LTE Wireless Virtualization and Spectrum

Management

Yasir Zaki
*
, Liang Zhao
*
, Carmelita Goerg
*
and Andreas Timm-Giel
**

*
TZI-ComNets/University of Bremen, Bremen, Germany
**
Institute of Communication Networks/TUHH, Hamburg, Germany
{yzaki/zhaol/cg}@tzi.de, timm-giel@tu-harburg.de


Abstract Many research initiatives have started looking
into Future Internet solutions in order to satisfy the ever
increasing requirements on the Internet and also to cope
with the challenges existing in the current one. Some are
proposing further enhancements while others are proposing
completely new approaches. Network Virtualization is one
solution that is able to combine these approaches and
therefore, could play a central role in the Future Internet. It
will enable the existence of multiple virtual networks on a
common infrastructure even with different network
architectures. Network Virtualization means setting up a
network composed of individual virtualized network
components, such as nodes, links, and routers. Mobility will
remain a major requirement, which means that also wireless
resources need to be virtualized. In this paper the Long
Term Evolution (LTE) was chosen as a case study to extend
Network Virtualization into the wireless area.
I. INTRODUCTION
The Internet started in the Sixties as a US Defense
Department project mainly to investigate possibilities for
robust non-centralized digital communications. The main
reason for the project was that the telephone system
(which was the only communication system at that time)
had major problems of being dependent on central
switching stations which were points of failure for the
system. So the question was whether it would be possible
to design a network that could quickly reroute digital
traffic around failed nodes?
The solution was to build a datagram network called
"catenet", and use dynamic routing protocols that can
adjust the traffic flow through the network. The US
Defense Advanced Research Projects Agency (DARPA)
launched the DARPA Internet Program [1]. Over the years
the Internet has continuously been improved to grow to a
network with 1.668 billion users [2]. Services have
evolved from data and message exchange to todays
worldwide web providing multi-channel communication
services and access to multimedia content. Further
challenges are visible: the Internet of Things [3] requiring
addresses and addressability of billions of sensors and
devices and multimedia services, such as IPTV offering
high quality video content over the Internet, which are just
some examples for future trends and requirements on the
Internets architecture and protocols. Within todays
research communities, some investigates new
architectures and protocols for the Future Internet, while
others work on adapting the Internet protocols to cope
with all these future challenges as extensions to the
current Internet. Network virtualization is one alternative
migration path from the existing Internet to a new Internet
and thus leads to the new Internet with isolated networks
sharing common resources, like routers and links.
II. NETWORK VIRTUALIZATION
Virtualization itself is not something new; it is a well
known technique that has existed for years, especially in
the computer world like the use of virtual memory and
virtual operating systems. The new idea is to use
virtualization to create complete virtual networks. This
involves applying the current operating system
virtualization experience for network components, leading
to virtual network resources like virtual routers, virtual
links, and virtual base stations.
A number of research initiatives and projects all over
the globe have started focusing on Network Virtualization,
e.g. GENI [4] [5], PLANETLAB [6], VINI [7], CABO
[8], Cabernet [9] in the United States; 4WARD [10] [11]
in Europe, AKARI [12], AsiaFI [13] in Asia and many
others. This shows that the current direction in designing
the Future Internet is going in favor of having multiple co-
existing architectures, instead of only one, where each
architecture is designed and customized to fit and satisfy a
specific type of network requirements rather than trying to
come up with one global architecture that fits all. That is
why Network Virtualization will play a vital role as it
helps diversifying the Future Internet into separate Virtual
Networks (VNets) that are isolated and can run different
architectures within.
III. WIRELESS NETWORK VIRTUALIZATION
Network virtualization will allow operators to share the
same physical infrastructure and have networks co-
existing in a flexible, dynamic manner utilizing the
available resources more efficiently. This implies that the
physical infrastructure needs to be virtualized into a
number of virtual resources being offered to the different
virtual networks. While virtualization for servers, routers
and wire line links has been extensively studied in the
literature [14] [15] [16] [17] [18] [19], the wireless part
has not yet received major consideration within todays
research.
Virtualization of wireless resources is a complex
challenge. First, the wireless resources at the Base Station
have to be shared and assigned to different virtual network
operators. This sharing needs to be fair. Fairness in
wireless systems can be defined differently: fairness in
terms of spectrum used, or power used, or a product of
these two or even fairness of Quality of Service (QoS)
delivered to end users. Furthermore, it is not sufficient to
look at the resources that are being shared, assigned or
scheduled at one base station, but also the interference
caused by the utilization of these resources need to be
considered as well. In this paper we limit our investigation
to only one base station and consider fair sharing of
spectrum by means of bandwidth that will be assigned to
virtual operators based on predefined contracts.
IV. MOTIVATION BEHIND MOBILE NETWORK
VIRTUALIZATION
The Mobile networks are one of the fastest growing
technologies that are influencing major aspects of our life.
The idea of being able to virtualize these mobile networks
and share their resources will lead to a more efficient
utilization of the scarce wireless resources. Additionally
network virtualization can reduce the amount of the
required base station equipment and thus reduce the
required energy to run wireless networks.
Network virtualization will also allow completely new
value chains. Smaller players can come into the market
and provide new services to their customers using a virtual
network. This also allows completely new future
networks, e.g. isolating one virtual network (e.g. banking
network) from a best effort Internet access network. In
addition to all of the previous, the idea of being able to
share the frequency resources among multiple operators is
very appealing. This gives operators the flexibility to
expand/shrink their networks and the air interface
resources they use, and this will lead to better overall
resource utilization and reduced energy consumption.
V. LONG TERM EVOLUTION (LTE)
As a case study for wireless virtualization we chose
LTE [20]. LTE is the latest evolution of 3GPP [21] mobile
phone standard. The LTE air interface is based on
Orthogonal Frequency Division Multiple Access
(OFDMA) in the downlink and Single Carrier FDMA
(SC-FDMA) in the uplink (which also supports the use of
multi-antenna technologies (MIMO)). Since LTE is one of
the most promising future mobile networks and because of
the nature of OFDMA, it is a very good candidate to be
considered for applying network virtualization being the
focus of this paper.

