A localization algorithm with learning-based distances

Existing range-based localization algorithms are superior only when a high accuracy node-to-node measured distance exists. This assumption is actually difficult to satisfy with current ranging techniques used in tiny sensor nodes. Meanwhile, range-free localization algorithms work independently of ranging error but can only produce limited node accuracy. In this paper, we propose a novel localization scheme that uses a learning-based distance function to estimate distances. The adaptation of distance function to ranging error and other network conditions, i.e., network density, number of anchor, results in better estimated distances. This leads to more accurate position calculation comparing to existing works, especially when ranging error is high.