Multi-Carrier Macrocell Femtocell Deployment-A Reverse Link Capacity Analysis

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Multi-Carrier Macrocell Femtocell DeploymentA

Reverse Link Capacity Analysis


Venkatesh Ramaswamy and Deepak Das
Airvana, Inc.
19 Alpha Road,
Chelmsford, MA 01824
Email: {vramaswamy, ddas}@airvana.com
Abstract In this paper, we study the Reverse Link (RL)
capacity of a co-channel macrocell-femtocell network, where the
macrocell and femtocell networks share the same carrier. The
analysis is done based on an outage probability criterion, and
we use this analysis further to compare the capacities of different
practical multi-carrier macrocell-femtocell deployment strategies.

I. I NTRODUCTION AND S YSTEM M ODEL


Several cellular operators are in the process of deploying
femtocells or home base-stations to improve indoor coverage
for both voice and data users. Femtocells are seen as a low
CAPEX/OPEX, scalable way to improve wireless user experience indoors, and additionally improve macrocell network
capacity by alleviating the load on the macrocell network. In
[1], we carried out RL capacity analysis of a network of femtocells assuming that femtocells were deployed on a dedicated
carrier. However, frequency spectrum is expensive and scarce
and co-channel macrocell-femtocell deployment could easily
be the norm. This motivates the need for models to analyze
RL capacity for mixed macrocell-femtocell deployment.
We employ the theory of Poisson point process to model the
spatial dynamics of macrocell users, femtocell BSs, and femtocell users within a femtocell. Several previous studies have
confirmed the appropriateness of Poisson model in modeling
the mobility of macrocell users and ad-hoc nature of femtocell
deployment [2] [3]. Specifically, we assume, similar to [2] that
macrocell users and femtocell BSs are distributed in a sector
as a homogeneous Spatially Poisson Point Process (SPPP)
with intensities m and f , respectively. Furthermore, each
femtocell has a Poisson distribution of actively transmitting
users with intensity u . A femtocell is assumed to be much
smaller than the macrocell. Femtocell BS is seen as a point
source of interference by the macrocell BS, and by other
femtocell BSs. For analytical tractability, all femtocell users
are assumed to be transmitting with maximum power, i.e., they
are assumed to be at the edge of the femtocell. This implies
that the capacity calculations are somewhat pessimistic, but,
we believe, indicative of key trends.
To compute the RL capacity of the system in terms of the
outage probability, we need to model different interference
components experienced at the macrocell BS and the femtocell
BS. We assume perfect power control, path-loss and lognormal shadowing effects and ignore fast multipath effects
as we are interested in performance over long time-scales.

The interference experienced at the macrocell BS can be


attributed to same cell macrocell users, out-of-cell macrocell
users and femtocells. These interference terms are denoted by
in
out
, Imm
, and If m , respectively. If the number of macrocell
Imm
users in a sector has Poisson distribution, then it follows that
in
will have a Poisson distribution. For the macrocell BS,
Imm
out-of-cell macrocell users and femtocells form a Poisson
field of interferers. Lowen et al. in [4] characterized the
interference due to Poisson field of interferers under several
different settings. If we assume that there are a large number
of interferers, located between points A and B from the
BS, where A > 0 and B < , then from Fig. 3 in [4],
out
, and If m will closely approximate a
the distribution of Imm
Gaussian distribution. Out-of-cell macrocell users have to be at
least Rm (radius of a macrocell) units away from the macrocell
BS, and we assume the interference from macrocell users
outside the third tier is negligible. Likewise, we assume that a
femtocell interferer can come no closer than the diameter of a
femtocell to a macrocell BS, and the interference experienced
at the macrocell BS due to femtocells outside the macrocell
radius is negligible.
It now remains to model the interference experienced at a
femtocell BS. Femtocells are deployed in an ad-hoc manner,
and the interference experienced by a femtocell can vary
widely depending on the location of the femtocell with respect
to the macrocell BS. The interference experienced at a femtocell BS is due to in-cell femtocell users, macrocell users and
other femtocells. Let us denote these three interference terms
in
in
by Ifinf , Imf , and Ifout
f , respectively. If f is similar to Imm , and
using the same argument, it is straightforward to observe that
it has a Poisson distribution. Again, assuming that macrocell
users can only be located outside a femtocell BS, and the
interference caused by macrocell users farther than Rm units
from the femtocell BS is insignificant, the distribution of Imf
will approximate a Gaussian form. Likewise, by restricting the
distance between two femtocell BSs to at least 2Rf (twice
the femtocell radius), and ignoring the interference caused by
femtocells that are more than Rm units away, we can safely
assume a Gaussian distribution for Ifout
f.
Figure 1 shows a macrocell-femtocell network layout and
illustrates various interference terms and their distributions.
The interference modeling and capacity analysis used in this
paper is similar to the more general model by Chandrasekhar

