Vous êtes sur la page 1sur 6

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

Soft Frequency Reuse Schemes for Heterogeneous


LTE Systems
Giovanni Giambene , Van Anh Le , Thomas Bourgeau , Hakima Chaouchi
University

of Siena, Via Roma 56, 53100, Siena, Italy


Email: giambene@unisi.it
Paris Sorbone University, 4 Place Jussieu, 75005 Paris, France
Email: thomas.bourgeau@upmc.fr
Institut Telecom, Telecom Sud Paris, 9 rue Charles Fourier, 91011 Evry, France
Email: hakima.chaouchi@it-sudparis.eu
AbstractWith the increasing demand of mobile data traffic,
cellular networks are evolving towards heterogeneous systems
involving cells of different sizes (e.g., pico, micro, and macro
cells). The presence of small cells is causing new interference
issues that require the introduction of frequency reuse techniques
suitable to manage both co-tier and cross-tier interference. In this
paper, we propose a novel Soft Frequency Reuse (SFR) scheme
for LTE-A, based on three frequency segments (named ThreeBand Improved SFR, 3B-ISFR) where the Frequency Reuse Pattern
(FRP) allocation scheme for micro cells is centralized at the level
of each macro cell. This scheme is based on the exchange of
measurements made by the micro cells with their macro cell
(cognitive-assisted approach) and includes an intra-cell offloading
algorithm to balance the load between cell-center and cell-edge of
each micro/macro cell. The goodness of 3B-ISFR has been proved
by means of performance comparisons with other schemes in the
literature in terms of outage probability and average cell capacity.
KeywordsLong Term Evolution, Fractional Frequency Reuse

I.

I NTRODUCTION

The large adoption of new-generation mobile terminals


(smart phones and other mobile Internet devices) offering
enhanced user experience with high-bandwidth-consuming applications (video streaming, mobile cloud) has caused an exponential growth of data traffic, reaching the theoretical limits
of network capacity and spectral efficiency of current cellular
systems. The exponential growth of mobile data traffic is a
pressing need that evolving 4G cellular networks (e.g., Long
Term Evolution - Advanced, LTE-A) need to address [1]. In this
context, Heterogeneous cellular Networks (HetNets) comprising of conventional high-power macro cell base stations (eNB)
overlaid by low power nodes, also known as small cells, are
considered to be an efficient solution to improve resource reuse
and spectral efficiency, while offloading user traffic from macro
cells towards small ones. We refer here to LTE-A cellular
systems and consider small cells such as pico/micro cells with
transmit power between 250 mW up to approximately 2 W that
are deployed in indoor/outdoor scenarios to offload the traffic
from outdoor macro cells. In such a multi-tier heterogeneous
environment, where high power and low power nodes (cells)
are overlaid by each other while sharing the same spectrum,
Inter-cell Cell Interference Coordination (ICIC) becomes a
central issue.

978-1-4673-6432-4/15/$31.00 2015 IEEE

3161

In classical (single-tier) cellular networks, there is inter-cell


interference due to the reuse of resources among sufficiently
far cells. Inter-cell interference is particulary significant at cell
edge (due to the closeness with neighboring cells) and less important at cell center. This is the reason why improved resource
reuse schemes have been proposed in the literature, such as
Soft Frequency Reuse (SFR), where cells are divided between a
central part and an edge one: cell-edge User Equipments (UEs)
are serviced with a fraction of the whole bandwidth according
to a classical reuse pattern; instead, transmissions to cell-center
UEs can use a complementary bandwidth (or even the whole
bandwidth), but at a lower power level. SFR can achieve better
outage conditions than classical reuse schemes and improves
the capacity with respect to Fractional Frequency Reuse (FFR)
schemes [2],[3]. However, with SFR some cell-center UEs,
which are usually very close to cell-edge UEs, might still
experience a low channel quality because they are closer to
interfering eNBs and a lower power is used to transmit to
them. We refer in this study to the direct communications from
eNBs to UEs, even if the use of relay nodes would be another
interesting scenario for HetNets [4].
Inter-cell interference can be further classified as co-tier
and cross-tier interference. In the first case, we refer to the
interference of macro cells among themselves or of small
cells among themselves. Instead, the second case concerns
the interference between macro and small cells when they are
sharing the same frequency bands. Recent research proposals
showed the limits of classical ICIC techniques in managing
co-tier and cross-tier interference in HetNets [5],[6],[7].
To avoid co-tier interference between neighboring micro
cells, the authors in [5] consider the bandwidth divided into
two segments, one for cell-center UEs with low transmission
power and another for cell-edge UEs with higher transmission
power. Moreover, neighboring micro cells rotate the use of
these bandwidth segments between central and edge parts in
order to reduce co-tier interference. The target is to minimize
the transmission power of micro cells under the constraint
of a minimum UE bit-rate. Unfortunately, this paper does
not consider the cross-tier interference due to macro cells.
The works in [6],[7] further elaborate this scenario including
macro cells and cross-tier interference, but do not consider that
close micro cells should use different frequency segments, thus
having worse Signal-to-Interference-plus-Noise Ratio (SINR)
experienced by cell-edge UEs in micro cells.

