Sie sind auf Seite 1von 5

Stochastic Geometry Modeling and Performance

Evaluation of Downlink MIMO Cellular Networks


Peng Guan(1) , Marco Di Renzo(1) , and T. Q. Duong(2)
(1)
Paris-Saclay University – Laboratory of Signals and Systems (L2S, UMR-8506)
CNRS – CentraleSupelec – University Paris-Sud XI
3 rue Joliot-Curie, 91192 Gif-sur-Yvette (Paris), France
e-mail: {peng.guan, marco.direnzo}@lss.supelec.fr
(2)
Queen’s University Belfast – School of Electronics, Electrical Engineering and Computer Science
University Road Belfast, BT7 1NN, Northern Ireland, UK
e-mail: trung.q.duong@qub.ac.uk

Abstract—In this paper, a mathematical framework for evalu- networks with biased cell association and for transmission
ating the error probability of downlink Multiple-Input-Multiple- over general fading channels. In [10], Multiple-Input-Multiple-
Output (MIMO) cellular networks is introduced. It is based on Output (MIMO) cellular networks are studied with the aid
the Poisson Point Process (PPP)-based abstraction for modeling
the spatial locations of the Base Stations (BSs) and it exploits of tools from stochastic geometry and stochastic ordering. In
results from stochastic geometry to characterize the distribution [11], the error probability of single-antenna cellular networks
of the other-cell interference. The framework is applicable to is investigated. Recent results are available in [12]-[15].
spatial multiplexing MIMO systems with an arbitrary number The papers mentioned above are focused on the computation
of antennas at the transmitter (Nt ) and at the receiver (Nr ). It of the coverage/outage probability and of the average rate of a
is shown that the proposed framework leads to easy-to-compute
integral expressions, which provide insights for network design typical Mobile Terminal (MT). In addition, with the exception
and optimization. The accuracy of the mathematical analysis of a few papers, e.g., [10], the mathematical analysis is limited
is substantiated through extensive Monte Carlo simulations for to single-antenna BSs and to single-antenna MTs. Compared
various MIMO cellular network setups. to [10], in particular, the present paper has three main differ-
ences/contributions: i) the performance metrics. We focus our
I. I NTRODUCTION
attention on the error probability, while the authors of [10]
Stochastic geometry is emerging as a powerful mathematical study coverage probability and area spectral efficiency; ii) the
tool for modeling, analyzing and optimizing cellular networks, downlink MIMO transmission scheme. We focus our attention
since it allows one to circumvent the mathematical intractabil- on open-loop spatial multiplexing MIMO transmission, while
ity of the abstraction models used in the past, e.g., the hexag- the authors of [10] study closed-loop MIMO transmission;
onal grid model. With the aid of stochastic geometry [1], [2], and iii) the method of analysis. In [10], the authors exploit
the locations of the Base Stations (BSs) are modeled as points stochastic ordering for performance analysis and comparison.
of a point process. For mathematical tractability, a Poisson Our mathematical framework, on the other hand, is based
Point Process (PPP) is usually adopted [3]. Recent results have on the Gil-Pelaez inversion theorem [16]. Compared to [11],
confirmed that the PPP-based abstraction model is capable of we consider multiple antennas at both the BS and the MT.
accurately reproducing the main structural characteristics of It is worth mentioning, finally, that the error probability
operational cellular networks [3]. in the presence of a Poisson field of interferers has been
Owing to its mathematical tractability and accuracy, the studied in the literature. In [17], in particular, the authors have
PPP-based abstraction model is now routinely used to the introduced a framework for computing the error probability of
analysis and design of wireless networks in general and a multi-antenna receiver in the presence of different models of
cellular networks in particular. Notable examples include [4]- network interference. This mathematical framework, however,
[11]. In [4], the coverage probability and the average rate is not applicable to cellular networks, since the BS-to-MT cell
of cellular networks for transmission over Rayleigh fading association is neglected and the interferers can be closer to the
channels are computed in closed-form. In [5], the framework in typical MT than the serving BS.
[4] is extended to heterogeneous cellular networks, which are In the present paper, motivated by these considerations, we
modeled as the superposition of many PPPs. In [6], the PPP- introduce a mathematical framework for computing the error
based abstraction model is exploited to study heterogeneous probability of spatial multiplexing MIMO cellular networks,
cellular networks with a biased cell association criterion. In where the locations of the BSs are distributed according
[7], the authors incorporate the load characteristics of the BSs to a homogeneous PPP and the cell association criterion is
into the mathematical framework by using a conditionally thin- explicitly taken into account. More specifically, the technical
ning approach. In [8], offloading strategies for heterogeneous contribution of the present paper is threefold: 1) we provide
cellular networks are investigated by relying on a stochastic an exact closed-form expression of the Characteristic Function
geometry framework. In [9], the authors introduce a frame- (CF) of the aggregate other-cell interference at the MT; 2)
work for computing the average rate of heterogeneous cellular by using the Gil-Pelaez inversion theorem [16], we provide

