Professional Documents
Culture Documents
-1-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
Table of Contents
1.
2.
3.
4.
5.
Appendix ........................................................................................................................................... 8
-2-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
1.
In line with its vision and philosophy for the development of the future 5G wireless network, HARP
promoted and organized a workshop titled Wireless optical network convergence in support of cloud
architectures, shortly
WONC, at IEEE Globecom 2014, a flagship conference of the IEEE
communications society.
As from the title, the workshop assigns a prominent role to the concept of radio access networks based
on cloud computing to enable disruptive technological and architectural changes required by the
challenging targeted Key Performance Indicators (KPI) of 5G wireless networks. For the technical and
economic success of this concept, the integration and convergence of radio access and optical networks
play a strategic role and the workshop focuses on this specific aspect of cloud radio access networks. In
the workshop vision, the cloud architecture is based on massively distributed antennas or remote radio
heads (RRHs) connected to a computing cloud through an optical fiber infrastructure that replaces the
traditional backhaul network.
It is worth noting that the original declared aim of HARP was to bring distributed multi-antenna wireless
access to reality by combining two powerful technologies: 1) radio remote heads (RRHs), which allow for
widely geographically distributed access via radio-over-fiber connections to a central processing unit;
and 2) electronically steerable passive array radiators, which provide multi-antenna-like functionality
with a single active RF chain only, and thus developing the RRH optical aggregation network, required
for efficient coordination between disparate access areas. Then, the core of HARP is completely aligned
to the scope of the workshop and HARP contributed to WONC by proposing solutions for high capacity,
low-cost, and environmentally friendly multi-antenna RRHs supported by a flexible optical aggregation
network.
The relevance of the topic proposed in WONC is corroborated by the fact that several European projects
are working on this subject from different and complementary perspectives as well as major players in
the worldwide communications arena. Then, primary objective of the workshop was to bring together
academic and industrial actors in the field to draw the current state of art, identify the technical
challenges, exchange results and strategic visions, foster discussions to identify the key driving concepts
and technologies, and, eventually, develop synergies. This objective was achieved by inviting paper
authors and keynote speakers from relevant flagship EU funded research projects such as TROPIC,
CONTENT, SODALES, iJOIN, and HARP and representatives of leading industries as Huawei (China) and
prominent American scientists very active in the field.
The workshop was structured into two sessions and covered a wide range of key issues on the wireless
optical network convergence in support of cloud architectures. In particular, they span from emerging
technologies for remote radio heads with low capital and operational costs to signal processing
techniques for distributed antenna systems taking into account the non-ideality of the front-haul and
the back-haul, to allocation and scheduling of radio, optical and computational resources. In addition,
-3-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
emphasis was given in the architectural challenges associated with the converged optical/wireless
networks to support cloud and mobile cloud services.
The rest of the deliverable is structured as follows. In Section 2 we provide general information about
the workshop. Section 3 provides information about the structure of the workshop, keynote speakers
and the invited paper authors that contributed to the workshop. Section 4 gives an overview of the
technical contributions. Section 5 describes the actions taken to promote the workshop. Finally, Section
6 contains the full contribution to WONC by the invited authors.
2.
3.
The workshop was structured into two sessions. The first session titled Radio Access Technologies for
Wireless Optical Networks (RAT) collected all the contributions related to the physical layer, from
hardware design to algorithms for transmission and resource allocation. The second session titled
Wireless Optical Networks for Cloud Architectures (WON) was focused on architectural issues raised
by the cloud architectures. Each of the sessions was open by a keynote speech and followed by the
presentation of invited papers.
-4-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
The RAT session was open by the keynote speech of Prof. Ralf Mueller, HARP member from University
Erlangen-Nrnberg, Germany. Prof. Mueller presented an innovative technologies for multiantenna
arrays with single active RF chain developed in the framework of HARP in a speech titled "Load
Modulated MIMO: A New Hardware Concept to Reduce Cost, Size, and Amplifier Back-off".
Followed the works of the invited authors:
1) Prof. Wei Yu from University of Toronto, Canada and holding a Canada Research Chair in
Information Theory and Digital Communications (directly presented).
2) Prof. Gerhard Fettweis, University of Dresden, Germany, holding a Vodafone chair and active
member of the European STREP project iJOIN (presented by the co-author Jens Bartelt).
3) Prof. Sergio Barbarossa from University of La Sapienza, Roma, Italy and active member of the
European STREP project TROPIC (due to health problems, presented by his colleague Adrian
Agustin, from UPC, Barcelona also member of TROPIC).
The WON session was open by an introductory keynote speech prepared by the invited speaker Dr.
Ganghua Yang, vice-director of the communication technology lab at Huawei, Shanghai and due to VISA
problem, the speech was given by his colleague Dr. Jian Li from the same lab. The keynote speech titled
Wireless Optical Network Convergence in 5G Access Network provides a broad vision at the cutting
edge of current technological developments from Huawei perspective.
The keynote speech was followed by the presentation of the works of the invited authors:
1) Prof. Lars Dittmann, Technical University of Denmark, Denmark, member of the European STREP
Project HARP (directly presented).
2) Dr. Markos Anastasopoulos University of Bristol, UK, member of the EU Project CONTENT
(directly presented).
3) Jordi Ferrer Riera from Fundaci i2CAT, Internet i Innovaci Digital a Catalunya, Spain, active
member of the EU Project SODALES (directly presented).
4) Dr. Jean Marc Kelif, Orange Labs, France and member of the European Project HARP.
4.
The keynote speech opening the RAT session and titled "Load Modulated MIMO: A New Hardware
Concept to Reduce Cost, Size, and Amplifier Back-off" presented recent results of the HARP project on
the architecture of massive antenna arrays. The proposed architecture enables multi antenna array with
single power amplifier chain and allows significant front-end hardware savings resulting in an appealing
enabler for future small and powerful cell nodes able to carry massive MIMO technology. The talk
reviewed and compared radio frequency (RF) hardware architectures with a single power amplifier and
showed that they are superior to conventional MIMO implementations in terms of cost, dissipated heat,
and physical size. This progress on the RF-side have relevant impact also on the future network
architectures since it allows to merge the two competing cellular concepts of virtual and massive MIMO
into a hybrid approach of remote radio heads with massive MIMO arrays.
-5-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
The keynote speech was followed by the presentation of two works that focus on the non-ideality of the
backhaul and the front-haul.
The article titled Optimized Beamforming and Backhaul Compression for Uplink MIMO Cloud Radio
Access Networks studied the optimization of transmit beamforming and backhaul compression
strategies for the uplink of cloud radio access networks (C-RAN), in which multi-antenna users
communicate with a cloud-computing based central processor (CP) through multi-antenna base-stations
(BSs). The BSs perform compress-and-forward strategy to quantize the received signals and send the
quantization bits to the CP via capacity-limited backhaul links for decoding. In contrast to the previous
work on the uplink C-RAN, which mostly focuses on the backhaul compression strategies only, in this
work the authors propose the joint optimization of the transmit beamformers and the quantization
noise covariance matrices at the BSs for maximizing the benefit brought by C-RAN. A novel weighted
minimum-mean-square-error successive convex approximation (WMMSE-SCA) algorithm was developed
for finding a local optimum solution to the problem along with suboptimal low-complexity
approximations. Simulations show that with optimized beamforming and backhaul compression, C-RAN
can significantly improve the overall performance of wireless cellular networks.
The impact of imperfect optical links in the front-haul is studied in the project iJoin and presented in the
work titled An Improved Decoder for Cloud-Based Mobile Networks under Imperfect Fronthaul. The
fronthaul of a cloud-RAN network forwards digitized samples to the central processing unit. However,
with increasingly dense deployments of small cells, the availability of high quality fiber fronthaul cannot
be expected at all sites. Longer fronthaul distances or the employment of wireless technologies renders
fronthaul links imperfect in the sense that they introduce bit-errors into the digital samples. The authors
build upon so-called hardware noise resilient decoders to investigate decoders for imperfect fronthaul
that can achieve very good performance even under high bit-error rates on the fronthaul.
The session was concluded by the presentation of results of the European project Tropic. The work titled
Distributed Mobile Cloud Computing: Joint Optimization of Radio and Computational Resources
focused on the challenging problem of jointly optimize the radio interface by beamforming design and
computational resource allocation to minimize the overall power consumption under latency constraints.
More specifically, the authors consider a scenario composed by multiple mobile users asking for
computation offloading of their applications to a set of cloud servers. They aim to find the optimal
assignment of each mobile user to a cloud server through the most convenient base station and,
contextually, the optimal MIMO precoding matrices and computational rates (virtual machines) to each
user, under latency constraints dictated by the users' Quality of Experience (QoE). The radio resources
assigned to users belonging to the same cell are orthogonal to each other, but users belonging to
different cells might interfere with each other. The latency constraint imposes a strict relationship
between the time spent for transferring the program execution from the mobile device to the fixed
server (and viceversa) and the time needed to execute the computation. The authors formulate the
computation offloading problem as a joint optimization of the radio and computational resources, with
the objective of minimizing the overall energy consumption, at the mobile terminal side, while meeting
the latency constraints. Numerical results show that the proposed joint optimization yields significant
-6-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
energy savings with respect to more traditional schemes performing a separate optimization of the
radio and computational resources.
