2-D RSSI-Based Localization in Wireless Sensor Networks

Transcription

1 2-D RSSI-Based Localization in Wireless Sensor Networks Wa el S. Belkasim Kaidi Xu Computer Science Georgia State University Abstract Abstract in large and sparse wireless sensor networks many nodes communicate data across other nodes to a base station for routing or immediate processing. These topologies are often deployed in a coarse-grained or random manner that does not take into account optimal node placement or node location. In the case of node failures, it is difficult to assess the source of the problem and thus correct it. In this regard a cost effective and flexible localization solution is needed. We implement a system that uses RSSI and trilateration for node localization within a wireless sensor network using TelosB motes and TinyOS. The proposed solution is implemented in nesc and JAVA and evaluated in terms of functionality and efficiency. 1. Introduction Wireless Sensor Networks (WSN) are composed of many sensor nodes, sometimes deployed in sparse configurations to maximize the coverage area of the WSN. The nodes are often prone to failure due to environmental hazards as well as the fact that sensor nodes are generally designed to be cheap, efficient, and replaceable, making fault tolerance a pillar of any good WSN. In such a scenario depending on absolute position systems for localization is not cost effective due to the power consumption and deployment requirements. GPS systems are large and costly to mount on a sensor node, significantly increasing the cost and size and power consumption of a mote. Although it offers precise positioning and localization it cannot be considered a good solution in terms of efficiency and simplicity. An alternative to that is using the on-board capabilities present on the majority of sensor mote radio platforms: Received Signal Strength Indicator (RSSI.) RSSI allows for a cost effective and simple solution to localization with the draw backs of being inaccurate and unpredictable especially in indoor environments. The challenge is creating a system which can exploit the strengths of RSSI localization while minimizing the weakness and unreliability of RSSI. Figure 1. Example WSN Deployment from the Harvard Sensor Lab [1]. 2. Related Work Currently WSN research focuses on the problems of localization, optimizing sensor coverage, increasing connectivity between nodes, or any combination there-of. One work aims to implement a sensor network in a coal mine environment with mobile robot nodes and reference nodes [2]. The paper uses a Received Signal Strength Indicator (RSSI) technique to triangulate and estimate the position of miner nodes within the coal mine WSN. Once a disaster occurs a robot can enter the WSN and proceed using static reference nodes for localization with the assumption that not all the nodes in the WSN are functional. As the robot moves into the WSN, the orphaned nodes can send location data to the robot node thereby relaying the latest coordinate information of a trapped/orphaned miner node. While the work focuses on navigating a robot

2 to trapped orphan miner nodes using RSSI, it does not focus on RSSI-based node localization in the sense of a positioning system. Another interesting paper which focuses on localization as well as robot navigation proposes that both problems are solved in tandem; that is to say simultaneously as the robot navigates it must also localize neighboring nodes [4]. As the robot moves, it communicates with its neighboring nodes by sharing its location and communication range, and in turn the neighbors calculate their estimated bounded location by using their own communication range and the previous position data broadcast by the robot. The nodes are localized on an event basis, so only when an event occurs will the robot attempt to navigate to the source of an event and thus localize the nodes along the path to that event. In comparison our approach does not use GPS but instead uses prepositioned or localized anchor nodes that can localize new nodes without the need for absolute positioning. request broadcast. Localization requests are sent by a node which has successfully collected RSSI replies to the base station to request that the base station localize it with the collected RSSI data. A localization reply is the acknowledgement of a successful or failure to localize the requesting node from the base station. Lastly the status message is sent by every node in the network except for the base station to allow the base station to monitor all the nodes in the WSN. 3. System Design Our solution aims to accomplish the following: node monitoring, node localization, and accurate graphical representation. To achieve these goals we approach the problem in layers. First we describe node communication and forwarding, and second we detail the methods used for localization. These subsystems are built on top of each other; each level acts as a foundation for the next. 3.1 Node Communication In our system we have two node types that exist within the WSN: Member Nodes and the Base Station. Nodes communicate via controlled multi-hop flooding. That is to say we do not allow duplicate packets and eliminate infinite packet transmission cycles or loops. Figure 2. Packet types used across our WSN. Figure 3. Contents of the packets used in our WSN Packet Model We employ several packet types in our system to facilitate communication between different nodes and different node types. RSSI requests are broadcast by un-localized nodes, or nodes that have lost communication to the rest of the WSN. RSSI replies are unicast messages sent in response to an RSSI Controlled Flooding Packets which may need to be forwarded are assigned sequence numbers and an array of visited nodes. The sequence numbers eliminate duplicates from being received in the case of a broadcast, and the array of visited nodes allows the packet to keep track of all the nodes which it has hopped across, or

