<<

. 73
( 87 .)



>>

ergy is not uniformly propagated in all directions, but has a directional effect [53]). Mod-
eling of asymmetric beams and obstacles in these scenarios might be quite complex and
computationally expensive, too.
To model interference effects, many additional factors need to be defined, and a realistic
model is quite hard if not impossible to obtain, without paying for a high computation over-
head. As a simple description of the wide set of problems encountered when dealing with
accurate interference modeling of physical wireless transmission, we present a short list of
system details that should be considered in theory, and that aren™t usually considered in
many models, given the complexity and extensive computation required to simulate their
effects. Most of the described problems are given by continuous physical phenomena,
whose approximate modeling in the discrete-event simulation field would be really hard
and expensive. Mobility is an additional source of problems. In MANET scenarios, the
model would be even more complex than in cellular and PCS networks, because both the
transmitter and the receiver usually can be mobile, and a distributed, relative-mobility pa-
rameter should be evaluated, instead of a local, absolute-mobility parameter; Physical prob-
lems to be modeled in wireless transmission include the following phenomena [53]:

Fading: a physical phenomenon, frequency dependent, inducing delay and phase
variations between the main transmitted signal (following a dominant path) and
many secondary signals (following alternative paths), caused by obstacles and mo-
bility. This causes long-term and short-term variations of the resulting reception
power of transmitted signals. The Additive White Gaussian Noise model is used to
represent ideal channel conditions under the signal fading viewpoint. Rayleigh and
Ricean Fading are widely accepted models [2, 35, 59, 53] used for fading-prone
scenarios. They can be applied to highly mobile scenarios, No Line of Sight
(NLOS) and Line of Sight (LOS) paths, respectively [59]. The K parameter of the
Ricean Fading Model can be used to control the composed effect of LOS and NLOS
signal powers [59]. A Coherence time parameter is adopted to control the time fre-
quency and duration of fading effects on the channel. As an example of the model-
ing complexity for fading effects, let us assume that there are M base stations and N
mobile hosts in the simulation scenario, and there are roughly L paths determined
by obstacles in each propagation direction. Then we would need approximatively up
to 2M * N * L instances of Rayleigh fading generators. Efficient implementation of
fading models is still an ongoing research activity [2, 35, 53].
Shadowing: attenuation of signal power propagation caused by physical obstacles.
This effect is mainly responsible for irregular coverage areas. The Shadowing Prop-
agation Model defined in the previous section gives a statistical approximation of
this effect [53].
Reflection: signal reflection caused by large obstacles and indoor walls. It is quite
important to model this effect for indoor scenarios.
Refraction: Marginal signal change and reflection caused by variation in the medi-
um density.
Scattering: signal diffusion caused by sharp obstacles.
Diffraction: signal deviation caused by large edges and corners.
383
14.2 DESIGN AND MODELING OF WIRELESS AND MOBILE AD HOC NETWORKS


Each one of the above-mentioned phenomena may have different characteristics, given
different physical implementations and different coding techniques adopted for wireless
transmissions. The whole effect of such a collection of complex phenomena is usually
modeled as a simple error probability for a given amount of information received (e.g., a
bit) on the physical channel. The idea is to enclose all this in a black box describing the
whole effect and call it the probability to obtain a bit error. Obviously, this may be a
strong, unacceptable approximation, depending on the aim of the simulation. In many
models, in order to approximate the real behavior of the wireless medium, the physical
medium (or its high-level abstraction, the channel) behavior can be described with more
accurate error models. Signal to interference and noise ratio (SINR) and signal to noise
ratio (SNR) are the key parameters adopted to model the signal composition of interfer-
ence effects described above. The generalized SINR and SNR values, together with RTX
and CTX thresholds, can be adopted to model with some detail the high-level effect of in-
terference resulting in bit error rates (BER) and frame error rates (FER). BER and FER
values model the generalized probability that a transmitted bit or frame is received with
errors, respectively. FER is a function of BER and the frame length (in bits). In order to
capture the bursty effect of wireless transmission errors, the Gilbert“Elliott Error Model
has been used to define the status of the wireless medium as a function of time [62]. This
model defines a random Markov process between the following two states: good and bad.
Good status is characterized by low probability of bit error (low bit error rate, BER),
whereas bad status is characterized by high bit error rate. The time in bad and good status
is usually sampled from an exponential (or its discrete counterpart, geometric) time distri-
bution, with respective parameter and average values. It is a good choice to implement it
in time-slotted models.
The coding techniques adopted for the wireless transmission and the frequency spec-
trum allocated for adjacent channels are additional parameters to be evaluated in order to
define opportune error models for wireless communication. New coding techniques allow
for interference effects™ cancellation, and interference models should be defined for adja-
cent-channel interference and co-channel interference [53]. We skip all the details for
space reasons.

