a constraint that states the binary nature of the storage and transport variables.
The simulation results are reported in the next two figures, for a network of 11 nodes and
considering 10 content objects to storage or/and deliver in a 15 slots frame. The performance
of our strategy is compared with the non-proactive strategy proposed in [LTV15], where the
content popularity does not affect the optimization process:
Figure 2.3.2-5 shows the transport gain in terms of number of hops reduction, defined
as 𝐺ℎ = 𝑛ℎ𝑛 − 𝑛ℎ
𝑝 , where 𝑛𝑛 and 𝑛𝑝 are the average number of hops, respectively, in
the non-proactive and in the proactive algorithms. It can be observed as 𝐺ℎ increases
when the transport cost λ decreases, since, in this case, the objects transport is favored
by the proactive caching.
Figure 2.3.2-6 shows the energy cost comparison between the two strategies, respect to
the transport cost parameter. Note that proactivity yields considerable energy savings
for low λ values, taking benefit from the optimal transport strategy.
Figure 2.3.2-5 Number of hops gain vs. transport cost parameter
Figure 2.3.2-6 Total average energy cost vs. transport cost parameter
Relationship to other tasks
T3.2 takes input from T1.2 about the use cases of interest.
The learning mechanisms implemented in T3.2 will play a key role in devising
proactive resource allocation strategies.
Some of the algorithms developed in T3.2 will also be incorporated in the system
performance evaluation to be done in Task T4.1.
T3.2 contributes to dissemination.
2.3.3 Task T3.3: User/application centric orchestration to realize 5G liquid edge cloud
Contributors: URom, CEA, TTech, Intel
Task period: M04 – M32
Task status: running
The user/application centric paradigm requires a joint optimization of communication /
computation / storage resource allocation, taking explicitly into account application layer
parameters. This task aims at providing an energy efficient design satisfying application-
specific latency constraints. This requires in some cases a proactive allocation, based on the
learning techniques developed in Task 3.2.
Output
Developed two algorithms for joint association and joint allocation of radio and
computation resources for computation offloading.
Developed an algorithm for clustering, useful to partition the network and find the
optimal location of services.
Developed an algorithm for resilient design.
Developed an algorithm for detection of network criticalities, e.g. anomalous traffic
injection.
The work has produced the following publications:
o S. Sardellitti, M. Merluzzi, S. Barbarossa, “Optimal association of mobile
users to multi-access edge computing resources”, Proc. of ICC 2018, Kansas
City, USA.
o E. Ceci, S. Barbarossa, “Small perturbation analysis of network topologies”,
Proc. of ICASSP 2018, Calgary, Canada.
o S. Barbarossa, S. Sardellitti, E. Ceci, “Learning from signals defined over
simplicial complexes”, Proc. of DSW 2018, Lausanne, Switzerland.
o S. Barbarossa, S. Sardellitti, E. Ceci, M. Merluzzi, “The edge cloud: a holistic
view of communication, computation and caching”, chapter 16 of
"Cooperative and Graph Signal Processing: Principles and Applications", P.
Djuric and C. Richard Eds., Academic Press, Elsevier, 2018.
Developed a joint dynamic ON/OFF and access resource optimization at edge clouds
useful for reduction of CAPEX/OPEX.
The work has produced the following publication:
o Gia Khanh Tran, Hidekazu Shimodaira, Kei Sakaguchi, "User Satisfaction
Constraint Adaptive Sleeping in 5G mmWave Heterogeneous Cellular
Network," IEICE Trans. Commun., IEICE, Vol. E100-B, No. 4, Oct. 2018.
Deliverable
Task 3.3 contributed to deliverable D3.1.
Del.no. Deliverable name Task no. Due
3.1 Architecture of mmWave edge cloud and
requirement for control signaling
T3.1,
T3.2,
T3.3
M18
3.4 User/application centric orchestration of
mmWave edge cloud
T3.3 M32
Highlights
In the second year of the project, we developed new algorithms for user/application centric
orchestration. Some results were included in deliverable D3.1, while others will be included in
upcoming deliverables of WP3. In particular, Task 3.3 is split into three subtasks, which will
be described in different subsections in the following. In subtask 3.3.1 we developed two
algorithms for user association to MiEdge resources (AP and MEH), in subtask 3.3.2 we
developed a clustering algorithm for optimal network partition, content and service placement.
A joint dynamic ON/OFF and access resource optimization at edge clouds was developed also.
In subtask 3.3.3, two algorithms for robust design of the network and detection of network
criticalities were developed. In the next three subsections we will briefly present all the
algorithms and the main results.
Subtask 3.3.1 – Jointly optimal allocation of radio/computation/storage resources
In this subtask, we considered the problem of user association for the offloading of
computationally heavy applications from mobile devices to a MEH through a mmWave AP.
The results of this subtask led to the publication [SMB18]. The problem of association of mobile
users to MiEdge resources is fundamental, since computation resources play a key role in the
assignment. A typical assignment problem is shown in Figure 2.3.3-1.
Figure 2.3.3-1 Example of association of mobile users to MiEdge Resources
We can notice that, due to the limited capacity of the backhaul and the busy status of MEH 1,
UE 4 could get better QoE if it accesses AP2 and MEH 2, since the longer delay of the
communication part is compensated by the much shorter computation time. Indeed, the latency
experienced by user 𝑘 accessing AP 𝑛 and MEH 𝑚 is given by
Δ𝑘 =𝑛𝑏𝑘
𝐵 log2(1 + ℎ𝑘𝑛𝑝𝑘𝑛) +
wk
𝑓𝑚𝑘+ 𝑇𝐵𝑛𝑚
where 𝑛𝑏𝑘 is the number of bits necessary to transfer the application, 𝐵 is the bandwidth, ℎ𝑘𝑛
is the channel gain between user 𝑘 and AP 𝑛, wk is the number of CPU cycles necessary to run
the application, 𝑓𝑚𝑘 are the computation resources allocated to user 𝑘 by MEH 𝑚, and 𝑇𝐵𝑛𝑚is
the backhaul delay between AP 𝑛 and MEH 𝑚 . In the equation, the first term is the
communication delay, the second term is the computation delay, and the last term is the time
needed to transfer the application over the backhaul. The coupling between computation and
communication time given by the above formula motivates the joint association and the joint
allocation of radio and computation resources. For each user 𝑘, we introduce the assignment
binary variable 𝑎𝑘𝑛𝑚, which is equal to 1 if user 𝑘 accesses AP n and MEH 𝑚 and 0 otherwise.
Formulating the problem of the offloading as the minimization of the total power consumption,
we can write it as follows:
where 𝐩, 𝐟 and 𝐚 are the vectors of allocated powers, computation resources and binary
assignment variables, respectively. Moreover, we defined the function
to write the latency constraint. The above constraints have the following meaning: 𝑖) the overall
latency of each user 𝑘 must be lower than the maximum value 𝐿𝑘; 𝑖𝑖) the total power spent by
each user must be lower than a fixed total power budget 𝑃𝑘; 𝑖𝑖𝑖) the sum of the computational
rates 𝑓𝑚𝑘 assigned by each server cannot exceed the server available computational capability
𝐹𝑚; 𝑖𝑣) each mobile user should be served by one AP-MEC server pair, so that we enforce the
constraint ∑ ∑ 𝑎𝑘𝑛𝑚𝑁𝑐𝑚=1
𝑁𝑏𝑛=1 = 1 , for each 𝑘 . This problem is a mixed integer nonlinear
programming (MINLP) problem and, in general, NP-hard. To solve it with reduced complexity
we proposed two alternative strategies
1) A Successive Convex Approximation strategy with a penalty function forcing the binary
solution;
2) A low complexity approach based on matching theory with transfers.
The first approach is based on our previous work [SBS14], where the recent tools of Successive
Convex Approximation (SCA) [SCU17] are used to deal with the non-convexity of the problem,
and a penalty function is used to force the binary solution in which each user is associated to
only one pair AP-MEH. This first approach efficiently solves the problem, but the complexity
can be further reduced by exploiting the power of matching theory, recently became a powerful
solution of many problems in wireless networks. For this second algorithm, the assignment
problem is formulated as a matching game in which users and AP-MEC pairs rank one another
using suitable preference functions associated to the transmit power used by each user to
implement computation offloading under latency constraints. Matching theory is a powerful
and simple tool to associate agents of two different sets using suitable preference lists. In this
particular case, we solved the problem of matching in two interdependent subgames: a Deferred
Acceptance algorithm [GAL62], and a coalitional subgame, in which some users can request to
be transferred to improve their utilities. Preference functions are built according to the objective
function of the problem, so that users prefer the pair AP-MEH that minimizes its power
consumption. To test the effectiveness of the proposed algorithms, we show in Figure 2.3.3-2
the performance in an exemplary scenario with 4 users, 2 AP’s and 2 MEH’s. The figure shows
the sum power of all users vs. the maximum latency, for different association strategies. The
first two strategies assign each user to the AP with the maximum SNR. Then, in the first case
resources are optimized disjointly, while in the second case a joint optimization is performed.
The other association strategies are based on our algorithms and on the exhaustive search, which
explore all the possible solutions and finds the global optimal one. It can be noted that both
proposed approaches yield considerable power savings with respect to SNR-based methods,
taking advantage of the optimal assignment of each user to a cloud through the most convenient
base station. It has to be remarked that the complexity of the matching-based algorithm has a
polynomial growth with the number of players (users and AP-MEC pairs), although the reached
final solution could be suboptimal, as the preference lists are built based on an approximate a
priori knowledge.
Figure 2.3.3-2 Total Transmit power vs. Latency constraint
To further test the effectiveness of the matching algorithm, in Figure 2.3.3-3 we show the ratio
𝜌 between the overall power consumptions achieved with two different association rules and
the global optimal solution, averaged over the channel realizations. It is interesting to note from
Figure 2.3.3-3 that 𝜌 keeps quite close to 1 for the proposed matching algorithm.
Figure 2.3.3-3 Average ratio 𝝆 vs. Latency constraint
Subtask 3.3.2 – Load distribution and clustering via distributed pricing mechanism
In this subtask, during the second year, we developed an algorithm for clustering nodes (i.e.
MEH). Clustering is essential to reduce complexity, latency, and overhead of control signaling
in the network. We analyze the problem of finding the optimal location of services in a network,
in the sense of finding the best place to storage contents, or to compute tasks, or to carry out
any other facilities needed by the nodes in the network. The goal of this work is to apply
clustering methods to partition the network in clusters in such a way that the cluster-heads (or
cluster-centers) can deliver the requested objects to the nodes inside its reference cluster, within
specified delays. We present the problem applied to the caching policy presented in section
2.3.4.3. However, this clustering technique is general and can be applied to other problems with
few modifications. In the caching problem view, presented in [SCM18], we set a strategy to
find the optimal clustering and then locate the best places to store (replicate) the most popular
contents. Given a network with N nodes, whose topology is represented by a graph 𝒢 = (𝒱, ℰ),
we denote by B the 𝑁 × 𝑁 matrix whose entry 𝐵𝑖𝑗 contains the length of the shortest path
between nodes i and j. If 𝑃𝑗(ℓ) is the probability that content ℓ will be requested in the area
served by access node j, the average (expected) energy cost needed to deliver content ℓ from
node i to the rest of the network is
where T is the unit cost to transport an object over one link and i is the selected node to replicate
the content, chosen as the one to minimize the previous formula. Considering the case to store
in multiple caches, with the aim to find the optimal trade-off between delivery and storage cost,
the question is how many times to replicate the same content and where to put these replicas.
A possible strategy is the following: if we want to replicate an object M times, we can split the
overall network in M clusters, individuate a cluster head and place the object in the cluster head.
Alternative clustering techniques can be used:
o k-center dominating set: fixed the number of clusters, find the cluster-centers, and the
related partitions, that minimize the maximum distance between each node in the cluster
and its cluster-head;
o k-distance dominating set: fixed a maximum distance D of each node from the cluster-
head, find the minimum number of clusters guaranteeing that in each partition the
distance between the nodes and the cluster-centers is at most D.
A dominating set of a graph 𝒢 = (𝒱, ℰ), is a subset 𝐷 ⊆ 𝒱, such that every node 𝑣 ∈ 𝒱 is
either in D or adjacent to a vertex of D. In this work, we used the first approach, proposed in
[RM05], to find the optimal value of M which minimizes the overall cost for delivering
content ℓ, for each possible cluster-head, adding the storage cost S times the total number of
clusters:
where j vary within the m-th cluster 𝐶𝑚, with m = 1,…, M. In Figure 2.3.3-4, a numerical
example is reported for a network of 300 nodes, where the optimal value of M = 60 clusters in
which replicate the most popular content among 200 different objects is found. It can be noted
that as M increases, the storage cost increases linearly, whereas the transportation cost
decreases. Then, we find a unique optimal value of M that minimizes the overall delivery cost
and depends on the ratio between the cost of storing and the cost of transporting a content
object.
Figure 2.3.3-4 Total Energy cost vs. the number of clusters M
Dynamic ON/OFF and resource optimization at access
Owing to the C/U-splitting architecture, C-plane is delivered via large coverage macro BS while
U-plane can be served on both conventional macrocell as well as novel edge cloud BSs such
that dual connectivity is maintained. This scheme is beneficial for guaranteeing both stable and
high throughput communication of UEs. For instance, UE can keep the C-plane connection
active via the long-range macro cell, and adaptively activate U-plane connection to either
conventional macro cell or novel high throughput edge cloud according to both UE and network
status. Owing to the common C-plane managed at macro BS, the proposed heterogeneous
network architecture enables centralized dynamic resource management (DRM) as depicted in
the below figure. When the UE is in idle state it is connected to the legacy macro base station
and disconnected from any edge clouds, which enables the smallcell BS to be turned off. This
reduces both UE power consumption and network power consumption. Based on an optimized
scheduler, the session is initiated by a request from the macro base station on the C-plane to
alert the UE. Then a U-plane connection is set up between the UE and an appropriate BS. In
the case the connecting BS is an edge cloud, data for the UE will be cached at the storage of the
edge cloud in advance. In the case numerous UEs are connected to an edge cloud, the C-plane
is also responsible for orchestrating access resources among the active UEs.
