[08]an overview of the femtocell concept_2

◆ An Overview of the Femtocell ConceptHolger Claussen, Lester T. W. Ho, and Louis G. Samuel

The femtocell concept aims to combine fixed-line broadband access withcellular telephony using the deployment of ultra-low-cost, low-power thirdgeneration (3G) base stations in the subscribers’ homes or premises. Itenables operators to address new markets and introduce new high-speedservices and disruptive pricing strategies to capture wireline voice minutesand to grow revenues. One of the main design challenges of the femtocell isthat the hierarchical architecture and manual cell planning processes used inmacrocell networks do not scale to support millions of femtocells. In thispaper, a user-deployed femtocell solution based on the base station router(BSR) flat Internet Protocol (IP) cellular architecture is presented thataddresses these problems, and several aspects of the proposed solution arediscussed. The overall concept and key requirements are presented in detail.The auto-configuration and self-optimization process from purchase by theend user to the integration into an existing macrocellular network isdescribed. Then the theoretical performance of a co-channel femtocelldeployment is analyzed and its impact on the macrocell underlay is assessed.Finally, a financial analysis of a femtocellular home base station deploymentin a macrocellular network is presented. It is shown that in urban areas, thedeployment of publicly accessible home base stations with slightly increasedcoverage can significantly reduce the operator’s annual network costs (up to70 percent in the investigated scenario) compared to a pure macrocellularnetwork. © 2008 Alcatel-Lucent.

where specific cellular data systems have been

deployed for some time.

In the past, cellular wireless operators have been

content with focusing on wide area voice coverage.

Reasons for this have been the lack of competition, a

rapidly increasing market, and perhaps regulatory

issues where, for example, the initial grants of a

license usually had some stipulations on coverage, as

a percentage of either landmass or population [20].

However, the means to sustain a profitable cellular

business can no longer rest on voice usage alone [13].

IntroductionCellular wireless systems have been undeniably

successful with over 1.4 billion subscribers [14]. The

initial cellular systems were designed for a single

application: voice. This is evidenced by early work [5]

in this area. However, cellular operators are experi-

encing an increasing commoditization of voice serv-

ices due to competition. In spite of the increasing

number of voice subscribers, the average revenue per

user (ARPU) for voice in saturated markets is declin-

ing [13]. Conversely, [13] also indicates that cellular

wireless data usage is increasing ARPU in markets

Bell Labs Technical Journal 13(1), 221–246 (2008) © 2008 Alcatel-Lucent. Published by Wiley Periodicals, Inc.Published online in Wiley InterScience (www.interscience.wiley.com). • 10.1002/bltj.20292

222 Bell Labs Technical Journal DOI: 10.1002/bltj

This has led to a search for the new “killer application”

for wireless access, which is taking operators

inevitably in the direction of packet data and Internet

Protocol (IP) based systems. The fastest method for

finding such an application is to do so “organically” by

enabling free access to the Internet, and to allow a

competitive “jungle” of service providers to compete.

For example, the business models behind NTT

DoCoMo’s very successful i-mode* system in Japan

[23] employed such a technique. Bounding the user

to a known subset of applications is counter to natu-

ral selection, in which the evolution of the system

will naturally find its own means and media. Allowing

access to the Internet would be one way of enabling

a fertile evolution. However, going down this path

requires solutions to new problems.

Data traffic has a different set of characteristics

that enables its flexibility and hence enhances its

appeal to future cellular operators. However, this

comes at a price. Data traffic tends to be very bursty in

nature and, as evidenced by the increase in local area

network (LAN) speeds, requires more bandwidth than

traditional voice service. Coupled to this is the user’s

expectation of receiving the same quality of service

on the move as that on a wireline IP connection. All

of these characteristics indicate that future cellular

wireless systems will have to be thought of and

designed in a different way.

There are many ways in which network operators

can address the need for more bandwidth on the air

interface. One approach is to enhance the signal pro-

cessing capability of the transmitters and receivers to

a point where the channel efficiency is close to the

Shannon bound. Other options are the application of

smart antenna or multiple input, multiple output

(MIMO) technologies and high-order modulations [11],

or improving access and scheduling mechanisms

[2]. Alternatively, simpler, more pragmatic approaches

can be used, such as increasing the sectorization of

the cell or reducing cell size. It has often been the case

that the pragmatic and simplest of solutions are the

ones that are chosen in evolution paths.

A different way of thinking provides the conclu-

sion that there is a need for smaller cells. Currently,

cellular systems are designed and planned with the

concept of shared resources. In this respect, a user in

a wireless cell shares radio resources with all those

users that cohabit the cell. This invariably means that

the amount of individual bandwidth that a user can

expect to have is limited; in a sense this can be termed

“public bandwidth.” While this may have been

acceptable for voice service, the same does not hold

true for data transfer in larger cells. For data services,

the user expects the same availability of bandwidth

and low delay as those of a wired connection to the

Internet. The user, in effect, is expecting “private

Panel 1. Abbreviations, Acronyms, and Terms

3G—Third generationARPU—Annual revenue per userAWGN—Additive white Gaussian noiseBS—Base stationBSR—Base station routerCAPEX—Capital expenditureCDF—Cumulative distribution functiondB—DecibelDSL—Digital subscriber lineGGSN—Gateway GPRS support nodeGPRS—General packet radio serviceHSDPA—High speed downlink packet accessHUS—Home unit serverIP—Internet ProtocolLAN—Local area network

MIMO—Multiple input, multiple outputOPEX—Operational expenditureQAM—Quadrature amplitude modulationQoS—Quality of serviceRF—Radio frequencyRNC—Radio network controllerSG&A—Sales, general and administrative costsSGSN—Serving GPRS support nodeSIM—Subscriber identity moduleSINR—Signal-to-interference-plus-noise ratioTV—TelevisionUE—User equipmentUMTS—Universal Mobile Telecommunications

System

DOI: 10.1002/bltj Bell Labs Technical Journal 223

bandwidth.” The move to smaller cells operating in

licensed spectrum but which are privately owned is a

step toward meeting this expectation.

However, once past this expectation, wireless

operators change their focus to the profitable areas of

their business. These areas are hot zones, in-building,

home, and enterprise areas. Naturally one can think

of simplifying the challenge of these kinds of picocell

and femtocell deployments by deploying into sec-

ondary frequency bands. This would, however, result

in a lower spectral efficiency per area compared to a

co-channel deployment, and a secondary frequency

band might not always be available. A more efficient

approach is co-channel deployment of macrocells and

femtocells on the same frequency band, but this is

more challenging due to the resulting interference

issues. Therefore, the two main questions that need to

be addressed in the context of small cells are inter-

ference mitigation and the economic feasibility of

small cell deployment. This paper addresses both the

technical and financial impacts of adapting small cells

for targeted market deployments and extends previ-

ous work in these areas [8, 10, 15]. It investigates

how these deployments can be made scalable and

explores a broader set of operator coverage scenarios.

This is done by first solving the problems posed by

co-channel deployments in terms of performance

impact on existing macrocellular coverage, and then

extending these techniques to permit the automatic

configuration of the deployed cells. Finally, it addresses

the financial benefits that these techniques bring to a

wider set of deployment scenarios.

The paper begins with an overview of the BSR

flat-IP architecture and discusses its advantages for

femtocell deployments. Key requirements for femto-

cell deployments are discussed next, including various

aspects of plug-and-play deployment, and the require-

ment of allowing public access for co-channel

deployments, i.e., when only one frequency is shared

between macrocells and femtocells. In the section fol-

lowing, the technical feasibility of co-channel deploy-

ment of femtocells is investigated. The achievable

performance of co-channel femtocells and their inter-

ference impact on an existing macrocellular network

on the same frequency are discussed based on simu-

lation results. In addition, the impact of femtocell

deployments on the call drop probability resulting

from an increased number of handovers is investi-

gated for different pilot power auto-configuration

schemes. A financial analysis follows for a case where

femtocell coverage is slightly increased to picocellular

coverage—in combination with public access, this can

effectively complement macrocellular coverage. The

paper concludes with a summary of the main results.

