You are on page 1of 8

GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 46

HETEROGENEOUS AND SMALL CELL NETWORKS

Automated Small-Cell Deployment for


Heterogeneous Cellular Networks
Weisi Guo, University of Warwick
Siyi Wang, University of South Australia
Xiaoli Chu and Jie Zhang, University of Sheffield
Jiming Chen and Hui Song, RANPLAN Wireless Network Design

ABSTRACT and the cell density in urban areas is in excess of


6 cells/km2/operator. This has yielded a system-
Optimizing the cellular network’s cell loca- level capacity that is largely interference-limited
tions is one of the most fundamental problems as opposed to propagation-limited.
of network design. The general objective is to Mobile data demands in cellular networks
provide the desired QoS with the minimum sys- occur predominantly (70 percent) in indoor
tem cost. In order to meet a growing appetite areas, while the traditional radio planning strate-
for mobile data services, heterogeneous net- gy is ill equipped to address this issue. The
works have been proposed as a cost- and energy- indoor coverage issue is especially challenging
efficient method of improving local spectral for large buildings such as shopping malls, hotels,
efficiency. While unarticulated cell deployments and enterprise and government offices, where
can lead to localized improvements, there is a multiple indoor surfaces of different electromag-
significant risk posed to network-wide perfor- netic properties impede signal propagation. The
mance due to the additional interference. The typical indoor subscriber density in the afore-
first part of the article focuses on state-of-the-art mentioned buildings is high, but the quality of
modelling and radio planning methods based on service (QoS) delivered to them is currently low.
stochastic geometry and Monte Carlo simula- Three factors motivate cell planning opti-
tions, and the emerging automatic deployment mization: interference, user location, and radio
prediction technique for low-power nodes, or propagation. While a lot of work has gone into
LPNs, in heterogeneous networks. The tech- signal processing and resource management
nique advises an LPN where it should be techniques for mitigating interference, there
deployed, given certain knowledge of the net- have been fewer efforts on the latter two issues.
work. The second part of the article focuses on In this article, we investigate how to optimize
algorithms that utilize interference and physical the cell location subject to the interference pat-
environment knowledge to assist LPN deploy- tern, given a certain user distribution and radio
ment. The proposed techniques can not only propagation model.
improve network performance, but also reduce Low-power nodes (LPNs), such as femto
radio planning complexity, capital expenditure, access points (FAPs) and relay nodes (RNs),
and energy consumption of the cellular network. have been proposed as low-cost and low-energy
The theoretical work is supported by numerical methods for improving local spectral efficiency
results from system-level simulations that employ [1]. Such LPNs are integrated into the existing
real cellular network data and physical environ- cellular network via wired broadband (e.g., asyn-
ments. chronous digital subscriber line [ADSL], optical
fiber), or wireless backhaul (e.g., in-band trans-
INTRODUCTION mission or microwave links). The resulting cellu-
lar network is known as a heterogeneous
Traditionally, cellular network deployment has network (HetNet).
been designed primarily for outdoor coverage A key challenge to such a HetNet is how to
and voice services, which are achieved by over- mitigate the excessive interference in areas that
coming the stochastic nature of the radio propa- traditionally would have good coverage, but now
gation environment. In the past decade, there suffer degradation due to the additional infer-
has been an unprecedented growth in mobile ence created by nearby LPNs [2, 3]. In Fig. 1,
data demand. This has led to revolutions in mul- the mean received signal-to-interference-plus-
tiple access technology, as well as an increase in noise ratio (SINR) in an example HetNet is
cell density and spectrum reuse. The third and shown. The HetNet consists of a sectorized
fourth generation cellular networks mostly macro base station (BS) with 12 LPNs deployed
employ full bandwidth reuse (reuse pattern one), within its coverage area. It can be seen that the

46 0163-6804/13/$25.00 © 2013 IEEE IEEE Communications Magazine • May 2013


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 47

SINR is high near the LPNs, but rapidly falls to


a level that is below the original macro BS serv-
ing SINR in regions surrounding the LPN cover-