978-1-4799-8091-8/15/$31.00 ©2015 IEEE


an exact expression of the Average Pairwise Frame Error the signal received at the MT is as follows:
Probability (APEP(F) ), which is averaged over both the fading  
y= E/Nt r0−b H0 s0 + E/Nt r −b Hi si +n (1)
distribution and the deployment of BSs. From the APEP(F) ,  
 i∈Ψ(\0) i


x
the Average Frame Error Probability (AFEP) is obtained by iagg (r0 )

using the Nearest Neighbor (NN) approximation [18]; and Nr ×1


where x ∈ C is the useful signal transmitted by BS0 ,
3) asymptotic frameworks are proposed, in order to provide
iagg (r0 ) ∈ CNr ×1 is the aggregate other-cell interference and
insights on the achievable error performance as a function of
n ∈ CNr ×1 is the Additive White Gaussian Noise (AWGN)
important system parameters.
defined as n(r) ∼ CN (0, N0 ). More specifically: i) E is the
The remainder of the present paper is organized as follows. BSs transmit-energy per transmission; ii) s0 is the vector of
In Section II, signal and system models are described. In Sec- (t)
information symbols emitted by BS0 , where s0 ∈ M. The
tion III, the error performance of spatial multiplexing MIMO set M has M modulated symbols denoted by μχ ∈ C; iii)
cellular networks is studied. In Section IV, the mathematical
H0 ∈ CNr ×Nt is the channel matrix of the BS0 -to-MT link;
framework is validated against Monte Carlo simulations and
iv) b > 1 denotes the amplitude path-loss exponent. A similar
relevant performance trends are highlighted. Finally, Section notation is adopted for all the interfering channels of iagg (·).
V concludes this paper.
At the MT, an interference-unaware Maximum-Likelihood
Notation: In the present paper, the following notation is (ML)-optimum demodulator is considered. It is assumed to
used. z ∗ , |z| and arg {z} denote the conjugate, the modulus have perfect Channel State Information (CSI) of the BS0 -
and the phase of a complex number z, respectively. C denotes to-MT link, while ignoring the other-cell interference. The
the field of complex numbers. x ∈ SN ×1 denotes a N × 1 decision metric Λ (·) can be formulated as follows [20]:
column-vector with entries belonging to the set S. The lth 2
entry is denoted by x(l) . X ∈ SN ×M denotes a N × M
E −b
Λ (s̃0 ) = y − r H0 s̃0
matrix with entries belonging to the set S. The (l, m)th entry Nt 0
(2)
is denoted by X(l,m) . card {S} denotes the cardinality of
−2b E −b E −b E
the set S. x denotes the norm of vector x. X denotes ∝ r0 U + 2r0 Re {I(r0 )} + 2r0 Re {N}
Nt Nt Nt
the Frobenius
 norm
 of matrix
 X.
 j denotes the imaginary

