Localization using Mobile Wireless Sensor Networks
LLocalization using Mobile Wireless Sensor Networks
Aaron John-Sabu
Dept. of Electrical EngineeringIndian Institute of Technology Bombay
Mumbai, [email protected]
Abstract —Wireless Sensor Networks (WSNs) are groups ofspatially distributed and dedicated autonomous sensors for moni-toring (and recording) the physical conditions of the environment(and organizing the collected data at a central location). Theyhave been a topic of interest due to their versatility anddiverse capabilities despite having simple sensors measuring localquantities such as temperature, pH, or pressure. We delve intounderstanding how such networks can be utilized for localization,and propose a technique for improving conditions of living foranimals and humans on the IIT Bombay campus.
Index Terms —Mobile Wireless Sensor Networks, localization,target tracking, sustainable living
I. INTRODUCTIONIn a society that is booming with collaborative and wide-spread activities, both human-based and autonomous, it isa vital requirement to incorporate similar networks that canmonitor and support human needs such as pollution control,navigation, industrial monitoring etc. Utilizing wireless sen-sors networks for such functionalities further helps in thereduction of power consumption, work time, and difficulties inimplementation. We study the work performed on the applica-tion of WSNs in localization. The learnings are conglomeratedto develop a mechanism for the maintenance of wildlife in theIIT Bombay campus, particularly dogs.II. PREVIOUS RESEARCHIn recent times, the incorporation of WSNs into popu-lated societies have improved conditions of life and work,from increasing crop yield and optimizing utility distributionsystems (electricity, water, etc.) to reducing wasted powerand avoiding traffic jams. [1] elucidates how WSNs havedeveloped throughout the years, from more expensive andfewer sensors to cheaper but larger quantities of sensors,from discrete circuits and multi-chip solutions to system-on-chip (SoC) devices, and from one-way communicationlinks to bidirectional links and mesh and star designs. Theseoptimizations have led to widespread applications of WSNsone of them being navigational localization of objects andbodies using methods such as trilateration.
A. Localization using Wireless Sensor Networks
Localization can be performed using multiple methods, avery common one being GPS. However, the accuracy of GPSdata is relatively low ( > m ). [2] proposes a more accuratelocalization method wherein a wireless sensor network ofZigBees is developed and their relative signal strengths are used for trilateration-based localization. The Mobile TargetTracking (MTT) problem intends to find the moving path of atarget in a field based on target locations that are sampled atrandom intervals. [3] discusses several algorithms for solvingthis problem using two aspects: determining the current loca-tion of the target (localization, path tracing), and processinginformation collaboratively among multiple sensor nodes. Tra-ditional methods involving the informed selection of sensors,binary sensor-based methods with centralized and distributedarchitectures, and other methods based on triangulation aresuggested for tracking, while information can be processedusing leader-based algorithms or distributed algorithms.Unlike open environments, locations with several obstructionsor jamming hinder the proper function of the Global PositionSystem (GPS). In such scenarios, it is necessary to developa positioning system that can complement GPS. [4] proposesa pedestrian navigation system (PNS) wherein heterogeneousagents are used for sensing, communication and relayinginformation. Here, sensors such as accelerometers and mag-netometers are used as part of a dead reckoning (odometry)mechanism to localize the user. The NavMote (sensor onthe user) exchange information with NetMotes (predeterminedsensors) when they are in range; otherwise it works on itsown. This system has been tested in both indoor and outdoorenvironments, giving satisfactory performance at a distanceaccuracy of ± and a heading accuracy of ◦ .Localization for mobile targets can be performed usingbeacon-based methods, wherein some beacons aware of theirpositions provide geographic information to ordinary sensornodes to localize, whose precision increases with the numberof beacons. [5] proposes an algorithm utilizing mobile beaconsthat traverse the network deployment area and broadcastbeacon packets to generate a number of virtual beacons. Thedistance between the sensor node and the beacon can becalculated using RSSI (Received Signal Strength Indicator).Since a single-mobile-beacon system can have issues such asco-linearity due to the straight line moving