Mean received SINR, dB


age areas. This is due to the excessive cross-tier 40
interference.
20

HETEROGENEOUS 0

NETWORK PLANNING -20


400
MODELING
In order to optimize the cellular network perfor- 200
mance through cell site planning and transmis-
sion techniques, different modeling approaches 0
have been taken over the years to characterize
Distance, m 400
the network performance:
-200 200
•Monte Carlo multicell model (simulation):
0
can include multiple effects, which are not easily
-200
describable by tractable mathematical functions, -400
Distance, m
-400
such as ray-traced path loss models, antenna pat-
terns, terrain, clutter, and cell-specific configura- Figure 1. HetNet with a macro BS and randomly deployed femtocells. Femto-
tion data. For specific models, a large volume of cells improve local signal strength but severely degrade the surrounding-area
data is required, and an example is shown in Fig. signal strength due to excessive interference.
2a, where the mean received downlink signal
power is from a major operator’s HetNet in a
European city, with 95 macro and pico BSs mod-
eled. For generic models, a hexagonal cell layout Stochastic models can yield insights on the
with wrap-around is typically employed to obtain impacts of cell density, transmit power, and path
an upper bound of network performance [4]. loss, but they are not well suited to analyze
•Stochastic geometry model (statistical): can effects that are not easily modeled by probability
capture the network-wide performance of a non- distributions such as vertical antenna patterns
uniform network deployment [5], but includes and terrain clutter. Furthermore, stochastic
only stochastic effects that are mathematically models only provide a statistical deployment
tractable. An example for a network with a cer- solution (e.g., the optimal average number of
tain cell density is shown in Fig. 2b. femtocells per macrocell), as opposed to a deter-
•Single-cell linear model (deterministic): can ministic deployment solution (e.g., the optimal
capture the specific performance variations number and locations of femtocells in a specific
across the coverage area of a single cell in a macrocell). The linear model offers a balance
multicell network [6]. An example is shown in between the aforementioned two approaches by
Fig. 2c, where the framework considers only a providing a deterministic deployment solution in
dominant interference source, which clearly has a way faster than simulations.
limitations. Provided that there is always a domi-
nant interference source, scalability in cell densi- CELL SITE PLANNING AND CHALLENGES
ty is not an issue. If scaling the network means Cell site planning has traditionally targeted cov-
that more and more locations suffer equal inter- erage percentage and traffic density. The latter
ference from multiple sources, the linearity of is difficult to characterize, especially given its
the model will break down. dynamic nature, and the shifting trends in usage
Figure 3 plots the cumulative distribution patterns and social mobility. Nonetheless, a
function (CDF) of the received downlink SINR great deal of traffic information is inferred and
across a network with the three modeling forecasted from:
approaches. Simulation parameters for the real- • Demographic data: the residential and busi-
istic network (in a European city) are: 96 realis- ness population distribution based on
tic macro and pico BSs in a 9 km ¥ 6 km area, demographic census
with ray-traced path loss models (PACE 3G) • Traffic data: vehicular data based on public
and realistic antenna patterns. Parameters for transport and private vehicle movement
the theoretical models employ the WINNER patterns
Urban statistical path loss model and omnidirec- • Fixed line data: based on correlation with
tional antenna patterns. This work was conduct- fixed line telephony records, given that
ed at the University of Sheffield with the Mobile most mobile data traffic occurs indoors
VCE (MVCE) and multiple industrial partners On a macro and statistical scale, the stochastic
[7]. framework introduced in [5] can calculate the
We can see that if the realistic European LPN density as a function of the transmit powers,
city’s network is taken as a reference, the statistical path loss exponent, and noise level.
stochastic geometry is quite accurate. The hexag- For radio planning on a micro scale, Monte
onal and linear models can use a backoff factor Carlo simulations are employed along with
to improve their accuracy. The relative merits of detailed urban terrain maps and ray-traced path
each modeling technique are beneficial for dif- loss models. This is recognized as an NP-hard
ferent purposes. Specific challenges typically problem. Given a set of possible cell site or LPN
warrant the use of simulation-based approaches, locations, iterative techniques are usually used to
where custom features can be accommodated. scan the optimal locations for cell sites and