Nr Nt 
 ∗
unit. CN μ, σ 2 and N μ, σ 2 denote a complex and a (r) (r,t) (t)
where Δ0 = s0 − s̃0 , I(r0 ) = iagg (r0 ) H0 Δ0 ,
real Gaussian distribution with mean μ and variance σ 2 , r=1 t=1

Nr  
 2 Nt  ∗
 
 H(r,t) Δ(t)  , N =  n(r)  H(r,t) Δ(t) .
respectively. (·)k denotes the Pochhammer symbol. (·)! de- Nt Nr
notes the factorial operator. Re {·} and Im {·} denote the real U=  0 0  0 0
r=1 t=1 r=1 t=1
and imaginary part operators, respectively. E {·} denotes the Assuming Rayleigh fading for mathematical
2
expectation operator. fX (·) denotes the Probability Density  (r,t) 
tractability, we have H0  ∼ Gamma (1, Ω) and
Function
  (PDF)
 Variable (RV) X. ΦX (ω) =
of a Random 
the channel phases are uniformly distributed. Thus,
E exp j ω1 X (re) + ω2 X (im) denotes the CF of the (r,t)
H0 ∼ CN (0, Ω) for t = 1, 2, . . . , Nt and r = 1, 2, . . . , Nr .
complex RV X = Re {X} + jIm {X} = X (re) + jX (im) ,
As a result, by conditioning  upon Δ0 , we have
where ω = (ω1 , ω2 ). For notational simplicity, the short-hand Nt (r,t) (t) 2
ΦX (ω) = E {exp {jωX}} is used. Pr {·} denotes probabil- t=1 H0 Δ 0 ∼ CN |Δ0 0, Ω Δ 0  . This implies
 
ity. Γ (·) is the Gamma function. p Fq (a1 , . . . , ap ; b1 , . . . , bq ; ·) 2
that U ∼ Gamma Nr , ΩNr Δ0  .
is the generalized hypergeometric function [19, Ch. IV].
III. AVERAGE E RROR P ROBABILITY
II. S YSTEM M ODEL A. Characteristic Function of the Other-Cell Interference
As a first step for computing the error probability, we pro-
Let us consider the downlink of a cellular network, where vide a new expression for the CF of the other-cell interference.
a probe Nr -antenna MT is located at the origin of the bi- Let us assume, similar to the probe link, that the interfering
dimensional plane and the Nt -antenna BSs are modeled as channels Hi for i ∈ Ψ(\0) are independent and identically

