Contexte
L'institut de recherche informatique I2R (Institute for Infocomm Research) est un membre de l’Agence pour la Science, la Technologie et la Recherche (A*STAR) à Singapore
A*STAR, an official agency under Singapore's Ministry of Trade and Industry, was established in 1991 to promote and enhance scientific and technological research Comprising seven distinct research institutes, A*STAR focuses on various fields such as data storage, materials science, chemistry, computer science, microelectronics, advanced manufacturing, and metrology Its mission is to address global technological challenges and foster the development of future industries.
Fusionopolis 1 is a prominent center in Singapore dedicated to scientific and technological research Established in 2002, the Institute for Infocomm Research (I2R) aims to drive innovation in computing through dedicated research that supports Singapore's economic success I2R focuses on research and development in various fields, including wireless and optical communication networks, interactive and digital media, and signal processing During this internship, we are engaged in a research program on wireless networks called S&S (Sense and Sensibilities) at this institute.
Le programme Sense and Sensibilities(S&S)
The increasing urbanization is putting pressure on land resources and quality of life As cities and roads become more congested, urban solutions must provide real-time environmental information to assist government agencies in making informed decisions for a safe and clean living environment Given the diverse demands, detection systems must also accommodate various sensor modalities, technologies, and end-user applications However, most current detection solutions are designed for specific problems and applications The A*STAR S&S program is researching to develop a generic platform that can be utilized across different applications with varying infrastructure needs in multiple domains.
Led by Dr Tan Hwee Pink, the program comprises over 20 full-time researchers and engineers with expertise in various scientific fields, including electrical and electronic engineering, computer science, telecommunications technology, physics, and mathematics This team specializes in diverse detection applications, such as noise monitoring, water quality assessment, air quality surveillance, and parking monitoring Recently, the team received the Award for Leading, Educating, and Nurturing Talent from A*STAR, highlighting their significant contributions to research and innovation.
Énoncé du problème
The deployment plan aims to identify the most suitable locations for monitoring significant physical phenomena and collecting meaningful data, while considering environmental constraints and application and network requirements Several questions may arise when deploying the nodes.
Où doivent être placé les nœuds, de sorte que la qualité du phénomène physique à mesurer donne une précision satisfaisante et acceptable ?
À quelle distance les nœuds doivent être placés entre eux, de sorte que le nombre de paquets reỗus soit suffisant et la connectivitộ de l'ensemble du rộseau puisse ờtre assurée ?
Comment le nombre de nœuds à déployée peut être minimisée, de telle sorte que la couverture de détection soit maximale et le cỏt de déploiement soit réduit au minimum ?
Si les nœuds utilisent l'énergie solaire pour recharger leurs batteries, quels sont les endroits ó ils peuvent recevoir assez de lumière pour recharger leurs batteries et se maintenir en vie ?
If nodes are positioned too closely together, the data they collect will be highly correlated, resulting in weak coverage Conversely, if the nodes are too far apart, connectivity between them will be poor Therefore, it is essential to consider various physical and environmental constraints to identify optimal node placements.
To meet these constraints, it is essential to observe and model the spatiotemporal variations of the communication channel and relevant physical phenomena, such as noise Furthermore, the observation of these phenomena should stem from the analysis of field-collected data to ensure that statistical models closely reflect reality.
Contributions
Our goal is to establish an efficient process for deploying sensor networks in urban areas Initially, we developed a site survey tool capable of collecting essential data.
(i) l’intensitộ du signal reỗu (RSSI) entre deux points quelconques ; (ii) le signal du phénomène physique à observer à un endroit donné
A thorough understanding of radio wave propagation is crucial for designing effective wireless network deployment strategies Unlike wired networks, wireless networks are significantly more complex due to the nature of the radio channel, which is influenced by deployment site characteristics and can vary greatly based on terrain, transmission frequency, interference sources, and other factors Characterizing the radio channel through mathematical models with specific parameters is essential for predicting signal coverage and identifying optimal node locations Our survey tool consists of an Android application installed on a tablet, an XBee module, an Arduino board with a sound sensor to measure noise, and an internet-accessible server for configuration management and user synchronization This tool enabled us to conduct multiple experiments to collect environmental data, which we used to develop a regression model representing the temporal variation of the wireless channel, alongside an autoregressive model for the temporal variation of the physical phenomenon of interest.
