Capacity and energy efficiency of picocell deployment in LTE-A networks

fearlessquickMobile - Wireless

Dec 12, 2013 (4 years and 6 months ago)

367 views

Capacity and energy efficiency of picocell
deployment in LTE
-
A

networks


L. Saker
1,2
, S.E. Elayoubi, Letian Rong

1
Orange Labs

38
-
40 rue du Général Leclerc

92130 Issy
-
Les
-
Moulineaux

France

Tijani Chahed

2
Institut Telecom; Telecom SudParis; UMR CNRS 5157

9
rue Charles Fourier

91011 Evry

France


Abstract


In this paper, we show the impact of deploying
picocells on the capacity and energy efficiency of LTE
-
A

networks. We analyze the Erlang
-
like capacity of a network
composed of macro networks only and study
the impact of
introducing a number of picocells per site. Knowing that the
capacity is not the only factor that will drive the evolution of the
network, we also consider the energy efficiency as a Key
Performance Indicator (KPI). Our results show that, in
some
scenarios, introducing picocells is a good network densification
method as they achieve a higher network capacity with good
energy efficiency
.

Keywords
-

small cells, capacity, energy efficiency, LTE
-
A.

I.


I
NTRODUCTION

3G wireless technology has kept imp
roving since the first
releases of the 3G standards. Each new release of 3GPP
specification comes with new features that bring progress both
to consumers and network operators. 3GPP Release 10
specifies LTE
-
A, which introduces several significant
improveme
nts over LTE Release 9 [1], such as relays, carrier
aggregation and Heterogeneous Networks (HetNets). This
paper investigates a special aspect of HetNets, which is the
deployment of picocells that offload a part of the traffic of the
macro network.

A larg
e interest in the literature has been dedicated to
interference management between the macro and pico layers
[2], and integrating this within the framework of Self
Organizing Networks (SON) [3]. We focus in this work on the
Erlang
-
like capacity gains that
are expected from deploying
picos. By Erlang
-
like capacity, we mean the maximal traffic
intensity that can be served by the network with a target
Quality ofr Service (QoS).

On the other hand, energy efficiency of the radio access
network has become a criti
cal issue in future wireless networks
(especially 3G LTE and LTE
-
Advanced). Macro base stations
are the most energy consuming nodes in the access network
and will be the center of any green radio scheme[4
-
5]. We are
thus interested, besides the capacity ga
ins, by the energy
consumption of the resulting heterogeneous network. We
calculate in this paper the energy efficiency, measured by the
traffic capacity divided by the energy consumption, as
recommended by [4], and consider it as a KPI for the quality of
the network. We show that, in some scenarios, picocells
increase the energy efficiency compared to a pure macro
network. This depends on the maturity of the equipments, and
their energy consumption.


The remainder of this paper is organized as follows.
Rel
ated works are presented in Section II. We develop, in
section III, capacity models for LTE
-
A with small cells. The
energy consumption issues are introduced in Section IV.
Numerical results in Section V illustrate the capacity and
energy gains in different

scenarios. Section VI eventually
concludes the paper
.

II.

R
ELATED WORK

An important set of works on green radio has been
dedicated to the reduction of the transmitted power of base
stations. The idea is to find the optimal transmission power
that ensures cove
rage and capacity (see for instance [6] and
[7]). The impact of reducing the cell size on the energy
performance of an HSDPA has been investigated in [8].
Combining a sleep mode with a reduction of the cell size from
R
1

to
R
2
<R
1
, it has been shown that the

energy consumption
gain of a cellular network increases with a factor
n=(R
1
/R
2
)
2

for a given user density and service area.

However, alone,
these schemes are not sufficient to reduce the energy
consumption of wireless networks and a sleep mode is thus
nec
essary for any optimal design of green base stations, as we
have shown in [9].


As for small cell deployments, a particular attention has
been dedicated to energy savings in indoor access points [10]
[11], where sleep mode mechanisms have been used. Out
-
o
f
-
band low consumption radio modules have been used in order
to awaken femtocells when a call arrives.


For outdoor small cells, low power base stations have been
shown to improve the energy efficiency [12]. Their approach
is based on a link budget analysi
s, where the average
throughputs, with and without small cells, are compared for
different cell ranges and deployment scenarios. Although the
obtained results give interesting insights about outdoor small
cell deployments, they do not integrate the dynamic

behavior
of users and its impact on the user
-
perceived QoS.


In this paper, we introduce both link budget and flow level
queuing theory analyses. The main constraint being to preserve
the QoS of users, we derive capacity and energy efficiency
figures that

consider the traffic intensity and the number of
activated pico cells
.

III.

M
ETHODOLOGY

A.

