Smart Sensors and Sensor Networks Lecture 5 Localization

Smart Sensors and Sensor Networks Lecture 5 Localization and positioning 1

Smart Sensors and Sensor Networks n n n Localization and positioning is useful, even necessary in certain applications; for ex. in tracking or event – detection applications the WSN must provide the information where the event happened; The most well known example of localization and positioning system is the GPS; it uses 28 satellites to enable global three – dimensional positioning services; It is not suitable for WSNs for several reasons: q q q n It is not operating in indoor environments; Sometimes, higher accuracy is necessary in the positioning result; SNs have battery constraints which requires special design; Location information can be used to improve the performance of WSNs and provide new types of services: q q It can facilitate routing to reduce routing overhead; geographic routing; through location – aware network protocols, the number of control packets can be reduced; New services: navigation, follow – me, geocast, temporal geocast; 2

Smart Sensors and Sensor Networks n Properties of localization and positioning procedures: q q Physical position vs. symbolic position; Absolute vs. relative coordinates: n n n q Localized vs. centralized computation: n n q An absolute coordinate system is valid for all objects and for any place on earth (e. g. UTM); to provide absolute coordinates, a few anchors (beacons, landmarks) are necessary (at least 3 for a 2 – dimensional system); Relative coordinates can differ for any located object or set of objects; Anchors can rotate, translate and scale a relative coordinate system to make it coincide to an absolute coordinate system; Computations might be done locally or measurements might be reported to a central station that computes locations or positions and sends them back to the participants; Sometimes, it might not be desirable for a node to reveal its position to a central entity; Limitations: n n For some positioning techniques, there are inherent deployment limitations: GPS, for example, does not work indoor; Other systems have only limited ranges over which they operate; 3

Smart Sensors and Sensor Networks q Accuracy and precision: n n q There are the most important parameters of a localization system; Positioning accuracy is the largest distance between the estimated and the true position of an entity; high accuracy indicates a small maximal mismatch; Precision is the ratio with which a given accuracy is reached, averaged over many repeated attempts to determine a position; For ex. a system could claim to provide a 20 cm accuracy with at least 95 % precision; they form the accuracy/ precision characteristic of a system; Scale: n n A system can be intended for different scales, for example, in indoor deployment: the size of a room, of a building, or, in outdoor deployment: a parking lot, a geographical area etc. Two important metrics are: q q q The area the system can cover per unit of infrastructure; The number of locatable objects per unit of infrastructure per time interval; Costs: n Positioning systems cost in time (infrastructure installation, administration), space (device size, space for infrastructure), energy (during operation) and capital (price of a node, infrastructure installation, maintenance); 4

Smart Sensors and Sensor Networks Localization and positioning approaches n There are 4 main approaches to determine a node’s position: q q n Proximity: information about a node’s neighborhood is used; Triangulation and trilateration: exploits the geometric properties of a given scenario; Fingerprinting: the received signal pattern is compared against the training patterns in a database; Scene analysis: tries to analyze characteristic properties of the position of a node in comparison with premeasured properties; Proximity q q The simplest technique is to exploit the finite range of wireless communication: it can be used to decide whether a node that wants to determine its position is in the proximity of an anchor; This only provides coarse – grain information but this can be sufficient; Proximity information from several overlapping anchors can be analyzed; One example is the natural restriction of the infrared communication by walls; it can be used to provide a node with simple location information, for example if it is in the room or not; 5

Smart Sensors and Sensor Networks q n Proximity – based systems can also be relatively robust to the uncertainties of the wireless channel; deciding whether a node is in the proximity of another node means connectivity, which is useful on long time scales, averaging out short – term fluctuations; Trilateration and triangulation q q q The communication between two nodes allows to extract information about their geometric relationship; When distances between entities are used, the approach is called lateration and when angles are considered, the approach is called angulation; Trilateration: n 3 anchors are needed for lateration in plane; 6