The BSR Flat-IP ArchitectureUbiquitous anywhere, anytime high-speed net-

work access has been an ideal of communications

networks for some time. Moreover, it is a paramount

need in an emerging information society, and the

development of cellular networks was the first step

in providing this ubiquity. However, it would be

prohibitively expensive in terms of both opera-

tional expenditures (OPEX) and capital expenditures

(CAPEX) for cellular wireless systems to provide ubiq-

uity with current technologies. Moreover, at this time,

cellular architectures also do not permit the neces-

sary architectural ubiquity. In the late 1990s and early

part of this decade, it became clear that in order to

achieve ubiquity, a cellular system would require

three key components/characteristics: scalability, the

ability to provide a cost effective architecture that

could sustain cellular mobility [12], and a mechanism

that would place such an architecture into an auto-

nomic paradigm [17]. These ideas are embodied in

the base station router (BSR) concept, which is an

enabling technology for a broader thematic area of

flat cellular architectures [4]. The BSR combines all

functions of a radio access network and core network

in a single network element (i.e., NodeB, radio net-

work controller [RNC], serving GPRS support node

[SGSN], gateway GPRS support node [GGSN]) as

shown in Figure 1. These ideas have been further

extended to include the concept of self-deployment

[6, 7, 18], and of fully “robotic” base stations [9]. This

architectural direction is now percolating into the

standards bodies, as evidenced by 3GPP TR 23.882

[1], where flat cellular architectures are being con-

sidered. In its purest form, a flat cellular architecture

is one in which all the wireless access specific func-

tions are pushed into an intelligent access node

and the core network remains access independent.


This enables the easy integration of a variety of wire-

less access types and takes the cellular system towards

the ubiquitous ideal mentioned earlier.

The trends mentioned above indicate the need

for solutions that could help satisfy these diverse

requirements. In the short term, the most pressing

need is to re-think the kind of cellular architecture

that would most likely yield a cost effective way of

addressing the coverage gaps/holes in an operator’s

market (hot zones, in-building, home, and enter-

prise). In the long term, operators need to empower

such a network with the ability to think for itself. In

pursuing such aims, the general issue of increasing

OPEX and CAPEX—the very reason that these gaps

have not been effectively addressed—can also be

resolved.

Following the above arguments, a picocellular

and femtocellular home base station deployment as

an overlay network is investigated. This deployment

addresses the profitable areas that require both voice

and high-speed data access (i.e., hot zones, in-build-

ing, home, and enterprise areas). These picocells and

femtocells combine the flat architecture of the BSR

with substantial auto-configuration capabilities to sup-

port simple plug-and-play deployment by customers,

and their flat, all-IP architecture allows them to use a

wireline Internet connection for backhaul. They are

deployed in conjunction with a wide area cellular net-

work for area coverage in an urban environment as

illustrated in Figure 2.

There is a commercial need to address specific

market areas, that of hot zone, in-building, home,

and enterprise, to satisfy the end user’s expectation

for delivery of packet data on a par with wired

Internet access. New emerging cellular wireless archi-

tectures designed specifically for cellular wireless

packet data will be capable of autonomic means to

manage and configure such a network. What remains

to be determined is whether the areas identified above

can be addressed cost effectively by these emerging

ATM—Asynchronous transfer modeBSR—Base station routerIP—Internet ProtocolGGSN—Gateway GPRS support nodeGPRS—General packet radio service

MSC—Mobile switching centerPSTN—Public switched telephone networkRNC—Radio network controllerSGSN—Serving GPRS support nodeUE—User equipment

Gn

NodeB

GGSNSGSN

RNC

Circuitvoice

Uu Iu-PS

lu-CS

GiIP

UE

UE

ATM/IPMSC

PSTN

NodeB

In a flat cellular system the BSR integrates functionalelements of these nodes all into a single box.

Figure 1.Overview of the BSR concept.


technologies and if such a radical redesign of cellular

systems can have additional positive benefits on exist-

ing cellular systems. In this paper it is argued that

such a positive effect can be engineered.

Key Requirements for Femtocell DeploymentsThere are a few key requirements for a successful

large scale femtocell deployment. One of the most

important is easy plug-and-play setup by the user to

make the deployment economically feasible by keep-

ing the OPEX and deployment costs low. Therefore,

extensive auto-configuration capability is essential

since many parameters depend on the local radio con-

ditions, neighboring cells, and distance from co-chan-

nel macrocells. In this section, the plug-and-play

deployment process from purchase to network inte-

gration is described. The configuration of transmit

powers of a femtocell for co-channel operation with

an existing macrocellular network is one aspect of

auto-configuration which is used for the performance

evaluation in this paper, and it is discussed in detail.

Also discussed is the requirement of public access

for co-channel operation if no alternative macrocell

frequency is available.

BS—Base stationIP—Internet ProtocolGGSN—Gateway GPRS support nodeGPRS—General packet radio serviceMSC—Mobile switching center

PSTN—Public switched telephone networkRNC—Radio network controllerSGSN—Serving GPRS support nodeUMTS—Universal Mobile Telecommunications System

GGSN

SGSN

MSC

RNC

PSTN

OperatorIP network

MacrocellNode B

Home cell

IP Internet

Traditional UMTS architecture Home BS architecture

Home cellcontroller/gateway

Figure 2.Overview of a macrocellular underlay network with femtocell home base station overlay deployment.


Plug-and-Play Deployment: From Purchase to NetworkIntegration

A high-level overview of the setup process for a

femtocell from purchase to integration into an exist-

ing macrocellular network is described below.

Purchase and registration. The end user buys a

femtocell and a femtocell ID and registers user infor-

mation including the address where the femtocell will

be deployed, as well as preferential access user equip-

ment (UE) for the femtocell. This registration process

is performed the same way as when buying a mobile

phone today. During this process, the user’s address is

automatically converted to longitude and latitude and

stored along with the other user data in a relational

database. This database is part of a central network

element called the home unit server (HUS), which is

part of the home-cell controller and resides in the

operator’s core network. Additionally, the user is pro-

vided with access details to a secure Web site where

he can update some personal information and add or

remove preferential access UEs compliant to the opera-

tor’s terms and conditions. The femtocell ID provided

can be a separate subscriber identity module (SIM)

card with configuration information or a preinstalled

digital certificate hard-coded in the femtocell. It con-

tains information for contacting the HUS in the opera-

tor’s network and for authentication.

Installation and authentication. At home, the user

plugs in power and backhaul connectivity (e.g., digi-

tal subscriber line [DSL], cable, metro Ethernet) and

switches the femtocell on. When a connection to the

Internet is established, the femtocell contacts a HUS in

the operator’s network and transmits the identifica-

tion code that is part of the user’s femtocell ID. The

HUS then authenticates the femtocell. If the HUS

detects an old femtocell firmware or pre-configured

data set, it performs an automatic online update.

Auto-configuration of initial parameters. After

successful authentication, the HUS automatically gen-

erates and downloads initial configuration data to the

femtocell. The following parameters are transmitted

for a Universal Mobile Telecommunications System

(UMTS) femtocell:

1. Frequency for uplink and downlink,

2. Femtocell scrambling code list,

3. Cell ID,

4. Location, routing and service area codes,

5. Initial pilot- and maximum transmit power based

on target range, and

6. Macro neighbor search priority list (if informa-

tion on macrocells is available).

Parameters 1 and 2 are provisioned directly by

the operator. All other parameters are selected or cal-

culated automatically, based on information on the

macrocell network provided by the operator and on

user information provided via the registration process.

Once the initial parameters are downloaded, the

femtocell re-tunes its receiver to the downlink frequency

band and performs measurements to “learn” about its

radio environment, detecting neighboring base stations

and their scrambling codes. This way, the femtocell can

auto-configure the neighbor list and, as a result, choose

a scrambling code that is not used in the area from the

list provided by the HUS. Finally, the auto-configured

parameters are transmitted to the HUS and stored in the

relational database, where they can be used to identify

configuration problems should they occur.

Location check. To ensure the femtocell is deployed

in compliance with the operator’s terms and condi-

tions as set out in the contract with the end user and

does not radiate in a location where the operator does

not own spectrum usage rights, the physical location of

the femtocell must be checked. This can be achieved

by several means, for example, based on radio fre-

quency (RF) measurements instructed by the HUS that

are compared with expected results for the registered

location, or based on information from the backhaul

connection. If the location check confirms an allowed

location, RF transmission is authorized by the HUS.

From this point the femtocell is fully functional.

Ongoing self-optimization during operation. To allow

adaptation to changes in the network environment

(i.e., configuration and properties of neighboring

macrocells/femtocells) an ongoing measurement and

self-optimization process is performed to adapt param-

eters such as scrambling code, pilot- and maximum

data transmit power, and the neighbor list. This

ensures minimal impact on the macrocellular network

and ensures that femtocell performance is maximized

under the given constraints. All parameter updates are


transmitted to the HUS and stored in the relational

database.

