See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/261087151

Analysis of CAPEX and OPEX benefits of wireless

access virtualization

Conference Paper June 2013

DOI: 10.1109/ICCW.2013.6649273

CITATIONS READS

13 264

3 authors:

M. Moshiur Rahman Charles Despins

cole de Technologie Suprieure cole de Technologie Suprieure
10 PUBLICATIONS 77 CITATIONS 108 PUBLICATIONS 1,038 CITATIONS

SEE PROFILE SEE PROFILE

Sofine Affes
Institut National de la Recherche Scientifique
376 PUBLICATIONS 3,386 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Advanced Signal Processing and Networking Techniques for Cost-Effective Ultra-Dense 5G Networks
View project

NSERC CREATE Training Program in Pervasive and Smart Wireless Applications for the Digital Economy
(PERSWADE) View project

All content following this page was uploaded by M. Moshiur Rahman on 31 March 2015.

The user has requested enhancement of the downloaded file.

 IEEE International Conference on Communications 2013: IEEE ICC'13 - Workshop on Energy Efficiency in Wireless Networks &
Wireless Networks for Energy Efficiency (E2Nets)

Analysis of CAPEX and OPEX Benefits of Wireless

Access Virtualization

M. M. Rahman Charles Despins Sofiene Affes

Dept. of Electrical Engineering Prompt Inc. AND INRS-EMT
ETS, University of Quebec ETS, University of Quebec University of Quebec
Montreal, Quebec, Canada Email: [email protected] Email: [email protected]
Email: [email protected]

AbstractVirtualization of wireless access networks dictates The information and communications technologies (ICT) in-
a new direction in the research of cost effective and energy dustry is, in fact, responsible for 2% to 3% [9] of the worlds
efficient network modelling that the telecommunications industry total carbon emissions and this carbon footprint is doubling
is seeking. Different research initiatives are dealing with virtu- every four years. At the same time, as a result of its potential
alization of wireless resources (e.g. nodes, wireless access cards, impact in all sectors of human activity, ICT is accepted to
wireless spectrum) and new ideas are being put forth to leverage
be one of the key enablers of a low carbon economy [10].
cloud computing in the wireless telecom domain. In this paper,
we analyze the CAPEX and OPEX of our proposed virtualized As such, energy efficient operation of cellular networks [11]
wireless access network framework and show its benefits with is appreciated from both the operators power expenditures
regards to traditional network architectures. We also examine and environmental conservation perspectives. If virtualized
the trade-offs associated in achieving such benefits. resources are located in data centers powered by renewable
energy sources, the networks can also contribute to reduce
the industrys carbon footprint as well as generate eventually
I. I NTRODUCTION further OPEX savings in jurisdictions with a price on carbon
emissions.
The process of combining hardware and software resources
Wireless access networks account for up to 60% to 80%
into a single software based entity is at the core of the notion
of the telecoms energy consumption [12]. It is therefore
of network virtualization. Computer researchers have done ex-
imperative to devise techniques that target energy efficient
tensive investigations on virtualization in the application layer
operation of telecom networks and at the same time reduce
[1] and the outcome is visible in todays network architecture
carbon emissions. Leveraging cloud computing and virtual
in the form of VLAN, VPN, overlay networks, etc.
networking can thus be significant drivers of so-called Green
Future network deployment strategies are driven by the omni
Communications [13], [14] in the telecom domain.
presence of wireless links. The advent of the smart-phone
The rest of the paper is organized as follows: Section II
culture has brought about a drastic change in the traditional
succinctly describes the proposed frameworks for the virtual
voice dominated cellular networks. Consumers any time any
wireless access network (VWAN). CAPEX and OPEX analysis
where demand for high speed data traffic is evolving the
of the VWAN appears in Sections III-A and III-B respec-
wired broadband internet towards the wireless internet. The
tively, including comparisons with corresponding values of a
conundrum situation of providing high capacity wireless ser-
traditional network. Finally, Section IV concludes the paper
vices at a rather decreasing cost is driving major telecom
with a brief discussion on the CAPEX and OPEX benefits of
operators to resort to virtualization of the networks [2]. While
virtualization and the associated trade-offs.
virtualization of the core network resources such as routers,
servers, etc are either well understood or the subject of current
II. V IRTUAL WIRELESS ACCESS NETWORK (VWAN)
investigations, virtualization of radio access which accounts for
40% (while the core network is responsible for 10%-30%) [3] Cellular access networks encompass base station con-
of the total operational cost of a cellular network is receiving trollers (BSCs) (in LTE the BSC is incorporated within the
very little attention. Wireless virtualization also promises to eNodeB), base stations (BSs), and the wireless medium be-
alleviate the ossification problem of costly radio spectrum by tween the BSs and the UEs. The virtualization concept stems
ensuring its efficient use [4] [5] [6]. Major cellular vendors from the efficient use of network resources and leveraging the
and operators have notably advocated wireless virtualization use of distributed computing sources. In this respect, wireless
for cost effective and energy efficient service provisioning. network virtualization should be an intelligent amalgamation
The cost structure of wireless networks has been studied for of wireless cloud computing technology; efficient spectrum
both capacity and coverage limited cases [7] [8] [3]. Cellular sharing techniques in time, frequency, space, code domain or
network operation demands extensive power consumption. any combination of these; shared use of hardware resources,
This huge power demand not only is responsible for greater etc.
OPEX for the operators but also contributes to increased The absence of a conceptual definition of wireless access
carbon emissions as many countries with the largest wireless virtualization in the existing literature has prompted us to
usage (in absolute terms) are also dependent on fossil-based propose three different frameworks to implement this concept.
energy sources to power these networks. They are: local, remote and hybrid virtualization frameworks.

