Professional Documents
Culture Documents
Towards Understanding The Fundamentals of Mobility in Cellular Networks
Towards Understanding The Fundamentals of Mobility in Cellular Networks
4, APRIL 2013
Abstract—Despite the central role of mobility in wireless to understand tradeoffs between optimum cell associations
networks, analytical study on its impact on network performance and the undesired overhead in session setup and tear downs.
is notoriously difficult. This paper aims to address this gap by Sojourn time is defined as the time that a mobile resides
proposing a random waypoint (RWP) mobility model defined
on the entire plane and applying it to analyze two key cellular in a typical cell. It represents the time that a BS would
network parameters: handover rate and sojourn time. We first provide service to the mobile user. In the case of a short
analyze the stochastic properties of the proposed model and sojourn time, it may be preferable from a system-level view
compare it to two other models: the classical RWP mobility model to temporarily tolerate a suboptimal BS association versus
and a synthetic truncated Levy walk model which is constructed initiating a handover into and out of this cell. This concept
from real mobility trajectories. The comparison shows that
the proposed RWP mobility model is more appropriate for is supported in 4G standards with a threshold rule [3], but it
the mobility simulation in emerging cellular networks, which is fair to say that the theory behind such tradeoffs is not well
have ever-smaller cells. Then we apply the proposed model developed. Though handover rate is inversely proportional to
to cellular networks under both deterministic (hexagonal) and expected sojourn time, their distributions can be significantly
random (Poisson) base station (BS) models. We present analytic different, which motivates us to study them separately in this
expressions for both handover rate and sojourn time, which have
the expected property that the handover rate is proportional paper.
to the square root of BS density. Compared to an actual BS To explore the role of mobility in cellular networks, par-
distribution, we find that the Poisson-Voronoi model is about as ticularly handover rate and sojourn time, mobility modeling
accurate in terms of mobility evaluation as hexagonal model,
though being more pessimistic in that it predicts a higher is obviously a necessary first step. In this paper, we focus
handover rate and lower sojourn time. on RWP mobility model originally proposed in [4] due to its
simplicity in modeling movement patterns of mobile nodes.
Index Terms—Mobility models, handover, sojourn time, cellu-
lar networks, Poisson-Voronoi tessellation. In this model, mobile users move in a finite domain A. At
each turning point, each user selects the destination point
(referred to as waypoint) uniformly distributed in A and
I. I NTRODUCTION chooses the velocity from a uniform distribution. Then the
of this work is not to solve this open problem. Instead, we in mobile networks. In contrast, random synthetic models,
aim to propose a tractable RWP mobility model, which can including random walk, the classical and the proposed RWP,
be more appropriate than the classical RWP mobility model Gauss-Markov, etc., are generic and more mathematically
for the study of mobility in emerging cellular networks, so tractable. In addition, some aspects of some random syn-
that it can be utilized to analyze the impact of mobility in thetic models (including the proposed RWP model) can be
cellular networks to provide insights on network design. fitted using traces, as done in [19]. Nevertheless, trace-based
After presenting the proposed mobility model, we analyze mobility models are more realistic and scenario-dependent,
its associated stochastic properties and compare it to two other which is crucial to perform reliable performance evaluation
models: the classical RWP mobility model and a synthetic of mobile networks. Hence, both random and trace-based syn-
truncated Levy walk model [6] which is constructed from real thetic models are important for the study of mobile networks.
mobility trajectories. The comparison shows that the proposed In particular, when evaluating a specific protocol in mobile
model is more appropriate for the mobility simulation in networks, researchers can utilize random synthetic models for
emerging cellular networks. Then the proposed RWP mobility example the proposed RWP model to quickly get informative
model is applied to cellular networks whose BSs are modeled results. Meanwhile, researchers can build specific scenario-
in two conceptually opposite ways. The first is the traditional dependent trace-based model by collecting traces if possible.
hexagonal grid, which represents an extreme in terms of Then based on the constructed trace-based model researchers
regularity and uniformity of coverage, and is completely can further perform more extensive and reliable evaluation of
deterministic. The second is to model the BS locations as the protocol beyond just using random synthetic models.
drawn from a Poisson point process (PPP), which creates a set
of BSs with completely independent locations [9], [10]. Under B. Related work
the PPP BS model, the cellular network can be viewed as a The proposed RWP model is based on the one originally
Poisson-Voronoi tessellation if the mobile users are assumed proposed in [4]. Due to its simplicity in modeling movement
to connect to the nearest BSs. As one would expect, most patterns of mobile nodes, the classical RWP mobility model
actual deployments of cellular networks lie between these has been extensively studied in literature [5], [25]–[27]. These
two models, both qualitatively – i.e. they are neither perfectly studies analyzed the various stochastic mobility parameters,
regular nor perfectly random – and quantitatively, i.e. the SINR including transition length, transition time, direction switch
statistics and other statistical measures are bounded by these rate, and spatial node distribution. When it comes to applying
two approaches [11]. Thus, both models are of interest and we the mobility model to hexagonal cellular networks, simulations
apply the proposed RWP mobility model to cellular networks are often required to study the impact of mobility since the
under both models. Analytic expressions for handover rate and analysis is hard to proceed [2], [3]. Nonetheless, the effects of
sojourn time are obtained under both models, some of which the classical RWP mobility model to cellular networks have
are quite simple and lead to intuitive interpretations. been briefly analyzed in [26] and a more detailed study can
In some aspects the proposed RWP mobility model is simi- be found in [28]. However, as remarked above, the classical
lar to the random direction (RD) mobility model. Though the RWP model may not be convenient in some cases. In contrast,
classical RD mobility model does not have a non-homogenous we analyze and obtain insight about the impact of mobility
spatial distribution as in the classical RWP mobility model, under a hexagonal model through applying the relatively clean
its transition lengths are still of the same order as the size characterization of the proposed RWP model, as [28] did
of the simulation area, which seems to deviate significantly through applying the classical RWP model.
