Contributions to radio frequency indoorpositioning and through-the-wall mapping

Abstract

The first part of this work introduces a novel machine-learning based location engine aiming at improving floor identification accuracy in radio frequency (RF) multi-floor indoor positioning while preserving a low two-dimensional positioning error. The location engine also reduces the position fix computational complexity. A large database of RF samples was collected in a 13-storey building to evaluate the proposal. The second part of this work studies in detail radio tomographic imaging, also referred to as RF-based through-the-wall mapping (TWM). Four different reconstruction algorithms – two projective and two algebraic – are compared using a path-loss model corrupted by Rayleigh noise in a parallel-beam acquisition geometry. After that, the thesis proposes applying the Finite Element Method (FEM) to simulate several parallel-beam geometry RF TWM setups, providing a more accurate simulation model. The meshing parameters of the FEM model geometry have been optimized, enabling a significant computational cost reduction while preserving accuracy. Reconstruction of two floor maps is carried out using the FEM model with different sampling rates, operational frequencies, and antenna models. Finally, a multi-sensor circular acquisition geometry (MCG) is defined to reduce the time required to acquire the RF samples in comparison to the parallel-beam geometry. The MCG scheme is evaluated using the proposed FEM framework.