1

Source Location

privacy in wireless

sensor networks using

data mules

2

Chapters

1. Abstract 3

2. Introduction 4

2.1. Source-Location Privacy 5

2.2. Why Source-Location Privacy? 5

3. Related Works 7

4. Solutions for providing source location privacy 9

5. System Model 12

6. Preliminaries 12

7. Attack Model 12

7.1. α-Angle Anonymity 13

8. Mules-Saving-Source Protocol(MSS) 14

9. Direct Delivery Protocol (DD) 16

10. Proposed Model: MDD Protocol 18

10.1. Working of MDD Protocol 19

11. Experiments and Results 22

11.1. Proposed Simulation Model 23

11.2. Results 27

11.3. Average time delay of MDD Protocol 36

12. Comparison with Direct Delivery Protocol 36

12.1. Result analysis of DD and MDD Protocol 37

12.2. Advantages of MDD protocol 38

12.3. Disadvantages of MDD protocol 38

13. References 39

3

1. Abstract

Wireless sensor networks (WSNs) have many promising applications for monitoring critical

regions, such as in military surveillance and target tracking. In such applications, privacy of the

location of the source sensor is of utmost importance as its compromise may reveal the location

of the object being monitored. Traditional security mechanisms, like encryption, have proven to

be ineffective as location of the source can also be revealed by analysis of the direction of traffic

flow in the network.

In this paper, we investigate the source-location privacy issue and discuss the attack model as

semi-global eavesdrop attack model being more realistic than the local or global eavesdropping

attack model. Additionally, we adapt the conventional function of data mules to design a new

protocol for securing source location privacy called the Modified-Direct-Delivery (MDD)

protocol and analyze its capabilities, limitations and drawback against two other protocols called

the Mules-Saving-Source (MSS) protocol and Direct-Delivery (DD) protocol.

We analyze the delay incurred by using data mules in our protocol and examine the association

between privacy preservation and data delay in our protocol through simulation.

Keywords: source location privacy, data mules, wireless sensor networks, mules saving source

protocol, direct delivery protocol

4

2. Introduction

In recent years, WSNs have played an important role in a number of security applications, like

remotely monitoring objects etc. In such applications, the location of the monitored object is

tightly coupled with the sensor that detects it, called the data source. Therefore, preserving the

location of data source is important for protecting the object from being traced. Such a

preservation cannot be simply accomplished by encrypting the data packets as the location of the

data source can be disclosed by analyzing the traffic flow in WSNs.

There have been extensive techniques proposed to preserve source-location privacy against

different attack models:

Local-eavesdropping model - Local-eavesdropping assumes the attacker’s ability to monitor the

wireless communication is limited to a very small region, up to very few hops.

Global-eavesdropping model - The attacker is assumed to be capable of monitoring the traffic

over the entire network.

Both being unrealistic, because the former stringently restricts the attacker’s ability, while the

latter exaggerates it, considering resources and cost required for launching such an attack.

Semi-Global eavesdropping model - A more practical attack model, in this semi-global

eavesdropping model, the attacker is able to eavesdrop on wireless communications in a

substantial area that is much smaller than the entire monitoring network. This attack model

allows the attacker to gather substantially more information than a local eavesdropper.

Under the semi-global eavesdropping model, we explore a novel protocol for preserving source-

location privacy by using data mules. Traditionally, data mules are used in WSNs for reducing

energy consumption due to the data transmission between sensors and facilitating

communication in disconnected networks. A data mule picks up data from the data source and

then delivers them directly to the base station. We adapt the functionality of data mules so that

they not only maintain their traditional functionality, but also facilitate the preservation of the

location privacy of data sources.

5

2.1. Source-Location-Privacy

The problem of preserving source-location privacy can be explained using the “Panda Hunter

Game”, in which the sensors are deployed in the forest to monitor the movement of pandas. Each

panda is mounted with an actuator which signals to the surrounding sensors in its communication

range. When the sensor close to the panda receives the signal, it creates and sends data reports to

the base station over the wireless network. A hunter who is monitoring the wireless

communication between the sensors will be able to identify the direction of incoming traffic flow

and trace back the data transmission path to locate the data source, thus catching the panda. In

fact, any WSNs used for such monitoring applications are vulnerable to such kinds of traffic

analysis based attacks.

2.2. Why source-location-privacy?

In a wireless sensor network, location information often means the physical location of the event,

which is crucially given some applications of wireless sensor networks. So if an attacker gets

location information by analyzing a message that was captured, he will move to the location and

monitor the event. Meanwhile, the attacker will collect a lot of private information. So the

information retrieved by these networks is of vital importance and must be properly secured not

only from curious eavesdroppers but also from more skilled adversaries. Messages traversing the

network can be protected using traditional confidentiality and integrity mechanisms. But, even if

an adversary cannot obtain the information contained in the payloads, he can still retrieve other

sensitive information by observing and analyzing the communications. For example, an attacker

can obtain the information from the network and the environment being monitored by simple

observation of the network traffic. Besides, an attacker can compromise users’ location privacy

by observing the wireless signals from user devices

Although many existing privacy techniques can be employed in sensor network scenarios, they

cannot effectively preserve the sensor location in a sensor network. The reason is that the

problems are different in fact and many of the methods introduce overhead which are too

6

burdensome for sensor networks. And many techniques do not consider the capacity, computing

power, and power of sensors, which are the limiting factors in wireless sensor networks. And

some techniques analyze privacy and anonymity issues and propose solutions by manipulating

the message contents. In contrast to their schemes, this paper addresses the location privacy

threat due to the physical wireless medium that allows the adversary to perform traffic analysis

to derive the message flows.