trajectory of themobile beacon, a three-mobile-beacon-assisted mechanism issuggested wherein the object (sensor node, S i ) is localized asthe weighted centroid of the three beacon positions: S i (ˆ x si , ˆ y si ) = (cid:80) mj =1 w ij V j ( x vj , y vj ) (cid:80) mj =1 w ij , w ij = 1( d ij ) g , where V j is the j th virtual beacon, d ij is the distance between S i and V j , and g is an adjustable degree. a r X i v : . [ c s . N I] F e b hile anchors in anchor node-based schemes acquire their po-sitions in advance using GPS systems or artificial arrangementto locate unknown nodes, unknown nodes in anchor node-freeschemes are located using the connectivity information be-tween unknown nodes and anchor nodes. The former achievesbetter localization accuracy but require a large number ofanchor nodes which increases the energy consumption andhardware cost of WSNs. [6] proposes using mobile anchornodes to maximize the localization accuracy while decreasingthe energy consumption of WSNs. An anchor node movesbased on an equilateral triangle trajectory in a WSN areaand broadcasts position and time messages periodically which,on reception, are used via RSSI-based trilateration to de-termine the position of unknown nodes. Though sensitiveto the standard deviation of noise, the algorithm reducesthe number of beacon positions, trajectory lengths and nodedensity. Moreover, it remains robust at high traveling speedsof the anchor node. B. Optimization of WSNs and Other Applications
The distributed detection process involves sensor nodesthat are deployed randomly in a field to collect data fromthe surrounding environment. [7] describes and comparesthree detection schemes: centralized, distributed, and quan-tized, based on parameters such as detection probability andoverall probability of error, measured against varying energyconsumption, time periods and distances from the sensorsto the command center. The robustness of each scheme isalso tested using node destruction and partial observationdeletion. The distributed scheme is observed to be superiorin energy consumption (especially over large distances) androbustness. Although the centralized scheme uses fewer nodes,the distributed option needs fewer than twice that number toachieve the same detection performance.The neighbor discovery problem deals with situations wheresensor nodes find their neighbors constantly in mobile sensornetworks by communicating with each other while in motionand forward the collected information to a central commandcenter. Here, the active and dormant status of sensors canbe controlled, hence reducing the energy consumption signif-icantly. However, this may cause additional discovery latencysince discovery is possible only when neighboring nodes haveoverlapping active slots. Previous research had introduced thegroup-based method where a third state for waking up activelyis used to communicate the schedule and verify the neigh-borhood of nodes. [8] proposes an algorithm that considersthe embedded spatial properties and actively modifies theactive time of nodes depending on the number of undiscoveredneighbors. This has been tested using simulations and thediscovery time has been found to be minimal when comparedto algorithms presented in existing literature.[9] proposes a wireless sensor network distributed in thephysical environment for sensing, actuation and communica-tion such that indoor lighting is automated, while also givingoverride capabilities to users based on individual preferences.The system is decomposed into smaller pseudo-static subnets and issues such as nodes belonging to multiple subnets aresolved using a multi-agent system approach wherein agentactions are averaged or voting can be used to determinethe action. Furthermore, a supervised learning model can beused to predict the action of the user and act accordinglywhile remaining bounded to physical constraints. This can bereplaced by model-free reinforcement learning algorithms sothat the agents are trained when a model is unavailable or toodifficult to compute in a complicated real-world environment.III. T HE E FFECT OF
TAG LINKAGES
FOR
MOBILEWSN-
BASED
LOCALIZATION
A. Introduction
The Indian Institute of Technology Bombay campus, spreadacross over 500 acres, is home not only to humans but also awide array of plants and animals from leopards and crocodilesto cows, dogs and cats. Unfortunately, in recent times, humanactivities in the institute have been disrupted by stray dogs.Also, survival for these dogs has also become difficult due tothe movement of heavy-duty construction vehicles along thesame routes posing a threat to their lives. This report proposesa technique to track and guide dogs while avoiding harm tothe human population as well as the dogs in the institute.
