Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
approach based on the area-ratio is much more mathematically 2) 23 ≤ d ≤ 1: In this case the circle is cut off by the
tractable, which can derive the same result as [3]. Moreover, edges of the hexagon, e.g., AB in Fig. 1 (b). The intersection
πd2
by minor extensions, this approach can be applied to deriving is the area of the sector,
6 , minus the
area thatis cut off by
√ √
πd2 3 3
the distance distributions from a non-center point, either inside AB, i.e., S = 6 − d2 cos−1 2d − 2 d2 − 34 . Therefore,
or outside a cell, to any point inside the hexagonal cell. For √
instance, the distance from any of the second-layer hexagon 2 πd2 3 √ 3
FD (d) = √ − 2d2 cos−1 + 3 d2 − .
centers, and from a hexagon vertex (Section III-C) to the 3 3 2d 4
interior of a hexagon, etc. (2)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
A A
O O O
O O
B B
√ √ √
3 (b) 1 ≤ d ≤ 2 (a) 0 ≤ d ≤ 1 (b) 1 ≤ d ≤ 3 (c) 3≤d≤2
(a) 2
≤d≤1
Fig. 3. Random Distances from One Vertex of a Hexagon.
D D
P P
A A
√ √
4) 3 2 3 ≤ d ≤ 7: S, as shown
R
√
in Fig. 2 (d),
√ √
is the sum
O O
3 3 3
= = ( d2 − 3 −
S
B B Q
of trapezoidal
√ area S ABCD 4 , S PQCD 4
Q
C C
1)(5 − d2 − 3), which are the same as the last case, and
√ √ √ the part that belongs to the sector
(c) 2 ≤ d ≤ 3 3
(d) 3 3
≤d≤ 7 √ but cut √ off by√ the edge of
hexagon, S = d2 sin−1 3− 2dd −3 − 43 (1+ d2 − 3)(3−
2 2 2
SQ √
Fig. 2. Random Distances from the Center of an Adjacent Hexagon. PR
√ √
d2 − 3) − d2 cos−1 32d3 − 3 2 3 d2 − 27 4 . Therefore,
√ √
Although the distribution function of d in (1)–(2) can also 2 d2 − 3 3−
2 −1 −1 3 3
be derived by the approach in [3], the approach used here FD (d) = √
d sin − cos
3 3 2d 2d
is much more convenient than the geometric integral in [3].
√
By fD (d) = FD
(d), the probability density function of the √ 3 27 5 3
random distances inside a hexagon is + 3 d2 − 3 + d2 − − .(7)
2 4 4
⎧ π √
⎪
⎨ 0 ≤ d ≤ 23
4d 3
√ √ Combining (4)–(7), and by fD (d) = FD (d), the probability
fDI (d) = √ − 2 cos −1 3 3
2 ≤d≤1
π . (3) density function of random distances, from the center of one
3⎪⎩ 3 2d
0 otherwise hexagon to an arbitrary node in an adjacent hexagon, is
⎧ √ √
B. Random Distances from the Center of an Adjacent Hexagon ⎪ −1 3 3
⎪
⎪ cos 2d 2 ≤d≤1
⎪
⎪
The derivation of the distance distribution from the center of ⎪
⎪ 6
π
1≤d≤2
⎪
⎪ √ √
an adjacent hexagon also can be treated as an area ratio. Here ⎪
4d ⎨ sin
√ −1 3− 2 −3
we need to divide S by the area of a hexagon, i.e., A = 3 2 3 .
d
2d 2 ≤ d ≤ 323
fDA (d) = √ √ .
Suppose two hexagons are adjacent to each other as shown in 3 3⎪ ⎪
⎪ sin−1 3− d2 −3
⎪
⎪
2d
√ √ √
Fig. 2, and let the center of one hexagon be the origin. If each ⎪
⎪ − cos −1 3 3 3 3
side of the hexagons has an unit length, S has the following
⎪
⎪ 2d 2 ≤d≤ 7
⎪
⎩ 0
four cases. otherwise
√
1) 23 ≤ d ≤ 1: S in this case is the area of the (8)
sector cut off √by edge√
AB, as shown in Fig. 2 (a). Then
C. Model Extensions
S = d2 cos−1 2d3 − 23 d2 − 34 , and
Similarly, the distance distribution from a non-center point,
√ √
S 2 3 3 3 e.g., a vertex of a hexagon, can be derived using the same
2 −1
FD (d) = = √ d cos − 2
d − . (4) area-ratio approach. According to Fig. 3(a)–(c), we have the
A 3 3 2d 2 4
following three subcases:
2
2) 1 ≤ d ≤ 2: S is the area of the sector minus √the area 1) √0 ≤ d ≤ 1: As shown in Fig. 3(a), S = πd3 , and
2
of triangle OAB, as in Fig. 2 (b). Thus, S = πd6 − 43 , and A = 3 2 3 , therefore,
√
2 πd2 3 2π
FD (d) = √ − . (5) FD (d) = √ d2 . (9)
3 3 6 4 9 3
√ √
3) 2 ≤ d ≤ 3 2 3 : As shown√ in Fig. 2 (c),√S is the sum 2) 1 ≤ d ≤ 3: The intersection area in this case, as
√ 2π
of trapezoidal
√ area SABCD = 3 4 3 , SPQCD = 43 ( d2 − 3 − shown in Fig. 3(b), is the sector with radius d and angle 3 ,
1)(5− d2 − 3), and the part that belongs to the sector S = minus two parts 2that extend
outside
√ √
the hexagon
on each side.