Smart Sensors and Sensor Networks n n n The anchors have known locations and they transmit signals; if a node can hear the signals it will be able to determine its position by trilateration, as the intersection of the three circles which have as centers the anchors and as radius the distances from the node to the anchors: The problem is that, in reality, distance measurements have errors, leading to location errors, because the intersections of the circles will not be a point; A solution is to use the maximum likelihood method: q Starting from the known positions of the three anchors, for any point (x, y) on the plane, a difference function is computed: q r. A, r. B, r. C are the estimated distances to A, B and C, respectively; The location of the object can be predicted as the point (x, y) among all points such that σx, y is minimized; q n Another solution is multilateration, meaning more than three anchors are used for determining a node’s position in plane; 7

Smart Sensors and Sensor Networks n There is a mathematical support for the trilateration technique: q q Lets consider three anchors with known positions, (xi, yi), with i = 1, 2, 3, a node with unknown position, (xu, yu) and perfect distance values ri, i = 1, 2, 3; The Pythagoras theorem gives a set of three equations: q To solve this set, it is more convenient to rewrite it as a set of linear equations in xu and yu: q Rearranging the term gives: q It results a linear matrix equation, where the matrix of the left side and the right side are made only of known constants: 8

Smart Sensors and Sensor Networks q Angulation: n n It can be the angle of a connecting line between an anchor and a position – unaware node to a given reference direction (e. g. 00 north); It can also be the angle between two such connecting lines if no reference direction is commonly known to all nodes; A solution consists in using directional antennas, rotating on their axis, similar to a radar station; such devices are inappropriate for sensor nodes, they are useful for anchors; Another solution is to exploit the finite propagation speed of all waveforms; with multiple antennas mounted on a device at known separation and measuring the time differences between a signal’s arrival at the different antennas, the direction from which a wavefront arrived at the device can be computed; the smaller the antenna separation, the higher the precision of the time differences has to be, which results in strenuous timing requirements given the desirable small size of sensors; 9

Smart Sensors and Sensor Networks n Fingerprinting: q q Instead of estimating the distance between an anchor and a device, this approach tries to compare the received signal pattern against the training patterns from a database; It relies on the fact that signal strength received at a fixed location is not a constant, but rather it can be modeled by a random variable; The main idea is to compare the received signals against those in the database and determine the likelihood that the device is currently located in a position; A comparing solution has two phases: n Off – line phase: in this phase, signals are collected from all training locations; the number of training locations is decided first, then, the received signals strength are recorded; each entry in the database shows the location coordinates and the signal strength pattern; a training location may contain more than one node and for each of them a different entry can be generated; for higher accuracy one may establish multiple entries in the database for the same training location; from the database, some positioning rules, which form the positioning model, will then be established; 10

Smart Sensors and Sensor Networks n n Real – time phase: the positioning model may determine a number of locations, each associated with a probability; the typical solution is to output only the location with the highest likelihood; Each entry in the database has the format (x, y, ), where (x, y) is the coordinate of the training location and ssi, i = 1, …, n is the signal strength received at the training location from node i; 11

Smart Sensors and Sensor Networks q Nearest Neighbor Algorithms n n n n The simplest approach is the nearest neighbor in signal space (NNSS) approach; In the first phase, only the average signal strength of each node from each training location is recorded; In the second phase, the NNSS algorithm computes the Euclidian distance in signal space between the received signal and each record in the database; Euclidian distance means the square root of the summation of square of the difference between each received signal strength and the corresponding average signal strength from the access point under consideration; The training location with the minimum Euclidian distance is then chosen as the estimated location of the device; Because this algorithm only picks existing locations in the database, to improve its accuracy, it is suggested that the remaining set be dense enough; NNSS – AVG: takes the uncertainty of a device’s location into consideration: q q It picks a small number of training locations (not just one) that closely match the received signal strength (such as those with smaller Euclidian distances); It infers the location of the device to be a function of the coordinates of the selected training locations; for example, one may take the average of the x and y coordinates of the selected training locations as the estimated result; 12

Smart Sensors and Sensor Networks q Probability – Based Algorithms n n n n This approach regards signal strength as a probability distribution; In NNSS, because the received signal strengths are averaged out, the probability distribution would disappear; The probability – based approach maintains more complete information of signal strength distribution; the prediction result is typically more accurate; The core of the probability – based model is the Bayes rule: p(l|o) is the probability that the device is at location l given an observed signal strength pattern o; p(l) is the prior probability that a device is resident at l; it may be inferred from history or experience; for example, people may have a higher probability to appear in a hallway; if this is not available, p(l) may be assumed to be a uniform distribution; L is the set of all training locations; The denominator p(o) does not depend on the location variable l, so it can be treated as a normalized constant whenever only relative probabilities are required; 13