Organisation du rapport
The remainder of this report is structured as follows: Chapter 2 focuses on existing literature regarding coverage and connectivity in sensor networks, as well as radio propagation mechanisms Chapter 3 provides a detailed overview of the technical implementations of the field survey tool In Chapter 4, we discuss the statistical models developed from the data collected during various collection scenarios Finally, Chapter 5 concludes the report with discussions on future work.
Couverture et Connectivité
In the field of literature, active research areas focus on the coverage and connectivity of sensor networks Coverage in a network can generally be categorized into three distinct types: (i) area coverage, (ii) point coverage, and (iii) barrier coverage Area coverage involves the distribution of nodes to cover a specified surface, while point coverage aims to monitor a set of specific points within the designated area Barrier coverage, on the other hand, pertains to the detection of events that may occur at various locations.
Connectivity refers to the ability of sensor nodes to reach the backend server A lack of available routes from a node to the server can lead to data loss Each node has a communication range that determines the proximity required for another node to receive data from it, which differs from the detection range that indicates the area a node can observe When deploying a sensor network, various proposals utilize popular regular patterns such as hexagons, square grids, diamonds, and equilateral triangles, as shown in Figure 1, due to their deployment convenience and connectivity potential The performance of these patterns varies based on the communication range (rc) and detection range (rs) A study on connectivity within a sensor network is discussed in the article.
When the ratio of rc to rs is between 2 and 3, the diamond-shaped model outperforms the other three models In cases where the ratio falls between 1.14 and 2, the square model is the most effective For ratios less than or equal to 1.14, the hexagonal pattern proves to be superior.
In regions with obstacles, research indicates the division of the area into sub-regions composed of lines, where a sufficient list of sensors can cover an entire line while ensuring connectivity The distances between nodes are influenced by communication and detection ranges In the presence of an obstacle, multiple rows surrounding it are treated as a single region, with efforts made to maximize the covered area using a circular pattern However, the proposed solutions primarily focus on environments within buildings.
Figure 1: Motifs réguliers (gauche) & Déploiement avec obstacles (droite)
To enhance the studies conducted in [3], a new research effort focuses on the arrangement of nodes in environments with obstacles, employing the Delaunay triangulation method [4] The authors initially position nodes around the obstacles, as illustrated in Figure 1 on the right Subsequently, Delaunay triangulation is applied to distribute nodes in the remaining areas This method partitions a surface into polygons such that points within a polygon are closer to its center than to the centers of other polygons It represents a specific decomposition of a metric space based on distances to a discrete set of objects, typically a set of discrete points [5], with sensors located at the centers of the polygons (Figure 2).
The work closely related to ours involves deploying a network of sensors to monitor carbon levels in urban areas Initially, ecologists place certain nodes, and subsequent nodes must be positioned to ensure network connectivity They tackle this challenge using a geometric version of the Steiner tree problem under hole constraints, assuming that nodes within the same component form a group However, our project differs as we do not have predefined optimal positions for the initial set of nodes, making their approach inapplicable to our context.
Deployment models using regular patterns, as discussed in the literature, face challenges in complex environments like urban areas In our project, the node positions may vary based on real physical phenomena and solar energy availability, which are not predetermined Additionally, urban obstacles can be mobile or vary in size and density, making these deployment models ineffective.
Mécanismes de propagation du signal radio
To collect accurate data, it is crucial to understand the various signal propagation mechanisms, which will aid in designing effective data collection scenarios Radio wave propagation at different frequencies refers to the behavior of waves traveling from one point to another In wireless sensor networks, three fundamental mechanisms of radio propagation are typically considered: reflection, diffraction, and scattering Reflection involves a sudden change in the direction of wave propagation when it encounters different surfaces Diffraction describes how waves behave when they meet an obstacle or pass through an opening, which can be interpreted as scattering around the object's points.
