Short Multipath Mitigation Technique Using Feedforward Neural Networks

Abstract:
In this paper a novel approach for mitigation of short multipath distortions is presented. The technique is applicable to FMCW-based distance measurement systems assuming distinctly separated targets. The effects of spectral distortion caused by short multipath propagation in conjunction with Fourier-based frequency analysis methods are mitigated. For this purpose the distorted spectrum is first analyzed and specific characteristics are calculated. A Feedforward Neural Network processes these characteristics and provides a correction frequency relative to the erroneously shifted maximum in order to estimate the true line-of-sight frequency. A comprehensive analysis was performed by means of a two tone test. Applicability to real world environments was investigated by means of a multi tone test using an IEEE 802.15.4a channel model.