Smart Sensors and Sensor Networks n n n The term p(o|l) is called the likelihood function; this represents the core of the positioning model and can be computed in the off – line phase; 2 ways to implement it: kernel method and histogram method; Kernel method: q For each observation oi in the training data, it is assumed that the signals strength exhibits a Gaussian distribution with mean oi and standard deviation σ, where σ is an adjustable parameter in the model; given oi, the probability to observe o is: q Based on the kernel function, the probability function p(o|l) can be defined as: where n 1 is the number of training vectors in L obtained at location l; q The probability function is a mixture of n 1 equally weighted density functions; q The formulas were derived assuming a single node in a training location; with multiple nodes, the probability will become the multiplication of multiple independent probabilities, each for a node from the training location; n Histogram method: continues values are transformed in discrete ones; by this the instability of signal strengths are smoothed out; 14

Smart Sensors and Sensor Networks n Scene analysis: q q n A well known implementation consists in analyzing pictures taken by a camera and trying to derive the position from the picture; It requires substantial computational effort and is hardly appropriate for sensor nodes; Techniques for determining distances: q Received Signal Strength (RSS) n n n The idea is to use the property of signal degradation while traveling in a space to determine the mutual distance; Assuming that the transmission power, Ptx, the path loss model and path loss coefficient α are known, the receiver can determine the distance d from the received signal strength Prcvd: No additional hardware is necessary and distance estimates can be derived without additional overhead from communication that is taking place anyway; 15

Smart Sensors and Sensor Networks n The problem is that RSS values are not constant but can heavily oscillate, even when sender and receiver do not move; n A trend for the relation between the distance and signal strength does exist, however, the curve is unstable in small regions; many uncontrollable environmental factors (e. g. rain, fog etc. ) are present; a solution is to repeat measurements and filter incorrect values by statistical techniques; Another cause may be the presence of obstacles in combination with multipath fading; the signal attenuation along an indirect path, which is higher than along a direct path, can lead to incorrectly assuming a longer distance than what is actually the case; this is a structural problem, so it cannot be solved by repeated measurements; It is necessary to model the error for signal attenuation; the path loss function will include a random variable corresponding to the errors; n n 16

Smart Sensors and Sensor Networks q Time of Arrival (To. A) n n n Signal travelling time is used to estimate the distance between a device and a reference point; usual, slow signals are used, such as ultrasound; A signal is sent from the transmitter to the receiver and, in turn, this sends back a signal; the transmitter can infer the distance from the round – trip delay of the signals: The error of this technique comes from the processing time of signals (computing latency) and the unknown delay T 2 – T 1 at the receiver’s side; Slow signals are needed because high resolution clocks are necessary; One disadvantage of the sound is that its propagation speed depend on external factors, such as temperature and humidity; 17

Smart Sensors and Sensor Networks q Time Difference of Arrival (TDo. A) n n n The technique uses two signals that travel at different speeds, such as the radio wave and the ultrasound; Transmission in one direction is sufficient; at T 0 the sender generates an RF signal, followed by an ultrasound signal at time T 2; the receiver can determine its distance to the transmitter by: In a simpler implementation, the arrival of the RF signal at the receiver starts a counter for measuring the time until arrival of the ultrasound; the propagation time of the radio communication is ignored which is a realistic approach taking into account the differences between them; Advantage: a better accuracy; Disadvantages: q q q The need of two types of senders and receivers; The receiver must know the precise value of T 2 – T 0 to determine the distance; Angle of Arrival (Ao. A) n The angle and orientation of received signals is determined and for that an antenna array or an array of ultrasound receivers are necessary; 18

Smart Sensors and Sensor Networks n An example; n Anchor nodes are used that generate narrow, rotating beams where the rotation speed is constant and known to all nodes; Nodes can measure the time of arrival of each such beam, compute the difference between two consecutive signals and determine the angles α, β and γ using geometric relationships; The challenge is mainly to ensure that the beams are narrow enough (less than 150 are recommended) so that the nodes have a clear triggering point for the time measurements and to handle effects of multipath propagation; Advantages: n n n q q q The results are not affected by the network density; No extra traffic is generated in the network; Simulations have given good accuracy: maximum 2 m error in a 75 m x 75 m area; 19