14.2.1.4 Link Definition and Network Topology
The modeling of coverage areas is really important because it is related to the link defini-
tion between any couple of MHs, or between a MH and a BS. This is really important in
MANETs, because a simple star topology given by a set of MHs around a BS is not a con-
crete and dynamic vision of the system. Moreover, in modeling and simulation of
MANETs, the “link-established” property between a couple of MHs is more complicated
than in most wireless cellular networks, mainly due to the management of relative mobili-
ty (as opposite to absolute mobility) of these mobile hosts [66]. For any couple of MHs
and/or fixed hosts (e.g., base stations) covering the area of interest for the simulation (call
them host X and Y), we have three possible expected scenarios related to reception and
carrier sensing thresholds, coverage areas, and link definitions [24]:

1. X is out of the transmission area of Y, and vice versa. This means that X and Y are
partitioned (e.g., see A and C in Figure 14.4). The network topology is not assumed
to have any direct communication link between X and Y. Maybe communication be-
tween X and Y is possible, at the upper routing layer, if supported by an intermedi-
ate-hosts chain of mutually reachable hosts (e.g., like host B and C in Figure 14.4).
384 SIMULATION AND MODELING OF WIRELESS, MOBILE, AND AD HOC NETWORKS



C™s transmission area

C™s detection area
A B


D C




Figure 14.4. Example of a collision domain.



2. X is within the transmission area of Y, and Y is out of the transmission area of X.
This means that Y can communicate with X, but not vice versa. In this scenario, a
monodirectional link exists from Y to X (e.g., see hosts A and D in Figure 14.4).
Monodirectional links exist in many real scenarios, mainly due to the different
transmission power and propagation characteristics of MHs (e.g., see host D in Fig-
ure 14.4). The obtained network topology is a direct graph based upon the monodi-
rectional links.
3. X is within the transmission area of Y, and vice versa. X and Y are mutually reach-
able via a wireless bidirectional link (e.g., see hosts B and C in Figure 14.4). De-
pending on the coding techniques and channel bandwidth allocated for the physical
channel, it may happen that the bidirectional link is not a symmetric link. A bidirec-
tional link is symmetric if the physical channel capacity (i.e., the maximum bit rate
obtained for wireless transmission) is the same for both link directions (otherwise it
is asymmetric). Many simulation models usually assume bidirectional and symmet-
ric links, for ease of implementation. The assumptions about these scenarios may
severely influence the modeling and performance results in the evaluation of net-
work protocols, for example, in routing protocols and multihop communication.

Dealing with discrete, event-based simulation, the critical question related to MANET
topology management in the simulation process is what is the simulated time of next link-
state-change event that will be expected, given the current relative mobility pattern and
coverage areas of mobile hosts? The answer to this question would require a little more
computation, based on the model and data structures defined to implement the simulation.
Every link-state-change event can be calculated, based on current coverage and mobility
conditions, and its execution scheduled in an ordered event list. Any intermediate change
in the speed, direction, or transmission power of any one of the involved hosts would re-
quire us to delete the causally dependent, scheduled link-state events and substitute them
with the updated ones. Moreover, this event-list management is at the basis of any discrete
event simulation. This indicates clearly how complex and computationally hard the mobil-
ity and link management in MANETs can be.
Once the link existence is established, many other conditions of the high-level link
properties should be managed. As an example, in wireless physical channels it is not pos-
sible to receive a communication on the channel while a simultaneous transmission is per-
385
14.2 DESIGN AND MODELING OF WIRELESS AND MOBILE AD HOC NETWORKS