points of a homogeneous PPP (Ψ) of density λ. The distance  (r,t) 2
from the ith BS to the MT is denoted by ri for i ∈ Ψ. distributed (i.i.d) Rayleigh RVs with E Hi  = Ω. By
The MT is assumed to be tagged to the nearest BS. The conditioning upon r0 and upon U given in (2), the CF of I(·)
serving BS is denoted by BS0 , and its distance from the in (2) can be formulated as follows:
MT is denoted by r0 , which  is a RV with PDF equal to ΦI (ω; r0 , U) = ΦI (|ω| ; r0 , U)
fr0 (ξ) = 2πλξ exp −πλξ 2 [4]. The set of interfering BSs    (3)
= exp −λπr02 T |ω|2 r0−2b EU
i ∈ Ψ\ {BS0 } is still a homogeneous PPP [2], which is
denoted by Ψ(\0) . Ψ(\0) has density pλ, where 0 < p ≤ 1 with T̄ (x) = pῩ (x) − p and:
denotes the activity factor of the BSs, i.e., the probability that ⎧  q ⎫
an interfering BS i ∈ Ψ(\0) transmits at the same frequency as 
⎨ +∞ 
1 q 1 (−1/b)q si 2 ⎬
q
Ῡ (x) = Esi − (Ωx)
BS0 . The setup with p = 1 corresponds to the full frequency ⎩ 4 q! (1 − 1/b)q Nt ⎭
q=0
 ! (4)
reuse case [4], [9]. 1 1 x si 2
In the depicted downlink MIMO cellular network model, = Esi 1 F1 − ; 1 − ; − Ω
b b 4 Nt
( +∞ −1
where the expectation Esi {·}, which is computed with respect exp {−y} (1 − exp {−Ky}) dy = 1 − (K + 1)
0
to the Nt -tuple of information symbols si ∈ CNt ×1 , can be for K > 0. 
formulated as follows: These asymptotic frameworks provide relevant insights on
⎛ ⎞
   M
M
M Nt the error probability as a function of important system param-
K 2 1 ⎝ K 2⎠
Esi g si  = · · · g |μ χ t | eters. For example, (10) shows that, in the interference-limited
Nt M Nt χ =1 χ =1 Nt t=1
1 2 χNt =1
regime, the error probability is independent of SNR = E/N0
(5)
and of the density λ of BSs. In the noise-limited regime, on
and g(·) is a generic function and K > 0 is a positive constant.
the other hand, the error probability decreases by increasing
Proof : See Appendix I.  SNR = E/N0 and λ. These trends are in agreement with
B. Average Pairwise Frame Error Probability intuition and are substantiated in Section IV with the aid of
Monte Carlo simulations.
The second step lies in the computation of the APEP(F) ,
which is defined as the probability that the transmitted vector D. Average Frame Error Probability
s0 is decoded as ŝ0 = s̄0 = s0 , by assuming that s0 and s̄0 are
Finally, the AFEP can be obtained from the APEP(F) in (6)
the only two information vectors possibly being transmitted.
by using the NN approximation [18, p. 138]. More specifically,
Let SNR = E/N0 . Then, the APEP(F) can be formulated
(F) the following holds:
as APEP(F) (Δ0 ) = Pr {Λ (s̃0 ) < 0} = APEPN (Δ0 ) +
(F)  ) M N
t (χ)  
APEPNI (Δ0 ), which are defined in (6) at the bottom of this Nt (χ) 2
AFEP ≈ 1 M NΔ 2 APEP(F) Δmin (10)
page, where the following definitions hold: χ=1
min

   2
√ SNRU  
where: i) Δ(χ)
(F)
QN (x; Δ0 , SNR/Nt , b) = EU U exp − µ − µ 2 , µ Nt ×1
4Nt x 2b min = min
µ χ̃ χ̃ χ χ̃=χ ∈ M is the
   (7)  
(F) 1√ 1 minimum Euclidean distance among all pairs μχ̃ , μχ of the
QNI (x; Δ0 ) = EU sin Ux 2
2 (χ)
Nt -dimensional constellation diagram and ii) NΔ 2 is the
min
Proof : See Appendix II.  number of nearest neighbors of μχ . If the standard square
With the aid of [21, Eq. (3.381.4), Eq. (3.952.7)] and of the Quadrature Amplitude Modulation (QAM) with M symbols is
Kummer’s transformation 1 F1 (1 − m; 3/2, x) exp {−x} = considered, for example, the following holds:
 
1 F1 (m + 1/2; 3/2, −x), for arbitrary K1 > 0 and • Δmin 2 , NΔmin 2 = (4, 2) if (M = 2, Nt = 2)
 
K2 > 0 positive constants  and a Gamma RV U ∼ •
2
Δmin  , NΔmin 2 = (2, 4) if (M = 4, Nt = 2)
2  
Gamma Nr , ΩNr Δ0  , the expectations in (7) can be •
(avg)
Δmin 2 , NΔ = (0.4, 6) if (M = 16, Nt = 2)
2
formulated in closed-form as follows:  min  
2
√  • Δmin  , NΔmin 2 = (4, 4) if (M = 2, Nt = 4)
 