The WON session was focused on the network architecture aspects and started with a keynote speech
titled "Wireless Optical Network Convergence in 5G Access Network". Huawei presented its vision on
5G wireless systems in term of service requirements and network architecture. Some scenarios
for wireless optical network convergence at the access network and at base-stations were analyzed.
Finally, hot research topics on wireless optical network convergence were suggested and discussed.
The following work reflected contribution in the field of the European project Conect and was titled
Dependable Multi-Tenant Infrastructures Supporting Cloud and Mobile Cloud Services. The authors
focus on the design of dependable Mobile Optical Virtual Networks (MOVNOs) to enable multi-tenancy
over converged wireless, optical and computational infrastructures supporting cloud and mobile cloud
services. Dependability is defined as the MOVNO's ability to offer resilience in case of network or
computing resource failures and security guarantees in terms of isolation between MOVNOs that share
the common physical infrastructure. The VI planning optimization objective is to minimize the overall
energy consumption satisfying at the same time specific end-to-end delay constraints. To address these
issues, together with the uncertainty introduced by the mobility of the end devices the authors
proposed a MOVNO planning scheme based on non-linear programming. This modeling results in
interesting trade-offs between the overall energy consumption of the MOVNOs, the mobility of the enddevices, as well as the dependability impact on the energy consumption and the utilization of the
infrastructure resources.
The subsequent presentation reflected the HARP vision on the network architecture with special focus
on the design of the fronthaul and the flow aggregation at that level. The work was titled Meeting
fronthaul challenges of future mobile network deployments - the HARP approach. This contribution
addresses the need to design new architecture to be able to perform a flexible network aggregation at
different levels in future mobile networks. The authors presented the design approach for a C-RAN
based mobile aggregation network used in the EU HARP project. With the proposed architecture fronthaul aggregation is performed which might be an option for future generation of mobile networks.
The SODALES (SOftware-Defined Access using Low-Energy Subsystems) activity on the field and its vision
on wireless optical networks supporting cloud architecture is presented in the work titled Softwaredefined wired-wireless access network convergence: the SODALES approach. In the SODALES vision an
active remote node (ARN) at an intermediate location between the central office and end-user premises
is introduced to solve the issues associated to bandwidth and wired-wireless convergence. The ARN
represents the key architectural design innovation of the SODALES network. The purely passive
approaches (e.g. as exemplified by passive optical networking (PON) access architectures) are shown to
be too restrictive for future 5G networks when compared the modularity and flexibility offered by an
active remote node. A performance analysis of the ARN node demonstrates the potential benefits in
terms network performance, operational efficiency, and flexible functionality of the proposed
architecture. Future ARN capabilities can also be expected to include hierarchical caching, customer
-7-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
premises equipment (CPE) virtualization, and nearer to the end-user location of software-defined
platforms supporting ubiquitous cloud services.
The WON session is concluded by an HARP contribution to facilitate simulations at a network level and is
titled Analytical Performance Model for Poisson Wireless Networks with Pathloss and Shadowing
Propagation. The authors propose a fluid network model to easily compute the SINR (signal to
interference plus noise ratio) for wireless networks analysis in scenarios including pathloss and
shadowing.
5.
The workshop was promoted by regularly announcing it in worldwide emailing lists of the
telecommunication society such as TCCC-announce (tccc-announce@COMSOC.ORG) and TCCN
(tccn@COMSOC.ORG). This allowed us to reach a very wide audience. A more targeted advertising was
achieved by announcing the workshop in European project mailing lists such as HARP CONTENT,
SODALES, iJOIN focused on the specific topic of our workshop.
Additionally, the workshop flier was shown on the EURECOM website and the websites of the above
mentioned European projects.
6.
Appendix
1) Presentation slides of the keynote speech: Load Modulated MIMO: A New Hardware
Concept to Reduce Cost, Size, and
Amplifier Back-off, Ralf Mller (University
Erlangen-Nrnberg, Germany).
2) Invited paper: Optimized Beamforming and Backhaul Compression for Uplink MIMO
Cloud Radio Access Networks, Yuhan Zhou and Wei Yu (University of Toronto, Canada).
3) Invited paper: An Improved Decoder for Cloud-Based Mobile Networks under Imperfect
Fronthaul, Jens Bartelt (Dresden University of Technology, Germany); Gerhard Fettweis
(Technische Universitt Dresden, Germany).
4) Invited paper: Distributed Mobile Cloud Computing: Joint Optimization of Radio and
Computational Resources, Sergio Barbarossa and Stefania Sardellitti (University of Rome
La Sapienza, Italy); Gesualdo Scutari (State University of New York at Buffalo, USA).
5) Invited paper: Dependable Multi-Tenant Infrastructures Supporting Cloud and Mobile
Cloud Services, Markos Anastasopoulos, Anna Tzanakaki and Dimitra Simeonidou
(University of Bristol, United Kingdom).
-8-
HARP deliverable: D2.1.1 Proceedings of first workshop/ special session at an international conference
6) Invited paper: Meeting fronthaul challenges of future mobile network deployments - the
HARP approach, Lars Dittmann, Henrik L. Christiansen (Technical University of Denmark,
Denmark), and Aleksandra Checko (MTI Radiocomp).
7) Invited paper: Software-defined wired-wireless access network convergence: the
SODALES approach, Jordi Ferrer Riera, Carlos Bock and Eduard Escalona (Fundaci i2CAT,
Internet i Innovaci Digital a Catalunya, Spain); Volker Jungnickel (Fraunhofer Heinrich Hertz
Institute, Germany); Kai Habel (Fraunhofer HHI, Germany); Michael Parker, Stuart D Walker
and Terry Quinlan (University of Essex, United Kingdom); Victor Marques (Portugal Telecom
Inovacao, Portugal); David Levy (Ethernity Networks, Israel).
8) Invited paper: Analytical Performance Model for Poisson Wireless Networks with Pathloss
and Shadowing Propagation, Jean-Marc Kelif and Stephane Senecal (Orange Labs, France);
Marceau Coupechoux (Telecom ParisTech, France); Constant Bridon (ENS Cachan, France).
-9-
FAU, Germany
NTNU, Norway
Introduction
1 / 18
12-Dec-2014
2 / 18
Increase user density by increasing the number of antennas at the same site.
Increase user density by partitioning of cells into subcells, i.e. increasing the
number of base stations.
12-Dec-2014
12-Dec-2014
Georg Fischer
Mohammad A. Sedaghat
Ralf R. Mller
Introduction
12-Dec-2014
2 / 18
12-Dec-2014
2 / 18
Increase user density by increasing the number of antennas at the same site.
Increase user density by partitioning of cells into subcells, i.e. increasing the
number of base stations.
More base station sites required.
Increase user density by increasing the number of antennas at the same site.
Larger distance to the users.
Increase user density by partitioning of cells into subcells, i.e. increasing the
number of base stations.
Introduction
Introduction
Massive MIMO:
Spread spectrum:
Massive MIMO
Spread spectrum:
Massive MIMO
Introduction
12-Dec-2014
12-Dec-2014
3 / 18
3 / 18
Introduction
Introduction
12-Dec-2014
Is massive MIMO
compatible with
RRHs?
12-Dec-2014
4 / 18
4 / 18
Introduction
12-Dec-2014
Size issue:
12-Dec-2014
Cost issue:
Open Issues
Size issue:
Cost issue:
Open Issues
Introduction
5 / 18
5 / 18
Promises
Promises:
Size issue:
12-Dec-2014
12-Dec-2014
Standard approaches require one linear amplifier and mixer per transmit
antenna.
RF-costs limit the number of antenna elements and thus the overall system
performance.
Cost issue:
Open Issues
Introduction
6 / 18
5 / 18
Promises
12-Dec-2014
6 / 18
12-Dec-2014
6 / 18
MIMO transmitters can be implemented with a single amplifier that serves all
antenna elements.
Promises:
MIMO transmitters can be implemented with a single amplifier that serves all
antenna elements.
Promises:
Promises
Promises
12-Dec-2014
6 / 18
12-Dec-2014
The amplifier efficiency is the higher the more antenna elements are used.
6 / 18
MIMO transmitters can be implemented with a single amplifier that serves all
antenna elements.
Promises:
MIMO transmitters can be implemented with a single amplifier that serves all
antenna elements.
Promises:
Promises
General Concept
9W
=W
!"#$""%
LW
12-Dec-2014
12-Dec-2014
7 / 18
6 / 18
MIMO transmitters can be implemented with a single amplifier that serves all
antenna elements.
Promises:
Promises
General Concept
!"#$""%
LW
V (t) s(t)
Z (t) = const.
