An Efficient Screening Method for Identifying Parameters and Interactions that Impact Wireless Network Performance

Wireless networks rely on a protocol stack for their operation. Not only are the protocols at each layer configurable; potential interactions arise among the protocol stack, operating system, hardware, and operating environment. Hence, there is a vital need for screening, i.e., to determine the parameters and interactions that significantly impact the performance of a system. In this paper, we propose (1) the use of a locating array (LA) as the design of a screening experiment and (2) an algorithm to analyze the resulting measured performance data. Compared to conventional designs, LAs grow logarithmically in the number of parameters making them efficient for screening complex engineered networks. The analysis uses a framework from compressive sensing and provides robustness to noise in the system through a breadth-first search strategy. We apply LAs for screening audio quality and radio frequency exposure in a Wi-Fi conferencing scenario in the w-iLab.t wireless network test bed and validate the results using the Dantzig selector and Lasso regression methods.