Figure 1. LTE network architecture
From the specification LTE is an all IP network that
supports a downlink data rate of more than 100 Mbps and
an uplink data rate of more than 50 Mbps. LTE system
can support a scalable bandwidth from 1.4 MHz up to 20
MHz enabling it to also operate in lower frequency
ranges. LTE supports a new network architecture based on
a cost efficient two nodes architecture, that consists of
Enhanced NodeB (eNodeB) and Access Gateway (AGW)
as can be seen in Figure 1.
Applying network virtualization to the LTE network
means to virtualize the infrastructure of the LTE system
(that is eNodeBs, routers, Ethernet links ) and let
multiple network operators create their own virtual
network depending on their requirements and goals, while
using a common infrastructure. The challenges of that are
mainly how to virtualize the physical infrastructure to
support such scenarios, and what kind of changes are
required to be introduced to the LTE system. We mainly
foresee two different types of virtualization processes: one
is virtualizing the physical nodes of the LTE system like
for example the eNodeBs, routers, Ethernet links; and the
second is virtualizing the air interface of the LTE system,
which is the focus of this paper.
A. LTE Air Interface Virtualization
In order to virtualize the LTE air interface, the eNodeB
has to be virtualized, since it is responsible for accessing
the radio channel and scheduling the air interface
resources. Virtualizing the eNodeB is similar to node
virtualization. In node virtualization a number of solutions
exist, where the physical resources of the virtual machine
(like the CPU, memory, I/O devices ) are shared
between multiple instances of virtual operating systems.
XEN [14] for example is a well known PC virtualization
solution that calls the entity that is responsible for
scheduling the physical resources a Hypervisor. Our
proposal follows the same principle, where a hypervisor is
added to the LTE eNodeB as shown in Figure 2.

