Technical report, IDE10XX, January 2010
Cooperative Quality-of-Service Prediction in Distributed Systems
Master’s Thesis in Computer Systems Engineering Wei Wang
School of Information Science, Computer and Electrical Engineering
Halmstad University
Cooperative Quality-of-Service Prediction in Distributed Systems
Master’s thesis in Computer Systems Engineering
School of Information Science, Computer and Electrical Engineering Halmstad University
Box 823, S-301 18 Halmstad, Sweden
January 2010
Description of cover page picture: Vehicles connected to each other through an ad hoc wireless network.
Preface
This thesis report entitled Cooperative Quality-of-Service Prediction in Distributed Systems has been written for the partial fulfilment of my Master Degree in Computer Systems Engineering at Halmstad University, Sweden.
I would like to thank my guide and supervisor Kristoffer Lidström for all the motivation and timely help he has put in. I would also like to thank Dr. Tony Larsson for his guidance, inspiration and motivation to help me complete this thesis.
I would like to extend my sincere gratitude to all those who have helped me in the successful completion of this project.
Wei Wang
Halmstad University, January 2010
Abstract
Because of the prospective advantages and high application value of vehicular ad-hoc networks (VANETs) in the near future, more and more companies and research institutes embark on VANET related research. Communication technologies such as dedicated short-range communications (DSRC) are promising to dramatically improve road safety by providing early warnings. In a VANET, vehicles share safety-critical information with their neighbouring nodes, exchange observations as well as perform various computations on sensor data; therefore, the quality and reliability of links between nodes in the network is essential.
In this thesis, an analysis is made of a proposed method for monitoring communication quality in VANETs where obstacles may interrupt the inter-vehicle connectivity. An empirical experiment was conducted and a simulation was designed to evaluate the approach.
Content
1 INTRODUCTION...1
1.1 MOTIVATIONAND OBJECTIVES...1
1.2 ORGANIZATIONOF THESIS...1
1.3 INTRODUCTIONTOTHEAPPROACH...2
1.4 CHALLENGES...3
2 RELATED WORKS...4
2.1 VANET...4
2.2 COOPERATIVE COMMUNICATIONIN VANETS...4
2.3 PREDICTIONOF LINK QUALITY...5
2.4 RADIOPROPAGATIONMODELS...6
3 DESIGN AND IMPLEMENTATION...7
3.1 THEPURPOSEOFEXPERIMENT...7
3.2 HARDWARE PLATFORM...8
3.3 CHOICEOF OBSTACLES...8
3.4 SOFTWARE PLATFORM...8
3.4.1 A brief introduction to TinyOS...8
3.4.2 NesC (Networked embedded system C)...8
3.5 IMPLEMENTATION DETAILS...9
3.5.1 Experimental Setup...9
3.5.2 Description of Scenarios...11
3.5.3 Result Conclusion...17
3.5.4 Simulation Evaluation...18
3.5.5 Simulation Conclusion...21
3.5.5.1 Simulation Analysis...21
3.5.5.2 Evaluation of the simulation results...21
3.5.5.3 Drawbacks of the simulation...21
4 APPLICATION...23
5 CONCLUSION...24
6 FUTURE WORKS...25
7 APPENDIX...27
8 REFERENCES...29
Introduction
1 Introduction
1.1 Motivation and Objectives
Drivers typically drive near and within some complex road conditions such as intersections, roundabouts, tunnels and dangerous bends which can lead to accidents. In the US, intersection crashes account for more than 45 percent of all reported accidents, and 21 percent of fatalities. In 2003, more than 9000 Americans lost their lives due to intersection-related crashes, a rate of more than one accident per hour [2]. If the drivers are however provided with precautions or prior information about the road conditions then they can avoid potentially dangerous accidents.
Especially, as the speed of the vehicles increases the importance of timely and correct warning information in hazardous situations also increases.
VANET (vehicular ad-hoc networks) and other technologies enable car-to-car communication, which is envisioned to greatly help in improving road safety. Vehicles can get information including location, speed and direction from positioning systems such as GPS which they can then communicate to other vehicles in the vicinity. In the same way they can also observe neighbouring nodes using on-board sensors or by receiving information transmitted from them wirelessly. Relying on these inputs, it is possible to process them to get more information about the current road situation and help drivers make more accurate decisions.
The main objectives of this thesis are twofold,
i) Analyzing the communication quality monitoring approach proposed by Lidström and Larsson [1] as well as related work. Checking whether it is feasible to model the wireless communication quality based on repeated measurements and how this can be done in a distributed network. Investigate the theoretical feasibility and performance of cooperative monitoring and prediction of quality-of-service parameters in highly mobile networks.
ii) Experiment measurement and simulation evaluation.
To explore the accuracy of the theoretical model by constructing a physical platform, in order to incorporate real measurements into the analysis. In this thesis we predict the outcome of the experiment in the way of statistical measurements, so a quantitative measurement of the quality of communications between nodes in the network has been implemented. We then design a simulation based on the experimental data, which is used to evaluate the proposed approach.
Since it is difficult to create a scenario where the number of vehicles is large, we think it is a relatively easy way to create more virtual nodes by simulation.
1.2 Organization of Thesis
The thesis is organized into six sections. First we summarize the proposed approach [1].