In wireless sensor networks, minimization of energy consumption is considered a major

performance criterion to provide maximum network lifetime. Ant colony optimization

algorithms simulating the behavior of ant colony have been successfully applied in many

optimization problems such as vehicle routing and the asymmetric traveling salesman as well as

routing in wireless sensor networks

7

3. Related Works

In wireless sensor networks, it is important to provide confidentiality to the sensor’s location. In

this section, we describe previous proposed technologies that were designed to preserve the

source location in wireless sensor networks.

For a more comprehensive taxonomy of techniques of preserving privacy in WSNs, readers may

refer to the state-of-the-art survey

Fan et al. [02] preserve location privacy by using homomorphic encryption operations to prevent

traffic analysis in network coding. In [03], each cluster header can filter the dummy packets

received from the sensor nodes of its cluster to reduce the number of dummy packets. However,

the scheme requires much computation overhead due to using asymmetric-key cryptography, and

the packet delivery delay is long because the cluster header sends packets with a fixed rate

regardless of the number of events it collects.

Mehta et al. [04] formalize the location privacy problem using a global adversary model and

compute a lower bound for the overhead required for achieving a given level of privacy

protection.

The proposed scheme by Alomair et al. [05] can guarantee event indistinguishability by achieving

interval indistinguishability, where the adversary cannot distinguish between the first, the

middle, or the end of the interval.

In [06], dummy packets can be filtered at proxy nodes, and the lifetime of the WSN is analyzed at

different proxy assignment methodologies. Hong et al. [07] propose a scheme that can thwart time

correlation attack. In this attack, the adversary exploits the time correlation of transmissions in

successive links to learn the end-to-end route.

Zhou and Yow [08] propose an anonymous geographic routing algorithm which includes three

components to avoid the explicit exposure of identity and location in communication.

For local-eavesdropping based attack, flooding based approach was first introduced in [10], where

each sensor broadcasts data that it receives to all its neighbors. However, this technique suffers

from high communication overhead for sensors.

In [09], each data packet is first relayed to a randomly selected intermediate sensor in the network

and then is forwarded towards base station along the shortest path.

8

In [01], FitProbRate is proposed to maintain source anonymity, which is an exponentially

distributed dummy traffic generation scheme. The Fitprob parameter decides the dummy traffic

generated at a dynamic rate, which differs from other similar works. It is a great improvement

over source simulation and fake sources but still has the drawback of having overhead due to

dummy packet generation.

9

4. Solutions for providing source location privacy

1. Random Walk: The aim of the “random walk” approach is to have packets follow a random

route through the network. The random walk should make a packet’s path look completely

random to an adversary in order to counter the adversarie’s traffic analysis and hop-by-hop

traces. Solutions in this category use either a technique derived from the random walk, as

described by Ozturk et al[10], or a technique that results in a similar pattern, such as rumor

routing from Braginsky et al and routing through randomly selected intermediary node from

Li et al.

The following solutions are part of this category: angle based multi-intermediate nodes

selection ,the directed random walk, the greedy random walk, the location privacy support

scheme, the mules saving- source protocol, opportunistic routing, phantom routing, phantom

routing with location angle, phantom single-path routing, random routing, the random routing

scheme, routing through randomly selected intermediary node, the self-adjusting directed

random walk, and a combination of different solutions.

2. Geographic Routing: Solutions in this category use the physical location of the nodes

together with geographic routing algorithms to route packets through the WSN. Geographic

routing algorithms take the position of a node, its neighbor’s, and the sink into account, in

order to route a packet from the source to the sink. The solutions in this section make use of

additional methods, such as the usage of synonyms, encryption, and random intermediary

node selection to hide the flow of the traffic against a local adversary.

3. Delay: This category consists of solutions that alter the flow of the traffic as follows. Each

node buffers incoming packets and holds a packet for a random time before forwarding. As a

result, nodes alter the chronological order of the packets: they send the packets in a different

order than how they received them. As the chronological order of the packet change, so does

the traffic pattern. The change of the traffic pattern makes it hard for a local adversary to

track the traffic to the actual source.

10

4. Using dummy data sources: Solutions in this category introduce dummy traffic to alter the

real traffic. The aim is that an adversary should no longer be able to see which part of the

traffic is real, and which part is fake. In this category, we found the following solutions :

aggregation-based source location protection scheme, a real and a fake cloud-based scheme

for protecting source location privacy, constant rate , the dynamic bidirectional tree,

distributed resource allocation algorithm, dummy wake-up scheme , fake sources 1 and fake

sources 2, fitted probabilistic rate , the group algorithm for fake-traffic generation, globally

optimal algorithm, the heuristic greedy algorithm , mixes , the optimal filtering scheme,

periodic collection, persistent fake source routing, the probabilistic algorithm , proxy-based

filtering , SECLOUD, short-lived fake source routing, source simulation , the timed efficient

source privacy preservation, the timing analysis resilient protocol, tree based filtering ,the

trusted computing enabled heterogeneous WSN , unobservable handoff trajectory , and the

zigzag bidirectional tree.

5. Cyclic Entrapment: The solutions in this category aim at confusing the adversary by

shaping the traffic between nodes in cyclic patterns. A local adversary, who tracks the traffic

between the nodes, will travel in circles without finding the actual source. This category

consists of two solutions: cyclic entrapment method and information hiding in distributing

environments.

