Previous Abstract Return to Session C1

Session C1: Navigation Using Environmental Features

Towards Real-Time Channel-SLAM Multipath Assisted Positioning
Rostislav Karásek and Christian Gentner, German Aerospace Center (DLR), Germany
Location: Atrium Ballroom
Alternate Number 2

This paper presents a novel algorithm that aims to allow a real-time application of a simultaneous localization and mapping (SLAM) approach for multipath assisted positioning in GNSS-denied environments based on signals implemented in current and future off-the-shelf devices. We show that the power and spectral properties of such signals allow us, in addition to estimating the position of a receiver, to infer valuable information on the locations of reflectors and scatterers.
In this paper, we consider a positioning approach using wireless signals propagated on line-of-sight, reflected, and scattered by objects. The signal collected by the receiver antenna consists of a sum of the delayed and distorted replicas of the transmitted signal propagated via different paths between the transmitter and the receiver, called multipath components. Especially in indoor or urban scenarios, the high number of non-line-of-sight multipath components decreases the accuracy of the position estimation based on standard methods. Mostly, the approach for the multipath propagation problem is to try to estimate the line-of-sight multipath component while mitigating all of the non-line-of-sight multipath components. However, the non-line-of-sight multipath components carry information which can be utilized to increase the precision of the estimated position. The family of algorithms that allows utilizing information carried by the non-line-of-sight multipath components is usually referenced as multipath assisted positioning algorithms.
It was shown that the Channel-SLAM, which belongs to the multipath assisted positioning family of algorithms, can utilize information carried by the received signal without any prior knowledge such as room geometry or a fingerprinting database. The existence of the multipath wireless radio channel together with the knowledge of the initial position of the receiver are the only conditions required for positioning using the Channel-SLAM algorithm. The Channel-SLAM works in two steps: The first step is a maximum likelihood estimation of the wireless multipath channel, which is realized by the space-alternating generalized expectation maximization to estimate amplitude and delay of all multipath components present in the received signal. Then, the estimated parameters are tracked over time using the Kalman filter. The second step uses the concept of virtual transmitters, where the Channel-SLAM treats each of the multipath components as a line-of-sight signal transmitted from a virtual transmitter which position is unknown and estimated using simultaneous localization and mapping approach. Also, we model the physical transmitter as a virtual transmitter. However, the current state-of-the-art Channel-SLAM algorithm uses a set of particle filters representing the posterior probability density functions of the position of the receiver and each of the observed virtual transmitters, which yields high calculation complexity preventing a real-time deployment.
In this paper, we present a novel approach for virtual transmitter posterior probability density function estimation based on a low order Gaussian mixture model. This model is significantly decreasing the computational load while preserving the precision of the position estimation provided by the state-of-the-art Channel-SLAM approach. We introduce a sequential learning technique estimating the parameters of the Gaussian mixture model representing the posterior probability of the virtual transmitter position. Furthermore, we present a comparison of the state-of-the-art particle filter approach with our Gaussian mixture model, which provides ten times speedup and at the same time, requires ten times less memory. The Gaussian mixture model does not suffer the weight collapse, which is one of the significant drawbacks of particle filters where all particles collapse to the wrong state due to weight disparity. The simulations show the advantage of the proposed method and provide a study on parameter settings and their influence on the position estimation.
Finally, to verify the proposed algorithm, we evaluate it on measurements using an ultra-wideband system. We use a low-cost off-the-shelf transceiver DecaWave DW1000 which allows measurement of the channel impulse response. The measurements are carried out in and indoor scenario inside an office building using only one ultra-wideband anchor as a transmitter and one ultra-wideband anchor as a moving handheld receiver for pedestrian navigation. Additionally, the pedestrian carries an inertial measurement unit. The measurements provided by the inertial measurement unit are tightly coupled with the Channel-SLAM movement model, improving the overall performance of the algorithm. With this paper, for the first time, we present a real-time algorithm for multipath assisted positioning, which does not require any prior knowledge of the environment nor a fingerprinting database.



Previous Abstract Return to Session C1