Section 2 introduces technologies related to VANET and gives a short overview on the related
work. Later in Section 3, we describe the hardware and software used in this thesis in more
details and the procedure of the experiment and simulation. Applications of the method are
Cooperative Quality-of-Service Prediction in Distributed Systems
introduced in Section 4. Finally, we conclude this paper and describe future works in sections 5 and 6.
1.3 Introduction to the approach
The approach is characterized by comparing observations and detecting link quality in circumstances where the influence of obstacles on communication quality may cause communication disturbances. Taking an example to illustrate this approach, in the case of the intersection collision warning application, we assume that all vehicles are within the indicated transmission range (ITR) of each other, i.e. in ideal conditions they would be able to communicate. However, for two cars approaching the intersection, direct communication is blocked by a building. Thus, the warning system will not detect that there is a car approaching from the other side, if the system is not able to get the information of such conditions, a traffic accident might happen. If the system can be notified that communication in that location has previously been difficult, steps may be taken to ensure graceful degradation or other adaptation to this situation.
In the proposed approach, a communication disturbance (CD) detection scheme helps solve the problem mentioned above. Assume that all vehicles are within ITR of each other, vehicles that cannot communicate directly can still obtain the observation information of each other by receiving secondary observations, i.e. observations relayed via other nodes in the network to which connectivity exists. After they obtain each other’s positions indirectly, by comparing and analyzing the primary and secondary context, the two vehicles can infer that they were unable to communicate directly at a point in a time, even if they were within ITR of each other. These discrepancies in information detected by comparing primary and secondary observations must be collected and aggregated from a sufficient amount of data to confirm this conclusion. The proposed approach estimates the availability and reliability of the wireless medium at hazardous locations by cooperatively detecting and then accumulating CDs in a central server over time.
The method is based on the fact that in VANETs vehicles need to periodically broadcast their local status to all neighbours within communication range. Such broadcasts contain the identity, position, direction, velocity as well as observation attributes, for example when an observation was made, the indicated transmission range (ITR), the observation’s transmission quality information like received signal to noise ratio and bit error rate [1], delivery ratio and packet delays. These transmission quality indices can also be added to the observation attributes to better illuminate the status of observations.
CDs can be caused by various static or dynamic circumstances, such as static obstacles (buildings, rocks, dense foliage and highway barriers), dynamic obstacles (other vehicles, pedestrians), environmental conditions (rain, snow, heavy fog condition), hidden-terminal problem, interference from other sources, different delays, and equipment properties such as unexpected antenna radiation patterns [1].
1.4 Challenges
Introduction
There are many research issues in the field of VANETs like channel conditions, unstable context, challenging wireless medium, link quality decay, random node failures, inaccuracy of node locations, etc. These issues impose a great challenge on evaluating and implementing VANET related protocols and applications. An important factor in VANETs is that the wireless channel is unreliable, it has a high packet loss rate and the capacity of the channel may change dramatically. Additionally the high node mobility and fast topology changes make efficient medium sharing and link quality prediction even more difficult.
In this thesis, we have given more emphasis on the city, or urban, environment. Urban
environments have certain unique characteristics, such as tall buildings that obstruct and interfere
with the transmission of signals and vehicles are closer together than in the country or highway
scenario. Due to these characteristics wireless medium congestion and interference due to the
congested traffic environment are common problems.
Related works
2 Related works
2.1 VANET
Vehicular ad-hoc networks (VANETs) are a special form of traditional mobile ad-hoc networks (MANETs). The main features of the VANET are that the nodes are the vehicles themselves. The high speed of vehicles makes the scale and density of the network vary rapidly and causes the network topology to change frequently.
One important characteristic of the VANET is that the movement of nodes is not random, to some extent it can be predicted, for example the mobility of vehicles is constrained by predefined road geometry. The speed of vehicles is constrained primarily by the level of traffic congestion, speed limits and traffic regulations. In some position-based routing protocols, the position of a node is predicted by its speed and route, for example, in [3] authors have approximated the connection lifetime between two nodes by using their distance and velocity. There is also research focusing on collecting real traces by installing wireless devices in vehicles [4]. A particular advantage for VANET is the low power constraints as a result of the rechargeable source of energy which is typically a challenging issue in MANETs and sensor networks. In a car, the virtually unlimited power supply makes communication with surrounding cars and devices over long periods of time possible, enables extended operating and stand-by times and extends the processing power. Moreover, several and more effectively placed antennas can be used to widen the communication range. These features combined influence the performance of routing protocols and applications in VANETs significantly.
There are two major categories of applications for VANETs [5]: one is applications dedicated to improving the safety on the roads, for example, vehicles can transmit warnings about accidents, traffic jams, ice-on-road and construction sites in order to warn the driver about dangerous situations. The other is related to improving the comfort of drivers. For instance, mobile internet access enables in-vehicle access to the Internet in order to obtain on-board entertainment, e-mail, games and file transfer.
In this thesis, we study a collaborative, context aware radio link monitoring approach which could be generally used in wireless communication networks. Vehicles can be seen as sensor platform; they sense the road condition, transmit and share the information with their neighbouring nodes. In this network, cooperation with other nodes plays an important role in getting more information about their immediate environment.