EU U exp {−K1 U} 2
• Δmin  , NΔmin 2 = (2, 8) if (M = 4, Nt = 4)
     −1  −(Nr + 12 )
NrNr Γ Nr + 12 (χ)
=   Nr K 1 + ΩΔ0  2
In addition, NΔ 2 ∈ {4, 5, 6, 7, 8} if (M = 16, Nt = 2).
min
ΩNr Δ0 2 Γ (Nr )
  √  (8)
IV. N UMERICAL AND S IMULATION R ESULTS
EU sin K2 U
&   In this section, some numerical examples are shown in order
ΩΔ0 2 Γ Nr + 12  
1 3 1 to substantiate the accuracy of the proposed mathematical
= K21 F1 Nr + ; , − ΩΔ0 2 K22
Γ (Nr ) 2 2 4 frameworks and to investigate related performance trends.
C. Asymptotic Analysis Monte Carlo simulations are obtained by using the procedure
For noise- (SNR → 0) and interference-limited (SNR → described in [9, Sec. V].
+∞) cellular networks, simpler asymptotic frameworks can In Fig. 1, we show Monte Carlo simulation results in order
be obtained from (6). In particular, the following holds: to compare the accuracy of the PPP-based abstraction model
  against the regular grid-based abstraction model. A similar
(F)
lim APEP(F) (Δ0 ) = APEPN (Δ0 ) comparison is available in [4, Sec. V.A] for the coverage prob-
SNR→0
  ' +∞ (F) (9) ability. We focus our attention, on the other hand, on the AFEP.
1 T (Nt x) QNI (x)
lim APEP(F) (Δ0 ) = dx Fig.1 shows that the PPP-based abstraction model (solid lines)
SNR→+∞ 2π 0 1 + T (Nt x)
provides an upper-bound of the error probability compared to
Proof : In the noise-limited regime, the simplification the grid-based abstraction model (dashed curves). This is due
follows by definition. In the interference-limited to the fact that some interfering BSs may be arbitrarily close
regime, the simplification follows from (6) and to each other in the former case. The PPP-based abstraction

'  2
(F) SNR/Nt +∞ e−πλx (F)
b
APEPN (Δ0 ) = √ (b+1)
QN (x; Δ0 , SNR/Nt , b) dx
2 π 0 x
' +∞ ' +∞ (6)
xy b Nt * +
(F) − (F)
APEPNI (Δ0 ) = (2π)−1 (1 − x) e−y e 4πb λb SNR 1 − e−yT(Nt x) QNI (x; Δ0 ) dxdy
0 0
(a) (b)
0
10
NtxNr=1x1 NtxNr=1x1, λ=λ0
0
0
NtxNr=2x2 10 NtxNr=2x2, λ=λ0/2
10 NtxNr=4x4 NtxNr=4x4, λ=λ0/4
−1
10

−1
−1 10
10
−2
10

AFEP

AFEP
AFEP

−2
−2 10
10
−3 0
10 p=10
−1
p=10
−2 −3 −3
p=10 10 10
−4 −3
10 p=10
−4
p=10
−4 −4
10 10
−5
10
120 140 160 180 200 220 240 120 160 200 240 120 160 200 240
E/N0 [dB] E/N0 [dB] E/N0 [dB]

Fig. 1. AFEP of a Nt × Nr = 2 × 2 system against the reference Fig. 4. AFEP of a system with Rate = 4 bpcu against the reference SNR.
SNR. Comparison between PPP- (solid lines) and grid-based (dashed lines) Setup: (a) b = 3, λ = 10−5 , p = 10−3 . (b) b = 3, λ0 = 10−5 , p = 10−3 .
abstraction models. Setup: 16QAM, λ = 10−5 , b = 3.

(a) (b) Carlo simulations (“black” curves) and the proposed mathe-
matical framework (the other curves). In particular, the NN
b=1.5 λ=10−2
b=2.0
λ=10−3 approximation provides a close estimate of the AFEP.
0 b=2.5 0
10
b=3.0
10 λ=10−4 In addition, the following performance trends emerge by
b=3.5 λ=10−5 direct inspection of the figures. Figure 1 shows that the
λ=10−6
AFEP decreases by decreasing the activity factor p, as this
10
−1
10
−1
significantly reduces the net contribution of the other-cell in-
AFEP