from those in human walks [12]. This inconvenience may be The application of the proposed RWP mobility model to
solved by a modified RD mobility model [13], in which the cellular networks modeled as Poisson-Voronoi tessellation re-
transition length distribution is co-determined by the velocity quires stochastic geometric tools, which are becoming increas-
distribution and moving time distribution. This may compli- ingly sophisticated and popular [29]–[32]. As far as mobility
cate the theoretical analysis in this paper. This motivates us to is concerned, [33] proposed a framework to study the impact
propose a mobility model with transition length distribution of mobility in cellular networks modeled as Poisson-Voronoi
directly specified. Besides, transition length data seems to be tessellation. In particular, the authors proposed a Poisson line
more readily available than moving time data in many real process to model the road system, along which the mobile
data sets [6], [8]. Thus, it might be easier to use the proposed users move. Thus, the mobility pattern in [33] is of large scale
RWP mobility model when fitting the mobility model to real while the RWP mobility model in this paper is of small scale.
mobility trajectories. Our study can be viewed as complementary to [33]. For exam-
Before ending this subsection, it is worth mentioning ple, our result indicates that if cells decrease in size such that
the many trace-based mobility models [14]–[24]. One main the BS density per unit area is increased by 4 times, then the
drawback of trace-based mobility models is that, due to the handover rate would be doubled. Note that recent work showed
differences in the trace acquiring methods, sizes of trace that the PPP model for BSs was about as accurate in terms of
data, and data filtration techniques, mobility model built on SINR distribution as the hexagonal grid for a representative
one trace data set may not be applicable to other network urban cellular network [11]. Interestingly, we find that this
scenarios. Moreover, trace-based mobility models are often not observation is also true for mobility evaluation, though the
mathematically tractable (at least very complicated), prevent- Poisson-Voronoi model yields slightly higher handover rate
ing researchers from analytically studying the performance and lower sojourn time and thus is a bit more pessimistic than
of various protocols and/or getting quick informative results the hexagonal model (which is correspondingly optimistic).
1688 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013
40 −50
II. P ROPOSED RWP M OBILITY M ODEL ON I NFINITE −100
P LANES 30
−150
We describe the proposed RWP mobility model in this 20
−200
section. As in the description of the classical RWP model 10 −250
(see, e.g., [5]), the movement trace of a node can be
0 −300
formally described by an infinite sequence of quadruples: 0 20 40 60 −200 −150 −100 −50 0
0.6
CDF
where f (r, θ) is the spatial node distribution without pause
time and is given in Theorem 1 , and fX 1 (r, θ) denotes the 0.4
Classical RWP
spatial distribution of the waypoint X 1 and is given in Lemma 0.2 Levy walk
1. Plugging f (r, θ) and fX 1 (r, θ) into (14), the expression (12) Proposed RWP
0
follows. 0 500 1000 1500
Transition length (m)
α=1
E. Model comparison
1
In this section, we perform simulations to study the dif-
0.8
ference between the proposed RWP mobility model and the
classical one. For validation, we also compare them to a 0.6
CDF
synthetic truncated Levy walk model [6], which is constructed 0.4
Classical RWP
from real mobility trajectories. Thus, we indirectly compare
0.2 Levy walk
the two RWP mobility models to real human walks. In Proposed RWP
the truncated Levy walk model, transition lengths have an 0
1 0 500 1000 1500
inverse power-law distribution: PL (l) ∼ l1+α , 0 < α < 2. Transition length (m)
The pause times also have an inverse power-law distribution: α = 1.9
1
PS (s) ∼ s1+β , 0 < β < 2. 1
We simulate the movement of a mobile node under the
0.8
three mobility models, respectively. The simulation area is a
1000×1000 grid.1 Since the main assumption in the proposed 0.6
CDF
β = 0.1 Note that E[Tp ] = E[T ] + E[S] where E[T ] has been
1 given in Prop. 1 and E[S] can be determined from the pause
0.8
time distribution. However, exact computation of N by (15)
is tedious. Thus, we propose the following approximation
0.6
CDF
formula
0.4 ∞ 2π (2n+1)R
Classical RWP
0.2 Levy walk E[N ]app = n fX 1 (r, θ)rdrdθ, (16)
Proposed RWP n=1 0 (2n−1)R
0
0 0.02 0.04 0.06 0.08 where R = C/π with C being the hexagonal cell size. In
Direction switch rate (s−1) other words, we approximate the n-th neighbouring layer by a
β=1 ring with inner radius (2n − 1)R and outer radius (2n + 1)R.