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

In our paper, we propose a novel approach, called ThreeBand Improved SFR (3B-ISFR), which is a modified SFR
scheme to support HetNets taking co-tier and cross-tier interference into account. In particular, our scheme divides the
available bandwidth into three segments to be assigned to edge
and central parts of both micro and macro cells with the aim
to differentiate their allocation between adjacent micro/macro
cells. 3B-ISFR differentiates the transmission power levels
between cell-center UEs and cell-edge ones for both macro
and micro cells. In addition to this, 3B-ISFR adopts a fair
mechanism to reassign critical cell-center UEs to the cell edge
(intra-cell offloading) to improve their throughput.
II.

S YSTEM D ESCRIPTION

A. Notations and Basic Assumptions


In this paper, we consider downlink transmissions in a
two-tier heterogeneous LTE cellular network, composed of Nc
macro cells each overlaid by micro cells. In each macro cell, a
central eNB is surrounded by Nr uniformly distributed micro
cells and Nu uniformly distributed UEs. The signaling communications among macro/micro cells can be handled through
X2 interfaces and all cells are equipped with a complete set
of radio resource control functions. Let i (from 1 to Nc )
denote the index of a generic macro cell. Let j (from 1 to
Nr ) denote the index of a generic micro cell in a macro
cell. Moreover, j = 0 is used to denote the macro cell itself.
Furthermore, we consider that the entire LTE bandwidth is
divided into three equal-size segments: Fm for m = 1, 2, 3.
According to SFR, each cell is divided into cell-edge and cellcenter regions, where UEs are serviced by different frequency
segments and transmission power levels by the corresponding
eNB. Moreover, we consider that one of these Fm segments
is used in the macro/micro cell-edge area and the other two
m
segments are used in the macro/micro cell-center area. Let Pi,j
denote the power per sub-channel of micro cell j in macro cell
i, transmitting on a sub-channel belonging to Fm .

but co-tier interference is not taken into account. Moreover,


micro cells have no differentiation in the transmission power
between cell-edge and cell-center areas, thus having higher
outage probability.
Three-Band SFR (3B-SFR): In this scheme [see Fig.1(b)],
we propose to use three bandwidth segments (i.e., F1 , F2 and
F3 ) to be assigned to macro/micro cells for their respective
edge/center parts. In particular, we first distribute F1 , F2 and
F3 to macro cells according to SFR with a 3-color reuse
pattern, considering that only one segment is used in the cell
edge. Then, we assign these segments to micro cells in a way
that takes co-tier and cross-tier interference into account. In
this scheme, both macro and micro cells have a transmission
power differentiation between cell edge and cell center. More
details are provided in Sec. III on the algorithm to assign
frequency segments to micro cells. A UE adopts a SINRbased criterion to select an eNB (it does not matter if this
eNB corresponds to a macro or to a micro cell); then, still
a SINR-based criterion is used to map this UE to the edge
region or to the central one of the corresponding eNB. As
anticipated in Sec. I, the problem with the 3B-SFR technique
is that some cell-center UEs, which are very close to celledge UEs, may suffer from very low signal quality (i.e., low
SINR values). This is the reason why we propose the modified
scheme described below.
Three-Band Improved SFR (3B-ISFR): This technique is
based on 3B-SFR, thus using the same algorithm to distribute
frequency segments to cells and basically the same scheme
to assign UEs to macro/micro eNBs and their respective cell
center and cell edge. However, the association of UEs to their
cell-center can be modified on the basis of a bit-rate threshold
criterion; see next Sec. III.
III.