AFEP

terference. Figure 2 shows the important impact that the path-


loss exponent b has on the AFEP. The trend is different for
10
−2
10
−2
noise- and interference-limited cellular networks. The figure
also confirms that the BS density λ affects the AFEP only
in the noise-limited regime. Figure 3(a) confirms that having
10
−3
10
−3
multiple antennas at the receiver leads to an improvement of
40 80 120 160 200 240 280 40 80 120 160 200 240 280
the AFEP. However, no receive diversity gain is obtained in the
E/N0 [dB] E/N0 [dB] interference-limited regime. Figure 3(b) shows that increasing
Fig. 2. AFEP of a Nt × Nr = 2 × 2 system against the reference SNR.
Nt provides a better AFEP but the performance difference as
Setup: (a) 16QAM, λ = 10−5 , p = 10−3 . (b) 16QAM, b = 3, p = 10−3 . a function of Nt is smaller compared to Nr . In this figure, the
rate of the MIMO system is defined as Rate = Nt log2 (M )
(a) (b)
bits per channel use (bpcu). Figure 4(a) shows the AFEP
Nr=1 Nt=1
10
0
Nr=2
0
10 Nt=2
by assuming that the density of BSs λ is kept the same
Nr=4 Nt=4 but the number of BSs antennas Nt is different and Fig.
4(b) shows the AFEP by assuming that the density of BSs
−1 −1
10 10 antennas λ0 = Nt λ is kept the same and the BSs density
λ depends on Nt . A symmetric MIMO setup is assumed,
AFEP