1 This approximation captures the essence of (15) and allows us
0.8 to derive closed form results on E[N ] and the corresponding
lower and upper bounds.
0.6
CDF
0.4
Proposition 3. Let d be the side length of the hexagonal
Classical RWP cell and λ the mobility parameter. The approximation of the
0.2 Levy walk
Proposed RWP expected number of handovers during one movement period is
0 given by
0 0.02 0.04 0.06 0.08
∞
√
Direction switch rate (s−1) 3 3 2 2
β = 1.9 E[N ]app = exp − (2n + 1) λd , (17)
n=0
2
1
E[N ]app
L
√ Q 3 3λd , 2 (18)
6 3λd2
0.4
Classical RWP
Levy walk π √
0.2 E[N ]Uapp √ 1−Q 3 3λd2 . (19)
Proposed RWP 6 3λd2
0
0 0.02 0.04 0.06 0.08
Moreover, the difference Napp (λd2 ) between the upper bound
Direction switch rate (s−1)
and lower bound is a strictly increasing function of λd2
Fig. 3. Simulated statistics of direction switch rate: V ≡ 1 and α = 1. and is within the range (0, 1). In particular, Napp (λd2 ) →
0 as λd2 → ∞, and Napp (λd2 ) → 1 as λd2 → 0.
Proof: See Appendix B.
IV. A PPLICATIONS TO H EXAGONAL M ODELED C ELLULAR Using Prop. 3, the approximate handover rate can be
N ETWORKS computed as Happ = E[T 1
· E[N ]app and is bounded as
p]
1
In this and the next section, we study the impact of mo- Happ ≤ Happ ≤ Happ , where Happ
L U L
= E[T p]
· E[N ]L U
app , Happ =
1
E[Tp ] · E[N ]app . As the size of the cells in cellular networks
bility on important cellular network parameters, particularly U
handover rate and sojourn time, using the proposed RWP becomes smaller, higher handover rates are expected. In this
mobility model. We focus on hexagonal cellular networks in regard, it is interesting to examine the asymptotic property of
this section. handover rate as in Corollary 2.
Corollary 2. Assume that any of the following asymptotic
A. Handover Rate conditions holds: 1) d → 0 with fixed λ; 2) λ → 0 with fixed d;
We assume the typical mobile user is located at the origin. 3) λd2 → 0. Then the asymptotic approximate handover rate is
Then the expected number of handovers during one movement given by
1 π 1
period can be computed as Happ ∼ √ . (20)
E[T ] + E[S] 6 3λ 2d
∞
E[N ] = n P (dA(r, θ)), (15)
n=1 Cn Though derivation by (15) is not tractable, the exact han-
dover rate in hexagonal model can be obtained using a gen-
where P (dA(r, θ)) is the probability distribution of the way-
eralized solution of Buffon’s needle [34] as in the following
point density fX 1 (r, θ) given in Lemma 1 and Cn denotes
Proposition.
the area covered by the n-th layer neighbouring cells. Now
we formally define the handover rate. Proposition 4. Let d be the side length of the hexagonal
Definition 1. The handover rate is defined as the expected num- cell and λ the mobility parameter. The expected number of
ber E[N ] of handovers during one movement period divided handovers E[N ] during one movement period is given by
√
by the expected period time. Mathematically, handover rate is 4 3 E[V ]
given by H = E[N ]/E[Tp ]. E[N ] = E[T ] · . (21)
3π d
1692 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013
9
Simulation (d = 5)
we only consider the simplified RWP mobility model where
8 Simulation (d = 10) the mobile nodes do not have pause time and constant velocity,
Simulation (d = 20)
Exact Analysis (d = 5)
i.e., V ≡ ν. Then sojourn time in any cell with coverage area
7
Exact Analysis (d = 10) C can be computed as
Exact Analysis (d = 20)
Number of Handovers
6 Approximate Analysis (d = 5)
Approximate Analysis (d = 10) ST = E[T ] · P (dA(r, θ)), (25)
Approximate Analysis (d = 20) C
5
where P (dA(r, θ)) is the probability distribution of the spatial
4
node density f (r, θ) given in Theorem 1. For brevity, we only
3 focus on the sojourn time in the cell where the connection is
initiated during one movement period in the sequel.
2
Assuming the mobile node is co-located with its currently
1 associated BS at the origin, the sojourn time can be computed
as
0
1 2 3 4 5 6 7 8 9 10 √ d/2 √3d/2
Mobility Parameter (λ) 4 λ
ST = E[T ] · ( f (x, y)dydx
π 0 0
Fig. 4. Handover in hexagonal model by analysis and simulation with d − 3x+√3d
√
velocity ν ≡ 1 and no pause time.
+ f (x, y)dydx), (26)
d/2 0
The handover rate H = E[N ]/E[Tp ] is then given by where f (x, y) = exp(−λπ(x2 + y 2 ))/ x2 + y 2 . However,
√ no closed form result is available for (26), though it can
E[T ] 4 3 E[V ] be evaluated using numerical methods. We provide explicit
H= · . (22)
E[T ] + E[S] 3π d formulas for the lower and upper bounds of (26) in Prop. 5,
whose proof is omitted for brevity.
