Anda di halaman 1dari 5

Spectrum Allocation in Two-Tier Networks

Vikram Chandrasekhar and Jeffrey G. Andrews


Wireless Networking and Communications Group
Department of Electrical and Computer Engineering
The University of Texas at Austin
Email: cvikram@mail.utexas.edu, jandrews@ece.utexas.edu

Abstract Two-tier networks, comprising a conventional cellular network overlaid with shorter range hotspots (e.g. femtocells,
distributed antennas, or wired relays), offer an economically
viable way to improve cellular system capacity. The capacitylimiting factor in such networks is interference. The cross-tier
interference between macrocells and femtocells can suffocate the
capacity due to the near-far problem, so in practice hotspots
should use a different frequency channel than the potentially
nearby high-power macrocell users. Centralized or coordinated
frequency planning, which is difcult and inefcient even in
conventional cellular networks, is all but impossible in a twotier network. This paper proposes and analyzes an optimum decentralized spectrum allocation policy for two-tier networks that
employ frequency division multiple access (including OFDMA).
The proposed allocation is optimal in terms of Area Spectral
Efciency (ASE), and is subjected to a sensible Quality of Service
(QoS) requirement, which guarantees that both macrocell and
femtocell users attain at least a prescribed data rate. Results show
the dependence of this allocation on the QoS requirement, hotspot
density and the co-channel interference from the macrocell and
surrounding femtocells. Design interpretations are provided.

I. I NTRODUCTION
Wireless operators are in the process of augmenting the
macrocellular network with supplemental infrastructure such
as microcells, distributed antennas and relays. An alternative
with lower upfront costs to the operator is to improve indoor
coverage and capacity using the concept of end-consumer
installed femtocells or home base stations [1]. A femtocell
serves as a low power, small range (10 50 meters) data
access point that provides high quality in-building coverage to
home users. The implication is that femtocells provide higher
spatial reuse and cause smaller interference to other users.
)IOLQNVIRUIHPWRFHOOV

)FOLQNVIRUPDFURFHOO

A. Related Work

)UHTXHQF\OLQNVDFFHVVHGUDQGRPO\E\LQGLYLGXDOIHPWRFHOOV
)UHHOLQN QRWDFFHVVHGE\DQ\IHPWRFHOO
)UHTXHQF\OLQNVDFFHVVHGE\PDFURFHOO

Fig. 1.

Spectrum partitioning in a two-tier network

Frequency division multiple access (FDMA)s resurgence

978-1-4244-2941-7/08/$25.00 2008 IEEE

in emerging OFDMA wireless standards such as the 3GPPs


LTE, WiMAX and 3GPP2s UMB enable the macrocell to
perform exible rate assignment across frequency sub bands to
users. In femtocell deployments, due to reasons of scalability,
security and limited backhaul, it is reasonable to assume no
coordination across femtocells and between femtocells and
the central macrocell. Further, femtocells are placed opportunistically or randomly by end users, rendering conventional
frequency planning strategies to be difcult.
Recent research [2] has shown that in the presence of crosstier interference, an interference avoidance strategy is superior
to interference suppression in a randomly deployed femtocell
network. Assigning orthogonal spectrum resources between
the central macrocell and femtocell BSs eliminates cross-tier
interference. This motivates the spectrum partitioning strategy
(Fig. 1) proposed in this paper. Next, to avoid persistent
collisions with neighboring femtocells, this paper proposes that
each femtocell accesses a random subset of the candidate frequency subchannels, wherein each subchannel is accessed with
equal probability. We term this spectrum access strategy as
F-ALOHA (Frequency ALOHA). We motivate F-ALOHA for
three reasons. First, F-ALOHA avoids transmission delays and
increased RF sensitivity requirements for carrier sensing in the
presence of fading [3]. Next, F-ALOHA provides randomized
interference avoidance, since neighboring femtocells are unlikely to consistently access identical frequency subchannels.
Finally, F-ALOHA provides a decentralized spectrum access
by femtocells and low complexity.
Using F-ALOHA, the paper determines the optimal fraction of spectrum accessed by each femtocell for maximizing
the spatial reuse provided by femtocells. The spatial reuse
is expressible using the Area Spectral Efciency (ASE) in
b/s/Hz/m2 , which is dened as the network-wide sum throughput per frequency subchannel divided by the product of the
subchannel bandwidth and the area per cell-site [4].