3 flooded to, thus far. Packets are only broadcast to a node s 1-hop neighbors if a unicast to the base station from the sender is not possible. A node first tries to transmit a packet via unicast to the base station, if the base station is not within communication range, then the member node floods the packet to the base station via broadcasting to its 1-hop neighbors, and its neighbors broadcast to their 1-hop neighbors and so on. As the packet travels across the neighbors, each visited node is added to the visited node array of the packet before being forwarded, this stops nodes from continuously flooding the packet. The sequence number is to stop duplicates of the same broadcast hierarchy from being received and forwarded by 1- hop neighbor nodes. For example, in Fig. 4 if the middle node broadcasts to its 1-hop neighbors after receiving a packet from the bottom node, the bottom node will ignore the packet because the packet has already visited it. Then the top two nodes will broadcast it to their 1 hop neighbors, which include the middle node and themselves. The middle node will ignore the packet because it has already been marked as visited in that packet; however the top two nodes broadcast simultaneously and thus each node of the top level will only have itself recorded in the visited array. At this point the sequence numbers matter because the last received sequence number will prevent the top level nodes from broadcasting a message they have already received. Member nodes are the workhorse of our WSN; they forward packets, and drive the localization mechanism. Member nodes have two states: 1) localized 2) un-localized. Localized nodes are considered Anchor nodes, because they already have their position established sometime during or after deployment time. Un-localized nodes exist only when new nodes are added to the network, or old nodes are reset or their signal (to the base station) is lost due to external environment factors or otherwise. Nodes which are in an un-localized state will broadcast RSSI requests periodically at a defined application interval of 1000ms. Any localized (Anchor) nodes within range of the broadcasts will respond in unicast to the RSSI requests within 250ms. Regardless of a node s state, it is always able to participate in the multi-hop forwarding process, the delay between receiving and forwarding a message is also 250ms. Figure 5. Communication between member nodes whom are localized (green) and un-localized (red.) RSSI request beacon from red node and RSSI replies from green nodes. 3.2 Localization of Nodes Figure 4. Packet flooding example: Bottom level node broadcasts in blue, middle level node broadcasts in orange, top level nodes broadcast in red Node States The simplest way to get a sensor s location is using a Global Positioning System (GPS). The GPS system is capable of triangulating its location via signal from four or more satellites, these signals contain time and other data that the system will use to eventually accurately calculate its current location. However, if we mount GPS systems on all of the nodes in a WSN deployment, it will significantly increase the energy consumption and shorten node