Figure 2.3.3-5 Position of the proposed orchestration algorithm marked in RED color.
On the other hand, in dense urban scenario, network densification is necessary because of the
high traffic volume generated. Typical environments are open squares, street canyons, stations,
etc., where users tend to gather and move as large and dynamic crowds while want to keep
connectivity to the cloud. Our measurement results of downlink mobile traffic in 2014 around
a famous station in metropolitan Tokyo revealed that the traffic distribution was not uniform
and there were several hotspots. Moreover, the locations of these hotspots vary along with time
and disappear at midnight, which emphasizes the necessity of a dynamic ON/OFF scheme for
reducing CAPEX/OPEX.
The proposed algorithm attempts to maximize the following evaluation function which takes
into account both system rate and consumed energy. Here, W [Hz] is the assigned bandwidth
of the respective base station, C [bps/Hz] denotes the data rate calculated from SINR, L [bits]
is the UE’s demanded traffic, T [s] is the frame length of the base station, t [s] is the time at
the access allocated to a specific UE, Sn denotes the total number of active edge clouds.
Furthermore, to guarantee that the demanded traffic is delivered within its latency constraint,
the following minimal constraint is also imposed to the optimization problem.
Numerical results are summarized in the two following figures. The first figure shows that the
proposed approach can significantly reduce the network consumed energy. The second figure
shows that our algorithm does not scarify the satisfaction ratio of UE, even when BSs are
deactivated partially. The black line in the figure show the bad result of HomoNet (a network
architecture of only macro cells without small cells) for reference.
Figure 2.3.3-6 Power reduction effect
Figure 2.3.3-7 Guarantee of UE satisfactory
Subtask 3.3.3 – Resilient design and detection of network criticalities
In this subtask, during the second year, URome developed two algorithms corresponding to the
two goals of the subtask. The first one, based on the small perturbation analysis of network
topologies presented in [CEB18], is used for the robust resource allocation on links of the
meshed backhaul network in order to keep connectivity in the mmWave backhaul network. The
second one, based on the tools of topological signal processing and presented in [BSE18], is
used for the detection of network criticalities, such as anomalous injection of traffic or nodes
in which traffic is dropped.
Resilient design
For this part of the subtask, we derived a small perturbation analysis for networks subject to
random changes of a small number of edges (links). The aim of the analysis is to derive an
efficient tool to identify the most critical links, i.e. those links whose failure will have the largest
impact on the propagation of information throughout the network. This analysis leads to a
consequent resilient resource allocation strategy. The importance of the analysis lies on the fact
that different links have different importance in the connectivity of the network. As an example,
if there is a single link whose failure will disconnect the network in two parts, a special care
must be devoted to protect this link, for example with the allocation of more resources. To
identify the criticality of each node, in [CEB18] we developed an algorithm based on the small
perturbation theory. The purpose of our algorithm is to allocate resources in such a way that the
graph connectivity perturbation is minimized. We applied our algorithm to both SISO and
MIMO systems, obtaining good performance in terms of average perturbation. For the result,
we considered an exemplary network with certain outage probabilities for each link, and used
our algorithm for the power allocation on all the links of the network.
Figure 2.3.3-8 Comparison between SISO (red curves) and MIMO (𝒏 = 𝟒) systems (blue curves) with
and without optimization
As a numerical example, in Figure 2.3.3-8 we plot the expected perturbation of the network
connectivity versus the total power spent to establish all links. In the figure, we compare the
average connectivity of the network obtained using our optimization procedure or allocating
the same power over all links, assuming the same overall power consumption. From the Figure
2.3.3-8, we can observe a significant gain in terms of the total power necessary to achieve the
same expected perturbation of the network connectivity, or equivalently, the gain in terms of
the expected perturbation of the network connectivity for the same power consumption. We can
also observe the advantage of using MIMO communications, at least in the case of statistically
independent links.
Detection of network criticalities
The objective of this investigation is to analyze data traffic in order to detect anomalous
injection of traffic or nodes in the networks where traffic is dropped. The results obtained for
this problem are presented in [BSE18], in which we built our results on the concept of simplicial
complexes and topological signal processing. Without going into the details, we considered a
network as in Figure 2.3.3-9 where on each link a certain amount of traffic is flowing. The
traffic is encoded by the color of the links. In particular, the darker is the link, the more traffic
is flowing on that link. It is clear that, only observing the traffic on the links, it is not possible
to highlight criticalities, such as an anomalous injection of traffic from some nodes. We
designed a filtering strategy able to emphasize the component of the traffic that highlights such
anomalies and to reduce the effect of data traffic measurement errors.
Figure 2.3.3-9 Observed flow
In Figure 2.3.3-10, we show the results of the proposed filtering strategy. Looking at the
intensities over the edges, after processing, we can clearly observe that there are two nodes that
inject much more traffic than the others.
Figure 2.3.3-10 Reconstruction of the anomalous traffic
Relationship to other tasks
T3.3 takes input from T1.1 and T1.2 on uses cases and architecture and aims to contribute
to the overall architecture.
Clustering algorithm based on content popularities uses parameters learned in T3.2
T3.3 provided an algorithm for the system level performance evaluation carried in WP4.
Other algorithms will be used in WP4 for performance evaluation
T3.3 is also strictly associated to the work performed in WP2. In particular, the data plane
methods and the caching strategies developed in WP2 will be closely followed and
incorporated in the development of T3.3.
2.4 Work package 4: 5G System Evaluation and Proof of Concept
Contributors: FHG, Intel, TI, URom, TTech, KLAB, PANA
Work package 4 is dedicated to the evaluation of the 5G system performance enhanced by the
MiEdge concepts using system level simulation tools and real world field tests in the 5G Berlin
Testbed for indoor and outdoor. The work includes numerical system simulations on high
performance computing clusters to capture system relevant KPIs and effects in repeated and
controlled modelling environments representing typical use cases under investigation. After all
necessary steps of development, integration and testing are completed the key MiEdge network
components we will conduct field trials for specific use cases and scenarios which were
evaluated by system level simulations before. With D4.1 the simulations and system level
performance evaluation was completed and the implementation and design for the testbed
deployments is currently in progress for the upcoming deliverables D4.2 and D4.3.
The Over-the-Air tests are being conducted in the 5G Berlin and Tokyo Tech testbeds in real
world environment for feasibility proof of concept and evaluation of the key achievements of
the 5G-MiEdge project.
Joint evaluations and demonstrations between EU and Japan are the central scope of WP 4.
2.4.1 Task T4.1: System level performance evaluation
Contributors: TTech, FHG, Intel, URom, PANA
Task period: M07 – M24
Task status: completed
Task 4.1 focuses on evaluation of the 5G-MiEdge technology components and their
orchestrated overall network performance using system level simulation methodology enabling
flexible mapping of scenarios and use cases to system level simulation tasks and extraction of
5G system performance relevant KPIs.
Output
Deliverable D4.1 has been finalized.
Construction of system level simulator to evaluate the performance of 5G-MiEdge
technologies in two specific scenarios i.e. data prefetching and computation
offloading.
The work has produced the following publication:
o H. Nishiuchi, G. K. Tran, K. Sakaguchi, “Performance Evaluation of 5G
mmWave Edge Cloud with Prefetching Algorithm,” VTC2018-Spring, IEEE,
Jun. 2018.
Deliverable
Task 4.1 released deliverable D4.1.
Del.no. Deliverable name Task no. Due
4.1 Performance evaluation of 5G-MiEdge based
5G cellular networks
T4.1 M24
Highlights
Conclusion of Deliverable D4.1.
T4.1 focused on developing tools for performance evaluation of the proposed architecture in
5G-MiEdge via numerical analysis.
System level simulator architecture
Figure 2.4.1-1 Figure Block diagram of the system level simulator
Figure 2.4.1-1 Figure shows the block diagram for design of the overall System Level
Simulator (SLS), which the core network is omitted to describe. It is separated into a
conventional Long Term Evolution (LTE) system and the novel mmWave small cells system
for both uplink (UL) and downlink (DL). For the LTE system, it contains macro cells where
one macro cell is responsible for the control (C)-plane of all base stations (BS) within the cell.
LTE system’s parameters are based on 3GPP specifications. For the novel mmWave small cells,
only the mmWave user (U)-plane access links with UE was implemented. The access link
parameters are adopted from IEEE 802.11ad (WiGig) as well as from parameters based on the
MiWEBA project measurement results [MWB-D5.1]. The mmWave small cells are assumed
to be connected to the LTE network (C-RAN) by backhaul or front haul. With the above
architecture, UE C-plane are only connected to the macro LTE, while UE U-plane can connect
to both systems (LTE macro or mmWave small cells). It should be noted that functionalities
written in RED color in Figure 2.4.1-1 Figure are novel ones which are quite different from
conventional 3GPP ones.
As all UE are commonly connected to the macro BS’s C-plane, the architecture allows
user/application centric orchestration, developed in T3.3, such as optimal user association and
scheduling, time/frequency/space resource assignment etc. Furthermore, novel models
imitating UE’s movements and generated traffics are developed to evaluate edge clouds’
capabilities e.g. prefetching/computation offloading effectiveness, packet latency etc.
To evaluate the real system level performance of LTE with mmWave overlay as HetNet, the
simulator calculates the instantaneous SINRs of the UE for both conventional LTE and
mmWave modes for each time slot of the user scheduler. The SINR metrics are then
recalculated into the modulation-coding schemes (MCSs) for both PHYs of LTE and mmWave
modes and the corresponding packet error rate (PER), by using LTE and mmWave PHY
abstraction methodologies inherited from the deliverable D4.1 of MiWEBA project [MWB-
D4.1], thus providing the effective user throughput for each PHY. This information may be
further used for dynamic resource management at C-plane. Long-term performances, e.g.
average throughputs, are calculated at the end of the scheduler by adapting to a time/frequency
varying channels implemented for each UE in the SLS.
Furthermore, the above architecture is extended to evaluate performance of data prefetching
using mmWave edge cloud (shown in Figure 2.4.1-2). The architecture is a HetNet that is
composed of several mmWave small cells overlaid on a conventional macro cell deployment.
The macro cell BS collects context information such as user mobility and traffic through the C-
plane and deals with small bandwidth and real-time traffic in the U-plane. The small cell BS
deals with large bandwidth traffic in the U-plane, that the utilization of mmWave high speed
access is needed. Based on the latest status of the fiber penetration in the world, it would not be
possible to broadly deploy the mmWave access, due to fact that the huge amount of data flows
through the mmWave access link. Such the huge amount of the data flow would not be properly
managed by the backhaul without broad fiber deployment. It would act in this case as the
bottleneck of the system. To address such limitation, storage could be installed in each small
cell BS, for allowing data prefetching. The prefeching is expected to release the backhaul
bottleneck and achieve high data rate and short delay to receive full potential of mmWave
access.
Figure 2.4.1-2 Illustration of 5G cellular network using mmWave edge cloud
The simulation procedure is composed of four sections, setting simulation parameters (yellow),
generating all status of users, BSs and channels (blue), measuring user statuses over time (red),
and saving the simulation results (green). In the first block, all parameters used in the simulation
are defined, such as total number of UEs and BSs, antenna configurations with respect to macro,
small cell BS and UE, carrier frequency and bandwidth etc. Simulation parameters are
summarized in Table 2.4-1. In the second block, deployments of macrocell, small cell BS and
UE are performed. We consider one macrocell with radius R, surrounded by six macrocells that
generate interference, and overlaid randomly by several hotspots with radius r. One small cell
BS is set at the center of the hotspot. A 3GPP hotspot model, based on the ratio between the
number of UE in the hotspot and outside of it, is used for defining the positions of the UE, most
of them are dropped within the hotspots [TR36.814].
Figure 2.4.1-3 SLS procedure extending for data prefetching.
Table 2.4-1 Parameters of deployment
Parameter Value
Macrocell radius 𝑅 250 m
Hotspot radius 𝑟 40 m
Number of hotspots 12
Number of small BSs 12
Number of users 200
User dropping model 3GPP hotspot
A detailed antenna structure and all channels between users and BSs are generated using the
QuaDRiGa channel generator. The channel generator enables the modeling of MIMO radio
channels for specific network configurations such as heterogeneous configurations. Table 2.4-2
shows parameters of UE and BS antenna configuration. Macro and small cell BSs basically
uses values of 3GPP and IEEE802.11ad standard. Each macro and small cell BS has 3 sector
antennas.
Table 2.4-2 Parameters of UE and BS antenna configuration
Parameter Value
Carrier frequency (macro/small) 2.1 GHz / 60 GHz
Bandwidth (macro/small) 10 MHz / 2.16 GHz
Number of sectors (macro/small) 3 / 3
Number of antenna elements (macro/small/UE) 4 / 128 / 2
Antenna height (macro/small/UE) 25 m / 4 m / 1.5 m
Antenna beam pattern (macro/small/UE) 3GPP macro / 11ad channel
model /Half wave dipole
Tx power (macro/small) 46 dBm / 10 dBm
Development of novel traffic model with mobility
Mobility model
Users are created following the 3GPP hotspot model. After the initial UE allocation, users’
movements are triggered. However, the 3GPP hotspot model does not support movements and
therefore we have developed a new mobility model to take into consideration the movements
and the time scale. Figure 2.4.1-4 shows an example of user’s mobility in the macro cell under
evaluation. The user’s movement process works as follows:
1. The user selects one hotspot area randomly as a destination when entering the evaluation
macro cell.
2. The user moves to the destination and stays there for a certain time.
3. The user goes outside of the macro cell.
4. The user enters the macro cell as a new user, then goes out of the macro cell again and
again until the end of evaluation time.