Auto-Configuration of Transmit Powers for Co-Channel Operation

One aspect of the auto-configuration process is

the configuration of the downlink and uplink transmit

powers for pilot and data. This is a particularly critical

issue when the femtocells reuse the same frequency

band as an existing macrocellular network, since the

powers define the femtocell coverage area and have

an impact on the interference, handover signaling,

and dropped call rate. This section gives an overview

of the power auto-configuration that allows a co-

channel deployment of femtocells.

Downlink. In the downlink, both the pilot power

(that defines the cell range) and the maximum trans-

mit power (to limit interference) must be configured.

This is described below for the maximum transmit

power and is used for the performance analysis in this

paper. The same approach is also applied for the auto-

configuration of the pilot power, which is typically

around 1/10 of the total power, to achieve the target

range.

Here, the transmit power for each femtocell is set

to a value that is on average equal to the power

received from the closest macrocell at the target

cell radius of r, subject to a maximum power of Pmax.

The femtocell transmit power can be calculated in

decibels as

, (1)

where Lfemto(r) � 38.5 � 20log10(r) is the line of sight

path loss at the target cell radius r (excluding any wall

losses), Pmacro is the transmit power of the sector in

which the femtocell is located, and G(u) is the antenna

gain in direction of the femtocell where u is the angle

to the femtocell with respect to the sector angle.

Panel 2 describes the assumed simulation parame-

ters. Lmacro(d) denotes the average macrocell path loss

at the femtocell distance d (excluding any additional

� Lfemto(r), Pmaxb

Pfemto � minaPmacro � G(u) � Lmacro(d)

wall losses). In reality, Lmacro can be estimated based on

an average channel model and the distance or based

on path-loss measurements at the femtocell, and

assumptions on the average wall loss of a house. This

achieves a constant cell range independent of the dis-

tance to the macrocell in such a co-channel hierar-

chical cell structure.

Note that due to the strong signal falloff at short

distances from the transmitter, the calculation of

Pfemto is not very sensitive to errors and incorrect

assumptions on Lmacro(d) and will result only in small

deviations of the cell radius. Dependent on the loca-

tion of the femtocell within the macrocell, typical

transmit powers resulting from equation 1 range

from a few mW at the edge of the macrocell

to the maximum of Pmax � 125 mW close to the

macrocell.

Uplink. In the uplink, the UE power is limited to a

value limiting the aggregate interference of all femto-

cell UEs to the closest macrocell to a pre-defined value

(subject to a maximum of PUE,max � 125mW), thereby

ensuring the uplink of the macrocell is not degraded

significantly.

Here, a very simple approach is considered where

the total interference allowance is shared equally

among all femtocells in each sector irrespective of

how many are actively receiving data from UEs.

Interference generated by UEs to neighboring sectors

or cells is not taken into account. The maximum

allowed UE power can be calculated as

(2)

where nfemto is the number of deployed femtocells in

the sector of interest, Pinterference,max is the maximum

allowed interference originating from those, and

Lmacro,measured is the current path loss from the macro-

cell in which the UE is located, which can be obtained

from the downlink measurement report. This can be

further optimized by sharing the interference

allowance only between femtocells with active links,

or even by adapting the total allowed interference

based on the current macrocell performance.

PUE � min° Pinterference,max

nfemto� Lmacro,measured, PUE,max¢

allowed interference

•estimate of recieved macro cell power

µ


Public Access for Co-Channel OperationA further requirement for co-channel operation

when only one macrocell frequency is available is to

grant public access to femtocells in order to prevent

excessive interference for UEs of the same operator

located close to a privately owned femtocell.

To highlight the issues, one can consider an exam-

ple where a femtocell owner’s visitors are using UEs

operating on the same frequency and are located very

close to the femtocell (e.g., at a one meter distance

inside of the house). If they are not allowed to con-

nect via the femtocell and no alternative macrocell

connection is available on a different frequency, it will

be a source of very strong interference in the down-

link, even when not active due to the continuous pilot

transmission. Therefore, a significant increase in

macrocell power would be needed by these users to

maintain an adequate signal-to-interference-plus-

noise ratio (SINR). This would, in the best case, sig-

nificantly reduce the macrocell capacity. At worst,

users would not be able to receive data from the

macrocell at all due to the limited transmit power of

the macrocell. Allowing public access for co-channel

operation solves this problem.

However, allowing public access also has other

implications, since any user can access wireless services

via any femtocell, using its owner’s backhaul.

Therefore, this must be allowed by the regulator and

tolerated by the backhaul provider, and the femtocell

owner should be compensated, for example, through

cheaper calls at home.

Performance AnalysisIn this section, the technical feasibility of the fem-

tocell concept is investigated by using simulations to

determine the achievable performance in a co-channel

scenario with an existing macrocellular network. The

achievable capacities and the impact on the perform-

ance of the macrocelluluar network are investigated

for both uplink and downlink. In addition, the impact

on the dropped call rate resulting from handovers due

to a large scale femtocell deployment is quantified for

a simulated scenario in central London.

Achievable Capacity in a Co-Channel ScenarioFigure 3 illustrates a scenario with seven macro-

cells, each with three sectors, and 100 femtocells per

macrocell sector. As shown in Figure 3a, the femto-

cells, which appear as solid dots, are deployed

randomly within the coverage area and reuse the

same frequency as the macrocells, shown as squares,

in a hierarchical cell structure. Figure 3b shows the

Panel 2. Simulation Parameters

Outdoor path loss is modeled as 28 � 35log10(d)dB where d is the distance from the base stationin meters.

Indoor path loss is modeled as 38.5 � 20log10(d)� Lwalls dB where the wall loss Lwalls is explicitlymodeled (15/10/7 dB for external/internal/lightinternal walls, respectively, 3 dB for doors, and 1 dB for windows).

Shadow fading is modeled as random processwith log-normal distribution (8 dB standarddeviation for outdoor where other houses andobstacles are implicitly modeled, and 4 dBstandard deviation for indoor).

The receiver noise power is modeled as 10 log10 (kT NF W ) where the effective noisebandwidth W � 3.84 � 106 Hz, and kT � 1.3804� 1023 � 290 W/Hz.

The noise figure at the UE and the femtocellis NF[dB] � 7 dB, at the macrocell receiver a valueof NF[dB] � 4 dB is assumed.

The macrocell antenna gain is calculated as

b � 70/180 angle where gain pattern is 3dBdown from peak

Gs � 20dB sidelobe gain level in dBGmax � 16dB maximum gain level in dB

dB—DecibelUE—User equipment

�p � u � p with

G(u)dB � Gmax � min c12a ubb2

, Gs d ,


floor plan for the explicit indoor wall model for the

modeled house, and Figure 3c provides an example of

the resulting path gains in dB for a femtocell located

in the center of the house. Key simulation parameters

including the propagation models for path loss,

shadow fading, and the antenna gain for the macro-

cell sectors assumed are shown in Panel 2.

The macrocell downlink transmit power is set such

that the received signal-to-noise ratio inside of a house

at the cell edge is 10 dB, assuming an additional wall

loss of 15 dB. As a result the obtainable throughputs

from the macrocell are interference limited. The fem-

tocell power is controlled as described in the previous

section, assuming a target cell radius of r � 10 m for

1000 500 0 500 1000

1000

500

0

500

1000

x[m]

y[m

]

Area shownin 3D plots

Area of interest forperformance statistics

(a) Femtocell/macrocell co-channel scenario.

10 5 0 5 1010

5

0

5

10

x[m]

y[m

]

(b) Floorplan.(c) Example path gain in dB for a femtocell

located in the center of the house.

10

�100

�50

10

8

8

x[m]

y[m]

6

6

4

4

2

2

0

0

�2

�2

�4

�4

�6

�6�8

�8

Figure 3.Snapshot of the simulated scenario with 100 femtocells per macrocell sector, and house model.


line-of-sight propagation subject to a maximum value

of 125 mW. The small cell radius of 10 meters (m) is

considered to cover only the user’s homes, in order

to minimize handovers from passing users outside. The

maximum power for UEs in the uplink is PUE,max � 125

mW. The femtocell UE power is additionally controlled

as described above to limit the total interference

received at the closest macrocell to a value 3dB above

the thermal noise level.