IEEE Communications Magazine • May 2013 47


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 48

LPNs. Optimization methods such as integer pro-


High-resolution gramming, simulated annealing, and multi-era
terrain map of a genetic programming algorithms are employed to
European city search for optimal solutions. Meta-heuristic
methods such as Tabu search [8] can accelerate
the process by ignoring previous negative search
results (within a certain iteration period) that are
stored in a memory. The ultimate deliverable
goal is to make the search complexity linearly
proportional to the number of BSs and user
equipment (UE) devices considered.
To give an idea of the scale and complexity of
the challenge, a typical developed urban
metropolis has approximately 2 BS sites/km 2 /
operator. This equates to approximately 100 BSs
per city, incorporating over 300 macrocells. In
order to deploy LPNs in a HetNet, investigations
carried out by the industry have shown that the
typical number of LPNs required to boost indoor
coverage to outdoor levels ranges from 30 to 100
per BS, yielding a lower bound of 60 cells/km 2
and 3000 cells/operator in a city. The resulting
radio planning complexity for the HetNet is
extremely high, primarily because:
• Cell densification: 30- to 100-fold increase in
(a) cells
• Coverage resolution: 100-fold increase (from
Cell boundary Cell site location 20 m to 2 m) in coverage resolution for
LPNs and indoor areas, and at least a three-
fold increase in coverage height resolution
• Indoor-outdoor path loss complexity:
unknown increase in computation time
which lead to at least a 10,000-fold increase in
the computation time for radio coverage analysis
or prediction. This would increase deployment
planning and, more important, system optimiza-
tion times to infeasible levels. Hence, there is a
temptation to deploy LPNs without articulated
radio planning and rely on signal processing
techniques to improve performance. The danger
with this approach is that in the absence of
effective interference mitigation techniques,
there might be zones of intense interference as
shown in Fig. 1.
(b) The complexity of deploying LPNs and pre-
dicting their performance can be reduced by
Capacity profile finding approximate deployment locations using
key network parameters. In order to avoid or
reduce the complexity of protracted simulations,
analytical methods such as the stochastic geome-
try model proposed in [5] can be used. While
stochastic geometry offers network-wide mean
performance bounds that relate to node density
and other parameters, the challenge of how to
plan each specific BS of a HetNet remains open.
Interfering
BS
BS TOWARD AUTOMATED DEPLOYMENT
In order to gain insight into LPN deployment
location on a single BS level, the latest develop-
ment in network performance modeling includes
the effects of:
• Interference: From the co-channel transmis-
Relay
sion of a dominant neighboring cell [6].
• Capacity saturation: Realistic transmission
(c) schemes suffer from mutual information
saturation in discrete modulation schemes
For example, in the Long Term Evolution
Figure 2. Heterogeneous network modeling methods: a) Monte Carlo simula- (LTE) physical layer, the maximum achiev-
tion of a realistic environment; b) stochastic geometry representation of a able spectral efficiency is 4.3 b/s/Hz for a
multicell network; c) linear model of a single cell in a multicell network. typical outdoor environment. Research in

48 IEEE Communications Magazine • May 2013


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 49

[9] has shown that existing solutions, which


do not consider spectral efficiency satura- 1
tion, lead to a significant waste in radio
0.9
resources.
The work in [6] shows that by jointly consid-
0.8
ering the effects of interference and capacity sat-
uration, the optimization solution is significantly 0.7
different from those of noise-limited channels
without capacity saturation [10]. Automated cell 0.6