Scattering is the phenomenon where radiation, such as light, sound, or moving particles, is deflected in multiple directions due to interactions with other objects This process is illustrated in Figure 3, which depicts the interior of a building, highlighting Tx as the transmission point and Rx as the signal reception point.
Figure 3 : Mécanismes de propagation du signal
During communications, there is a gradual loss of signal intensity known as fading, which is caused by the signal traveling along multiple paths or by obstruction from hills or large buildings In urban areas, the propagation of radio waves can also be affected by noise and interference on the wireless channel Noise refers to any unwanted signal that exists within the desired channel's bandwidth, while interference occurs when additional signals are introduced alongside the original signal.
To collect data, we developed a tool called SmartPloy, which consists of an Android application connected to an XBee module and/or an Arduino board, along with an online server This chapter focuses on the design and implementation of both the application and the server We will detail the SmartPloy survey tool and the various features we have integrated.
Nous allons expliquer dans la première partie les différents outils matériels utilisés avant de détailler l’architecture logicielle dans la deuxième partie.
Architecture matérielle
To gather data on signal strength between two points or to measure noise levels at a specific location, we can utilize the SmartPloy tool This tool comprises several components, including a phone, an XBee module, an Arduino board, and a server.
Our objective is to utilize a mobile phone for data storage, as the sensor nodes we employ lack sufficient memory capacity for large data sets Additionally, since data collection will occur in the field with mobile human users, a phone is an ideal choice due to its portability Each phone we use is equipped with 3G internet connectivity, enabling access to online information Furthermore, a phone can broadcast packets to other phones participating in the experiment.
An XBee module is a versatile tool that enables communication with heterogeneous devices equipped with wireless connectivity by sending and receiving data packets These modules utilize the IEEE 802.15.4 networking protocol for effective communication Designed for high-speed applications that require low latency and cost-effectiveness, XBee modules also offer a significant communication range, depending on the surrounding environment Additionally, they can be integrated with an Arduino board for enhanced functionality.
Arduino is an open-source printed circuit board that enables the creation of various objects for robot control and home device management Our sensor nodes are built using these circuits, integrating a noise sensor that measures sound levels Once the noise is detected, the node transmits the data through its built-in XBee module to other devices and an online server.
We have an online Java server that facilitates the configuration of connected devices and coordinates online users Upon startup, the server receives configuration parameters, as illustrated in Figure 5 These parameters are then sent to online users upon their request When the server receives a message, it broadcasts this information to all connected users, notifying them of events such as new user connections or incoming messages.
Figure 4 illustrates the server configuration interface, which prompts users to input the number of users, transmission interval, data storage directory on the phone, and the channel to be used Additionally, there is an option to utilize previous configurations Once the configuration phase is complete, the server can be launched.
Figure 4 : Configuration et Lancement serveur
A phone equipped with an XBee module can send and receive packets to other phones also fitted with XBee modules, as illustrated in the following figure It can also receive data on physical phenomena, such as temperature, transmitted from an Arduino board Each user will be equipped with this device for data collection and will have the capability to send packets to other users Additionally, a server can facilitate better coordination among users.
Figure 5 : L'outil SmartPloy sans le serveur
In Figure 5, we showcase a Nexus 7 phone with the Android application installed The phone is connected to an XBee module, enabling the transmission and reception of data packets Additionally, the Arduino board can detect physical phenomena, such as sound, and relay this information to the phone through its integrated XBee module.
Architecture Logicielle
Our mobile application will implement several key features to achieve our objectives, including configuration settings, map display, real-time user chat, and data collection (measurements) These functionalities will facilitate a data collection campaign, ensuring that information is stored in memory while remaining synchronized with other users.
Définir des paramètres de configuration
This feature allows users to configure various application settings, including the packet transmission delay, the channel for sending packets, the directory for storing collected data, and the maximum number of packets to send Users can manually specify their location or utilize GPS (Global Positioning System) In addition to local parameter modifications, the application also provides the option to retrieve configuration settings from the server.