978-1-4673-5753-1/13/$31.00 2013 IEEE 436

 number of users (UEs) served by all the operators are kept
the same for both the traditional and the virtualized networks.
The number of slices in a SuperBS (SBS) corresponds to
the number of operators in the traditional network. It is also
assumed that the allocated spectrum per virtual operator [16]
(in a virtualized network) is the same as that of an operator
(in traditional network).

A. CAPEX analysis of VWAN

Let us assume that, in the area A, nop operators are serving
their customers. The number of BSs required to cover an area
of A square unit depend on the following parameters:
Maximum coverage radius of a BS, Rmax .
Fig. 1. Local virtualization framework Total number of users in the area A, NU Et
Assuming a coverage limited case, the total number of required
In a local virtualization approach (Fig-1), the individual PHY BSs/operator is
BSs are sliced (slicing is the process of allocating a coherent A
NBSO = 2
(1)
subset of physical resources of a typical PHY BS to the Rmax
created virtual BSs) vertically or horizontally to create multiple Let the user density per unit area be . We assume that user
virtual BSs(VBSs). The hypervisor, a supervising entity, is density is the same per operator for both the traditional and
in charge of synchronous allocation of physical resources virtualized networks, i.e., op = sl = , where, op and sl
between different virtual instances. are the user density/BS (/slice) for the respective networks.
A data center approach is adopted for remote virualization, The total number of users per operator is
where the radio equipment of the BS is segregated from the
baseband processing unit. The processing of the baseband NU EO = A
signal is centralized either in conventional telecom network 2 (2)
= NBSO Rmax
equipments or in software (VBS pools) in standard IT platform
[15]. Remote radio heads (RRHs) serve to interact with the 1) For traditional network: The traditional network dimen-
user equipments (UEs). sioning parameters are:
Hybrid virtualization is a combination of the local and re-
mote virtualization approaches. Here, baseband processing Number of operators, nop
is distributed among VBS pools (the data center) and the
Total number of users per operator in peak period,
local enhanced remote radio head (E-RRH) which will have
NU EO
augmented capacity for processing delay sensitive data (e.g.
voice, live video traffic); this can alleviate the QoS problem Cost per cell site, ccs
for the delay sensitive traffic of the remote virtualization
approach. Analysing the CAPEX and OPEX of all the three Cost per BS, cbs
frameworks are beyond the scope of this paper. Hence, our The total cost for cell site construction by all the operators is
analysis concentrates on the local virtualization approach, with
subsequent works devoted to the other two frameworks. ccst = nop NBSO ccs (3)
The cost of all the BSs operated by the operators is
III. CAPEX AND OPEX ANALYSIS OF VWAN
cbst = nop NBSO cbs (4)
We have analysed the capital expenditure (CAPEX) and
operational expenditure (OPEX) trends of a locally virtualized The total infrastructure cost is
network. CAPEX refers to the cell-site construction cost as cinf ra = ccst + cbst
well as the cost of the radio BSs needed to cover a certain NU Et (5)
geographical area. OPEX on the other hand, is highly influ- = 2
(ccs + cbs )
Rmax
enced by the power consumption of the network. It is worth
noting that, while in the CAPEX analysis, the parameter of The total number of users in the network is then
concern was relative infrastructure cost per user, in the case NU Et = nop NU EO (6)
of OPEX, the parameter of interest was power per bit. The
cost of site rent is not considered as it would merely add a Now, the infrastructure cost per user becomes
constant value that is same for both the traditional and the ccs + cbs
virtualized networks (at least for local virtualization). In the cinf rau = 2
(7)
process of the investigation, we also compare the results with Rmax
the corresponding values for a traditional (LTE) network. Here, it is interesting to note that the infrastructure cost per
In the scenario analyzed, a certain geographical area A is user is independent of the total number of operators in the
considered with a certain number of operators. The total region.