2.2 Cooperative Communication in VANETs
Cooperative communication systems have recently received significant attention in the
VANET research field. Cooperative communication based on vehicle-to-vehicle and vehicle-to-
infrastructure communications can greatly improve the efficiency of transportation, increase
safety, widen the communication range and enhance the quality and reliability of information [6].
Cooperative Quality-of-Service Prediction in Distributed Systems
In an ad-hoc wireless network, cooperative communication can adopt multi-hop communication, where the message from the source to the destination could pass several mobile nodes. Two or more mobile nodes in a network share their information and mutually transmit their messages, which is helpful when nodes are widely distributed.
Context aware cooperative communication in VANETs can make vehicles better aware of their surrounding environment; they can get information about the presence and location of obstacles and other vehicles by exchanging information between nodes. By cooperatively sensing and reacting to detected context content, information can be provided to evaluate the communication situation.
2.3 Prediction of Link Quality
The disconnection of nodes may cause dangerous problems in cooperative vehicular safety systems, and the disconnection shows at almost any power and distance settings, especially in complex places where communication is interrupted by circumstances such as big buildings, high mountains, bad weather, etc. The existence of bad links and blind spots is impossible to totally eliminate, but we can detect such spots and record the communication quality in order to give advance information about the quality degradation so as to avoid accidents caused by these interruptions.
The proper prediction of link quality can significantly reduce the problems emerging from imperfect connection and uses predicted link quality to pre-emptively take actions before accidents happened. Link quality is a critical part of most sensor network routing protocols, but current protocols do not provide many assessment of the quality between nodes. In [7], authors took a first step in proposing a probabilistic connectivity maintenance protocol based on the nodes deployment schemes. The protocol explicitly explains the probabilistic nature of communication links between nodes; they use the network packet delivery rate as a quantitative metric for communication quality and build a model to quantify the quality of communication between nodes in wireless sensor networks, finally they designed a distributed Probabilistic Connectivity Maintenance Protocol (PCMP). In this thesis, we develop an empirical experiment and build a simulation platform to model the environment where communication between vehicles is blocked by a building and evaluate how reliable the approach is as well as analyze its performance.
The current field of wireless communication quality prediction is filled with a variety of technologies. New approaches for the prediction of wireless medium quality and communication security should be designed to meet the specific network needs and to guarantee reliability and safety requirements. In order to better measure and predict the communication quality, statistical models for site-specific radio propagation can be used to provide a quantitative and a qualitative measure. Examples of research efforts include modelling and simulating specific issues involved in vehicular communication, such as medium access, mobility of vehicles, traffic flow and routing [8][9].
In some link quality prediction algorithms, the link quality between the transmitter and
receiver is mainly determined from the positions of the nodes in an ideal open environment. In
Related works real urban environments signal propagation will cause the interruptions due to interrupted line-of- sight path (shadowing) and significant signal fading effects [10], [11].
2.4 Radio propagation models
Previous work that predicts the received signal strength of packets relies on radio propagation models and there are different levels of detail associated with these models. A simple propagation model only takes into account the distance from sender to receiver with a signal loss function, e.g. the free space model assumes ideal propagation conditions between transmitter and receiver. A detailed model might use near and far receiver, the two-ray ground reflection model ignores effects such as radio frequency attenuation due to buildings and other obstacles [12].
A more general model uses statistical approximation of shadowing, which considers fading
effects. A more detailed model would consider signal attenuation from static or dynamic
obstacles, model line of sight or non line of sight communication from direct or indirect
communication. An elaborate model will consider antenna effects like orientation, length and
distance with ground to estimate reflection [13].
Design and Implementation
3 Design and Implementation
3.1 The purpose of experiment
In networking research, protocols and algorithms should be validated based on as accurate as possible models of the network. In this thesis, we think it is essential to create specific models that accurately model this approach to predict and understand its availability and reliability.
Although simulators can be used to implement the models, the fidelity of the simulation may not be good enough when compared to real measurements. Besides there are still some problems that could not be addressed only by simulation, some specific scenarios are hard to simulate. For example, it is difficult to consider antenna attributes which is an important factor in some models and to simulate various detailed radio propagation models, hidden terminal, etc. Some research has pointed out that the discrepancies between simulated and real-world network performance has been primarily attributed to the inappropriate level of detail at which these simulations are performed including the use of overly-simplified propagation models [13]. Empirical experiments are more relevant than simulation and the results match more closely with the real world results than simulation results.
To get a relatively good quality and reliable evaluation result, it is necessary to use a test-bed and scientific empirical experiment in different scenarios. In addition, in order to minimize the effect of the noise in the measurements we have chosen to do our experiments outdoors in an area free of multipath propagation. In our experiments, a quantitative measurement is provided to evaluate the communication between two nodes by recording received signal strength indicator (RSSI) values.
The requirement of experiment is that a set of RSSIs should be measured under different
transmission power settings, antenna types and geographical setup with one sender and one
receiver.
Cooperative Quality-of-Service Prediction in Distributed Systems
3.2
Hardware PlatformThe hardware platform of this experiment is two MICA2 motes, MIB510 interface programming board, a laptop and an obstacle. An introduction to the MICA2 motes and MIB5110 can be found in the appendix.