3B-ISFR R ESOURCE A LLOCATION A LGORITHM

A. Frequency Reuse Pattern Allocation

In this paper, we compare three different SFR schemes


applied to our two-tier HetNet scenario.

One of the major novelty of HetNets compared with


conventional cellular networks is the deployment of micro
cells in an unplanned way. In both 3B-SFR and 3B-ISFR
schemes, it is necessary to define the Frequency Reuse Pattern
(FRP) of each macro/micro cell. In particular, we have three
possible FRP configurations for a macro/micro cell as: {edge:
F1 ; center: F2 , F3 }, {edge: F2 ; center: F3 , F1 }, and {edge: F3 ;
center: F1 , F2 }. In this work, we use a honeycomb model for
macro cells that contain micro cells according to a uniform
distribution. The FRP of each macro node is easily defined
according to SFR with a 3-color reuse pattern as shown in [3].
The task of the algorithm below is to define the FRP for the
Nr micro cells of a macro cell. This algorithm that we are
going to describe is centralized at the level of macro cells and
assisted by their micro cells (cognitive-assisted approach).

Two-Band SFR (2B-SFR): This scheme is analogous to that


proposed in [7], where the LTE frequency band is divided
into two segments (i.e., F1 and F2 ) assigned to cell edge and
cell center of both macro and micro cells and where different
power levels are used for each segment. In this technique, one
F1 F2 pattern is used in all macro cells and another F2
F1 pattern is used in all micro cells, as shown in Fig.1(a).
The purpose of this frequency pattern allocation is to avoid
cross-tier interference between micro cells and macro cells,

We assume that each micro node can sense the signal


strength from neighboring macro and micro cells to compute
the Reference Signal Receive Power (RSRP) and the Reference
Signal Receive Quality (RSRQ), the same LTE measurements
used by a UE (downlink) for cell selection, reselection and
handover procedures, as explained in 3GPP TS 36.304. In
particular, RSRP is the average power of the resource elements
that carry cell-specific Reference Signals (RS) over the entire
bandwidth, so RSRP is measured only on the symbols carrying

If Fk is the segment used for the edge region of the i-th


macro cell, the corresponding macro cell transmission power
m6=k
k
for cell-edge UEs is Pi,0
= Pi,0
, where > 1 is a

m6=
power control factor. Moreover, Pi,j>0 = Pi,j>0
represents
the transmission power on the F frequency segment for celledge UEs belonging to the j-th micro cell of the i-th macro
cell, where > 1 is another power control factor. The total
transmission power over all sub-channels of a macro cell eNB
(or of a micro cell eNB) is Pa (Pb ).
B. SFR Schemes for HetNets Scenarios

3162

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

F32

F2

F2
F2

F2

F33
F2

F1
F{ 1F1 + F2 }
F2

F2

F1
F32

F1

F2

F2

F1

F33

F2

F3

3)

4)

F3

F3

F2

(a) 2B-SFR

(b) 3B-SFR and 3B-ISFR

F2

Different frequency reuse examples for 2B-SFR, 3B-SFR and 3B-ISFR in HetNets

We consider that the X2 interface of LTE is used to


send RSRQ measurements from micro cell eNBs towards the
corresponding macro cell eNB, which is assumed to know
its FRP according to a macro cell-level SFR with a 3-color
reuse pattern. Moreover, during the process for the FRP
definition of micro cells, the macro cell eNB can send via
X2 interface matrix P, containing the currently-defined FRP
of its micro cells. Hence, the macro cell eNB coordinates the
FRP allocation of its micro cells according to the following
algorithm that permits to improve SINR (especially in very
dense micro-cell deployments) by assigning the FRP to micro
cells according to their interference criticality.