The user moves from the initial position (circle) to the end position (square). Destination (star)
in Figure 2.4.1-4 is the final destination, which changes depending on the state, i.e. if the user
goes to the hotspot area or outside of the macro cell. The user stays at nomadic position
(triangle). Table 2.4-3 shows parameters of the mobility model. Figure 2.4.1-5 Figure
2.4.1-5shows user’s spatial distribution. It can be derived from this figure that the ratio between
the average number of UE in the hotspot and that outside the hotspot matches the 3GPP hotspot
model.
Traffic model
The size and interval of user demanding traffic are generated by the new developed traffic
model for the simulation. As a conventional model, a traffic model is used whose size is decided
following a gamma distribution, which is created by fitting a measurement data, and the
obtained average parameter is multiplied by 1000, in anticipation of 10 years future. Table 2.4-4
shows the parameters of the traffic model. However, in realistic environment, it should be
considered that user traffic depends on user position and status. Accordingly, based on
conventional model, we developed a new traffic model. In the model, walking users demand
small data (e.g. mail, web), whereas users that do not move demand large data (e.g. video,
application update). Figure 2.4.1-6 shows the cumulative distribution function (CDF) of the
new traffic model. Red and blue traffic are generated by the staying and the moving users,
respectively. It can be seen that large traffic and small traffic are separated depending on user
state while the distribution confirms a good match with the conventional model.
Table 2.4-3 Parameters of mobility model
Parameter Value
User speed 1 m/s
Grid width 25 m
Avg. staying time (exp dist.) 500 s
Table 2.4-4 Parameters of traffic model
Parameter Value
Traffic size Gamma distribution with shape and scale parameters
of 0.2892 and 2.012 ×105 respectively
Traffic bias 4 kbit
Traffic interval Exponential distribution of avg. 8 s
Timeout 60 s
Figure 2.4.1-4 Illustration of user’s mobility
Figure 2.4.1-5 Illustration of user’s spatial distribution
Figure 2.4.1-6 Illustration of CDF of traffic size
Backhaul resource allocation (Prefetching)
We assume heterogeneous networks considering limited backhaul resource. In the environment,
the way to allocate the limited resource greatly affects the performance.
Prefetching process
In this section, it is explained how to prefetch user data in reality. For prefetching, it is important
to select appropriate indices of small cell BS s, user u and traffic n. They are decided based on
context information collected via C-plane of macrocell BS. Figure 2.4.1-7 shows illustration of
a periphery of the small cell BS s. The process is as follow:
1) Get user destination information by any application (e.g. calendar)
2) Predict traffic and connectable small cell BS 𝑠 at the destination
3) Regard user as target to prefetch demanding data within time window Tp when user
approaches the destination
4) Select user 𝑢 and traffic 𝑛 by prefetching algorithm
In the process 2, connectable small cell BS means a BS with the highest SINR prediction. We
assume it is possible to predict SINR (communication area) using a power map made via
measurements beforehand. In the process 3, Tp is a time window to start prefetching. In the
process 4, prefetching algorithm means an algorithm which selects proper user u and traffic n
in the users regarded as target to prefetch demanding data in the process 3. The algorithm is
explained in the next section.
Figure 2.4.1-7 Illustration of a periphery of the small cell s
Prefetching algorithm
In the last step in the prefetching process, combination of user u and traffic n is selected by
prefetching algorithm. Figure 2.4.1-8 illustrates traffic demand at the small cell s. The
horizontal and vertical axes show time and traffic demand, respectively. A user u demands data
n whose size is Lu,n at time tu,n. Now, we consider how much backhaul resource CB should be
allocated at time t. In this simulation, we compare two algorithms. The one is round robin (RR)
which randomly selects user u and traffic n. The other is weighted proportional fairness (WPF)
which sets an objective function considering user context information and selects user u and
traffic n to maximize it. The objective function Ou,n(t) is defined as
,
, ,
p
,
,
( ) ( )( )
( )
u n
u n u n
u
u n
u n
LO t w t
B t
Tw t
t t
where Bu(t) is the allocated backhaul resource to user u until time t, wu,n(t) is weight coefficient
taking into account the traffic generation time tu,n for user u and traffic n at time t. wu,n(t) is a
ratio of Tp against margin time defined as the difference between tu,n and t. α is called PF
coefficient which changes priority of weight coefficient. There are two reasons why we set the
function form like this. First, large traffic should be accommodated to small cell BS to utilize
fully mmWave high speed capacity (conversely it is possible to deal with small traffic with
macrocell). Second, wu,n(t) increases as traffic generation time tu,n approaches time t, that means
to allocate data in advance as much possible in the storage before users really download the
traffic.
Figure 2.4.1-8 Illustration of traffic demand at the small cells
Performance evaluation on data prefetching
The simulation results are shown in this section. Figure 2.4.1-9 shows the system rate defined
in previous section for each backhaul capacity on the condition that the number of users, the
storage limit and the time window are 200, infinity and 500s respectively, against all the
considered prefetching algorithms (RR, WPF and no prefetching). In the figure, red circle,
green triangle and blue square show WPF algorithm, RR and without prefetching, respectively.
In the case of zero or too small backhaul capacity, small cell BSs do not function and the system
rate is equivalent to only macro cell rate which has about 100 Mbps. In the case of 10 Gbps or
more for backhaul, the system rate achieves the maximum rate even without prefetching. This
means prefetching function is not needed with sufficient backhaul. However, it is unlikely from
the viewpoint of current low optical fiber penetration rate in the world. The most noteworthy
point is at 1Gbps (109) backhaul. The system rate with prefetching is much bigger than that
without prefetching. The result proves that it is possible to improve deterioration of system rate
thanks to effect of prefetching and storage. In addition, the system rate with WPF algorithm is
bigger than that with RR and achieves about 95% of maximum rate without prefetching at
10Gbps backhaul. The result validated superiority of including the algorithm.
Figure 2.4.1-9 System rate for each backhaul capacity
Performance evaluation on computation offloading
In this section, the focus is on offloading computationally heavy tasks from mobile devices to
edge cloud resources through mmWave small cells. The 5G cellular network architecture with
mmWave edge cloud we consider is shown in Figure 2.4.1-10. For our evaluation with this
architecture, U-plane is handled by mmWave small cell BS due to the high-speed access
necessary to meet strict latency constraints. The responsible entity for running users’
applications is a mobile edge host (MEH), accessible from the mmWave small cells through a
backhaul (blue line in the figure). The MEH enables mobile users to offload heavy applications
for running them within low latency constraints, with reduced power consumption. Given this
architecture, in the next sections we will formulate the problem of computation offloading and
show an exemplary scenario considering user mobility and multi-link communications as a way
to counteract blocking events, typical in mmWave communications. Finally, we will show
performance evaluation.
Figure 2.4.1-10 5G cellular network with mmWave edge cloud for computation offloading
Problem statement
In our evaluation, we assume that the applications the users are running (e.g. real time object
recognition) are too computationally heavy to be run at the UE side, so that offloading is
necessary. As performance metrics, we use computation queues, communication queues, and
the transmit power for the offloading. It has been extensively studied that it is convenient, in
terms of transmit power, to jointly allocate communication and computation resources. Here,
radio and computation resources are jointly optimized in a slot fashion, based on the current
queue states, with the objective of minimizing the total transmit power under latency
constraints. We consider time as divided in slots, and in each time slot a certain amount of
computation requests is generated, with a corresponding number of bits to be transmitted to
transfer the execution to the MEH. Each user is also capable, when possible, of communicating
with multiple links (two in our scenario) at the same time, to reduce the power consumption
and to counteract blocking events typically in mmWave communications. Each AP is then
connected to a MEH through a high capacity backhaul, as shown in Figure 2.4.1-11, where
vehicles represent obstacles, i.e. typical blocking objects.
Figure 2.4.1-11 Computation Offloading Scenario
We denote by 𝑄comp,𝑘(𝑡) the computation queue of user 𝑘 at time slot 𝑡 (CPU cycles) and by
𝑄comm,𝑘(𝑡) the communication queue (bits), by 𝑝𝑘,𝑖(𝑡) the transmit power of user 𝑘 over link
𝑖, 𝑎𝑘,𝑖(𝑡) the channel gain, 𝐵 the bandwidth, 𝑓𝑘 the computation resources allocated to user 𝑘
by the MEH (CPU cycles/s). Then, the optimization problem in time slot 𝑡 can be written as
follows:
where
is a decision variable, based on the availability of the 𝑖-th AP for user 𝑘. In particular, user
𝑘 will transmit over link 𝑖 only if this is not blocked. Constraint 𝑖) is the latency constraint, so
the one that couples the communication and the computation time. Constraint 𝑖𝑖) and 𝑖𝑖𝑖)
ensure that the transmit power is in the feasible region of the optimization problem, i.e. the
transmit power of a device is positive and less than a power budget. Constraint 𝑖𝑣) is relative
to the feasible region for the computational resources. In particular, computation resources
allocated to a user cannot exceed the total computation power of the MEH. Finally, constraint
𝑣) ensures that the sum of all computation resources allocated to the users admitted to the
offloading does not exceed the total computation resources of the MEH. Indeed, in each time
slot the algorithm also performs an admission control phase, based on the feasible set of the
problem. Of course, if a user finds both links blocked, it is not admitted to the system. In
particular, when a certain user is not admitted to the offloading procedure, its computation and
communication queues are cumulated in its buffer during all time slots, until it is not admitted
again. Then, it will have to transmit all the information that has not been transmitted and the
MEH has to perform all the computations that have not been performed. This condition will
affect the overall transmit power. The evolution of the computation queues can be written as
follows
where 𝐴𝑘(𝑡) is the amount of computation requests arrived at the previous time slot, modeled
as a random process with mean �̅�𝑘
Scenario Definition
In Figure 2.4.1-11 we show an exemplary scenario, composed by a road intersection and three
users wishing to offload their applications to a MEH through two different mmWave APs. We
consider that, due to obstacles (represented by vehicles in this case), the communication toward
a specific AP can be interrupted. When both links are available, each user sends information to
both APs, which are linked to a MEH through a high capacity backhaul, so that they send
information bits necessary to transfer the applications. In this scenario, UE 3 is moving while
the other two users have no mobility, but all users continuously send information bits to be
processed (e.g. images for object detection). While UE 3 moves, a blocking event may occur,
as in Figure 2.4.1-12, where we show some snapshots of the evolution of the scenario.
Figure 2.4.1-12 Evolution of the scenario
In this setting we compare the performance of the double link case with a single link case in
which, when a blocking event occurs, the user has to wait until the link is again available. We
evaluate the performance in terms of power consumption, while looking at the evolution of the
computation queues.
Simulation parameters and performance evaluation
In this section we show the performance of computation offloading with simulation in
MATLAB environment. U-plane is handled by mmWave small cells, with parameters as
defined in Table 2.4-1. We assume that both mmWave AP and the UE are capable of exploiting
beamforming techniques thanks to multi-antenna systems. Other simulation parameters are
related to MEH capabilities (edge cloud related parameters) and UE’s radio.
Table 2.4-5 Simulation parameters
Parameter Value
Average computation request arrival rate �̅�𝑘 107
Bits to be transmitted in each time slot 5*105
Computation power of MEH
[CPU cycles/s]
3*109
Slot duration
[ms]
100
Latency constraint [ms] 10
Number of UE 3
UE mobility 1 m/s
Carrier frequency [GHz] 60
In Figure 2.4.1-13 and Figure 2.4.1-14 we show the performance evaluation results of the
proposed algorithms in terms of cumulative transmit power and computation queues,
respectively. Each evaluation is done for a single link case, in which a user is capable of
communicating with a single link at a time, and for the double link case. Looking at Figure
2.4.1-13, it is possible to see the gain of exploiting double link communications, in terms of
transmit power, both during non-blocking conditions (from time slot 1 to around time slot 200)
and when a blocking occurs (around time slot 200). Indeed, in this case, the UE pauses the
transmission until blocking is resolved and holds cumulating computation requests and bits to
be transmitted. When the blocking ends, the UE transmits all the cumulated requests, increasing
the transmit power due to the latency constraint. In Figure 2.4.1-14, we show the evolution of
the computation queues, i.e. the amount of CPU cycles left for running the application. In
particular, since both UE1 and UE2 do not experience blocking events, their computation
queues are stable for the whole period of the evaluation. On the other hand, the computation
queue of UE3 is comparable to UE1 and UE2 queues during all the evaluation, except around
slot 200, where it grows, due to the occurrence of the blocking event. This happens only in the
single link case, since UE3 is unable to communicate during all the duration of the blocking
event. After time slot 200, UE3 is able to communicate and sends all the cumulated bits with
the corresponding computation requests to the MEH. This causes an increase of the transmit
power and a consequent drop of the computation queues to the values similar to UE1 and UE2.
Instead, in the double link case, the computation queues of UE3 are comparable to UE1 and
UE2 for the whole duration of the evaluation.
Figure 2.4.1-13 Overall cumulative Transmit power vs. Time slot
Figure 2.4.1-14 Evolution of the computation queues vs. Time slot
To summarize, in this section we evaluated the performance of computation offloading in a
urban scenario with the mmWave edge cloud, showing the advantages of multi-link
communications and the dynamic evolution of computation queues. Multi-link
communications are shown to be advantageous in terms of transmit power and computation
queue stability.
Relationship to other tasks
Realistic context prediction determined in WP3 needs to be considered.
2.4.2 Task T4.2: Development of common/joint 5G-MiEdge Testbed
Contributors: FHG, TI, TTech, KLAB, PANA
Task period: M13 – M30
Task status: running
Task 4.2 focuses on the setup of a joint testbed, this includes development and testing of 5G-
MiEdge technology components and orchestration in a real world deployment. We finished the
initial setup of a testbed in both Fraunhofer Heinrich-Hertz Institute in Berlin and Tokyo
Institute of Technology. Having almost identical setups on both sites enables us to transfer
progress and the latest achievements. With to the tight collaboration between the two partners,
they even visited the other sites to help on integrating our latest technology. Thanks to the
efficient partnership, we can now work on realization of the planned features with maximum
efficiency.
