Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren
Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren
Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren Autor: Miguel A. Iribarren
Lehrstuhl Telekommunikation
Introduction
In recent years, Wireless Sensor Networks enjoy a great interest both from the scientific community and the industrial sector. Advances in wireless communications and electronics have made viable the development of multifunctional sensor networks low-cost and low consumption. These tiny sensors are capable of measuring physical parameters, process information and communicate with other devices through a wireless channel. The fields of application of WSNs are very varied: telematics, precision agriculture, exploration of habitats, disaster detection and performance, intelligent buildings, government facilities management, health and medical monitoring. It is clear that one of the critical aspects in most applications is to relate the physical data measured with a concrete position. That is, each device must know its exact position at any instant of time. That is why localization algorithms for WSNs have become a very popular research area of study. The cooperation and information exchange between the nodes of the network make possible to execute distributed algorithms for estimating the location of the nodes. However, the hardware limitations of the devices do not allow reaching a precision required for many applications. To that end, through increased memory and processing capacity of an external entity, the use of centralized proposals might implement more complex algorithms that achieve greater accuracy in estimated locations.
Performance Evaluation
When information about the possible trajectories of the end device is available, the system can estimate much better positions and the tracking of the mobile node. Now Particle Filter infers the estimated positions by using Fingerprint database as well as the knowledge of the end devices dynamic. One parameter of the algorithm is responsible for assigning the percentage for every method. The figure below show the results when the percentage associated to end devices dynamic method decreases. In this case, it can be appreciated that the accuracy is greater when the knowledge of the end devices user is computed with a greater importance.
40% 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 5,5 6 6,5 7 7,5 8 8,5 9 9,5 10 10,5 11 11,5 12 12,5 13 13,5 14 Error (m) 30% 20% 10% 0%
CDF
Localization in WSNs
A Wireless Sensor Network consists of a large number of these devices capable of collecting information from their environment. These devices or nodes process information locally and can communicate cooperatively through a wireless channel. Sensor nodes are scattered in different locations of the network. These nodes may be fixed or mobile. A fixed node, also called anchor, is a device whose position is permanent and usually known. Instead, mobile nodes or end devices can move around the network. A crucial task is to estimate the location of every device in the network so that the system knows anytime where information collected by nodes is being sensed. Localization algorithms carry out this task and they can be classified attending to several criteria.
The operation of the filter can be evaluated step by step through the study of particles. Figure 5 plots an example of the particles computation after the Map Filtering stage. Here it is indicated the situation of particles on the building map. Some particles will die because of its trajectory is not possible. For example, particles cannot cross the wall between Room I43 and Room I44. All invalid particles are represented by red dots.
valid invalid
The Particle Filter is executed in the online phase of the system. The available information to implement the algorithm is: building maps, the comparison between Fingerprint database and online RSSI measurements and the knowledge of likely trajectories for the end device. The idea behind Particle Filter is to represent the possible locations for mobile node through particles. These particles will have an associated weight that describe the probability that the end device is located at this position. The next picture shows the different steps of the algorithm:
1000
750
500
250
Conclusions
Particle Filter implements a centralized scheme to estimate the position of mobile nodes in a WSN. The higher memory and processing capacity that allows this centralized algorithm, results in a greater accuracy in estimated positions in comparison with a distributed algorithm. The really power of the Particle Filter is the possibility to locate an end device under different criteria such as RSSI, temperature, measurements from inertial sensors, filtering by building maps and so on. The challenge would be design an algorithm which integrate the computation of all criteria. The fingerprint database supplies useful information about the stationary conditions of the scenario. However, the own RSSI variability together with limitations in the devices hardware make that the evaluation of this algorithm is not as suitable as it is expected. Furthermore, an exhaustive knowledge of the motion model of the mobile node involves the improvement of the algorithm. If information about the likelier rooms, trajectories that it usually covers and in general its dynamic motion are available, that leads to a better performance of the algorithm.
This project report describes the design of a centralized localization algorithm for an indoor scenario to estimate the position as well as the tracking of mobile nodes in a Wireless Sensor Network. A solution based on Particle Filter algorithm together with Fingerprinting Technique has been designed and implemented for a network of 18 nodes. Fingerprinting has also been carried out by forming a database of more than 30 references. Building maps, RSSI measurements and the knowledge of the nodes dynamic have been used to perform these tasks.
Figure 3. Particle Filter Stages