3.3 Choice of Obstacles
In the real world the existence of various obstacles significantly influences the communication quality between vehicles in a VANET. Therefore, it is necessary for VANET model evaluations to take the influence of obstacle model into account. Moreover, the influence of different obstacles are various, the shape and size, types of materials, the layout of obstacle and the nature of physical and communication [14] besides their changing nature with time, all of them can affect radio signal propagation through attenuation, reflection, diffraction, and refraction.
The obstacle that interrupts direct communication in the approach described in section 1 is a static obstacle, which represents a building in our experiment. In order to reflect the presence of a building, a 1 meter high, 1.5 meters long, mixed-material rectangular board is used. Obstacles of this shape are not supposed to be too big, compared with the whole network field and the size of MICA2, the 1 meter height of obstacle is suitable.
3.4 Software Platform
3.4.1 A brief introduction to TinyOS
MICA2 motes run under the TinyOS which is an open source event-driven component-based operating system designed especially for embedded sensor networks [15]. The TinyOS system, libraries and applications are implemented in NesC programming language, which is designed for programming structured component-based applications. TinyOS supplies a set of reusable components which can be used to program conveniently and easily to acquire and process data from sensors.
3.4.2 NesC (Networked embedded system C)
NesC is an extension to the C programming language designed to embody the structuring concepts and execution model of TinyOS [16]. It is a component architecture language for embedded systems. A NesC application includes three parts: a list of C declarations and definitions, a set of interface types and a set of components [16].
Programs in NesC are built of components, which are assembled to form whole programs.
There are two types of components in NesC, one is modules which contain application code, and
Design and Implementation
another is configurations which assemble other components together, connecting interfaces used by components to interfaces provided by others.
3.5 Implementation Details
In this section we will explain the experiment procedure, datasets we collected, and give some statistical characterization of the data.
3.5.1 Experimental Setup
We evaluate the proposed approach using practical experiments and simulations to build a
model of the real world application. The experiments consist of environment dependent empirical
link quality measurements in a statistical manner. RSSI is used as the link quality estimator in this
experiment. A set of RSSI values are collected using MICA2 motes, the RSSI parameter acts as a
wireless link quality estimator to evaluate the signal strength in different situations. To simulate
different mobility patterns of vehicles, several movement scenarios were designed. The proposed
link quality prediction method is based on statistical analysis of measurements; therefore repeated
experiments were carried out to get a relatively accurate result. The communication quality index
between two locations is altered with various factors. The communication quality index
associated with the CD detection, for instance, is influenced by network density. Different vehicle
can have different transmission properties due to different wireless communication devices
installed on them, which also can influence the CD detection. Additionally, the radio environment
is unstable from day to day, various environmental conditions have effect on radio propagation,
and the attenuation of signals is different in the damp cold early morning or night compared to the
dry hot day. In the experiment, the power of battery is decreasing gradually, it is impossible to
keep in a fixed level. Due to these uncontrollable factors, the reproduction of several kinds of
situations is necessary. The experiments need to be carried out over time, to get an average result
between two specific locations. We use MICA2 motes as the experimental platform. It has been
widely used in many wireless sensor network researches. Table1 shows the configuration used in
the experiments.
Cooperative Quality-of-Service Prediction in Distributed Systems
Attribute Value
Terrain 15m X 20m
Number of Nodes 2
Node Placement A set of different positions
Max date rate 38k Baud
Antenna Length 8 cm / 4 cm
Transmission power –20dBm to +10 dBm
Outdoor range 30m*20m
Radio Frequency 915MHz
Radio Bandwidth 200Kb/s
Experiment area Parking lot in Halmstad University
Table1. Experiment Configuration
The MICA2 provides a measurement of the received signal strength, referred to as RSSI,
which is an acronym for Received Signal Strength Indication. It is a measurement of the signal
power on the radio link. Different units are often used for RSSI readings, usually in the units of
dBm for RF signal strength. The CC1000 has a built-in RSSI that provides an analogue outputsignal at the RSSI/IF pin [17]. The RSSI reading is associated with the distance between the
transmitter and receiver and other factors such as the frequency of the radio signal. Higher
frequencies have higher attenuation; signals slowly lose their ability to go through obstacles, go
around obstacles and to reflect off obstacles. Thus, radio frequency also plays a role in the result
of RSSI [18].
Design and Implementation
3.5.2
Description of Scenarios
Picture 1: Outdoor Environment of the experiment on the campus of Halmstad University
In this experiment, MICA2 motes model the vehicles, a board as the obstacle was placed within an area to model the location of buildings. We designed several scenarios to simulate different vehicle conditions. All experiments were conducted in a part of the parking lot of Halmstad University, which is approximately 10m x 30m in open area. The ground is flat. No surrounding buildings for nearly 30 meters in all directions. Except experimenter, Mica2 and the laptop on the field in this environment, there is no line of sight obstruction for nearly 20 meters in all directions. However, it is inevitable for the radio attenuation from various reflections even if in the outdoor environment and there are some unknown effects on exact radio propagation.