The problem considered in this paper is related to Yeung


and Nanda [5], who propose frequency partitioning in a microcell/macrocell system based on mobile speeds and the loading
of users in each cell, but ignore co-channel interference (CCI).
In contrast, the proposed spectrum allocation maximizes the
area spectral efciency in a two-tier network considering CCI,
path losses and prevailing channel conditions.

1583

Asilomar 2008

B. Contributions
This paper employs a stochastic geometry framework for
modeling the random spatial distribution of femtocells according to a Spatial Poisson Point Process (SPPP) [6]. Assuming that (1) downlink transmissions from the macrocell
and femtocells are frequency orthogonal and (2) femtocells
transmit using F-ALOHA, the paper derives the expected pertier throughputs as a function of the QoS requirements and the
number of cellular and hotspot users. Due to the larger spatial
coverage of the macrocell and fewer users per femtocell,
providing equal throughputs per user in each tier requires
nearly 90% of the spectrum to be allocated to the macrocell.
Conversely, partitioning spectrum evenly necessitates relaxing
the QoS constraints to favor signicantly higher data rates at
femtocells. Next, with low femtocell densities, each femtocell
should access the entire spectrum to ensure highest spatial
reuse. In contrast at high densities (eg. 100 femtocells/cellsite), femtocells should access a decreasing fraction (eg.
< 30%) of available spectrum. Finally, while benets of
channel-aware macrocell scheduling permeate to femtocells in
improving overall spatial reuse, the propagation environment
between neighboring femtocells ultimately impacts spectrum
requirements in a two-tier infrastructure.
II. S YSTEM M ODEL
The setup consists of a hexagonal region H of radius Rc
with
a central macrocell BS C providing coverage area |H| =
3 3/2Rc2 , which is surrounded by two rings of interfering
macrocells. Each macrocell is overlaid with femtocell hotspots
of radius Rf , which are randomly distributed on R2 according
to a homogeneous SPPP f with intensity f [6]. The mean
number of femtocells per cell site equals Nf = f |H|.
Macrocell users are assumed to be uniformly distributed inside
each macrocell. Femtocells are assumed to provide closed
access only to licensed indoor users who fall within the radio
range Rf of their respective home BSs. Let U = Uc + Nf Uf
denote the average number of users in each cell site. These
U users are distributed into Uc uniformly distributed tier 1
mobile outdoor users and Uf users per femtocell hotspot.

If each femtocell accesses k subchannels among their allotted Ff subchannels, the net portion of accessed spectrum
per femtocell equals f (1 ) where f  k/Ff . Provided
femtocells choose their subchannels independently and with
equal probability, F-ALOHA thins [6] the intensity of interfering femtocells in each subchannel by their subchannel
access probability. The probability
 p of choosing a given

subchannel equals: p = 1 Ffk1 / Fkf = Fkf = f .
Consequently, the set of interfering femtocells per frequency
subchannel is a Marked SPPP [6] f with intensity f f =
f k/Ff . When f = 1, all femtocells in f access the entire
spectrum but mutually interfere in all subchannels. For f 
1, femtocells transmit in a small region of spectrum and avoid
causing mutual interference. This strategy provides a higher
spectral efciency, but incurs reduced spectrum utilization
because femtocells do not transmit over their entire spectrum.
B. Channel Model and Variable Rate Transmission
The channel between each BS and its users is composed
of a xed distance dependent path loss, a slowly varying
component modeled by lognormal shadowing and Rayleigh
fast fading with unit average power. For simplicity, thermal
noise is neglected at the receiver.
AS 3: Each user is assumed to perfectly estimate and feedback their Signal-to-Interference Ratio (SIR) per frequency
subchannel to their BS.
AS 4: BSs allocate transmission power uniformly over all
their active frequency subchannels.
Each BS assigns rate adaptively based on the received SIR
per user. Let G denote the Shannon Gap with variable rate
M-QAM transmission. Assume an instantaneous transmission
rate of bi bps/Hz if the instantaneous SIR lies in [i , i+1 ).
Using adaptive modulation with L discrete rates, the instantaneous rate W b in a W Hz wide subchannel is chosen
as b = bi , when SIR [i , i+1 ), 1 i L where
bi  log2 (1 + Gi )bps/Hz. Assuming identical statistics over
all frequency subchannels, the long term expected throughput
(in b/s/Hz) per macrocell/femtocell per subchannel is given as:
T =

