Efficient and complete data collection is one of the most important tasks in wireless ad-hoc sensor networks. Additionally, the collection of the full data set should be performed in the most resource efficient way, thus prolonging the battery lifetime of the network. We introduce a new approach for energy efficient data collection through the use of staggered sampling. Staggered sampling means that at each sampling moment (epoch) only a small percentage of sensors collect (sample) data. The proposed approach leverages on statistical relationships between samples taken from different sensors and/or at different epochs for the prediction of the non-sampled sensor data.

The main goal of the approach is to ensure complete collection of data during a periodic cycle while minimizing the number of sensor readings collected at any point in time. Complete data collection is confirmed by ensuring that each sensor is either sampled at each epoch or the data sample can be accurately recovered though model prediction of the sampled sensors. The proposed approach consists of two main phases. First, efficient modeling of the prediction relationship between two sensors using kernel smoothing over different time lags is performed. Second, the selection of epochs at which each sensor is to sample the data is determined. A 0-1 integer linear programming formulation is used to address this NP-complete assignment problem optimally on relatively large instances. We demonstrate the effectiveness of the approach on traces from actually deployed networks for sensor of two modalities: temperature and humidity.