Voltage modulation:
9W
=W
!"#$""%
LW
9W
=W
General Concept
V (t)
Z (t) + R
V (t)
Z (t) + R
12-Dec-2014
i(t) =
The current
12-Dec-2014
i(t) =
The current
7 / 18
7 / 18
V (t)
Z (t) + R
i(t) =
The current
General Concept
V(t)
2
V(t)
1
iN(t)
i2(t)
i1(t)
Antenna
Antenna
Antenna
FWBMVBUJPO
R
V(t)
1
Voltage modulation
12-Dec-2014
12-Dec-2014
8 / 18
7 / 18
V (t) = const.
1
s(t)
Z (t) + R
Load modulation:
!"#$""%
LW
Z (t) = const.
V (t) s(t)
Voltage modulation:
9W
=W
General Concept
V(t)
2
V(t)
1
iN(t)
i2(t)
i1(t)
Antenna
Antenna
Antenna
Implementation
FWBMVBUJPO
V
Z1(t)
Z2(t)
Z1(t)
R
Antenna
Antenna
i2(t)
i1(t)
Load modulation
FWBMVBUJPO
8 / 18
=(t)
Antenna
12-Dec-2014
9 / 18
The load modulators shall use only reactive components, e.g. banks of PIN-diodes
acting as variable capacitors, to avoid power dissipation into heat.
12-Dec-2014
V(t)
1
Voltage modulation
General Concept
Antenna
L1
D1
D3
L2
L3
D3
Load
Modulator
io1
"
&
'
Implementation
The data signal does not pass through the power amplifier (PA).
Block Structure
12-Dec-2014
12-Dec-2014
Pulse shaping after the load modulator by surface acoustic wave (SAW) filters.
=(t)
10 / 18
9 / 18
The load modulators shall use only reactive components, e.g. banks of PIN-diodes
acting as variable capacitors, to avoid power dissipation into heat.
Implementation
w
"
%
&
'
(
1
2
6
.
Implementation
"
&
'
" "
!"
&
'
!"
!"
!"
8029/-992.
!"$
!"
!"%0+:;<<;9=0<,3>2>;:;/?
12-Dec-2014
Let the current per antenna be complex Gaussian distributed. For N transmit
antennas, the transmitted power is 2 distributed with 2N degrees of freedom.
!"
12-Dec-2014
The data signal does not pass through the power amplifier (PA).
The PA provides kinds of RF-power sockets where the antennas are plugged.
Implementation
+,-./012+/3,04567
10 / 18
10 / 18
Cost Reduction
12-Dec-2014
11 / 18
12-Dec-2014
11 / 18
In 3G and LTE, the installed peak power is much higher than the average radiated
power.
Cost Reduction
In 3G and LTE, the installed peak power is much higher than the average radiated
power.
Cost Reduction
Cost Reduction
20
Average power
40
60
Output power (%)
0
0
10
20
30
40
50
60
70
80
40
60
Output power (%)
Crest factor=8dB
80
100
20
Average power
100
Cost Reduction
Crest factor=8dB
80
Cost Reduction
Amplifier Efficiency
0
0
10
20
30
40
50
60
70
80
Amplifier Efficiency
12 / 18
12-Dec-2014
F = 80%.
12 / 18
Pmean
A/B =
4 Pmax
12-Dec-2014
Pmean
A/B =
4 Pmax
Size Reduction
Multiple-PAs
100
Class A/B
8 dB
31%
1.9 W
4.2 W
6.1 W
12.0 W
Single-PA
100
Class F
1.2 dB
61%
1.9 W
1.2 W
3.1 W
2.5 W
12-Dec-2014
12-Dec-2014
Array Densification
Transmitter
# antennas
Amplifier
Crest
Efficiency
Radiated power
Dissipated power
Consumed power
Peak power
Example
Cost Reduction
14 / 18
13 / 18
Size Reduction
12-Dec-2014
12-Dec-2014
Array Densification
Array Densification
Size Reduction
14 / 18
14 / 18
Size Reduction
12-Dec-2014
14 / 18
12-Dec-2014
14 / 18
These two theoretical limits are neither related to frequency nor wavelength.
The maximum antenna gain is limited by the physical size of the antenna.
The maximum directivity does not depend on the physical size of the antenna.
Array Densification
Array Densification
Size Reduction
Size Reduction
12-Dec-2014
14 / 18
YES!
(in particular, for voltage modulation)
12-Dec-2014
14 / 18
These two theoretical limits are neither related to frequency nor wavelength.
The maximum antenna gain is limited by the physical size of the antenna.
The maximum directivity does not depend on the physical size of the antenna.
Array Densification
The maximum antenna gain is limited by the physical size of the antenna.
These two theoretical limits are neither related to frequency nor wavelength.
The maximum directivity does not depend on the physical size of the antenna.
Array Densification
Size Reduction
=W
=W
1
9W
9W
1
LW
1
LW
LW
antennas
coupled
FWBMVBUJPO
Size Reduction
=W
=W
1
9W
9W
1
LW
1
LW
antennas
coupled
FWBMVBUJPO
=W
9W
LW
LW
12-Dec-2014
=W
1
=W
=W
LW
1
LW
antennas
coupled
15 / 18
12-Dec-2014
15 / 18
9W
=W
9W
Size Reduction
=W
=W
1
9W
9W
1
LW
1
LW
antennas
coupled
=W
1
=W
=W
LW
LW
1
LW
antennas
coupled
9W
FWBMVBUJPO
Conclusions
15 / 18
12-Dec-2014
16 / 18
12-Dec-2014
Load modulation:
Voltage modulation:
Conclusions
Load modulated antennas can be spaced more closely without much effort.
=W
9W
LW
Size Reduction
+ Low backoff
+ Class F amplifier sufficient
Nonlinear relation between signals
and loads
Load modulation:
Related Work
12-Dec-2014
Parasitic element
"
Active element
Load Modulated MIMO
Varactor
12-Dec-2014
Goal: Try to reduce the number of RF-chains and the size of the array:
ESPAR MIMO
17 / 18
16 / 18
Load modulation moves the border between digital and analog circuitry one step
further towards the antenna.
Load modulation is well-suited for massive MIMO and allows for dense arrays.
Voltage modulation:
Conclusions
Conclusions
=W
1
=W
=W
LW
LW
LW
1
LW
Related Work
FWBMVBUJPO
antennas
coupled
12-Dec-2014
=W
1
=W
=W
FWBMVBUJPO
9W
LW
1
LW
certain
antennas
coupled
17 / 18
12-Dec-2014
17 / 18
Goal: Try to reduce the number of RF-chains and the size of the array:
ESPAR MIMO
FWBMVBUJPO
9W
Goal: Try to reduce the number of RF-chains and the size of the array:
ESPAR MIMO
Related Work
=W
1
=W
=W
LW
LW
1
LW
certain
antennas
coupled
Literature
FWBMVBUJPO
12-Dec-2014
12-Dec-2014
R. C. Hansen.
Electrically Small, Superdirective, and Superconducting Antennas.
John Wiley & Sons, 2006.
18 / 18
17 / 18
9W
Goal: Try to reduce the number of RF-chains and the size of the array:
ESPAR MIMO
Related Work
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
I. I NTRODUCTION
To meet the exponential increasing data demand in wireless communication driven by smartphones, tablets, and video
streaming, modern cellular systems are moving towards heterogenous networks consisting of base-stations (BSs) deployed
with high density, covering progressively smaller areas. As a
consequence, inter-cell interference management becomes a
challenging problem. Cloud radio access network (C-RAN)
is a promising network architecture for dealing with this
interference problem. In a C-RAN architecture, the encoding/decoding functionalities of the BSs are migrated to a cloudcomputing based central processor (CP). By taking advantage
of high backhaul links between the CP and the BSs, C-RAN
allows coordination and joint signal processing across multiple
cells, and the possibility for inter-cell interference cancellation,
which has the potential to significantly improve the overall
throughput of the wireless cellular network [1].
978-1-4799-7470-2/14/$31.00 2014 IEEE
One of the main problems arising in the practical implementation of C-RAN is how to optimally utilize the backhaul when
the backhaul link capacities are finite. Substantial research
works have made progress towards this direction, either from
information theoretic point of view [2][4] or algorithm design
point of view [5], [6]. In [7], a virtual multiple access channel
(VMAC) scheme, which is a two-stage compress-and-forward
strategy based on successive decoding, is proposed for the
single-input single-output (SISO) C-RAN architecture. It is
shown that the VMAC scheme can achieve the sum capacity of
C-RAN to within a constant gap under a sum backhaul capacity
constraint.
This paper studies the linear transceiver and backhaul compression design in the VMAC scheme for the uplink multipleinput-multiple-output (MIMO) C-RAN model as shown in
Fig. 1. As a generalization of [7] which considers the SISO
case only, this paper considers the MIMO case where both the
users and the BSs are equipped with multiple antennas. The
main difference between the SISO case and the MIMO case is
the impact of transmit optimization at the user terminals. In the
SISO case, since most of the intra-cluster interference has been
nulled by multicell decoding, it is near optimal for the users
to transmit with their maximum powers. In the MIMO case,
the users are capable of doing transmit beamforming, so the
optimal transmit beamforming design is more involved. This is
especially so when the backhaul links in C-RAN have limited
capacities.