√ √ PQ 2 3 4d 2 −3−1
√ √ That is, S = 3 − 2 2 d − 4
πd θ
, where and θ =
d2 sin−1 3− 2dd −3 − 43 (1 + d2 − 3)(3 − d2 − 3). Then,
2 2
√ √ √
−1 3 4d2 −3−1
√ √ sin 4 d . By simplification, 3 − θ = sin−1 2d3 , so
π
2 3 − d2 − 3 √ 2 5 3 √ √
FD (d) = √ d2 sin−1 + 3 d −3− . 2d 2
3 4d2 − 3 − 1
3 3 2d 4 FD (d) = √ sin −1
+ . (10)
(6) 3 3 2d 6
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
√ 2
√ Circle, Inscribed
2 2
Circle, Enclosing
d2
0.01
π d 2−3
FD (d) = √ − 2 sin−1 + d − 3. (11)
3 3 3 d 3 0
0 10 20 30 40 50 60
Distance − d (m)
70 80 90 100
⎪ 3
⎪ √ √ 0
2d ⎨ −1 3
2 sin 2d 1≤d≤ 3
50 100 150 200
Distance − d (m)
250 300
fDV (d) = √ √ √ .
3 3⎪ − 2 sin−1 dd−3
2 0.01
⎪ π
3≤d≤2
(12) 0.004
0.002
This is the scenario where the BS is located at the in-
0
tersection point of three cells, using directional antenna for 0 20 40 60 80 100 120
Distance − d (m)
140 160 180 200
Simulation Simulation
1.7 Simulation
E[log (1+P d−α/N )]
0.8
2 t 0 0.7 E[log2(1+Ptd−α/N0)] −α
E[log2(1+Ptd /N0)]
1.6
0.7
Circle, Same Area Circle, Same Area
0.6
Circle, Same Area
1.5 Circle, Inscribed Circle, Inscribed
1 Hexagon Distribution 0.6 Circle, Inscribed
Circle, Enclosing Circle, Enclosing
Capacity (bps/Hz)
Circle, Same Area Circle, Enclosing
Capacity (bps/Hz)
Capacity (bps/Hz)
PDF (One Cell)
1.4
Min−Man Avg 0.5 Min−Man Avg
Circle, Inscribed 0.5 Min−Man Avg
Circle, Enclosing 1.3
0.4
0.4
1.2
0.5 0.3
1.1 0.3
0.2
1
0.2
0.1
0.9
0 0.8 0.1 0
0 0.5 1 1.5 2 2.5 3 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 −30 −28 −26 −24 −22 −20 −18
−log D SNR (dBm)
2 Cell Size (m) Path Loss Exponent α ref
(a) Distribution of log2 D (b) Capacity vs. Cell Size (c) Capacity vs. Path Loss Exponent (d) Capacity vs. Transmission Power
Fig. 5. Per-User Link Capacity.
where B is the channel bandwidth. C is determined by the to the path loss exponent, where the error grows with α.
following parameters in a high SNR environment, SNRref in Fig. 5(d) is the received signal strength for nodes
at the cell border, corresponding to Pt from 4 to 64 Watt.
C ≈ B (log2 Pt − α log2 D − log2 N0 ) ,
From the figures, as we anticipate, the link capacity decreases
which is a function of D when Pt , α and N0 are constants. with the cell size and path loss exponent, but increases with
Different values of Pt and N0 shift the range of C, and α the transmission power. The min-max average model deviates
expands or squeezes C. Let Y = − log2 D, with the distance from the simulation results further when the cell size is
distribution function fDI (d) in (3), we have smaller, or when α is higher. However, it is interesting to
see from all three figures that, using a circle with the same
P (Y ≤ y) = P (− log2 D ≤ y)
area size as a hexagon for approximation, the circle provides
= P (D ≥ 2−y ) = 1 − FDI (2−y ). the best approximation, as we can expect from Fig. 5(a). Our
Therefore, hexagon model is accurate for all the parameter settings.