Figure 6 : Interface utilisateur pour configuration
Cette figure 6 montre une interface utilisateur ó nous pouvons renseigner les paramètres de configuration avant de commencer la collection de données
Solliciter les paramètres du serveur
Utiliser les paramètres sur le téléphone
Cette fonctionnalité nous permet de voir sur une carte les endroits déjà visités par chaque utilisateur qui participe à l’expérience
Figure 7 illustrates the application sending a request message to the server and subsequently awaiting a response Once the server returns the list of locations previously visited by users involved in the experiment, all these locations are displayed on a map for the user.
Discuter avec les utilisateurs en ligne
Avec cette fonction, l'utilisateur peut envoyer et recevoir des messages de chat à aux autres utilisateurs en ligne
In Figure 8, upon the user's initial login, the application prompts them to enter their username, enabling them to send and receive messages with other users The screenshot below illustrates a communication example between a user named "cheikg" and the current user.
Figure 9 : Capture des messages de chat
With SmartPloy, users can send and receive packets to assess the signal strength between two locations It also enables data reception from an Arduino node that gathers noise data Additionally, users can synchronize through the online server.
When a user is ready to start or restart the experiment, they send a message indicating their readiness In response, the server notifies other connected users that a specific user has joined This exchange is illustrated in the accompanying figure, which shows three phones equipped with XBee modules, referred to as "device 1," "device 2," and "device 3." We assume that devices 2 and 3 are already connected to the system Device 1 then connects and sends a message (1) to the server indicating its readiness to begin the experiment The message (1) consists of the string "id Ready," while disconnection is indicated by "id Not Ready," with "id" representing the identifier of the phone that sent the message Once the server receives this message, it informs all other users of the new connection.
A message is sent to inform users about the list of individuals ready to begin data collection This message consists of a series of character strings indicating whether a user ID is "Ready" or "Not Ready," with "ID" representing the user's identifier Upon receiving this message, each user can choose to initiate data collection and send or receive packets that include the phone's identifier, GPS coordinates, and the radio channel used.
Figure 10 : Syncronisation entre Téléphones de SmartPloy
Figure 11 illustrates the data flow during the collection process When the user decides to start sensing, they send data to two recipients: the server and other phones The message sent to the server can either be a chat message, as shown in Figure 13, or a synchronization message that occurs every two minutes, as previously detailed in Figure 10.
The system will send a packet to other phones every five seconds, containing the GPS latitude, longitude, accuracy, phone ID, and packet sequence number The transmission will cease either when the user manually stops it or when a predefined limit on the number of packets is reached Configuration settings allow users to specify this packet limit before the sending process halts.
Simultaneously with the transmission, it listens and receives data from either other telephones, the Arduino board, or the server The message received from the server has already been illustrated in the figure.
The received packets from other phones enable the measurement of signal strength and contain key information such as the source address, arrival time of the packet, sender ID, Received Signal Strength Indication (RSSI), packet sequence number, communication channel, and GPS coordinates This data is stored in a text file within the phone's memory.
If the received packet originates from an Arduino board, the following information is recorded in a separate text file: the sender's source address, arrival time, node identifier, battery charge percentage, energy measured across ten frequencies, and GPS coordinates.
L’interface utilisateur développé pour recevoir et envoyer des paquets peut être observée avec la figure 12 suivante:
Figure 12 : Interface pour envoyer ou recevoir des données
By checking the "wireless networking" box, users can send and receive packets to estimate the signal strength (RSSI) between two points To receive data from an Arduino board, users must also check the "Inter-sensor-sampling interval" box; otherwise, they will only receive packets from other devices Currently, the "inter-solar-sampling interval" option for solar energy data collection is not yet implemented Once users are ready, they can send a message to the server to find out the number of connected users, allowing them to decide whether to press the "Start" button to initiate the process.
Donnộes reỗues des autres téléphones équipés de xbee
Donnộes reỗues d’une carte Arduino
Une fois ô ready ằ, il reỗoit du serveur le nombre d’utilisateur en ligne le programme et commencer l’expérience
To summarize the various messages exchanged among the features of SmartPloy, we present the following figure (Figure 13) Each feature interacts with the server at different times using an internet connection, exchanging string-type messages For displaying the map of previously visited locations, the application sends a request (request list all locations), to which the server responds (reply list all locations) The server sends a list to clients containing objects defined by latitude and longitude, representing previously visited positions Additionally, the application includes a feature that periodically sends the device's geographical location to the server using GPS coordinates (send device location) Each time the server receives this message with the user's latitude and longitude, it saves the information (save device location), allowing users to request and visualize these positions.
