Professional Documents
Culture Documents
AbstractThe need to carry out analytical studies of wireless station deployment [5], it becomes important to re-evaluate
arXiv:1412.5065v1 [cs.IT] 16 Dec 2014
systems often motivates the usage of simplified models which, the performance of these network deployments using more
despite their tractability, can easily lead to an overestimation of accurate channel models.
the achievable performance. In the case of dense small cells net-
works, the standard single slope path-loss model has been shown A. Related Work
to provide interesting, but supposedly too optimistic, properties
such as the invariance of the outage/coverage probability and of A well-known stochastic geometry framework for wireless
the spectral efficiency to the base station density. cellular networks can be found in [1]; by assuming a single
This paper seeks to explore the performance of dense small slope path loss model, the authors have observed the indepen-
cells networks when a more accurate path-loss model is taken
into account. We first propose a stochastic geometry based dence of the Signal-to-Interference-plus-Noise-Ratio (SINR)
framework for small cell networks where the signal propagation and Spectral Efficiency (SE) from the BS deployment density,
accounts for both the Line-of-Sight (LOS) and Non-Line-Of- where the main consequence is the linear dependence of the
Sight (NLOS) components, such as the model provided by the ASE on the deployment density.
3GPP for evaluation of pico-cells in Heterogeneous Networks. In some recent work, the effect of cell densification while
We then study the performance of these networks and we show
the dependency of some metrics such as the outage/coverage assuming different models than single slope has been con-
probability, the spectral efficiency and Area Spectral Efficiency sidered. For instance, in [2] it was shown that by using
(ASE) on the base station density and on the LOS likelihood a combined Line-Of-Sight (LOS)/Non-Line-Of-Sight (NLOS)
of the propagation environment. Specifically, we show that, with path loss model the ASE becomes superlinear at low densities
LOS/NLOS propagation, dense networks still achieve large ASE and sublinear at high densities. However that work was based
gain but, at the same time, suffer from high outage probability.
Index TermsSmall cells, dense deployment, LOS/NLOS, on a simulation study, lacking an analytical framework to back
stochastic geometry, Area Spectral Efficiency. the reached conclusions.
In [3], the authors have studied the non-linear behaviour of
I. I NTRODUCTION the ASE with the deployment density in indoor scenarios, and
The evaluation of wireless communications systems com- have included an exponential component in the path loss to
monly resorts to the use of simplified channel models, with account for the wall attenuation. As a result, the ASE was
the purpose of simplifying the analytical formulation of the shown to scale as N with the number N of cells.
associated system model. While simplified channel models are In [4], the authors extended the stochastic geometry frame-
more tractable, they can easily lead to inaccurate results and work proposed in [1] to a multi-slope path loss model. They
consequent wrong conclusions. focused their analysis on a dual-slope propagation model
A concrete example is the case of cell-splitting in cellular that considers two propagation regimes, namely the near-field
networks, where the network performance has been assessed and the far-field, with different attenuation exponents. The
using a single slope path loss model. By considering this proposed model is particularly suitable for millimeter wave
model, it has been shown that the Area Spectral Efficiency communications, but is essentially different from the one
(ASE) increases linearly with the Base Station (BS) deploy- proposed by 3GPP [6] for the assessment of heterogeneous
ment density whereas the outage probability is independent networks in lower frequency bands (e.g., 2 GHz).
of the BS density [1]. Moreover, the energy efficiency turns Finally, the effect of NLOS propagation on the outage prob-
out to be a monotonic increasing function of the cell density ability has been studied in [7], where the authors propose a
[2]. Nonetheless, when the assumption of single slope path- function that gives the probability to have LOS at a given point
loss is dropped and different path-loss models are used, those depending on the distance from the source, on the average size
aforementioned properties do no longer hold [2][4]. of the buildings and on the density of the buildings per area.