437
 2) For locally virtualized network: A virtual network is set
up in the same area A, consisting of SuperBSs (SBSs) serving
the same number of users as in the traditional network. Those
operators will now be called virtual operators (VOs) in this
scenario. The number of slices (virtual base stations (VBSs)) of
a SBS is the same as the number of operators in the traditional
case. The parameters for the locally virtualized network are as
follows:

Number of slices (VBSs) per SuperBS, nsl

User density per area, SBS
Number of SuperBS in area A, NSBS
Cost per cell site of a SBS, ccssbs (same as the
traditional case)
Fig. 2. Relative infrastructure cost vs user density (R=2 unit)
Cost of a SBS unit, csbs = cbs [1 + 0.2 (nsl 1)].
We assume that the cost of a SBS increases linearly TABLE I. R ELATIVE INFRASTRUCTURE COST REDUCTION
(with a slope of 20%) with the number of VBSs it
contains. #slices in the SBS Rel cost reduction (%)
2 46
4 68.33
Let us assume each SBS has the same maximum coverage 6 76.7
radius of Rmax as the traditional BS. Now, the total number 8 80.83
of SBS required to cover the area A is

A
NSBS = 2
(8) B. OPEX analysis of a VWAN
Rmax
Channel capacity with modified Shannons formula [17] is
The total number of users in area A is SN R
R = w wef f log2 (1 + ) (14)
NU Et = SBS A SN Ref f
2 (9)
= nsl Rmax NSBS Here, w is the allocated BW/UE, SN Ref f is the SNR effi-
ciency, for our analysis we set it to 1. The bandwidth efficiency
The total cost for cell site construction for the SBS network is is

ccssbst = NSBS ccssbs (10) wef f = ef fac ef fcp ef fpace ef fso = 0.56 (15)
where, ef fac = 0.9, due to adjacent channel leakage and the
The total cost for the SBSs of the network is practical filter issue;
ef fcp = 0.93, due to cyclic prefix;
csbst = NSBS csbs (11) ef fpace = 0.94, due to pilot assisted channel estimation;
ef fso = 0.715, due to signalling overhead.
Now, the infrastructure cost for the SBS network becomes
1) Traditional network power consumption model: The
cinf rasbs = ccssbst + csbst power consumption of different elements of a BS are:
NU Et (12) Transceiver power, Ptrans
= 2
(ccssbs + csbs )
nsl Rmax
Rectifier power, Prect
Finally, the infrastructure cost per user is Digital signal processor power, PDSP
(ccssbs + csbs ) Power amplifier power, PP A
cinf raSBSu = 2
(13)
nsl Rmax MW transmission power, Pmw
We can see that the infrastructure cost per user decreases Air cooling power, Pair
with the number of slices in the SuperBS. Fig-2 shows how
Power consumption for different components of a BS appears
the relative infrastructure cost varies with the user density
in Table-II [18]. Each antenna is associated with a transmission
for both the traditional network (TN) and locally virtualized
chain that consists of the transceiver, rectifier, digital signal
(LV) network. While the relative infrastructure cost for a TN
processor and the power amplifier. So, the power consumption
remains constant for a varying number of operators in area A,
of a BS is
it decreases with the increase in the number of slices (VOs) in
the LV network. The analytical results shown in table-I display PBS = na (Ptrans + Prect + PP A + PDSP ) + Pair + Pmw
the cost reduction variation with the number of slices. (16)

438
TABLE II. P OWER CONSUMPTION FOR DIFFERENT PARTS OF THE BS

BS Parts Power Consumption (W)