The same experiment was repeated at different transmission power levels. I carried out the
experiment at 3 different power levels: 5, -7, and -20 dBm. The motes radio frequency runs at
915 MHz. Two MICA2 motes were used in this experiment. One acted as the receiver which
recorded the signal strength it received from the other node and forwarded it to the laptop. The
results were saved in text files that recorded the positions of two nodes and the RSSI readings
between two nodes in each scenario. Because of multipath interference and other factors, the link
between two motes is unstable; all the experimental results are statistical average values of at
least 50 samples, which help with calibration and gets relatively meaningful and reliable data for
the motes. Several parameters were changed in the experiment, such as the distance between the
two nodes and with the ground, transmission power and antennae height. Scenarios are repeated
with the same devices. A simple application (SendingMote.nc) is uploaded to the transmitter that
periodically sends messages over the radio at a fixed transmission power. Another application
(RssiBase.nc) that includes the RSSI data is put in the receiver that connected to a laptop using a
MIB510 serial programming board and sampled the RSSI. The receiver measures the signal
strength and forwards it to the laptop. The laptop was running cygwin, which commanded the
node to transmit messages. This procedure was repeated under different scenarios.
Cooperative Quality-of-Service Prediction in Distributed Systems
Figure 3.1: Model of the sender and receiver
The constant parameters in all experiments are:
1) MICA2 Motes are used.
2) Fully charged batteries are used for two motes.
3) The operating frequency of the motes is 915Mhz.
4) Use the same antenna in all experiments to avoid intra-antenna variation as a result of manufacturing differences.
5) Make all antennas pointing towards the sky in the whole process of experiment.
6) Obstacle held constant.
Since the RSSI readings can be affected by transmission power, distance, antenna gain and height, some parameters are changed in this experiment.
The dynamic parameters are as follows:
1. Outdoor and Indoor: The difference of indoor environment measurement and outdoor environment are as follows. In the indoor environment the signal attenuates much faster than outdoors due to effects by walls and physical obstacles such as floors, furniture which absorbing some of the signal energy, in addition, reflection from obstacles, multipath ways, diffraction, scattering and other interference effects associated with indoor environments can influence the signal strength. The figure below shows the influence of indoor environment.
Figure 3.2 Multipath radio effect, transmitter signal are reflected or diffracted by structures
Design and Implementation
I collected some values under the indoor environment to compare with the outdoor values.
There is obvious variation between data sets. As we would expect, the measurements in outdoor environments exhibit less noise due to less clutter, since the indoor values are very unstable, I mainly describe the outdoor scenario, and the RSSI readings collected indoor is not used in the simulation phase.
2. Different altitudes: Put the MICA2 motes on the ground or elevated. It can get better transmission when the nodes are elevated above the ground.
3. With or without LOS obstacles: The strength of the signal dropped when in the presence of the obstacle in the same scenario compared to the absence of obstacle, the obstacle significantly influences the signal strength.
4. Antenna length: The antenna length can greatly affect the transmission range of the MICA2 motes, if without antennas the transmission range will be greatly reduced, thus, the antenna length is an important parameter that should be taken into account when analyzing the data in this experiment.
5. Node Position: Nodes placed in array of different positions. Because of the attenuation of signal strength with distance, the RSSI value decreases with the increase of distances.
Scenario1:
The two MICA2 motes were placed directly onto the ground level. Antenna length of the motes is kept at the original length of 8cm. Node A is in a fixed position, just opposite to the obstacle, at a distance of 1 meter. Moving node B along another side of the obstacle, firstly the position of node B is in the line of sight of node A, after node B is moved to the behind of the obstacle, which obstructed the two nodes, making them in the non line of sight of each other. The two nodes are in a line and vertical to the obstacle, the distance of the two nodes is 2 meters, at this pair of positions, the RSSI value has a sharp drop compared with the positions in the line of sight. Then node B is moved gradually further away from the obstacle until the two nodes is in line of sight again. We chose five positions along a line for node B. Figure 3.3 shows the deployment of the two nodes and obstacle. Figure 3.4 shows the fluctuation of RSSI with distance and obstacle. After finishing all the distances, the transmitting output power was changed and everything else repeated.
The MICA2 provides 26 different kinds of power levels from -20 dBm to 5 dBm. The
register in the radio that controls the power level is designated at PA_POW register [19]. The
default address is 0X0B, the PA_POW should be set to 0x00 for minimum, to 0xFF for highest
transmission power. We repeated this scenario at three transmission power levels: 5, -7, -20
randomly. There was little variation in RSSI with transmit power changed.
Cooperative Quality-of-Service Prediction in Distributed Systems
Figure 3.3 Deployment of the two nodes and obstacle
Figure 3.4 Fluctuation of RSSI with the distance and obstacle
Scenario 2:
Two nodes were placed on each side of the obstacle; at the beginning the two nodes were in
line of sight of each other. They were then moved along the two sides of obstacle simultaneously,
keeping the distance of the two nodes at 2 meters, the trace of the nodes is thus two parallel lines,
see figure 3.5. When the nodes are not in the line of sight due to the interruption of obstacle, the
RSSI value would decrease dramatically, after moving out from behind the obstacle, the RSSI
value would increase. We repeated this scenario by enlarging the distance of the two nodes from
2 meters to 3 and 4 meters. Figure 3.6 demonstrates how the RSSI fluctuates with the two nodes
in relation to the obstacle.