formed on the same physical channel [24]. A bidirectional (full duplex) link can be ob-
tained by adopting time-division duplex (TDD) or frequency-division duplex (FDD).
TDD consists in splitting the transmission and reception phases over adjacent, non-over-
lapped time intervals on the same physical channel. FDD consists in adopting two physi-
cal channels: one for transmission and one for reception. In MANET modeling and simu-
lation, time-division duplex is commonly adopted. All data transmissions and receptions
have to be in the same frequency band, since there are no “bridge” nodes (perhaps except
base stations) to translate the transmissions from one physical channel to another one.
This usually requires strict time synchronization in the system, and Medium Access Con-
trol (MAC) protocols definition [24]. Frequency-division duplex may be adopted (togeth-
er with TDD) in centralized networks (like cellular) characterized by up-link and down-
link channels [24].
The coverage area management in the model can be used to simulate additional MAC
level details relevant for MANETs, such as collision domains. A collision domain can be
defined as the coverage area shared by a set of MHs mutually connected by a single
shared communication channel (i.e., a single logical channel). Collision of concurrent sig-
nals transmitted on the same collision domain would cause a destructive interference for
detected signals on the receiver. The main task of a MAC protocol policy is to avoid such
collisions, mainly by avoiding the start of new transmissions while another transmission is
being detected. A detection-based policy for a MAC implementation may result in some
problems whose investigation would require an accurate coverage-model definition. As an
example, let us suppose A is within the detection area of B and vice versa, B is within the
transmission area of C and vice versa, and C is outside the detection area of A (see Figure
14.4). In this scenario, A senses the transmission of B, that is, it senses the channel as
busy, but it cannot decode the transmission. This condition is often modeled in order to
obtain a real performance investigation about exposed terminals [24]. Exposed terminals
are terminals, (e.g., A) whose transmission is not allowed (e.g., by a MAC policy over a
collision domain) due to exposure to irrelevant transmissions, (e.g., B to C). A similar
problem is given by hidden terminals: due to shadowing effects and limited transmission
ranges, a given terminal C could start a transmission toward another terminal B (C and B
are within each other™s transmission area; see Figure 14.4) while B is receiving signals
from a hidden (with respect to C) terminal A. This means that B cannot complete any re-
ception, due to the destructive collision of signals from A and C. It may also happen that
B can detect and isolate one of the colliding transmissions, (e.g., from C to B). In this
case, we model a capture effect of transmission from C to B, despite A™s interference and
collision. A discussion of details for hidden and exposed terminals and modeling of cap-
ture effects can be found in [24, 52]. A rough modeling, based on a shared, global,
Boolean variable, Channel = Busy or Idle, would not describe with required accuracy the
previous scenarios.
This discussion was given to illustrate how, in the simulation plan, the definition and the
structure of the coverage area, topology, and interference models may include the informa-
tion required to perform a realistic simulation. Anyway, detailed models are quite complex,
and simplifying techniques and assumptions are widely adopted. In the following, we are
going to describe another relevant characteristic of MANET and wireless network models
adding additional complexity to the model definition and management: host mobility.

14.2.1.5 Mobility Models. User mobility is the main added value of wireless net-
works. Accurate simulation results would require accurate details to be modeled, and
386 SIMULATION AND MODELING OF WIRELESS, MOBILE, AND AD HOC NETWORKS