System level simulations were performed using

MATLAB* to identify the possible downlink and uplink

throughputs at any location of the scenario both for

the macrocell and for femtocells operating simulta-

neously in the same frequency band. Throughput sta-

tistics are obtained for femtocells located in the three

sectors covered by the central macrocell, so that inter-

ference from the adjacent cells can be taken into

account. It is assumed that each UE connects to the

base station (femto or macro) with the best downlink

signal-to-interference ratio (i.e., public access).

Transmissions to multiple users are scheduled in time

as in high speed downlink packet access (HSDPA).

The theoretically obtainable throughput can be cal-

culated based on a well known capacity relationship,

the Shannon-Hartley theorem [19], which can be

written as

(3)

where W denotes the channel bandwidth and SNR is

the signal-to-noise ratio where the thermal noise

is modeled as additive white Gaussian noise (AGWN).

To be able to take interference into account in equa-

tion 3, both inter-cell and inter-sector interference are

modeled as a zero-mean circularly symmetric com-

plex Gaussian process whose variance equals the sum

of the powers received from adjacent co-channel cells

and sectors, respectively. Assuming an operation point

of a[dB] from the capacity limit, the obtainable through-

put in each location of the scenario can be calculated

depending on the SINR at this location as

(4)

In this paper W � 3.84 � 106 Hz and a[dB] � 3 dB

are assumed. In the downlink, interference resulting

from all macrocell sectors and all femtocells is taken

Ca � W log2 a1 �SINR

10a[dB]�10b.

C � W log2 (1 � SNR),

into account. In the uplink, interference results from

all currently active UEs. Macrocell UEs are located

randomly in the coverage area of each sector (one

active UE per sector at each time instant). Femtocell

UEs are located randomly in each house (one active

UE per femtocell at each time instant).

The simulated resolution is 40 meters for the

macrocell throughput. Areas where femtocells are

deployed were simulated with a higher resolution of

0.6 meter; 100 simulation iterations were performed

with different shadow fading values, house locations,

and femtocell locations for each scenario to collect

throughput statistics from which cumulative distri-

bution functions (CDFs) of the possible through-

puts were derived. They describe the probability

distribution of the possible macro- and femtocell

throughputs over the area covered by the three

sectors of the central macrocell, or the area covered by

femtocells, respectively. In order to evaluate the

impact of different numbers of deployed femtocells

on the macrocell performance, the scenario was also

simulated with 0, 10, and 100 femtocells per sector.

The simulation results show that co-channel

deployment of femtocells on the same frequency as an

existing macrocellular network is feasible with only

minor impact on macrocell performance when using

the proposed femtocell power auto-configuration. In

addition, due to the small distance between transmit-

ter and receiver, and the additional shielding due to

walls in typical situations, the femtocells can provide

very high throughputs, much higher than those

supported by the modulation and coding schemes

specified in current standards. A more detailed

description of the results is discussed below.

Downlink results. The throughput CDFs for the

downlink of both macrocells and femtocells with dif-

ferent numbers of N active femtocells and examples of

downlink throughput distributions around one

macrocell and femtocell are shown in Figure 4.

As Figure 4a shows, the drop in macrocell

throughput as a result of the additional interference

caused by 10 or 100 deployed femtocells is only mini-

mal. This is essential for a successful co-channel fem-

tocell deployment since any significant degradation in

coverage or capacity of the existing macrocellular net-

work would be unacceptable. In the downlink, this


results from the low femtocell transmit power, the

strong signal falloff at the femtocell boundary due to

the path loss at such short ranges, and, in most cases,

an additional wall separation. It is also shown that due

to high indoor SINRs, the theoretically achievable fem-

tocell throughput is very high, so that in most of the

cases 64 quadrature amplitude modulation (QAM) or

higher order modulation would be required to achieve

it, assuming 1/2 rate coding. The high indoor SINRs

are a result of the small distance to the femtocell and

the wall separation which shields the signal from inter-

ference. It is also shown that increasing the number

0 2 4 6 8 10 12 14 16 18 200

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

64 QAMrate 1/2

16 QAMrate 1/2

4 QAMrate 1/2

(a) Downlink throughput 3dB from the capacity limit (Mbps).

CD

F

macrocell, N = 0macrocell, N = 10macrocell, N = 100femtocell, N = 10femtocell, N = 100

Adding femtocells has littleimpact on macrocellperformance

Very high femtocell throughputsare possible, 64 QAM or highermodulation required

CDF—Cumulative distribution functionQAM—Quadrature amplitude modulation

(b) Example throughput distribution.

600

480

360

240

120

�120

�240

�360

�360 �240 �120120 240 360 480 600

0�480

�480

y[m]

x[m]

50

0

Thro

ug

hp

ut

[Mb

ps]

0

(c) Example throughput distribution in thearea around one femtocell.

10

8

6

4

2

�2

�4

�6

�6 �4 �22 4 6 8 10

0�8

�8

y[m]

x[m]

50

0

Thro

ug

hp

ut

[Mb

ps]

0

Figure 4.CDF and distribution of downlink throughputs for locations within the central macrocell for both, macrocellsand femtocells.


of femtocells from 10 to 100 does not significantly

affect the downlink femtocell throughput due to the

small cell size and a typically strong wall separation

between houses. Therefore, large numbers of femto-

cells can be deployed in the same frequency band of

existing macrocell networks without a significant

downlink performance degradation of the macrocells,

when public access to the femtocells is granted.

Figure 4b shows one example of a corresponding

downlink throughput distribution around the central

macrocell with 10 deployed femtocells per macrocell

sector, for clarity without shadow fading. As expected,

the macrocell throughput is highest close to the base

station and in the directions of the main lobe of the

directional antenna for each sector and falls off toward

the edge of each sector due to the lower signal levels

and increased interference from neighboring sectors.

The spikes of higher throughput shown are in loca-

tions where femtocells are deployed.

Figure 4c shows one example of a downlink

throughput distribution around one of the femtocells

located at the coordinates [�100 m, �300 m] within

the central macrocell, for clarity without shadow fad-

ing. As a result of the short distances to the femtocell

and the walls that shield the house from interference,

the achievable indoor throughputs are very high.

Outside of the house, throughput is reduced since

those areas are served primarily by the macrocell

which provides the strongest signal, and interference

caused by other macrocells and femtocells is higher

than indoors.

Uplink results. Throughput CDFs and examples of

throughput distributions for the uplink are shown

in Figure 5.Similarly to that of the downlink, the drop in

macrocell uplink performance resulting from the addi-

tion of 10 or 100 active femtocells is minimal, as shown

in Figure 5a. This is a result of the fast maximum power

control in the uplink for femtocell UEs that limits the

interference caused to macrocell UEs to a pre-defined

level. Despite this maximum uplink power limitation,

it is shown that the possible throughput of femtocell

UEs is very high as a result of high signal-to-interference

ratios due to the small distance to the femtocell and

the wall separation to the macrocell. When the num-

ber of femtocell UE transmissions is increased from

10 to 100 per macrocell sector, the maximum allowed

interference per UE is reduced by a factor of 10.

Therefore, the SINR can drop by up to 10 dB, resulting

in a reduction in uplink throughput. However, since

the uplink throughput performance is very high, this is

not problematic. As for the downlink, the theoretically

achievable uplink throughput is so high that in most of

the cases 64 QAM or higher order modulation would

be required to achieve it.

Figure 5b shows one example of a corresponding

uplink throughput distribution for 10 deployed fem-

tocells per macrocell sector, for clarity without shadow

fading. As expected, the achievable uplink throughput

is high in the direction of the main lobe of the direc-

tional antenna of each sector close to the base station

and falls off with increasing distance. The levels of

achievable throughput are different for each sector,

since they are impacted by the interference from cur-

rently transmitting mobile devices in the neighboring

sectors, which are randomly located within each of

the sectors. The spikes of high throughput are in loca-

tions with femtocell coverage.

Figure 5c shows one example of an uplink

throughput distribution around one of the femtocells

located at the coordinates [�100 m, �300 m] within

the central macrocell, for clarity without shadow fad-

ing. It is shown that the achievable indoor throughput

is very high. This is also the case for areas outside with

good channels to the femtocell such as in front of win-

dows within the range where the received downlink

signal from the femtocell is still the strongest. This is

due to the assumption that in the uplink, the mobile

device is connected to the same base station as in the

downlink (i.e., the base station from which it receives

the strongest signal power).

