Improving Estimation of the Nearest Neighbor in Mobile Wireless Sensor Networks

In wireless networks, every node needs to keep its neighbors’ link connectivity originated from periodic beacon messages in order to maintain the best route and forward packets along the path. Especially, wireless sensor networks should be working under low power constraint, so retransmission should be restricted in WSN as far as it can [2]. Moreover, low power wireless link makes the connectivity too volatile since low power wireless radio is vulnerable due to temporal interference from higher power radio, e.g., 802.11, Bluetooth, and etc. [3], [4]. Additionally, mobility makes the link connectivity more unstable [5]. These intrinsic wireless link behaviors make the link estimation for best routes far more difficult. Thus, it would be of great significance to estimate the best neighbor by efficiently exploiting the previous link status, i.e., history of beacon time and power so as to forward packets successfully. In our project, we exploit the locally weighted linear regression method to estimate signal power when a node assumes to receive a packet from neighbors at an arbitrary time and rank the proximity of neighbors according to the estimated power. More specifically, the node which has the highest power would be the closest neighbor. We use NS-2 network simulator to model the mobility of nodes and obtain training data from their own learning tables. We compare the estimation method with naive deterministic schemes such that a node, which has the largest received power in the last beacon time, is selected as the 1st rank. We expect our learning algorithm will reduce the error rate to estimate the best candidate for forwarding compared to the naive methods.