6. In Network Location Anonymization: Solutions in this category hide either the identity or

the location of a node. The following solutions are part of this category: the anonymous

communication scheme, anonymous path routing, the cryptographic anonymity scheme,

destination controlled anonymous routing protocol for sensor nets, hashing based ID

randomization, max query aggregation, phantom ID, the probabilistic destination controlled

anonymous routing protocol for sensor nets, the reverse hashing ID randomization, and the

simple anonymity scheme.

7. Cross-layer Routing: With cross-layer routing, the nodes use the beacon frames, which are

normally only used for network maintenance, to share information on sensed events. Local

adversaries, which normally only listen to the network level packets, miss part of the

11

information exchange, and do not find the real source. This category has two solutions: the

cross-layer solution and the double cross-layer solution.

8. Separate Path Routing: A local adversary often needs multiple packets along the same

route to track the actual source. The solutions based on separate path routing make sure that

the packets travel via different nonintersecting paths from source to sink. Using separate

paths leads to fewer packets per path, which delays the local adversary in its tracking, or even

makes the adversary unable to track the actual source at all. This category consists of random

parallel routing, weighted random stride routing, and weighted random stride routing towards

a global viewing adversary.

9. Network Coding: In network coding, each node cuts up its message and sends it out in

smaller pieces. These pieces are then forwarded via different routes towards the sink. This

category consists of the solutions from Fan et al as they propose to use network coding to

provide SLP.

10. Limit Node Delectability: This category consists of solutions that limit the transmission

power of the nodes to make them harder to detect. We have identified the following solutions

in this category: anti localization by silencing, context-aware location privacy, hidden

anchor, hyberloc, lowering radio transmission power, and multi cooperator power control.

From the above mentioned techniques we will be presenting two already developed privacy

protocols and then propose the modified protocol for privacy in wireless sensor networks.

12

5. System Model[10]

The terrain of our underlying network is a finite two-dimensional grid, which is further divided

into cells of equal size. The network is composed of one base station, static sensors, and mobile

agents, called data mules.

Static sensors - All static sensors are homogeneous with the same lifetime and capabilities of

storage, processing as well as communication. They are deployed uniformly at random in the

cells, and assumed to guarantee the connectivity of the network.

Data mules - Data mules are the mobile agents which can be artificially introduced in the

network [10]. We assume they move independently and do not communicate with each other.

Also, they are assumed to know their own locations when they are moving all the time. Their

mobility pattern can be modeled as a random walk on the grid, whereby in each transition it

moves with equal probability to one of the horizontally or vertically adjacent cells. After a data

mule moves into a cell, it stays there for tpause time period before its next transition.

At the beginning of the pause interval, the data mule announces its arrival by broadcasting Hello

Message. Only data source will respond and relay buffered data to the data mule. We assume the

data mule does not communicate with sensors when moving. The data mule’s communication

range is larger than that of a sensor, thus a data source which cannot directly transmit data to

the data mule will use multi-hop routing.

6. Preliminaries

In this section, we will first introduce our attack model and then propose a linear-regression

based approach for analyzing data traffic. Furthermore, we will demonstrate the effectiveness of

out attack model by compromising the phantom routing protocol [11]. Finally, we will define the

α-angle anonymity model for studying the location privacy preservation of data source.

7. Attack Model

We assume the attacker is capable of launching only passive attacks, in which he can only

monitor the traffic transmission but not decrypt or modify data packets. Suppose the attacker

monitors the radio transmissions between sensors in a circular area of radius Ratt. Larger the

13

monitoring area, stronger the attacker. If the monitoring area is large enough to cover the whole

network, it is global eavesdropping; on the other hand, if the area is limited only to a few hops, it

is local eavesdropping. However, we define semi-global eavesdropping as whose strength lies in

between the two extreme attack models. In addition, we believe without any prior knowledge of

source location, the attacker is inclined to launch the attack by collecting traffic data from around

the base station. Intuitively, since the whole network traffic converges to the base station, it

serves as the ideal point for starting the attack. Admittedly, the attacker can make an initial

estimation of the direction of data source and move in that direction. Meantime, he can keep

updating his estimation with more traffic observed as he moves, until he finds the data source.

However, in this paper we aim to discourage the attacker even from making a good initial

estimation before he starts moving.

Traffic flow in phantom routing

7.1. α-Angle Anonymity[10]

In order to anonymize source location privacy under semi-global eavesdropping attack, we

introduce α-angle anonymity model. This model ensures the preservation of source location

privacy by enlarging the inference space from which the attacker estimates the real direction of

the data source. The inference space is determined by the system variable α. The value of α can

be open to the public, even including the attacker, however, this should not threaten the privacy

14

of source location. According to the definition, we can see that the larger the value α, the larger

the inference area. The shaded area in Fig. 2 represents the attacker’s inference space. Given a

larger inference space, the attacker cannot deterministically estimate the real direction or location

of data source, thereby the source location privacy being preserved.

A protocol is α-angle anonymous if the real direction of data source is equally likely distributed

in the angle range [β − α, β + α], where β is the angle of the direction inferred by the attacker

based on his observation.

8. Mules-Saving-Source Protocol(MSS) [10]

To protect the source location privacy against a semi-global eavesdropper, we design a protocol,

called Mules-Saving-Source protocol, achieving α-angle anonymity.

Our protocol exploits the random mobility of data mules to establish a data transmission pattern

which effectively preserve the location privacy of data source. Specifically, we modify the

traditional function of data mules by having them hand data to regular sensors at only specific

locations in the network, from where data will be further routed towards base station along the

shortest paths[10]. The specific sensors will be selected so as to bias the direction of composite

traffic to be derived by the attacker based on data transmission he observes around base station.