There were already some experiments on both sites, the results are explained in the following
sections and the progress to D4.2 looks very promising.
Output
The work in Task 4.2 is split in four subtasks. In the following sections, the output of those
subtasks is presented.
Development of MiEdge AP with integration of mmWave wireless access and edge cloud
Contributors: FHG, TTech, PAVC
In this subtask, a MiEdge AP is developed. It offers millimeter-wave backhaul and access
connectivity and edge cloud features. This task is based on all the concepts from work the
other work packages. The resulting AP will be used in the testbed for evaluation of all the
developed scenarios and therefore has very strict requirements that need to be met:
Latency below 1 ms
Data rate over 1 Gbps
Backhaul range over 100 m
Running edge cloud applications
Dynamic reconfiguration capabilities
Central orchestration
In order to match these requirements, there is an urgent need for millimeter-wave
communication in backhaul and access. This will allow data rates of multiple Gbps and a low
round-trip latency of merely 0.5 ms.
In Figure 2.4.2-1, a prototype of the developed MiEdge AP node is displayed. It consists of
one computation device for the system and edge cloud capabilities and multiple mmWave
backhaul links. On top is the first movable mmWave backhaul link, a Panasonic 802.11ad
(WiGig) module, enhanced with a beam forming array to increase the range to over 200
meters. Below is another mmWave backhaul link. Followed by the computation device, an
Intel NUC. For the access part, we use commercial routers and clients, not visible in this
picture.
Due to the multiple backhaul links, the deployed system can form a mesh topology with
redundant links. Additionally the links can move and align dynamically, allowing a flexible
reconfiguration of the whole topology.
These features allow the network to adapt to changing demands of our developed scenarios.
When looking at e.g. the dynamic crowd scenario, there is a large network with a high
number of users, concentrated in a small area of the network and moving through the
deployed network.
In order to match this dynamic load on the cells, the network needs to adapt. First it can
change the traffic routes, make sure all links are used and the load is distributed evenly. When
this is not enough, the topology has to be adapted, break up existing links to areas with less
traffic and align them to the heavily loaded areas. A similar approach is made for the mobile
edge computing. Pre-fetched/cached content and computation task running on one node can
Figure 2.4.2-1 MiEdge AP Node
be accessed through multi-hop routing when the user moved to another node, but the content
can also be transferred to the next node. When using smart algorithms, the movement path
can be predicted and those operations are performed before the user actually moves between
nodes, drastically increasing the overall performance of the network.
And for smooth operation, every aspect is centrally orchestrated by a SDN. Automatic rules
are applied to known scenarios, other scenarios are learned and continuously improved by the
orchestrating algorithms.
Integration and functional testing
FHG in Berlin and TTech in Tokyo are doing the development of the MiEdge AP, both
locations have almost identical setups. With our tight cooperation, we share the latest
developed algorithms and results.
The topology of our two existing testbeds is shown in Figure 2.4.2-2. It consist of four
MiEdge APs, like the one presented in Figure 2.4.2-1 MiEdge AP NodeFigure 2.4.2-1,
interconnected with four dynamic mmWave backhaul links and offering edge cloud
computing and power state control. Users can connect via mmWave access. The orchestration
is using the designed liquid RAN control plane to issue commands and receive node, link and
position information from the nodes and users.
The integration and functional testing was completed successfully. The orchestration by a
SDN controller is working perfectly. User traffic can be dynamically routed through the
network using one or more intermediate hops. Links can be re-aligned towards other nodes to
dynamically adapt the network topology. The power states are set according to the current
network load.
Figure 2.4.2-2 Testbed architecture
In both labs we conducted several tests for system performance evaluation, measuring
different KPIs like throughput while utilizing the implemented features.
Figure 2.4.2-3 shows the topology for a series of tests we conducted in Berlin earlier this year
[BMSB2018]. There is a sender on the one side of the network; a receiver on the other side
and two possible routes in between, both routes use two hops. This setup was used to stream a
video from the sender to the receiver and measure latency and throughput while evaluating all
the possible reconfiguration options. The results are shown in Figure 2.4.2-4, the experiment
starts with N3 being powered off. First the links N1-N2, N2-N4 are initialized and the
measurements start. Then N3 is powered on, the interfaces are aligned and links N1-N3, N3-
N4 are initialized. With the distribution of flow rules, the data between sender and receiver is
now routed via the second path and there is no noticeable decrease in throughput or increase
Figure 2.4.2-3 Testbed performance evaluation architecture
Figure 2.4.2-4 Testbed performance evaluation results
in latency. The experiment continues with powering down N2, powering it on again,
configuring the links N1-N2, N2-N4 and switching back to the first route. Like the first time,
this reconfiguration has no noticeable effect on throughput and latency. Then the
measurements are stopped and the experiment finishes.
This experiment shows that the testbeds SDN controlled dynamic routing, alignment and
power control performs perfectly and can be used in future experiments.
Our second experiment, a collaboration of FHG and TTech in Tokyo, is currently in progress.
It includes edge cloud containers that can be dynamically instantiated, moved and
orchestrated by the SDN controller. Implementation and initial tests were successful and now
we’re working on an outdoor deployment for more realistic test results.
Those will be published in the upcoming deliverable D4.2.
Development of mmWave MiEdge Shower with integrated mmWave massive antennas
Contributors: TI, TTech
Wireless communication characteristics of mmWave MiEdge Shower were evaluated. The
wireless terminal consists of 60-GHz band 32x32-element waveguide slot array. The
measurement environment is shown in Figure 2.4.2-5. 3.4 Gbps was confirmed with QPSK &
1.728 Gsample/symbol. Figure 2.4.2-6 shows propagation distance characteristics of BER and
SINR. It is confirmed that BER < 10-12 is realized up to 14 m. The fabricated antennas are
going to be characterized by compact antenna test range described in deliverable 2.3 [D2.3].
Figure 2.4.2-5 Measurement environment
(a)
(b)
Figure 2.4.2-6 Propagation distance characteristics (a) Bit error rate, (b) SINR
Development of liquid RAN C-plane over 5G cellular networks and 5G terminals
Contributors: FHG, KLAB
The developed and implemented architecture features an overlaying control channel, using
liquid RAN C-plane over another radio technology (RT). This is a key component of the
deployed setup, the mmWave links for backhaul and access are not always available, can be
re-aligned and reconfigured and there has to be a reliable control channel for this
orchestration, like shown in Figure 2.4.2-2, we have implemented the liquid RAN C-plane on
all used devices:
mmWave Aps
mmWave Backhaul nodes
mmWave UEs
SDN controller
User entities
Other devices for e.g. power management
While using mmWave for data transmission between the devices.
The used liquid RAN C-plane RT is easily replaceable by another technology, for
development inside the FHG and TTech labs, was started with Ethernet, then replaced with
Wi-Fi, for the first large-scale outdoor tests conducted on the TTech campus we used
WiMAX for the liquid RAN c-plane.
Those RT technologies were resilient and all showed excellent reliability, we are confident to
easily replace them with 5G macro cellular - once it becomes available.
Deliverable
Task 4.2 did not have any deliverables planned.
Del.no. Deliverable name Task no. Due
4.2 5G-MiEdge Testbed integrating mmWave
access, liquid RAN C-plane, and
user/application centric orchestration
T4.2 M30
0 2 4 6 8 10 12 1410-1210-1110-1010-910-810-710-610-510-410-310-210-1100
Propagation Distance z (m)
Bit
Err
or
Rate
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
16
18
20
Propagation Distance z (m)
SIN
R (
dB
)
Relationship to other tasks
Task 4.2 is the actual implementation and realization of the designs developed in WP2.
2.4.3 Task T4.3: Field trials toward Tokyo Olympic
Contributors: FHG, TTech, KLAB, PANA
Task period: M19 – M36
Task status: running
Task 4.3 targets field trials with the developed MiEdge network elements in the real world
scenario of the 5G Berlin Testbed. Selected 5G-MiEdge scenarios and use cases will be
evaluated under real-world constraints in order to prove feasibility and gain insight into
optimization potential of SDN based orchestration. The field trials will create global visibility
and convincing feasibility arguments for standardization contributions.
Output
Task 4.3 has just started in January and builds upon the implemented design of task 4.2. For
this task we implemented all of the designed characteristics and features from work package 2
on the testbed and published the first results on performance evaluation [BMSB2018].
In July, partners from FHG and TTech plan to perform large-scale outdoor experiments on the
Tokyo Institute of Technology campus. These experiments will include mobile edge
computing where a UE can interact with a container running on one of the nodes. Figure
2.4.3-1 shows the planned deployment, with four nodes and a link range of 50 to 70 meters.
This deployment will offer mmWave access and a mobile UE, moving between the mmWave
Figure 2.4.3-1 Outdoor experiment on TTech campus
Aps and accessing the edge cloud container that can also be migrated between nodes to
increase responsiveness.
During the experiment runtime, we will capture important KPIs:
On UE
Throughput to container
Latency to container
mmWave RSSI
GPS position
On each node
Backhaul throughput
Container statistics
With these measurements, we can evaluate the overall performance of the implemented
system. The following experiments will take up on the achieved results and try to improve
them by e.g. adding algorithms to predict the UE movement and begin to reconfigure the
backhaul links, migrate the container in time, so once the UE is associated to the next
mmWave AP, the whole network is already prepared.
Afterwards, we can do a thorough evaluation of the selected scenarios and release them in the
final deliverable D4.3 in June 2019.
Deliverable
Task 4.3 did not have any deliverables planned.
Del.no. Deliverable name Task no. Due
4.3 5G-MiEdge field trials integrated in 5G-Berlin
Testbed toward Tokyo Olympic 2020
T4.3 M36
Relationship to other tasks
This task takes the results of Task 4.2 to further evaluate the 5G-MiEdge network
elements and implemented concepts.
2.5 Work package 5: Standardization, spectrum regulation, dissemination and
exploitation
Contributors: CEA, FHG, Intel, TI, URom, TTech, KLAB, PANA
The goal of this work package is to create awareness about the 5G-MiEdge project and its
specific objectives and technical results. Several activities will ensure that the project is
presented across various communication channels. Thereby the consortium is addressing
European and international 5G research activities (e.g. 5GPP association activities), research
societies, industry forums (e.g. Small Cell Forum and NGMN), and standardization and
regulation bodies (3GPP, IEEE, and ETSI). Finally, in order to maximize the impact of 5G-
MiEdge to the scientific community we target to publish the main results in high quality
journals and magazines.
2.5.1 Task 5.1: Dissemination and exploitation of project results
Contributors: CEA, FHG, Intel, TI, URom, TTech, KLAB, PANA
Task period: M04 – M36
Task status: running
Output
Task 5.1 led to several notable dissemination activities targeting the academia the wireless
industry, and the general public, which are resumed in this Section. The overall list of
activities is summarized in Appendix 1. It is worth to highlight that six patents have been filed
so far by the project members.
Deliverable
Task 5.1 contributed to deliverable D5.2.
Del.no. Deliverable name Task no. Due
5.1 First report on dissemination, standards,
regulation and exploitation plan
T5.1,
T5.2
M12
5.2 Second report on dissemination, standards,
regulation and exploitation plan
T5.1,
T5.2
M24
5.3 Final report on dissemination, standards,
regulation and exploitation plan
T5.1,
T5.2
M36
Highlights
5G-MiEdge gives a particular importance to the dissemination activities, in order to ensure
that all main results of the project will achieve the broadest possible impact on the global
telecommunication eco-system, both at the academic and at the industrial level.
Dissemination activities is mainly addressing the following four main target groups:
General public
Dissemination through the general public, is realized mainly through the project website and
press releases. During the second year, we have prepared a flyer that we have distributed to
the public of the different international events where 5G MiEdge partners have participated.
The project also aims to have a presence in other webpages beside it main webpage http://5g-
miedge.eu/. Here we highlight two recent articles and press releases concerning 5G-MiEdge:
Article from the German blog “electronica” on tactile internet that, amongst other 5G
activities, present 5G-MiEdge technologies and goals [electronica]
French article of the Data Analytics Post (DAP) discussing the edge computing, where
Dr. Emilio Calvanese Strinati discusses about 5G-MiEdge [DAP]
Scientific community
5G-MiEdge targets to disseminate its results in international journals and the presentations of
papers to the most important conferences, as well as via the participation to workshops and
panels in major IEEE and ACM events: At the current stage, we have five accepted
journals/magazines and one book chapter. Amongst them, there are two invited contributions
to IEEE Wireless Communications Magazine and IEEE Transaction on Wireless
Communication, respectively. At the end of the second year of the project, we have
submitted/presented thirty-eight contributions to notable international and domestic
conferences, such as IEEE ICASSP 2018, IEEE ICC 2018, IEEE GLOBECOM 2017,
EUCNC 2017, IEEE SPAWC 2017, and IEEE CSCN 2017. 5G-MiEdge partners have also
been invited to present the results of our project in several international conferences, such as
IEEE Globecom 2016, IEEE ICC 2017, and ICC 2018. The detailed list of submitted/accepted
papers can be found in D5.2 [ref].
In addition, during the second year of the project the following events have been co-
organized:
5G Test-Beds & Trials – Learnings from implementing 5G (5G-Testbed) in IEEE
Globecom 2017
SmartCom 2017 -The 4th International Workshop on Smart Wireless Communications
Economics and Adoption of Millimeter Wave Technology in Future Networks in
IEEE WCNC 2018
WDN-5G: The 11th International Workshop on Evolutional Technologies &
Ecosystems for 5G Phase II in IEEE ICC 2018
2nd Workshop on business models and techno-economic analysis for 5G networks in
EUCNC 2018
Industrial community
5G-MiEdge targets a constant attendance to the most important industrial events, participating
through the creation of stands, showroom and booths. During the second year of the project,
we have disseminated our activities during three notable industrial events, i.e., the Industry
Panel on V2X for Automated Driving in 4G, 5G and Beyond at IEEE GLOBECOM 2017, the
Industrial Panel on Economics and Adoption of Millimeter Wave Technology in Future
Networks at IEEE WCNC 2018, and the LETI Innovation Days.