978-1-4244-2515-0/09/$25.00 2009 IEEE

If m Gaussian
out
Gaussian
Imm
Ifout
f Gaussian
in
Poisson
Imm
Imf Gaussian
in
If f Poisson

If m
Femtocell

out
Imm

or more walls. The APL term is only relevant to the


interference terms If m , Imf and Ifout
f;
Rm and Rf : Radius of a macrocell and a femtocell,
respectively; Rm = 1 unit and Rf = Rm /25;
: The total thermal noise power in the spread bandwidth.
III. C OMPUTATION OF R EVERSE L INK C APACITY

Ifout
f
Ifinf

Imf

Macrocell user

in
Imm

Fig. 1. Layout of a macro-femto cellular network. The different interference


components and their distribution are also highlighted.

et al. [2]. In particular their model assumes a Levy-stable distribution for Ifout
f and If m . While their primary focus is to use
their capacity model to analyze different interference reduction
techniques such as time-hopped CDMA, our objective is to
evaluate different spectrum allocation policies for multi-carrier
femtocell-macrocell deployments. In a more recent work [5],
the authors propose frequency band assignment for multicarrier femtocell-macrocell deployments based on forward link
path-loss criteria.
II. D EFINITIONS
In this section we define all the parameters used in our
model highlighting their preferred values.
(Eb /N0 )m and (Eb /N0 )f : Bit energy to noise density
ratio of macrocell users and femtocell users, respectively;
tar
tar
(Eb /N0 )m and (Eb /N0 )f : Target Eb /N0 required for
macrocell users and femtocell users, respectively. We
assume a value of 3 dB for these two parameters;
PG: CDMA processing gain; assumed to be 128 in
numerical examples;
m , f , and u : Poisson intensities of macrocell users in
a sector, femtocell BSs in a sector and users in a femtocell
BS;
r
r
Sf and Sm : Received signal power from femtocell users
to their controlling femtocell BS and the received signal
power from a macrocell user to its controlling base
station. Without loss of generality, we assume they are
r
= S = 1 mW;
equal. That is, Sfr = Sm
and : Outdoor and indoor path loss coefficients; =
4 and = 2 from [2];
T and S : Log-normal shadowing for an interferer to
the base station of interest and to its controlling base
station, respectively. Each of these random variables has
zero mean, and standard deviation = 8 dB;
APL: Aggregate Penetration Loss, i.e., loss due to signal
passing through walls of a house. We assume a loss of
18 dB in signal power if the signal passes through one

We define the reverse link capacity of the system as the


maximum number of macrocell users and femtocell BSs
the network can support such that the outage probability
experienced by macrocell and femtocell users is below a
threshold. The macrocell (femtocell) outage probability is
defined as the probability that the bit energy-to-noise density
ratio (Eb /N0 )m ((Eb /N0 )f ) of macrocell (femtocell) users
is below a minimum (Eb /N0 ) required for reliable system
performance.
The received Eb /N0 for macrocell users can be specified as
 
r
Eb
PG Sm
.
=  in
(1)
out
N0 m
Imm + Imm + If m +
Outage probability is then given by
 
 tar 
Eb
Eb
m
<
.
Pout = P
N0 m
N0 m

(2)