In fact, solely allowing data mules to directly deliver data to base station can thoroughly preserve

source location privacy against a semi-global eavesdropper. This is because the data transmission

between data source and base station is completely hidden by the random movement of the data

mules which ferry data. However, its disadvantage is the non-trivial delay caused by data mules,

which may not be tolerable especially in large-scale wireless sensor networks.

In this section, we first describe our protocol and then prove it to be α-angle anonymous. Note

that we predefine a coordinate system with the base station as the origin, which is assumed to be

known by data mules. Our protocol includes three phases:

15

I. picking a fake direction at source,

II. carrying and unloading data by data mules, and

III. routing data to the base station

Phase I. Picking a fake direction at source - When a target is detected by the sensors, they

coordinate among themselves and let the one closest to the target become the data source. The

coordination protocol has been well studied in literature [13] and its discussion is out of the scope

of this paper. The data source periodically generates and sends data reports towards base station.

Additionally, it generates a value of β as the fake direction of data source to be used for biasing

the attacker’s observation in the traffic flows coming towards base station. Specifically, the data

source selects β from the range [θ − α, θ + α] uniformly at random, where θ is the absolute angle

between the direction of data source and the direction of x-axis in our coordinate system, and α is

a value preset to configure the privacy preservation level. The β angle is known only by the data

source[10] initially.

Phase II. Carrying and unloading data by data mules - When a data mule moves into a cell, only

the data source in its communication range responds with the buffered packets. Along with the

data, the data source also sends the value of β angle the data mule. After getting the data, the data

mule roams around the network until reaching certain location, called dropping point[10].

Dropping point is referred to as any point located on the dropping line drawn from base station at

an angle β in the coordinate system. Upon arriving in a cell intersecting with the dropping line,

the data mule unloads the data to the sensor closest to the dropping line present within the cell.

Phase III. Routing data at sensors - After data packets are offloaded to a sensor by the data

mule, they are routed towards base station along the shortest path. Ideally, the transmission path

is along the dropping line. However, due to the nonlinear multi-hop routing[10], data transmission

16

may have trivial deviation from the dropping line, which should not affect the privacy

preservation. One can see the traffic flow will go towards base station roughly along the

direction with a β angle, thereby successfully biasing the attacker’s inference of data source

direction.

9. Direct Delivery Protocol (DD)

Mule Saving Source protocol hand over the data to regular sensors at only specific locations in

the network, from where data will be further routed towards base station along the shortest paths.

But in Direct Delivery protocol the data is collected by the data mules and is directly delivered to

the basestation. In this protocol, the data delivery performance is not satisfactory and the delay of

delivery is usually very long than mule-saving-source protocol.

This protocol works in a simple and single module i.e picking up the data from the sensor nodes

and delivering it directly to the basestation or the sink. The main advantage of the protocol is DD

protocol guarantees the complete preservation[10] of the location privacy of data source, however

the disadvantage is the high delay[10], as compared to MSS protocol.

According to the attack models[10] we have discussed in this paper, the traffic in the network can

be used to detect the source node, however with direct delivery protocol the data is carried by the

data mules and their path are random, and they collect data from the source and deliver it to the

sink when in range. This protocol guarantees complete security because traffic analysis[10]

cannot be performed over the network when data mules carry the data from the source to the

sink.

In direct delivery protocol, the performance of the protocol can be improved with the

improvement in mobility pattern algorithm used for each data mule.

The path vector can be altered or improved by considering the following factors:

(1) Path selection: which trajectory the data mule follows

17

(2) Speed control: how the data mule changes the speed while moving along the path

(3) Job scheduling: from which sensor the data mule collects data at each time point

Path selection is to determine the trajectory of the data mule in the sensor field. To collect data

from each particular sensor, the data mule needs to go in the sensor’s communication range at

least once.

Speed control is to determine how the data mule changes its speed along the chosen path. The

data mule needs to change the speed so that it stays within each sensor’s communication range

long enough to collect all the data from it.

Job scheduling is done once the time-speed profile is deter- mined, we get a mapping from each

location to a time point. Thus we get a scheduling problem by regarding data collection from

each sensor as a job. Each job has one or more intervals in which it can be executed. Job

scheduling is to determine the allocation of time to jobs so that all jobs can be completed.

However, there is a tradeoff between source privacy and time delay for data transmission from

source to the basestation or the sink. When privacy of source location is the primary objective we

need to deploy direct delivery protocol considering no constraint on time delay of data delivery,

in other cases, mule-saving-source (MSS) protocol delivers data with better efficiency than the

direct delivery protocol.

18

10. Proposed Model : Modified-Direct-Delivery (MDD) Protocol

The proposed solution for the source location privacy has the following elements, properties and

terminologies:

I. Sensor Field: Sensor field is a square block with dimension of 400m x 400m.

This area is the working field of sensors and the mules in the simulation.

II. Sensors: Sensors are homogeneous with the same lifetime and capabilities of storage,

processing as well as communication. They are deployed uniformly at random in the

cells, and assumed to guarantee the connectivity of the network.

III. Zones: Sensor field is divided into zones, each zone contain different number of

sensors and data mules. These zones play a major role in fast delivery of data in the

proposed protocol.

IV. Data Mules: Data mules are the mobile agents which can be artificially introduced in

the network. In this protocol, they move independently and do communicate with

each other by establishing an ad-hoc network when two mules are in transmission

range. They communicate to deliver data. They are assumed to know their own

locations when they are moving all the time. Their mobility pattern can be modeled as

a random walk on the grid. They may be equipped with GPS facility so that they can

calculate the relative location over the sensor field.