Figure 2. Virtualized LTE eNodeB protocol stack
The LTE Hypervisor is responsible for virtualizing the
eNodeB into a number of virtual eNodeBs (each is used
by a different operator); the physical resources are
scheduled among the different virtual instances via the
hypervisor. In addition, the LTE hypervisor is also
responsible for scheduling the air interface resources (i.e.
OFDMA sub-carriers) between the virtual eNodeBs. In
this paper, we only focus on the latter functionality that is
the air interface scheduling; the first part is not the focus
of this paper, but similar solutions are commercially
available like for e.g. VANU MultiRAN [22] which is a
solution used to support multiple virtual base stations all
running on a single hardware platform.
B. LTE Virtualization Hypervisor
The Hypervisor collects information from the
individual virtual eNodeB stacks, like user channel
conditions, loads, priorities, QoS requirements and
information related to the contract of each of the virtual
operators. This information is used to schedule the air
interface resources between the virtual operators. These
air interface resources are actually the Physical Radio
Resource Blocks (PRB); this is the smallest unit that the
LTE Medium Access Control (MAC) scheduler can
allocate to a user. A PRB consists of 12 sub carriers as
well as 7 OFDMA symbols as can be seen in Figure 3.

Figure 3. LTE Downlink Physical Resources Structure
Scheduling the PRBs between the virtual eNodeBs
means splitting the spectrum between the different
eNodeBs of the different operators. The hypervisor can
make use of apriori knowledge (e.g. channel conditions,
operator contract, load ... etc.) to schedule the PRBs.
OFDMA scheduling has been extensively studied in the
literature [23] [24] [25] [26], but what is new here is the
scheduling of the spectrum among the different operators.
This is even more challenging because of the additional
degree of freedom that has been added. A number of
possibilities exist here, where the scheduling could be
based on different criterias: bandwidth, data rates, power,
interference, pre-defined contracts, channel conditions,
traffic load or a combination of them. At the end the
hypervisor has to convert these criteria into a number of
PRBs to be scheduled to each operator, but the challenge
is to make sure that the allocated PRBs would be fair and
enable the operators to satisfy their requirements.
In our paper, we concentrate on a contract based
hypervisor algorithm. In the algorithm the spectrum is
divided between the virtual operators based on predefined
contracts that the virtual operators have made with the
infrastructure provider. We mainly define four types of
contracts and these are:
a. Fixed guarantees: the operator requests a fixed
bandwidth that would be allocated all the time
whether it will be used or not
b. Dynamic guarantees: the operator requests a
guaranteed max bandwidth that would be allocated if
required; otherwise only the actual need would be
allocated. The operator might only pay based on the
used bandwidth which could save cost
c. Best effort (BE) with min guarantees: the operator
specifies a min guaranteed bandwidth which will be
allocated at all time; and a max value as an upper
bound. The allocation will be done in a BE manner.
d. Best effort with no guarantees: the operator would
only be allocated part of the bandwidth if the current
load permits i.e. in a pure BE manner
In order for the hypervisor to allocate the spectrum an
estimation of the current bandwidth need has to be
provided. Each operator will send back his bandwidth
estimations at frequent time intervals. The estimated
bandwidth is calculated as follows:
) 1 ( 1 ) ( n PRBsTTI n E n E (1)
Where E(n) is the averaged estimate count of PRBs
over n time interval that are either additionally required by
the operator or are not required that the operator wants to
give back; PRBsTTI(n) is the current estimate count of
PRBs at the nth TTI, this is calculated by summing the
PRBs that were additionally needed to schedule the un-
served users within this TTI minus the number of left
PRBs that were not used; n is the number of TTIs in the
hypervisor interval; is the averaging factor, it indicates
how much history is taken into the averaging function.
E(n) can take both positive and negative values,
because the operator would either need more PRBs or
would like to give some PRBs back that are not required.
The hypervisor would use this to calculate and allocate
each virtual operators spectrum, and this would be used
specifically for contract types b, c and d. For contract type
b, this would serve as the allocated bandwidth for that
operator, upper bounded by the contract max value. For
the BE contracts (contract c and d) the hypervisor will first
allocate the c operators with their min guaranteed value,
and whatever PRBs left at the end would be distributed
between the BE operators. The split is based on a fairness
factor that is calculated as follows:
Est Total n E FF
i i
_ / ) ( (2)
Where FFi is the fairness factor of operator i; E(n)i is
the PRBs estimate of operator i; and the Total_Est is the
total BE PRBs estimate over all BE operators, and is
calculated as follows:

s BEoperator
i
i
n E Est Total
#
1
_ (3)
The number of PRBs allocated for each BE virtual
operator is calculated as follows:
PRBs Left FF alloc PRBs
i i
_ int _ (4)
Where the Left_PRBs is the number of PRBs left for the
BE operators after allocating the guaranteed operators
VI. LTE SIMULATION MODEL
The LTE simulation model used in this paper was
developed using OPNET [27]. The model is designed and
implemented following the 3GPP specifications. The
focus of this work was not on the node or link
virtualization, but rather on the air interface virtualization
and how to schedule the air interface resources among the
virtual operators, no node/link virtualization was
simulated, instead a perfect node/link virtualization was
assumed. Figure 4 shows an example scenario.