Proof: See Appendix C. Proposition 5. The sojourn time ST can be bounded as STL ≤
We remark that the insights obtained by either approximate ST ≤ STU where
or exact approach are the same. Let us consider the simplified
π ν
Happ ∼ √ , (23)
6 3d
Assuming that E[T ] is finite2 , ST tends to 0 since both the
and the exact handover rate H is given by
√ lower bound STL and upper bound STU tend to 0 as d → 0.
4 3ν This is an intuitive result. So cellular networks with small cells
H= , (24)
3π d need to be equipped with fast algorithms. Otherwise, other
which is consistent with the one given √ in [35]. Note that the handover strategies may be needed if the sojourn time in a
hexagonal cell size sH is given by 3 3d 2 cell is shorter than the time needed to complete the handover
π/2 νin hexagonal procedure. The analysis of the impact of mobility parameter
tiling. So (23) can be written as Happ ∼ 4 √sH . Similarly,
√ λ is more involved. To obtain insight, consider the constant
the exact method yields H = π4 2
3 √ν
sH . Both results imply velocity case which yields the following result.
that handover rate is inversely proportional to the square
root of cell size sH . In other words, if we deploy more Proposition 6. Assume that V √≡ ν . Then as λ → 0, sojourn
small cells and increase the BS density say by 4 times in time ST ∼ αd/ν , where α ∈ ( 23 , 1) is a constant.
current cellular network, we would expect the handover rate Proof: See Appendix D.
to be roughly increased by 2 times. Interestingly, the mobility
From Prop. 6, we observe the interesting fact that sojourn
parameter λ does not play a role, while the velocity and the
time converges to αd/ν as the mobility parameter λ goes
cell size affect the handover rate in a trade-off manner. Fig. 4
to zero. Then what matters is the velocity: Sojourn time is
compares the number of handovers obtained by simulation to
inversely proportional to the velocity. Also, we can express ST
2 1 √
the counterparts evaluated by analytic formula (17) and (21)
respectively. It is shown that the exact analytic result closely as ST ∼ 3√ 3ν
· sH where recall that sH is the hexagonal
matches the simulation while the approximation approach cell size. This shows that sojourn time is proportional to the
tends to underestimate the real number of handovers. square root of cell size, which is contrary to the asymptotic
result for handover rate.
B. Sojourn Time
Sojourn time is defined as the expected duration that the 2 This is true for both the case with constant velocity and the case with
mobile node stays within a particular serving cell. For brevity, uniformly distributed velocity on [vmin , vmax ], as shown in Section III-B.
LIN et al.: TOWARDS UNDERSTANDING THE FUNDAMENTALS OF MOBILITY IN CELLULAR NETWORKS 1693
25
V. A PPLICATIONS TO P OISSON -VORONOI M ODELED
Simulation (λ = 10−2)
C ELLULAR N ETWORKS Simulation (λ = 10−3)
In this subsection, we apply the proposed RWP mobility 20
Simulation (λ = 10−4)
−2
model to analyze handover rate and sojourn time in cellular Analytic Result (λ = 10 )
Number of Handovers
−3
Analytic Result (λ = 10 )
networks modeled by Poisson-Voronoi tessellation [32], [36]. −4
Analytic Result (λ = 10 )
We first give a brief description on Poisson-Voronoi tessella- 15
0.45
important. For Poisson-Voronoi tessellation modeled cellular
Poisson−Voronoi handover rate
Hexagonal exact handover rate network, the random closed set Z of interest is the union of
0.4 Hexagonal approximate handover rate all cell boundaries ∪xi ∈Φ ∂Cxi (Φ), where ∂Cxi (Φ) denotes the
boundary of the cell Cxi (Φ).
0.35
With a slight abuse of notation, let Co (Φ) denote the Voronoi
cell containing the origin o. We can further simplify Hl (r) as
Handover Rate
0.3
follows:
0.25 P (Z ∩ rl = ∅)
Hl (r) = 1 − = 1 − P (rl ⊆ Co (Φ)). (33)
1 − P (o ∈ Z)
0.2
The last equality follows because 1) the origin o is contained
0.15
in the interior of Co (Φ) almost surely and thus P (o ∈ Z) = 0
where Z = ∪xi ∈Φ ∂Cxi (Φ), and 2) the event {Z ∩ rl = ∅} is
0.1 equivalent to that {rl ⊆ Co (Φ)}. Now we are in a position to
characterize the sojourn time ST .
0.05
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 Proposition 8. Let μ be the intensity of the homogeneously
Intensity of Base Station (μ) PPP distributed BSs and λ the mobility parameter. The pdf of
sojourn time ST is given by
Fig. 6. Comparison of handover rate in Poisson-Voronoi model, exact
and approximate handover rate in hexagonal model with velocity ν ≡ 1, 1 1 (−1) 1 t 2
fST (t) = √ exp (Q ( (1 − ))) hl (x) ,
normalized BS’s intensity μ = 1, mobility parameter λ = 1 and √ no pause 2 λE[T ] 2 2 E[T ]
time. They all imply that the handover rate grows linearly with μ.