AFEP

10
−2 −2
10
i.e., Nt = Nr . By comparing Fig. 4(a) and Fig. 4(b), we
conclude that it is possible to reduce the density of BSs by
increasing Nt without a significant performance degradation in
−3 −3
10 10 the interference-limited regime and with a small performance
degradation in the noise-limited regime.
−4 −4
10 10
V. C ONCLUSION
100 150 200 120 140 160 180 200 220 240
E/N0 [dB] E/N0 [dB] In this paper, a new mathematical framework to the compu-
tation of the average error probability of downlink MIMO cel-
Fig. 3. (a) AFEP of a Nt × Nr = 2 × Nr system against the reference
SNR. Setup: 4QAM, λ = 10−5 /2, b = 3, p = 10−3 . (b) AFEP of a
lular networks has been proposed and has been validated with
Nt × Nr = Nt × 2 system with Rate = 4 bpcu against the reference SNR. the aid of Monte Carlo simulations. The analysis has unveiled
Setup: b = 3, λ = 10−5 , b = 3, p = 10−3 . important performance trade-offs that may emerge depending
on the SNR operating regime, the path-loss exponent and the
model, on the other hand, provides mathematical tractability number of antennas available at BSs and MTs. An extended
and insight as a function of several system parameters. version of the present paper is available in [24], to which the
In Figs. 2–4, we observe a good accuracy between Monte reader is referred for further information, results and insights.
ACKNOWLEDGMENT R EFERENCES
This work is supported in part by the European Commission [1] F. Baccelli, M. Klein, M. Lebourges, and S. Zuyev, “Stochastic geometry
and architecture of communication networks”, J. Telecommun. Syst., vol.
under the auspices of the FP7-PEOPLE MITN-CROSSFIRE 7, no. 1, pp. 209–227, 1997.
project (grant 317126) and by the Royal Academy of Engi- [2] F. Baccelli and B. Blaszczyszyn, Stochastic Geometry and Wireless
neering (UK) under the auspices of the 2014-2015 Distin- Networks, Part I: Theory, Now Publishers, Sep. 2009.
[3] A. Guo and M. Haenggi, “Spatial stochastic models and metrics for the
guished Visiting Fellowship program. structure of base stations in cellular networks”, IEEE Trans. Wireless
Commun., vol. 12, no. 11, pp. 5800–5821, Nov. 2013.
A PPENDIX I – P ROOF OF (3) [4] J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to
coverage and rate in cellular networks”, IEEE Trans. Commun., vol. 59,
By inserting iagg (·) of (1) √ in I(·)
 of (2), the latter no. 11, pp. 3122–3134, Nov. 2011.
can be formulated as I(r0 ) = E i∈Ψ(\0) ri−b z̃i , where [5] H. S. Dhillon, R. K. Ganti, F. Baccelli, and J. G. Andrews, “Modeling
 Nr Nt Nt  ∗ and analysis of K-tier downlink heterogeneous cellular networks”, IEEE
(r,t̃) (t̃) (r,t) (t)
z̃i = 1/Nt r=1 t̃=1 t=1 H i s i H 0 Δ 0 . J. Sel. Areas Commun., vol. 30, no. 3, pp. 550–560, Apr. 2012.
[6] H.-S. Jo, Y. J. Sang, P. Xia, and J. G. Andrews, “Heterogeneous cellular
The proof follows the same line of thought of [11, Sec. III]. networks with flexible cell association: A comprehensive downlink SINR
The main difference is that Zi in Iagg (·) of [11, Eq. (5)] analysis”, IEEE Trans. Wireless Commun., vol. 11, no. 10, pp. 3484–
needs to be replaced by z̃i in I(·). In addition, and similar 3495, Oct. 2012.
[7] H. S. Dhillon, R. K. Ganti, and J. G. Andrews, “Load-aware modeling
to Zi , the RVs z̃i are still circularly symmetric and i.i.d. for and analysis of heterogeneous cellular networks”, IEEE Trans. Wireless
i ∈ Ψ(\0) . Thus, the CF of I(·) can be formulated as shown in Commun., vol. 12, no. 4, pp. 1666–1677, Apr. 2013.
(re)
[11, Eq. (7)–(9)], by replacing Re {Zi } with z̃i = Re {z̃i }. [8] S. Singh, H. S. Dhillon, and J. G. Andrews, “Offloading in heteroge-
(re) neous networks: Modeling, analysis, and design insights”, IEEE Trans.
Accordingly, the moments of z̃i conditioned upon H0 Wireless Commun., vol. 12, no. 5, pp. 2484–2497, May 2013.
and Δ0 need to be computed. Let us denotethese condi-
2q 
[9] M. Di Renzo, A. Guidotti, and G. E. Corazza, “Average rate of downlink
 heterogeneous cellular networks over generalized fading channels – A
(re)
tioned moments by ηq (H0 , Δ0 ) = E z̃i |H0 ,Δ0 z̃i . stochastic geometry approach”, IEEE Trans. Commun., vol. 61, no. 7,