Link budget calculations

For the analytical calculations proposed in this paper, a
uniform network confi
guration is considered. A User
Equipment (UE)
M

in the target cell
0

is characterized by a
distance
R

to base station
0

and an angle

θ

with a re
ference
axis as shown in

Figure

1. The points at

the center of all
surrounding cells

indicate the interfering eNBs.
For each
point
of the cell
, we derive the
SINR

received from the

target eNB
and the different picocells. The aim is to calculate the
throughput of a Resource Block (RB) that is allocated to a UE
located in this point and that is served by either the macro cell
or the pico one. For these calculations, we take into accou
nt
interferences

from surrounding pico as well as macro

cell
s
.



Figure
1
: Network configuration with 3 picos per site.


W
e consider
, for each node
i

(pico or macro cell),

an
average load

in the interfering cells equal
t
o

. This means
that we have the
same average proportion of occupied
resources in each of the cells,
i
tot
i
K
K
,


,

where
K
tot
,i

is
the total number
of download resource blocks available for
cell
i

and
K
i

is the num
ber of occupied resou
rce blocks.


The same formula describes thus

the distribution of the
number
of collisions at each RB
, be it allocated to a user
served by an eNB or a Picocell.
W
e define the vector
X
of
zeros and ones whose dimension is equal to the number of
neighboring
cells and whose elements take the value 1 if there
is a collision with the corresponding cell.
As for macro
networks studied in [13], t
he probability distribution of the
number of collisions is calculated by:

X
X
X
X
X
t
t



1
)
1
(
)
,
Pr(




where
X
t

is the transpo
se of
X
.


For a given vector of collisions X, the average SINR at each
point of the cell, characterized by distance r to the macro eNB
and angle

θ with a reference axis (as shown in Figure 1) can be
calculated for the two different links. These are the direct link
between

the eNB and the UE (
SINR
eNB
-
UE
), and the second link
between the picocell and the UE (
SINR
pico
-
UE
)
:






X
X
n
i
i
i
i
N
r
q
P
X
r
q
P
r
SINR
1
0
0
0
)
,
(
)
,
(
/
)
,
Pr(
)
,
,
(






wher
e
P
0

is the emitted power of the serving node,
q
0

is the
path loss between the receiver and the serving node,
P
i

is the
emitted power of the interfering node,
q
i

is the path loss

between the receiver and the interfering node and
N
0

is the
thermal noise.


T
he UE will then be connected to the link offering the best
quality, leading to the final SINR equation:










)
,
,
(
),
,
,
(
max
)
,
,
(






r
UE
pico
SINR
r
UE
eNB
SINR
r
SINR

Using the derived SINR values,
link level curves can be used to
obtain the throughput
of an RB allocated to a UE in the
differe
nt points of the cell. An example of these curves is
shown in Figure 2
.
W
e obtain
, by multiplying this throughput
by the number of RBs available for each cell,
K
tot
,i

(
i
=eNB or
PicoCell), the peak throughput
over all the areas covered by
node
i
, assuming t
hat all
the
resources are allocated to one user.

0
5
10
15
20
25
30
35
0
100
200
300
400
500
600
SINR (dB)
RB data rate (kbps)

Figure
2
-

Downlink LTE link curve


Let
D
c

be this throughput, calculated at a given point,
c
, of the
cell and
N

the number of points in the considered grid.

B.

Flow level capacity ana
lysis

In the previous section, we studied, using link budget
analysis, the throughput that can be achieved at the different
positions of the cell, for a stand
-
alone user. This is not
sufficient as, in high traffic situations; there are several users
that a
re scheduled in parallel. A flow
-
level capacity analysis,
with users that arrive and depart dynamically to the cell, is
thus necessary.


Figure
3

Queuing model

(macro cell with k picocells)

A

network carrying elastic traffic

and c
omposed only of
macro cells

can be modeled as a Processor Sharing queue, and
its evolution described by the overall number of users i
n the

cell




C
c
c
n
n
1
[14]
. The cell load

is

calculated by

ρ
(
λ
)
=
λ
F/D
,

where
F

is the average file size,

λ

is th
e overall arrival rate
and


1
1












C
c
c
c
D
D




is
the harmonic mean of the throughput.
The probability of
having
n

active users in the cell is given by:

)
1
(
)
(





n
n


If a target throughput
T

is sought,
we show that
the
probability, for us
ers
in position

c
, to achieve this throughput
is given by:

1
1
)
1
(
1
)
(
)
(








c
c
c
k
c
k
c
k
m
c
k
k
m
QoS






k
c

= D
c

/T

is the maximal allowable number of users in the cell
such that the throughput of a user in position
c

is acceptable
.
The average QoS over the cell

is thus:

)
(
)
(
1




c
C
c
c
QoS
QoS




When picos are introduced, the cell site is modeled as a
network of processor sharing queues as shown in Figure 3,
where a part of the traffic is served by each of the nodes
(macro or picos).

