Analysis of large-scale experimental data from wireless networks

The experimental design and subsequent analysis of the data collected from experimentation are tightly coupled. This paper considers techniques for the analysis of data collected from experimentation based on a locating array. Because locating arrays can be highly unbalanced, new analysis techniques are required. In order to cope with noise in the measured performance data, a search is conducted to identify the significant parameters and interactions that impact the data. A novel "heavy hitters" bounded breadth-first search (BFS) tree algorithm is proposed for analysis. It is used to validate data collected from large-scale experimentation with a physical wireless network testbed and a wireless network simulator, varying 24 and 75 parameters, respectively