Although the results obtained with single slope path loss Although it allowed the authors to show the dependence of the
can be considered reliable if the BS density is within a limited outage probability on the density of base stations and of the
range of values, the same might not be true if one needs to density of buildings per area, the propagation model proposed
extend this result to a wider BS density range. Therefore, as the in [7] lacks a validation based on a measurement campaign
trend of future networks is shifting towards ultra dense base with real data.
B. Our Contribution propagation,2 respectively; pL (d) is the probability of having
In this paper, we focus on a dense network of small cells LOS as a function of the distance d. We further assume
where the signal propagation considers a path loss model that the propagation is affected by Rayleigh fading, which is
with LOS and NLOS components, and we use stochastic exponentially distributed exp().
geometry tools to evaluate common performance metrics (such The incorporation of the NLOS component in the path loss
as coverage/outage probability, SE and ASE) as a function of model accounts for possible obstructions of the signal due
the base station deployment density. Specifically, we consider to large scale objects (e.g. buildings), which will result in a
the propagation model provided by 3GPP for heterogeneous higher attenuation of the NLOS propagation compared to the
networks at 2 GHz, as we believe this is a reliable model for LOS path.
the performance evaluation of small-cells and dense networks. B. LOS probability function
Our contribution is threefold: (i) we propose a tractable model
for the LOS probability function which well fits the 3GPP In our study, we refer first to the LOS probability function
model and, at the same time, it enables the study of the proposed by 3GPP [6, Table A.2.1.1.2-3] to assess the network
performance depending on the LOS likelihood of the envi- performance in scenarios with pico-cells deployment; this
ronment; (ii) we derive a stochastic geometry framework that function is the following:
incorporates the proposed LOS probability function; finally d0
pL,3G (d) = 0.5 min 0.5, 5e d + min 0.5, 5e d1 . (2)
d
The PGFL in (13) along with (10) and (A.3) can be plugged proposed by [7] to check whether this function provides results
in (4) to obtain the SINR ICDF through numerical integration. that are with in line with those obtained with the 3GPP model.
The validation of (3) is only carried out for L = 82.5
F. Average Spectral Efficiency and Area Spectral Efficiency m, whereas the other two values of the parameter L will
Similarly to [1, Section IV] we compute the average spectral solely be used to study the effect of different propagation
efficiency and the ASE of the network. First, we define the environments on the network performance. To benchmark
ASE as: the LOS probability function we compute the SIR CDF by
numerical integration of (4), while assuming (3) as pL (d);
#BS BW C A BW C
ASE = = = C, (14) we then compare this curve with the SIR CDF evaluated
A BW A BW through Monte Carlo simulations assuming (2) as pL (d). The
where BW is the bandwidth, C is the average spectral effi- benchmark is shown in Fig. 2. As we can from this plot, the
ciency, A is the area, #BS is the number of base stations two CDFs well match for different values of BS densities;
within the area A. this implies that our model is well suited to study the effect
1 4.5
3GPP model L = 120m
0.9 2 L = 82.5m
exp(d/L) , L = 40 m 4
L = 40m
0.8 exp(d/L)2, L = 82.5 m
0.6
0.5 2.5
0.4
2
0.3
1.5
0.2
1
0.1
0 0.5
0 1 2 3 4
0 50 100 150 200 250 300 10 10 10 10 10
Distance [m] BSs density [#BSs/km2]
Fig. 1. The function exp(d p) [7] has been plotted to intersect the Fig. 3. Spectral efficiency vs base station density.
3GPP model at the point corresponding to probability 0.5. The corresponding
values and p are 8.59 103 and 1.01 101 , respectively. 2
10
100
= 100 BS/km2
90 3
10
50
40 5
10
30 L = 120m
L = 82.5m
20 3GPP model L = 40m
6
10 exp(d/L)2 10
0 1 2 3 4
10 10 10 10 10
exp( dp)
BSs density [#BSs/km2]
0
30 20 10 0 10 20 30 40 50
SIR [dB]
Fig. 4. ASE vs base station density.