Substituting (1) into (2), and ignoring thermal noise, we can


rewrite (2) as
 out

I
If m
I in
m
> mm ,
= P mm +
Pout
(3)
S
S
S
out
. In our model, Imm
and If m are
where = (EbPG
/N0 )tar
m
in
Gaussian and Imm is Poisson. Conditioning on the number
of active in-cell macrocell users, and using the fact that sum
of two Gaussian random variables is also a Gaussian random
variable, we can compute the macrocell outage probability as

m
Pout

exp(m )
km
1 exp(m )
k!
k=1


(k 1) (mm m + f m f )

, (4)
Q
2 + 2 )
(mm
m
fm f

where Q() is the Q-function for the standard Gaussian


2
m are the mean and variance
distribution, mm m and mm
out
of Imm /S, and f m f and f2 m f are the mean and variance
of If m /S.
We can compute the outage probability using the above
equation once we determine the first and second moments of
out
/S and If m /S. In order to compute these moments, we
Imm
need to first characterize the associated interference terms. Let
rT be the distance of an arbitrary macrocell user to the BS
of interest, and rS be its distance to its controlling BS. The
interference experienced due to all macrocell users in all other
cells, denoted by region A0 is given by
 

out
rS
Imm
(T S )/10
=
10
m dA0 . (5)
S
rT
A0

Now it is straightforward to compute the mean and variance


as follows:
 out 


I
E mm
= mm m = E 10(T S )/10
S
   
rS
m dA0 .
(6)
rT
A0

interest. This outage probability is given by


f
Pout

exp(u )
ku
1 exp(u )
k!
k=1


(k 1) (f f f + mf m )

, (12)
Q
2 )
(f2 f f + mf
m

where f f f and f2 f f are the mean and variance of Ifout


f /S,
2
and mf m and mf m are the mean and variance of Imf /S,
out
Imm
I out
I out
2
m = E
Var mm = mm
E2 mm . (7) respectively. Now what remains is the specification of Ifout
f /S,
S
S
S
and Imf /S. Let ro be the distance of an arbitrary femtocell
from the femtocell of interest. The interference due to all the
As mentioned earlier, in order to characterize the interferfemtocells is given by
ence due to a femtocell, we consider the femtocell as a point

Ifout
source of interference transmitting at a power equal to the
f
=
u 10(T S )/10 10(APL/10)
sum of transmit powers of all its users. Further, for simplicity,
S
A2

every femtocell is assumed to have u users with all the
Rf
users located at the femtocell edge causing maximum possible
(13)
f dA2 ,
ro
interference. Therefore, the equivalent interference a femtocell
creates is equal to u times the maximum possible interference where A2 specifies the entire region where femtocells that
a femtocell user can create. With this simplifying assumption, can cause interference on the femtocell of interest are present.
we can characterize the interference caused by femtocells on Likewise the interference due to all macrocell users is given
a macrocell BS as
by


If m
Imf
=
u 10(T S )/10 10(APL/10)
=
10(T S )/10 10(APL/10)
S
S
A1

 A 3 
Rf
rm
m dA3 .
(14)
(8)
f dA1 ,
ro
ro




2 

where ro refers to the distance of an arbitrary femtocell BS


to the macrocell BS, and A1 refers to the entire region where
interfering femtocells can be located. Proceeding in a similar
way as (6) and (7), we can compute f m f and f2 m f .
Thus far, we were concerned with computing the outage
probability of macrocell users. We now turn our attention to
computing the outage probability of femtocell users. We start
by defining the bit energy-to-noise density ratio of femtocell
users as
 
PG Sfr
Eb

=  in
.
(9)
N0 f
If f + Ifout
f + Imf +
The outage probability now takes the following form
 
 tar 
Eb
Eb
f
<
.
Pout = P
N0 f
N0 f

(10)

Substituting (9) into (10), and ignoring thermal noise, we can


write (10) as
 out

If f
Ifinf
Imf
f
+
>
.
(11)
Pout = P
S
S
S
where =

PG
.
(Eb /N0 )tar
f

In our model, we assumed that Ifout


f

and Imf are Gaussian and Ifinf is Poisson. We can compute


the femtocell outage probability directly by conditioning on
the number of active femtocell users within the femtocell of