Design and Implementation
Figure 3.5 Deployment of the two nodes and obstacle
Figure 3.6 Fluctuation of RSSI with the distances and obstacle
Scenario 3:
This scenario is similar to scenario 2, but the distance between the two nodes is enlarged
gradually. Initially they are in line of sight of each other and the distance between them is about
1.5 meters. At this point the RSSI value is very high. The two nodes are then moved along the
obstacle and the distance between them is gradually increased. As the two nodes are moved
further apart and become interrupted by the obstacle, the RSSI has a sharp jump. As the nodes are
moved even further and come into line of sight of each other, the RSSI goes up again. As they are
Cooperative Quality-of-Service Prediction in Distributed Systems
moved further the RSSI falls slightly with the increase in distance between the two nodes. Figure 3.7 shows the trace of the nodes and deployment of obstacle. Figure 3.8 indicates the RSSI fluctuation with the distribution of nodes and obstacle.
Figure 3.7 Deployment of the two nodes and obstacle
Figure 3.8 Fluctuation of RSSI with the distance and obstacle
Design and Implementation
Scenario 4:
Nodes were in the same environment and deployment in a way similar to that of the first scenario, except the two nodes were lifted 50cm above the ground which provides better signal strength and transmission range between nodes. Figure 3.9 demonstrates the variation.
Figure 3.9 Fluctuation of RSSI with the distance and obstacle
3.5.3 Result Conclusion
The experiment collected a set of RSSI readings using MICA2 motes in an outdoor environment which involves multiple physical locations of the two nodes. By conducting the experiment, we obtained the signal strength between a pair of nodes at different distances with and without the obstacle obstructing line of sight communication. We can observe that when the communication of nodes is interrupted by the obstacle, the RSSI has a dramatic drop; that is the communication disturbance mentioned in the approach. This step finished the collection of communication disturbances caused by a static obstacle. The experimental test cases and statistical data are useful for generating subsequent simulations to further assess the feasibility of the approach.
Because there are many factors that could influence the experiment and some uncertainties of
the environment which make it difficult to recreate variables like weather and antenna
orientation, the field environment can produce inconsistent results. Radio wave transmission
irregularity is a common problem in wireless communication. It has been experimentally shown
that communication ranges of nodes are not nice regular disks [20]. In addition, due to random
errors, natural variations in the process, interference, noise, hardware miscalibration etc., there
must be certain measurement or statistical error so the experimental data is impossible to be
perfectly.
Cooperative Quality-of-Service Prediction in Distributed Systems
From the theoretical point of view, the RSSI values should decrease with the drop of transmission power and increase of distance, but the RSSI values did not follow this rule strictly in this experiment, occasionally, irregular phenomena may occur. For example, wind-induced irregularities can cause an irregular fading of radio waves, which has been mentioned in [21].
When the wind is blowing, it moves dust where the signal is reflected, putting the same two nodes at same positions in different time can cause different results, which may cause inconsistencies for the experimental data collected in different time. Moreover, with the decrease of power of two AA batteries, the signal strength will become to be weaker and weaker; the power cannot guarantee to be consistent in the entire process, which also brings the inaccuracy of the experiment.
3.5.4
Simulation EvaluationSince there are not enough real nodes to model a large-scale VANET, we choose the simulator as a tool to enlarge the number of nodes in the network to observe relatively large-scale phenomena. The simulator enables an amount of virtual nodes be build. The simulator is coded in Java and is based on the empirical data collected in the experiment phase.
The simulator is used to create more virtual nodes to play the role of real nodes and simulate their attributes and mobility. The simulator enables observing the behavior of how the nodes send and receive messages and deal with the observations. In this simulator, there are seven nodes created which act as seven vehicles, four nodes’ positions correspond to coordinates recorded in the experiment phase, the others are synthetically defined in an array. There is an assumption that the communication between two nodes is performed if the RSSI value is greater than a certain threshold, otherwise the two nodes are unable to communicate directly with each other. After receiving secondary observations from each other, they compare and analyze the primary context and secondary context. The algorithm of CD collection is adopted from [1].
The implementation of the simulator is as follows. First each node is moved to their positions which include the measured experimental positions and synthetically created positions for five time points. To exemplify, five time points and the positions at these points:
t1(2,1);t2(3,2);t3(4,3);t4(5,4); t5(6,5). At time t1, the seven nodes move to their new positions,
then calculate the distance between the positions with (0,0); If the distance equal or greater than
ITR, then this node is deleted, otherwise each node broadcasts a message that contains its ID and
coordinate to other nodes one by one. Next, the RSSI between the transmitter (Node A) and
receiver (Node B) is checked. If the RSSI between them is equal or less than the threshold, Node
B will not receive the message. The receiver will compare the ID of this message, if the ID of
message is in its primary context and if it is also the ID of sender, then this message is put into a
primary list and is relayed to other nodes. Nodes received the relayed message will compare the
ID included in the message with the ID of sender, if the ID is identical and the node ID is in the
primary context, then put the receiver to the primary list, otherwise put into secondary list. The
flowchart below shows the process of this simulation.