2)

{F1 , F3} F2
{F1 , F2} F3

F2

RS. According to 3GPP 36.211, the location of the resource


elements carrying the RSRP signal is different from cell to
cell according to the physical cell ID. In this way, RSRP can
differentiate a cell (downlink) from another. In our case, we
can consider that RS signals are transmitted at a reference
power level. Moreover, the Received Signal Strength Indicator
(RSSI) is a parameter that provides information on the received
power across the whole bandwidth, including the useful signal,
all interfering ones, and even the thermal noise. Then, RSRQ
is determined as Ns RSRP/RSSI, where Ns is the number of
sub-channels in the bandwidth. We can assume that a micro
cell eNB during the FRP algorithm phase is able to measure
RSRQ (a sort of SINR parameter) for each macro/micro cell
in its surroundings with RSRP above the receiver sensitivity
level.

1)

F2

{F2 , F3}

F2

Fig. 1.

{F +F }

F3 1 {F 2, F }
1
2

F2

F1

F2

{F1 , F3} F2

F2

Initialize matrix P, stored at the macro cell eNB,


and containing the FRP of its micro cells. At the
beginning, we assume that none of the Nr micro
nodes has its FRP assigned.
Each micro cell measures RSRQ from all neighboring macro cells (i.e., all macro cells except its
macro cell). Then, it sends a message containing the
strongest RSRQ to the macro cell eNB by means of
X2 interface.
The macro cell eNB collects RSRQ values from
micro nodes and selects the micro cell j with the
biggest RSRQ, since this micro cell is also the closest
one to its macro cell border (potentially representing
the micro cell with the most critical interference
conditions). Then, the macro cell sends a message
with the current matrix P to micro cell j asking it to
select its FRP value.
Micro cell j adopts its FRP (out of the three available
combinations), which is different from that of its

3163

5)

6)

macro cell and from that of the neighboring node


(be it another macro cell or another micro cell inside
the same macro cell1 ; we have this second case only
if the FRP has been allocated to the interfering micro
cell at a previous step), having the strongest RSRQ.
Then, micro cell j sends a message to the macro cell
eNB to update its FRP entry in matrix P.
Each remaining micro cell (i.e., a micro cell for which
the FRP is not yet assigned) selects the strongest
RSRQ from micro nodes having a valid entry in
matrix P (new RSRQ measurements) and from macro
cells (we can reuse the RSRQ measurements already
taken at previous step #2) and notifies the macro cell
eNB about this selection.
Out of the remaining micro nodes, the macro cell
eNB selects the micro cell j with the biggest RSRQ
value, since this micro cell is the next one in terms of
interference criticality. Hence, the macro cell sends a
message to micro cell j containing the current matrix
P and asking it to select its FRP. Then, go to step #4
until all micro cells have their FRP defined.

This scheme can manage to allocate FRP to new micro cells


that are added later to the cellular system when the previous
allocation has been decided. Of course, selecting the FRP for
these new cells alone could be sub-optimal.
The complexity of this FRP algorithm can be determined
as follows in terms of number of signalling messages Sm
exchanged between the Nr micro nodes and the corresponding
macro cell eNB. In the first round of the algorithm (step #2),
Nr signaling messages are sent to the macro cell eNB. Then,
the macro cell eNB sends a feedback broadcast message to
notify the selected micro cell to set its FRP (on the basis of
the current contents of matrix P conveyed by this message)
and to tell the other micro cells that they should continue
to perform RSRQ measurements. The selected micro cell in
turn sends back its FRP to the macro cell. Hence, the total
number of messages exchanged in this first round from steps
#2 to #4 is Nr + 2. At the next run from steps #4 to #6
we have Nr 1 + 2 message exchanges, then Nr 2 + 2
1 In this algorithm, we do not consider to coordinate the FRP with micro
cells of adjacent macro cells: micro cells belonging to different macro cells
are decoupled in the FRP allocation phase to simplify the problem and not to
consider the feedback effects in a macro cell due to the FRP decisions taken
in micro cells of adjacent macro cells. This limitation will be removed in the
simulations in Sec. V, where interference from micro cells of adjacent macro
cells is accounted for.