The traffic of each node is gi
ven by the
proportion of cell points whose best server is that node. Let
S
pi

and
S
m

be the area of surface covered by the picocells
i

(i=1…
k)

and the macro cell. These areas are calculated by the
proportion of cell points covered by each node (macro or
pi
cos). The arrival rate λ
pi

of picocell
i

is equal to

S
S
pi
pi




and the arrival rate λ
m

of macro equals to

S
S
m
m




with
S

the area of cell.


IV.

E
NERGY CONSUMPTION

A.

Macro base stations

E
mpirical studies give the total energy consu
mption of the
base stations function of the cell load for different
configurations (number of sectors, installed baseband
processing capacity)
, as shown in [5]:


)
/
)
(
(
)
(
c
P
P
m
P
P
TRX
p
macro
max







where

m

is
the number of installed sectors,
P
p

is the power
con
sumptio
n of the processing unit,
P
TRX


is

the fixed power
consumption of the radio

module,

P
max

is the m
aximum output
power and

c

is the direct current to Radio Frequency (RF)

conversion factor.

The load
ρ

is calculated, as shown in the
previous section, function of the offered traffic.

Following
practical
values for
the eNB model parameters
are
proposed in the table below:




P
p

110W

P
TRX

100W

P
max

40W

c

0.32

Table
1

-

BTS mo
del
parameters


B.

Pico base stations

While 2/3 of macro eNB power consumption could come
from the power amplifier, picocells BTS use low RF power
.
As a result, p
ico cells power consumption is load
-
independent.
We consider a constant power consumption (
P
pico
)

that may
vary from 10 to 70 W (actual consumption of picocells is
about 60 W, but this may decrease as the maturity of
equipments is constantly enhanced).


Considering both the macro and the
K

deployed picocells, t
he
energy consumption

of the network nod
es

is given by:

pico
P
K
P
P
macro
cell
.
)
(
)
(





The energy efficiency of the network can be defined as the
ratio between the capacity of the network and the energy
consumption. This is expressed in a capacity/W/unit surface.
Note that the term "capacity" has to b
e precisely defined. From
an operator point of view, the capacity is the maximal amount
of traffic that can be served by the network, under a target
QoS. This is expressed by:

target
max
max
QoS
)
(
;




QoS
C


V.

R
ESULTS

In order to evaluate the

picocells

deployment a
nd the sleep
mode scheme
, we show here some numerical results in which
the

energy consumption and

QoS gains are illustrated
.

We consider an LTE
-
A network in a dense urban region with
an inter
-
site distance of 800 meters
. Picocells have a
transmission power

of 30 dBm and are deployed near the cell
edge (at 450 meters from the macro eNB)
.



A.

Picocells deployment

In order to compare the effect of different picocell
densities,

four strategies are considered:



Macro only
:

no pico sites are deployed.



Light
-
pico
:

deployi
ng 3

pico
s/
sites

(1 pico per cell)
.



Medium
-
pico
:
deployi
ng 6

pico
s/
sites

(2 picos per cell)
.



High
-
pico
:

deployi
ng 9

pico
s

/
sites

(3 picos per cell)
.


10
15
20
25
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Traffic (Mbps)
Outage rate


Macro only
Light-pico
Medium-pico
High-pico

Figure
2:

Cell o
utage rate
(
P
pico

= 30W)


In figure 2, we plot the cell outage rate,

i.e. the pr
obability
that the throughput is less than the target 500 Kbps. In this
figure, the traffic increases from 10 to 25 Mbps/cell, for the
four deployment scenarios. Note that this outage rate is
inversely proportional to the user QoS.

From this figure, if a
t
arget satisfaction rate of 95% is sought, we can derive the
Erlang
-
like capacity (first line of Table 2).

As of the energy consumption shown in Figure 3, it
naturally increases when picocells are deployed in the
network. From Figures 2 and 3, we can obtai
n the energy
consumption corresponding to the maximal capacity that is
supported by the network (second line of Table 2). Note that,
even if the energy consumption is increased, the deployment
of picocells is driven by capacity needs as they increase the
o
verall Erlang capacity.