The backhaul compression problem for the uplink C-RAN
model has been considered extensively in the literature. Various
algorithms such as the gradient projection method [5], the
robust backhaul compression approach [6], and the alternating
convex optimization approach [7] are developed for maximizing the (weighted) sum rate under the bachaul constraints.
All of these algorithms focus only on the optimization of
quantization noise covariance matrices across the BSs, while
keeping the transmit beamformers fixed. This paper goes one
step further by considering the joint transmit beamformer and
quantization noise covariance matrix optimization problem.
Accounting for both the transmit beamforming and the quantization design problem together in the optimization framework
is nontrivial because the two are coupled through the backhaul
constraints. To tackle this problem, this paper proposes a
weighted minimum-mean-square-error successive convex approximation (WMMSE-SCA) algorithm for efficiently finding
1472
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
..
x1
z1
H11
H21
z2
..
x2
1
y1 : y
H12
H22
C1
2 C2
y2 : y
1
y
2
y
H1G
H2G
HLG
Hik Vk sk + zi
for
i = 1, 2, . . . , L,
L
yL : y
k=1
Central
Processor
L
y
HL1
HL2
..
xG
yi =
CL
zL
y
i = yi + qi
Fig. 1. Uplink of a cloud radio-access network with capacity-limited backhaul
a local optimum solution to the weighted sum rate maximization problem. Moreover, a simple approximation scheme
consisting of transmit beamformers matching to the strongest
channel signal vector and per-antenna scalar quantizers with
uniform quantization noise levels across the antennas at each
BS is also developed for the sum-rate maximization problem
when the signal-to-quantization-noise ratio (SQNR) is high and
when successive interference cancelation (SIC) is applied at
the receiver. This low-complexity scheme designs the transmit
beamforming and the backhaul compression separately. Numerical simulations show that with SIC implemented at the
receiver the proposed separate design is near optimal in the
SQNR regime of practical interest.
Hj Vj VjH HH
j + Kz + KQ ,
(3)
j=k
(1)
Jk = JSIC
=
k
Hj Vj VjH HH
j + Kz + KQ .
(4)
j>k
I
+
K
k=1 Hik Vk VkH HH
Qi
i
ik
log
Ci
(5)
|KQi |
for i = 1, 2, . . . , L, where Hik is the complex channel gain
matrix between kth user and ith BS.
To characterize the tradeoff between the achievable rates for
the users and the system resources, we formulate the following
weighted sum-rate maximization problem:
1473
max
Vk ,KQi
s.t.
1
k log I + VkH HH
k Jk Hk Vk
(6)
k=1
Jk = JLE
or Jk = JSIC
k
k
G
2
+
I
+
K
k=1 Hik Vk VkH HH
Qi
i
ik
Ci ,
log
|KQi |
KQi 0, for i = 1, 2, . . . , L,
(
)
Tr Vk VkH Pk , for k = 1, 2, . . . , G,
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
where k s are the weights representing the priorities associated with the mobile users.
Due to the non-convexity of both the objective function and
the backhaul capacity constraints in problem (6), finding the
global optimum solution of (6) is challenging. We point out
that the present formulation (6) can be easily extended to a
more general case where the user scheduling strategy is also
included. More specifically, one can consider a weighted sum
rate over all the users in the network, where the beamformers
for the users are set to be the zero vector if they are not
scheduled. In this paper, we focuses on the active uses only
and assume that the user scheduling is done prior to solving
problem (6).
2
Hik Vk VkH HH
(9)
i =
ik + i I + KQi .
k=1
H
H
Ek = (I UH
k Hk Vk )(I Uk Hk Vk )
Uk .
+ UH
Hj Vj VjH HH
k
j + Kz + KQ
j=k
2
Hik Vk VkH HH
ik + i I + KQi ,
k=1
1
H H
2
log |i | + Tr i
Hik Vk Vk Hik + i I + KQi
k=1
log |KQi | Ci + N
(11)
(12)
H
H
Hj Vj Vj Hj + Kz + KQ Hk Vk .
Uk =
(13)
j=k
Vk ,KQi ,Uk ,
Wk ,i
k=1
+
s.t.
(10)
Uk
log |i | +
i i F
( i=1 )
Tr 1
i i
(14)
log |KQi | Ci ,
KQi 0, for i = 1, 2, . . . , L,
(
)
Tr Vk VkH Pk , for k = 1, 2, . . . , G,
where
is some positive constant1 , Ci = Ci + N , and
G
2
i = k=1 Hik Vk VkH HH
ik + i I + KQi . Note that the last
term in the objective function which involves a summation of
Frobenius norms is a quadratic regularization item. It makes
problem (14) a strongly convex optimization with respect to
each optimization variable.
It is easy to verify that problem (14) is convex with respect
to any one of the optimization variables when the other
(8)
1474
1 In
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Vk ,KQi
s.t.
k=1
k Tr(Wk Ek ) +
i i F
(15)
i=1
(
)
Tr 1
i i log |KQi | Ci log |i | ,
KQi 0, for i = 1, 2, . . . , L,
(
)
Tr Vk VkH Pk , for k = 1, 2, . . . , G,
(
)
Initialize KQi = I, and Vk such that Tr Vk VkH = Pk .
G
2
i k=1 Hik Vk VkH HH
.
ik + i I + KQi )
(
1
H H
Uk
Hk Vk .
j=k Hj Vj Vj Hj + Kz + KQ
H
Wk I + Uk Hk Vk .
Fix i , Uk , and Wk , solve the convex optimization
problem (15). Set (Vk , KQi ) to be its optimal solution.
Repeat Steps 25, until convergence.
The WMMSE-SCA algorithm yields a nondecreasing sequence of objective values for problem (6). So the algorithm is
guaranteed to converge. Moreover, it converges to a stationary
point of the optimization problem. The convergence result is
stated in Theorem 2. Due to space limit, we omit the proof.
(0)
(0)
Theorem 2: From any initial point (Vk , KQi ), the limit
i=1
2
i log Gi KX GH
i + i I + KQi
i=1
G
i log |KQi |
j Tr (Kxk )
(17)
j=1
I
+
K
0.
Then
the
optimality
conQi
i
(i 2
)1
1
dition becomes i I + KQi
i KQi , i.e.,
KQi
i
2 I
1 i i
(19)
1475
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1.0
Pk
Vk =
k (1 : d)
(20)
d
where k (1 : d) is a matrix formed by d columns of k
which corresponding to the d largest singular values of Hk .
This channel-matching beamformer together with per-antenna
scalar quantizer provide us a low-complexity approximation
scheme, which separately designs the transmit beamforming
and the backhaul compression. Simulation results in the later
section show that with SIC such a separate design performs
very close to the optimized joint design in the SQNR regime
of practical interest.
Define
G
2
+
I
+
I
k=1 Hik Vk VkH HH
i
i
ik
.
(21)
C(i ) = log
|i I|
We summarize the approximation scheme as below
Algorithm 2 Approximate Scheme
1: Perform
SVD on Hk , i.e., Hk = k k H
k . Set Vk =
Pk
d k (1
2:
3:
: d) for k = 1, . . . , G.
Use bisection in [min , max ] to solve for i in C(i ) =
Ci , for i = 1, . . . , L.
Set KQi = i I for i = 1, . . . , L.
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
60
58
56
54
52
50
48
46
44
42
120
140
160
180
200
220
240
260
280
300
320
Fig. 3. Per-cell sum rate vs. average per-sector backhaul capacity of the
proposed schemes without SIC.
1476
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1.0
0.9
0.8
0.7
0.6
V. C ONCLUSION
0.5
This paper studies transmit beamforming and backhaul compression design for an uplink MIMO C-RAN system. We propose a novel WMMSE-SCA algorithm to efficiently optimize
the transmit beamformer and quantization noise covariance
matrix for maximizing the weighted sum rate. Further, it is
shown that with SIC at the receiver, channel-matching beamforming with uniform power allocation at the users together
with scalar quantization with uniform quantization noise levels
across the antennas at each BS is near optimal for maximizing
the sum rate when SQNR is high. This leads to a lowcomplexity algorithm, with separate transmit beamforming and
backhaul compression design. Simulation results show that
with the optimized transmit beamformer and quantization noise
covariance matrix, the C-RAN architecture can significantly
improve the performance of wireless cellular networks, and
majority of the gain can be achieved by the proposed lowcomplexity scheme.
0.4
0.3
0.2
0.1
0.0
0
72
68
64
60
56
R EFERENCES
Baseline: No central processor
Approx. scheme W/ SIC
WMMSE-SCA scheme W/ SIC
52
48
44
120
140
160
180
200
220
240
260
280
300
320
Fig. 5. Per-cell sum rate vs. average per-cell backhaul capacity of the proposed
scheme with SIC.
system with the proposed beamforming and backhaul compression schemes implemented without SIC at the receiver. It
is shown that both of the WMMSE-SCA and approximation
schemes significantly outperform the baseline scheme without
multicell processing. The figure also shows that WMMSE-SCA
scheme always outperforms the approximation scheme when
SIC is not implemented. As the backhaul capacity increases, the
performance gap between these two schemes becomes smaller.