Call Drop Probability as a Result of Handovers in an Urban Residential Scenario

Widespread deployment of co-channel femtocells

can result in an increase in the number of macrocell-

femtocell handover requests by underlay macrocell

users (i.e., users walking outside and connected to

the macrocell) due to the leakage of the femtocell’s

pilot outside the home it is deployed in. To minimize

the effect of this, the femtocell’s pilot power needs to

be adjusted so that leakage outside the home is


reduced. This section describes simulation studies to

obtain a larger-scale insight into the effect of imple-

menting pilot auto-configuration schemes for the

femtocell in a real-world residential scenario, specifi-

cally on the number of macrocell-femtocell handover

requests coming from macrocell users.

Scenario description. A femtocell deployment in a

London residential area, as shown in Figure 6, is

chosen. A 500 m � 500 m area is chosen since this

is in line with the area typically covered by a singe

macrocell sector in a dense residential area. The area

consists of densely packed rows of compact terraced

64 QAMrate 1/2

16 QAMrate 1/2

4 QAMrate 1/2

Adding femtocells haslittle impact on macrocellperformance

Very high femtocell throughputsare possible, 64 QAM or highermodulation required

Performance drop sincemaximum uplink power isreduced by factor 10

0 5 10 15 20 25 30 35 400

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

(a) Uplink throughput 3dB from the capacity limit (Mbps).

CD

F

macrocell, N � 0macrocell, N � 10macrocell, N � 100femtocell, N � 10femtocell, N � 100

CDF—Cumulative distribution functionQAM—Quadrature amplitude modulation

(c) Example throughput distribution in thearea around one femtocell.

10

8

6

4

2

�2

�4

�6

�6 �4 �22 4 6 8 10

0�8

�8

y[m]

x[m]

100

50

0

UL

Thro

ug

hp

ut

[Mb

ps]

0

(b) Example throughput distribution.

600

480

360

240

120

�120

�240

�360

�360 �240 �120120 240 360 480 600

0�480

�480

y[m]

x[m]

100

50

0

UL

Thro

ug

hp

ut

[Mb

ps]

0

Figure 5.CDF and distribution of uplink throughputs for locations within the central macrocell for both, macro- and femtocells.


houses with similar layouts, as shown in Figure 6b.

The houses do not have front gardens and thus are

located very close to the sidewalk. They also have

windows in the front room directly facing the side-

walk. The high density of the homes, combined with

the high likelihood of femtocell signal leakage onto the

sidewalk, provides a “worst case” scenario of a co-

channel femtocell deployment.

The femtocells are randomly deployed in 10

percent of all households in the area. This figure is

approximately equivalent to half of all the subscribers

of a market-leading UK mobile operator deploying a

femtocell in their homes [21]. This translates to 130

femtocells deployed in the area. The femtocells are

placed randomly within the houses. Figure 6a shows

the location of the deployed femtocells, which appear

as solid dots.

Each femtocell-deployed home is assumed to be

occupied at all times by one to four femtocell users.

Each femtocell user has a 100 milli-Erlang voice traf-

fic call model, and an evening mobile television (TV)

usage model derived from the mobile TV trial results

presented in [3], with TV sessions lasting on average

24 minutes, with a 16 percent busy hour mobile TV

session attempt.

The macrocell user has a voice call model with

an exponentially distributed call duration with a mean

of 100 seconds. The mean call inter-arrival is 3600

seconds, but this parameter is not important since

the drop probability is measured on a per-call basis.

The macrocell users are modeled to walk through this

area starting from one entry point and moving to

an exit point. The entry points are randomly chosen.

The exit point is also randomly chosen, with the con-

dition that it must not be in the same direction as the

entry point to simulate typical transitional movement:

i.e., a macrocell user coming in from the east will not

have an exit point to the east as well.

For the investigation of the handover probabili-

ties, simulations were performed using the same

assumptions as in the previous section. A model of a

terraced house, typical for London in the simulated

Katherine Road

South Esk Road

Bristol Road

Stafford Road

Derby Road

Rutland Road

Strone Road

Monega RoadHalley Road

Rothsay Road

Lansdown Road

Prestbury Road

Jephson Road

Femtocell Sidewalk

13.5 m

1m

4.5 m

Macrocell user

House

(a) Location of deployed femtocells. (b) Typical home layout.

Figure 6.Simulated residential area with femtocell placement and layout of the simulated houses.


area is used, and shown in Figure 6b. In addition, it is

assumed that the pilots are 1/10 of data power for

both macrocells and femtocells, which is common

practice in UMTS. Handover probabilities are calcu-

lated for users walking past the front of a house with

a femtocell at a speed of 1 meter per second (m/s), at a

distance of 1 meter from the house boundary. It

is assumed that a handover is triggered when a pilot

signal of a new cell is 4 dB better than the pilot signal

of the current cell for a time of 500 milliseconds (ms).

When a handover is triggered, the time of the hand-

over procedure is assumed to be 650 ms.

Pilot auto-configuration schemes. Three pilot adjust-

ment schemes are examined:

• Fixed pilot power (scheme A). The femtocell pilot is

set at a fixed power level that allows 90 percent of

the femtocells within a macrocell coverage area to

obtain a target cell radius of 10 meters (free space)

in the investigated scenario.

• Pilot adjusted to distance from macrocell (scheme B). The

pilot power is adjusted relative to the macrocell’s

pilot signal. The femtocell pilot is adjusted to obtain

a target cell radius of 10 meters (free space) if it

can, limited by its maximum pilot transmit power,

as described in the section on auto-configuration of

transmit powers for co-channel operation.

• Pilot adjusted according to femtocell usage (scheme C).

The pilot is adjusted based on the femtocell usage.

When there are femtocell users in active mode (in

a call), the pilot is set to the same level as scheme

B. When all users are in idle mode (not in a call),

the pilot power is reduced by 10 dB. This scheme

assumes that the femtocell user’s idle mode cell

reselection threshold is increased by 10 dB: i.e.,

the UE will remain camped on the femtocell when

in idle mode despite the reduced power.

Handover probabilities are detailed in Figure 7.

For all three cases, the handover probabilities as a func-

tion of the distance from the femtocell were calculated

using simulations. A total number of 50 iterations were

performed per distance with random location of the

house within the three sectors of the central macro-

cell. The resulting handover probabilities as a function

of the femtocell distance from the footpath are shown

in Figure 7a. For the mobility simulations, scheme A

uses the fixed power curve, scheme B uses the curve

for active mode auto-configured power, and scheme C

uses the curve for idle mode auto-configured power

when all femtocell users are in idle mode and the active

mode auto-configured power curve when at least

one femtocell user is in active mode. One example of

the resulting coverage for a femtocell deployed in the

center of the house using scheme B is shown in Figure

7c. The complementary areas covered by the macrocell

for this example are shown in Figure 7b.

Results of handover impact. The resulting macrocell

user handover probability per call when scheme A is

implemented is 80.77 percent. This result shows that

using a fixed power would be not feasible in this sce-

nario due to the very high handover probabilities.

Assuming that there is a 2 percent probability that a

handover results in a dropped call, the drop proba-

bility during a handover event is 3.96 percent of the

handover probabilities (i.e., 2 percent for macrocell-

to-femtocell handover, plus 98% � 2% for femtocell-

to-macrocell handover). This means that each

macrocell call has an approximately 3.2 percent

chance of dropping due to handovers resulting from

the deployment of the femtocells alone.

For scheme B, the user handover probability per

call is 22.9 percent. This is a significant improvement

compared to scheme A and translates to a 0.9 percent

call drop probability due to macro-femto handovers.

Using scheme C, the handover probabilities are

dependent on the usage of the femtocells. Simulations

were run with one to four users in all femtocells. The

results of the simulation in Table I show a significant

improvement over schemes A and B, particularly

when the number of femtocell users is low. The drop

probabilities due to handovers range from 0.3 percent

with one femtocell user to 0.45 percent with four

Macrocell user Users per femtocell handover probability

1 user 7.5 percent

2 users 9.4 percent

3 users 10.3 percent

4 users 11.3 percent

Table I. Handover probabilities for scheme C.

BSR—Base station router


femtocell users, which are acceptable given that the

average household size in that neighborhood is 2.36

persons [22].

The simulation results for the three schemes rein-

force the need for auto-configuration abilities to pro-

vision the femtocells properly. Using a default fixed

setting, as with scheme A, would render widespread

deployment impractical, because of not only the

resulting unacceptably high number of dropped calls,

but also the substantial increase in signaling over-

head required to support the higher handover rate.