(34)
where E[T ] is the expected transition time,
1
4.5
Analytic result by hexagonal model (approximate)
x = √2πλ Q(−1) ( 12 (1 − E[Tt
] )) and h l (r) is given by
Analytic result by hexagonal model (exact) π π−α
sin2 α sin β
4
Analytic result by Poisson−Voronoi model
Simulation based on data from real network hl (r) = 4πμ2 r3 4 b0 (β)
0 0 sin (α + β)
Number of Handovers
3.5
1.8
In this paper, we study the critical mobility issue in cellular
networks. To this end, we first propose a tractable RWP
1.6
mobility model defined on the entire plane. The various
1.4 properties of the mobility model are carefully studied and
μ=1 μ = 0.01
simple analytical expressions are obtained. Then we utilize
The pdf of Sojourn Time
1.2
this tractable mobility model to analyze the handover rate
1 and sojourn time in cellular networks. The analysis is carried
out for cellular networks under both hexagonal and Poisson-
0.8
Voronoi models. We derive closed form expressions and/or
0.6
μ = 0.1 bounds for the performance metrics in question. These an-
0.4
μ = 0.025 alytical results are instrumental for mobility management in
cellular networks.
0.2
Note that the proposed RWP mobility model represents the
0 real movement of mobile nodes in a simplified manner. Thus,
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Sojourn Time it does not capture some other mobility characteristics such
as temporal and spatial dependency of the mobility pattern. It
Fig. 8. Plot of pdf of sojourn time in Poisson-Voronoi model with mobility
is desirable to extend the current model further to incorporate
parameter λ = 0.01 and velocity ν ≡ 1. these extra mobility characteristics while maintaining a certain
degree of tractability. It would be rather interesting (and
challenging) to characterize the intuitive tradeoff between the
complexity and tractability of the mobility models. Besides,
6
Hexagonal Upper Bound (λ=0.005) it is also interesting to evaluate the performance of the many
Hexagonal Upper Bound (λ=0.05)
5.5
Hexagonal Lower Bound (λ=0.005) wireless protocols by applying the proposed tractable model
5
Hexagonal Lower Bound (λ=0.05)
Poisson−Voronoi (λ=0.005)
in theory and/or simulation.
Poisson−Voronoi (λ=0.05)
4.5
4 ACKNOWLEDGMENT
Expected Sojourn Time
2
A PPENDIX
1.5
1
A. Proof of Theorem 1
0.5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
We first derive the pdf of the random waypoint X 1 =
Intensity of Base Station (μ) (R1 , Θ1 ) as follows:
P (R1 ≤ r + Δr) − P (R1 ≤ r)
Fig. 9. Comparison of sojourn time in hexagonal tiling and Poisson-Voronoi
fX 1 (r, θ) = lim
2π
r+Δr
Δr→0 xdxdφ
tessellation with velocity ν ≡ 1. Hexagonal model yields larger sojourn time 0 r
than that of Poisson-Voronoi model. Besides, the upper and lower bounds for exp(−λπr ) − exp(−λπ(r + Δr)2 )
2
hexagonal model are tight. = lim
Δr→0 2πrΔr
2πλ(r + Δr) exp(−λπ(r + Δr)2 )
= lim
Δr→0 2πr
integral, which is tractable for efficient numerical evaluation. = λ exp(−λi πr2 ).
We plot the pdf of sojourn time in Fig. 8 for illustration. As
This result is summarized in Lemma 1 for ease of reference.
expected, the smaller μ is, the more likely the mobile node
stays longer within the cell. This is intuitive since a smaller Lemma 1. Given X 0 is at the origin, the pdf of the random
μ implies larger cell sizes on average. As a result, it is less waypoint X 1 = (R1 , Θ1 ) is given by
likely that the mobile node would move out of the cell.
fX 1 (r, θ) = λ exp(−λπr2 ). (36)
We also compare the analytic result about sojourn time in
the Poisson-Voronoi model to its deterministic counterpart in
the hexagonal model in Fig. 9. It is shown that the analytic We next derive the pdf of the spatial node distribution. The
result in the Poisson-Voronoi model is more conservative and main technique of the following proof is inspired by [26],
yields smaller mean sojourn times than its counterpart in the which is also adopted in [25]. Consider a small set dA located
hexagonal model. Besides, we can see that the upper and lower 1 denote the vector X 1 − X 0 and |L
at (r, θ). Let L 1 | = L1 .
bounds of the sojourn time in hexagonal model are pretty tight. Then the pdf of the spatial node distribution can be interpreted
1696 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013
exp(−λπr2 ) λ exp(−λπr2 ) π1 t2
f (r, θ) = = , (40) lim (1 − 2Q(t)) = lim exp(− ) = 1,
t→0 2t t→0 2
2πrE[L1 ] πr
√ π1
where we use the result that E[L1 ] = 1/2 λ in the last lim (1 − 2Q(t)) = 0.
t→∞ 2t
equality. This completes the proof.
The desired results follow by further observing t is a strictly
increasing continuous function of λd2 .
