K-Nearest-Neighbor analysis of received signal strength distance estimation across environments

Abstract — Prior studies investigating the use of non-parametric models for ranging and localization via received signal strength have been restricted usually to one or two relatively similar environments, and have used primarily to 802.11 network in-terfaces. This paper discusses methods for and results of ranging experiments with signal strength measurements in a sensor network as the environment is varied. The K-nearest-neighbor distance estimation error is computed for both ranging and localization scenarios. Degradation of the mean 80th percentile error from 30 feet to 50 feet is seen when multiple environments are considered; however, the standard deviation improves from 16 feet to 10 feet. The localization results are similar. I