Target Localization using Complex Support Vector Machines


M. Palaniswami;
Post Doctoral Research Fellow: Alistair Shilton.

Student: Bharat Sundaram.
Introduction: Developing localization algorithms for Distributed Sensor Networks is an open research problem. The localization problem for Sensor Networks can be defined as finding the physical location of an 'entity of interest' (to the network) using data being collected during runtime of the network. The 'entity of interest' could be a sensor node in the network or it could be a target that the network is aiming to localize.
Significance: The algorithms being developed for Sensor Network localization can be broadly classified under Graph theoretic approaches. The features used for localization usually include Signal Strength, Time of Arrival, Time Difference of Arrival, Angle of Arrival and Node Transmission Radius. In this project, a sensor node is an abstraction used to describe any entity capable of sensing the environment, collecting data and communicating over the network. Thus the sensor node could represent an inexpensive mote with attached sensors, a gateway node acting as a hop point, an underwater node capable of sensing and communicating under water or even an Unmanned Aerial Vehicle.
Applications: This work has now been modified for better performance using Complex Support Vector Regression (CSVR). It will present our current approach and the ramifications for terrestrial Wireless Sensor networks.
Challenges: The fundamental aim of the project is to model the localization problem as a function estimation problem. Thus the research task is to find features in the sensor node data that can be used to perform the required localization. Given the set of suitable features and training data set of locations of the 'entity of interest', we propose to use Support Vector Regression (SVR) to estimate the function relating the two. As a first step, we implemented this technique to localize hostile RADAR using a formation of 3 UAVs.
top of page