B. Proof of Proposition 3
From Lemma 1, fX 1 (r, θ) = λ exp(−λπr2 ). We compute C. Proof of Proposition 4
E[N ]app as follows. We use (generalized) argument for Buffon’s needle problem
∞ 2π (2n+1)R in this proof. Consider the typical node located in area A of
E[N ]app = n λ exp(−λπr2 )rdrdθ size |A| and cell boundaries BA of length |BA |. Then the prob-
n=1 0 (2n−1)R ability that this node crosses the small boundary Δb within a
∞
1
small time interval Δt is Δp = |A| ·Δb·νΔtE[| sin Θ|], where
= n exp(−π(2n − 1)2 R2 λ) − exp(−π(2n + 1)2 R2 λ) Θ is uniformly distributed on [0, 2π] (following from our
n=1
∞ mobility model) and thus E[| sin Θ|] = π2 . So the probability
= exp(−π(2n + 1)2 R2 λ) that this node crosses BA is p(A) = Δp · |B| 2 |BA |
Δb = π |A| · νΔt.
n=0 Thus, conditioned on moving, the handover rate
√
C 4
27d p(A) 2 |BA |
Substituting R = = √ , we obtain H = lim = ν · lim
π Δt
|A|→∞ π |A|→∞ |A|
2π √
∞
√ 2 9d 4 3ν
3 3 = ν· √ = .
E[N ]app = exp − (2n + 1)2 d2 λ . π 9 3d2 /2 3π d
n=0
2 √
4 3 E[V ]
L
Then the lower bound Napprox is derived as follows. Correspondingly, E[N ] = E[H ] · E[T ] =
· E[T ]
√ 3π d
∞ √ E[T ] 4 3 E[V ]
3 3 and H = .
E[N ]app ≥ exp − (2x + 1)2 d2 λ dx E[T ] + E[S] 3π d
0 2
π √ D. Proof of Proposition 6
= √ Q 3 3λd2 E[N ]L app . Applying the general expression (26) for ST yields
6 3λd2
4
The upper bound E[N ]U app can be derived
in√a similar manner. ST = f (ξx d, ξy d)·
πν
For the remaining proof, denote t = 3 3λd2 . Note that d/2 √3d/2 d √ √
− 3x+ 3d
1 2 μ
(3 − k)-facet. Now let Φk be the set of all configurations = . (47)
x(3−k) ⊆ Φ. That is, the intersection of the Voronoi cells E[T ] π λ
associated with any configuration x(3−k) ∈ Φk is a k-facet. where in (46)
∞ we use Lemma 1, and in (47) we apply the
For each configuration x(3−k) = {x1 , ..., x3−k } ∈ Φk , we result that 0 rα−1 exp(−γπrβ ) dr = Γ(α/β)
βγ α/β
for α, β, γ >
associate a point z x(3−k) (Φ) called “centroid” such that for all 0. Plugging E[T ] yields the desired results.
y ∈ R2 , z(x(3−k) + y|Φ + y) = y + z(x(3−k) |Φ). Note that
there are several degrees of freedom in choosing the centroids. R EFERENCES
We refer to [36] for more details.
Consider the intersection of the Poisson-Voronoi tessellation [1] T. Camp, J. Boleng, and V. Davies, “A survey of mobility models for
ad hoc network research,” Wireless Commun. Mobile Comput., vol. 2,
with a fixed line L. Without loss of generality, we assume no. 5, pp. 483–502, 2002.
that L contains the origin. Then the nonempty sectional cells [2] I. Akyildiz, J. Mcnair, J. Ho, H. Uzunalioglu, and W. Wang, “Mobility
C¯xi (Φ) = Cxi (Φ) ∩ L satisfy that (i) L = ∪xi ∈Φ C¯xi (Φ), management in next-generation wireless systems,” Proc. IEEE, vol. 87,
no. 8, pp. 1347–1384, Aug. 1999.
and (ii) ri(C¯xi (Φ)) ∩ ri(C¯xj (Φ)) = ∅, ∀i = j where ri(·) [3] “Evolved universal terrestrial radio access (E-UTRA): mobility enhance-
denotes the relative interior [36]. So the sectional cells C¯xi (Φ) ments in heterogeneous networks,” 3GPP TR 36.8de V0.0.1, Technical
constitute a tessellation of L, denoted by VL (Φ). Now consider Report, May 2011.
[4] D. Johnson and D. Maltz, “Dynamic source routing in ad hoc wireless
the intersection FL (x(2) |Φ) = F (x(2) |Φ) ∩ L which can be networks,” Mobile Comput., pp. 153–181, 1996.
either empty or a singleton. Clearly, FL (x(2) |Φ) is the 0- [5] C. Bettstetter, H. Hartenstein, and X. Pérez-Costa, “Stochastic properties
facet of tessellation VL (Φ) if FL (x(2) |Φ) is nonempty. Now of the random waypoint mobility model,” Wireless Netw., vol. 10, no. 5,
(2) pp. 555–567, Sept. 2004.
let ΦL0 = {x = {x1 , x2 } ∈ Φ1 : FL (x(2) |Φ) = ∅} which [6] I. Rhee, M. Shin, S. Hong, K. Lee, S. Kim, and S. Chong, “On the Levy-
parametrizes the 0-facets of VL (Φ). Then the intensity of the Walk nature of human mobility,” IEEE/ACM Trans. Netw., vol. 19, no. 3,
0-facets of VL (Φ) is well defined as pp. 630–643, 2011.