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

message exchanges, etc. Hence, the total number of messages


exchanged to define micro cells FRP within a macro cell is
equal to the sum of integer numbers from 1 to Nr plus 2Nr
as follows:
Nr (Nr + 1)
+ 2Nr .
(1)
Sm =
2
2
Hence, the FRP scheme signalling complexity is O Nr .
B. Cell Selection, SINR Model, and Identification of Cell
Center/Edge UEs
Once determined the FRP for macro and micro cells, we
consider that a UE selects the macro/micro cell eNB that
guarantees the maximum RSRQ value2 . We denote the channel
gain between a macro cell eNB i and UE u as hu,0,i and the
channel gain between micro cell eNB j inside macro cell i
and that UE as hu,j,i . Let N denote the noise power level
on a generic sub-channel. Note that the power level of a subchannel in a cell depends on its FRP. Then, SINR is modeled
as detailed below.
If UE u connects to macro cell k (k = 1 to Nc ), its SINR for
a sub-channel belonging to Fm (m = 1 to 3) can be expressed
as follows:
2

m
SIN RF
u,0,k

m
Pk,0
|hu,0,k |
= N
N
Nr
c
P m
Pc P
2
2
m |h
Pi,0 |hu,0,i | +
Pi,j
u,j,i | + N
i=1
i6=k

i=1 j=1

(2)
If UE u connects to a micro node g (g = 1 to Nr ), its
SINR for a sub-channel belonging to Fm (m = 1 to 3) can be
expressed as follows:
2

m
SIN RF
u,g,k

m
Pk,g
|hu,g,k |
= N
Nr
Nc
P
P
Pc m
2
2
m |h
Pk,j
Pi,0 |hu,0,i | +
u,j,i | + N
i=1

i=1

j=1

Robin (RR) scheduler over the assigned downlink resources for


either cell center or cell edge; with RR, the Physical Resource
Blocks (PRBs) of the cell-center (cell-edge) area are equally
divided among the active cell-center (cell-edge) UEs.
C. Enhancing UE Experience: Intra-Cell Offloading
Once the PRBs to be allocated to a UE are defined (RR
scheme), then, the average throughput of a UE u, R(u), is
determined according to its SINR as shown in the following
equation (6) of sub-Sec. III-D [1]. As stated in Sec. II, our
proposed 3B-ISFR scheme is able to address the issue of some
critical cell-center UEs suffering from low R(u) values. In
particular, we consider a threshold bit-rate value, Rs , to decide
whether to move a generic cell-center UE u to the group of
cell-edge UEs. This algorithm is as follows:
1)
2)
3)
4)

If the average bit-rate R(u) assigned to cell-center


UE u is lower than Rs and
If, after moving UE u to the group of cell-edge UEs,
all cell-edge UEs have an average bit-rate greater than
or equal to Rs ;
Then, perform intra-cell offloading by assigning UE
u to the group of cell-edge UEs.
Otherwise, if the above second condition is not satisfied, UE u remains in the group of cell-center UEs,
but its Rs requirement is not met; this UE is counted
in the group of critical UEs, but not considered in
outage conditions.

This method allows us to improve the service for some


critical cell-center UEs that are moved to the group of celledge UEs, but also increases the bandwidth available to other
cell-center UEs, since the number of cell-center UEs is reduced
as a consequence of this algorithm. The Rs value should be
set depending on the number of UEs in a cell (see Sec. V).
D. Planning Metrics: Outage Probability and Cell Capacity

(j,i)6=(g,k)

(3)
After the cell selection process has completed, we propose
to use the following scheme in each macro/micro cell to differentiate between cell-center and cell-edge UEs. In particular,
we assume that UEs communicate to the corresponding eNB
their RSRQ values (averaged on a certain interval). Then, all
UEs connected to a (macro/micro) cell are sorted in RSRQ
ascending order by the corresponding eNB. Let denote the
cell-center UE ratio, representing the percentage of UEs that
are in the central part of the cell. In 2B-SFR, 3B-SFR and
3B-ISFR, the first 1 portion of sorted UEs is assigned
by the corresponding eNB with cell-edge sub-channels (e.g.,
frequency band denoted by F1 with higher transmission power
level) and the remaining portion is assigned with cell-center
sub-channels (e.g., frequency bands denoted by F2 and F3
with lower transmission power level). In addition to this, in
the 3B-ISFR scheme, a UE originally-assigned with cell-center
resources can be reassigned with cell-edge resources according
to the intra-cell offloading scheme described in the next subSection. When the UEs of a macro/micro cell are assigned to
cell center or cell edge, they are serviced by means of a Round
2 The use of a bias scheme would allow more UEs to be connected to micro
cells. This aspect will be the subject of a future study.