Smart Sensors and Sensor Networks Localization and positioning systems n Global Positioning System q q q The GPS determines one’s exact location and precise time anywhere on earth at any time; It relies on 28 satellites orbiting on six different planes so that a minimum amount of 4 planes can be seen from any point on the earth; Each satellite vehicle, SV, transmits exact position and precisely synchronized on – board clock time in a spread spectrum signal; A GPS receiver measures the signal transit times between its point of observation and at least 4 different satellites whose positions are known to be able to solve for the 4 unknowns: longitude x, latitude y, altitude z and time deviation Δt; No reverse uplink communication is necessary between the transceiver and the satellites; the GPS system references the earth centered geoid; Typical performance is given by the following data: 160 m. W power consumption, 3 m accuracy, 20 ns timing precision, 41 s cold start, 3. 5 s hot start and a 4 Hz update rate at a size (without antenna) of 1. 9 cm 3; 20

Smart Sensors and Sensor Networks q q n GPS receivers are integrated in handheld devices and mobile phones; GPS is not appropriate for SNs because of high power consumption, high cost, rather high dimensions and because of the requirement for direct line of sight to satellites; The Lighthouse Location System q q q It is based on direct line of sight between fixed infrastructure laser transmitters and the mobile unit; Each transmitter emits a laser beam that is rotated in two perpendicular axes, thus scanning a whole room; The mobile unit registers the phase and duration of light flashes and uses this information to intersect three hyperboloids, each in reference to the transmitter’s position; This approach is unique because high precision can be achieved with relatively low system, communication and computational complexity on the sensor nodes; The line of sight requirement and the extensive calibration necessary prior to usage make this system rarely used in SNs; 21

Smart Sensors and Sensor Networks Localization and positioning systems for WSNs n They can be divided in single – hop and multihop localization systems; n Single – hop positioning systems q Active Badge: n n q Is a cell – based location system in which objects are attached with a badge that periodically emits infrared signals with a unique ID; Infrared receivers mounted at known positions collect these signals and relay them over a wired network; The system knows in which infrared cell a badge currently stays; The disadvantages are that it is hard to deploy in a large – scale environment and that infrared is sensitive to external light; also, the infrared waves are stopped by walls; Active Bat: n n It consists in a collection of wireless transmitters, a matrix of receiver elements and a central RF base station; The wireless transmitters, called bats, can be carried by a tagged object or attached to an equipment; 22

Smart Sensors and Sensor Networks n n n n q The sensor system measures the time of flight of the ultrasonic pulses emitted from a bat to receivers installed in known and fixed positions; It uses the time difference to estimate the position of each bat by trilateration; The RF base station coordinates the activity of bats by periodically broadcasting messages to them; upon hearing a message, a bat sends out an ultrasonic pulse; A receiver that receives the initial RF signal from the base station determines the time interval between receipt of the RF signal and receipt of the corresponding ultrasonic signal; it then estimates the distance from the bat; These distances are sent to the computer, which performs data analysis; By collecting enough distance readings, it can determine the location of the bat within 3 cm of error in a three – dimensional space at 95 % accuracy; Disadvantage: the deployment cost is high; RADAR n n Takes advantage of the already existing RF data network formed by IEEE 802. 11 access points; IEEE 802. 11 networks are now becoming more prevalent in many office and public areas, so no extra hardware is necessary; The received signal characteristics from multiple anchors is compared with premeasured and stored characteristic values; 23

Smart Sensors and Sensor Networks n q Both the anchors and the mobile device can be used to send the signal, which is then measured by the counterpart device; Cricket n n n Is a system that can provide location – dependent applications; Unlike the Active Badge system in which the infrastructure determines the position of a mobile device, in the Cricket System the mobile device computes its own position; this is useful when privacy issues become relevant; Cricket does not rely on any centralized management or control and no explicit coordination occurs between beacons (anchors); To obtain information about a space, every object is attached to a listener, a small device that listens to messages from beacons mounted on ceiling and walls; Similar to the Bat system, Cricket uses a combination of an RF signal and ultrasound to evaluate the distances between beacons and listeners; A beacon sends the space information over an RF and an ultrasonic pulse at the same time; when the listener hears the RF signal, it uses the first few bits as training information and then turns on its ultrasonic receiver; it waits for the ultrasonic pulse and upon its arrival will compute the distance from the beacon using the time difference; 24