CDF of SINR
deployment is a concept that attempts to deter-
ministically find the optimal location of a new 0.5
cell, subject to knowledge about the locations of
existing cells, users, and the propagation envi- 0.4
ronment. This is in contrast to random deploy-
ment or optimization using brute force search 0.3
methods in simulations. While some of the auto-
mated deployment solutions are known to expe- 0.2
Hexagonal wrap-around
rienced radio-planning engineers, the availability Real European city network
of the deployment location in closed-form as a 0.1 Stochastic geometry
function of transmit power, transmission scheme Linear model
and pathloss parameters, is novel and signifi- 0
-10 -5 0 5 10 15
cantly beneficial. The work has been applied to: SINR, dB
outdoor wireless relays [6], and access-points
(APs) for indoor areas [11, 12]. The automated Figure 3. CDF of the network-wide SINR with different modeling approaches.
deployment model has been validated against an
industrially bench-marked multi-cell system sim-
ulator. The following sections provide an the RNs relatively close to the parent BS so that
overview of the automated deployment model the BS-RN channel could be good enough to not
and its impact on the future of HetNet planning. limit the RN-UE channel. However, this may
create two problems in a realistic network:
• UE devices close to the BS already experi-
AUTOMATED OUTDOOR ence close to saturated performance and do
DEPLOYMENT not require relaying.
• RNs are likely to degrade that saturated
MOTIVATION AND METHODOLOGY performance through in-band interference,
For the outdoor cellular network, one of the while offering very little improvement.
largest sources of operational expenditure is the In fact, it was found that unarticulated or
tethered back-haul rental cost. Furthermore, the miscalculated deployment of LPNs may cause
dense nature of outdoor LPNs requires the oper- network-wide spectral efficiency degradation.
ator to balance the optimal coverage LPN loca- As shown in Fig. 4b, in an interference-limit-
tions with the availability of backhaul cabling. ed and saturation-aware channel, the optimal
These have motivated the deployment of wireless RN location is approximately 0.7–0.8 of the
RNs. However, the challenge with allocating macrocell coverage radius (dBS) away from the
scarce spectrum to relaying is not only difficult to BS. The optimal BS-RN distance (dBS-RN* ) that
manage, but also complex to optimize. maximizes the mean network spectral efficiency
For automated cell deployment, the optimal can be expressed as [6]
location of a cell (femtocell or RN) is determin-
1
istically found using an algorithm. A linear −
⎛ P ⎞ α
model proposed in [6, 12] uses the estimated sig- *
dBS-RN ∝ dBS ⎜ γ Sat. RN ⎟ , (1)
nal power received from each cell. The estima- ⎝ P ⎠
BS
tion process considers the transmit power,
statistical path loss, and cell location. The effects where g Sat. is the signal-to-noise ratio (SNR) at
of terrain clutter and antenna patterns have not which the capacity saturates, a is the path loss
yet been considered. However, a realistic system distance exponent, and P RN and P BS are the
can also measure the real signal power received transmit power of the RN and BS, respectively.
from different BSs. The measurements can be The expression shows that the optimal distance
used instead of the estimation method. The from the RN to the BS is inversely proportional
measurements can then be used to optimize the to the transmit power ratio of the RN and the
locations of cells in accordance with the formu- BS, and the constant of that proportionality is
las devised in [6, 12]. the path loss distance exponent. The optimal
BS–RN distance is also transmission-scheme-
THEORY aware, since a lower-order transmission scheme
In [6], the proposed theoretical framework for such as binary phase shift keying (BPSK) (with
wireless RNs accounts for the effects of interfer- lower g Sat. value) leads to the RNs being
ence and capacity saturation. The optimal loca- deployed further away from the BS, in order to
tions of RNs from their parent BS are protect UE devices already experiencing saturat-
fundamentally different to those in a Gaussian ed performance. The proposed RN deployment
noise channel [10]. As shown in Fig. 4a, in a yields an optimal balance between improving the
noise-limited and saturation-free channel, the BS-RN channels and improving the performance
optimal parent BS to RN distance is to deploy of cell edge UE.