A. Per-Tier Spectrum Access


The available spectrum comprises F frequency subchannels
each with bandwidth W Hz. We wish to determine the optimal
partitioning (Fc , Ff ), where Fc subchannels are available for
macrocell transmissions and Ff = F Fc subchannels are
available for femtocell transmissions (Fig. 1). Denote =
Fc /F as the fraction of spectrum assigned to the macrocell
BS with the following key assumptions:
AS 1: Each femtocell schedules its users in a Channel Blind
Round-Robin (RR) fashion. The macrocell schedules its users
according to either a RR or a Channel Aware Proportional Fair
(PF) scheduler.
AS 2: The fraction is a positive real number in [0, 1].
The long term per-tier throughput is assumed to be equally
divided among the Uc outdoor and Uf indoor users.

L1


l Pr[l SIR < l+1 ] + L Pr[SIR L ] (1)

l=1

The average throughput provided by each macrocell [resp.


femtocell] is obtained multiplying the expected throughput in
(1) by their respective spectrum allocation [resp. f (1 )].
III. P ER -T IER S PECTRUM A LLOCATION
Let spectrum W F be allocated such that the macrocell
BS transmits over a portion , while femtocell BSs transmit
over the remaining fraction 1 . Let Tc (, Uc ) be the long
term throughput (in b/s/Hz) in each subchannel provided by
the macrocell. Let each femtocell access a portion f of
its allotted spectrum using F-ALOHA transmission. Dene
Tf (f f ) as the expected femtocell throughput over each
frequency subchannel. With universal frequency reuse across

1584

all macrocells, the ASEs provided by each tier are given as:
Tc (, Uc )
,
|H|

ASEc =

ASEf =

Nf f Tf (f f )
|H|

(2)

The factor Nf f represents the mean number of transmitting


femtocells in each subchannel in the set f . The network-wide
ASE is consequently given as:
ASE = ASEc + (1 )ASEf

(3)

The expected network throughput (in b/s) over the W F wide


spectrum is obtained by multiplying (3) by W F |H|.
Before determining the spectrum allocation, we stipulate a
QoS requirement , which ensures that users in either tier are
guaranteed a minimum expected throughput. By implication,
regulates the maximum spectrum that any tier can receive.
Denition 1: The QoS parameter 0 < 0.5 guarantees
that the expected throughput per user in one tier is at least
/(1 ) w.r.t the other tier.
Choosing different assigns different priorities (QoS) to one
tier relative to the other. For example, setting = 0.5 ensures
that users in both tiers obtain identical expected rates. On the
other hand, decreasing favors assigning greater spectrum to
the tier providing a higher expected throughput per active user.
Given a total available spectrum of 1 Hz, the problem
is to determine the optimal spectrum allocation over all
possible spectrum partitioning strategies [0, 1] between
the macrocell and femtocells, as shown below:
= arg max Tc (, Uc ) + (1 )Nf f Tf (f f )
01

(4)

macrocell user 0 at position r w.r.t C0 in a given subchannel


is given as:
0
2

|h0 | ||r/Rc || c
(7)
I (r)


  r bk c
2


|h0k | 0k
(8)
where I (r) =
  Rc  
kB
Here, Rc is the cell radius, c represents the outdoor path
loss exponent and |h0k | is the exponentially distributed channel
power between interfering BS Ck and the user of interest. The
RV 0 [resp. 0k ] is the lognormal shadowing between the
central BS [resp. interfering BSs] and the desired user, which
2
are distributed as LN(c,dB , 2 c,dB
), where = 0.1 ln 10 is
a scaling constant.
Using (1), the macrocell subchannel throughput Tc has
been derived analytically for a channel blind Round-Robin
(RR) scheduler [7] and estimated using numerical simulation
for a channel aware Proportional Fair (PF) scheduler. In the
simulation, the number of subchannels is set as Fc = 1 with
a link bandwidth W = 15 KHz. Each mobile is moving
at v = 13.34 m/s (30 mph) and the throughputs/user are
averaged over 500 drops, with 8000 trials/drop for modeling
time-varying fading. The Rayleigh fading is held xed over
a duration Tc = 0.4/fd where fd = vfc /(3 108 ) is the
Doppler frequency at a carrier fc = 2 GHz. Fig. 2 compares
the performance of PF versus RR scheduling for different
Uc . Exploiting channel variations through proportional fairness
roughly doubles the expected subchannel throughput.
SIRc (r) =