V. Base Station: Base Station or the sink is the central repository of the data transmitted

by source in the sensor field. The data collected by the data mules is delivered to base

station, which is then organized and analyzed for fruitful results.

VI. Active Sensors or Source Sensors: In a simulating environment, every sensor in the

field need not transmit data; sensors are chosen randomly that have data and they

19

needed to transmit it to the data mules. Such sensors are active sensors. They are

ready to transmit and can sense the presence of data mule.

VII. Ad-hoc Networks: These networks are established when two data mules either from

same zone or different are in transmission range. This network is used to transmit the

collected data from one zone to other and subsequently to the zone with basestation.

10.1. Working of MDD Protocol

To ensure source location privacy and ensure it with better performance i.e. with minimum time

delay we have designed the protocol assuming all sensor nodes deployed have same energy and

the load over the data mule is balanced. In this protocol we have used a major part of direct-

delivery (DD) protocol, the data from the source node is delivered directly to the basestation but

the technique and process of delivery of data is improved by introducing zones and ad-hoc

network in between the data mules. The inter mule communication improves the time delay in

data delivery while the direct delivery of data from source to the sink is guaranteed.

The working of MDD protocol starts with an active node or source node that have some data that

it sensed based on the type of sensor; it may be temperature sensing, motion sensing, heat

sensing or any other sense property.

In this protocol we have three types of static sensors based on their state:

I. OFF Sensors: These are the sensors that are in a state in which their radio are in

OFF_STATE; they cannot sense process or transmit data in this state.

II. SLEEP Sensors: Sensors have a lot of energy consumption while they are sensing,

processing or transmitting. To reduce the frequent use of energy, when no event is

20

scheduled; the sensors fall into sleep state for some time and wake up immediately

when event get scheduled. This saves energy.

III. ACTIVE Sensors: Sensors that are ready to sense or have sensed and need to

transmit data to the mules are in ACTIVE state. These sensor nodes are considered as

source nodes and their location is need to preserved.

However the sensors are deployed in a field that is further divided into zones. The data mules

present in zone can collect data from the same zone only. The data mule collects data from active

sensors and the data is retransmitted to the other data mule through ad-hoc networks established

when these agents are in transmission range of each other.

The data transmission between the mules carry data from one zone to other as data transmission

takes place with the mule from other zone closer to the base station. This process has a major

impact over the reduction in time delay of data transmission. This time delay can be further

reduced by introducing path vector algorithms that improve efficiency of data transmission by

altering the parameter and properties of the data mule.

The efficiency is improved by following factors:

(1) Path selection[12]: which trajectory the data mule follows

(2) Speed control[12]: how the data mule changes the speed while moving along the path

(3) Job scheduling[12]: from which sensor the data mule collects data at each time point

Path selection is to determine the trajectory of the data mule in the sensor field. To collect data

from each particular sensor, the data mule needs to go in the sensor’s communication range at

least once.

21

Speed control is to determine how the data mule changes its speed along the chosen path. The

data mule needs to change the speed so that it stays within each sensor’s communication range

long enough to collect all the data from it.

Job scheduling is done once the time-speed profile is deter- mined, we get a mapping from each

location to a time point. Thus we get a scheduling problem by regarding data collection from

each sensor as a job. Each job has one or more intervals in which it can be executed. Job

scheduling[12] is to determine the allocation of time to jobs so that all jobs can be completed.

The intercommunication between data mules is continued till the data is received by the mule in

zone containing the sink or basestation. The mule in this zone carries the data to the base station

which is then delivered and used for analysis purpose.

22

11. Experiments and Results

Modified-Direct-Delivery protocol is simulated and the results are obtained for analysis and

comparison purpose. The simulator used for the experiment purpose is OMNeT++. OMNeT++ is

a very efficient and powerful wireless sensor network simulator. OMNeT++ is an object-oriented

modular discrete event network simulation framework. It has a generic architecture, so it can be

(and has been) used in various problem domains:

Modeling of wired and wireless communication networks

protocol modeling

modeling of queuing networks

modeling of multiprocessors and other distributed hardware systems

validating of hardware architectures

evaluating performance aspects of complex software systems

in general, modeling and simulation of any system where the discrete event approach

is suitable, and can be conveniently mapped into entities communicating by exchanging

messages.

OMNeT++[13] itself is not a simulator of anything concrete, but rather provides infrastructure

and tools for writing simulations. One of the fundamental ingredients of this infrastructure is

component architecture for simulation models. Models are assembled from reusable components

termed modules.

Modules can be connected with each other via gates (other systems would call them ports),

and combined to form compound modules. The depth of module nesting is not limited. Modules

communicate through message passing, where messages may carry arbitrary data structures.

Modules can pass messages along predefined paths via gates and connections, or directly to their

destination.

OMNeT++ [13]also supports parallel distributed simulation. OMNeT++ can use several

mechanisms for communication between partitions of a parallel distributed simulation, for

example MPI or named pipes. The parallel simulation algorithm can easily be extended, or new

ones can be plugged in. Models do not need any special instrumentation to be run in parallel – it

23

is just a matter of configuration. OMNeT++ can even be used for classroom presentation of

parallel simulation algorithms, because simulations can be run in parallel even under the GUI

that provides detailed feedback on what is going on.