Other collaboration activities
One of the 5G-MiEdge project goals during this first year has been to establish links with
other related collaborative projects in the 5G ecosystem. Several activities have in been
carried through several collaboration actions with EU (Superfluidity, SPEED-5G, VirtuWind,
5G-EX, 5G MonArch, and 5GCity, TWEETHER, and ULTRAWAVE), EU-JP (5G! Pagoda),
and JP (MiEdge+) research projects. In addition, some collaborations have been developed
with EU-Kr and EU-Br projects 5G Champion and Futebol.
Relationship to other tasks
Task 5.1 receives inputs from all the task of WP1, WP2, WP3, and WP4.
2.5.2 Task 5.2: Standards and regulatory bodies
Contributors: Intel, FHG, CEA, TI, URom, TTech, KLAB, PANA
Task period: M04 – M36
Task status: running
Output
During the second year, we did not have any concrete chances to effectively impact the work
in standardization and normative groups; however, in order to increase these changes in the
third year, we have also started to monitor an additional ETSI group (the ISG ENI), whose
activities can be likely influenced by 5G-MiEdge results. In addition, during the last year of
the project, we plan a meeting with European regulators (likely Ofcom and ANFR, with
which the first contact already started), to present the major 5G-MiEdge results. The high
number of patents filled so far from the consortium (6) highlight the quality of the innovations
produced so far, which may lead to additional contributions in the last year of the project.
It is worth to recall that at this stage of the project, several contributions have been presented
to the 3GPP SA1 group (during the first year).
As a result of these contributions 5G-MiEdge studies have impacted two 3GPP documents.
3GPP TR22.886, V15.1.0, “Study on enhancement of 3GPP support for 5G V2X
services,” March 2017.
3GPP TS22.186, V15.1.0, “Service requirements for enhanced V2X scenarios,” June
2017.
Deliverable
Task 5.2 contributed to deliverable D5.1.
Del.no. Deliverable name Task no. Due
5.1 First report on dissemination, standards, regulation
and exploitation plan
T5.1,
T5.2
M12
5.2 Second report on dissemination, standards,
regulation and exploitation plan
T5.1,
T5.2
M24
5.3 Final report on dissemination, standards, regulation
and exploitation plan
T5.1,
T5.2
M36
Highlights
5G-MiEdge is monitoring the progress and the outcome of standardization bodies working on
topics relevant for the project. When possible the consortium tries to push into standards the
main results coming out of the project. Regular reports of the main outcomes of standards
bodies are verbally shared among the project partners, so to ensure that the directions taken by
the project are aligned with the on-going 5G standardization activities.
The most relevant standards group for the topics in focus in the project, and therefore the
groups that the project aims at impacting most, are listed in the table below.
Table 2.5-1 Target standardization bodies and current impact
Forum Groups 5G-MiEdge contributions Partners Status
3GPP SA1 Use cases and scenarios for the
forthcoming 5G system architecture.
Intel, FHG 6 contributions.
3GPP RAN
groups
Impact of the newly proposed control plan
on the access strata of the 3GPP access
technologies.
Intel, FHG Under
Monitoring.
3GPP CT groups Potential impact on the upper layers of the
wireless protocol stacks coming out of the
newly proposed orchestration procedures.
Intel Under
Monitoring.
ETSI ISG MEC Improvements on ETSI Architecture.
Potential definition of a new PoC.
Intel Under
Monitoring.
ISG ENI Propose new AI solutions for mmWave
communication technologies. Propose new
use cases.
CEA Under
Monitoring.
IEEE 802.11 Propose invented schemes of mmWave as
functionalities for 5G architecture.
PANA Under
Monitoring.
In addition to standards, another important aspect for achieving an effective deployment of a
new technology is the alignment and potentially the impact on regulatory bodies. The
availability of new spectrum and a different use of the existing one are key enabling aspects
for mmWave technologies. Therefore, 5G-MiEdge will tightly monitor and support the work
that is starting in regulatory bodies in preparation of the WRC 2019.
Relationship to other tasks
Task 5.2 receives inputs from all the task of WP1, WP2, WP3, and WP4.
2.6 Work package 6: Project management
Lead beneficiary: FHG (EU) / TTech (JP)
Work package 6 includes all the specific functions assigned to the Project Coordinator and the
Technical Manager to ensure that the project successfully achieves its stated objectives on time,
within budget, and with the expected high level of quality of the technology developed. FHG
is the project leader in the European consortium, whereas this role is taken by TTech in the
Japanese consortium. FHG and TTech co-coordinate the project for both the administrative and
technical parts.
2.6.1 Task 6.1: Administrative project management
Contributors: FHG, all
Task period: M01 – M36
Task status: running
The basic purpose for project management is to ensure the correct level of coordination and
cooperation amongst the project consortium members. In this context, the successful
administrative coordination of a project is an important side of the project management, which
includes also relevant legal issues and policies for proper management of IPR. The project
administrative management covers the following activities:
Preparation and coordination of General Assembly meetings (one kick-off meeting at
the beginning of the project, at least two general assembly meetings per year, project
technical committee teleconference on a monthly basis, WP meetings as necessary) and
of technical reviews.
Setup of partners’ communication means, internal rules, common document templates;
setup of the project quality framework allowing deliverables quality supervision, risk
analysis and contingency planning, scientific outputs monitoring.
Budget administration in order to match the plans approved by the General Assembly;
management of legal and contractual obligations.
Communication and reporting with the Commissions (European Commission (EC) for
Europe and Ministry of Internal Affairs and Communications (MIC) in Japan), interface
with related projects and other parties.
Reporting technical and financial status of the project to Commissions within 60 days
after the end of each 12-month period. A final report will include all outputs over whole
period of the project.
Secure resources for administrative and financial procedures.
Output
In the second period, Task 6.2 coordinated the cooperation between the project consortium
members and managed the meetings.
Deliverable
Task 6.1 contributed to this annual report AR6.2.
Del.no. Deliverable name Task no. Due
AR6.1 First annual status report of 5G-MiEdge
project
T6.1,
T6.2
M12
AR6.2 Second annual status report of 5G-MiEdge
project
T6.1,
T6.2
M24
AR6.3 Final annual status report of 5G-MiEdge
project
T6.1,
T6.2
M36
Highlights
Third General Assembly: October 2017 in Rome, Italy
In October 2017, the third general assembly took place in Rome. The host institute was
Sapienza University of Rome, who was also the local organizer of SmartCom2017, an
international conference of which 5G-MiEdge is a technical sponsor. After the two-day
conference held at Grand Hotel Palatino, the general assembly was organized at
Department of Information Eng., Electronics and Telecommunications (University of
Rome). Presentations and discussions about our plans for the project in the 2nd year were
conducted for two days and one sub-project-manager of the consortium from Japan side
was delegated.
Fourth General Assembly: May 2018 in Tokyo, Japan
The fourth general assembly, hosted by Tokyo Institute of Technology, took place in
May 2018 at Tamachi campus of the hosting institute. It was characterized by two
intense days of interesting talks and discussions, including a discussion with 5G!Pagoda
on PoC towards the end of the two projects. The General Assembly was particularly
useful to coordinate our efforts, strengthen the bond among the partners, and optimize
the approach to meet our goals for the final fiscal year of the project.
Relationship to other tasks
Task 6.1 oversees all other work packages and the general progress of the whole project
2.6.2 Task 6.2: Technical coordination
Task leader: TTech
Task period: M01 – M36
Task status: running
The success of a medium-scale focused research project strongly depends on the quality of its
consortium. The stance (history of past experiences and international relevance) of the partners
is an early indication of such quality, but excellence can only be materialised by ensuring that
each partner is fully dedicated to the project’s objectives, which in turn requires constant
oversight. This task is designed to provide a framework within which such quality oversight
will be exercised. The leader of this task promoted two technical directors each from EU and
Japan i.e. Prof. Sergio Barbarossa (URom) and Prof. Makoto Ando (Tokyo Tech). The two
directors supervise the following activities:
Supervise the content and quality of deliverables and quality of technical and scientific
output.
Perform quality checks by reviewing and pre-approving deliverables and defining best
practices in order to ensure that the planned objectives are achieved at the highest of the
partners’ abilities.
Report and discuss quality issues and risks at project meetings and whenever needed.
Risks will be early identified and will be immediately reported to the coordinators for
management and mitigation.
Management of the technical activity, coordinating WP leader activity to guarantee
coherent research, deliveries and full achievement of the project technical objectives.
This includes cross-WP work organization, the definition of topics to address in
meetings and conference calls, technical progress monitoring. Activities also involve
the technical coordination related to studies and developed technologies, ensuring
coherent research under common working assumptions, hypothesis, parameters, and
implementation of proper measures to ensure integration of the WP results in the proof
of concept.
Risk Assessment
The identified risks are shown in Table 2.6-1. We are continuously monitoring the possible
risks, the work package leader of each work package is responsible for bringing newly
identified risks up for discussions in our monthly meetings. Those risks are identified by
monitoring the scientific publications, conferences and technical innovations related to the
work packages. The period for those risks does vary with the identified risks. Some risks exist
throughout the entire project, they were identified early and constantly monitored, while some
risks, like implementation of certain features, only manifest once the corresponding task has
started.
In our monthly meetings, we discuss possible plans to either avoid those risks entirely or to
utilize them to our advantage.
Table 2.6-1 Risk Assessment: Technical Risks
WP Risk description Impact/
Probability
Comment
ALL Cultural conflicts
between EU and
Japanese partners
Low Many of the partners have already cooperated in other projects
and activities.
ALL Specific expertise or
resource is missing
Medium The 5G-MiEdge consortium brings together leading European
and Japanese organizations with complementary skills for the
successfully achievements of the technical objectives.
Advisory boards of experts from both sides were also
established to supervise the project. All partners have
committed to mobilize all necessary resources for achieving
the project objectives and experimental facilities have been
identified based on their accessibility.
WP1 Requirements and use
cases not identified or
detailed insufficiently
to design the 5G-
MiEdge architecture
Low Use cases and preliminary requirements have been already
identified D1.1 and D1.2. The first architecture was defined
very early in the 2nd year of the project, and by adapting to
outputs of WP2 and WP3 activities.
Output
In the second period, Task 6.2 coordinated the technical consensus of the project
consortium members in all submitted deliverables.
Deliverable
Task 6.2 contributed to this annual report.
Del.no. Deliverable name Task no. Due
AR6.1 First annual status report of 5G-MiEdge
project
T6.1,
T6.2
M12
AR6.2 Second annual status report of 5G-MiEdge
project
T6.1,
T6.2
M24
AR6.3 Final annual status report of 5G-MiEdge
project
T6.1,
T6.2
M36
Highlights
Monthly teleconferences among all partners are regularly established to discuss about the
project’s progress. Also ad-hoc meetings per WP or deliverable were done on a regular basis
to keep the project on schedule.
T6.2 also invited external advisors, who are experts of the field to supervise the project.
Especially, external experts were invited to present and share their views at our general
assembly:
o Dr. Yoshiaki Kiriha (Tokyo University, Japan)
o Prof. Michele Zorzi (The University of Padova, Italy)
o Prof. Antonio Capone(Politecnico di Milano, Italy)
o Prof. Stefano Salsano(University of Rome Tor Vergata, Italy)
WP2
WP3
Solution design is not
finished in time or
includes design
mistakes
Low Reviews of technical work is conducted by work package
leaders and steering committee to ensure consistency among
WPs and fulfillment of the project objectives. When necessary,
solution options to this risk are developed by a team of
multiple partners and feedback is received from the design,
implementation, and validation activities in WP1 and WP4.
WP2
WP3
Emergence of
competitive
technologies
Low A continuous technological watch in standards and major
conference/journal proceedings will highlight the
shortcomings, and enable a fast reaction to such threats.
WP4 Unavailable hardware
resources
Medium This project was carefully planned and drafted. WP1 and WP2
laid a solid foundation for the tasks in WP4 and we share
similar resources on both testbeds, in Berlin and Tokyo. This
allows us to have frequent discussions on which hardware to
use.
WP4 Serious delays on
implementation of
planned features
Medium The shared hardware on both testbeds allows us to work on the
implementation more efficiently, we can detect possible
problems early and either avoid them entirely, or fix them
without losing precious time or resources.
Relationship to other tasks
Task 6.2 oversees all other work packages and the general progress of the whole project.
Impact
3.1 Impact on Academia and Research Centers
Contributors: FHG, CEA, URom, TTech
To maximize its impact to Academia and Research Centers, 5G-MiEdge targets to disseminate
its results in international journals and the presentations of papers to the most important
conferences, as well as via the participation to workshops and panels in major IEEE and ACM
events. In addition, to guarantee our impact in the 5G ecosystem, and in particular to the
academia, the project has co-organized (together with other collaborative research projects) five
workshops. It is worth to highlight that three of these workshops have been hosted in well
known international IEEE conferences like ICC, WCNC, and Globecom. Finally, we have
attempted to strengthen the EU-JP collaboration by organizing a workshop at EuCNC, which
is a major EU event, and at the 2018 SmartCom event, which is historically one of the most
important workshop in the Japanese ecosystem.
We are continuing to make available on our website public deliverables, scientific articles, and
advertising the project participation to international events related to the mobile community.