Financial Analysis of Picocellular Home BaseStation Deployment With Public Access

The previous sections focused primarily on the

performance impact to an existing network when

0 2 4 6 8 10 12 140

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

House boundary

Pro

bab

ility

of

han

do

ver

to f

emto

cell

and

bac

k

(a) Femtocell distance from footpath

Fixed powerAuto-configured power, activeAuto-configured power, idle

(b) Macrocell coverage

8

�30

�40

�50

�60

�70

�80

�90

�100

Nomacrocellcoverage

Passing users��

y[m

]

�110

�120

�130

6

4

0

2

�2

�4

�6

�8

�4 �2 2 4

x[m]

0

(c) Femtocell coverage

8

�30

�40

�50

�60

�70

�80

�90

�100

No femtocellcoverage

Passing users��

y[m

]

�110

�120

�130

6

4

0

2

�2

�4

�6

�8

�4 �2 2 4

x[m]

0

Figure 7.Handover probabilities as function of distance from the femtocell and example of coverage around one housewith a femtocell.


home-deployed femtocells reuse the same frequency

as network macrocells. This section explores a new

opportunity for operators to provide outdoor coverage

in densely populated areas by increasing the transmit

power of home-deployed femtocells (i.e., femtocells

become picocells) and allowing public access. This

can reduce operator expenses for their macrocellular

networks, since a large portion of the traffic is served

by picocells, and macrocells would only be required to

fill coverage gaps.

In order to evaluate the financial impact of pico-

cellular home base station deployments in con-

junction with a macrocellular network for area

coverage, a cost analysis is performed for an urban

area with the size of 10 km � 10 km. The user

demand for the scenario under investigation is based

on real measurements of voice traffic, extrapolated

to the different operator market shares considered.

An example of a user demand map is shown in

Figure 8 in terms of maximum concurrent calls

within an area of 100 m � 100 m for an operator

market share of 40 percent. The following assump-

tions are made: A total population of 200,000 people

and 65,000 homes are assumed; the home distribu-

tion is equivalent to the user distribution for evening

demand; 95 percent of the population are mobile

users; the usage is assumed to be 740 minutes per

user per month with an average call duration of

3 minutes.

Picocellular home base stations are deployed ran-

domly in homes, which are distributed according to an

evening user demand distribution. It is assumed that

each home unit is able to serve up to eight concurrent

users with 384 kbps each, within a 100 m � 100 m

area where it is deployed. All users that cannot be

served by a home unit need to be served by the macro-

cellular network.

For the macrocellular network, it is assumed that

all base stations are deployed such that each is able to

serve on average a pre-defined number of umacro active

users. For a small numbers of users, this assumption

results in a best case scenario for the macrocellular

20

10

010

8

6

4

2

00

24

68

10

y [km]

x [km]

User Distribution

use

rs in

an

are

a o

f 10

0 m

�10

0 m

Figure 8.Maximum user demand scenario for an operator with 40% market share in a 100 m � 100 m plot.


deployment, since the method does not take transmit

power limitations into account. Since the macrocel-

lular network needs to serve the remaining m users

which are not covered by the home network, the

required number of base stations can be easily esti-

mated as nmacro � m/umacro.

In order to account for different base station

performance levels and voice/data traffic mixes, cases

with different numbers of users per base station are

investigated.

Economic ModelsThis section describes the estimated economic

models, parameters, and assumptions used in the cost

calculations of deploying, operating, and maintaining

the cellular network. Two different models are used,

one for the macrocells and another for the user-

deployed picocells. Note that the models are meant to

find the main differences in cost between the conven-

tional macrocells and the home cells. Therefore, costs

that would exist in both macrocell and home cell cases

(such as billing, bad debt, or certain essential network

nodes) will not be included here.

Macrocell deployment case. All costs associated

with the provision of macrocellular coverage would

be taken up directly by the network operator, which

is the typical arrangement in current networks.

Generally, the costs of a network operator fall into

two categories: CAPEX, such as the cost of equipment

purchase and installation, site purchase, and associ-

ated expenses; and OPEX, such as the day-to-day

recurring costs of maintenance, customer support,

and administration.

The CAPEX can be easily calculated as the sum of

the network equipment costs,

where n is the number of base sta-

tions, cBS is the cost of a node of the base station, and

cRNC, ccore, and crouter are the CAPEX per base station of

the RNC, core UMTS, and core packet routers, respec-

tively. It should be noted that if an operator purchases

its sites, that cost needs to be included in the base sta-

tion cost as well.

The estimation of OPEX can be divided into

network-related costs; SG&A costs such as sales, mar-

keting, and handset subsidies; and non-networking

related costs such as customer care and billing. As

stated above, costs that would also exist in the picocell

� ccore � crouter),

CCAPEX � n(cBS � cRNC)

deployment can be ignored, leaving only network-

related costs to be considered.

The network related costs can be further divided

into three parts:

• Maintenance. This is the cost per annum of main-

taining equipment and is generally given as a per-

centage of the cumulative CAPEX. Cumulative

here implies the cost of the equipment rollout to

date. It includes the costs of physical care, hard-

ware replacement, and software updates.

• Site running costs. These include utility bills such

as electricity and site leasing costs. These costs are

given as cost per site per annum.

• Backhaul costs. These are given as the cost per cir-

cuit per annum (e.g., dollar per annum per E1/T1

leased line). They can also be represented as cost

per unit bandwidth per annum.

The assumptions made here are (a) that all

CAPEX is incurred prior to the commencement of

operations and (b) that capital was raised using a loan

that is repaid with an interest rate r over N years. The

annual repayment as a portion of the total loan is

straightforward to derive. If P is the principal bor-

rowed, then after N years, the amount owed would be

assuming no repayments were made.

However, each repayment R reduces, in part, the

lump sum and hence the interest repayments. A

repayment made at the end of year i has the net effect

of reducing the debt by the sum of R and the interest

that would have to have been paid on that amount

during the remaining years i � 1,. . ., N, i.e.,

After N years, the entire debt would

have been repaid; hence

(5)

which, rewritten, gives the annual repayments:

(6)

The annual CAPEX can then be calculated by

replacing P with the total CAPEX. Therefore, the total

annual network costs for the macrocellular network

(CAPEX and OPEX) is given by:

R �rP

1�(1�r)�N

(1 � r)N P � aN

i�1

(1 � r)N�i R � 0

R(1 � r)N�i.

(1 � r)N P,


(7)

where amaintenance is the fraction of the CAPEX

assumed as the maintenance cost, Asite is the annual

site cost per base station site, Abackhaul is the annual

backhaul cost per unit bandwidth to a single site,

and BW is number of units of bandwidth per site. A

“unit” of bandwidth in this context would be, for

example, a single E1 leased line or 100baseT connec-

tion, rather than the usual measures of bits per second

or hertz.

Table II provides the assumed values of the

parameters, based on figures used in [16]. The num-

ber of base stations, n, is not listed in the table as this

parameter is varied in the simulation studies. Note

that these are meant to give indicative values, and

the actual costs can vary significantly.

Picocell deployment case. In the home-cell case, it is

assumed that the use of auto-configuration reduces

CAPEX by reducing installation, deployment, and

optimization costs. OPEX cost savings would also be

significant, as they would not include site leasing and

utility costs.

There is a wide range of business arrangements

between the end user and the operator for the deploy-

ment of the picocell as part of a service bundle. This

includes, on top of the provision of mobile services,

the home broadband service that serves as the back-

haul to the picocell. Examples of some business

arrangements include those where:

• The picocell is fully subsidized by the operator.

• The cost of the picocell is not subsidized and taken

up fully by the end user, but with free/reduced

price for calls made by the user at his picocell used

as an incentive.

In this study, the first business arrangement

described is used, although it can be argued that the

costs to the operator for both options would probably

be similar. One more vital assumption used is that the

picocell is used to provide coverage to other subscribers

(i.e., not only to the subscriber who “owns” the pico-

cell), but only if it does not impact the quality of serv-

ice (QoS) expected by the picocell subscriber. Another

�amaintenance CCAPEX � n(Asite � Abackhaul BW),

Amacro �rn(cBS � cRNC � ccore � crouter)

1 � (1 � r)�N

possible alternative is for a portion of the picocell’s

capacity to be permanently reserved for serving “out-

side” traffic. The picocell subscriber can receive some

form of compensation proportional to the amount of

“outside” traffic his or her picocell serves. This feature

has a significant impact in this study, as it provides a

mechanism whereby operators are able to cut costs

significantly by reducing (or even eliminating) major

costs such as site and backhaul costs.

The CAPEX associated with picocell deployments

can be calculated the same way as in the macrocell

case, but due to the flat cellular architecture used, it

does not include the costs associated with RNC and

core UMTS nodes.

