Indoor Positioning Using Correlation Based Signal Analysis and Convolutional Neural Networks

Positioning algorithms are designed based on location related information contained in received signals, these can be propagation delay, angle of arrival, and received power. However, regardless of the positioning parameter, low-complexity linear position estimators provide reliable and accurate results only under line-of-sight propagation conditions. Hence, this paper proposes an alternative position information parameter based on the correlation of signals received at several sensing units. A low-complexity convolutional neural network uses this novel parameter for estimating the source coordinates. A simulated indoor environment based on ray tracing has been employed to compare the localization performance of the proposed approach against classical positioning schemes under a common simulation framework. The results indicate that an accurate yet low-complexity positioning solution can be achieved in multipath propagation scenarios where traditional schemes based on time-difference-of-arrival and received signal strength usually present limited performance. Furthermore, guidelines for selecting system parameters that improve the positioning accuracy of the proposed scheme are presented.