The region denoted by A3 defines the entire region over which


macrocell users that can cause interference on the femtocell
out
/S, we
of interest are present. As shown in the case of Imm
2
2
can compute f f f , f f f , mf m and mf f easily from
(13) and (14).
IV. D IFFERENT D EPLOYMENT S CENARIOS AND THEIR
C APACITY R EGIONS
Channel assignment for coexisting macrocells and femtocells can be done in several ways. Using two carriers f1
and f2 as example, we study four different channel assignment schemes to highlight factors that influence the capacity
of channel allocation policies for mixed macrocell-femtocell
deployment. For the purpose of comparison, we introduce a
metric called capacity region C, defined as all pairs (m , f )
that simultaneously satisfy the macrocell and femtocell outage
criterion on an average, without leaving any coverage holes,
where a coverage hole is when either a femtocell or macrocell
does not meet its outage performance target. That is, capacity
region is the region specified by
f
m
0.1, Pout
0.1 femtocell BSs}. (15)
C = {(m , f ) | Pout

Note that by our definition the outage performance within


the capacity region described above is, on average, at least as
good as the outage target.
The numerical results provided in the subsequent sections
are obtained as follows. For each carrier, we compute the set of
all pairs (m , f ) that can be supported such that the macrocell

A. Dedicated Carrier Deployment


Fig. 2 depicts the case where a dedicated carrier is assigned
to all femtocells and another to the macrocell network. In our
example we allocate the entire carrier f1 to the macrocell
network and carrier f2 to the femtocell network. In Fig. 3,
we show the outage capacity plot (no. of users versus the
outage probability) for macrocell users in f1 and that for
femtocells in f2 , and the capacity region (shaded in green)
for the deployment.

Fig. 2. Dedicated Carrier Deployment. Carrier f1 is allocated for macrocells


and carrier f2 for femtocells. For f1 we show multiple sectors with macrocell
users. For f2 , a single macrocell area is expanded to show the femtocells with
femtocell users.

On carrier f1 , we can support up to 28 macrocell users


per sector at Eb /N0 =3 dB at an outage probability of 0.1.
Similarly, at 0.1 outage probability, we can support 139
femtocell BSs per macrocell sector. Note that each femtocell
supports five users, each at Eb /N0 =3 dB, accounting to a total
of 695 femtocell users per macrocell sector. The area shaded
in the third plot shows the capacity region for the deployment.
As long as the number of macrocell users and the femtocell
BSs are within the green region, we can guarantee that both
femtocell users and macrocell users will satisfy the Eb /N0
requirements.
B. Shared carrier deployment
In shared carrier deployment, macrocell and femtocells
share both f1 and f2 as shown in Fig. 4. As noted earlier,
the outage experienced at a femtocell BS depends on its
distance from the macrocell BS. A femtocell at the cell
edge will experience the highest amount of interference from
surrounding macrocell users, whereas the one at the macrocell
BS will experience the least amount of interference from
macrocell users. We show the outage experienced at celledge and cell-center femtocell BSs in first two plots in Fig.
5. With two macrocell users per sector, we can support up

Outage probability for macro users in f

Outage probability for femtocell users in f

10

10

Pm

Pfout

10

out

10

10

10

10

20

40

60

80

10

100

No. of macrocell users/sec

50

100

150

No. of femtocell BSs/sec


Operating region for split frequncy deployment

150

No. of femtocell BSs/sec

outage is less than 0.1 using (4). Similarly, we use (12) to


compute the set of all pairs (m , f ) that can be supported
such that the femtocell outage is less than or equal to 0.1
for two femtocellsone that is farthest from the macrocell
BS and the other that is closest to the macrocell BS, i.e.
the two femtocells that capture two extremes of femtocell
outage. Finally, the set of all pairs (m , f ) that form the
intersection of the above three sets form the capacity region.
The capacity region for f1 and f2 together is obtained by
adding the capacity regions of f1 and f2 .

100

50

50

100

150

No. of macrocell users/sec

Fig. 3. Outage capacity for each carrier and the capacity region for dedicated
carrier deployment.