Digital signal processor 100
Power amplifier (SISO) 156
Power amplifier (MIMO) 10.4
Transceiver 100
Rectifier 100
Air conditioner 225
Microwave link 80

where, na is the number of antenna per BS. If there are NBSO

per operator, then the total power consumption per operator is

PBSO = NBSO PBS (17)

If nop operators are operating in the area A, then the total

power consumption in the network is Fig. 3. Total power consumption vs #slices

PBStot = nop PBSO (18)

The user density per unit area is , hence, the number of users
2
per BS is, Rmax . And from equation-(2) the number of
users per operator is
2
NU EO = NBSO Rmax

So, the total number of users served by nop operators in service

area A is
2
NU Et = nop NU EO = nop NBSO Rmax (19)

The capacity of a UE is R Mbps. The total capacity of the

whole network is

RBStot = NU Et R (20)
Fig. 4. Total power consumption vs #slices (shared antenna case)
Now, the consumed power per bit becomes
PBStot Now, the consumed power per bit is
PbitBS =
RBStot
(21) PSBStot
PBS PbitSBS =
= 2 RSBStot
Rmax wwef f log2 (1 + SN R)
PSBS
=
which is independent of the number of operators in area A. 2
nsl Rmax wwef f log2 (1 + SN R)
(25)
2) Locally virtualized network power consumption model:
Let us assume that the number of SBSs operating in area A is So, it is seen that the power consumption per bit decreases with
NSBS and the number of slices in each SBS is nsl . An SBS the increase of slices in the SBS. Total power consumption in a
will have more processing capabilities than a regular BS and locally virtualized network is also less than that of a traditional
hence, will require more power for its cooling system as well network while serving the same number of UEs, as is evident
as the MW (Microwave) link. in Fig-3. We can see that the power consumption of a LV
We model the power consumption on cooling and MW trans- network is less than the traditional network as virtualization is
mission to increase linearly with the number of slices in the implemented (2 or more slices).
SBS; we assume it to increase by 20% with each additional 3) Super-BS with antenna sharing by the VOs: We consider
slice. So, the power consumption would have the form the case where each antenna is shared by different VBSs
PairSBS = Pair [1 + 0.2 (nsl 1)] (22) (slices). In this configuration, each slice will have its own DSP
but the antennas as well as the RF chains will be shared by
the existing VBSs. The RF power of a traditional BS would
PmwSBS = Pmw [1 + 0.2 (nsl 1)] (23)
be
The power consumption of a SBS is Prf = Ptrans + Prect + PP A (26)
For a SBS, the RF power will have the following form
PSBS = nsl na (Ptrans +Prect +PP A +PDSP )+PairSBS +PmwSBS
(24) Prf SBS = Prf [1 + 0.2 (nsl 1)] (27)