The OMNeT++ simulation IDE[13] is based on the Eclipse platform and extends it with new

editors, views, wizards, and other functionality. OMNeT++ adds functionality for creating and

configuring models (NED[13] and INI[13] files), performing batch executions and analyzing the

simulation results, while Eclipse provides C++ editing, SVN/GIT[13] integration and other

optional features (UML modeling, bug-tracker integration, database access, etc.) via various

open-source and commercial plug-in. The environment will be instantly recognizable to those at

home with the Eclipse platform.

11.1. Proposed Simulation Model

The proposed simulation model for analyzing the proposed modified-direct-delivery protocol we

have designed an event based simulation. We are especially interested in large-scale sensor

networks where there is a reasonably large separation between the source and the sink. The main

area of interest is the time delay in the delivery of the data from the source to the sink. From the

results obtained in the simulation i.e. the time between the creation of a message from a source to

the time of delivery to the sink is calculated.

In this model, the sensor field is divided into four different zones identified by four different

colors. The number of data mules deployed in each zone is based on the total number of sensor

nodes deployed in a zone i.e.

Number of Data Mules in a zone α Number of Sensor Nodes

Every sensor node has different states; to describe state of every sensor in the simulation

different colors are used.

I. Off State : Grey

II. Sleep State : Black

III. Active State : Red

24

Sleep time and Idle time are defined accordingly where Sleep time is the maximum time limit

when the sensor can go offline and the Idle time is the time limit taken by the sensor before

entering into the sleep time. To record the events generated through the simulation process

“record-eventlog = true[13]” is set.

Messages are created in forms of packets. Packet created by each sensor is forwarded

accordingly to mules which this then forwarded to other mules and finally to the basestation.

TransferMsg is the packet that contain numRecData that contain the timestamp of creation of the

packet by each sensor and then it is recorded throughout the mobility.

Following are the specification used in the simulation model for MDD protocol:

I. Sensor field dimension: 400m x 400m.

II. Number of Sensors used: Variable (50-300).

III. Number of Data Mules used: 8

IV. Range of each Data Mule: 50m

V. Simulation Time Limit: 200 sec

VI. CPU Time Limit: 1000 sec

VII. Sleep Time of Sensor Nodes: 120sec

VIII. Idle Time of Sensor Nodes :30sec

25

A Sensor Field with Data Mules, Sensors, Zones, Basestation

Creation of ad-hoc connection and transmission of data from one zone to other

26

Result obtained in the event log is displayed as in the figure below. Every sensor node transmit

some data in form of packets which is finally received by the basestation

Average Packet Transfer Delay is the average of source delay and carrier delay.

Source delay is the time between the creation of the packet and the time at which the packet is

delivered to the data mule in range. Carrier delay is the time between the time of packet received

by the mule and the time at which packet is received at the base station.

Average Packet Delay = Source Delay + Carrier Delay

Event log on completion of method - finish()