(5)
s.t min {Tc,u (), Tf,u ()} (Tc,u () + Tf,u ())
Tc (, Uc )
(1 )f Tf (f f )
, Tf,u () 
& Tc,u () 
Uc
Uf

A. Macrocell Throughput
2

RR Scheduling
PF Scheduling
3

Macrocell Throughput T (b/s/Hz)

Proposition 1: If the expected macrocell throughput per


subchannel is independent of the total spectrum allocated to
the macrocell , i.e. Tc (, Uc ) = Tc (Uc ) [0, 1], the
optimizing in (4) satises the QoS constraint with equality,
belonging to a set with two candidate spectrum allocation
assignments given as:

1
Uf
1 Tc (Uc )

{x, 1 x}, x  1 +

Uc f Tf (f f )
(6)
Proof: Since Tc (, Uc ) = Tc (Uc ) [0, 1], the optimization problem in (4) is to determine the optimal which
maximizes a convex combination of Tc and Nf f Tf (f f )
with a linear constraint (5). Hence, the argument maximizer
is located at the extremal points of (5). Solving for the
satisfying (5) with equality yields (6).

3.5

2.5

1.5

10

15

20

25

30

Users Per Macrocell

Fig. 2.

Tc with RR and PF Macrocell Scheduling, c = 4

B. Femtocell Throughput
Using F-ALOHA, the interfering femtocells w.r.t a (typical)
reference femtocell F0 form a marked SPPP f f with
intensity f f . In a given frequency subchannel, the SIR of
user 0 at distance Rf w.r.t femtocell F0 is given as:


Pf2 0k |x0k |f
SIRf = 0 Rf f /
kf

Let |h0 | denote the exponentially distributed channel power


between the central macrocell BS C0 and its scheduled user 0.
Denote B as the set of two rings of interfering macrocell BSs.
Denoting the Euclidean norm by ||||, the received SIR for

Here, |x0k | are the distances of the interfering femtocells


Fk . Using Turkmanis [8] approximation, the composite
lognormal-exponential distributed channel powers from the
desired (0 ) and interfering BSs (0k ) are modeled as

1585

lognormal rvs LN[(f i,dB 2.5), 2 (f2 i,dB + 5.572 )] and


I LN[(f o,dB 2.5), 2 (f2 o,dB + 5.572 )] respectively.
Note that = 0.1 ln 10 is a scaling constant. Further, f
[resp. f ] designate the path-loss exponents from the desired
femtocell [resp. interferers] to user 0. Finally, Pf is the xed
wall penetration loss between neighboring transmissions.
The closed form distribution
from neigh of interference
2
f
is known
boring femtocells If,f =
kf Pf 0k |xk |
only for f = 4. Dening f  2f , the following lemma
[9, Theorem 3] provides an asymptotically tight lower bound.
Lemma 1: With randomized transmissions and lacking
power control, the distribution of If,f is given as:
2

Pr(If,f > y) 1 exp [f f E[If ]Pf f y f ] (9)


For xed y, as f 0, the tail probability Pr(If,f > y)
0 in (9) indicating that selecting fewer subchannels using FALOHA provides greater resilience against persistent collisions from nearby femtocells. Using Lemma 1 and dening

f  f E[If ](Pf2 |Rf |f )f , the femtocell SIR distribution


can be lower bounded as:

Pr (SIRf ) 1 E0 [exp (f f f 0

)]

To justify (11), there are an average of |H|f f transmitting


femtocells per subchannel. With F-ALOHA access of 0 <
1, each femtocell obtains an average subchannel throughput of
Tf (), resulting in ASEf equaling f Tf (f ).
Remark 1 (Boundedness of the ASE): The
maximum
ASE obtained in (11) depends only on the effective
intensity f f . With increasing f , provided f < 1, the
maximum femtocell ASE per subchannel is unchanged (or
f f = constant), implying that the optimal f decreases
inversely as f . Consequently, with increasing number
of femtocells, the network-wide femtocell throughput
|H|W F (1 )ASEf grows linearly with (1 ).
Fig. 4 plots (11) for different Nf with f = 3.5 and
Pf,dB = 2. In all cases, the highest ASE is xed at nearly
0.000121 b/s/Hz/m2 validating Remark 1. With a low femtocell density (Nf = 10), the best strategy is to access the entire
allocated spectrum. In a dense network (Nf = 100), the ASE
is maximized when each femtocell accesses approximately
30% of the available spectrum. Further, in (11), as long as

(10)
4

The lower bound in (10) can be easily evaluated numerically.