However, as shown in Fig. 3 such a gap does not vanish even
in the high SQNR regime, which demonstrates that the joint
design is strictly better than the separate design when the CP
does not implement SIC.
Fig. 4 and Fig. 5 show the performance of the proposed
schemes with SIC implemented at the receiver. As compared
to the case without SIC, the proposed WMMSE-SCA and
approximation schemes can achieve more gain on the overall
performance with SIC applied. The performance improvement
is more significant for the users with low rate (e.g. 10percentile users), as shown in 4. The figures also show that the
approximation scheme performs very close to the WMMSESCA scheme, especially when backhaul capacity is large. As
shown in Fig. 5, when the backhaul capacity goes to infinity,
1477
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Dept. of Information Engineering, Electronics and Telecommunications, Sapienza University of Rome, Rome, Italy
2
Dept. of Electrical Engineering, State University of New York at Buffalo, Buffalo, USA
I. I NTRODUCTION
The main motivation underlying this work comes from two
big challenges of current development of wireless networks:
the scarcity of energy resources at the mobile terminals
necessary to run increasingly sophisticated applications under
latency constraints dictated by the users Qualty of Experience
(QoE) and the need of optimizing allocation of communication/computation resources in a dense network deployment
scenario. The way to overcome the limitations of cellphones
batteries is to offload computations from the mobile devices
to a cloud service provider. Computation offloading is a key
strategy to improve the computational capabilities of mobile
handsets or to reduce energy consumption at the mobile
terminal. A strong impulse to computation offloading has come
The work of Barbarossa and Sardellitti was funded by the European
Commission, within the 7th Framework Programme, Project ICT-TROPIC,
under grant no. 318784. The work of Scutari was supported by the USA NSF
under Grants CMS 1218717 and Career Award no. 1254739.
1484
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1485
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
(1)
where: tkn is the time necessary to send the program state and
input (encoded with bk bits) necessary to transfer the program
execution from the k-th MUE to the n-th base station;
exe
mk = k /fmk
(2)
Q,f ,a
s.t.
E(Q, a) ,
K
X
k Ek (Q, ak )
(P)
k=1
iv)
Nb X
Nc
X
k, n, m
n=1 m=1
(6)
where
we
have
introduced
the
function
wk
ck
+ TBnm .
+
gknm (Q, fmk , aknm ) , aknm
rkn (Q) fmk
The constraints listed above have the following meaning:
i) the overall latency for each user k must be lower than the
maximum tolerable value Lk ;
ii) the total power spent by each user must be lower than its
total power budget Pk ;
iii) the sum of the computational rates fmk assigned by each
server cannot exceed the server computational capability Fm ;
iv) each mobile user should be served by one base stationcloud pair. Then to drive the solution towards the situation
where each user is served by a single base station and a
Nb X
Nc
X
single cloud, we enforce the constraint
aknm = 1,
n=1 m=1
1486
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
E(Q, a)
s.t.
Q,f ,a
hm (f , a) Fm , m, f 0,
k, n, m
(7)
Q Q, a A.
We denote by X the feasible set of the optimization problem
in (7). Even by relaxing the binary variables a, problem in
(7) is still non-convex, since the objective function and the
constraints gknm and hm are not convex. In what follows,
we exploit the structure of (7) and building on some recent
advances on Successive Convex Approximation (SCA) techniques, as proposed in [10], [11], we devise an efficient iterative approximation algorithm converging to a local optimal
solution of (7).
III. A LGORITHMIC DESIGN
To solve the non-convex problem (7) efficiently, we adopt an
SCA-based algorithm where the original problem is replaced
by a sequence of strongly convex problems. To do this, we need
to find, at each iteration, a suitable convex approximation of
the nonconvex objective function E(Q, a) and the constraints
gknm (Q, fmk , aknm ), which are preliminarily discussed next.
where hA, Bi , tr(AH B) and Qk Ej (Z) denotes the conjugate gradient of Ej (Z) with respect to Qk evaluated at Z .
Note that in (8) we also added quadratic regularization terms,
making Ek (Zk ; Z ) uniformly strongly convex with respect
to Zk .
Based on each Ek (Zk ; Z ), given Z 0, we can now define
E(Z;
Z ) ,
K
X
k Ek (Zk ; Z ).
A. Approximant of E(Q, a)
Let Z , (Q, f , a), Z , (Q , f , a ) and Zk ,
c
(Qk , fk , ak ) with fk , (fmk )N
m=1 . Following [10], [11]
the main idea is to approximate around the current (feasible) iterate Z X the original nonconvex nonseparable
objective function E(Q, a) with a strongly convex func
tion, say E(Z;
Z ), that is separable in the MUEs variables and has the same first order behaviour of E(Q, a)
at Z . To motivate our choice observe preliminary that i)
for
P any
P given Qk = tQk each term Ek (Qk , ak ; Qk ) =
a
tr(Q
)
(Q
;
Q
)
is
biconvex
in
Q
k
k
k and
kn
k
n
m knm
convex
in
a
given
Q
;
and
ii)
the
other
terms
of
the
k
P
sum j6=k Ej (Qk ; aj , Qk ) are not convex in Qk . Thanks
to this structure a convex approximation of E(Q, a) can
be obtained for each MUE k by convexifying the term
E
by linearizing the non-convex part
Pk (Qk , ak ; Qk) and
,
Q
).
E
(Q
;
a
k
j
k
j6=k j
(9)
k=1
(11)
+
and Rkn (Qk ) defined in (4). Note that
Q rkn () and rkn () are
concave on Q and convex on Qk , m6=k Qm , respectively.
Using (10), and observing that at any feasible Q, f , a, it must
be rkn (Q) > 0 and fmk > 0 for all k, n and m, the constraints
gknm (Q, fmk , aknm ) in (7) can be rewritten as
+
gknm (Q, fmk , aknm ) = rkn
(Q) rkn
(Qk )
(12)
1487
+
gknm (Q, fmk , aknm ; Qk ) , rkn
(Q) rkn
(Qk )
E
X D
Qj rkn
(Qk ), Qj Qj + qknm (fmk , aknm ),
jI,j6=k
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1
with Qj rkn
(Qk ) = HH
Hjn . Finally the
jn Rkn (Qk )
non-convex constraint hm (f , a) may be replaced by the following convex approximation
m (f , a; Z ) ,
h
Nb
K X
X
(13)
k=1 n=1
) , argmin E(Q,
f , a; Z )
Z(Z
Q,f ,a
s.t.
k I,
aA
(P )
) , (Q(Z
), f (Z ), a
(Z )) the
where we denoted by Z(Z
unique solution of the strongly convex optimization problem.
The proposed solution method consists in solving iteratively
problem P , starting from a feasible Z0 . The formal description of the algorithm is given in Algorithm 1 below. Note that
in Step 2 of the algorithm we allow a memory in the update of
the iterate Z , (Q , f , a ). A practical termination criterion
in Step 1 is to stop the iterates when |E(Z+1 ) E(Z )| ,
where > 0 is the prescribed accuracy. The convergence proof
of Algorithm 1 is omitted because of the space limitation; it
consists in showing that all the conditions in [10, Th.1] are
satisfied.
Algorithm 1 : Inner SCA Algorithm for P
Data: Z0 , (Q0 , f 0 , a0 ) X ; { } (0, 1]; k , f > 0;
a > 0. Set = 0.
(S.1): If Z satisfies a suitable termination criterion, STOP
) , (Q(Z
), f (Z ), a
(Z )) [cf. P ];
(S.2): Compute Z(Z
) Z ;
(S.3): Set Z+1 = Z + Z(Z
(S.4): + 1 and go to (S.1).
IV. N UMERICAL
RESULTS
1488
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
10
cloud2
MUE3
MUE4
BH12
BS2
cloud1
BH23
BH42
BH13
MUE2
MUE1
BS1
BH14
cloud4
cloud3
BH43
MUE5
10
BS3
BS4
MUE6
10
0.5
1.5
2.5
d1
10
K =4
0.1
SCA algorithm
exhaustive search
0.09
0
10
0.08
E1
0.07
E1
0.06
1
10
0.05
0.04
0.03
10
0.5
1.5
2.5
0.02
d1
0.01
2
Nb
[5] FP7 European Project. (2012). Distributed computing, storage and radio
resource allocation over cooperative femtocells (TROPIC). [Online].
Available: http://www.ict-tropic.eu.
[6] S. Barbarossa, S. Sardellitti and P. Di Lorenzo, Joint allocation of
computation and communication resources in multiuser mobile cloud
computing, Proc. of IEEE 2013 Int. Workshop on Signal Process.
Advances in Wireless Commun. (SPAWC 13), Darmstadt, Germany, 16
19 June 2013.
[7] S. Sardellitti, G. Scutari, and S. Barbarossa,Joint optimization of radio
and computational resources for multicell MIMO mobile cloud computing, Proc. of IEEE 2014 Int. Workshop on Signal Process. Advances in
Wireless Commun. (SPAWC 14), Toronto, Canada, 2225 June 2014.