IEEE Communications Magazine • May 2013 49


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 50

Another parameter of concern is the number which shows that the optimal number of RNs
of RNs per BS sector that maximizes the spec- per BS sector is inversely proportional to the
tral efficiency of the network. The interference- transmit power ratio of the RNs to the BS, and
and saturation-aware theoretical framework in the constant of that proportionality is the path
[6] shows that the optimal number of RNs per loss distance exponent.
BS sector (N*RN) is upper bounded by Furthermore, due to the radial nature of the
RN deployment framework, the result can be
1
− extended to non-uniform cell geometries with
⎛ P ⎞ α
*
N RN ≤ π ⎜ 2 RN ⎟ , (2)
azimuth antenna patterns, as shown in Fig. 4c.
⎝ P ⎠BS The network-wide spectral efficiency improve-
ment achieved by the proposed automated RN
deployment over the random deployment is
approximately 55 percent for outdoor RNs [6].
Max. QoS While the automated RN deployment solutions
Parent BS capacity relay are known to experienced radio planning engi-
neers, the availability of the solution in closed
form as a function of transmit power, transmis-
sion scheme and path loss parameters is novel
and of benefit by reducing radio planning time.
ius
rad
Cell VALIDATION USING REAL NETWORK DATA
In order to validate the automated LPN deploy-
ment solutions in a realistic outdoor cellular net-
work, the automated deployment algorithm is
applied to data from a real cellular operator’s
network in a developed urban city [7]. Figure 5a
shows the area of focus (data extraction), which
Max. mean
Macrocell edge capacity relay is a 4 km2 central urban area, including approxi-
mately 4 macro BSs and 40 LPNs. The interfer-
(a) ence from 92 other BSs in the city area is also
considered.
The results in Fig. 5b show that the unarticu-
lated random LPN deployment actually
Max. QoS degrades the network performance from that of
capacity relay the homogeneous deployment, which was also
predicted in [6]. Articulated auto-deployment of
LPNs, on the other hand, achieves a significant
improvement in network-wide spectral efficiency
against both the random LPN deployment and
the conventional homogeneous cellular network.
In terms of mean spectral efficiency, the
improvement is approximately 50 percent, which
Max. mean closely matches the theoretical predictions
capacity relay
found in [6].

(b) AUTOMATED INDOOR DEPLOYMENT


Irregular Antenna azimuth MOTIVATION
macrocell edge pattern
In indoor areas, the availability of tethered back-
haul makes the deployment of LPNs or APs an
attractive solution. While the locations of out-
door cells are controlled by operators to meet
network performance targets, there is less under-
standing or control on where indoor LPNs
should be placed. Conventionally, indoor APs
are deployed at locations of convenience.
Indeed, the end user cannot always arbitrarily
decide where an AP can be placed. Recent
research shows that in a strong interference
environment, some regions of a room are more
beneficial than others. The optimal placement of
APs was previously investigated in [13], whereby
iterative computation techniques were used to
find the optimal locations of multiple nodes in
an indoor environment [11]. More recently,
(c) there has been development on how to use sta-
tistical path loss and user distribution parame-
Figure 4. Optimal RN deployment for: a) noise-limited and saturation-free ters to predict the optimal location of an AP [12]
channels; b) interference-limited and saturation-aware channels; c) interfer- without using exhaustive computational algo-
ence-limited and saturation-aware channels with irregular cell coverage. rithms.

50 IEEE Communications Magazine • May 2013


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 51

In indoor areas, the


availability of teth-
ered backhaul makes
the deployment of
LPNs or APs an
attractive solution.
While the locations
of outdoor cells are
controlled by opera-
tors to meet network
performance targets,
there is less under-
standing or control

a)
on where indoor
LPNs should be
Empirical CDF
placed.
1

0.9

0.8

0.7

0.6
CDF of capacity

0.5

0.4

0.3

0.2

0.1 Homogeneous network


Random HetNet
Automated HetNet
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Capacity, b/s/Hz
b)

Figure 5. European city’s spectral efficiency profile for a HetNet: a) location and spectral efficiency map; b)
CDF of spectral efficiency profile [7].

THEORY AND VALIDATION ernment cell databases. Based on spectral


efficiency maximization, the optimal loca-
Recent work in [12] considered both 802.11n * ) can be explicitly found [12]:
tion (dFAP
WiFi APs and LTE FAPs for single- and multi- −1
room buildings. The key findings are that for ⎡ −
2
⎛ dbuilding ⎞ ⎤
maximizing mean spectral efficiency, the APs
*
dFAP ≈ dbuilding ⎢1 + W α ⎜⎝ 1 + d ⎟
⎥ ,
should be deployed with the following steps ⎢
⎣ FAP-BS ⎠ ⎥⎦ (3)
(using FAPs as the example):
• A single FAP should be deployed adjacent where the optimal location dFAP* is taken as
to the external wall that faces the closest the distance from the external wall nearest
outdoor macro BS, as shown in Figs. 6a and to the outdoor macro BS, W is the aggregate
6b, and on the building floor that most penetration loss of the internal walls, dbuilding
closely matches the height of the outdoor is the length of the building, and dFAP-BS is
macro BS [11]. The location knowledge of the distance between the serving FAP and
the macro BSs can be found through gov- the nearest dominant interfering BS.