Figure 4. Virtualized LTE model (example with 3 Virtual Operators)
VII. SIMULATION SCENARIOS AND CONFIGURATIONS
Two different scenarios are investigated within this
paper, one without virtualization which is called legacy
and one with virtualization which is called virtualized.
In the virtualized scenario, 4 different virtual network
operators are configured each with a different contract
configuration as follows:
1. Video streaming operator: configured with a fixed
guaranteed contract of 33 PRBs
2. VOIP operator: configured with a dynamic
guaranteed contract, with a max value of 33 PRBs
3. VOIP + BE Video on demand operator: configured
with the best effort with min guarantees contract,
with min and max value of 25 and 45 consecutively
4. Small VOIP operator: configured with BE and no
guarantees contract
As for the legacy scenario, only the first three operators
are configured, because there will be no chance for the 4
th

operator since the spectrum cannot be shared. Each of the
three operators will get 33 PRBs. The rest of the
simulation is configured according to Table 1 and Table 2.
Table 1: Simulation parameters
Parameter Assumption
Nr. of virtual operators 4 virtual operators each with one eNB
eNodeB coverage area Circular with one cell, R = 375 meters
Total Number of PRBs 99 (corresponds to about ~ 20 MHz)
Mobility model Random Way Point (RWP), Users are
initially distributed uniformly in the cell
Users speed 5 km/h
Number of active users VO1: 12 video users
VO2: 40 VOIP users
VO3: 16 VOIP + 16 video users
VO4: 3 VOIP users
Path loss model 128.1 + 37.6 log10(R) dB, R in km [28]
Slow Fading model Lognormal with zero mean value
Standard deviation = 8 dB
Correlation distance = 50 meters
Fast Fading model Jakes model
CQI reporting Ideal
Modulation schemes QPSK, 16 QAM, 64 QAM
Link-2-System interface Effective Exponential SINR mapping [29]
Hypervisor resolution 1 sec
Estimate 0.5
Simulation run time 1000 sec
Table 2: Traffic models configurations
VOIP traffic model
Silence/Talk Spurt length neg. exponential with 3 sec mean
Encoder scheme GSM EFR
Conversational envir. land phone quiet room
Call duration Uniform (1, 3 min)
Inter-repetition time neg. exponential with 90 sec mean
Video streaming traffic model
Incoming/Outgoing stream inter arrival time Const (0.01 sec)
Incoming/Outgoing stream frame size Const (80 Bytes)
VIII. SIMULATION RESULTS
In order to show the performance gains that could be
achieved from virtualizing the LTE air interface two
different scenarios are compared against each other. One
scenario is similar to today mobile network setup, where
three operators are configured each with his own
frequency band. The second scenario is the futuristic
approach, where the operators share the frequency band
dynamically based on predefined contracts. In addition, a
4
th
operator (small operator) is also configured.
Figure 5 shows the number of PRBs that each virtual
operator has been allocated over time. It can be noticed
that for the 1
st
operator the PRBs allocation is fixed to 33
PRBs, since it is using the fixed guaranteed contract. As
for the other three operators we can notice that the
allocated number of PRBs changes with time depending
on the traffic load and the contract of each operator.

Figure 5. Cell1 per virtual operator (VO) allocated bandwidth (or PRBs)

The average user air throughput for operator 1 users
can be seen in Figure 6. The result shows that the
operator has the same performance with and without
virtualization; this is because of the guaranteed fixed
allocation contract, which means the operator have the
same bandwidth in both scenarios. Figure 7 shows the
user average application end-2-end delay, and again the
operator has similar performance in both scenarios.