At the current stage, we have five accepted journals/magazines and one book chapter. Amongst
them, there are two invited contributions, one IEEE Wireless Communications Magazine and
one IEEE Transaction on Wireless Communication. We do also participate to the most relevant
technical conferences such as IEEE Globecom 2017 and IEEE ICC 2018, so to present the
objectives and the preliminary results of the 5G-MiEdge project. The impact of the 5G-MiEdge
project is increasing in the second project year, and after 11 conference papers in the first year,
we submitted/presented 27 contributions to notable international and domestic conferences in
the second year.
To conclude, CEA has filed a patent related to the studies carried out in 5G-MiEdge in WP2
(in the first year of the project) and another one (in the second year of the project) related to the
studies on WP3 and during the second year of the project. TTech has filed one patent based on
its results on WP1 and WP3. Also, CEA is starting to participate to the ETSI ENI ISG, where
CEA is planning to exploit the results from 5G-MiEdge related to artificial intelligence for
wireless networks. Similarly FHG and INT are successfully disseminating results related to 5G-
MiEdge in the 3GPP framework.
3.2 Impact on Industry and ecosystem
Contributors: Intel, TI, PANA, KLAB
In the second year of the project, 5G-MiEdge has extended its reach to the international
communities, conferences, venues and events.
The presence in the consortium of industrial players and of operators, active in two important
geographical areas like Europe and Japan, among the most active ones for the first trials of 5G
enabling technologies, gives 5G-MiEdge the opportunity to best leverage on the technical
project results, which start to be more and more important as the project proceeds in its plan of
record.
The overall final users and social impact of the project is still to come, as some technical results
appeared, but most of them will be finalized in the last year of the project.
With regard to the business aspects of the project, progress was made w.r.t. the results achieved
in the first year of the project, especially
- Use cases and scenarios have been better defined and finalized,
- A system architecture has been finalized and relevant requirements have been derived.
The mentioned above main achievements are key to provide project partners with means to
create more impacts to the standardization bodies and to have a much better understanding of
the forthcoming 5G deployments.
During the second year of the project important results have been achieved. But as even more
significant results are expected during the third and last year of the project, the 5G-MiEdge
team has decided that it is advisable to wait for a more detailed performance evaluation of the
technologies addressed in the project to be available, before performing a sound business
analysis of the considered use cases.
From the point of view of equipment providers, the new 5G network architecture description
has recently been finalized (June 2018, with more small change to happen till end of 2018).
Therefore the race to 5G has moved from standardization to pre-deployment and has entered
the productization phase. All the ecosystem is currently focusing on delivering the first versions
of equipment that can provide 5G connectivity and related KPIs.
From the terminal provider perspective, the 3GPP Release 15 set of features, as defined by
standards bodies, will just focus on the basic functionalities of the new technology, leaving to
the forthcoming 3GPP Release 16 some more room for more advanced features to appear.
Never the less it is possible to already draw some interesting comments based on the use cases
in focus in 5G-MiEdge.
Prior to 5G-MiEdge, PANA has conducted experimental trials of mmWave high-speed content
delivery at Narita International Airport. In these trials, the content data were pre-stored in the
standalone WiGig signages. The technologies related to Omotenashi service in 5G-MiEdge will
best fit for this use case, enabling efficient content update using the edge-cloud network
architecture. In addition, mmWave access technologies such as MU-MIMO, hybrid
beamforming, ultra-lean signalling etc. will provide significant increase of throughput even in
the highly populated area like a boarding gate at the airport.
The technologies related to the moving hotspot use case can be applied for content update of
the in-flight entertainment (IFE) system. The total volume of the content, which includes video,
multimedia content, games etc., reaches 500+ GB on average. Currently the content is updated
manually by replacing a HDD unit. With 5G-MiEdge outcomes, this can be done wirelessly,
which will help reduce labor costs as well as provide more frequent and efficient update
optimized for the flight destination.
V2X is a key technology for automated driving, which is one of the most
promising growth markets. The technologies developed in 5G-MiEdge are essential for
enabling massive data upload, 3D map download, cooperative sensor sharing etc. Further study
is required to check feasibility of technologies as well as validity of business ecosystem. As an
example, experimental validation of mmWave ITS system has started in Japan under a funding
from MIC, Japan. The knowledge acquired by 5G-MiEdge will be utilized in these future
projects.
In order for mobile operators to utilize the project results widely and generally, it is necessary
that the results need to be implemented in the system as standardized technologies. Technical
outcomes of control signalling and algorithms for optimizations are emerging from the project,
but further dissemination is necessary so that the necessity of the methods is understood and
they are implemented as standard technologies. Related standardization activities will be led by
WP5.
The technologies deployed in 5G-MiEdge are focusing on topics that will most probably be
discussed in 3GPP starting with Release 17, i.e. not before the second half of 2019. The project
partners are ready to provide by then some contribution especially w.r.t. the main results coming
out of WP3 and WP4.
Prior to 5G-MiEdge, PANA has conducted experimental trials of mmWave high-speed content
delivery at Narita International Airport. In these trials, the content data were pre-stored in the
standalone WiGig signages. The technologies related to Omotenashi service in 5G-MiEdge will
best fit for this use case, enabling efficient content update using the edge-cloud network
architecture. In addition, mmWave access technologies such as MU-MIMO, hybrid
beamforming, ultra-lean signalling etc. will provide significant increase of throughput even in
the highly populated area like a boarding gate at the airport.
The technologies related to the moving hotspot use case can be applied for content update of
the in-flight entertainment (IFE) system. The total volume of the content, which includes video,
multimedia content, games etc., reaches 500+ GB on average. Currently the content is updated
manually by replacing a HDD unit. With 5G-MiEdge outcomes, this can be done wirelessly,
which will help reduce labor costs as well as provide more frequent and efficient update
optimized for the flight destination.
V2X is a key technology for automated driving, which is one of the most promising growth
markets. The technologies developed in 5G-MiEdge are essential for enabling massive data
upload, 3D map download, cooperative sensor sharing etc. Further study is required to check
feasibility of technologies as well as validity of business ecosystem. As an example,
experimental validation of mmWave ITS system has started in Japan under a funding from
MIC, Japan. The knowledge acquired by 5G-MiEdge will be utilized in these future projects.
Update of the plan for exploitation and dissemination of results
The diversity in expertise of the 5G-MiEdge consortium increases the probability of a
successful impact on the global market of the technologies proposed, devised, and developed
within the project lifetime.
We have updated the identified main exploitation streams from the DoW.
4.2.1 Equipment vendors (Intel, PANA)
During the first and the second year of the project the main achieved results have been
constantly disseminated in internal meetings within the consortium participants R&D divisions,
especially focusing on the system architects communities.
Now that the 3GPP 5G phase 1 of features has been finalized and the 5G Phase 2 has been
defined and work is ongoing throughout all 2019 to finalize the related specs, the project
partners can leverage on the concluded work on use cases, scenario, enhanced 5G architecture
and related requirements.
As a matter of fact, the individual exploitation plans, as defined in the first year of the project,
have not changed much during the second year of the project. Industrial partners are waiting
for the final substantial results to be made ready by the project in the third and final year, so to
bring the main learnings into internal discussions with the system engineering teams, in order
to check the feasibility of adding project-related innovations into future products, to hit the
market after 2020.
Finally, the technological know-how build-up in the project will provide a competitive
advantage versus other companies and potential or real competitors, not involved in the project.
Indeed the possibility of filing new patents to protect the newly developed technologies is an
additional key aspect in the exploitation of the project results in the mid and long term.
4.2.2 Telecom operators (TI, KLAB)
Around this second year period, many 5G related trials have been announced in public all over
the world. Major mobile operators and equipment providers compete in showing excellent
results of those stakeholders to demonstrate technological competences. Those trials were
utilized to show 5G showcases which demonstrate not only higher bandwidth but also new
kinds of use scenarios in mobile communications like remote control of construction machines,
ultra-high-definition (4K/8K) broadcasting, virtual/augmented reality (VR/AR), automated
driving and so on. Those will look for new business with various potential partners who will be
service providers utilizing the 5G network in the future. However, the 5G infrastructures are
just deployed for specific trial purposes in limited venues yet. Therefore a lot of technical and
business challenges need to be addressed in order to realize such use scenarios of the 5G era in
general. The project results will be used as a guidance to solve parts of issues found for future
deployment of the 5G infrastructures and new services. Also the results are disseminated
broadly in academic conferences, papers and panels for better understanding. Some activities
for standardization will be tried.
In particular, TI started pre-commercial trials in the cities of Bari and Matera, with the plan to
provide full 5G coverage and services by the end of 2019. TI also plans to provide 5G service
in San Marino by 2019, with the aim to be the first European country served by 5G. The first
pre-commercial base stations have been deployed in 2018 in the above-mentioned locations,
and service concepts are being developed with local and international companies to demonstrate
the full capabilities of 5G. In such context the availability of fully standardized mmWaves
solutions is of key importance to ensure the required throughput and capacity.
The project results are very important to better understand how to design and operate a network
in the mmWaves.
4.2.3 Research institutes (FHG, CEA)
The research institutes Fraunhofer Heinrich-Hertz-Institute and CEA-LETI, the Laboratory for
Electronics & Information Technology, working in 5G-MiEdge will contribute to the project
topics to increase their knowledge on the future network design, for better understanding of
current industry needs and to identify new ideas, challenges and focus for future research
activities. The obtained knowledge will be used to invent solutions for future challenges,
transfer new technological solutions to their industrial partners, in particular network vendors,
telecom operators, service providers and others. At this stage of the projects CEA and FHG
have already produced notable scientific papers targeting both international conferences and
journals/magazines. Thanks to the results obtained in 5G-MiEdge, CEA is starting new research
topics for enabling ultra- broadband communications in frequency bands higher than those
considered by 5G-MiEdge. CEA has filed a patent related to the studies carried out in 5G-
MiEdge in WP2 (in the first year of the project) and another one (in the second year of the
project) related to the studies on WP3. CEA has showcased the first results originated in the
framework of 5G-MiEdge in the CEA LETI Innovation Days yearly event. Also, CEA is
starting to participate to the ETSI ENI ISG, where CEA is planning to exploit the results from
5G-MiEdge related to artificial intelligence for wireless networks.
4.2.4 Universities (URom, TTech)
Our plan is to continuously develop new theoretical methods and technologies to improve the
efficiency of 5G networks. The acquired knowledge will be disseminated to industry and
society through major international peer-reviewed conferences, workshops and journals. We
also plan to patent the most innovative results. The possibilities to introduce latest project
research results in the education courses of students at the university are also paramount
aspects of the exploitation of the project results. PhD and master students whose research
topics are related to 5G-MiEdge are involved in the project and they can exploit the state-of-
the-art PoC hardware and equipment developed in the project for conducting experiments and
demonstrating 5G-MiEdge's achievements both in their theses and in external publications.
Update of the data management plan (if applicable)
This does not apply.
Follow-up of recommendations and comments from previous review(s) (if
applicable)
Regarding the remarks and recommendations from our review meeting in Tokyo in
September 2017, we focused on the following aspects.
There were additional risks to take into account on the risk assessment, shown in Table 2.6-1,
taking care of WP4, which started in the second period. For each work package, the WP
leader is responsible for identifying risks and take actions to avoid or exploit them to our
advantage.
We are constantly working on increasing our exploitation process; the results are detailed in
Section 2.5.
As the final period is starting now, we are working on identifying our quality of impact,
evaluating which KPIs have the most potential.
For the deviations from Annex 1 and Annex 2 in Section 7, we provide more detailed
numbers and explanations.
Deviations from Annex 1 and Annex 2 (if applicable)
During the first period there was only one deviation from Annex 1 and 2, detailed in section
7.2.1.
7.1 Tasks
All tasks were implemented as planned.
7.2 Use of resources
The following three tables contain the actual and planned use of resources for each partner,
including differences and a sum of all WPs.
Table 7.2-1 shows the revised data for the first period, Table 7.2-2 for the second and
Table 7.2-3 for the combined past project duration.