3164

In the case a UE has SINR from its (macro/micro) cell that


is lower than the threshold SINR of the most robust modulation
and coding level [3], this UE experiences outage conditions.
The resulting cell-level outage probability is as follows:
OutageP rob. =

N umber of U Es in outage condition


N umber of U Es in cell
(4)

According to Table 1 in [1], let us assume that the eNB uses


the i-th LTE modulation and coding level for the transmissions
to a generic UE u depending on its SINR value (downlink)
above the threshold of the i-th mode. The corresponding
efficiency i is written below where ri and Mi denote the
code rate and modulation size of the i-th mode:

bit
i = ri log2 (Mi )
(5)
resource element
The average throughput of UE u results as [1]:

bit
P RB u
R(u) = 12 7 i T T I
s

(6)

where P RBu is the average number of PRBs assigned to


UE u on a Transmission Time Interval (TTI) of 1 ms and 127

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

represents the number of resource elements per PRB. The


number of PRBs assigned to a generic cell-center (cell-edge)
UE u in a TTI, P RBu , is obtained by dividing the number of
PRBs of the cell-center (cell-edge) area by the number of cellcenter (cell-edge) UEs according to an RR scheme. Finally, the
macro/micro average cell capacity is obtained by summing the
throughput contributions R(u) for each UE in the cell.

N UMERICAL R ESULTS

In this Section, we provide simulation results and comparisons of 2B-SFR, 3B-SFR, and 3B-ISFR in terms of cell
outage probability and cell capacity, differentiating between
macro and micro cells and also providing aggregate values
(macro cell plus its micro cells). Let us refer to the outage
results in Fig. 2 as a function of for Nr = 4 micro cells per
macro cell, = 2/3, Nu = 1000 UEs/macro cell, and different
threshold bit-rate values Rs that only apply to the 3B-ISFR
scheme. We can see that 3B-ISFR has always a lower macro
cell outage probability than 3B-SFR. The macro cell outage
probability of 3B-ISFR depends on the selected Rs value: the
best results are achieved with Rs = 3 kb/s for 1000 UEs/macro
cell, when 3B-ISFR outperforms 2B-SFR for a wide range of
3 SINR has deterministic values since in this study we do not consider fading
variations, but just a shadowing margin is included in the path loss model [1].

3165

Macro Cell Outage [%]

We have implemented the LTE scenario and the SFR


schemes described in Sections II and III in a Matlab simulator.
This simulator takes several aspects of a real LTE system
into account, using different modules for path loss, SINR
calculation, etc. In particular, we use the Stanford University
Interim (SUI) model [8] for the propagation loss in macro
nodes (path loss exponent of 4.16 for the intermediate scenario
and shadowing margin of 15.8 dB [1]) and the WINNER
II model [9] for micro nodes (path loss exponent of 2.5
and extra attenuations due to walls so that there is a 15 dB
wall attenuation at 150 m from micro cell eNB). The basic
settings of our simulator are [1]: honeycomb regular macro
cell tessellation (macro cell radius R = 1664 m and
distance
between two neighboring macro cell eNBs d = R 3), macro
cell (micro cell) transmission power Pa = 46 dBm (Pb = 30
dBm), and system bandwidth of 20 MHz. The simulated area
consists of one central reference cell and two complete rings
of adjacent interfering cells. The number of UEs per macro
cell is Nu = 1000, the macro cell (micro cell) antenna height
hmacro (hmicro ) is equal to 40 m (2 m). Moreover, we consider
the following parameters for our system: the cell-edge-to-cellcenter power ratio , the center UE ratio , the number of
micro cells per macro cell Nr , and the threshold UE bit-rate
Rs . For the sake of simplicity, , , and Rs values are the
same for both macro and micro cells (then, = ). We have
validated our Matlab simulator comparing its SINR values3
with those of the LTE Ns-3 simulator [3] in different parts
of macro and micro cells. Simulation results are only taken
from the central reference macro cell since this cell has a
significant set of interfering adjacent cells. We repeated 30
times each point of the simulations (Monte Carlo method) with
different randomized UE positions and verified that the results
are reliable with maximum 95%-level confidence intervals of
4% around the mean central value.
V.