Design and Implementation
When we think of vehicle-to-vehicle communication scenario, we have to face different
situations, various traffic densities: high traffic density and low traffic density, both of them could
influence the performance of the approach. When the number of vehicles in the surrounding area
is large, more reflecting objects degrade the strength and quality of the receiving signal, the
timely propagation of observations for accident avoidance is degraded; vehicles will compete to
send their message and contest the medium resources, the background data traffic is heavy, it may
cause the overhead and congestion of wireless medium and the influence of hidden terminal
interference. In another case, when the number of neighbouring vehicles is not enough to build a
one-hop communication, to relay observations is far more impossible. This type of environment
is too poor to judge the availability and quality of communication in these situations. Roadside
equipment can be used as support to achieve successful communication. It is also necessary to
adjust the transmission power according to the density of network.
Cooperative Quality-of-Service Prediction in Distributed Systems
Design and Implementation
3.5.5 Simulation Conclusion
3.5.5.1 Simulation Analysis
Simulation is needed as a tool to support the scalability of the nodes in this approach in my thesis. In the simulation phase, simulation presents a model of what the entire approach might look like. Simulation results show the performance of the approach. The idea behind simulated nodes is to enable large scale and low cost tests with a number of participating nodes. The simulation model is based on measured empirical data obtained from the MICA2 platforms. The virtual nodes built in this simulation are used to test the site-specific radio disturbance detection mechanism. The communication disturbances are aggregated in the simulation; the communication disturbances demonstrate that there is an obstacle interrupting the communication of the nodes
3.5.5.2 Evaluation of the simulation results
Our result shows that the communication discrepancies changes with the variation of RSSI threshold used in the proposed method. If all RSSI values are above a specific threshold, there is no communication discrepancies detected at all. While if the RSSI values are all below a specific threshold, none of nodes can communicate.
In theory the communication performance can be degraded due to congestion of the wireless medium and hidden terminal interference [1] in the urban environment. Research also indicates that the delivery ratio and packet delays in VANETs are more sensitive to the clustering effect of vehicles at intersections and their accelerations-deceleration [22].
In our simulation, it is possible to add or delete any number of nodes by changing parameters in a file. In order to detect site-specific disturbances at least three nodes are needed in this approach. In this simulation when the number of surrounding nodes is three, the communication discrepancies detection is still working.
When the number of discrepancies detected in a short term is low, it is also possible to detect the obstructions, but the result is not quite accurate because the discrepancy might be caused by unexpected incident or some unstable factors which are not due to long-term obstructions. In order to guarantee the precision, it is necessary to detect discrepancies over a long period of time.
3.5.5.3 Drawbacks of the simulation
This simulator enlarges the network size in order to detect site-specific radio disturbances cooperatively. On the negative side, some effects that influence the approach are hard to
illustrate, some events are missed in the simulations: signal interferences, noise problems, etc. In
Cooperative Quality-of-Service Prediction in Distributed Systems
addition, it is not reliable to evaluate some complex scenario using the simulation. The network
load statistics and how the efficiency of this network will be highest are not derived accurately by
the simulation. It doesn’t evaluate some details with some quality parameters; we can think over
it by using other tools in the future work and evaluate the sensitivity of the simulation results
toward some details.
Application
4 Application
The approach is not only useful in vehicle communication nearby an intersection environment where wireless communication is blocked by buildings; it can be applied to detect link quality by cooperative communication in hazardous spots like nearby a junction or in a dangerous bend. In some critical situations there exist recurring communication weak spots where communication is often interrupted by some circumstances, which may cause accidents. However, if in an early stage there are some timely and proactively information related to the presence of weak spot provided to drivers appearing in these areas, it will be beneficial for drivers to take appropriate actions.
This approach can develop a wide range of new application scenarios. Promising usage cases
including in some places, such as tunnel, highway, railway, there are some circumstances, like
frequently bad climates, will interrupt communication in some extent. By sensing these
circumstances and cooperatively exchanging the communication conditions in such areas in a
long time, after uploaded to centralized infrastructure; it will obtain a link quality map in these
locations.
Conclusion
5 Conclusion
VANETs provide many possibilities to improve the road safety and comfort. Cooperative communication in such network plays an important role; vehicles exchanging information with each other directly or indirectly can detect communication weak- or blind-spots, in order to build maps of link quality.
The surveyed approach [1] attempts to solve such problems. In this thesis, we have surveyed the approach by the design of experiment and simulation to model and simulate some scenarios using on the approach. In the experimental phase, we have presented an empirical measurement of the signal strength with mica2 based on a series of measurements about RSSI values with different configurations.
In the simulation phase, more virtual nodes were created to check the feasibility of the communication disturbance detection mechanism.
In conclusion, in some critical locations where wireless communication is interrupted due to
obstructions, by estimating the link quality relying on the cooperating nodes over a relatively long
time, it is possible to proactively react to some communication problems using information
collected by nodes that have previously encountered these situations. The approach should be
applicable to other scenarios besides traffic intersections. This report also contains test results and
the simulator data code.
Future works
6 Future works
No model is able to simulate the complex world of VANETs completely. In this thesis, there is work that remains to be improved in the design and analysis of the experiment and simulation in the future work.
1. More detailed measurement (Optimize QoS parameters): The RSSI readings can be refined by getting RSSI noise floor readings to estimate the channel background noise. Besides RSSI, packet delivery ratio, packet loss and bit-error rate (BER) are also important link estimation parameters in predicting the link quality. Other types of parameters could provide an even more fine-grained measurement than just measuring RSSI. For instance, some research indicates that link quality indicator (LQI) is regarded as a better indicator than RSSI [24].