The phone's measurement feature allows it to communicate its readiness status to the server by sending either a "Ready" or "Not Ready" message The server then broadcasts this information to other users For the chat functionality, users can send messages to the server, which subsequently relays them to other participants Upon launching the application, users must first configure their settings before proceeding to other tasks like chatting Users can either manually input configuration parameters or request information from the server, such as the communication channel, packet transmission intervals, maximum packet count for experiments, data storage directory on the device, and the timing for sending the phone's location to the online server The server responds with the requested details.
Procédure générique
To establish a statistical model for characterizing communication in sensor networks, we begin by selecting an appropriate statistical model If data is unavailable, we must first collect the necessary data Once the data is gathered, we attempt to identify the model parameters that align with the available data If the model does not adequately fit the collected data, we repeat the process by selecting a new statistical model This iterative process continues until the proposed model accurately reflects the data.
Figure 14 : Processus pour la mise en place d'un modèle
Modélisation statistique de la puissance du signal
Collection de données sur la puissance du signal
We aim to measure the signal intensity between two arbitrary points within a specified area This data will be utilized to validate a statistical model that will assist in optimally deploying our nodes, thereby enhancing network connectivity Consequently, we are designing a scenario to collect the necessary data and assess the applicability of the proposed model in our context.
To determine user localization accuracy, we utilized GPS, which can be imprecise due to urban interferences such as large buildings Our goal was to assess the consistency of data provided by SmartPloy We aimed to verify if the results from SmartPloy align with expected outcomes in an unobstructed environment, where we anticipate signal strength (RSSI) to decrease with distance, as previously described The selected area for this experiment measured 167m x 100m and was located at Fusionopolis Way, Singapore Two participants were involved: one remained stationary (green icon in Figure 15), while the other moved in a circular pattern within the designated area (red icon) Both devices transmitted 5 packets per second over a duration of 30 minutes.
Figure 15 : Collection de données dans une zone sans obstacles
L'icône verte indique l'emplacement d'un utilisateur qui est statique pendant la collecte L'icône rouge indique la position du deuxième utilisateur qui se déplace, en utilisant les coordonnées GPS Scenario 1
In this scenario, our objective is to collect real data in a real-world environment, specifically where we plan to deploy sensors We aim to observe the behavior of signal strength in an urban setting, so we traveled to Dover, a locality in Singapore, for this data collection One user remains stationary at the roadside, while another moves along the stretch of road Our goal is to analyze signal behavior in an area heavily frequented by vehicles and pedestrians We utilize GPS coordinates to calculate the distance between the two points.
Figure 16 : Collection de données autour d'une route
Analyse Temporelle
Pour les données du scenario 0
In our open space experiment, we collected Received Signal Strength Indicator (RSSI) data to assess its conformity with expectations, anticipating lower RSSI values as distance increased As expected, the RSSI values, plotted on the vertical axis, decreased with increasing distance on the horizontal axis, as illustrated in Figure 17.
Figure 17 : RSSI sur Distance, en zone sans obstacles Pour les données du scenario 1
Comme nous l'avons dit précédemment, nous avons recueilli des données le long d'une route et affichons cette figure montrant le RSSI sur la distance
Figure 18 : RSSI sur Distance autour d'une route
Regression to the mean is a statistical phenomenon where an extreme measurement on the first assessment tends to be closer to the average on the second assessment Conversely, if a variable is extreme on the second measurement, it is likely to have been closer to the average on the first This concept indicates that repeated measurements within the same population will gravitate towards the mean, illustrating the tendency to move away from extremes The observable results are depicted in the following figure.