[8] S. Sardellitti, G. Scutari, and S. Barbarossa,Distributed joint optimization
of radio and computational resources for mobile cloud computing, Proc.
of IEEE Int. Conf. on Cloud Networking (CloudNet), Luxembourg, 810
October 2014.
[9] G. Scutari, F. Facchinei, P. Song, D.P. Palomar, and J.-S. Pang, Decomposition by partial linearization: Parallel optimization of multi-agent
systems, IEEE Trans. on Signal Process., vol. 63, no. 3, pp. 641656,
Feb. 2014.
[10] G. Scutari, F. Facchinei, L. Lampariello, and P. Song, Parallel and
distributed methods for nonconvex optimization, Proc. of IEEE Int. Conf.
on Acoustics, Speech, and Signal Process. (ICASSP 14), Florence, Italy,
49 May 2014,.
[11] G. Scutari, F. Facchinei, L. Lampariello, and P. Song, Parallel and
distributed methods for nonconvex optimization-Part I&II: Theory &
Applications, IEEE Trans. on Signal Process., (under review) 2014.
1489
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Fig. 1.
links.
1478
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Sec. II. as well as introduce our own system model. Sec. III.
describes the theoretic background and the improved decoder,
for which numerical results are given and discussed in Sec.
IV. The paper closes with our conclusions in Sec. V.
II. P REVIOUS W ORK AND S YSTEM M ODEL
In todays mobile networks, reliable communication is
ensured by modern forward error correcting (FEC) channel
codes such as turbo codes, which are, e.g., specied for the
3GPP LTE standard [9]. These codes can be decoded with
high performance thanks to the use of soft-information in
algorithms like the soft-Viterbi [10] or BCJR [11] algorithm.
The soft-information, often given in the form of log-likelihood
ratios (LLRs), is computed during demodulation/detection and
requires knowledge of the transmissions underlying random
process, which is usually assumed to be that of an additive
white Gaussian noise (AWGN) channel. This assumption
usually holds with sufcient precision, since other channel
effects of mobile channels such as multipath propagation and
Doppler effect are compensated by an equalizer.
In [12], the authors investigated how the distribution of
Gaussian received samples changes, if these sample are stored
in digital form in a faulty buffer memory. The background to
their work is that, due to aggressive voltage scaling in CMOS
technology, the samples will be incorrectly read from the
memory, resulting in uniformly distributed bit-errors. These bit
errors change the distribution of the samples in a deterministic
way. Building on this, the authors of [7], [8] developed error
resilient decoders that rely on a soft-information calculation
that no longer assumes a pure Gaussian, but the modied distribution. In this work, we now adapt their idea for application
in the uplink of cloud-based mobile networks.
RAN
UE
RRH
FH
BBU
Fig. 2.
1479
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
(x 1)2
,
2 2
(1)
2 2
which is depicted as the green curve in Fig. 3. The corresponding LLRs are calculated as
f (x |x = 1)
.
(2)
L(x ) = ln
f (x |x = +1)
However, we want to determine the soft information not of x
but of x , which cannot be assumed to be Gaussian distributed
due to the effect of the BSC. To model its effect, we assume
in the following the use of a uniform linear mid-rise quantizer
with a step-width of x. The bit vector w is corrupted by
between 0 and B bit ips on the BSC, which means that after
dequantization the amplitude will be shifted depending on the
number and position of bit ips. For illustration, we assume
that the bit vector w received a single bit error on the b-th bit.
From here on we also skip the condition for x in our notation,
as the effect is similar for both values of x. The resulting
distribution can be calculated as [12]
f (x + 2b ), m even
1
(3)
fb (x ) =
f (x 2b ), m odd
with
m=
and
w
2Bb1
(6)
so that we can write the new distribution that takes all possible
bit errors of one quantized received symbol into account as
f (x ) =
k=0
P (k)f (x ).
(7)
1.2
1
(5)
B
1.4
(4)
= 2B1 x.
PDF(x)
f (x |x = 1) =
0.8
0.6
0.4
0.2
0
0001 0010 0011 0100 0101 0110 0111 1000 1001 1010
x
1480
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1.4
f(x')
f(x'')
1.2
PDF(x)
0.8
0.6
0.4
0.2
0
-5
-4
-3
-2
-1
0
x
Fig. 4.
Pure Gaussian distribution of a BPSK signal (green) and after
forwarding the signal over a BSC with Pe = 101
10
-1
Pe=10 , conv.
-1
Pe=10 , adapted
-1
10
-2
Pe=5*10 , conv.l
-2
Pe=5*10 , adapted
BER
-2
10
-2
Pe=10 , conv.
-2
Pe=10 , adapted
-3
10
-3
Pe=10 , conv.
-3
Pe=10 , adapted
-4
Pe=0, conv.
10
4
6
SNR in dB
10
Pe=0, adapted
Fig. 5. End-to-end bit error rate under different SNRs of the radio access
channel and different bit error probabilities of the fronthaul channel for a
conventional decoder assuming a Gaussian distribution (solid lines) and the
adapted decoder (dashed lines)
1481
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
10
-1
P e,est=P e=5 10
-2
BER
10
10
10
-3
P e,est=10
-2
P e,est=10
-3
P e,est=10
-4
P e,est=10
-5
P e,est=10
-6
-2
V. C ONCLUSION
P e,est=0.999
-4
3.5
4.5
SNR in dB
5.5
Fig. 6. End-to-end bit error rate under different SNRs of the radio access
channel for the adapted decoder and Pe = 5 102 when Pe is worngly
estimated
1.4
1.2
SNR loss in dB
1
0.8
0.6
0.4
0.2
0 -5
10
10
-4
10
-3
10
-2
10
-1
10
10
SN R
1
Pe erfc
(9)
2
2
To achieve the Pe of 1012 required by CPRI, an SNR of
about 16.5 dB is needed as compared to only 7.3 dB for a Pe
of 103 , which would have the same performance according
to our simulations. Thus, by using a lower transmit power and
accepting a higher Pe , 9.2 dB or 88 % of transmit power could
be saved. Conversely, using the same transmit energy and
assuming a ber attenuation of 0.2 db/km [15], the range could
be extended by 46 km. Using wireless FH and assuming that
directive mmWave links are subject to free-space path loss, the
transmission distance could be increased by a factor of 2.9.
All these options would result in tremendous cost savings for
network operators.
Of course, the provided results apply only if the data transmitted over the fronthaul is pre-equalized user data. Stricter
1482
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1483
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
I.
INTRODUCTION
1490
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
SOVNO/VDC #2
MOVNO#1
MOVNO#1
Second. Path
Primary Path
Metro
Wireless Access
Metro Optical
Metro
Wireless Access
Data Centers
Metro Optical
Data Centers
a)
b)
Fig. 1 MOVNO Planning with a) Resilience considerations: b) Security Considerations for the case where no common links are allowed between any two
MOVNOs
MODEL DESCRIPTION
A. Problem Description
As already discussed, this paper focuses on planning of
MOVNOs over an integrated converged infrastructure
comprising a cellular LTE system for the wireless access part
and an optical metro network supporting frame based sub-
1491
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
wavelength switching granularity that interconnects the endusers with the computing resources. The objective of the
proposed planning problem is to identify the virtual
infrastructure topologies and resources required by the various
fixed and mobile tenants to facilitate them to provision services
and execute their applications over the common physical
infrastructure.
The VI planning model addresses also the requirements for
dependability of the planned VIs, offering at the same time
seamless connectivity to the mobile users. In this paper
dependability is supported through the adoption of resilience and
security mechanisms. In the VI planning phase, these
mechanisms are considered it terms of the suitable VI topology
selection and the additional resource requirements they
introduce. More specifically resilience in case of single link
failures is supported through the allocation of additional
resources required to facilitate 1:1 protection for all virtual
resources per MOVNO (Fig.1a). On the other hand, security is
supported according to [18]-[19], by keeping security sensitive
VIs disjoint to the rest of the VIs. By disjoint VIs we mean that
security sensitive VIs are allowed to have no or a limited number
of common physical links that they share with other VIs
(Fig.2a). The number of acceptable common links can be
defined depending on the security level that needs to be
supported. The rationale behind this policy is that, at any time
instance VIs will be realized through full or partially disjoint
physical layer paths, thus, minimizing possible effects from
other malicious VIs sharing the same physical substrate.
subject to
x scheduling constraints described through (see eqs.1, 2)
1492
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Wireless Access
Wireless Backhaul
Metro Optical
Data Centers
9000
8000
10000
7000
6000
5000
4000
DC
Metro Optical
Wireless Backhaul
Wireless Access
100
80
60
40
20
0
wo
with
wo
with
1
Call to mobility factor
wo
with
a)
b)
Fig. 2 a) Impact of traffic load on power consumption with and without network protection, b) Impact of mobility factor on the requested resources
1493
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Resource Efficiency
1
0.9
0.8
0.7
0.6
6
0.04
0
0.06
0.3
0.2
0.1
0
2
1
Common links
2
1
Common links
Wireless Access
Wireless Backhaul
Metro Optical
0
W/O protection
0.02
2
Call to mobility factor
0.4
WIth protection
0.5
Fig. 4 Additional resources required (over the dynamic scheme with full
information) for different availability requirements
1494
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
[7]
[8]
[9]
V. CONCLUSIONS
The increasingly deployed cloud and mobile cloud
computing services are emerging as one of the fastest growing
business opportunities. To enable this emerging business
opportunity, there is a clear need for suitable network solutions
to interconnect distributed DCs with fixed and mobile end users.