Figure 6. Operator1 DL average per user air interface throughput

Figure 7. Operator1 DL average per user application end-2-end delay

The average per user air interface throughput for
operator 2 can be seen in Figure 8. One can notice that the
users have similar performance in both scenarios. This is
also applicable for the average per user application end-2-
end delays shown in Figure 9. The results confirm that all
users are being served with no problem and hence no
additional buffering delay is being introduced.

Figure 8. Operator 2 DL average per user air interface throughput


Figure 9. Operator 2 DL average per user application end-2-end delay

The results showed that operator 2 has the same
performance in both scenarios. But, in the virtualized
case operator 2 is not wasting the air interface resources
and is only using the required number of PRBs to serve
his users as can be seen in Figure 10. This is a big
advantage because the operator will be able to cut cost
since the he will only pay based on the used resources.

Figure 10. Operator 2 DL used number of PRBs vs. time

Figure 11 shows the voip users air throughput, one can
notice that operator 3 has similar performance in both
scenarios. Figure 12 shows the video users air interface
throughput, what could be seen is a slightly better
performance in the virtualized scenario, this is because
operator 3 is using more resources compared to the
legacy scenario due to his BE contract. Figure 13 and
Figure 14 show the end-2-end delay results of operator 3
users. The voip users end-2-end results are similar in both
scenarios. As for the video users one can notice huge
delay values in the legacy scenario as compared to the
virtualized scenario. This is because in the latter
scenario operator 3 is using the resources left from
operator 2. The reason why the voip users are not affected
in the legacy scenario is because these users are served
with higher priority and the resources are enough to serve
them, but not enough to serve the video users.

Figure 11. Operator 3 DL average per voice user air throughput

Figure 12. Operator 3 DL average per video user air throughput


Figure 13. 3 DL average per voice user app. end-2-end delay

Figure 14. Operator 3 DL average per video user app. end-2-end delay

This is a big advantage for the virtualization scenario,
because operator 3 makes use of the available PRBs left
by the 2
nd
operator to serve his users. This is one of the
motivations to apply virtualization: being able to share the
spectrum more efficiently, have some multiplexing gain
and cut cost. One additional advantage is the ability to
serve small operators in a pure best effort manner with
whatever resources left rather than wasting them. Operator
4 results shown in Figure 15 confirms that.