Figure 19 : Régression vers la moyenne
Comme nous pouvons le voir sur la figure 19, nous avons un intervalle de confiance de 95%, l'intervalle d'incertitude Afin de comparer avec la distribution gaussienne, nous traỗons la figure 20 suivante
Figure 20 : Comparaison avec la densité gaussienne
In this figure, we compare our empirical density of residuals with the normal (Gaussian) density, revealing that our model aligns well with the real data collected in our second experiment This suggests that the model may be suitable for our objectives and context Moving forward, it is essential to gather more data and perform cross-validation by dividing the sample size (n) into a training set (over 60% of the sample) and a test set The model is developed using the training set and validated against the test set to ensure its reliability.
Modélisation statistique du phénomène physique
Collection des données de bruit
We conducted this experiment to observe our tool's behavior in a real-world environment and to perform preliminary data analysis Data collection took place near a construction site and a restaurant, each for a duration of 40 minutes Our analysis of the data gathered from both locations reveals temporal variations across ten different frequencies The results indicate a significant difference in energy levels, with lower frequencies at the construction site exhibiting much higher energy compared to those recorded at the restaurant This is further illustrated in the subsequent figure, which provides clearer insights into the energy variations over time.
The aim of this experiment is to gather noise data in a real-world environment where we will deploy the sensor network After collecting the data, we will conduct a temporal analysis to explore how we can make recommendations for the deployment strategy.
The experiment was conducted alongside a road in Dover, involving three users positioned as shown in Figure 23 In this scenario, the noise source can be identified: the first user is located near a construction site producing significant noise, while the second user is situated 20 meters away from the first, and the third user is also 20 meters from the second The experimentation lasted for 30 minutes, beginning at 10 AM, with measurements taken every 5 seconds.
Figure 23 : Collection de données de bruit
Analyse temporelle
To explore correlations between data, we utilize partial correlation measures, which assess the degree of association between two random variables while controlling for the influence of one or more other variables Figure 24 illustrates the energy measured over time by user 1 in the first histogram at the top In the middle, we observe the autocorrelation curve, which identifies regularities and repeated patterns within a signal, such as a periodic signal disrupted by noise or a fundamental frequency absent from the signal This analysis reveals correlations among the variables, with the final histogram in Figure 24 displaying the coefficients of correlated variables.
Ainsi, à la position de l'utilisateur 1, avec le bruit X ( t -1 ) enregistre au temps (t – 1) , nous pouvons prédire le bruit X ( t ) au temps t (le temps est en seconde):
Figure 24 : Times series pour l'utilisateur 1
Partial Correlation Function Noise as a Function Time
Par la même méthode décrit précédemment, nous avons déduit, pour la position de l'utilisateur 2, l'expression suivante :
X (t) la valeur du bruit à l’ instant t
Pour la position de l’utilisateur 3, on a l’expression suivante :
Based on these results, it is possible to predict the measured value over a three-second period By gathering more representative data, we could estimate the maximum time required to measure a data point, thereby conserving energy and enhancing the network's lifespan.
The deployment of wireless sensor networks, as examined in the literature, often lacks a genuine optimization strategy While authors tend to position their nodes to ensure network redundancy, this approach significantly increases financial costs Additionally, we have demonstrated that the use of patterns is not feasible in urban environments.
During this internship, we developed an application for collecting noise data and signal strength information, which are used to build statistical models This mobile application allows users to gather data efficiently and utilizes an online server for user synchronization Technologically, this project enables Android phones to send requests to sensors, allowing users to obtain various types of information or configure the sensors remotely This functionality is highly beneficial for managing sensor networks.
Statistical models enable short-term predictions, allowing for the identification of time intervals during which the node can gather environmental data, such as noise levels Once validated, these models can yield even more significant and relevant insights from the research.
During this project, we faced several challenges, particularly in data collection from over ten experiments that ultimately proved unusable We overlooked critical factors such as the optimal start time, duration, number of packets to send, sending speed, and whether to remain static or dynamic As we progressed, we refined our data collection scenarios, leading to improved results For future work, we aim to plan more effective scenarios and gather additional data to enhance our statistical models, yielding more relevant outcomes for deployment Although we intended to conduct a spatial analysis of the collected data, we were unable to do so, making it a priority for future efforts.