Our proposal involves a converged network integrating optical
metro and wireless access technologies with DCs in a common
infrastructure providing ubiquitous access to fixed and mobile
end users. To effectively and efficiently utilize the physical
infrastructure the model of multi-tenancy is proposed and is
facilitated through the adoption of infrastructure virtualization
and the creation of dependable Mobile Optical Virtual Networks
(MOVNOs). Dependability is defined as the MOVNOs ability
to offer resilience in case of network or computing resource
failures and security guarantees in terms of isolation between
MOVNOs that share the common physical infrastructure.
MOVNOs are planned with the objective to minimize the overall
energy consumption satisfying at the same time specific end-toend delay constraints. To address these issues, together with the
uncertainty introduced by the mobility of the end devices a
MOVNO planning scheme based on NLP has been developed.
Our modeling results illustrate trade-offs between the overall
energy consumption of the MOVNOs, the mobility of the enddevices, as well as the dependability impact on the energy
consumption and the utilization of the infrastructure resources.
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
ACKNOWLEDGMENT
This work was supported by the CONTENT (FP7-ICT318514) project funded by the EC through the 7th ICT FP and
the EPSRC grant EP/L020009/1: Towards Ultimate
Convergence of All Networks (TOUCAN) .
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
1495
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Aleksandra Checko
MTI Radiocomp
DK-3400 Hillerd, Denmark
aleksandra.checko@mtigroup.com
C-RAN,
mobile,
aggregation,
SDN,
traffic
I. INTRODUCTION
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Late aggregation
aggregat ion
aggregat ion
Fig. 1. The balance between early vs. late aggregation from an optimisation
perspective (1 stage vs. 2 stage)
Early aggregation
aggregat ion
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
Client
aggregat ion
1496
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
RNC
RNC
RNC
RNC
Packet
mobile
core
NodeB
RNC
NodeB
RNC
RNC
RNC
RNC
circuit
mobile
core
NodeB
Traffic level aggregation
eNodeB
eNodeB
eNodeB
Mobile core
RRH
eNodeB
Front-haul link
eNodeB
RRH
RRH
Front-haul link
Front-haul link
RRH
RRH
RRH
BBUs
Front-haul link
Traffic level
aggregation
Service level
aggregation
Fig. 4. Simple fronthaul link system with separation of RRH and BBU.
1497
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
(typically line rates in the Gbps range is required for a LTE cell
with 2x2 MIMO) [5].
Legend
RF
Logical
connection
RF
RF
RRH Aggregation
network
Access
network
BBU
Base
Base
Base
Base
pool
band
band
band
band
S1
R
F
X2
RF
RF
RRH Aggregation
network
RF
R
F
Fronthaul
BBU
Base
Base
Base
Base
pool
band
band
band
band
Aggregation
network
EPC
MME
S1
S5
SGW
Application server
Internet
S11
PGW
Application server
Backhaul
RRH
Front-haul link
RRH
RRH
Front-haul link
Front-haul link
RRH
RRH
RRH
Front-haul switch
with BBU pool
RRH
Front-haul link
Traffic level
aggregation
Service level
aggregation
1498
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Load
balancing
switches
RRHs
Virtual
BBU Pools
GPS?, 1588?,
new solution?
CPRI
BBU
RRH
Variable delay
BBU Pool++
Variable delay
BBU Pool++
RRH++
Aggregation
network backhaul
BBU
Pool
CPRI2Eth
Gateways
CPRI /
OBSAI
CPRI2Eth
Gateway
1499
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
VI. CONCLUSION
By introducing C-RAN (in addition to directional oriented
antenna system) the EU project HARP demonstrates a cost
efficient and scalable solution for a front-haul network
supporting current and future mobile networks.
Even though the EU project HARP mainly is addressing
currently defined mobile networks standards, the C-RAN based
front-haul network might be even more relevant for the next
generation 5G network that currently is being research as a
concept to support an approach that can meet the demand in
terms of capacity, coverage, mobility, service support etc.
1500
REFERENCES
[1]
[2]
[3]
[4]
[5]
www.cpri.info
A. Checko, L. Dittmann, L. Scolari, H. Christiansen, HARP - High
capacity network Architecture with Remote radio heads & Parasitic
antenna arrays, in proc. of EuCNC 2013, Lisbon, Portugal.
A. Checko, H.L. Christiansen, Y. Yan, L. Scolari, G. Kardaras, M.S.
Berger, L. Dittmann, Cloud RAN for Mobile Networks - a
Technology Overview accepted for publcation in IEEE
Communications Surveys and Tutorials
A. Checko, H. Holm, and H. Christiansen, Optimizing small cell
deployment by the use of C-RANs, in European Wireless 2014
(EW2014).
L. Dittmann, presentation ITU JCA-SDN meeting 11/7/14, Geneva,
Swiss.
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Fundaci i2CAT
Barcelona, Spain
{jordi.ferrer,carlos.bock,eduard.escalona}@i2cat.net
University of Essex
Colchester, United Kingdom
{mcpark,stuwal,quinlan}@essex.ac.uk
Victor Marques
David Levi
Ethernity Networks
Israel
vidi.levi@ethernitynet.com
AbstractIt is envisaged that end-user access bandwidth
requirements will notably increase in the coming years; at the
same time, the number of connected mobile devices will also
exponentially grow as the fully-digital connected homes (Internet
of Things) becomes an everyday reality. This article presents an
active remote node (ARN) at an intermediate location between
the central office and end-user premises, as a flexible and futureproofed infrastructure topology approach for solving the
associated bandwidth and wired-wireless convergence issues. The
ARN represents the key architectural design innovation of the
SODALES (SOftware-Defined Access using Low-Energy
Subsystems) network. The rise of mobile communications, and
the trend for seamless convergence between fixed and wireless
networks is tending to make the purely passive approaches (e.g.
as exemplified by passive optical networking (PON) access
architectures) too restrictive, considering the modularity and
flexibility offered by an active remote node. We present a
performance analysis of the ARN node, to support the costeffectiveness of the proposed SODALES solution and to
demonstrate the potential benefits in terms network
performance, operational efficiency, and flexible functionality.
Looking forward, future ARN capabilities can also be expected to
include hierarchical caching, customer premises equipment
(CPE) virtualization, and nearer to the end-user location of
software-defined platforms supporting ubiquitous cloud services.
Keywordsaccess network convergence, active remote node,
performance analysis
I.
1501
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1502
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1503
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
1504
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
25"
20"
15"
10"
5"
Tr
a
c"i
n
"( G
bp
s) "
1,
00
"
2,
00
"
3,
00
"
4,
00
"
5,
00
"
6,
00
"
7,
00
"
8,
00
"
9,
00
10 "
,0
0"
11
,0
0
12 "
,0
0
13 "
,0
0
14 "
,0
0
15 "
,0
0
16 "
,0
0
17 "
,0
0
18 "
,0
0
19 "
,0
0
20 "
,0
0
21 "
,0
0
22 "
,0
0
23 "
,0
0
24 "
,0
0
25 "
,0
0
26 "
,0
0
27 "
,0
0
28 "
,0
0
29 "
,0
0"
0"
Fig. 5. Network traffic of a real FTTH network offering 100 Mb/s to 500
subscribers. Note that downstream traffic tends to be limited by the server
side, while upstream reaches maximum customer capacity.
Fig. 8. Traffic load of a real FTTH deployment used for the simulations
1505
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
1506
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
I. I NTRODUCTION
Mobile services demand is becoming more and more important. Therefore, the estimation of performance and quality
of service (QoS) has to be more and more precise in the
way to answer with accuracy to the demand. Performance
and quality of service evaluations of wireless networks can be
analyzed by using simulations or analytical models. Several
QoS parameters (like throughput, outage probability) can be
derived from the SINR distribution [1] [2] [3] [4].
Moreover the knowledge of the signal to interference plus
noise ratio (SINR) reached by a UE (User Equipment) allows to better perform the radio resource allocation and the
scheduling policy. Analytical models thus try to derive simple
SINR formula in order to quickly evaluate the performance of
a cellular network. Therefore, their analysis need tractable and
accurate models of networks. Two factors play an important
role in the evaluation of the SINR : the localization of BS
and the propagation phenomena. Among the network model
usually considered, the Hexagonal one is the most used.
