Abstract
This paper proposes a new Kalman filter based online framework to estimate the spatial wireless connectivity in terms of received signal strength (RSS), which is composed of path loss and the shadow fading variance of a wireless channel in autonomous vehicles. The path loss is estimated using a localized least squares method and the shadowing effect is predicted with an empirical (exponential) variogram. A discrete Kalman Filter is used to fuse these two models into a statespace formulation. The approach is unique in a sense that it is online and does not require the exact source location to be known apriori. We evaluated the method using real-world measurements dataset from both indoors and outdoor environments. The results show significant performance improvements compared to state-of-the-art methods using Gaussian processes or Kriging interpolation algorithms. We are able to achieve a mean prediction accuracy of up to 96% for predicting RSS as far as 20 meters ahead in the robot s trajectory.
Links
BibTeX (Download)
@conference{Parasuraman2018d,
title = {Kalman filter based spatial prediction of wireless connectivity for autonomous robots and connected vehicles},
author = {Ramviyas Parasuraman, Petter Ögren, Byung-Cheol Min},
url = {https://www.researchgate.net/publication/326235283_Kalman_Filter_Based_Spatial_Prediction_of_Wireless_Connectivity_for_Autonomous_Robots_and_Connected_Vehicles},
year = {2018},
date = {2018-08-27},
abstract = {This paper proposes a new Kalman filter based online framework to estimate the spatial wireless connectivity in terms of received signal strength (RSS), which is composed of path loss and the shadow fading variance of a wireless channel in autonomous vehicles. The path loss is estimated using a localized least squares method and the shadowing effect is predicted with an empirical (exponential) variogram. A discrete Kalman Filter is used to fuse these two models into a statespace formulation. The approach is unique in a sense that it is online and does not require the exact source location to be known apriori. We evaluated the method using real-world measurements dataset from both indoors and outdoor environments. The results show significant performance improvements compared to state-of-the-art methods using Gaussian processes or Kriging interpolation algorithms. We are able to achieve a mean prediction accuracy of up to 96% for predicting RSS as far as 20 meters ahead in the robot s trajectory.},
keywords = {networking, robotics},
pubstate = {published},
tppubtype = {conference}
}