IEEE Communications Magazine • May 2013 51


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 52

• If more than one FAP is deployed, one FAP initial location input for more protracted simula-
should be deployed as described above, and tion-based deployment optimization software.
the other FAPs should be placed at maxi-
mum mutual distance so that interference
between FAPs is minimized [11], as shown
ENERGY AND COST SAVINGS
in Fig. 6c. A key benefit of deploying a more spectrally effi-
The optimal number and locations of FAPs cient network is so that the carbon footprint and
should be determined sequentially, from the expenditures are reduced. There is already a sig-
lowest number to the highest. In a building with nificant commitment from major wireless opera-
a small number of rooms, inter-FAP interfer- tors to cut their carbon footprint and reduce
ence dominates the indoor network performance operational expenditures.
[11]. While the total power consumption of an
The above theoretical automated deployment LPN is typically small (10–25 W), there are
algorithm employs statistical path loss expres- already hundreds of millions of LPNs across the
sions. The results have been validated against an world, and this figure is set to grow rapidly.
outdoor-indoor system simulator known as Therefore, it is important to consider their eco-
iBuildNet [14], using ray-traced path loss models logical and economical impact. Using bench-
both outdoors and indoors. The simulation con- marked system simulation tools, it was found
figuration is shown in Fig. 6a, and the results that the network-wide spectral- and transmit-
found strongly agree with those predicted by the energy-efficiency improvement achieved by the
theoretical automated deployment algorithm. proposed automated deployment over the ran-
The theory therefore allows two potential bene- dom deployment is approximately 20–50 percent,
fits: automated deployment of cells that are con- depending on the environment [6, 11, 12]. This
scious of mutual interference, and providing leads to a carbon footprint reduction of 7–16
percent and a small operational expenditure
(OPEX) saving of 5–12 percent [7].
Furthermore, as a result of deploying LPNs
Building more efficiently, it can be argued that fewer
Macro BS
LPNs need to be deployed to achieve the same
mean network performance than the reference
system (random deployment). In that case, both
the energy and cost savings are more profound
and can reach 40–50 percent [4, 7].

FUTURE WORK
To the best of our knowledge, relevant cell self-
deployment and self-organization work has been
conducted mainly by Bell Labs and other Euro-
(a)
pean researchers for BSs that can fly or at least
reposition themselves in some way [15]. Howev-
er, it is not yet clear from their work how and
where the cells will reposition themselves and
how the mutual optimization works. The work
conducted in self-deployment provides that
insight. Coupled with certain automated mecha-
nisms, in the future cells can reposition them-
selves in accordance with user patterns, traffic
Interference loads, and interference conditions.
BS One of the key challenges with deployment
optimization generally is that the optimal capaci-
ty location of a cell may not be available for
(b) practical and economic reasons. In that case,
each node should be equipped with certain self-
optimization features such that suboptimal
placement does not exacerbate the network per-
formance. Another reality is that there is a com-
plex balancing act between profit margins from
capacity improvements and those from savings
made on site rental costs.

Interference CONCLUSIONS
BS
This article has given a survey and tutorial of
emerging work on deploying LPNs in a HetNet.
Due to the high node density of future HetNets,
(c) there is a demand for solutions that can reduce
radio network planning and potentially allow
Figure 6. Optimal FAP deployment for maximum uniform coverage in: a) both outdoor and indoor nodes to be deployed
investigation setup in iBuildNet; b) a single room; c) multiple rooms. autonomously or with very little guidance.