many fine-grained, low-level causal effects to be taken into account in the simulation
process. Mobility has a central role, and is a relevant background effect to be modeled in
almost every simulation analysis of wireless systems (see Figure 14.1). The effect of mo-
bility on the system policies and protocols is relevant at many layers: dynamic topologies,
due to simulated hosts™ mobility, map causality effects in the “areas of influence” of each
mobile device, resulting in dynamically shaped causality domains [6, 24]. The effect of
mobility introduces adaptive behaviors of users, protocols, and applications. Moreover, it
may happen that mobility models are related to the physical scenario under consideration
[55, 60]. Mobility patterns may sometimes be application dependent [13, 55]. Most of
Medium Access Control, Routing, and Transport protocols proposed for MANET scenar-
ios are customized and designed for specified mobility models, and behave better than a
general-purpose protocol for that given scenario. The evaluation of user positions can be a
computationally relevant task in a wireless mobile system™s simulation, due to the mobili-
ty and high number of events related to the user position. In two or more neighbor hosts
simply sharing the wireless medium (without any end-to-end communication session on),
the causal effect of signal interference, due to mobility, could result in a chain of local-
state events from Medium Access Control (MAC) up to the Transport and Application lay-
ers [23, 59]. In MANETs, given the infrastructureless architecture, some of the mobility
models adopted for cellular systems are not appealing. As an example, Markov models
(random walks) described by cell-to-cell migration probabilities, or fluid-flow models,
whose characteristic is to describe host mobility in terms of “the mean number of users
crossing the boundary of a given area,” are not considered as relevant as random- and re-
stricted-mobility models, gravity models, or group-mobility models.
In general, two types of mobility models can be adopted in the simulation of wireless
mobile networks, and specifically MANETs: motion traces and synthetic models [13].
Motion traces provide accurate and realistic information about user mobility patterns
and behavior, in particular when user mobility is related to real users in a bounded sce-
nario (e.g., downtown streets, highways) [3, 13, 55, 60]. Unfortunately, traces require
large log files, depending on the number of tracked hosts and the time granularity of sam-
ples. Traces are significant descriptions about the steady-state mobility of a user only if
the motion samples are collected for significant time intervals. If the sample frequency is
low, approximated solutions (e.g., interpolation and dead reckoning) can be used, but this
requires additional computation, and may result in strange behavior (like users walking
through obstacles instead of turning around them). Moreover, traces can be collected only
for existing systems, and MANET traces are still hard to find mainly because large
MANETs scenarios have still to be implemented and user applications have to be defined.
Motion traces solve the problem of defining the initial and run-time position distribution
of MHs in a deterministic way. Another interesting characteristic of motion traces is their
ability to capture the real correlation effect between user mobility and real application/
user needs. It may happen that user movement is driven by application needs, for example,
in order to reach good coverage areas. Also, it may happen that users move by showing a
correlated group behavior [13, 28, 61]. Synthetic models trying to define a similar corre-
lation effect, for single MHs and MH groups, will be defined in the following.
Synthetic models are defined to represent the mobility of users in a realistic way, with-
out using traces. Many synthetic models have been defined in the past to be adopted as
analytical models [3, 13]. The main requirement for such analytical models was mathe-
matical tractability instead of realism. Such models also survived in many simulation
studies, mainly due to validation possibilities they offer with respect to the simulation
387
14.2 DESIGN AND MODELING OF WIRELESS AND MOBILE AD HOC NETWORKS


counterpart [5]. In other scenarios, random-motion models far from reality can be adopted
in order to stress a given mechanism or protocol, emphasizing worst-case scenario results
[13, 27]. Random-motion models have been recently extended by introducing correlation
effects, restrictions, and group behaviors, in order to meet the requirements of mobile sys-
tems™ modeling and simulation [3, 13]. We can distinguish between three degrees of “ran-
domness” in the classification of random models [3, 5]: (1) models that allow users to
move anywhere in the simulation area, following pseudorandom processes to select speed
and direction, (2) models that bound the movement of users (like streets, walls, etc.) but
still allow for pseudorandom selection of direction and speed at crossings (like City Sec-
tion and Manhattan Model [42]), and (3) models based on predefined paths (deterministic
paths).
In the following, we define and discuss a list of synthetic models used for MANET
simulation.

Random Mobility Model. This is a discrete interpretation of the Brownian motion

<<

. 73
( 87 .)



>>