Fig. 2. Comparison of SIR CDFs for different pL (x). for observing a non constant behaviour of the SE curve as a
of the LOS/NLOS propagation on the performance of dense function of the BS density is the LOS/NLOS propagation.
networks, as it yields results in line with those obtained with The peak of the curve is observed for a given value of the BS
the 3GPP model. On the contrary, the SIR obtained with (17) density which we will refer to as optimal BS density for the SE,
as pL (d) only matches the 3GPP benchmark for high cell denoted by SE . At the optimal BS density, the user is likely
densities; in fact, at high densities the strongest interferers to be in LOS with the serving BS, while it is in NLOS with
are in LOS with the user regardless of the shape of the most of the interferers, meaning that the interference power is
pL (d) functions we have tested. Using (17) as probability LOS low. For BS densities lower than SE , the serving BS is likely
function would give us accurate results only for a limited range to be in NLOS with the users with a consequent reduction of
of BSs density and, therefore, would not be suitable for the the received power, of the SINR and therefore of the SE. On
kind of study we carry out in our paper. the contrary, at BS densities higher than SE , the interfering
BSs are likely to be in LOS with the user, causing an overall
B. Spectral efficiency, ASE and outage probability interference growth and thus a reduction of the SIR and SE.
To study the performance of the dense networks as a The behaviour of the SE curve is also influenced by the LOS
function of the BS density, we evaluated numerically the probability as a function of the distance. In fact, looking at the
SE (eq. (16)), the ASE (eq.(14)) and the outage probability plots in Fig. 3, we notice that the optimal SE density depends
(obtained from eq. (4)). on the LOS likelihood parameter L. In sparser propagation
We first present the SE results, which are shown in Fig. 3. environments (e.g., L = 120m) the propagation is likely to be
From this plot, we can see that the SE curve is not constant of the LOS kind at longer distances from the user, compared
but exhibits a peak for a given value of base station density. to the case of dense propagation environments; this means that
This result is in contrast with what we would obtain if we SE will be reached at a lower BS density.
used a single slope propagation model, for which the SE has Fig. 4 shows the ASE curves for different LOS propaga-
been shown to be constant and thus independent of the base tion environments. Unlike the SE, the ASE is an increasing
station density [1]. Therefore, we can infer that the reason function of the BS density and this is due to the effect of
0.7 is the current standard reference model when studying wireless
L = 40m, th = 5dB networks.
L = 40m, = 10dB
0.6 th
L = 82.5m, th = 5dB A PPENDIX A
L = 82.5m, th = 10dB
N EAREST NEIGHBOUR DISTANCE PDF
Outage probability P[ ]
th
0.5
L = 120m, th = 5dB
L = 120m, th = 10dB If we assume the LOS probability to be given by (3), we
0.4
can further develop (10) as follows:
Z
0.3
P [r > R] = exp pL (kxk)dx
0.2 B(0,R)
Z
0.1 exp (1 pL (kxk)) dx . (A.1)
B (0,d1
eq (R))
0
10
0 1
10 10
2 3
10
4
10 By solving the integrals in (A.1) and with further symbolic
BSs density [#BSs/km2] manipulation we obtain:
2 R2
eq
Fig. 5. Outage vs base station density. R2
L2 e L2 e 2
L2
P [r > R] = e L
e
(A.2) eReq ,
the cell densification, as shown in (14). Nonetheless, the SE where R = d1 (R). Let us define the functions f (R),
eq eq 1
has an impact on the ASE, as we can notice from the lower f (R), f (R) and their first derivatives f1 (R), f2 (R), and
2 3
steepness of the ASE curve. In particular, the ASE switches f (R), respectively, as follows:
3
from a superlinear gain at low cell densities to a sublinear
2 R2
eq
gain at high cell-densities. To complete the analysis of the R
L2 e L2 L2 e L2
f1 (R) = e , f2 (R) = e ,
performance of cell densification, we also need to evaluate the
2 R2
outage probability curves, shown in Fig. 5. From this plot we f3 (R) = eReq , f1 (R) = 2Re L2 f1 (R),
can see that, at higher BS densities, the outage starts growing Keq2 R2eq
drastically and, depending on the LOS likelihood, it can reach f2 (R) = Keq 2
2eq R2eq 1 e L2 f2 (R),
22-32% for -10dB SIR threshold and 50-60% for -5dB SIR 2 2eq 1
threshold. This is due to the effect of the interfering BSs which, f3 (R) = Keq 2eq R f3 (R).