2. We can try to use more efficient and large-scale simulation tools of VANETs to evaluate the approach, such as Corsim and Transims.
3. In this experiment, the MICA2 motes were put in a nearly obstacle free environment. To provide more accurate prediction of the link quality, more details of the environment can be taken into account the scenario model, we can change the surrounding objects and deploy little obstructions to make the surrounding infrastructure more representative of an urban outdoor deployment, a more accurate deployment that includes street direction, traffic signs, moving pedestrians, building shapes, etc. The MICA2 is static and moved by experimenter; maybe we can put the MICA2 motes on some type of moving platform to make the nodes move automatically.
4. The CDs can be caused by the poor quality of wireless medium, not only due to the high buildings. CDs can indicate a number of circumstances, mentioned in chapter 1.2. Different models can be designed to simulate other environmental conditions that may cause CDs.
5. Antenna orientation: By orienting antennas in a suitable way, much more connectivity can be retained. Each antenna has its own radiation pattern that is not uniform. In practice, this means that when the pair-wise antenna orientations of the transmitter and the receiver are changed, the RSSI values collected at the receiver varies for a given distance between a given pair of communicating nodes. We can test different types of antenna orientation to check the influence of the antenna.
6. It is not precise to state that the two wireless nodes is simply connected or disconnected according to its RSSI value is above or below a certain threshold, we can design a more sufficient and accurate quality map by considering different RSSI values into the quality map.
7. A hurdle in making this method applied into real world is that not all of the vehicles will be
equipped radio transceiver and other system deployment to be composed to the cooperative
objects, participate in wireless communication.
Cooperative Quality-of-Service Prediction in Distributed Systems
8. Maybe the quality map of CD’s distribution will follow some law, it is probable to be linear, or
some curves in other shapes. Therefore, maybe we do not need to get all the CDs nearby the
obstacle, after a relatively long-term data gathering, we can conclude a law about the CD
distribution.
Appendix
7 Appendix
Vehicular Ad Hoc Network: Vehicles connected to each other through an ad hoc formation form a wireless network.
Primary Observation: An observation of an object either communicated by that object itself or deduced through firsthand information from on-board sensors.
Secondary Observation: An observation of an object relayed by a third party able to observe the object directly.
Primary Context: The sets of primary observations.
Secondary Context: The sets of secondary observations.
ITR: Indicated Transmission Range
Communication Disturbance (CD): Discrepancy between primary and secondary observations.
Signal Strength: Relative power of received signal compared to power at a reference distance from transmitter.
MICA2: The sensor node uses MICA2 motes manufactured by Crossbow to conduct the
experiments in my thesis. The ChipCon model CC1000 as the RF transceiver is utilized in the
MICA2. The central processor of MICA2 is ATmega128L running at 8 MHz[23]. It consumes
8mA power in normal mode and less 15μA in sleep mode. The low power property suits to
develop a sensor node. The MICA2 Motes come in three models according to their RF frequency
band: the MPR400(915 MHz), MPR410 (433 MHz), and MPR420 (315 MHz), we use
MPR400(915 MHz in this experiment, with a wire antenna on each node. It is powered by two
AA batteries. The radio of the MICA2 has 26 output power levels. MICA2 can provide some
information about the signal strength, RSSI, etc.
Cooperative Quality-of-Service Prediction in Distributed Systems
MICA2 MPR400 Motes (borrowed from [15])
MIB510 Board: The MIB510 interface board is a multi-purpose interface board used with MICA family and other sensor products. The MIB510 serial interface board is used to program the MICA2 Mote in my experiment. It supplies power to the devices through an external power adapter option, and provides an interface for a RS-232 Mote serial port and reprogramming port.
The MIB510 can collect the sensor network data on a PC as well as other computer platforms.
The MIB510 has an onboard processor that programs the Mote processor. When the MICA2 node is connected to the MIB510 serial interface board, it can act as a base station [15].
MIB 510 board (borrowed from [15])
References
8 References
[1] Lidström K. and T. Larsson, "Cooperative communication disturbance detection in vehicle
safety systems", Proc. 10th IEEE International. Conf. on Intelligent Transportation Systems,Seattle, Washington, USA, Sept.30 - Oct.3, 2007.
[2] U.S. federal highway administration, http://safety.fhwa.dot.gov.
[3] William Su, Sung-Ju Lee, and Mario Gerla, “Mobility prediction and routing in ad hoc wireless networks”, International Journal of Network Management, Volume 11, Issue 1, Jan.
2001.
[4] C. Lochert, A. Barthels, A. Cervantes, M. Mauve, and M. Caliskan, “Multiple simulator interlinking environment for IVC", In 2nd ACM International Workshop on Vehicular Ad hoc Networks (VANET 2005), Poster Session, Cologne, Germany, pp. 87, 88, Sept. 2005.
[5] Menouar, H. , Lenardi, M. and Filali, F. “Improving Proactive Routing in VANETs with the MOPR Movement Prediction Framework” Telecommunications, 2007. ITST '07. 7th
International Conference on ITS, pp. 1-6, ISBN: 1-4244-1178-56-8, Jun. 2007.