Fig. 4. Shared carrier deployment. Both f1 and f2 are shared by macrocell


and femtocell users.

to approximately 52 femtocell BSs per sector without any


coverage hole. Each macrocell user has an outage probability
no greater than 0.1, and each femtocell user in every femtocell
has an outage probability no greater than 0.1. With four or
more macrocell users, we cannot avoid a coverage hole, i.e.,
femtocell BSs cannot operate on the macrocell boundary. So
we show the femtocell BS capacity as zero. There seems to
be a dramatic deterioration due to the presence of macrocell
users on femtocell BS capacity. Essentially a cell-edge macrocell user seems to have a devastating effect on the performance
of a nearby femtocell BS. The outage capacity for macrocell
users is captured in the third plot.
In this deployment, the capacity region for f2 is the same as
that for f1 . So the capacity region for f1 and f2 together will
be twice the capacity region for carrier f1 . The green shaded
area in Fig. 6 is the RL capacity region (without coverage
holes) for the entire deployment.
C. Mixed Carrier Deployment (Option I)
Fig. 7 illustrates this deployment scenario. In this deployment we partition the sector into two regions (shown as shaded
and unshaded regions in the figure.) This partition is shown at
a distance of Rm /2 from the macrocell BS (the partition can

Outage probability for celledge femtocells in f

Outage probability for cellcenter femtocells in f

10

10

macrocell users/sec = 0
macrocell users/sec = 100
macrocell users/sec = 200
macrocell users/sec = 400
macrocell users/sec = 600

Pfout

Pfout

macrocell users/sec = 0
macrocell users/sec = 1
macrocell users/sec = 2
macrocell users/sec = 3
macrocell users/sec = 4
1

10

10

10

50

100

10

150

50

100

No. of femtocell BSs/sec

150

Fig. 7.

No. of femtocell BSs/sec

10

300

Capacity of macrocell users


Capacity of femtocells at R /2
m

250

10

10

15

20

25

30

35

40

45

50

No. of macrocell users/sec

Fig. 5.

No. of femtocell BSs/sec

Pm
out

Femtocell BSs/sec = 0
Femtocell BSs/sec = 100
Femtocell BSs/sec = 200
Femtocell BSs/sec = 300
Femtocell BSs/sec = 400

10

Capacity of cellcenter femtocells

200

150

100

50

Outage capacities for shared carrier deployment.

10

Capacity of celledge femtocells


Capacity of cellcenter femtocells
Capacity of macrocell users

700

20

30

40

50

60

No. of macrocell users/sec

800

No. of femtocell BSs/sec

Mixed carrier deployment (option I).

Outage probability for macrocell users in f1

Fig. 8.

Capacity region for mixed carrier (I) deployment.

600

500

Fig. 8 shows the capacity region for this deployment. For


sake of brevity, we have not shown the steps to arrive at the
capacity region, but the procedure is same as that for previous
deployment scenarios.

400

300

200

D. Mixed Carrier Deployment (Option II)

100

10

20

30

40

50

60

70

80

90

100

No. of macrocell users/sec

Fig. 6.

Capacity region for shared carrier deployment.

be further optimized, but we chose that as half the cell radius


for this example). In f1 , macrocell is deployed in the entire
sector, but femtocells are restricted to the shaded region. We
dedicate f2 only for femtocells. The main advantage with this
deployment is that operators do not need to change existing
macrocell deployment. Simulations have shown that femtocells
at macrocell-edge are the ones most affected by macrocell
user interference. In this deployment, we control the extent
to which femtocells are affected by macrocell-edge users.
Another advantage of this deployment is its flexibility. In
addition to deploying femtocells on carrier f2 , the operator
can flexibly pack more femtocells on f1 . The operator may
dynamically adjust the number of femtocells and/or area of
femtocell coverage on f1 depending on the macrocell user
loading. Alternatively, they can afore-hand fix the number
of femtocells and/or area of femtocell coverage on f1 with
a macrocell user capacity in mind. If number of macrocell
users goes up, operator can dynamically move femtocells from
carrier f1 to carrier f2 .

This deployment scenario shown in Fig. 9 is an extension of


the idea in option I. We have again ensured that a macrocell
user at the macrocell-edge does not come near a femtocell
on either f1 or f2 . However, there are some deployment
challenges for macrocell users on f2 . Due to mobility, macro

Fig. 9.

