<<

. 72
( 87 .)



>>


1
Note that polar coordinates can also be used.
377
14.2 DESIGN AND MODELING OF WIRELESS AND MOBILE AD HOC NETWORKS


(MHs) when, due to the motion process, they reach the boundaries of the simulated area.
Many possible solutions have been proposed depending on the simulation and model re-
quirements. The modeler should be careful about the effect that the boundary behavior,
composed with a mobility model, may have on the resulting spatial node distribution [3,
4, 5, 13]. In what follows, we shall distinguish between three boundary policies widely
used by the wireless and MANET networks™ modelers (Figure 14.2):

The “bouncing boundary” solution requires mobile hosts (MHs) to bounce back to-
ward the simulated area when they are going to move outside [3, 5, 13] (see Figure
14.2a). This simple solution can be used if the number of MHs in the simulation is
required to be constant (e.g., a closed system). Different “bouncing” rules can be
adopted, for example, mirror reflection (e.g., preserving the angle of incidence in
a bouncing angle “ or “ , respectively) and random reflection (i.e., a random re-
flection angle is generated). This boundary policy can be considered as quite unre-
alistic for large areas, and quite approximated for simulation of indoor mobility
(e.g., people moving in a room). A modeler has the choice to either preserve the
state of a bouncing MH or to create a new instance of MH when it virtually “leaves”
the area (i.e., the MH hits the area boundaries). This choice may be useful if the
state information of the MH is relevant for the simulation process or for the protocol
to be tested (e.g., protocols based on MH™s history and evolution state).
The “leave and replace” variant of the bouncing boundary policy is to delete a
“bouncing” (leaving) MH and clone it in a randomly chosen position within the
simulated area, following any node-position distribution (see Figure 14.2b). As in
the “bouncing boundary” solution, the cloning of the leaving MH can preserve state
and history information on the new instance or not, depending on modeler choice.
This policy may result in a nonuniform steady-state spatial node distribution, with a
node concentration around the center of the simulated area [3, 5, 13]. Intuitively,
this is due to the fact that MHs leave from the boundaries and reenter by choosing a
randomly distributed position “inside” the area. This also results in a biased (i.e., re-
duced) probability of finding MHs moving from the borders toward the middle of
the simulated area. This solution is rarely considered useful in MANET simulation.
The “torus boundary” solution is another policy widely adopted by many re-
searchers [3, 13]. In this policy, when a mobile host reaches the north, west, south,
and east boundaries of any rectangular area, it simply leaves the area and reenters
with the same direction and speed from the south, east, north, and west bound-


N
(end)
rand(x,y)
(start)
(end) E
(start) (end) W

(start)


S
a) Bouncing b) Leave & replace c) torus

Figure 14.2. Three boundary policies.
378 SIMULATION AND MODELING OF WIRELESS, MOBILE, AND AD HOC NETWORKS


aries, respectively (see Figure 14.2c). Intuitively, the rectangular area is wrapped
around itself, north with south, and east with west, like a ring. The reason why the
torus policy is widely used is given by its simple implementation, and because it
simplifies the management of uniformly distributed host densities and directions
(accordingly with the implemented motion models). There is a full correlation ef-
fect balancing leaving and reentering hosts™ directions and velocities. Again, leav-
ing MHs can preserve state and history information when they reenter the area,
depending on modeler choice (except for velocity and direction, which should be
maintained).

14.2.1.2 Host Sources and Position Distributions. Host mobility is the main
factor in determining the “arrival” and physical presence of a set of MHs within a fixed
simulated area. The physical presence of hosts does not necessarily represent the relevant
factor to be modeled for the simulation goals. This mainly depends on the MH roles to be
modeled and on the performance indices required. As an example, a switched-off MH in
the simulated area may not be considered relevant to determine transmission-based per-
formance indices. We denote as “active” the role of a MH that can be considered effective
in determining the value of a performance index whose evaluation is the goal of a simula-
tion run. Every performance analysis for a wireless mobile system can be performed by
assuming a (fixed or variable) number of “active” mobile hosts (MHs) implemented in a
limited area of interest. In what follows, we focus upon active MHs, and consider only the
motion-related “presence” of these MHs. The modeler should evaluate the opportune and
realistic definition of the average density of “active” MHs, with attention to the policies
for the creation and position allocation of new MH instances, both at the simulation start
and at run time [54].
Dealing with the sources and creation policies in accordance with the factors influenc-
ing the performance metrics of interest, one possible choice is to require the number of
active objects in the simulated area to be constant. This choice may be useful, for exam-
ple, when performance metrics to be obtained are related to the number of MHs, or to the
MHs™ density (e.g., existence of route path, network partition, average degree of MHs). In
the following, we denote MHs™ presence in the simulated area as “active presence,” what-
ever meaning this would imply:

In closed systems, the initial number of “active” simulated MHs is constant, and
every MH lives (i.e., it maintains its functionalities) for the whole simulation run.
Balanced systems realize a simple hybrid solution that can be useful for some simu-
lation analysis. Every time a MH leaves the simulated world (e.g., moving outside
the simulation area, or switching off the network interface), a new instance of the
MH is causally introduced in the simulated area, following the selected position dis-
tribution. Bouncing, torus, and “leave and replace” boundaries allow a natural im-
plementation of a closed or balanced system under the MHs™ mobility viewpoint (if
MHs™ sources and sinks are missing).
In open systems, one or more sources of active simulated MHs are defined (e.g., in-
terarrival or activation processes for MHs). In such scenarios, the modeler should
evaluate and select interarrival time distributions for the sources (e.g., exponential
or Poisson distribution), sink policies (e.g., MHs with no battery energy or moving
outside the simulated area are discarded), and the initial-position distribution for in-
379
14.2 DESIGN AND MODELING OF WIRELESS AND MOBILE AD HOC NETWORKS


coming MHs (e.g., at random uniformly distributed coordinates in the area). If the
arrival process is too fast, the system is unstable, that is, the asymptotical number of
MHs in the simulated area is not upper bounded. This can lead to a biasing problem
if we are interested in the evaluation of performance metrics that may be related to
the average MH density (e.g., multihop link reliability in routing protocols, network
connectivity and partitions, average next-hop distance, average transmission power,
etc.) [3, 12, 13, 23]. To obtain an open, stable system, the rate of MHs leaving the
area should be statistically balanced by the sum of arrival rates of the MH sources.
This means that the number of MHs in the area is not a constant value (as in closed
and balanced systems), but converges asymptotically.

One possible choice for initial allocation of a new MH™s position is a random selection
of its position coordinates within the simulated area. Uniform or normal distributions are
widely used in the literature, depending on the host density to be modeled (e.g., uniform
vs. hot-spot density, respectively) [13, 55]. This choice does not provide any best-effort
guarantee about network partitions. One possible solution to this problem is to divide the
simulation area into a grid of square cells, with a size that could be determined by the
minimum range of connectivity among the MHs, and distribute MHs™ positions such that
at least one MH is in every cell of the grid.
On the other hand, a real, sampled distribution snapshot of MH positions can be used,
if available. This is a common way to model hotspot MHs™ distribution [55]. In many sce-
narios, depending on the mobility models and their respective parameters adopted for
MHs, the initial distribution of MHs is less or more relevant in determining the steady-
state MHs™ distribution [3, 13]. When the “memory effect” of the initial distribution is not
preserved by the motion model characteristics, every possible transient effect of the initial
distribution should be evaluated and eliminated to collect unbiased steady-state results.
The choice of runtime position-allocation policies for newly generated MHs is a little
bit more subtle. Random allocation of MHs may result in nonuniform distribution and bi-
ased node density, for example, if hosts leave the system only from the boundaries [3]. A
possible solution for this scenario would be to “delete” hosts selected randomly in the
simulation area, and to adopt distribution-balancing boundary policies such as bouncing
and torus borders [3, 5, 13].

14.2.1.3 Coverage Areas, Physical Propagation, Transmission Errors, and
Interference Models. Usually in MANETs, every host can be considered as a poten-
tially mobile host. As a consequence, hybrid MANETs under analysis today may include
fixed, static base stations (BSs), with their respectively managed coverage areas, as in cel-
lular and PCS systems [18]. A detailed physical model for wireless transmission, includ-
ing propagation, mobility, error, and interference models, is one of the most difficult and
computationally expensive tasks to do, and strong approximation and assumptions are
usually introduced [2, 3, 12, 35, 42, 62, 66]. Many models and solutions have been pro-
posed, at different levels of detail [26, 59]. We will skip most of the details, for space rea-
sons, and we will just point out some of the modeling issues related to MANETs.
The physical wireless transmission is based on the emission of electromagnetic waves
coding information with many possible modulation and coding techniques. The natural
decay of transmitted signals can be modeled following simple analytical approximations.
If the residual signal power of the receiving network interface is above the detection
380 SIMULATION AND MODELING OF WIRELESS, MOBILE, AND AD HOC NETWORKS