[7] M. Gonzalez, C. Hidalgo, and A. Barabási, “Understanding individual
E[ x(2) ∈ΦL χ{z L (x(2) |Φ) ∈ BL }] human mobility patterns,” Nature, vol. 453, no. 7196, pp. 779–782,
L
μ0 = 0
, (43) 2008.
|BL |1 [8] C. Song, T. Koren, P. Wang, and A. Barabási, “Modelling the scaling
properties of human mobility,” Nature Physics, vol. 6, no. 10, pp. 818–
where χ{·} is indicator function taking value 1 if the event 823, 2010.
in its argument is true and 0 otherwise, z L (x(2) |Φ) is the [9] P. J. Fleming, A. L. Stolyar, and B. Simon, “Closed-form expressions for
other-cell interference in cellular CDMA,” Technical Report 116, Univ.
centroid of the 0-facets of VL (Φ), BL ⊆ L is any arbitrary of Colorado at Boulder, Dec. 1997. Available: http://ccm.ucdenver.edu/
Borel set and |BL |1 is the volume of BL with respect to R. reports/
1698 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013
[10] T. X. Brown, “Cellular performance bounds via shotgun cellular sys- [38] L. Muche, “Contact and chord length distribution functions of the
tems,” IEEE J. Sel. Areas Commun., vol. 18, no. 11, pp. 2443–55, Nov. Poisson-Voronoi tessellation in high dimensions,” Adv. Appl. Probabil-
2000. ity, vol. 42, no. 1, pp. 48–68, 2010.
[11] J. Andrews, F. Baccelli, and R. Ganti, “A tractable approach to coverage
and rate in cellular networks,” IEEE Trans. Commun., vol. 59, no. 11, Xingqin Lin is currently a Ph.D. student at the Uni-
pp. 3122–3134, Nov. 2011. versity of Texas at Austin. He received the B.Eng.
[12] E. Royer, P. Melliar-Smith, and L. Moser, “An analysis of the opti- degree in Electronic Information Engineering from
mum node density for ad hoc mobile networks,” in Proc. 2001 IEEE Tianjin University (China) in 2009 and the M.Phil.
International Conf. Commun., vol. 3, pp. 857–861. degree in Information Engineering from the Chinese
[13] P. Nain, D. Towsley, B. Liu, and Z. Liu, “Properties of random direction University of Hong Kong in 2011, respectively. His
models,” in Proc. 2005 IEEE INFOCOM, vol. 3, pp. 1897–1907. research interests are mainly in the broad area of
[14] X. Hong, M. Gerla, G. Pei, and C. Chiang, “A group mobility model for wireless communications and networking. During
ad hoc wireless networks,” in Proc. 1999 ACM International Workshop his M.Phil. studies, he focused on analyzing and
Modeling, Analysis Simulation Wireless Mobile Syst., pp. 53–60. designing distributed wireless resource allocation
[15] A. Jardosh, E. Belding-Royer, K. Almeroth, and S. Suri, “Towards algorithms using the tools of game theory and op-
realistic mobility models for mobile ad hoc networks,” in Proc. 2003 timization. He currently works on device-to-device communications, carrier
ACM Mobicom, pp. 217–229. aggregation and mobility modeling and analysis for HetNets. He received the
[16] M. Musolesi, S. Hailes, and C. Mascolo, “An ad hoc mobility model Microelectronics and Computer Development (MCD) fellowship from UT
founded on social network theory,” in Proc. 2004 ACM International Austin in 2012, and held a summer internship at Alcatel-Lucent Bell Labs in
Symp. Modeling, Analysis Simulation Wireless Mobile Syst., pp. 20–24. Murray Hill, NJ in summer 2012.
[17] A. Saha and D. Johnson, “Modeling mobility for vehicular ad-hoc
networks,” in Proc. 2004 ACM International Workshop Veh. Ad Hoc Radha Krishna Ganti (S’01-M’10) is an Assis-
Netw., pp. 91–92. tant Professor at the Indian Institute of Technology
[18] W. Hsu, K. Merchant, H. Shu, C. Hsu, and A. Helmy, “Weighted Madras, Chennai, India. He was a Postdoctoral
waypoint mobility model and its impact on ad hoc networks,” ACM researcher in the Wireless Networking and Commu-
SIGMOBILE Mobile Comput. Commun. Rev., vol. 9, no. 1, pp. 59–63, nications Group at UT Austin from 2009-11. He
2005. received his B. Tech. and M. Tech. in EE from
[19] M. McNett and G. M. Voelker, “Access and mobility of wireless pda the Indian Institute of Technology, Madras, and a
users,” ACM SIGMOBILE Mobile Comput. Commun. Rev., vol. 9, no. 2, Masters in Applied Mathematics and a Ph.D. in
pp. 40–55, 2005. EE from the University of Notre Dame in 2009.
[20] M. Kim, D. Kotz, and S. Kim, “Extracting a mobility model from real His doctoral work focused on the spatial analysis
user traces,” in Proc. 2006 IEEE Infocom, vol. 6, pp. 1–13. of interference networks using tools from stochastic
[21] K. Lee, S. Hong, S. Kim, I. Rhee, and S. Chong, “SLAW: a new mobility geometry. He is a co-author of the monograph Interference in Large Wireless
model for human walks,” in Proc. 2009 IEEE Infocom, pp. 855–863. Networks (NOW Publishers, 2008).