4 life span. The GPS system will also increase the sensor s cost and the mounting of the GPS system will increase the sensor s size. These are the disadvantages for GPS localization in WSN, and we need to find a more cost efficient and practical way to get a sensor s location RSSI Distancing Research has shown that there is a strong relationship between Received Signal Strength (RSSI) and distance [5] [6]. RSSI is the relative received signal strength in a wireless environment, the higher the RSSI value, the stronger the signal. Along with RSSI we can use Trilateration techniques to determine, or approximate a node s position using 3 other known nodes location and the distances from the unlocalized node to the known nodes. The relationship between distance and RSSI is shown in (1) where A is the transmission power at 1 meter away and d is the distance and n is the path loss coefficient. RSSI = - (A+10n (log 10 (d))) (1) The major problem with RSSI distancing that makes it viable only as an approximation is due to uncontrollable environmental factors that heavily undermine the distance relationship modeled by (1). In our tests we found that RSSI is not a reliable measurement of distance especially in indoor environments mostly because of multi-path, fading, node proximity, temperature, humidity and other factors. Nevertheless we used RSSI distancing to approximate distance between nodes and eventually localize a node using trilateration D Trilateration Assuming we know the location of at least 3 anchor nodes within 1-hop of an un-localized target node. If we also know the distance from the target node to the 3 anchor nodes, we can effectively calculate the position of the target using trilateration [7]. Figure 6. Trilateration of an un-localized (red) node using three anchor nodes (green) with position and distance [7]. Assuming we know the location of at least 3 anchor nodes we can formulate the position of the target node into (2) using triangle geometry and then solve the resulting system of equations: r 2 1 = (x - a 1 ) 2 + (x - b 1 ) 2 r 2 2 = (x a 2 ) 2 + (x b 2 ) 2 r 2 3 = (x a 3 ) 2 + (x b 3 ) 2 (2) After which we can obtain the X, Y coordinates: y = [(a 2 - a 1 ) + (a 2 3 +b 2 3 -r 2 3 ) + (a 1 a 3 ) (a 2 2 +b 2 2 -r 2 2 ) + (a 3 a 2 ) (a 2 1 +b 2 1 -r 2 1 )] / [2[b3 (a 2 -a 1 ) + b2 (a 1 -a 3 ) + b3 (a 3 -a 2 )]] x = [r a b 2 1 r a b 2 2 2(b 1 b 2 ) y] / [2(a 1 a 2 )] (3) 4. Implementation & Results The system was implemented on TinyOS in nesc and in Java to facilitate WSN control and graphical representation. The topology communication works as such: TelosB nodes TelosB Base Station Java GUI and Controller. For the Java application, the MoteIF interface was used to communicate via serial with the TelosB base station, with the Java application acting as the bridge between user interaction and the

5 WSN. The application is robust enough to catch basic user errors such as not positioning nodes when an unlocalized node is requesting localization as well as adding node IDs which do not exist in the WSN topology. The left-hand screen displays the status of nodes as well as if/when nodes are added and where they are added on the grid, it also shows the intermediate steps of distance calculations before the trilateration and localization is completed. Any nodes which cannot communicate with the base station will time out and be marked in red so that the user can react and fix/replace those nodes. 4.1 Cartesian vs. Real vs. Screen Coordinates To correctly implement the proposed system we had to model actual ground coordinates into screen coordinates. Because our solution aims to localize and pinpoint nodes on a physical map, we had to create a custom GUI interface that actually meant something in the real world. We implemented an animated grid in Java Swing that modeled a 2D horizontal plane, particularly each grid square of 25x25 pixels equated to a real world square of 15x15 centimeters. Another distinction is that screen coordinates are not the same as Cartesian coordinates, so before and after trilateration we added a conversion between Raster and Cartesian coordinates, as well as a conversion between pixels to distance (cm) and vice versa. Our grid on screen is 400 by 400 pixels with 16 square boxes across its width and height. Thus our actual grid in the real world is 240cm by 240cm. This can of course be changed in our program; in fact the GUI grid can be re-sized in real time however that is meaningless unless the physical grid and node locations are also changed to match. 4.2 RSSI-Distance Regression Modeling To model the distance progression for RSSI we determined that our best option was to measure RSSI & distance pairs across several test environments and then fit the data to a linear or logarithmic regression. In our tests we used a transmission power rate of dBm which is power level 5 for the CC2420 in TinyOS. The following figures show the measurements in a classroom environment as well as our own test environment. We use the results from Table for our evaluation. Table Classroom Data [5] 19dBm [3] 25dBm CM RSSI CM RSSI CM RSSI CM RSSI Figure 7. The grid is 16x16 squares of 25x25 pixels each that represent a 240x240cm real world grid. -19dBm set in Logarithmic (natural log) Regression