3BISFR Rs=1kb/s

20

S IMULATOR M ODEL D ESCRIPTION

3BISFR Rs=3kb/s
15

3BISFR Rs=6kb/s

10

3BSFR
2BSFR

5
0

10

10

15
Micro Cells Outage [%]

IV.

values. Moreover, the micro cell outage probability of 3BISFR is not sensitive to the Rs value and is always better than
2B-SFR, having excellent values for 4. On the basis of
these outage results, we can conclude that 3B-ISFR with Rs
= 3 kb/s and 4 is the best configuration for Nu = 1000
UEs/macro cell and = 2/3.

10

Fig. 2. Macro cell and micro cell outage probability versus for Nr = 4
and = 2/3

Fig. 3 shows the average capacity for both macro and micro
cells as a function of for Nr = 4 micro cells per macro cell,
= 2/3, Nu = 1000 UEs/macro cell, and different threshold bitrate values Rs for the 3B-ISFR scheme. The performance of
3B-ISFR is always better than 3B-SFR because of the 3B-ISFR
capability to improve the capacity available for most critical
cell-center UEs. Moreover, 3B-ISFR outperforms 2B-SFR in
terms of micro cell capacity on a wide range of values. As
for macro cells, 3B-ISFR outperforms 2B-SFR only for
4 (note that 2B-SFR achieves the highest macro cell capacity
for very small values, but this is not a good cell planning
strategy since the corresponding micro cell capacity and outage
probability values are quite poor), the same range of values
determined on the basis of outage considerations. In particular,
we notice that the adoption of Rs = 3 kb/s allows 3B-ISFR
to achieve the highest capacity for = 4.5 that is considered
as an optimal configuration. We have also performed other
simulation runs for different numbers of UEs per cell (Nu
from 200 to 1000) and different numbers of micro cells per
cell (Nr up to 10), obtaining that the optimal value for 3BISFR is still 4.5 for = 2/3 and that the optimal Rs values
are inversely proportional to Nu . Moreover, we have verified
that the optimized value around 4 (depending on outage and
capacity conditions) is not sensitive to Nu and values.
Fig. 4 shows the average total cell capacity (macro plus all
micro cells of the same macro cell), the overall cell outage
probability, and the overall percentage of critical UEs as
functions of the center UE ratio that governs the distinction
within a cell between cell-center and cell-edge UEs. This graph
has been obtained for = 4.5, Rs = 3 kbit/s, and Nu = 1000
UEs/macro cell. The proposed 3B-ISFR scheme outperforms
2B-SFR and 3B-SFR in terms of both the overall cell outage
probability and the total cell capacity. 3B-ISFR achieves the
maximum capacity for = 0.2, but the adoption of small

IEEE ICC 2015 - Mobile and Wireless Networking Symposium

5.5

1.8

1.6

4.5

3BSFR
2BSFR
Reuse 1

1.4
1.2

10

3BISFR Rs=1kb/s

Mirco Cells Capacity[b/s]

15
3BISFR
Rs=3kb/s

2.2

x 10

3BISFR R =3kb/s
s

3BISFR Rs=6kb/s

3BSFR
2BSFR

4
3.5
3
2.5

1.8

1.6

1.5

1.4

10

Fig. 3. Macro cell and micro cell capacity versus for Nr = 4 and = 2/3

central areas in macro/micro cell is not a good choice because


the percentage of critical UEs would be too high, thus preventing many UEs to attain a reasonable throughput. A trade-off
has to be achieved between cell capacity maximization and
reduction of the percentage of critical users, depending on Nu
and . Basically, we should adopt = 2/3 that allows the
same proportionality between number of UEs and bandwidth
assigned to them at both cell center and cell edge.
4

x 10

3
2
3BISFR R =3kb/s

6
Nr

10

10

6
Nr

10

Fig. 5. Impact of the number of micro cells per macro cell Nr for = 4.5
and = 2/3

VI.