Combining (1) and (10) yields an upper bound on the mean
femtocell throughput Tf per frequency subchannel.

Area Spectral Efficiency (b/s/Hz/m2)

1.4

Mean Femtocell Throughput (b/s/Hz)

7
HA, N =10
f

HA, N =50

HA, Nf=100
LA, N =10

LA, N =50
f

LA, N =100

x 10

Highest Attainable ASE

1.2
1
0.8
0.6
0.4
Nf=10
0.2

N =50
f

N =100
0
0

0.2

0.4

0.6

0.8

Transmission probability per frequency link

Fig. 4.
0

0.2

0.4

0.6

0.8

Femtocell ASE Vs F-ALOHA spectrum access

Transmission probability per frequency link

Fig. 3.

Theoretical and Empirical femtocell throughput f Tf

Fig. 3 plots the mean throughput per femtocell (1 )f Tf


(in b/s/Hz) assuming the entire spectrum is allocated to
femtocells ( = 0). Black dash-dotted curves show results
of numerical simulation. Two cases are considered namely
(1) high attenuation (marked HA with f = 4, Pf,dB = 10)
and (2) low attenuation (marked LA with f = 3.5, Pf,dB =
2) from neighboring femtocells. With f = 1 and Nf = 50
femtocells/cell site, f Tf falls from approximately 4.5 b/s/Hz
in a HA scenario to nearly 0.5 b/s/Hz in a LA scenario.
To calculate the optimum f , we determine the optimal fraction of subchannels accessed by each femtocell for maximizing
spatial reuse, which is given as:
f = f arg max Tf (f ), ASEf = f f Tf (f f ) (11)
0<1

f = 1, each femtocell accesses the entire spectrum (1 ).


Consequently, as f increases, Tf decreases due to higher
interference from neighboring femtocells. However, if f < 1,
Tf = ASEf /(f f ) remains constant with increasing f .
One may then explore the dependence of the mean overall
femtocell throughput Tf f (1 ) on the spectrum allocation
and F-ALOHA access f . Equivalently: With increasing
femtocell density f , can increasing allocation (1 ) to
femtocells counterbalance decreasing F-ALOHA access f
to cause a net increase in mean femtocell throughput? This
question is answered by the following condition: Given a
spectrum allocation l at mean femtocell density f , let Tf,l
and f,l be the mean subchannel throughput and the optimal
F-ALOHA access respectively. On increasing f by f with
allocation h , let the corresponding quantities equal Tf,h and
f,h . The femtocell network is dened as fully-utilized [resp.
sub-utilized] if a marginal increment in the femtocell density

1586

scenario with = 0.01 and Nf = 50 femtocells/cell site,


nearly 70% of spectrum is allocated to femtocells.

reduces [resp. improves] the mean femtocell throughput:


(1 l )f,l Tf,l (1 h )f,h Tf,h
1 h
ASEf,l f + f
Tf,l f,l

=
1 l
ASEf,h
f
Tf,h f,h

(12)

90
LA, RR, = 0.01

Equation (12) reects the competing effects of increasing allocation (1) and decreasing F-ALOHA spectrum access f
(or increasing f ) in determining the net femtocell throughput.

Channel Bandwidth WF, MHz

HA, RR, = 0.5

IV. N UMERICAL R ESULTS


Results are presented in Figs. 5 and 6 with the system
parameters in Table I. To model higher data rates at femtocells
relative to the macrocell, QoS parameters of = 0.5 (equal
mean throughputs to all users) and = 0.01 (or 100x higher
throughput per femtocell user relative to a cellular user) are
considered. Two scenarios are considered namely 1) High
Attenuation (HA) of neighboring femtocell transmissions with
parameters f = 4 and Pf,dB = 10 and 2) Low Attenuation
(LA) scenario with f = 3.5 and Pf,dB = 2.
TABLE I
S YSTEM PARAMETERS
Description
Macrocell/Femtocell Radius Rc , Rf
Total users per cell site U
Users per femtocell Uf
Wall penetration loss Pf,dB
Shannon Gap G, Modulation Levels L
Path-loss exponents c , f , f
Lognormal Parameters c,dB , f i,dB , f o,dB

HA, PF, = 0.5

70

HA, RR, = 0.01


HA, PF, = 0.01

60

50

40

30

20

10

0
10

20

30

40

50

60

70

80

90

100

110

Average Femtocells Per Cell Site

Fig. 6.