6 Table Test Environment Data [5] 19dBm Standard Using Reflectors CM RSSI CM RSSI Standard set in Logarithmic (natural log) Regression Reflectors set in Logarithmic (natural log) Regression 4.3 RSSI Accuracy & Multi-path In many cases RSSI is simply not a viable solution for localization due to the dreaded multi-path issue. RSSI is just not stable enough to be a reliable measurement, coupled with trilateration and requiring the target node to be placed within the confines of a right triangle our solution s ability to localize nodes in unpredictable environments is limited (Figures 8/9 show an accurate localization while Figures 10/11 show an inaccurate one.) However, if we can assume that there will be at least 4 anchor nodes that surround 1 un-localized node, and that the RSSI remains stable and conforms to a linear or logarithmic progression, then we can safely assert that a node would always be accurately localized within a ~15-45cm rate of error. The reason we require 4 nodes is that with 4 nodes in every corner of a grid square an un-localized node can always be surrounded by more than 1 node, whereas with 3 nodes, an un-localized node can fall on the outer edges of the triangle, near the anchor nodes that are furthest away from the 90 degree angle. To test this theory we tried to eliminate the issue of multi-path, or at least limit it to some degree by using

7 parabolic reflectors behind the radio of each anchor node. At the same time we directed all the anchor nodes towards the un-localized node when localization began, we saw a steady progression in RSSI with little to no spikes in the measurements. Likewise, the localization became more accurate more often (as long as it stayed within the triangle and was nearer to the anchor node at the 90 degree angle (shown in Figures 12 & 13.) Figure 8. The red square shows where the node is in the real world, it corresponds to figure 9. Here the node was accurately localized. Overall our proposed system is able to effectively monitor, communicate with, and maintain the nodes within a WSN. Localization is an approximation using an inaccurate means; however we can still approximate the distance and location of a node within ~10 centimeters at best and ~100cm at the worst. The main issue with RSSI is its instability regardless of the localization algorithm used. Nothing can make up for the inherent unpredictability of RSSI except if we constrain the signal to a certain direction and angle of propagation thereby reducing multi-path and path loss, but at the same time limiting the field of radio communication. It is a trade-off which only further proves that RSSI simply cannot do more than what it s meant to do. That being said RSSI is still cost efficient because it is ubiquitous on most radio platforms so it can be a simple and quick solution for approximate localization. We can conclude that under certain conditions and environmental constraints RSSI is still an effective localization tool unless very precise positioning is needed. Figure 10. The red square shows where the node is in the real world, it corresponds to figure 11. Here the node was incorrectly localized due to RSSI instability. Figure 9. This figure corresponds in scale to figure 8. The bottom node here is the same node at the origin in the GUI (top left) of figure Conclusion & Discussion

8 Figure 11. This figure corresponds in scale to figure 10. Figure 12. This figure shows a 2 nd trial of the same test as Figure 10 and Figure 11 but in this case the anchor nodes have reflectors as seen in figure 13. Figure 13. This figure corresponds in scale to figure References [1] [2] Pei, Z.; Deng, Z.;, "A distributed location algorithm for underground miners based on rescue robot and coal-mining wireless sensor networks," Robotics, Automation and Mechatronics, 2008 IEEE Conference on, vol., no., pp , Sept [3] Houaidia, C.; Idoudi, H.; Saidane, L.A.;, "Improving connectivity and coverage of wireless sensor networks using mobile robots," Computers & Informatics (ISCI), 2011 IEEE Symposium on, vol., no., pp , March 2011 [4] Shenoy, S.; Jindong Tan;, "Simultaneous localization and mobile robot navigation in a hybrid sensor network," Intelligent Robots and Systems, (IROS 2005) IEEE/RSJ International Conference on, vol., no., pp , 2-6 Aug [5] 1. J. Xu, W. Liu, F. Lang, Y. Zhang and C. Wang, "Distance Measurement Model Based on RSSI in WSN,"Wireless Sensor Network, Vol. 2 No. 8, 2010, pp [6] 4. Saxena, M. Gupta, P. ; Jain, B.N. Experimental analysis of RSSI-based location estimation in wireless sensor networks in Communication Systems Software and Middleware and Workshops, COMSWARE rd International Conference on [7] Chris Grover; Radhika S. Grover (2 June 2011). Programming With Java: A Multimedia Approach. Jones & Bartlett Publishers. pp. 79