Practical modeling and prediction of radio coverage of indoor sensor networks

The robust operation of many sensor network applications depends on deploying relays to ensure wireless coverage. Radio mapping aims to predict network coverage based on a small number of link measurements. This problem is particularly challenging in complex indoor environments where walls significantly affect radio signal propagation. Nevertheless, we show that it is feasible to accurately predict coverage through a two-step process: a propagation model is used to predict signal strength at a recipient node, which is then mapped to a coverage prediction. Through an in-depth empirical study, we show that complex models do not necessarily produce accurate estimates of signal strength: there is an important tradeoff between model accuracy and the number of parameters that must be estimated from limited training data. We find that the best performance is achieved by a family of models which classify walls based on their attenuation into a small number of classes and develop an algorithm to perform this classification automatically. Based on these insights, we build a novel Radio Mapping Tool (RMT) for predicting radio converge in indoor environments. Experimental results demonstrate RMT's effectiveness in two buildings: RMT reduces the number of locations where coverage is erroneously predicted to exist by as much as 39% and 54% compared to the classic log-normal radio propagation model.