| Popis: |
Network-wide localization in wireless sensor networks leverages the location information of a small set of nodes, called anchors, for estimating the location coordinates of non-anchor nodes. Traditionally triangulation based methods are used in an iterative manner to gradually localize all non-anchor nodes in the network. The challenge in these methods is how to control the propagation of initially small localizing errors that later magnify during the localization of remaining nodes. Machine learning based localizing algorithms, on the other hand, do not work in an iterative manner. However, one of the crucial challenges in this approach is dealing with relatively smaller training data sets typically derived from anchor nodes and the number of such nodes is normally small in the network. In order to overcome that problem, this paper proposes to use additional sampling locations, called virtual nodes, in the deployment area for generating additional training data. No physical nodes are required to be placed at these sampling points. Rather a suitable automated tool can be used for generating additional training data. Specifically, a wireless sniffer device can be used for generating additional feature vectors by way of recording the Received Signal Strength Indicator (RSSI) values at every sampling location. Our findings from simulations on how the localizing accuracy improves by exploiting different trade-offs regarding the use of virtual nodes are reported here. |