fY (y) = 1 − FDI (2−y ) = ln(2)2−y fDI (2−y ). (17) B. Co-Channel Interference
On the other hand, the circular approximations replace Besides the signal from the intended transmitters in the
fDI (2−y ) in (17) by fDI (2−y ) in (14), with different values of same cell, interferences in the same frequency channel (e.g.,
r. Figure 5(a) shows the distribution of (17) with the hexagon in CDMA systems) also arrive at the BS from the undesired
model and circular approximations, where the circle with the transmitters in other cells, and lead to the degradation of the
same area size as the hexagon best approximates the hexagon system performance. In a CDMA system, while a mobile
distribution, because their PDFs are mostly overlapping. This user connects to its serving BS, the interference it causes
gives us a hint that this particular circle may give good results simultaneously at an adjacent BS is proportional to Pt d−α ,
for statistical moments, after being integrated. However, the where d follows the distribution fDA (d) in (8). Let Z = D−α ,
exact distribution, instead of the expectation and other mo- where D is the random variable for the distance d, we have
ments, determines the important system parameters especially
when an individual user moves inside a cell. For instance, P (Z ≤ z) = P (D−α ≤ z)
1 1
users at different locations have different capacities when = P (D ≥ z − α ) = 1 − FDA (z − α ).
communicating with the BS.
For the convenience of modeling and simulation, we give Therefore,
the analysis of the first moment, the expectation of link 1
1 1 1
capacity. To compute this expectation, in the unit of per user fZ (z) = 1 − FDA (z − α ) = z − α −1 fDA (z − α ). (19)
α
per Hz, with the hexagon distribution fDI (d), we have 1
The circular approximations replace fDA (z − α ) in (19) by
Pt x−α Pt x−α
fDA (z − α ) in (15). Figure 6(a) shows the distribution of (19)
1
E log2 1 + = log2 1 + fDI (x)dx,
N0 N0 with the hexagon model and circular approximations, where
(18)
the circle with the same area, as well as the inscribed circle,
by integrating over the cell area. Figure 5(b) compares the
are both close to the distribution of D−α . But the former
simulation results with the hexagon distribution model in (18),
again best approximates the hexagon distribution, with respect
the circular approximations and the min-max average, assum-
to the overlapping region of the corresponding PDFs. The
ing the path loss exponent α = 2 and the BS transmission
expectation of the received interference strength from a mobile
power Pt = 50 Watt. The figure shows that (18) is the most
user, at the BS of any six adjacent cells, using the distance
accurate model, whereas the other models either overestimate
distribution function fDA (d) in (8), is
or underestimate the link capacity.
In Fig. 5(c), where α is used as the control parameter,
the min-max average model is clearly much more sensitive Pt E[d−α ] = Pt x−α fDA (x)dx, (20)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
3.5
Simulation 0.045 Simulation
0.2 0.093
PtE[d−α] 0.0215
−α
PtE[d ]
0.06 Simulation 0.0492
0.092
0.04 P E[d−α]
3 t
Hexagon Distribution 0.18 Circle, Same Area 0.021 Circle, Same Area 0.05
Circle, Same Area 0.035 Circle, Same Area
0.091 Circle, Inscribed Circle, Inscribed
PDF (Adjacent Cells)
0.0205 0.0488
2.5 Circle, Inscribed Circle, Inscribed
Interference (Watt)
Interference (Watt)
Interference (Watt)
0.16
1447 1450 1453 Circle, Enclosing 0.03 Circle, Enclosing −19 −18.95
Circle, Enclosing Min−Max Avg 2.245 2.25 2.255
Min−Max Avg
0.04 Circle, Enclosing
2 Min−Max Avg
0.14 0.025
0.03
1.5 0.02
0.12
0.015 0.02
1
0.1
0.01
0.5 0.01
0.08
0.005
0
10
−0.9 −0.7
10
−0.5
10 10
−0.3
10
−0.1
10
0.1
0.06 0 0
1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 −30 −28 −26 −24 −22 −20 −18
D−α Cell Size (m) Path Loss Exponent α
SNR (dBm)
ref
(a) Distribution of D−α (b) Interference vs. Cell Size (c) Interference vs. Path Loss Exponent (d) Interference vs. Tx Power
Fig. 6. Co-Channel Interference.
by integrating over the adjacent cell area. Figure 6(b) compares interference factor. Furthermore, the distance distributions
the simulation results of the average interference with different when both endpoints inside a cell, or between adjacent cells,
models, with respect to the cell size. It is obvious that become random, are more mathematically challenging [16],
the model using the min-max average largely underestimates [17]. However, such distributions will be very useful, e.g., in
the average co-channel interference. While the two circular analyzing hidden terminal problems and cooperative commu-
approximations in the zoom-in plot, i.e., using a circle with the nications. We believe the probabilistic model proposed in this
same area as the hexagon and using an inscribed circle, both paper and its future extensions will provide guidelines for a
slightly underestimates the average interference. As shown in more accurate network dimensioning and protocol design.