Table 7.2-1 First period: Revised actual and planned use of resources per partner
Partner WP1 WP2 WP3 WP4 WP5 WP6 Total
Fraunhofer
Actual 2.00 3.76 0.00 0.00 1.00 2.00 8.76
Planned 1.71 3.42 0.00 1.67 1.09 1.33 9.22
Difference 0.29 0.34 0.00 -1.67 -0.09 0.67 -0.46
CEA
Actual 2.23 4.10 4.59 1.60 2.30 0.00 14.82
Planned 2.14 3.96 4.03 0.44 1.36 0.00 11.95
Difference 0.09 0.14 0.56 1.16 0.94 0.00 2.87
Intel Actual 3.36 0.00 2.12 0.00 1.16 0.00 6.64
Planned 3.86 0.00 2.48 1.00 1.36 0.00 8.70
Difference -0.50 0.00 -0.36 -1.00 -0.20 0.00 -2.06
TI
Actual 2.00 2.29 0.00 0.00 1.00 0.00 5.29
Planned 0.86 3.42 0.00 0.11 0.55 0.00 4.94
Difference 1.14 -1.13 0.00 -0.11 0.45 0.00 0.35
URom
Actual 1.70 2.20 6.20 1.00 0.80 0.00 11.90
Planned 1.71 2.28 6.21 0.33 0.82 0.00 11.35
Difference -0.01 -0.08 -0.01 0.67 -0.02 0.00 0.55
Table 7.2-2 Second period: Actual and planned use of resources per partner
Partner WP1 WP2 WP3 WP4 WP5 WP6 Total
Fraunhofer
Actual 1.07 2.44 0.00 6.06 0.75 1.07 11.39
Planned 1.43 4.56 0.00 8.33 1.45 1.33 17.10
Difference -0.36 -2.12 0.00 -2.27 -0.70 -0.26 -5.71
CEA
Actual 2.23 4.10 4.59 1.60 2.01 0.00 14.53
Planned 1.79 5.28 5.38 2.22 1.82 0.00 16.49
Difference 0.44 -1.18 -0.79 -0.62 0.19 0.00 -1.96
Intel Actual 3.29 0.00 3.52 2.00 1.97 0.00 10.78
Planned 3.21 0.00 3.31 0.00 1.82 0.00 8.34
Difference 0.08 0.00 0.21 2.00 0.15 0.00 2.44
TI
Actual 1.50 2.62 0.00 0.00 0.50 0.00 4.62
Planned 0.71 5.14 0.00 0.67 0.73 0.00 7.25
Difference 0.79 -2.52 0.00 -0.67 -0.23 0.00 -2.63
URom
Actual 1.43 2.88 8.27 2.00 1.10 0.00 15.68
Planned 1.43 2.88 8.28 2.00 1.09 0.00 15.68
Difference 0.00 0.00 -0.01 0.00 0.01 0.00 0.00
Table 7.2-3 Overall: Actual and planned use of resources per partner
Partner WP1 WP2 WP3 WP4 WP5 WP6 Total
Fraunhofer
Actual 3.07 6.20 0.00 6.06 1.75 3.07 20.15
Planned 3.14 7.98 0.00 10.00 2.54 2.66 26.32
Difference -0.07 -1.78 0.00 -3.94 -0.79 0.41 -6.17
CEA
Actual 4.46 8.20 9.18 3.20 4.31 0.00 29.35
Planned 3.93 9.24 9.41 2.66 3.18 0.00 28.44
Difference 0.53 -1.04 -0.23 0.54 1.13 0.00 0.91
Intel Actual 6.65 0.00 5.64 2.00 3.13 0.00 17.42
Planned 7.07 0.00 5.79 1.00 3.18 0.00 17.04
Difference -0.42 0.00 -0.15 1.00 -0.05 0.00 0.38
TI
Actual 3.50 4.91 0.00 0.00 1.50 0.00 9.91
Planned 1.57 8.56 0.00 0.78 1.28 0.00 12.19
Difference 1.93 -3.65 0.00 -0.78 0.22 0.00 -2.28
URom
Actual 3.13 5.08 14.47 3.00 1.90 0.00 27.58
Planned 3.14 5.16 14.49 2.33 1.91 0.00 27.03
Difference -0.01 -0.08 -0.02 0.67 -0.01 0.00 0.55
7.2.1 Explanations for resource deviation
Deviations from the planned resources are explained in the following section.
Fraunhofer Heinrich-Hertz Institute
In the second period, the Fraunhofer Heinrich-Hertz Institute was suffering an unexpected
personal shortage, one long-time member has left us and we were unable to quickly fill the
gap. This lead to major resource deviations across all work packages, especially in work
packages 2 and 4 with deviations of over 2 pm.
However, we did not change the planned goals and managed to complete all the planned
tasks. We are looking forward to come closer to the overall planned resource allocations for
the final period of the project.
CEA
In the first year, we overconsumed resources (+2.87) and we were able to achieve notable
results faster than expected. This overconsumption has been balanced in the second year (-
1.96), and it is mainly highlighted in the effort related to WP2. However, overall, we are in
line with the planned effort and we are successfully achieving the expected results.
Intel
In the second year of the project Intel slightly overspent, mostly in WP4. The overall
deviation according to the original budget plan remains very small and no further
explanations are required.
TI
The overall effort spent by TI in Year 2 has been lower than planned (2.63 pm lower) and this
is not compensated by the small extra effort spent during Yr1 (0.35 pm extra effort). Anyway,
this is not an issue because it is due to the allocation of the experimental activities involving
TI in WP2 and WP4 to Year 3. In the following more details on the specific WPs are
provided as well as the need for a minor shift of resources from WP2 to WP1.
WP1: also in Year 2 the actual effort spent was higher than planned. This because the effort
required to the discussions and to the contributions to D1.3 was higher than expected (also
because it was necessary to elaborate some architectural concepts developed within WP3, WP
where TI is not involved. Since also during Year 1 the actual effort was higher than the
planned (2 pm actual vs 0,86 pm planned) and taking into account that, still are due
contributions toward the final WP1 deliverable D1.4 it comes out that TI effort on WP1 has to
be increased from the original 2 pm to 4 pm. The resources can be taken from WP2 where the
actual effort appears lower than planned.
WP2: also in Year 2 the actual effort spent was lower than planned. This because the actual
measurement activity on the antenna prototypes, also due to external unexpected constraints
(i.e., custom issues) shifted to beginning of Year 3. In addition, the updated estimation of the
needed resources at this phase of the project leads to a likely total WP2 needs around 7 pm for
TI. The actual Yr1+ Yr2 effort is 4.91 pm so, about 2 pm can be moved toward WP1 where
(see above item) such resources are needed. The remaining 2 pm on WP2, will be used to
perform the measurements.
WP4: on the basis of the project evolution, despite of the original planning, the actual
involvement of TI on this WP turned out to be suitable on Year 3.
WP5: Yr1+ Yr2 actual effort is 1.5 pm. The remaining 0.5 pm appears adequate for the 3rd
Year dissemination activities of an industrial partner like TI.
URom
There was no deviation in the second period.
7.2.1 Unforeseen subcontracting
This does not apply.
7.2.2 Unforeseen use of in kind contribution from third party against payment or free
of charges (if applicable)
This does not apply.
References
[ADI12] B. Ahlgren, C. Dannewitz, C. Imbrenda, D. Kutscher, and B. Ohlman,
“A survey of information-centric networking,” IEEE Commun. Mag.,
vol. 50, no. 7, pp. 26–36, July 2012.
[AFK17] A. Abouaomar, A. Filali, and A. Kobbane, “Caching, device-to-device
and fog computing in 5th cellular networks generation: Survey,” in
Proc. WINCOM, Rabat, Morocco, 2017, pp. 1-6.
[AKS18] I. Ahmed, H. Khammari, A. Shahid, A. Musa, K. S. Kim, E. D.
Poorter, and I. Moerman, “A survey on hybrid beamforming
techniques in 5G: architecture and system model perspectives,” IEEE
Commun. Surveys Tuts., DOI: 10.1109/COMST.2018.2843719.
[BAI14] T. Bai, R. Vaze, and R. W. Heath, Jr., “Analysis of blockage effects on
urban cellular networks,” IEEE Trans. Wireless Commun., vol. 13, no.
9, pp. 5070-5083, Sep. 2014.
[BAR17] S. Barbarossa, E. Ceci, M. Merluzzi and E. Calvanese-Strinati,
"Enabling effective Mobile Edge Computing using millimeterwave
links," 2017 IEEE International Conference on Communications
Workshops (ICC Workshops), Paris, 2017, pp. 367-372.
[BBD14] E. Bastug, M. Bennis, and M. Debbah, “Living on the edge: The role of
proactive caching in 5G wireless networks,” IEEE Commun. Mag., vol.
52, no. 8, pp. 82-89, Aug. 2014.
[BBZ15] E. Bastug, M. Bennis, E. Zeydan, M. A. Kader, I. A. Karatepe, A. S.
E., and M. Debbah, “Big data meets telcos: A proactive caching
perspective,” J. Commun. Netw., vol. 17, no. 6, pp. 549–557, Dec.
2015.
[BCM17] S. Barbarossa, E. Ceci and M. Merluzzi, "Overbooking radio and
computation resources in mmW-mobile edge computing to reduce
vulnerability to channel intermittency," 2017 European Conference on
Networks and Communications (EuCNC), Oulu, 2017, pp. 1-5.
[BMSB2018] K. Koslowski et al, “SDN Orchestration to Optimize meshed
Millimeter-Wave Backhaul Networks for MEC enhanced eMBB use
cases”, BMSB 2018
[BSC18] S. Barbarossa, S. Sardellitti, E. Ceci, M. Merluzzi, “The edge cloud: a
holistic view of communication, computation and caching”, Chapter
16 of "Cooperative and Graph Signal Processing: Principles and
Applications", P. Djuric and C. Richard Eds., Academic Press,
Elsevier, 2018
[BSE18] S. Barbarossa, S. Sardellitti, and E. Ceci, “Learning from signal over
simplicial complexes”, Data Science Workshop (DSW 2018),
Lausanne, Switzerland
[CEB18] E. Ceci and S. Barbarossa, “Small perturbation analysis of network
topologies”, International Conference on Acoustic, Speech and Signal
Processing (ICASSP 2018), Calgary, Canada
[CGK12] N. Choi, K. Guan, D. C. Kilper, and G. Atkinson, “In-network caching
effect on optimal energy consumption in content-centric networking,”
in Proc. IEEE Int. Conf. Commun. (ICC 2012), June 2012, pp. 2889–
2894.
[D2.3] 5G-MiEdge public deliverable D2.3. Available online at 5g-miedge.eu
[D5.2] 5G-MiEdge public deliverable “Second report on dissemination,
standards, regulation and exploitation plan”, June 2019. Available
online at 5g-miedge.eu
[DAP] https://dataanalyticspost.com/edge-computing-cles-traitement-
donnees-de-smart-city/
[EBS17] M. S. Elbamby, M. Bennis, and W. Saad, “Proactive edge computing in
latency-constrained fog networks,” in Proc. EuCNC, Oulu, Finland,
2017, pp. 1-6.
[electronica] http://blog.electronica.de/en/2017/07/04/trend-tactile-internet-the-
magic-of-real-time/
[GAL62] D. Gale and L. S. Shapley, “College asmissions and the stability of
marriage”, The American Mathematical Monthly, Vol. 69, No. 1,
1962, Mathematical Association of America
[HMR17] M.Hajimirsadeghi, N.B.Mandayam, A.Reznik, “Joint caching and
pricing strategies for popular content in information centric networks,”
IEEE J. Sel. Areas Commun., vol. 35, no. 3, pp. 654–667, Mar. 2017.
[IEEETAP16] M. Zhang, J. Hirokawa and M. Ando, "A Four-Corner-Fed Double-
Layer Waveguide Slot Array With Low Sidelobes Developed for a 40
GHz-Band DDD System," in IEEE Transactions on Antennas and
Propagation, vol. 64, no. 5, pp. 2005-2010, May 2016.
[IW16] A. Ioannou and S. Weber, “A survey of caching policies and
forwarding mechanisms in information-centric networking,” IEEE
Commun. Surveys Tut., vol. 18, no. 4, pp. 2847-2886, Fourthquarter
2016.
[JST09] V. Jacobson, D. K. Smetters, J. D. Thornton, M. F. Plass, N. H.
Briggs, and R. L. Braynard, “Networking named content,” in Proc. 5th
Int. Conf. Emerg. Netw. Exper. Technol. ACM, 2009, pp. 1–12.
[KSK18] A. Khreishah, H. B. Salameh, I. Khalil, and A. Gharaibeh, “Renewable
energy-aware joint caching and routing for green communication
networks,” IEEE Syst. J., vol. 12, no. 1, pp. 768–777, Mar. 2018.
[LTV15] J.Llorca, A.M.Tulino, M.Varvello, J.Esteban, and D.Perino,“Energy
efficient dynamic content distribution,” IEEE J. Sel. Areas Commun.,
vol. 33, no. 12, pp. 2826–2836, Oct. 2015.
[MAO17] Y. Mao, J. Zhang, S. H. Song and K. B. Letaief, "Stochastic Joint Radio
and Computational Resource Management for Multi-User Mobile-Edge
Computing Systems," in IEEE Transactions on Wireless
Communications, vol. 16, no. 9, pp. 5994-6009, Sept. 2017.
[MB17] P. Mach, Z. Becvar, “Mobile Edge Computing: A Survey on
Architecture and Computation Offloading”, IEEE Communications
Surveys & Tutorials, vol. 19, Issue 3, pp. 1628-1656, 2017
[MEC003] ETSI GS MEC003, Mobile Edge Computing (MEC); Framework and
Reference Architecture, v1.1.1, Mar. 2016.
[MWB-D4.1] System Level Simulator Specification, MiWEBA Deliverables, 2014.
[MWB-D5.1] Channel Modeling and Characterization, MiWEBA Deliverables,
2014.
[NEE10] M. J. Neely, “Stochastic network optimization with application to
communication and queueing systems”, Synth. Lect. on Comm. Netw.s,
vol. 3, no. 1, pp. 1–211, 2010.
[OC15] J. Oueis and E. Calvanese Strinati, “Procédé et dispositifs associés de
formation d’un nuage informatique stockant le résultat de l’exécution
déportée d’une tâche informatique,” French Patent FR1562546A, Dec.
16, 2015.
[OC17] J. Oueis and E. Calvanese Strinati, “Computation caching for local
cloud computing,” presented at WCNC, 5G TACNET Workshop, San
Francisco, USA, 2017.
[Oue16] J. Oueis, “Joint communication and computation resources allocation
for cloud-empowered future wireless networks,” Ph.D. dissertation,
Univ. Grenoble Alpes, Grenoble, France, 2016.
[RM05] B. Robicˇ and J. Mihelicˇ, “Solving the k-center problem efficiently
with a dominating set algorithm,” J. Comput. Inform. Technol., vol. 13,
no. 3, pp. 225–234, 2005.
[SBS14] S. Sardellitti, S. Barbarossa and G. Scutari, "Distributed mobile cloud
computing: Joint optimization of radio and computational resources,"
2014 IEEE Globecom Workshops (GC Wkshps), Austin, TX, 2014, pp.
1505-1510
[SCM18] S. Sardellitti, F. Costanzo, M. Merluzzi, “Joint optimization of caching
and transport in proactive edge cloud”, 2018 EUSIPCO, Rome, Italy,
September 2018.
[SCU17] G. Scutari, F. Facchinei and L. Lampariello, "Parallel and Distributed
Methods for Constrained Nonconvex Optimization—Part I: Theory,"
in IEEE Transactions on Signal Processing, vol. 65, no. 8, pp. 1929-
1944, April15, 15 2017.