Figure 15. Operator 4 DL user air throughput & app. end-2-end delay
IX. CONCLUSION AND OUTLOOK
The goal behind applying network virtualization in
LTE systems is the expectation that a better performance
can be achieved; the simulation results confirmed that.
Based on the contract configurations and the traffic load
of each virtual operator the air interface resources are
shared among the operators. In this way, the overall
resource utilization is enhanced and the performance of
both network and end-user is better. Although the
simulation results are quite specific, the basic findings are
representative and show the advantages of applying
network virtualization to the LTE system. Both operator 2
and 3 benefited from virtualization mainly by being able
to cut costs and providing better performance for the
users. The paper also showed the possibility of opening
the market to new players (small operators) that can serve
a specific role and have small numbers of users.
This work is only a starting point, and more issues still
need to be addressed: interference coordination among
multiple virtual operators, signaling overhead due to the
hypervisor, defining diverse hypervisor scheduling
disciplines based on enhanced contracts. Nevertheless,
with LTE wireless virtualization, operators can expect not
only lower investments for flexible network deployments
but also lower costs for management and maintenance.
ACKNOWLEDGMENT
We would like to thank all members of the 4WARD
project, especially the virtualization work package.
REFERENCES
[1] The Internet history:
http://www.lincoln.edu/math/rmyrick/ComputerNetworks/InetRef
erence/57.htm
[2] www.internetworldstats.com Copyright 2001 - 2009, Miniwatts
Marketing Group. All rights reserved worldwide.
[3] ITU Internet Reports 2005: The Internet of Things,
http://www.itu.int/osg/spu/publications/internetofthings/
[4] GENI Planning Group; "GENI: Conceptual Design, Project
Execution Plan". GENI Design Document 06-07, January 2006.
http://www.geni.net/GDD/GDD-06-07.pdf
[5] GENI: Global Environment for Network Innovations.
http://www.geni.net/
[6] A. Bavier, M. Bowman, D. Culler, B. Chun, S. Karlin, S. Muir, L.
Peterson, T. Roscoe, T. Spalink, and M. Wawrzoniak; "Operating
System Support for Planetary-Scale Network Services" Mar. 2004.
[7] Andy Bavier, Nick Feamster, Mark Huang, Larry Peterson, and
Jennifer Rexford; "In VINI Veritas: Realistic and Controlled
Network Experimentation". ACM SIGCOMM'06, Sept. 2006.
[8] N. Feamster, L. Gao, J. Rexford; "How to lease the Internet in
your spare time". ACM SIGCOMM Computer Communications
Review, Vol. 37, No.1, January 2007.
[9] Yaping Zhu, Rui Zhang-Shen, Sampath Rangarajan, and Jennifer
Rexford, "Cabernet: Connectivity architecture for better network
services"; Workshop on Rearchitecting the Internet, Dec. 2008.
[10] Niebert N, Baucke S, El-Khayat I, et al.; "The way 4WARD to the
creation of a Future Internet". ICT Mobile Summit, Stockholm,
June 2008.
[11] 4WARD project page at www.4ward-project.eu
[12] AKARI Architecture Conceptual Design for New Generation
Network [translated version 1.1]
(http://akari-project.nict.go.jp/eng/concept-
design/AKARI_fulltext_e_translated_version_1_1.pdf)
[13] Asia Future Internet (AsiaFI) (www.asiafi.net)
[14] David E. Williams and Juan Garcia; "Virtualization with Xen:
Including Xenenterprise, Xenserver, and Xenexpress". Syngress
Publishing, Inc... ISBN-13: 9781597491679. May 2007.
[15] S. Bhatia, M. Motiwala, W. Muhlbauer, V. Valancius, A. Bavier,
N. Feamster, L. Peterson, and J. Rexford; "Hosting virtual
networks on commodity hardware,''. Georgia Tech. University.,
Tech. Rep. GT-CS-07-10, January 2008.
[16] E. Kohler, R. Morris, B. Chen, J. Jahnotti, and M. F. Kasshoek.
"The Click Modular Router". ACM Transaction on Computer
Systems, vol. 18, no. 3, 2000, pp. 263297.
[17] VROUT: http://nrg.cs.ucl.ac.uk/vrouter
[18] VMware Server, http://www.vmware.com/products/server/
[19] Cisco VN-Link: Virtualization-Aware Networking, white paper.
http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns2
24/ns892/ns894/white_paper_c11-
525307_ps9902_Products_White_Paper.html
[20] LTE Air Interface White Paper by Motorola at
http://business.motorola.com/experiencelte/pdf/LTETechnicalOve
rview.pdf.
[21] 3rd Generation Partnership Project at www.3gpp.org.
[22] VANU website: http://www.vanu.com/solutions/multiran.html
[23] R. Agrawal, R. Berry, Huang Jianwei, V. Subramanian; "Optimal
Scheduling for OFDMA Systems. Signals, Systems and
Computers, 2006". ACSSC '06. Fortieth Asilomar Conference on.
[24] Rajiv Agarwal, Vinay Majjigi, Rath Vannithamby, John Cioffi;
"Efficient Scheduling for Heterogeneous Services in OFDMA
Downlink". IEEE Globecom 2007, Washington D.C.
[25] M. Einhaus, O.Klein; "Performance Evaluation of a Basic
OFDMA Scheduling Algorithm for Packet Data Transmissions".
ISCC 2006, Cagliary, Italy.
[26] M.l Einhaus, O. Klein and B. Walke; "Comparison of OFDMA
Resource Scheduling Strategies with Fair Allocation of Capacity".
2008 5th IEEE Consumer Communications & Networking
Conference (CCNC 2008), Las Vegas, NV, January 2008.
[27] OPNET website: http://www.opnet.com
[28] M. Anas, F.D Calabrese, P.E Mogensen, C. Rosa, K.I Pedersen;
"Performance Evaluation of Received Signal Strength Based Hard
Handover for UTRAN LTE". Vehicular Technology Conference,
2007. VTC2007-Spring IEEE 65
th
.
[29] Erik Westman; "Calibration and Evaluation of the Exponential
Effective SINR Mapping (EESM) in 802.16". Master Thesis.
Stockholm, Sweden 2006.