This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Neural Network based Geo-Regioning Luciano Leins and Christoph Steiner Communication Technology Laboratory, Swiss Federal Institute of Technology, CH-8092 Zurich Email: firstname.lastname@example.org, email@example.com Abstract The large bandwidth of Ultra-Wideband (UWB) signals induces a high temporal resolution of the multipath components of the propagation channel. Due to the variation of the propagation paths for different positions of the transmitter, a CIR can act as a fingerprint for the position of the transmitter, if the position of the receiver is known. UWB Geo-Regioning is a coarse localization or clustering method based on channel im- pulse response fingerprinting . In this paper, neural networks are used to perform Geo-Regioning. As input vector, either the complex values or the power values of the CIR are used. In order to test the performance of the Geo-Regioning with neural networks, the CIRs from the measurement campaign in  are used. I. INTRODUCTION The very broad transmission bandwidth of Ultra-Wideband (UWB) technology leads to convenient frequency diversity and a high temporal resolution of the propagation channel, which is useful for localization techniques. Most of the localization techniques use a time of arrival (ToA) approach with trilateration to determine the position of the transmitter. At least three synchronized reference receivers with known positions are necessary to estimate the position, which implies an additional signaling overhead in a network. These methods work very accurate for line-of-sight (LOS) con- ditions. Difficulties occur in non-line-of-sight (NLOS) condi- tions, since the strongest path does not necessarily correspond to the direct path, which lead to large positioning errors. Geo-Regioning as introduced in  is a different localiza- tion technique based on CIR fingerprinting with some distinct advantages over geometrical positioning techniques. In order to obtain a position estimate, only one receiver is required. A coherent receiver estimates the CIR for data decoding purposes. Such receivers have already all information at hand to estimate the position of the transmitter. This means that the position information is obtained during communication without any additional signaling overhead or receiver hardware requirements. However, before being able to start the localiza- tion algorithm, a priori information about the environment or the propagation channel is used to generate a reference map consisting of the geographical regions. An observed CIR is then mapped to an entry in the reference map to determine the position of the transmitter. Several mappings are possible, such as nearest neighbor mapping or a probabilistic approach, where parameters for a channel model are estimated from the a-priori information . In this paper, the applicability of neural networks for the mapping is investigated....
View Full Document
- Spring '10
- Neural Networks