at high cell densities, are likely to be in LOS with the user. Then we can write P [r > R] as P [r > R] =
This will be perceived by the user as a strong interference, f1 (R)f2 (R)f3 (R). The PDF of the minimum distance
which causes an SIR reduction. In addition, let us notice that from the serving base station is given by:
in environments with higher NLOS probability, the optimum d d
of the outage curve is achieved at higher BS densities. fr (R) = (P [r > R]) = [f1 (R)f2 (R)f3 (R)]
dR dR
In the light of our study, we make the argument that, despite
the high gain provided in terms of cell ASE, cell densification = f1 (R)f2 (R)f3 (R) + f1 (R)f2 (R)f3 (R)
+ f1 (R)f2 (R)f3 (R) .
can severely affect the outage probability and hence the (A.3)
network coverage. Therefore, if ones needs an extremely dense
R EFERENCES
cell deployment to meet some given capacity requirements,
networks will require interference handling techniques in order [1] Coverage
J. G. Andrews, F. Baccelli, and R. K. Ganti, A Tractable Approach to
and Rate in Cellular Networks, IEEE Trans. Wireless Commun.,
to operate in a satisfactory manner. vol. 59, no. 11, pp. 31223134, 2011.
[2] C. Galiotto, I. M. Gomez, N. Marchetti, and L. Doyle, Effect
V. C ONCLUSIONS of LOS/NLOS Propagation on Area Spectral Efficiency and Energy
Efficiency of Small-Cells, To appear in the proceedings of Globecom
In this paper we have proposed a stochastic geometry frame- 2014, 2014. [Online]. Available: http://arxiv.org/abs/1409.7575
work to study the coverage/outage probability, the Spectral [3] J. Ling and D. Chizhik, Capacity Scaling of Indoor Pico-Cellular
Efficiency (SE) and the Area Spectral Efficiency (ASE) of Networks via Reuse, IEEE Commun. Lett., vol. 16, no. 2, pp. 231233,
Feb. 2012.
dense small cell networks where the signal path-loss model [4] X. Zhang and J. G. Andrews, Downlink Cellular Network Analysis
includes both LOS and NLOS components. Through our for- with Multi-slope Path Loss Models, 2014. [Online]. Available:
mulation, we have investigated how the network performance http://arxiv.org/abs/1408.0549
[5] N. Bhushan, L. Junyi, D. Malladi, R. Gilmore, D. Brenner, A. Damn-
scales with the density of the base station deployment and we janovic, R. Sukhavasi, and S. Geirhofer, Network Densification: the
studied the effect of the propagation environment in terms of Dominant Theme for Wireless Evolution into 5G, IEEE Communications
LOS likelihood on SE, ASE and outage probability. Magazine, 2014.
[6] 3rd Generation Partnership Project (3GPP), Further Advancements for E-
Finally, we have shown that propagation environments with UTRA Physical Layer Aspects (Release 9), Mar. 2010, 3GPP TR 36.814
high NLOS probability allow for a denser cell deployment. V9.0.0 (2010-03).
Overall, in this paper we sought to uncover insights on how [7] T. Bai, R. Vaze, and R. W. Heath, Analysis of Blockage Effects on Urban
Cellular Networks, IEEE Trans. Wireless Commun., vol. 13, no. 9, pp.
dense networks should be deployed while taking into account 50705083, Sep. 2014.
a model that more accurately describes the signal propagation [8] M. Haenggi, Stochastic Geometry for Wireless Networks. Cambridge
compared to the more common single slope path-loss, which Press, 2013.