C ONCLUSIONS

One of the major novelty of HetNets compared with


conventional cellular networks is the deployment of micro cells
in an unplanned way. However, this approach may lead to
increase both co-tier and cross-tier interference. In this paper,
we have proposed the 3B-ISFR scheme for LTE-A to address
at the same time co-tier and cross-tier interference. 3B-ISFR
achieves a good performance in terms of average total cell
capacity and overall cell outage probability. A further work
will be needed to refine the cell association scheme, to study
the user-level performance, to define a distributed version of
the 3B-ISFR scheme, and to perform a system optimization.

3BSFR
2BSFR

1
0

0.1

0.2

0.3

R EFERENCES
0.4

0.5

40
30
20
10
0

3BISFR
0

0.5

Critical UEs Percentage [%]

Overall Cell Capacity [b/s]

Overall Cell Outage Probability [%]

x 10

Overall Cell Outage Probability [%]

x 10

Overall Cell Capacity [b/s]

Macro Cell Capacity [b/s]

0.6

0.7

0.8

0.9

40
30
20
10
0

0.5

Fig. 4. Impact of cell-center UE ratio in terms of total cell capacity, overall


outage probability, and critical UEs percentage for = 4.5 and Rs = 3 kb/s

Finally, Fig. 5 shows the average total cell capacity and the
overall cell outage probability as functions of Nr for = 4.5,
Rs = 3 kb/s, and = 2/3. These graphs also show the results of
the full frequency reuse scheme used in both macro and micro
cells (no distinction between cell center and cell edge), thus
improving the bandwidth available in each cell (no bandwidth
subdivision among cells) at the expenses of much worse outage
conditions. The advantage of 3B-ISFR with respect to 3BSFR and 2B-SFR increases with the number of micro cells
per macro cell. The full frequency reuse scheme can achieve a
slightly higher capacity, but outage probability is too high. We
can conclude that our proposed 3B-ISFR scheme attains better
performance than other schemes, thus representing a promising
approach for HetNets.

3166

[1] G. Giambene, T. A. Yahiya, V. A. Le, K. Grochla, K. Poys, Resource


management and cell planning in LTE systems, in Wireless Networking
for Moving Objects. Springer, 2014, pp. 177197.
[2] M. Qian, W. Hardjawana, Y. Li, B. Vucetic, X. Yang, J. Shi, Adaptive
soft frequency reuse scheme for wireless cellular networks, Vech. Tech.,
IEEE Transaction on, vol. 64, no. 1, pp. 118131, January 2015.
[3] G. Giambene, V. A. Le, Performance evaluation of different fractional
frequency reuse schemes for LTE, in Proc. of the FITCE International
Conference, 53th, Naples, 12-15 November 2014.
[4] L. Zhou, R. Q. Hu, Y. Qian, H. H. Chen, Energy-spectrum efficiency
tradeoff for video streaming over mobile ad hoc networks, Selected
Areas in Communications, IEEE Journal on, vol. 31, no. 5, pp. 981
991, May 2013.
[5] D. Lopez-Perez, X. Chu, A. V. Vasilakos, H. Claussen, Power minimization based resource allocation for interference mitigation in OFDMA
femtocell networks, Selected Areas in Communications, IEEE Journal
on, vol. 32, no. 2, pp. 333344, 2014.
[6] R. Q. Hu, Y. Qian, An energy efficient and spectrum efficient wireless
heterogeneous network framework for 5G systems, IEEE Communications, vol. 52, no. 5, pp. 94101, May 2014.
[7] Q. Li, R. Q. Hu, Y. Xu, Y. Qian, Optimal fractional frequency reuse and
power control in the heterogeneous wireless networks, Wireless Comms.,
IEEE Transactions on, vol. 12, no. 6, pp. 26582668, 2013.
[8] V. Erceg, et al., An empirically based path loss model for wireless
channels in suburban environments, Selected Areas in Communications,
IEEE Journal on, vol. 17, no. 7, pp. 12051211, July 1999.
[9] P. Kyosti, et al., WINNER II Channel Models, Deliverable ISTWINNER D1.1.2 ver 1.2, 2007 [online] Available: http://projects.celticinitiative.org/winner+/WINNER2-Deliverables/D1.1.2v1.2.pdf

Vous aimerez peut-être aussi