Smart Sensors and Sensor Networks q Overlapping connectivity: n n It is an outdoor positioning system that operates without any numeric range measurements; it uses only the observation of connectivity to a set of anchors to determine a node’s position; The underlying assumption is that transmissions, of known and fixed transmission power, from an anchor can be received within a circular area of known radius; Anchor nodes periodically send out transmissions identifying themselves, or, equivalently, containing their positions; once a node has received these announcements from all anchors of which it is in reach, it can determine that it is in the intersection of the circles around these anchors; The estimated position is then the arithmetic average of the received anchors’ positions; 25

Smart Sensors and Sensor Networks n n n q Moreover, assuming that the node knows about all the anchors that are deployed, the fact that some anchor announcements are not received implies that the node is outside the respective circles; this information further allows to restrict the node’s possible position; The achievable absolute accuracy depends on the number of anchors, more anchors allow a finer – grained resolution of the area; At 90% precision, the relative accuracy is one – third the separation distance between two adjacent anchors, assuming that the anchors are arranged in a regular mesh and that coverage area of each anchor is a perfect circle; In a 10 m x 10 m area, the average error is 1. 83 m; in 90% of the cases, positioning error is less than 3 m; Accuracy degrades if the real coverage range deviates from a perfect sphere, as it usually does in reality; Approximate point in triangle: n n n The technique is based on connectivity information; The idea is to decide whether a node is within or outside of a triangle formed by any three anchors; Using this information, a node can intersect the triangles and estimate its own position, similar to the intersection of circles; 26

Smart Sensors and Sensor Networks n n n The node has detected that it is inside the triangles BDF, BDE and CDF and also that it is outside the triangle ADF, ABF, AFC and others; Hence, it can estimate its position in a restricted area, for example in the area’s center of gravity; The problem is how to decide whether a node is inside or outside the triangle formed by any three arbitrarily selected anchors; The intuition is to look at what happens when a node inside a triangle is moved: irrespective of the direction of the movement, the node must be closer to at least one of the corners of the triangle than it was before the movement; conversely, for a node outside a triangle, there is at least one direction for which the node’s distance to all corners increases; Moving a sensor node to determine its position is hardly practical; but one possibility to approximate movements is for a node to inquire all its neighbors about their distance to the given 3 corner anchors, compared with the enquiring node’s distance; 27

Smart Sensors and Sensor Networks n n If, for all neighbors, there is at least one corner such that the neighbor is closer to the corner than the enquiring node, it is assumed to be inside the triangle, else outside; deciding which of two nodes is closer to an anchor can be approximated by comparing their corresponding RSS values; Both the RSS comparison and the finite number of neighbors introduce errors; for example, for a node close to the edge of the triangle, there is a chance that the next neighbor in the direction toward the edge is already outside the triangle, incorrectly leading the enquiring node to assume this also; Therefore, the approach is likely to work better in dense networks where the probability of such kinds of errors is reduced; Nonmonotonic RSS behavior over distance is another source of error; 28

Smart Sensors and Sensor Networks n Multihop positioning systems: q Multihop range estimation n n n The basic multilateration approach requires a node to have range estimates to at least three anchors to allow it to estimate its own position; One considers the problem when anchors are not able to provide such range estimates to all nodes in the network, but only to their direct neighbors, because of, for example, limits on the transmission power; The idea is to use indirect range estimation by multihop communication to be able to reuse the multilateration algorithms; At least three possibilities are presented in the literature ; all of them are based on flooding the network with information, independently starting from each anchor, similar to the operation of a distance vector (DV) routing protocol; The simplest possibility is the DV – hop method; The idea is to count the number of hops, along the shortest path, between any two anchors and to use it to estimate the average length of a single hop by dividing the sum of distances to other anchors by the sum of the hop counts; every anchor computes this estimated hop length and propagates it into the network; 29