Required Spectrum W F meeting target data rates of Dc and Df

Fig. 6 shows the minimum required spectrum W F , satisfying a target average data rate of Dc = 0.1 Mbps for each
macrocell user. For each femtocell, we consider QoS values
of = 0.5 and = 0.01, corresponding to target average data
rates Df = Dc (1 )/ equaling 0.1 and 10 Mbps/hotspot
user. From (4), the necessary spectrum W F = Dc Uc /(Tc ) =
Df Uf /[(1 )f Tf ].
Two key observations are: First, a macrocell with channel
aware PF scheduling provides signicant spectrum savings to
guarantee Df and Dc ; eg. with = 0.01 and Nf = 50
femtocells/cell site in a HA scenario, the spectrum savings are
nearly 40% relative to RR scheduling. Next, with = 0.01
and a LA scenario, adding femtocells requires a linear increase
in spectrum W F indicating the femtocell network is fullyutilized in (12). Conversely, in a HA scenario, the femtocell
network is sub-utilized in (12), hence adding femtocells increases the mean throughput per femtocell.

Value
288 m, 40 m
300
2
2 dB, 10 dB
3 dB, 8
4, 3, 3.5 & 4
8 dB, 4 dB, 12 dB

1
0.9
0.8

Spectrum Allocation

LA, PF, = 0.01

80

0.7
0.6
0.5
0.4

R EFERENCES

0.3

[1] V. Chandrasekhar, J. G. Andrews, and A. Gatherer, Femtocell networks:


a survey, IEEE Comm. Mag., vol. 46, no. 9, pp. 5967, Sept. 2008.
[2] V. Chandrasekhar and J. G. Andrews, Uplink capacity and interference
avoidance in two-tier femtocell networks, To appear, IEEE Trans. on
Wireless Comm., 2009.
[3] N. Hoven and A. Sahai, Power scaling for cognitive radio, in International Conf. on Wireless Networks, Communications and Mobile
Computing, vol. 1, June 2005, pp. 250255.
[4] M. S. Alouini and A. J. Goldsmith, Area spectral efciency of cellular
mobile radio systems, IEEE Trans. on Veh. Tech., July 1999.
[5] K. Yeung and S. Nanda, Channel management in microcell/macrocell
cellular radio systems, IEEE Trans. on Veh. Tech., Nov. 1996.
[6] J. Kingman, Poisson Processes. Oxford University Press, 1993.
[7] V. Chandrasekhar and J. G. Andrews, Spectrum allocation in two-tier
networks, Submitted, IEEE Trans. on Comm., 2008, [Online] Available
at http://arxiv.org/abs/0805.1226.
[8] A. M. D. Turkmani, Probability of error for m-branch macroscopic
selection diversity, in Comm., Speech and Vision, IEE Proc. I, Feb. 1992.
[9] S. P. Weber, J. G. Andrews, and N. Jindal, The effect of fading, channel
inversion, and threshold scheduling on ad hoc networks, IEEE Trans. on
Info. Theory, Nov. 2007.

HA, =0.5
HA, =0.1
HA, =0.01
LA, =0.5
LA, =0.1
LA, =0.01

0.2
0.1
0

50

100

150

Average Femtocells Per Cell Site

Fig. 5.

Optimal spectrum allocation for varying

Fig. 5 shows the allocation using (6) and RR scheduling.


With = 0.5, nearly 90% of spectrum is assigned to the
macrocell. The central macrocell serves a higher number
of users who experience relatively poor reception over a
bigger area. Ensuring equal cellular and hotspot throughput
consequently requires assigning nearly the entire spectrum to
cellular transmissions. As decreases, femtocells require more
spectrum for providing greater indoor capacity; eg. in a LA

1587