This paper addresses the research question: can feedforward neural network (FFNN)-based path loss modeling improve the accuracy of Kriging? Radio propagation factors, which consist of path loss and shadowing, can accurately be obtained via crowdsourcing with Kriging. In most works on Kriging-aided radio environment mapping, measurement datasets are first regressed via linear path loss modeling to ensure spatial stationarity of the shadowing. However, in practical situations, the path loss often contains an anisotropy owing to terrain and obstacle effects. Thus, Kriging may not perform an optimal interpolation because of the errors in path loss modeling. In this paper, an FFNN is used for path loss modeling. Then, ordinary Kriging is applied to interpolate the shadowing. We first evaluate the performance of this method in a case where the transmitter is fixed. It is shown that this method does not improve Kriging in a large-scale and fixed transmitter system; although the FFNN outperforms OLS in path loss modeling. Then, this method is extended to distributed wireless networks where transmitters are arbitrarily located, such as in mobile ad hoc networks (MANETs) and vehicular ad hoc networks (VANETs). The results of a measurement-based experiment show that the FFNN is capable of improving Kriging in such a distributed network case.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.