Parameter Description Value Units

r Rate of return 5 percent/year on capital

N Term of loan 15 years

CBS CAPEX of one 40,000 $base station

CRNC RNC CAPEX 3,800 $per base station

Crouter Core network 154 $packet router CAPEX per base station

Ccore Core network 830 $UMTS CAPEX per base station

amaintenance Maintenance 12 percent/year cost as a fraction of CAPEX

Asite Annual site cost 6,000 $/yearfor base station

Abackhaul Annual cost of 10,000 $/line/yearbackhaul per E1 line

BW Number of 4 lines/sitebackhaul lines per base station site

Table II. “Typical” parameter values for estimating costsof macrocellular deployment.

CAPEX—Capital expenditureRNC—Radio network controllerUMTS—Universal Mobile Telecommunications System


The OPEX for picocell deployments are different

because they exclude the costs directly associated

with the picocells. Site costs are not considered as

this is taken up by the end users. The home base sta-

tions (BSs) use the end user’s broadband connection

for backhaul, which is why backhaul costs are also

not included. The total annual network costs for the

picocellular home network (CAPEX and OPEX) can

thus be calculated as:

(8)

The values used for the picocell are given in

Table III. CCAPEX here includes the cost of both the

picocell and the core network routers.

Results of the Financial AnalysisFigure 9 shows the fraction of users served by

home cells as a function of different operator market

shares, based on the percentage of customers with

installed home base stations. It becomes evident that

even though the home base stations are placed ran-

domly with the home distribution, a relatively small

fraction of installed units already achieve significant

total user coverage. For example, for an operator that

has 40 percent market share and only 20 percent of its

customers have home base stations deployed, those

Apico �rn(cBS � crouter)

1 � (1 � r)�N � amaintenance CCAPEX

home base stations alone can satisfy approximately

80 percent of the total demand from all its customers.

With smaller market share, the fractional user cover-

age becomes smaller for the same fraction of installed

home base stations, which is expected since the

number of installed home base stations is also smaller,

proportional to the market share.

Figure 10 illustrates the annual costs for various

parts of the network as a function of the percentage

of customers with installed home base stations for an

operator with a market share of 40 percent, and 64

supported active users per macrocell. At 0 percent,

all users are covered by macro cells, and at 100 per-

cent, all homes have a home base station installed.

Annual OPEX for the home network are slightly

higher than associated capital expenses. For the

macrocellular network, the operational expenses

clearly dominate the total annual cost. The results

show that the costs for the macrocellular network

initially decrease very quickly, as a consequence of

the high user coverage that can be achieved with a

relatively small fraction of homes with installed home

base stations, as shown before in Figure 9. This has a

significant effect on the total network costs for instal-

lations that include both macrocellular and home net-

work-based equipment. For this example, the total

annual network costs are minimal in the case where

approximately 30 percent of the customers have a

home base station installed.

Figure 11 shows how the number of active users

supported per macrocell impacts total annual network

costs. Not surprisingly, costs increase significantly with

a reduced number of supported users. The results also

show that macrocellular coverage becomes less

economically viable as demand for high data rate serv-

ice increases, as is expected in the near future.

For example, for requested data rates of 384 kbps, a

UMTS system using HSDPA with a downlink through-

put of approximately 5 Mbps per sector is only able to

sustain up to 39 active concurrent links for a three-

sector cell. Furthermore, a backhaul of four E1 lines

(2 Mbps each) per cell would limit the number of

active high data rate links to 20. In this case, a mixed

deployment of macrocells for area coverage and home

picocells for the main demand can reduce annual

network costs by up to 70 percent compared to a

Parameter Description Value Units

r Rate of return 5 percent/yearon capital

N Term of loan 15 years

CBS CAPEX cost of 350 $one picocell

Crouter Core network 77 $packet router CAPEX per BS

amaintenance Maintenance 12 percent/yearcost as a fraction of CAPEX

Table III. “Typical” parameter values for estimatingcosts of picocell deployment.

BS—Base stationCAPEX—Capital expenditure


0 20 40 60 80 1000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Percentage of customers with home base stations

Frac

tio

n o

f u

ser

cove

rag

e

40% market share30% market share20% market share10% market share

Figure 9.Fractional user coverage for different operator market shares.

0 20 40 60 80 1000

0.5

1

1.5

2

2.5

3

3.5

4� 106 Results for 40% market share and 64 users per macrocell


An

nu

al n

etw

ork

co

st (

$)

Home BS CAPEXHome BS OPEXHome BS totalMacrocell CAPEXMacrocell OPEXMacrocell totalTotal network costs

BS—Base stationCAPEX—Capital expenditureOPEX—Operational expenditure

Figure 10.Network costs for an operator with 40 percent market share and 64 users per macrocell.


network with macrocells only. Even in a case where

voice services predominate, and where each macrocell

typically supports up to 64 users per sector, the total

annual network costs can be reduced by up to 30 per-

cent when a mixture of macrocells and home base

stations are deployed. It is also shown that the optimal

fraction of customers with home base stations varies

between 10 percent and 60 percent, depending on

the average number of supported concurrent users

per macrocell.

Given the results above, it becomes clear that a

random deployment of picocellular home units by the

end user within a macrocellular network can reduce

an operator’s total annual network costs in urban

areas significantly. Indeed, results predict that the

potential financial benefit will increase significantly

in the future with the widespread adoption of high

data rate services.

ConclusionsExperience over the last 10 to 15 years with

macrocellular deployments has shown that while it

was technically feasible to plan and deploy small cells

manually, such deployments have been generally

held to be economically unfeasible due to the OPEX

intensity of planning for such deployments. This has

prevented operators from fully addressing markets

that such small cell technology could effectively tar-

get. This paper focused on two of the major issues

for small cell deployment in UMTS: the ability to

deploy small cells autonomically in macrocellular co-

channel environments, and given that solutions have

been found for such deployments, the financial

impact of user-deployed publicly accessible picocel-

lular home base stations in macrocellular networks.

In developing solutions for co-channel deployments,

essential requirements such as auto-configuration and

public access were discussed and a power control

method was proposed that achieves a constant fem-

tocell range and ensures a low pre-definable impact

on the uplink performance of the macrocell network

for co-channel operation. Moreover, it was shown

that co-channel deployment of femtocells can be

achieved with only minor impact on the macrocell

0 20 40 60 80 1000

5

10

15� 106


An

nu

al n

etw

ork

co

st (

$)

16 users/macrocell32 users/macrocell

64 users/macrocell128 users/macrocell256 users/macrocell

Figure 11.Network costs for an operator with 40 percent market share and different numbers of supported active users per macrocell.


throughput. This allows efficient spatial frequency

reuse, resulting in a significantly higher spectral effi-

ciency per area. Despite the low femtocell transmit

powers, the short distance between femtocell and UE,

and the wall separation to other interference sources,

results point to very high achievable theoretical fem-

tocell throughputs for both uplink and downlink in

most of the covered femtocell area. In reality, those

data rates can be achieved only by using 64 QAM or

higher order modulation schemes. It is therefore rec-

ommended that such schemes should be pursued in

the relevant cellular standards bodies.

The simulation results of the impact on call drop

probabilities have further reinforced the need for

auto-configuration capabilities, since otherwise both

the signaling overhead and the call drop rate are

increased substantially as a result of a highly increased

rate of handovers.

In analyzing the financial impact, simulations

have shown that macrocellular coverage becomes less

economically viable with the increasing demand for

high data rate services that is expected in the near

future, mainly due to the high operational expenses

associated with macrocellular networks. This problem

can be addressed in urban areas with the use of a mix-

ture of picocellular and femtocellular home base sta-

tions, randomly placed by customers. The picocellular

home base station is introduced as a marginally higher

power and higher-capacity home base station. It was

shown that a large fraction of user demand could be

covered quickly by installing picocells in only a small

fraction of the customers’ homes. As a result, picocel-

lular home base station deployment in combination

with a macrocellular network for area coverage has

the potential of significantly reducing total network

costs. Annual cost reductions in the range of 30 per-

cent to 70 percent have been predicted for voice and

high-speed data services, respectively. Therefore, the

use of home base stations can provide a major financial

advantage now, and one that will grow significantly in

the near future, with the widespread adoption of high

data rate mobile services.

*Trademarksi-mode is a trademark of NTT DoCoMo.MATLAB is a registered trademark of The Mathworks, Inc.