10
15
20
25
200
220
240
260
280
300
320
340
360
380
Traffic (Mbps)
energy consumption per cell (W)


Macro only
Light-pico
Medium-pico
High-pico

Figure
3:

Energy consumption in the cell (P
pico

= 30W)



Macro
only

3 picos

6 picos

9 picos

Capacity
(Mbps/cell)

15

16

17

19
.5

Energy
(W/cell)

252.66

284.6

314.74

346.67

Table
2

C
apacity and energy co
nsumption of different
strategies with
P
pico

= 30W

As discussed in the previous section, the energy efficiency is
measured by a capacity/W/unit surface. As we consider a
constant inter
-
site distance for macro cells, we can measure
this efficiency Kbps/W/ce
ll. Table 3 compares the energy
efficiency for the four strategies described above with
different values of
P
pico
, the energy consumption of one
picocell.

We observe that the energy efficiency increases when
deploying picocells if they do not consume a lot

(
P
pico
< 30W).
Starting from this point, the efficiency of the resulting
heterogeneous network starts decreasing.



P
pico

10

W

20

W

30

W

40

W

50

W

60

W

70

W

Macro
only

60

60

60

60

60

60

60

3 picos

60.5

58.5

56.5

54.5

52.5

51

49.5

6 picos

62

58

54.5

5
1

48

45.5

43

9 picos

6
8

62

56

5
1

48

44
.5

41
.5

Table
3

-

Energy efficiency

(Kbps/W/cell)

with different
values of
P
pico


VI.

C
ONCLUSION

In this paper, we studied the impact of picocells on the
performance of LTE
-
A networks
.

We calcul
ated the capacity
gains and the resulting energy efficiency in different
deployment scenarios. We showed that picocells represent a
good network densification option, in terms of achieving a
higher throughput at reasonable energy cost, provided that the
en
ergy consumption of picocells is not too high.


As of future works, we aim at designing distributed sleep
mode techniques that enable activating/deactivating picocells
in a dynamic way, following the traffic fluctuations.

A
CKNOWLEDGMENT


Th
is work was par
tially supported by the European Celtic
project OPERANET
.

R
EFERENCES


[1]

3GPP TS 25.
306

V
9
.
2
.0
,

“Physical layer procedures (FDD)”, release
9.

[2]

H.

Viswanathan
, A. L. Stolyar, “
Interference Management in
Femto/Small Cell and Macro Environments
," IEEE CTW 2010
.

[3]

A
. L. Stolyar, H. Viswanathan,, “
Self
-
organizing Dynamic Frac
tional
Frequency Reuse for Best Eff
ort T
raffi
c Through Distribut
ed Inter
-
cell
Coordination,"
INFOCOM'2009,

Rio
-
de
-
Janeiro, April 19
-
25, 2009.

[4]

ETSI TS 102

706, “
Energy Efficiency of Wireless Access

Network
Equipment
”, 2009.

[5]

R. Rinaldi, G. M. Veca, ”The hydrogen for base stations”, in Proc. of
Telecoms. Energy Conference, INTELEC’07, pp. 288
-
292, Sept. 2007.

[6]

J. Kelif, M. Coupechoux and F. Marache, Limiting power transmission


of green cellular networks: Impact on coverage and capacity, IEEE ICC


2010.

[7]

J.Palicot, Cognitive Radio: An Enabling Technology for the Green
Radio

Communications Concept, IWCMC, June 2009.

[8]

B. Badic, T. O’Farrell, P. Loskot, J. He,



Energy Efficient Ra
dio Access
Architectures for Green Radio
:

Large versus Small Cell Size
Deployment
,” IEEE VTC
-
fall 2009.

[9]

L.
Saker
, S.
Elayoubi1
, T.

Chahed
“M
inimizing energy consumption via
sleep mode in green base stations
,

”, IEEE WCNC
-
2010 Sydney.

[10]

I. Haratcherev, M. Fi
orito, C. Balageas,
Low
-
power sleep mode and out


of
-
band wake
-
up for indoor Access Points, IEEE GreenComm09,





December

2009.

[11]

H. Claussen, I. Ashraf, L.T.W. Ho, Dynamic Idle Mode Procedures for

Femtocells, Bell Labs Technical Jour
nal, in press, 2010.

[12]

F. Richter, A.
J. Fehske, and G. P. Fettweis, “Energy effi
ciency aspects
of base station

deployment strategies in cellular networks,


in
Proceedings of the 70th Vehicular

Technology Conference (VTC Fall),
September 2009.

[13]

S.E. Elayoubi
, O
. B. Haddada, and B. Fourestie, “
Performance
Evaluation of Frequency Planning
Schemes in OFDMA
-
based
Networks”
, IEEE Trans. on Wireless Communications, vol.7, no.5, May
2008.

[14]

T. Bonald and A. Proutière, "Wireless downlink data channels: User
Performance

and cell dimensioning", ACM Mobicom 2003.