threshold, a communication is possible. Otherwise, to allow a communication (link estab-
lishment) between the intended sender and receiver, it would be necessary to increase the
transmission power of the sender and/or reduce their relative distance d.
One of the most used propagation models, adopted in MANET simulation is the sim-
ple Free Space Propagation Model [12, 53]: if Pt is the transmission power (i.e.,
energy/time) used for the signal transmission, then the receiving power Pr is proportional
to 1/d2, where d is the distance between sender and receiver in open space (see Figure
14.3).
The Free Space Propagation Model can be extended to better describe the effects over
near and far receivers, with the Two-Ray Ground Reflection Model [12, 53]. This model is
the same as Free Space Propagation Model, except when the distance d is greater than a
crossover point, called the reference distance (around 100 meters). For such long dis-
tances, the receiving power Pr is modeled as proportional to 1/d , > 2.
The Free Space and Two-Ray propagation models assume ideal propagations over a
circular area around the transmitter. To model irregular coverage areas, the Shadowing
Propagation Model [45, 53] is defined with two components: a component similar to the
Free Space Propagation Model, and a random component to make randomly variable (and
statistically controlled) the edge of the communication range. For a complete discussion
of the Free Space Propagation Model and other models, see [26, 53].
A modeling choice to define if a transmission can be detected by a tagged receiver is
to define a receiving threshold (RTX) and a carrier-sense threshold (CTX) for every de-
vice [12, 59, 53]. For every simulated transmission, it would be required to scan every
MH in the system and to apply the propagation model to the transmitted signal. This re-
quires evaluation, for each receiver, if the receiving power perceived for the ongoing
transmission is sufficient for reception (i.e., greater than RTX), if it is sufficient for de-
tection and carrier sensing (i.e., greater than CTX), or if it is simple interference.
Reception and carrier sensing events can be passed to the model components devoted to
manage events at the upper layers of the model, for example, Medium Access Control
policy implementation. This scan-based computation may require a long time if per-
formed for a large set of MHs.



transmission power


propagation model law

Receiver RTX
Thresholds CTX
d

Transmitting Host

Transmission area

Detection area
simulated area

Figure 14.3. Transmission power, propagation, and coverage areas.
381
14.2 DESIGN AND MODELING OF WIRELESS AND MOBILE AD HOC NETWORKS


The system model can be extended with coverage areas, in order to reduce the trans-
mission-detection overhead and to model much more complex propagation models, de-
pending on the modeling and simulation requirements. Transmission coverage area defin-
ition can be directly associated with every transmitter, but the area size and shape is
relative to the receiver thresholds. For ease of management, the area-size definition would
require the assumption of common threshold levels (i.e., common CTX and RTX values)
for every MH in the system (see Figure 14.3).
The transmission (coverage) area of a wireless transmitter can be defined as the area
where the transmitted wireless signal propagates and can be correctly detected and decod-
ed (i.e., transmission is possible with few/no errors due to interference). This area depends
on the transmission power of the transmitter, on the propagation model, on the reception
threshold (sensitivity) of the receiving network interface (RTX), and on the amount of in-
terference (noise) caused by many possible factors (described in the following). The trans-
mission area should be defined and managed in the model, for each MH, in order to dy-
namically evaluate a communication capability (i.e., a direct link) between every
candidate transmitter and receiver.
The detection area (see Figure 14.3) of a wireless transmission device is the area
where the signal propagates, and where it can be detected by a carrier sensing mechanism,
without being necessarily decoded (i.e., CTX Received_Signal_Power RTX). This
means that a mobile host can sense the wireless medium as busy without being able to de-
code received signals. The definition of this area in the model, for each MH, may be rele-
vant for the evaluation of detailed carrier sensing and MAC-level effects, such as exposed
terminals, hidden terminals, and capture effects (described in the following) [24, 52].
The interference area of a wireless transmission device is defined as the area where the
transmitted wireless signal propagates, without being detected or decoded by any receiver,
adding interference and noise to any possible ongoing transmission for intended receivers.
The cumulative effect of noise (i.e., interference) might add errors to the transmission of
bits of information. The definition and management of this area in the model, for each
MH, may be relevant for the evaluation of detailed interference and error models. Trans-
mission areas are included in detection areas, and detection areas are included in interfer-
ence areas, given the propagation properties of wireless signals in open spaces. Possible
choices to model coverage and interference areas in open spaces are given by regular
polygons centered in the transmitter™s position. Circular coverage areas can be defined for
open-space propagation models, and are a simple common choice for MANET modeling
and simulation purposes. The circle radius, centered on the transmitter position, can be
made proportional to the transmission power in an adaptive way, mapping to real power
control management and policies implemented in simulated MHs. If fixed (static) trans-
mitters are present, with a constant transmission power, as in cellular and PCS networks, a
common choice is to approximate the coverage areas by hexagons, squares, and Voronoi
diagrams. This choice can simplify the link management, because connectivity between a
MH and a fixed transmitter can be evaluated with no ambiguity (i.e., only one reference
transmitter is defined in every point). This choice is simple to define and to manage for
the simulation purpose, but it realizes a strong approximation of the real behavior of wire-
less transmissions. The circular coverage model is quite realistic in open spaces, but it
would require some changes if obstacles were present to interfere with signal propagation.
When obstacles are present in the simulation area, the coverage and interference areas
may be severely affected, in almost unpredictable ways [35, 47, 53]. This is even worse if
mobility of wireless sources (or, equivalently, wireless receivers) is present. Models to
382 SIMULATION AND MODELING OF WIRELESS, MOBILE, AND AD HOC NETWORKS


deal with obstacles have been defined [32, 47, 53, 55]. Moreover, if the antennas are not
omnidirectional and the transmission beam is not isotropic, the regular polygon choice for
coverage areas is even more approximate (e.g., with smart antennas, the transmission en-

<<

. 72
( 87 .)



>>