[22] W. Hsu, T. Spyropoulos, K. Psounis, and A. Helmy, “Modeling spatial
and temporal dependencies of user mobility in wireless mobile net- Philip J. Fleming is Head of the Radio Technol-
works,” IEEE/ACM Trans. Netw., vol. 17, no. 5, pp. 1564–1577, 2009. ogy and Engineering group in the Advanced Tech-
[23] N. Aschenbruck, A. Munjal, and T. Camp, “Trace-based mobility nologies organization at Nokia-Siemens Networks
modeling for multi-hop wireless networks,” Comput. Commun., vol. 34, (B.V.) His current research interests are wireless
no. 6, pp. 704–714, 2011. networking and communication theory, particularly
multi-antenna cooperative transmission and recep-
[24] E. Cho, S. A. Myers, and J. Leskovec, “Friendship and mobility:
tion algorithms. He was a Distinguished Member of
user movement in location-based social networks,” in Proc. 2011 ACM
Technical Staff at Bell Labs and a Dan Noble Fellow
SIGKDD International Conf. Knowledge Discovery Data Mining, pp.
and Fellow of the Technical Staff at Motorola and
1082–1090.
has worked on all major cellular communications
[25] E. Hyytia, P. Lassila, and J. Virtamo, “Spatial node distribution of
systems, most recently LTE and LTE-A. He was also
the random waypoint mobility model with applications,” IEEE Trans.
Motorola Scientist-in-Residence at the MIT Media Lab in 2004-5. He has a
Mobile Comput., pp. 680–694, June 2006.
Ph.D. degree in Mathematics from the University of Michigan and has 15
[26] C. Bettstetter, G. Resta, and P. Santi, “The node distribution of the issued patents and numerous engineering publications.
random waypoint mobility model for wireless ad hoc networks,” IEEE
Trans. Mobile Comput., pp. 257–269, Sept. 2003. Jeffrey G. Andrews (S’98, M’02, SM’06, F’13)
[27] G. Resta and P. Santi, “An analysis of the node spatial distribution of the received the B.S. in Engineering with High Dis-
random waypoint mobility model for ad hoc networks,” in Proc. 2002 tinction from Harvey Mudd College in 1995, and
ACM International Workshop Principles Mobile Comput., pp. 44–50. the M.S. and Ph.D. in Electrical Engineering from
[28] E. Hyytiä and J. Virtamo, “Random waypoint mobility model in cellular Stanford University. He is a Professor in the De-
networks,” Wireless Netw., vol. 13, no. 2, pp. 177–188, 2007. partment of Electrical and Computer Engineering
[29] D. Stoyan, W. S. Kendall, and J. Mecke, Stochastic Geometry and its at the University of Texas at Austin. He developed
Applications, 2nd edition. John Wiley and Sons, 1995. CDMA systems at Qualcomm from 1995-97, and
[30] M. Haenggi, Stochastic Geometry for Wireless Networks. Cambridge has consulted for entities including Verizon, the
University Publishers, 2012. WiMAX Forum, Intel, Microsoft, Apple, Samsung,
[31] M. Haenggi, J. Andrews, F. Baccelli, O. Dousse, and M. Franceschetti, Clearwire, Sprint, and NASA. He is a member of
“Stochastic geometry and random graphs for the analysis and design the Technical Advisory Boards of Accelera and Fastback Networks. Dr.
of wireless networks,” IEEE J. Sel. Areas Commun., vol. 27, no. 7, pp. Andrews is co-author of Fundamentals of WiMAX (Prentice-Hall, 2007) and
1029–1046, Sept. 2009. Fundamentals of LTE (Prentice-Hall, 2010), holds the Earl and Margaret
[32] F. Baccelli and B. Blaszczyszyn, Stochastic Geometry and Wireless Brasfield Endowed Fellowship in Engineering at UT Austin, is an elected
Networks - Part I: Theory. Now Publishers Inc., 2009. member of the Board of Governors of the IEEE Information Theory Society
[33] F. Baccelli and S. Zuyev, “Stochastic geometry models of mobile and is an IEEE Fellow. Dr. Andrews received the National Science Foundation
communication networks,” Frontiers Queueing, pp. 227–243, 1997. CAREER award in 2007 and has been co-author of five IEEE best paper award
[34] L. Schroeder, “Buffon’s needle problem: An exciting application of recipients.
many mathematical concepts,” Mathematics Teacher, vol. 67, no. 2, pp.
183–186, 1974.
[35] V. Casares-Giner, V. Pla, and P. Escalle-Garcı́a, “Mobility models for
mobility management,” Netw. Performance Eng., pp. 716–745, 2011.
[36] J. Møller, Lectures on Random Voronoi Tessellations. Springer-Verlag,
1994, vol. 87.
[37] F. Baccelli and S. Zuyev, “Poisson-Voronoi spanning trees with ap-
plications to the optimization of communication networks,” Operations
Research, pp. 619–631, 1999.