52 IEEE Communications Magazine • May 2013


GUO LAYOUT_Layout 1 4/29/13 1:31 PM Page 53

Recent advances in interference- and saturation- heterogeneous networks, self-organization, energy efficien-
cy, and cooperative communications.
aware deployment algorithms can potentially The results show
enable LPNs to be deployed while minimizing S IYI W ANG (siyi.wang@mymail.unisa.edu.au) received his
intercell interference and maximizing network- B.Eng. in communication engineering from Shanghai Uni-
that deploying LPNs
wide spectral efficiency. The theoretical work in versity in 2006 and his M.Sc (with distinction) degree in without location
electrical engineering from the University of Leeds in 2007.
this area is validated with simulation results He is currently working toward a Ph.D. with the Institute optimization can
employing realistic network and environmental for Telecommunications Research at the University of South
data. Australia. The research for his Ph.D. investigates communi- degrade network-
The results show that deploying LPNs with- cating data via the diffusion of nano-particles and is fund-
ed by University President’s Scholarships. His research
wide spectral effi-
out location optimization can degrade network- interests include indoor-outdoor network interaction, small
wide spectral efficiency, while automated ciency, while
cell deployment, machine learning, stochastic geometry,
deployment optimization techniques can provide theoretical frameworks for complex networks, and nano- automated deploy-
a low-complexity solution to intelligent HetNet based particle communications.
ment optimization
rollout. XIAOLI CHU (x.chu@sheffield.ac.uk) is with the Electrical and
Electronics Engineering Department of the University of
techniques can pro-
REFERENCES Sheffield. She received her Ph.D. degree from the Hong vide a low
Kong University of Science and Technology in 2005. From
[1] V. Chandrasekhar, J. Andrews, and A. Gatherer, “Fem- 2005 to 2012, she was with the Centre for Telecommuni- complexity solution
tocell Networks: A Survey,” IEEE Commun. Mag., vol. cations Research at King’s College London. She has pub-
46, no. 9, 2008, pp. 59–67. lished more than 50 journal and conference papers. She is to intelligent
[2] D. Lopez-Perez et al., “Enhanced Intercell Interference Coor- the leading Editor of the book Heterogeneous Cellular Net-
dination Challenges in Heterogeneous Networks,” IEEE works (Cambridge University Press, May 2013), and was
HetNet roll out.
Trans. Wireless Commun., vol. 18, no. 3, 2011, pp. 22–30. Guest Editor of the Special Issue on Cooperative Femtocell
[3] X. Chu et al., “On Providing Downlink Services in Collo- Networks of the ACM/Springer Journal of Mobile Networks
cated Spectrum-Sharing Macro and Femto Networks,” & Applications in 2012.
IEEE Trans. Wireless Commun., vol. 10, no. 12, 2011,
pp. 4306–15. JIMING CHEN (jiming.chen@ranplan.co.uk) received his M.S.
[4] C. Khirallah and J. Thompson, “Energy and Cost Impact and Ph.D degrees in communication and information sys-
of Relay and Femtocell Deployments in LTE-Advanced,” tem from the University of Electronic and Science Technol-
IET Commun., vol. 5, Dec. 2011, pp. 2617–28. ogy, China, in 2003 and 2006, respectively. From 2006 to
[5] M. Haenggi et al., “Stochastic Geometry and Random 2011, he was a research scientist and member of the Alca-
Graphs for the Analysis and Design of Wireless Net- tel-Lucent Technical Academy in Bell Labs at Alcatel-Lucent
works,” IEEE JSAC, vol. 27, no. 7, 2009, pp. 1029–46. Shanghai Bell Ltd.. Currently, he is a senior research fellow
[6] W. Guo and T. O’Farrell, “Relay Deployment in Cellular at Ranplan Wireless Network Design Ltd. (http://www.ran-
Networks: Planning and Optimization,” IEEE JSAC, vol. plan.co.uk), which produces the world leading HetNet
PP, no. 99, 2012, pp. 1–10. planning and optimization tool, iBuildNet.