439
 TABLE III. P OWER SAVING PER BIT FOR DIFFERENT NUMBER OF
SLICES IN THE SBS
R EFERENCES
[1] N. Chowdhury and R. Boutaba, Network virtualization: state of the art
# of slices Psav %(No antenna sharing) Psav % (Antenna sharing) and research challenges, IEEE Communication Magazine.,vol. 47, no.
2 16.0 35.0 7, pp. 20-26, July 2009.
3 21.4 46.3 [2] M. Chiosi, D. Clarke, P. Wilis, et al., Network Functions Virtualisation:
4 24.0 52.12
5 25.7 55.6
An Introduction, Benefits, Enablers, Challenges & Call for Action, SDN
6 27.0 58.0 and OpenFlow World Congress, Darmstadt, Germany, October 2012.
[3] K. Johansson, A. Furuskar, P. Karlsson and J. Zander, Relation Between
Base Station Characteristics and Cost Structure in Cellular Systems,
Proc. IEEE PIMRC, October 2004.
[4] L. Zhao, M. Li, Y. Zaki, A. Timm-Giel and C. Gorg, LTE Virtualization:
So, the power consumption of a SBS is from Theoretical Gain to Practical Solution, Proc. ITC 2011.
[5] L. Zhao et al. Virtualization approach: Evaluation and Integration,
PSBS = na Prf SBS + nsl PDSP + PairSBS + PmwSBS 4WARD (Architecture and Design for the Future Internet) report FP7-
(28) ICT-2007-1-216041-4WARD/D-3.2.1, June 2010.
Now, the consumed power per bit becomes [6] S. Perez, J. Cabero, and E. Miguel, Virtualization of the wireless
medium: A simulation-based study, IEEE Spring Vehicular Technology
PSBStot Conference, April 2009.
PbitSBS = (29) [7] J. Zander, On The Cost Structure of Future Wideband Wireless Access,
NSBS nsl R IEEE VTC, May 1997.
Antenna sharing gives better total power performance than the [8] J. Zander, Affordable Multiservice Wireless Networks - Research Chal-
lenges for the next decade, IEEE PIMRC, September 2002.
non-sharing case, as can be seen from Fig-4. Table-III shows
[9] Smart 2020, Enabling the Low-Carbon Economy in the Information Age,
the comparison of power saving per bit for both the no antenna The Climate Group, London, U.K., 2008, www.smart2020.org
sharing and the antenna sharing cases. [10] L. Neves, J. Krajewski, P. Jung, M. Bockemuehl, GeSI SMARTer 2020:
The Role of ICT in Driving a Sustainable Future, gesi.org.
[11] E. Oh, B. Krishnamachari, X. Liu and Z. Niu, Towards dynamic energy-
IV. C ONCLUSION efficient operation of cellular network infrastructure, IEEE Communica-
tions Magazine, vol.49, no.6, June 2011.
The emerging prospect of wireless virtualization with [12] C. Han, T. Harrold and et al.Green radio: radio techniques to enable
CAPEX and OPEX benefits has prompted investigation in energy efficient wireless networks, IEEE Communications Magazine,
this area. A significant infrastructure cost reduction is possible vol.49, no.6, June 2011.
from SBS virtualization and the gain grows with an increasing [13] C. Despins, F. Labeau, R. Labelle, T. Le Ngoc, J. McNeil, A.
Leon-Garcia, M. Cheriet, O. Cherkaoui, Y. Lemieux, M. Lemay, C.
number of slices. But it should be noted that each additional Thibeault, F. Gagnon, Leveraging green communications for carbon
slice will add to the complexity level of the required hardware emission reductions: techniques, testbeds and emerging Carbon Footprint
implementation; the existing hardware technology can also set Standards,IEEE Communication Magazine, August 2011.
a hard limit on the achievable cost gain in a high capacity SBS. [14] K. Nguyen, M. Lemay, B. St. Arnaud and M. Cheriet, Convergence of
Virtualization will require extensive processing capabilities on Cloud Computing and Network Virtualization: towards a Zero-Carbon
the part of the SBS. Hence, highly efficient multi-core and Network, IEEE Internet Computing, Issue 99, 2011.
multi-thread processors are required for baseband processing. [15] Y.Lin, L. Shao, Z.Zhu, Q Wang, R.K. Sabhikhi,Wireless network cloud:
Architecture and system requirements,IBM Journal of Research and
Adept design of a hypervisor is also critical for ensuring fair Development, 2010.
resource sharing among residing VBSs in a SBS. [16] B. Ishibashi, N. Bouabdallah and R. Boutaba, QoS Capacity of Virtual
Wireless Networks, Computer Networks Journal. Elsevier, Vol. 55(7), pp.
Power consumption is a major contributor to the OPEX in a 15921613, April 2011.
cellular network. Hence, a power saving of 27% (for 6 slices in [17] P. Mogensen, W. Na, et al., LTE Capacity compared to the Shannon
no antenna sharing case) to 58% (for 6 slices in antenna sharing Bound, IEEE VTC, April 2007.
case) is very compelling in this regard. For SBS cost and power [18] M. Deruyck, E. Tanghe, W. Joseph, L. Martens, Modelling and op-
consumption modelling, a linear increase with respect to the timization of power consumption in wireless access networks,Elsevier
traditional BS was assumed. While it gives an intuitive figure Computer Communications, April 2011.
of the corresponding quantities, the absolute figures might be
different but we do not expect it to change the above analytical
behaviour significantly.
From the perspective of abating GHG emissions, wireless
virtualization not only promises to lower power consumption
but centralizing baseband processing in wireless data centres
(for remote and hybrid virtualization) can stimulate the use of
green energy by powering those sites with air, water or solar
sources.
As future work, we will carry out detailed analysis of the
remote and hybrid virtualization models which is expected to
provide improved CAPEX and OPEX performance as well
as an edge in mitigating inter-cell interference and better
management of cellular handoffs.

View publication stats

440