References[1] 3rd Generation Partnership Project, “3GPP

System Architecture Evolution (SAE): Report onTechnical Options and Conclusions,” 3GPP TR23.882, �http://www.3gpp.org/ftp/Specs/html-info/23882.htm�.

[2] 3rd Generation Partnership Project, “HighSpeed Downlink Packet Access (HSDPA):Physical Layer Aspects,” 3GPP TR 25.858, 2002,�http://www.3gpp.org/ftp/Specs/html-info/25858.htm�.

[3] Arqiva and O2, “Oxford Mobile TV Trial FinalResults,” May 2006, �http://www.dvb-h.org/PDF/060626.Oxford-Final-Results.pdf�.

[4] M. Bauer, P. Bosch, N. Khrais, L. G. Samuel,and P. Schefczik, “The UMTS Base StationRouter,” Bell Labs Tech. J., 11:4 (2007), 93–111.

[5] R. A. Chaney, “Automatic Mobile RadioTelephone Switching System,” U.S. Patent3,355,556 (Nov. 28, 1967).

[6] H. Claussen, “Autonomous Self-Deployment ofWireless Access Networks in an AirportEnvironment,” Autonomic Communication:Second Internat. IFIP Workshop, WAC 2005,Athens, Greece, Oct. 2005: Revised SelectedPapers (I. Stavrakakis and M. Smirnov, eds.),Lecture Notes in Computer Science (LNCS),Vol. 3854, Springer, Berlin, Heidelberg, NewYork, 2006, pp. 86–98.

[7] H. Claussen, “Distributed Algorithms for RobustSelf-Deployment and Load Balancing inAutonomous Wireless Access Networks,” Proc.IEEE Internat. Conf. on Commun., (ICC ‘06)(Istanbul, Tur., 2006), vol. 4, pp. 1927–1932.

[8] H. Claussen, “Performance of Macro- and Co-Channel Femtocells in a Hierarchical Cell Structure,” Proc. 18th IEEE Internat.Symposium on Personal, Indoor and MobileRadio Commun. (PIMRC ‘07) (Athens, Gr., 2007), paper 626.

[9] H. Claussen, L. T. W. Ho, H. R. Karimi, F. J.Mullany, and L. G. Samuel, “I, Base Station:Cognisant Robots and Future Wireless AccessNetworks,” Proc. 3rd IEEE Consumer Commun.and Networking Conf. (CCNC ‘06) (Las Vegas,NV, 2006), vol. 1, pp. 595–599.

[10] H. Claussen, L. T. W. Ho, and L. G. Samuel,“Financial Analysis of a Pico-Cellular HomeNetwork Deployment,” Proc. IEEE Internat.Conf. on Commun. (ICC ‘07) (Glasgow, Scot.,2007), pp. 5604 –5609.

[11] H. Claussen, H. R. Karimi, and B. Mulgrew,“Low Complexity Detection of High-Order


Modulations in Multiple Antenna Systems,” IEEProc. Commun., 152:6 (2005), 789–796.

[12] S. Das, T. E. Klein, K. K Leung, S. Mukherjee,G. Rittenhouse, L. G. Samuel, H. Viswanathan,and H. Zheng, “Distributed Radio Link ControlProtocol Facilitating Fast Cell Site Selection inan All-IP Cellular Network,” Proc. IEEE 62ndVeh. Technol. Conf. (VTC Fall ‘05) (Dallas, TX,2005), vol. 3, pp. 1692–1696.

[13] E. Fox, North American ARPU GrowthOutpaces the World: A Look at WirelessForecast Drivers, Yankee Group, Mar. 2006.

[14] GSM Association, “GSM World SubscriberStatistics,” �http://www.gsmworld.com/news/statistics/substats.shtml�.

[15] L. T. W. Ho and H. Claussen, “Effects of User-Deployed, Co-Channel Femtocells on the Call Drop Probability in a ResidentialScenario,” Proc. 18th IEEE Internat.Symposium on Personal, Indoor and MobileRadio Commun. (PIMRC ‘07) (Athens, Gr., 2007), paper 292

[16] L. T. W. Ho, F. J. Mullany, H. Claussen, andL. G. Samuel, “Autonomous Organization ofWireless Network Transport in a Multi-Provider Environment,” Proc. 7th Internat.Symposium on Autonomous DecentralizedSystems (ISADS ‘05) (Chengdu, China, 2005),pp. 719–724.

[17] L. T. W. Ho, L. G. Samuel, and J. M. Pitts,“Applying Emergent Self-Organizing Behaviorfor the Coordination of 4G Networks UsingComplexity Metrics,” Bell Labs Tech. J., 8:1(2003), 5–25.

[18] F. J. Mullany, L. T. W. Ho, L. G. Samuel, and H.Claussen, “Self-Deployment, Self-Configuration: Critical Future Paradigms forWireless Access Networks,” AutonomicCommunication: First Internat. IFIP Workshop,WAC 2004, Berlin, Ger., Oct. 2004: RevisedSelected Papers (Michael Smirnov, ed.), LectureNotes in Computer Science (LNCS), Vol. 3457,Springer, Berlin, Heidelberg, New York, 2005,pp. 58–68.

[19] C. E. Shannon, “A Mathematical Theory of Communication,” Bell Sys. Tech. J., 27:3 (1948), 379–423 and 27:4 (1948),623–656.

[20] United Kingdom, Department for Business,Enterprise and Regulatory Reform (BERR),Broadband: FAQs, July 2004, �http://www.berr.gov.uk/files/file13254.pdf�.

[21] United Kingdom, Office of Communications(Ofcom), The Communications Market 2006,Annual Communications Market Report, Aug.2006, �http://www.ofcom.org.uk/research/cm/cm06/�.

[22] United Kingdom, Office of National Statistics,“Neighbourhood Statistics, Newham 014A,”�http://www.neighbourhood.statistics.gov.uk�.

[23] G. Vincent, “Learning from i-mode,” IEE Rev.,47:6 (2001), 13–18.

(Manuscript approved November 2007)

HOLGER CLAUSSEN is a research engineer in the End-to-End Wireless Networking ResearchDepartment at Bell Labs, in Swindon,United Kingdom. He received his Dipl.-Ing.(FH) and M.Eng. in electronic engineeringfrom the University of Applied Sciences in

Kempten, Germany, and the University of Ulster, UnitedKingdom, respectively. He received a Ph.D. in signalprocessing for digital communications from theUniversity of Edinburgh, United Kingdom, for his workon low complexity MIMO receiver architectures. At BellLabs, Dr. Claussen has been working on auto-configuration and dynamic optimization of networks,self-deploying networks, distributed algorithms, flatcellular network architectures, fourth generation (4G)systems, mobility, resource management, and end-to-end network modeling. He is currently involved intechnology transfer for auto-configuration and self-optimization algorithms required for future basestation router (BSR) products.

LESTER T. W. HO is a research engineer in the End-to-End Wireless Networking ResearchDepartment at Bell Labs in Swindon, UnitedKingdom. He studied at Queen Mary andWestfield College, University of London,where he received a B.Eng. in electronic

engineering with first-class honors, and a Ph.D. on thetopic of self-organization in wireless networks. At BellLabs, Dr. Ho has been working on auto-configurationand dynamic optimization of wireless networks,protocols, and algorithms for flat cellular networkarchitectures, dynamic spectrum allocation, andtechno-economic modeling of future wireless business models. His current research interests includeself-organizing behavior in wireless networks, end-to-end network modeling and analysis, resourcemanagement, and novel business models.


LOUIS G. SAMUEL is a vice president in the ChiefTechnology Office (CTO) of Alcatel-Lucent’sEurope and North region. He is based inSwindon, United Kingdom. Before joiningAlcatel-Lucent, he served in the Royal Navyas a nuclear reactor specialist. He studied at

Queen Mary and Westfield College, University ofLondon, receiving a master’s of communicationengineering and a Ph.D. in the application of non-linear dynamics to teletraffic modeling. At Bell Labs hehas been involved in the development of advancedprotocols and network architectures for wirelesscommunications. His current research interests includenon-linear dynamics, complexity theory, agent basedsystems, software architectures and infrastructures,software protocols, fourth generation (4G) systems,mobility, and resource management. More recently hedirected the initial research that led to the base stationrouter (BSR) and the development of flat cellular IParchitectures, including heavy involvement in thetechnology transfer required to make the BSR aproduct, as well as promotion of the BSR to Alcatel-Lucent customers. ◆

[08]an overview of the femtocell concept_2

Documents