B. Procedure and Solution
Based on the concept of mobile target tracking as discussedin [3], a network of distributed sensors may be placed atsuitable locations in the institute. However, owing to the largesize of the campus, this is not a scalable solution since thenumber of sensors and the corresponding energy consumptionwill be huge. Moreover, the distribution of dogs across theinstitute need not be uniform, due to which placing sensorsat certain locations will not be efficient although there is apossibility for a small number of dogs to visit these areas.Incorporating ideas from [6], the movable sensors on the dogscan behave as anchors using which other movable sensorscan localize themselves via trilateration. The number of fixedsensors can hence be reduced if the number of dogs is large.This is demonstrated using a simulation wherein the numberof fixed sensors has been reduced to in a m × m areawhere each sensor can sense sensors within a radius of m .Three movable sensors are placed such that the size of theirpoint represents the uncertainty of finding their location.We define the neighbors of a sensor at a particular moment tobe the sensors capable of connecting to and providing infor-mation about position and time to the sensor for trilateration atthat instant. A simple geometric uncertainty principle is usedwherein the uncertainty of the location of the sensor increasesgiven the number of neighbors is less than (with a lower ratefor neighbors) and decreases otherwise at an exponential rateproportional to the number of neighbors.Free-ranging dogs generally exhibit territoriality as studied in[10] and those in the institute have been observed to do so inconsiderably small pieces of land such as individual hostelsand small strips of roads. Similarly, the three sensors in thesimulation are constrained within particular regions. . Results The mechanism is simulated in two-dimensional spacewithin the given dimensions. We initialize the sensors withsome positional uncertainty as shown in Fig. 1. On deacti-
Fig. 1. Initial Configuration - Without Tag Links vating links between tags (i.e., they do not behave as mutualanchors), there are larger periods when the number of links is due to which the uncertainty of some sensors is observed todiverge with time as depicted in Fig. 2.The second simulation involves links between neighboring Fig. 2. Final Configuration - Without Tag Links tags. This starts from the beginning of the simulation as shownin Fig. 3. For simplicity, we assume that the range of the tag
Fig. 3. Initial Configuration - With Tag Links sensors is equal to that of the anchors although, in a practicalscenario, this might not be the case since the former willconsist of simpler circuits and transceivers, hence resulting insmaller ranges. It is observed from Fig. 4 that the uncertaintyof the previously mentioned sensors does not diverge sincethe sensors are in contact with each other for several periodswhich previously had led to the divergence of uncertainty.IV. FUTURE SCOPEWearing traditional sensors for a long duration maybe detrimental to the health of the dog, and the sensing
Fig. 4. Final Configuration - With Tag Links quality may degrade from the environmental and hygienicconditions. Biosensors and flexible electronics may be moreappropriate long-term options. The long-term effects ofbiologic cybernetics and electronic stimulation on dogsis open to research although there has been progress inshort-duration studies on several animal species as shown in[11], [12] and [13].The simulation code related to this project is availableat github.com:aaronjohnsabu1999/Localization-Using-WSNsV. ACKNOWLEDGMENTSI am grateful to Prof. Siddharth Tallur for the guidance andsupport throughout this project as the course instructor for EE617 (Sensors in Instrumentation) at IIT Bombay.Rquality may degrade from the environmental and hygienicconditions. Biosensors and flexible electronics may be moreappropriate long-term options. The long-term effects ofbiologic cybernetics and electronic stimulation on dogsis open to research although there has been progress inshort-duration studies on several animal species as shown in[11], [12] and [13].The simulation code related to this project is availableat github.com:aaronjohnsabu1999/Localization-Using-WSNsV. ACKNOWLEDGMENTSI am grateful to Prof. Siddharth Tallur for the guidance andsupport throughout this project as the course instructor for EE617 (Sensors in Instrumentation) at IIT Bombay.R