
1. Introduction
Opportunistic networks (OppNets) (Conti et al., 2010)are
emerging paradigms of human-associated ad hoc networks in
which mobile users interact with each other based on their
geographical proximity. The communication in OppNets is per-
formed in a peer-to-peer fashion using short-range and low-cost
mobile devices (such as smartphone and tablet) via Bluetooth or
Wi-Fi technologies. In this setting, humans are the main carriers of
mobile devices and hence, mobility of devices mirror movement
patterns of their owners. This raises the problem of how to
generate realistic human mobility traces in order to evaluate the
performance of networking protocols in OppNets accurately.
The first generation of networking protocols in traditional
mobile ad hoc networks was mainly evaluated using synthetic
movement models, such as random way point (RWP) (Bettstetter
et al., 2003) or random walk models such as Brownian motion
(Groenevelt et al., 200 6). However, several research efforts such as
Jungkeun et al. (2003) validate that human mobility is rarely
random and random models often fail to analyze the performance
of encounter-based protocols in OppNets accurately. However, it
should also be noted that human movement and random walks
contain some statistical similarities (Injong et al., 2011).
In reality, human mobility is strongly dependent to users'
personal and social characteristics and behaviors as well as
environmental parameters (Aschenbruck et al., 2011). For instance,
mobile carriers which are attracted to specific locations or indivi-
duals, may have significant correlation between their respective
localization constrains and movement patterns. These different
dimensions of human mobility patterns have been characterized
in the recent studies. For example, an experimental analysis in
Phithakkitnukoon et al. (2012) demonstrate that users frequently
visit the locations with which they have strong social ties.
Furthermore, mobile users tend to visit just a few locations, where
they spend the majority of their time (Song et al., 2010a). In most
cases, they often travel over short distances and rarely migrate
long distances (Gonzalez et al., 2008).
Characteristics of human movement can be explored in three
main categories: spatial, temporal, and connectivity. The spatial
features refer to the trajectory patterns in physical space. Temporal
aspects are related to the time-varying features of user mobility,
whereas connectivity properties concern the contact information
of users. Quite recently, several statistical analysis have been
carried out in an effort to better comprehend the properties of
human mobility and uncover hidden patterns. Comparatively, it
can be seen that there is not a general consensus on the
characteristics of human mobility, even on some fundamental
features such as the distribution of travel distance. Consequently, it
is of paramount importance to obtain insight into these attributes
and study the latest findings in this area.
Considering the characteristics of mobility features, an appro-
priate trace(s) and model(s) for the simulation and evaluation of
protocols in OppNets should be wisely selected. As several traces
and models that have been proposed are highly similar in nature,
it serves a great purpose to have a clear comprehension of the
plethora of existing models and data sets. Broadly, human mobility
datasets can be obtained in two main methods: realistic and
simulation-based models. Majority of the real traces have been
registered in bounded environments such as campuses and con-
ferences using Bluetooth or Wi-Fi technologies. In order to gen-
erate large scale traces, mobility information has also been
acquired using location-based social networks. The simulation-
based mobility models are alternative approaches to generate
mobility traces synthetically. The main motivation to employ such
mathematical methods is to generate scalable and flexible mobility
traces. However, the key statistical properties of simulation-based
models can be validated using real traces. Despite the fact that the
simulation-based models have some spatio-temporal and connec-
tivity dependencies, they can be re-parameterized to be applicable
for various scenarios in OppNets.
Human mobility prediction is another challenging issue that
has attracted significant attention recently. Undoubtedly, forecast-
ing users' future walks, stay duration and contact properties, based
on their mobility characteristics and history has many applications
in OppNets. For example, it is remarkably important for an
opportunistic data forwarding algorithm to predict the next venue
a mobile user will visit, her stay duration and the even the
individuals she will contact. By forecasting human mobility,
networking protocols can take advantage of the expected informa-
tion in order to streamline the performance of the algorithms
significantly.
In this paper, we study recent solutions for human mobility
challenges in OppNets with respect to three major aspects: human
movement characteristics, mobility models and prediction meth-
ods. Firstly, we categorize the fundamental features of human
mobility along the three aspects of spatial, temporal, and con-
nectivity properties. Secondly, commonly used real mobility traces
which have been captured using Bluetooth/Wi-Fi technologies and
on-line location-based social networking services are summarized.
We also present a thorough survey on recently proposed
simulation-based human mobility models for OppNets. Thirdly,
we categorize human mobility prediction methods into three
classes and explore some new techniques in each class. Based on
our discussion, we point out some improvements that can be
made in the different aspects of human mobility models.
The three topics we study in this paper are closely related to
each other. Analyzing different characteristics of human mobility
(such as travel distance, contact time) could result in useful
indicators and metrics. These measurements are of significant
value to uncover meaningful mobility patterns and also to validate
available mobility traces. On the other hand, spatio-temporal and
contact characteristics of human mobility can be used as useful
estimation criteria to predict users' future trajectories.
The remainder of this paper is organized as follows. Section 2
provides an overview on the related work. Section 3 covers
definitions and terminologies related to characteristics of human
mobility. Human mobility models which are categorized into two
major classes: trace-based models and simulation-based mobility
models are introduced in Section 4. Recently proposed human
mobility prediction methods are presented in Section 5. Some
major open research issues are presented in Section 6. Section 7
concludes the paper.
2. Related work
There have been some general surveys as well as a few specific
surveys for human mobility models. Camp et al. (2002) study
mobility models for ad hoc networks in two categories: entity
models and group models. In addition, they provide a performance
evaluation concerning the impact of the different models on
multi-hop routing protocols. However, this work mainly focus on
random mobility models which are not appropriate movement
models for human motion.
Musolesi and Mascolo (2009) categorize human mobility mod-
els into real world traces and synthetic mobility trace and study
advantages and disadvantages of both types. They also, for the first
time, introduce the concept of social networks into mobility
models. Similarly, Aschenbruck et al. (2011) provide a survey of
real world and simulation-based traces as well as synthetic
mobility models for multi-hop wireless networks. The focus of
this paper is on mobility traces/models that include position
P. Pirozmand et al. / Journal of Network and Computer Applications 42 (2014) 45–5846