However this model is based on a regular deployment of BS on
1 Jean-Marc
Constant Bridon4
1507
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Du = log2 (1 + u ).
(2)
Fig. 1.
Poisson Network
Continuum
of base stations
Rnw
B. SINR of a user
We consider a single frequency network composed of N
base stations, transmitting at power P on each subcarrier. We
define gi (u) the path gain between BS i and user u on a
given subcarrier. The SINR u of user u served by BS i on
the considered subcarrier is given by:
u = P
P gi (u)
.
P gj (u) + Nth
Rc
2Rc
(1)
j6=i
Fig. 2.
1508
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
r
= P ,
rj
j6=i
2
r
2BS (2Rc r)2
(4)
If BS is known, this analytical model allows to significantly simplify the calculation of the SINR: the only required
variable is the distance of the mobile to its serving BS. This
model has been proven to be reliable and close to the reality
for homogeneous hexagonal networks [14] [15], as well as for
heterogeneous networks [6].
The cell edge throughput can particularly be calculated
by setting r = Rc in (2) and (4). Therefore, the minimum
performance and quality of service offered to UE is evaluated
in a simple way. Moreover, a simple integration over the cell
range allows to calculate the average throughput of the cell.
These results on the fluid model are valid for a constant
inter site distance 2Rc . In the following section, we remind
that we established [17] a correspondence between a stochastic
Poisson network and the Fluid network model.
C. Comparison of the SINR CDF for the Poisson and Fluid
models
Since the CDF of the SINR characterizes the performance
and the quality of service of wireless systems, we established
a modified expression of the SINR given by the fluid model,
which allows to calculate the CDF of SINR for the Poisson
network reminded hereafter [17]:
ied
SIN RFmodif
= SIN RF luid (a + b)
luid
By considering expression (3), we can see that the calculation of the SINR of a user depends on its location. We notice
that in this case, users are connected to their best serving
station which is also their nearest BS.
B. Pathloss and Shadowing impact on the SINR
1) Propagation: Considering the power Pj transmitted by
the BS j, the power pj,u received by a mobile u can be written:
Yj,u ,
pj,u = Pj Krj,u
(7)
j,u
(8)
pY (s) =
as
2a
p0,u = P0 Kr0,u
Y0,u .
(9)
For the sake of simplicity, we now drop index u and set r0,u =
r. The interferences received by u coming from all the other
base stations of the network are expressed by:
(5)
pext =
N
X
Pj Krj Yj .
(10)
j=1
which yields
ied
CDFP oisson CDFFmodif
luid
(6)
j=1
P0 Kr Y0
Pj Krj Yj + Nth
(11)
maxN
j=0 (Pj Krj Yj )
= PN
.
(12)
1509
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
Fig. 3. CDF of the SINR with = 2.6 (left) and = 3 (right), for a Poisson
model network and standard deviations of the shadowing values comprised
between 0 dB (no shadowing) (curve in the left of each figure) and 8 dB
(curve in the right of each figure). For readability, only the curves drawn with
= 0 and 8 dB are indicated in the figure. The curves between these two
curves represent the cases = 3 dB, .6 dB. They are in the line thickness
Remark
We consider a model proposed in [18]. The Rayleigh fading,
not considered in this model, may be indistinguishable from
shadowing if the fading is sufficiently slow [18], as for
example if the mobile travels through a region of deep fades
at a very slow speed.
C. Cumulated Distributed Function of the SINR with Shadowing
We consider a Poisson model network. And we calculate
the CDF of the SINR, in the zone covered by this network,
by taking into account the shadowing. The UE are connected
to their best serving station, i.e. the BS which offers the best
signal. The SINR is calculated by using (12).
The figures 3 and 4 show the CDF of the SINR for different
values of the pathloss parameter and for standard deviation
of values 0 dB (no shadowing), 3 dB, 6 dB, and 8 dB. It
can be observed that for values of 3.5 and 4 (Fig. 4), the
curves between 0 and 8 dB are indistinguishable. Only for
= 2.6 (Fig. 3) it is hardly possible to distinguish between the
different curves. These curves show that for typical values of
the pathloss parameter , comprised between 2.6 and 4, and
values of the standard deviation of the shadowing ( 6 dB)
the impact of the shadowing is negligible : the difference
Fig. 4. CDF of the SINR with = 3.5 (left) and = 4 (right), for a Poisson
model network and standard deviations of the shadowing values comprised
between 0 dB (no shadowing) (curve in the left of each figure) and 8 dB
(curve in the right of each figure). For readability, only the values of 0 and 8
dB are indicated in the figure. The curves between these two curves represent
the cases 3 dB, 6 dB. They are in the. line thickness
1510
Globecom 2014 Workshop - Wireless optical network convergence in support of cloud architectures
taking into account the shadowing. Therefore it becomes possible to analyze realistic wireless networks without considering
the shadowing, in the standard range of values of propagation
pathloss parameter , and standard deviation of shadowing
. As a consequence the CDF of the SINR, established by
Monte Carlo simulation for the Poisson model network is very
close to the one calculated with the analytical expression of
the SINR given by the Fluid model network when a linear
function of the propagation parameter is applied. Therefore,
the analysis of performance, outage probability, throughput,
the radio resource allocation, the scheduling policy can be
done in an easy and quick way, by using the expressions
established with the Fluid Model network.
Fig. 5. CDF of the useful signal and the interferences, for = 2.6 (left)
and = 4 (right), without shadowing ( = 0 dB, blue continuous curves) and
with shadowing ( = 6 dB, red dotted
. .curves)
ACKNOWLEDGMENT
This work was supported by the Seventh Framework Program for Research of the European Commission under grant
number HARP-318489.
R EFERENCES
[1] X. Lagrange, Principes et e volutions de lUMTS, Hermes, 2005.
[2] K. S. Gilhousen, I. M. Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver,
and C. E. Wheatley, On the Capacity of Cellular CDMA System, IEEE
Trans. on Vehicular Technology, Vol. 40, No. 2, May 1991.
[3] S. E. Elayoubi and T. Chahed, Admission Control in the Downlink of
WCDMA/UMTS, Lect. notes comput. sci., Springer, 2005.
[4] A. J. Viterbi, A. M. Viterbi, and E. Zehavi, Other-Cell Interference in
Cellular Power-Controlled CDMA, IEEE Trans. on Communications,
Vol. 42, No. 2/3/4, Freb/Mar/Apr. 1994.
[5] J-M. Kelif, M. Coupechoux, On the Dimensioning of Cellular OFDMA
Networks, Physical Communication Journal, Ref : PHYCOM118, doi :
10.1016/j.phycom.2011.09.008, online October 2011,
[6] J-M. Kelif, W. Diego and S. Senecal, Impact of Transmitting Power
on Femto Cells Performance and Coverage in Heterogeneous Wireless
Networks, Proc. of WCNC, 2012.
[7] F. Baccelli, B. Blaszczyszyn, and F. Tournois Downlink admission/congestion control and maximal load in CDMA networks, in Proc.
of IEEE INFOCOM03, 2003.
[8] A. Agarwal and A. K. Jagannatham, Distributed Estimation in Homogenous Poisson Wireless Sensor Networks, IEEE Wireless Communications Letters, Vol. 3, Feb. 2014
[9] J-M. Kelif, M. Coupechoux and P. Godlewski, Effect of Shadowing on
Outage Probability in Fluid Cellular Radio Networks, WiOpt 2008
[10] J-M. Kelif and M. Coupechoux, Impact of Topology and Shadowing on
the Outage Probability of Wireless Networks, ICC 2009
[11] M. Minelli, M. Coupechoux, J.-M. Kelif, M. Ma, SIR Estimation in
Hexagonal Cellular Networks with Best Server Policy, Wireless Personal
Communications Journal,2012
[12] B. Blaszczyszyn and M. Karray, Wireless Cellular Networks with
Shadowing, NetGCooP, Paris, 12-14 October 2011
[13] L. Rong, S. E. Elayoubi, and O. Ben Haddada. Performance Evaluation
of Cellular Networks Offering TV Services. IEEE Trans. on Vehicular
Technology, 60(2):644655, Feb. 2011.
[14] J-M. Kelif and E. Altman, Downlink Fluid Model of CDMA Networks,
Proc. of IEEE VTC Spring, May 2005.
[15] J-M. Kelif, M. Coupechoux, P. Godlewski, A Fluid Model for Performance Analysis in Cellular Networks, EURASIP Journal on Wireless Communications and Networking, Vol. 2010, Article ID 435189,
doi:10.1155/2010/435189
[16] M. Maqbool, P. Godlewski, M. Coupechoux and J-M. Kelif, Analytical
Performance Evaluation of Various Frequency Reuse and Scheduling
Schemes in Cellular OFDMA Networks, Performance Evaluation Journal, 2009.
[17] J-M. Kelif, S. Senecal, C. Bridon, M. Coupechoux, A Fluid Approach
for Poisson Wireless Networks, arXiv:1401.6336, 2014
[18] A. J. Viterbi, CDMA - Principles of Spread Spectrum Communication,
Addison-Wesley, 1995.
1511