Neural Network Based Path Detection for an FMCW Positioning System

摘要

Multipath propagation is a major source of error for runtime detection based positioning systems. For the case of an FMCW-based positioning system, the overlap of pulse shapes in the frequency domain restricts exact measurement of frequency, and thus of runtime. Choosing a measurement point on the slope of the pulse is a way to mitigate measurement errors. In this paper, we present a neural network as a means of estimating the ideal measurement point. The network is shown to outperform fixed level measurements even with very sparse training data.