[7] W. Guo et al., “Energy Consumption of 4G Cellular Net-
works: A London Case Study,” IEEE VTC 2013-Spring, H UI S ONG (hui.song@ranplan.co.uk) is currently the R&D
June 2013. manager at Ranplan Wireless Network Design Ltd.
[8] M. Diallo, S. Pierre, and R. Beaubrun, “A Tabu Search (http://www.ranplan.co.uk), United Kingdom. He contribut-
Approach for Assigning Node-Bs to Switches in UMTS ed significantly to Ranplan’s in-building wireless network
Networks,” IEEE Trans. Wireless Commun., vol. 9, no. 4, design and optimization tool iBuildNet. He obtained his
2010, pp. 1350–59. Ph.D in wireless communications from the University of
[9] A. Lozano, A. Tulino, and S. Verdú, “Optimum Power Bedfordshire in April 2010. His research interests are in the
Allocation for Parallel Gaussian Channels with Arbitrary fields of wireless network planning and optimization tech-
Input Distributions,” IEEE Trans. Info. Theory, vol. 52, niques, next generation wireless systems, HetNet and small
no. 7, 2006, pp. 3033–51. cells (femtocells, picocells and metrocells etc.), MIMO-
[10] B. Lin et al., “Optimal Relay Station Placement in IEEE OFDM link adaptation, propagation modeling, and system-
802.16j Networks,” Proc. ACM Int’l. Conf. Wireless level simulation.
Commun. and Mobile Computing, 2007, pp. 25–30.
[11] S. Wang, W. Guo, and T. OFarrell, “Low Energy Indoor JIE ZHANG (jie.zhang@sheffield.ac.uk) is a full professor and
Network: Deployment Optimisation,” EURASIP J. Wireless holds the Chair in Wireless Systems at the Department of
Commun. and Net., vol. 2012:193, June 2012, pp. 1–15. Electronic and Electrical Engineering, University of Sheffield,
[12] W. Guo and S. Wang, “Interference-Aware Self- United Kingdom. His research interests are focused on radio
Deploying Femto-Cell,” IEEE Wireless Commun. Letters, propagation, indoor-outdoor HetNet planning and optimiza-
vol. 1, no. 6, Dec. 2012, pp. 609–12. tion, small/femtocell, self-organizing network (SON) and
[13] D. Stamatelos and A. Ephremides, “Spectral Efficiency smart environments. Since 2006, he has been an Investigator
and Optimal Base Placement for Indoor Wireless Net- of over 20 research projects worth over £17 million (his share
works,” IEEE JSAC, vol. 14, no. 4, 1996, pp. 651–61. is over £5 million) by the Engineering and Physical Science
[14] RANPLAN, “iBuildNet — Indoor Radio Network Plan- Research Council (EPSRC), the European Commission (EC)
ning Tool,” http://www.ranplan.co.uk/. FP6/FP7, industry, and so on. He was/is one of the investiga-
[15] H. Claussen et al., “Enhanced Intercell Interference tors of some of the earliest projects on Femtocell, wireless
Coordination Challenges in Heterogeneous Networks,” friendly building, and green communications. Since 2007 he
IEEE Consumer Commun. and Net. Conf., 2006, pp. has published over 100 papers in referreed journals and con-
595–99. ferences (e.g., IEEE Transactions on Wireless Communica-
tions/Communications/Antenna and Propagation/Microwave
BIOGRAPHIES Theory and Techniques, IEEE Transactions on Selected Areas
in Communications, and IEEE Communications Magazine. He
WEISI GUO (weisi.guo@warwick.ac.uk) received his M.Eng., is a lead author of the books Femtocells: Technologies and
M.A., and Ph.D. degrees from the University of Cambridge. Deployment (Wiley, Jan. 2010) and Small Cells: Technologies
He is currently an assistant professor and co-director of the and Deployment (Wiley, 2013). He and his colleagues pub-
Cities research theme at the School of Engineering, Univer- lished one of the most widely cited femtocell papers,
sity of Warwick. He is the author of the VCEsim LTE System “OFDMA Femtocells: A Roadmap on Interference Avoidance”
Simulator, and his research interests are in the areas of and some early work on femtocell self-organization.

IEEE Communications Magazine • May 2013 53