27

11.2. Results:

Number of Sensors in the simulation = 50 Number of Active Sensors = 29

Sensor ID [50] Time delay

9 0.371403

12 0.69743

14 52.3333

16 53.1111

20 0.314237

21 0.601479

25 0.297465

26 31

27 0.576412

28 0.866664

30 31

31 26.7775

33 12.981344

35 0.898583

38 0.650935

41 0.324784

Sum 212.802636

Average 13.30016475

Active Sensors 29

Number of Sensors in the Simulation = 100 Number of Active Sensors = 59

Sensor ID[100] Time Delay

11 0.44108

12 0.190687

14 13.3705

15 12.8797

20 0.206121

21 0.915134

22 0.604211

23 0.5144

0 10 20 30 40 50 60

9

14

20

25

27

30

33

38

Average

Time delay

Time delay

28

24 0.0952684

25 13.6327

26 14.7245

29 0.724588

30 0.747607

31 0.480331

32 18.45653

33 0.448478

34 33.7823

35 0.0483317

36 0.892575

43 0.759089

43 37.0345

43 18.9901

44 21.9895

45 0.248859

46 0.972774

52 0.13836

53 35.6679

54 0.0662034

55 0.442271

56 51.9416

61 0.978298

63 0.513878

64 54.4541

65 36.7167

74 0.308547

75 0.20511

76 2.1762

84 0.303599

85 0.310919

86 0.59431

95 0.254127

96 0.479442

Sum 378.7014285

Average 9.016700679

Active Sensors 59

0 20 40 60 80

11

14

20

22

24

26

30

32

34

36

43

44

46

53

55

61

64

74

76

85

95

Average

Time Delay

Time Delay

29

Number of Sensors in the Simulation = 150 Number of Active Sensors = 92

Sensor ID [150] Time Delay

14 0.81875

20 0.423232

21 0.746915

22 0.144944

23 26.8929

24 0.710244

25 0.334637

31 0.459475

32 0.584495

33 10.4623

34 24.8694

35 13.65949

36 12.86319

41 0.44637

42 0.930044

43 14.4535

44 0.472867

45 6.59396

46 0.556379

52 12.1956

53 0.660337

54 0.972362

55 0.278245

56 0.430616

61 0.493681

62 1.55513

63 46.6049

64 0.597943

65 0.552439

66 0.984333

71 62

72 0.522178

73 0.993358

74 83.7282

75 0.576774

76 0.174956

80 0.723687

82 0.239248

0 50 100

14

21

23

25

32

34

36

42

44

46

53

55

61

63

65

71

73

75

80

83

93

114

116

125

135

Avg Time Delay

Time Delay

Time Delay

30

83 0.264427

84 18.951

93 5.46409

105 43.5938

114 0.896177

115 0.534452

116 9.344324

121 3.132454

125 4.422342

126 0.00641925

135 0.345644

136 5.4232342

Avg. Time Delay 8.461708849

Active Sensors 92

Number of Sensors in the Simulation = 200 Number of Active Sensors = 140

Sensor ID [200] Time Delay

16 0.736903

17 0.196749

18 24.6814

19 0.42355

20 0.345234

22 0.45345324

27 0.2345223

28 0.457654

29 20.8814

30 52.4855

31 0.891192

32 0.610015

33 4.22606

34 3.435463

35 1.43534

39 0.546675

40 0.96785

41 0.756757

42 0.724689

43 20.8976

44 47.0378

45 2.2346712

31

46 53.8947

47 2.51829

49 0.58753

53 0.9422347

54 0.72842

55 1.823585

56 4.9185

57 0.132662

58 51.7726

59 8.19248

60 0.346248

61 0.974222

62 65.5214

65 0.927264

66 0.729562

68 0.8462841

69 0.200878

70 0.411577

71 53.4185

72 0.331979

73 0.522826

74 0.223097

75 40.0565

79 25.54924

80 0.4095

82 2.5984

84 0.769848

85 0.740943

86 0.571967

87 0.613141

88 60.8519

89 0.258981

90 0.21825

94 0.83465

97 0.883426

98 0.438728

99 0.694709

100 6.98392

101 0.76734

103 12.9582

104 6.23746

109 8.34872

110 2.87438

0 50 100 150

16

18

20

27

29

31

33

35

40

42

44

46

49

54

56

58

60

62

66

69

71

73

75

80

84

86

88

90

97

99

101

104

110

115

117

128

130

137

141

143

153

174

Avg Time Delay

Time Delay

Time Delay

32

114 0.131903

115 9.2938

116 0.101484

117 23.0955

125 0.235982

128 0.45735

129 0.925274

130 0.247636

133 0.83748234

137 3.4759231

140 0.892347

141 0.3487238

142 0.8748123

143 0.7734872

144 0.873453

153 3.4298351

155 0.95892341

174 31

186 0.353487

Sum 686.5929228

Avg Time Delay 8.173725271

Active Sensors 140

Number of Sensors in the Simulation = 300 Number of Active Sensors = 188

Sensor ID [200] Time Delay

6 0.854366

7 1.88643

11 0.844316

14 0.736903

15 0.543435

16 32.67454

17 0.42355

20 0.345234

21 0.733287

23 0.228765

24 0.643254

25 2.658814

27 5.764543

31 0.891192

32 1.6663243

33

33 24.22606

34 33.435463

35 1.43534

38 0.546675

39 0.96785

40 0.756757

41 0.567753

42 19.433786

44 45.96336

45 2.2346712

46 50.59347

47 2.51829

49 0.58753

53 0.763443

54 0.72842

55 1.0966543

56 4.877655

57 19.68678

58 23.9756477

59 8.19248

60 0.346248

61 0.974222

62 36.85521

65 0.927264

66 0.729562

67 0.8462841

68 0.200878

70 0.411577

71 53.4185

73 0.331979

74 0.522826

76 0.88601

77 16.833878

78 22.54924

79 0.7331

80 27.8333687

83 2.00135

84 0.740943

86 0.571967

87 0.613141

88 60.8519

89 0.258981

90 0.21825

0 20 40 60 80

6 14 17 23 27 33 38 41 45 49 55 58 61 66 70 74 78 83 87 90 93

101 109 115 120 123 130 136 139 145 150 160 167 171 174 199 219 234 249 254 269 298

Time Delay

Time Delay

34

91 0.83465

92 0.883426

93 0.438728

94 0.694709

99 6.8647

101 0.46431022

103 13.88542

104 16.2446

109 8.34872

110 2.87438

114 0.131903

115 29.2938

116 0.101484

117 23.0955

120 0.235982

121 0.45735

122 0.925274

123 0.247636

125 0.83748234

127 45

130 0.892347

131 0.3487238

135 0.8748123

136 0.7734872

137 0.873453

138 3.4298351

139 0.95892341

140 31

142 0.353487

145 7.59292279

146 16.62571583

148 12.883643

150 0.87346

155 0.6153332

157 0.773624

160 18.66496

161 22.009465

166 12.83465

167 0.48573

168 0.874754

170 20.736543

171 0.28929847

172 1.3765321

35

173 31

174 12.67527643

180 0.389463

192 0.98901

199 0.0347238

200 0.0874823

213 1.985793

219 1.84793

220 29.985794

229 2.9849

234 19.8465982

247 34.9934652

248 0.87642

249 0.7652734

250 0.99764852

252 0.50047332

254 0.87346549

260 0.46287475

261 0.92464875

269 8.7648583

270 0.747457683

278 0.48935752

298 1.3487572

Sum 984.7131429

Avg Time Delay 7.941235024

Active Sensors 188

36

11.3. Average time delay of Modified-Direct-Delivery Protocol:

No of Sensors Average Time Delay

50 13.30016475

100 9.0167003

150 8.4617088

200 8.173725

300 7.941235024

Average Time Delay of Modified-Direct-Delivery Protocol: 9.378706775

12. Comparison with Direct Delivery Protocol

Direct Delivery Protocol delivers packets directly to the basestation, so in this type of delivery

data mule have to carry data for a long time all through the path way till basestation is in range of

the mule containing the data. In this protocol, the probability of collision or carrying data from

same sensor node is more thus reducing the efficiency.

Certain improvements in this protocol with a better path selection and speed control with inter

mule communication brings out a lot of improvements in the average time delay of the protocol.