Mixed carrier deployment (option II).

users starting on f2 may move from the shaded to the unshaded region. As per the picture, these macro users need to
be moved from f2 to f1 . In practice, it may be necessary to
allow a macro user in the middle of an active connection to
remain on f2 . Only idle macro users could be moved from f2
to f1 , when it is determined from Round Trip Delay (RTD)
estimation that they have gone out of the shaded region
such an overhead may be worthwhile based on RL capacity
improvement.

150

Capacity of macrocell users


Capacity of femtocells at R /2
m

No. of femtocell BSs/sec

Capacity of cellcenter femtocells

100

50

10

15

20

25

30

35

40

45

50

No. of macrocell users/sec

deployment has the potential to be significantly better than


dedicated carrier deployment in terms of the total number of
users (macro plus femto) that can be supported. Depending
on the desired operating point in terms of macro-users per
sector that need to be supported, the operator can choose a
particular multi-carrier deployment that maximizes the total
number of users that can be supported on the macro plus femto
deployment. One can alternatively argue that by compromising
a little on the capacity in terms of macrocell users, one can
significantly increase the number of femtocells (and hence
total number of users) that can be supported.

Capacity region for carrier f1 for mixed carrier deployment (II).

Fig. 10.

1400

Dedicated carrier deployment


Shared carrier deployment
Mixed carrier deployment (Option I)
Mixed carrier deployment (Option II)

1200
400

No. of femtocell users/sec

Capacity of macrocell users


Capacity of femtocells at R /2
m

350

No. of femtocell BSs/sec

Capacity of cellcenter femtocells


300

250

200

1000

800

600

400

150
200
100
0

10

15

20

25

30

35

40

45

50

No. of macrocell users/sec

Capacity region for carrier f2 for mixed carrier deployment (II).

Fig. 11.

Macrocell and femtocell outage and the capacity region for


f1 is captured in Fig. 10, and that for f2 is captured in Fig.
11. Based on the capacity regions for f1 and f2 , we deduce
the capacity region for this deployment in Fig. 12.
Operating region for mixed carrier deployment (Option II)
200

180

160

No. of femtocell BSs/sec

10

15

20

25

30

No. of macrocell users/sec

50

140

Fig. 13. Comparison of capacity regions for different deployment scenarios

VI. C ONCLUDING R EMARKS


In this paper, we have presented an analysis of the reverse
link capacity of a macrocell-femtocell network in terms of
the number of macrocell users and femtocell BSs that can be
supported such that a certain outage criterion is satisfied for all
users. We have used this analysis to understand the capacity
of four practical mixed-carrier macro-femto deployment scenarios. The numerical results provided indicate that there is
capacity gains to be had in deploying femtocell network and
macrocell network intelligently on multiple frequencies.
R EFERENCES

120

100

80

60

40

20

10

15

20

25

30

No. of macrocell users/sec

Fig. 12. Capacity region for mixed carrier deployment (II). This is deduced
from capacity regions for f1 and f2 .

V. C OMPARISON OF D IFFERENT D EPLOYMENT S CENARIOS


For the purpose of comparison, we collectively show the
capacity regions of all the four deployment scenarios in Fig.
13. We change the y-axis from number of femtocell BSs to
number of femtocell users assuming five users per femtocell
BS. It is evident from the figure that intelligent shared carrier

[1] D. Das and V. Ramaswamy, On the reverse link capacity of a CDMA


network of femto-cells, Proc. of IEEE Sarnoff Symposium, April 2008.
[2] V. Chandrasekhar and J. G. Andrews, Uplink capacity and interference
avoidance for two-tier cellular networks, Proc. of IEEE Global Telecommunications Conference, pp. 33223326, November 2007.
[3] C. C. Chan and S. V. Hanly, Calculating the outage probability in
a CDMA network with spatial poisson traffic, IEEE Transactions on
Vehicular Technology, vol. 50, no. 1, pp. 183204, January 2001.
[4] S. Lowen and M. Teich, Power-law shot noise, IEEE Transactions on
Information Theory, vol. 36, no. 6, pp. 13021318, November 1990.
[5] I. Guvenc, M. R. Jeong, F. Watanabe, and H. Inamura, A hybrid
frequency assignment for femtocells and coverage area analysis for
co-channel operation, IEEE Communications Letters, vol. 12, no. 12,
December 2008.

You might also like