[SMB18] S. Sardellitti, M. Merluzzi, S. Barbarossa, “Optimal association of
mobile users to multi-access edge computing resources”, 2018
International Conference on Communications (ICC), Kansas City,
USA, 2018, pp. 1-6
[TR36.814] 3GPP TR 36.814, "Further advancements for E-UTRA physical layer
aspects".
[TS23.502] 3GPP TS23.502, Procedure for the 5G system (Release-15), Dec.
2017.
[WAN14] W. Wang and et al., “CRCache: Exploiting the correlation between
content popularity and network topology information for ICN
caching,” in Proc. IEEE Int. Conf. Commun. (ICC 2014), 2014, pp.
3191–3196.
[WZZ+17] S. Wang, X. Zhang, Y. Zhang, L. Wang, J. Yang, and W. Wang “A
survey on mobile edge networks: Convergence of computing, caching
and communications,” IEEE Access, vol. 5, pp. 6757-6779, 2017.
[ZBB16] E.Zeydan, E.Bastug ̆, M.Bennis, M.A.Kader, I.A.Karatepe,A.S.Er, and
M. Debbah, “Big data caching for networking: Moving from cloud to
edge,” IEEE Commun. Mag., vol. 54, no. 9, pp. 36–42, Sept. 2016.
[ZLZ15] M. Zhang, H. Luo, and H. Zhang, “A survey of caching mechanisms in
information-centric networking,” IEEE Commun. Surveys Tut., vol. 17,
no. 3, pp. 1473–1499, 2015.
Appendix: Dissemination activities during the project first year
Table A1 Journals/Book Chapters produced during the second year of the project. Publication Title Author(s) Status Related
WP
1 IEEE Trans. On Wir. Coms, 17 (5), 3154-
3169, 2018.
Coverage Analysis and Load Balancing in
HetNets with mmWave Multi-RAT
Small Cells
G. Ghatak, A. De Domenico, and M. Coupechoux
Accepted WP2
2
IEICE journal IEICE Trans. Electron.,
vol. E100-C, no. 10, pp. 790-808, Oct.
2017
Where, When, and How mmWave is Used
in 5G and Beyond
K. Sakaguchi, T. Haustein, S. Barbarossa, E. Calvanese Strinati,
A.Clemente, G. Destino, A. Pärssinen, I. Kim, H. Chung, J. Kim,
W. Keusgen, R. J. Weiler, K.Takinami, E. Ceci, A. Sadri, L. Xain, A. Maltsev, T. Gia Khanh, H. Ogawa,
K. Mahler, R. W. Heath Jr
Accepted All
3
Journal of IEICE, vol. 100, no. 8, pp. 825-830, Aug. 2017
How Do We Use Spectrum More
Efficiently
Kei Sakaguchi Accepted All
4
ICL Journal #173 Meshed 5G Millimeter-wave Backhaul Networks with
centralized SDN Orchestration
Konstantin Koslowski, Wilhelm Keusgen, Thomas Haustein
Accepted WP2
6
Book Chapter in "Cooperative and
Graph Signal Processing," Eds.,
Elsevier, 2018
The edge cloud: A holistic view of
communication, computation and
caching
S. Barbarossa, S. Sardellitti, E. Ceci and M. Merluzzi
Published WP2/WP3
Table A2 Conference papers produced during the second year of the project. Event Date Location Type of
contribution
Title Author(s) Status Related WP
1 IEICE Society Conference
09/2017
Rome, Italy
Conference
60 GHz Multi User Gigabit
Data Transfer System Based on WiGig/IEEE
802.11ad
Koji Takinami, Tomoya
Urushihara, Masashi
Kobayashi, Naganori Shirakata
Accepted WP2
2 SmartCom201
7 10/
2017 Rome, Italy
Poster
Cooperative WiGig/Wi-Fi Multi-User
Content Delivery
M. Kobayashi,
T. Urushihara,
N.
Accepted WP2
System with Application-
Centric Connection
Management
Shirakata, K. Takinami
3 SmartCom201
7 10/
2017 Rome, Italy
Poster
10 Gbps Ultra-High-Speed
Wireless Data Transfer
System with Adaptive Data Stream Control for 5G Mobile
Edge Cloud
T. Urushihara,
M. Kobayashi,
N. Shirakata,
K. Takinami
Accepted WP2
4 SmartCom201
7 10/
2017 Rome, Italy
Poster
Architecture Study on 5G
Networks with Edge
Computing and D2D for
Improving Communicatio
n Quality
Katsuo Yunoki, Koichiro
Kitagawa, Takashi
Fujimoto, Bingxuan
Zhao, Ryochi
Kataoka, Hiroyuki Shinbo
Accepted WP3
5
IEEE Globecom201
7 WS-5GTestbed
Dec. 2017
Tokyo, Japan
Conference
Turning the knobs on
OpenFlow-based
resiliency in mmWave small
cell meshed networks
Ricardo Santos, Hiroaki
Ogawa, Gia Khanh
Tran, Kei Sakaguchi,
Andreas Kassler
Presented
WP4
6 IEICE SR
Technical Committee
May. 2017
Honululu, Hawai,USA
technical report
5G-MiEdge -- Millimeter-wave Edge Cloud as an
Enabler for 5G Ecosystem --
K. Koslowski
et al.
Presented
All
7 IEEE IMS2017 Jun. 2017
Okinawa, Japan
Conference
MiEdge: Fusion of mmWave Access and
Mobile Edge Computing for
5G
K. Sakaguchi
Presented
All
8
Japan-China Workshop on
the Next Generation
Mobile Communication Technology
and Application
Jun. 2017.
Okinawa, Japan
Conference
Integration of mmWave
Access and Backhaul
Networks for 5G & Beyond
Gia Khanh Tran, Kei
Sakaguchi, Makoto
Ando
Presented
All
9 5G Taipei Summit
Sep. 2017
Taipei, Taiwan
Conference
The Challenge and
Proposition of mmWave in 5G
K. Sakaguchi
Presented
All
10
IEEE ATC2017 Oct. 2017
Quynhon, Vietnam
Conference
[Invited] mmWave
Heterogeneous Networks in 5G
& Beyond
Gia Khanh Tran, Kei
Sakaguchi, Makoto
Ando
Presented
All
11
IEICE RCS2017-185
Oct. 2017
Sendai ,Japan
technical report
Applications of mmWave
Heterogeneous Network in 5G
-- mmWave meshed network,
mmWave Edge Cloud,
mmWave V2V/V2X --
Kei Sakaguchi
Presented
WP1
12
IEICE Gen. Conf.'2018
Mar. 2018
Tokyo, Japan
Conference
Proposal of Prefetching
Algorithm for 5G mmWave HetNet under
Backhaul Constraint
Hiroaki Nishiuchi, Gia Khanh Tran, Kei
Sakaguchi
Presented
WP3, WP4
13
IEICE Gen. Conf.'2018
Mar. 2018
Tokyo, Japan
Conference
Study on Interference Management
for Millimeter-Wave Mesh
Backhaul Networks
Makoto Nakamura, Gia Khanh Tran, Kei
Sakaguchi
Presented
WP2, WP4
14
IEICE SmartCom
2017
Oct. 2017
Rome, Italy
technical report
[Invited] 5G-MiEdge:
Introduction of Millimeter-Wave Edge
Cloud
Koji Takinami,
Antonio De Domenico, Konstantin Koslowski, Gia Khanh
Tran, Yuyuan Chang, Hiroaki Ogawa, Mattia
Merluzzi, Sergio
Barbarossa
Presented
WP2
15
IEICE SmartCom
2017
Oct. 2017
Rome, Italy
technical report
[Invited] Use cases and scenario
resulting from merging
mmWave access with
Valerio Frascolla,
Kei Sakaguchi,
Sergio Barbarossa, Antonio De
Presented
WP1
Multi-access Edge
Technologies
Domenico, Sergio
Barberis, Gia Khanh Tran, Koji Takinami, Thomas Haustein
16
IEICE SmartCom
2017
Oct. 2017
Rome, Italy
Poster
Performance Evaluation of Prefetching
Algorithm for 5G mmWave Edge Cloud
Hiroaki Nishiuchi, Gia Khanh Tran, Kei
Sakaguchi
Presented
WP3, WP4
17
IEEE BMSB 2018
Feb. 2018
Valencia, Spain
Conference
SDN Orchestration to Optimize
Meshed Millimeter-
Wave Backhaul Networks for
MEC-enhanced eMBB Use
Cases
Konstantin Koslowski,
Ricardo Santos, Kei Sakaguchi,
Thomas Haustein, Wilhelm Keusgen, Andreas Kassler, Hiroaki Ogawa, Makoto
Nakamura and Yu Tao
Presented
WP2/WP4
18
ICC Workshops (WDN-5G)
2018
May 2018
Kansas City, MO,
USA Conference
Optimal Association of Mobile Users
to Multi-access Edge
Computing Resources
Stefania Sardellitti,
Mattia Merluzzi,
and Sergio Barbarossa
Presented
WP2/WP3
19
ICC Workshops (WDN-5G)
2018
May 2018
Kansas City, MO,
USA Conference
Architecture of mmWave edge
cloud in 5G-MiEdge
Gia Khanh Tran,
Hiroaki Nishiuchi,
Valerio Frascolla,
Koji Takinami,
Antonio De Domenico,
Emilio Calvanese Strinati, Thomas
Haustein, Kei
Sakaguchi, Sergio
Barbarossa, Sergio
Presented
All
Barberis, Katsuo Yunoki
20
IEEE_DSW2018
June 2018
Lausanne, Switzerlan
d Conference
Learning from Signals Defined over Simplicial
Complexes
Sergio Barbarossa,
Stefania Sardellitti, and Elena
Ceci
Presented
WP2/WP3
21
IEEE ICASSP 2018
15-20 April 2018
Calgary, Alberta, Canada
Conference
Small Perturbation Analysis of Network
Topologies
Elena Ceci and Sergio Barbarossa
Presented
WP2/WP3
22
IEEE WCNC 2018 WS
April 2018
Barcelona Spain
Conference
Millimeter-waves, MEC, and network
softwarization as enablers of
new 5G business
opportunities
Valerio Frascolla et
al. Accepted WP1
23
EUSIPCO 2018 Sept. 2018
Rome, Italy
Conference
Proactive computation
caching policies for 5G-and-
beyond mobile edge cloud networks
N. di Pietro and E.
Calvanese Strinati
Accepted WP3
24
IEEE VTC fall 2018
Aug. 2018
Chicago, USA
Conference
Analytical Characterization of Cell Load in mm-wave
Gourab Ghatak,
Antonio De Domenico,
and Marceau
Coupechoux
Accepted WP2
25
IEEE VTC spring 2018
Jun. 2018
Porto, Portugal
Conference
Performance Evaluation of 5G mmWave Edge Cloud
with Prefetching Algorithm
H. Nishiuch, G.K. Tran,
K. Sakaguchi
Presented
WP3/WP4
26
IEEE VTC fall 2018
Aug. 2018
Chicago, USA
Conference
Interference Management
for Millimeter-wave Mesh
Backhaul Networks
M. Nakamura, G.K. Tran,
K. Sakaguchi
submitted
WP2
27
EUSIPCO 2018 Sept. 2018
Rome, Italy
Conference
Joint optimization of
caching and transport in
proactive edge cloud
S. Sardellitti,
F. Costanzo,
and M. Merluzzi
Accepted WP3
Table A3 Presentations given during the second year of the project.
Event Date Location Type of contribution
Title Author(s)
1 The IEICE Communications Society Conf. ,BP-4-1
9/2017 Tokyo, Japan Panel discussion
Millimiter-wave Application in 5G and Beyond
K. Sakaguchi
2 IEICE Mobile Communication Workshop
2/2018 Kanagawa,Japan Keynote mmWave V2V2X for Automated Driving
Kei Sakaguchi
3 SmartCom2017 10/ 2017
Rome Presentation
5G-MiEdge: Introduction of Millimeter-Wave Edge Cloud, a Key Technology for 5G Phase II Deployment
Koji Takinami, Antonio De Domenico, Konstantin Koslowski, Khanh Tran Gia, Yuyuan Chang, Hiroaki Ogawa, Mattia Merluzzi, Sergio Barbarossa
4 SmartCom2017 11/ 2017
Rome Presentation
Research Challenges of Computation Resource Integration into Mobile Network
Katsuo Yunoki, Sergio Barbarossa, E. Calvanese Strinati, Valerio Frascolla
5
SPEED-5G WS on Advanced spectrum management in 5G+ networks
3/2018 London England Presentation
5G-MiEdge Bridging the gap between 5G and B5G
Valerio Frascolla
6 ICC Workshops (WDN-5G) 2018
5/2018 Kansas City, MO, USA
Keynote
Cooperative Perception realized by Millimeter-wave V2V2X
Kei Sakaguchi
7 IEICE General Conference 2018
3/2018 Tokyo, Japan Presentation
Control method for resource utilization of edge computing in 5G
Katsuo Yunoki, Hiroyuki Shinbo
8
EuCNC 2018 WS: Optical and Wireless Network Convergence: An Enabler for 5G
6/2018 Ljubljana, Slovenia
Panel Partecipation
5G use cases for Olympic games The visions of the projects 5GCHAMPION and 5G-MiEdge
Valerio Frascola
9
EuCNC 2018 WS: 2nd Workshop on business models and techno-economic analysis for 5G networks
6/2018 Ljubljana, Slovenia
Presentation
Limitations and issues of 5G deployments from the standards and regulatory point of view
Valerio Frascola
10
EuCNC 2018 WS: 2nd Workshop on business models and techno-economic analysis for 5G networks
6/2018 Ljubljana, Slovenia
Presentation
Techno-economic aspects of mmWaves and MEC, two key enablers for 5G networks
Valerio Frascola
11
EuCNC 2018 WS: 2nd Workshop on business models and techno-economic analysis for 5G networks
6/2018 Ljubljana, Slovenia
Panel Partecipation
Closing panel of the WS
Valerio Frascola