When two mules are interconnected they transmit the data as well in case of duplicate values of

packet id, collision is avoided. This reduces the probability of collision and thus improving the

efficiency of the protocol. Also, mule need not travel all through the path to the basestation for

transmission. Zones in the sensor field is a major improvement in the protocol and using zones

0 100 200 300 400

1

2

3

4

5

Average Time Delay

No of Sensors

37

maximum number of sensor nodes can be covered in much smaller time thus reducing the source

delay.

MDD Protocol have following improvements over Direct Delivery Protocol:

I. Introduction of zones: Mules can cover maximum number of nodes in small amount

of time; reducing source delay

II. Inter mule communication: Mules in MDD protocol create ad-hoc networks when

in range, this assures faster delivery of data in small time; reducing carrier delay.

12.1. Result analysis of DD and MDD Protocol

With the result analysis above, we can conclude that for every simulation with varying number

of sensor nodes MDD protocol gives a better and efficient result in delivery of data with reduced

time delay in every case scenario.

23.923628

16.893394

14.008463

13.944558

10.788331

15.9116748

13.30016475

9.0167003

8.4617088

8.173725

7.941235024

9.378706775

50

100

150

200

300

Avg. Time Delay of DD

Comparision of DD and MDD Protocol

ATD (MDD-Protocol) ATD (DD-Protocol)

38

12.2. Advantages of MDD protocol

1. Low source delay and carrier delay in delivery of data.

2. Guarantees full security of source location.

3. Reduces probability of collision of data at basestation.

4. In case of collision, the redundancy is removed at real time, hence reduces load on

modules

5. Independent of transmission trough sensors, so eavesdrop attack model is inefficient.

Network traffic analysis is inefficient attack model.

6. Energy efficient protocol, sensor need not transmit to a long distance(1kb for every 331

feet).

7. Localization efficient, mules are equipped with GPS facility reducing localization issues.

12.3. Limitations of MDD protocol

1. Efficiency is dependent on mobility pattern of data mule. Path Selection and speed

control algorithms improves efficiency.

2. Mules need high data storage and processing power for data storage and collision control.

3. Dependent on network configuration, sensor deployment, zones and mobility pattern of

mules.

4. No intelligent selection of mobility pattern in data mules.

5. Mules assisted with GPS and other infrastructure increases cost.

39

13. References

01 N. Li, M. Raj, D. Liu, M. Wright, and S. K. Das, “Using data mules to preserve source location privacy in

wireless sensor networks,” at 13th International Conference on Distributed Computing and Networking

(ICDCN ’12), vol. 7129 of Lecture Notes in Computer Science, p. 309, 324, Hong Kong, China, 2012.

02 Y. Fan, Y. Jiang, H. Zhu, and X. Shen, “An efficient privacy-preserving scheme against traffic analysis

attacks in network coding,” in Proceedings of the 28th Conference on Computer Communications (IEEE

INFOCOM '09), pp. 2213–2221, April 2009. View at Publisher · View at Google Scholar ·

03 R. Lu, X. Lin, H. Zhu, and X. Shen, “TESP2: timed efficient source privacy preservation scheme for

wireless sensor networks,” in Proceedings of the IEEE International Conference on Communications

(ICC '10), May 2010. View at Publisher · View at Google Scholar · View at Scopus

04 K. Mehta, D. Liu, and M. Wright, “Protecting location privacy in sensor networks against a global

eavesdropper,” IEEE Transactions on Mobile Computing, vol. 11, no. 2, pp. 320–336, 2012. View at

Publisher · View at Google Scholar · View at Scopus

05 B. Alomair, A. Clark, J. Cuellar, and R. Poovendran, “Statistical framework for source anonymity in

sensor networks,” in Proceedings of the 53rd IEEE Global Communications Conference (GLOBECOM

'10), December 2010. View at Publisher · View at Google Scholar · View at Scopus

06 K. Bicakci, H. Gultekin, B. Tavli, and I. E. Bagci, “Maximizing lifetime of event-unobservable wireless

sensor networks,” Computer Standards and Interfaces, vol. 33, no. 4, pp. 401–410, 2011. View at

Publisher · View at Google Scholar · View at Scopus

07 X. Hong, P. Wang, J. Kong, Q. Zheng, and J. Liu, “Effective probabilistic approach protecting sensor

traffics,” in Proceedings of the Military Communications Conference (MILCOM '05), vol. 1, pp. 169–

175, October 2005. View at Publisher · View at Google Scholar · View at Scopus

08 Z. Zhou and K. C. Yow, “Anonymizing geographic Ad Hoc routing for preserving location

privacy,”International Journal of Network Security, vol. 2, no. 3, pp. 210–218, 2006.

40

09 Li, Y., Ren, J.: Source-location privacy through dynamic routing in wireless sensor networks. In:

Proceedings of INFOCOM (2010)

10 Ozturk, C., Zhang, Y., Trappe, W.: Source-location privacy in energy constrained sensor network routing.

In: Proceedings. 2nd ACM workshop on Security of ad hoc and sensor networks, SASN (2004).

11 Mayank Raja, Na Lia, Donggang Liub, Matthew Wrightb, Sajal K. Das : Using Data Mules to Preserve

Source Location Privacy in Wireless Sensor Networks (CReWMaN),The Information Security Lab (iSec)

12 Ryo Sugihara and Rajesh K. Gupta. University of California, San Diego: Path Planning of Data Mules in

Sensor Networks. In: ACM Journal Name, Vol. V, No. N, Month 20YY.

13 Manual- OMNeT++