Smart Sensors and Sensor Networks n n n A node with unknown position can then use this estimated hop length and the known number of hops to other anchors, to compute a multihop range estimate and perform multilateration; When range estimates between neighboring nodes are available, they can be directly used in the same framework, resulting in the DV – distance method; In presence of range estimates and a sufficient number of neighbors, a node can try to compute its Euclidian distance to a faraway anchor; q n n Assuming that the distances AB, AC, BC, XB, XC are all known, it is possible to compute the unknown distance XA; actually there are 2 situations depending on the position of node X to line BC; node X can distinguish these two situations based on local information; The obtainable accuracy depends on the ratio of anchors relative to the total number of nodes; the Euclidian method increases accuracy as the number of anchors increases; the DV – like methods are better suited for a low ratio of anchors; The DV – like methods perform less well in anisotropic networks than in uniformly distributed networks; the Euclidian method is not very sensitive to this effect; 30

Smart Sensors and Sensor Networks q Iterative and collaborative multilateration n This solution uses normal nodes, once they have estimated their positions, just like anchor nodes in a multilateration algorithm; n Nodes A, B and C are unaware of their position; node A can triangulate its own position using 3 anchors; once node A has a position estimate, node B can use it and 2 anchors for its own estimate; node C will need only 1 anchor; A centralized implementation is trivial, starting with the node having the most connections to anchors and iteratively computing the rest of the positions; n 31

Smart Sensors and Sensor Networks n n n n In a distributed implementation, nodes can compute a position estimate once at least 3 neighbors can provide position information, resulting in an initial estimate of a node’s position; When more information becomes available, for example because more neighbors have estimated their own positions, it is possible to improve the position estimate and propagate an uploaded estimate to a node’s neighbor; The hope is that this algorithm will offer the correct positions for all nodes; The average position error after such an iterative refinement depends on the accuracy of the range estimation, the initial position estimate, the average number of neighbors and on the number of anchors; Also, it is not guaranteed that the refinement algorithm converges at all; there are situations where the position error increases the longer the algorithm runs; An improvement can be to add confinement weights to all position estimates and to solve a modified weighted optimization problem; One particular challenge to this class of algorithms occurs when not all nodes in the network will have 3 nodes with position estimates near them; this is easy to detect for a single node, but difficult to do for an entire group of nodes; however, depending on the topology, it might still be possible to estimate at least some positions by collaborative multilateration; 32

Smart Sensors and Sensor Networks n n n In fig. , it is impossible to determine node 2’s and node 4’s locations even if the locations of nodes 1, 3, 5 and 6 are known; Two potential locations for node 2 can be obtained from beacons 1 and 3; Similarly, two potential locations for node 4 can be obtained from 5 and 6; Collaborative multilateration allows estimation of the distance between nodes 2 and 4; three situations are possible depending on the distance (RSS) from nodes 2 and 4; Another example: 33

Smart Sensors and Sensor Networks q CSIE/ NCTU Indoor Tour Guide n n This is a prototype indoor tour guide system developed at the Department of Computer Science and Information Engineering, National Chiao Tung University, Taiwan; The hardware platforms include several; Compaq i. PAQ PDAs and laptops; Each mobile station is equipped with a wireless card (Lucent Orinoco Gold); Signal strengths are used for indoor positioning and the probability – based pattern – matching algorithm is used; 34

Smart Sensors and Sensor Networks n n n The concept of logical areas is used to identify offices, rooms etc. The manager is the control center responsible for monitoring each user’s movements, configuring the system and planning logical areas and events; The location server takes care of the location discovery job and the service center is in charge of message delivery; The database can record users’ profiles; The gateway can conduct location – based access to the Internet; One of the innovations is that an event – driven messaging system has been designed: q q n A short message can be delivered to a user when he enters or leaves a logical area; The event – driven message can also be triggered by a combination of time, location and property of location (such as who is in the location and when is it reserved for meetings); A user can set up a message and a corresponding event to trigger the delivery of the message; Messages can be unicast or broadcast; Another innovation is to provide location – based access control: q q In certain rooms, such as classrooms and meeting rooms, users may be prohibited from accessing certain sensitive Web pages; These rules can be organized through the manager and set up at the gateway. 35