A data-driven framework for rapid modeling of wireless communication channels

Abstract

Accurate estimation and prediction of wireless signal strength holds the promise to improve a wide variety of applications in network-ing and unmanned systems. Current estimation approaches use either simplistic attenuation equations or detailed physical models that provide limited accuracy and may require a lengthy period of environmental assessment and computation. This dissertation presents a new, data-driven, stochastic framework for rapidly building accurate wireless connectivity maps. The framework advances the state of the art in three aspects. First, it augments the classic spatial interpolation procedure known as Kriging with a complementary additive approach to capture the typical anisotropic nature of wireless channels in cluttered environments. Second, it includes a technique for rapidly creating and maintaining a connectivity map in near real-time through the use of a spatial Bayesian recursive filter. Third, it introduces a novel methodology to adapt the resolution of a connectivity map based on the spatial characteristics and the quantity of available sample measurements. Detailed analyses, using several datasets collected recently in the Monterey Harbor, have confirmed the power and agility of the proposed approach.