  pp. 3050–3071, July 2013.
(r,t) (r,t)
Since Hi ∼ CN (0, Ω), Re Hi ∼ CN (0, Ω/2). [10] H. S. Dhillon, M. Kountouris, and J. G. Andrews, “Downlink MIMO
hetnets: Modeling, ordering results and performance analysis”, IEEE
(re)
Thus, z̃i turns out to be a Gaussian RV by con- Trans. Wireless Commun., vol. 12, no. 10, pp. 5208–5222, Oct. 2013.
[11] P. Guan and M. Di Renzo, “Stochastic geometry analysis of the
ditioning upon H0 , Δ0 and si , which we denote by
 2  average error probability of downlink cellular networks”, IEEE Int. Conf.
(re)   
Nt  (t̃)  Comput. Netw. Commun., pp. 1–5, Feb. 2014.
z̃i ∼ N|H0 , Δ0 ,si 0, (Ω/2) U (1/Nt ) t̃=1 si  . The [12] M. Di Renzo and W. Lu, “The equivalent-in-distribution (EiD)-based ap-
proach: On the analysis of cellular networks using stochastic geometry”,
proof follows by computing the moments of Gaussian RVs IEEE Commun. Lett., vol. 18, no. 5, pp. 761–764, May 2014.
with the aid of [22, Eq. (13)]: [13] M. Di Renzo and P. Guan, “Stochastic geometry modeling of coverage
and rate of cellular networks using the Gil-Pelaez inversion theorem”,
ηq (H0 , Δ0 ) = ηq (U) IEEE Commun. Lett., vol. 18, no. 9, pp. 1575–1578, Sep. 2014.
⎧ ⎛ ⎞ ⎫
⎨ √ Nt   q ⎬ [14] M. Di Renzo and W. Lu, “Stochastic geometry modeling and perfor-
π 1  (t̃) 2
= Esi (−1) q ⎝ s  ⎠ Ω U q q mance evaluation of MIMO cellular networks using the equivalent-in-
⎩ Γ (1/2 − q) Nt  i  ⎭ distribution (EiD)-based approach”, IEEE Trans. Commun., vol. 63, no.
t̃=1
(11) 3, pp. 977–996, Mar. 2015.
[15] M. Di Renzo, “Stochastic geometry modeling and analysis of multi-tier
as well as of the series representation of the generalized millimeter wave cellular networks”, IEEE Trans. Wireless Commun., to
appear. [Online]. Available: IEEE Early Access.
hypergeometric function
 [19, Ch. 5, Eq. 2], and of the iden- [16] J. Gil-Pelaez, “Note on the inversion theorem” Biometrika, vol. 38, pp.
Nt  (t̃) 2 2 481–482, Dec. 1951.
tities s
t̃=1  i 
= si  and 2 F2 (1/2, α; 1/2, β; −x) = [17] M. Di Renzo, C. Merola, A. Guidotti, F. Santucci, and G. E. Corazza,
“Error performance of multi-antenna receivers in a Poisson field of
1 F1 (α; β; −x). interferers: A stochastic geometry approach”, IEEE Trans. Commun.,
vol. 61, no. 5, pp. 2025–2047, May 2013.
A PPENDIX II – P ROOF OF (6) [18] A. Goldsmith, Wireless Communications, Cambridge Univ. Press, 2005.
[19] A. Erdelyi, W. Magnus, F. Oberhettinger, and F. G. Tricomi, Higher
The proof follows by applying the Gil-Pelaez inversion Transcendental Functions – Vol I, McGraw-Hill, Inc., 1953.
theorem [16] to (2). In particular, the APEP(F) conditioned [20] M. K. Simon and M.-S. Alouini, Digital Communication over Fading
upon H0 and r0 can be formulated as follows: Channels, John Wiley & Sons, Inc., 2st ed., 2005.
[21] I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series, and
APEP(F) (Δ0 ; H0 , r0 ) = APEP(F) (U, r0 ) = Products, San Diego, CA: Academic, 7th ed., 2007.
' +∞   [22] A. Winkelbauer, “Moments and absolute moments of the normal distri-
1 r0−b E d |ω| bution”. [Online]. Available: http://arxiv.org/pdf/1209.4340.pdf.
− sin U |ω| ΦN(re) (|ω| ; U) ΦI(re) (|ω| ; r0 , U) [23] E. Biglieri, G. Caire, G. Taricco, and J. Ventura-Traveset, “Computing
2 0 2 N t |ω|
error probabilities over fading channels: A unified approach”, European
where: i) N(re) = Re {N} ∼ N (0, (N0 /2) U) follows Trans. Telecommun., vol. 9, no. 1, pp. 15–25, Jan./Feb. 1998.
[24] M. Di Renzo and P. Guan, “A mathematical framework to the compu-
from the AWGN assumption and from (2); ii) I(re) = tation of the error probability of downlink MIMO cellular networks by
Re {I} is defined in (2); iii) ΦN(re) (|ω| ; U) = ΦN (|ω| ; U) using stochastic geometry”, IEEE Trans. Commun., vol. 62, no. 8, pp.
and ΦI(re) (|ω| ; r0 , U) = ΦI (|ω| ; r0 , U), since both N 2860–2879, July 2014.
and I are circularly symmetric
 RVs; iv) ΦN (|ω| ; U) =
2
exp − (1/4) |ω| N0 U , since it is a Gaussian RV [23]; and
v) ΦI (·; ·, ·) is given in (3). The rest of the proof follows the
same line of thought of [11, Sec. IV].

Das könnte Ihnen auch gefallen