Fig. 6(a), their interference distribution functions are different
R EFERENCES
from that of a hexagon. Although using a circle with the
same area size still gives the best approximation, the slightly [1] Y. Zhuang, J. Pan, and L. Cai, “Minimizing energy consumption with
probabilistic distance models in wireless sensor networks”, Proc. IEEE
different distribution (even with the same average value) will INFOCOM’10, pp. 2453–2461, 2010.
affect important system parameters such as the link outage [2] D. P. Chu, “Distance between random points in two rectangular cities”,
probability. Comm. in Statistics-Simulation and Computation, 35(2):257–276, 2006.
[3] A. M. Mathai, “Random distances associated with triangles”, Int. J. of
Figure 6(c) plots the average co-channel interference with Mathematical and Statistical Sciences, 7(1):77–96, 1998.
different values of α, and Fig. 6(d) plots the results with [4] J. S. Evans and D. Everitt, “On the teletraffic capacity of CDMA cellular
different transmission power. The min-max average again networks”, IEEE Trans. on Veh. Tech., 48(1):153–165, 1999.
[5] Z. Lei, D. J. Goodman and N. B. Mandayam, “Location-dependent other-
underestimates the interference greatly in both figures. For cell interference and its effect on the uplink capacity of a cellular CDMA
instance, when α = 2.5, the min-max model underestimates system”, Proc. IEEE VTC’99, pp. 2164–2168, 1999.
the interference by 50%. Note that even if the system is [6] M. G. Jansen and R. Prasad, “Capacity, throughput, and delay analysis of
a cellular DS CDMA system with imperfect power control and imperfect
not CDMA-based, frequency reuse is unavoidable in such sectorization”, IEEE Trans. on Veh. Tech., 44(1):67–75, 2002.
systems. In these situations, we can extend the above results by [7] J. S. Wu, M. T. Sze and J. K. Chung, “Uplink and downlink capacity
considering the co-channel interference from the cells which analysis for two-tier CDMA cellular systems”, Proc. IEEE INFOCOM’97,
pp. 626–633, 1997.
are non-adjacent to the tagged base station, and derive the [8] J. Liu, X. Jiang, H. Nishiyama, N. Kato, “Reliability assessment for
co-channel interference distribution accordingly. wireless mesh networks under probabilistic region failure model,” IEEE.
Trans. on Veh. Tech. 60(5):2253-2264, Jun 2011.
V. C ONCLUSIONS [9] L. E. Miller, “Distribution of link distances in a wireless network”, J. of
Research-National Inst. of Standards and Tech., 106(2):401–412, 2001.
By deriving the closed-form distance distribution functions [10] L. E. Miller, “Joint distribution of link distances”, Proc. Conf. on
Information Science and Systems, The John Hopkins University, 2003.
between cellular BSs and users inside the same, or adjacent [11] A. M. Mathai and G. Pederzoli, “Random points with reference to a
hexagonal cells, this paper have developed a geometric prob- circle, revisited”, Rendiconti del Circolo Matematico di Palermo, Serie
ability model for analyzing the per-user link capacity and co- II, Suppl., 50, pp. 235–258, 1997.
[12] K. B. Baltzis, “Hexagonal vs circular cell shape: a comparative analysis
channel interference in cellular systems. The derivation tech- and evaluation of the two popular modeling approximations”, Cellular
nique, which is based on the area-ratio that is not limited by Networks—Positioning, Performance Analysis, Reliability, Ch. 4, 2011.
node coordinates, is simpler when compared with the existing [13] H. Haas and S. McLaughlin, “A derivation of the PDF of adjacent
channel interference in a cellular system”, IEEE Comm. Letters, 8(2):102–
methods in geometric probability, and the accuracy of the 104, 2004.
model has been validated through simulations. The approach [14] P. Pirinen, “Cellular topology and outage evaluation for DS-UWB sys-
can also be applied to deriving the distance distributions from tem with correlated lognormal multipath fading”, Proc. IEEE PIMRC’06,
pp. 1–5, 2006.
a non-center point. [15] S. W. Oh and K. H. Li, “Effect of circular-cell approximation on the
Our future work includes deriving distance distributions forward-link BER performance of a power-controlled CDMA system”,
under general user distributions, such as Gaussian distribu- Proc. IEEE GLOBECOM’99, pp. 2472–2476, 1999.
[16] Y. Zhuang and J. Pan, “Random Distances Associated with Rhombuses”,
tions, etc. The conditional probability of distances from an ArXiv Technical Report, arXiv: 1106.1257, 2011.
arbitrary point to both the serving BS and the interfering BS, [17] Y. Zhuang and J. Pan, “Random Distances Associated with Hexagons”,
can be derived and utilized to model the location-dependent ArXiv Technical Report, arXiv: 1106.2200, 2011.