Previous Abstract Return to Session B3 Next Abstract

Session B3: Precise GNSS Positioning Applications

Robust Multipath Detection by Intra- and Inter-Domain Fusion with Real-Time Capability
Artem Gostishchev, Friederike Fohlmeister, Andriy Konovaltsev, German Aerospace Center (DLR), Germany
Location: Cypress


The ranging errors induced by multipath echoes in the receivers of Global Navigation Satellite Systems (GNSSs) are of big concern in applications with safety aspects. Even with an increased number of GNSS constellations, available frequencies and signals, the phenomenon of multipath propagation will still be an open issue when it comes to accuracy and integrity of the positioning and timing services of GNSS. Specifically in maritime applications, the estimations of the position and velocity of a vessel provided by a GNSS receiver are used by the Automatic Identification System (AIS) in order to share this information with other maritime users and allow for a better assessment of the traffic situation. For future GNSS applications, such as autonomous docking in a harbor, significantly higher positioning accuracy and integrity will be required. Therefore, the recognition of any positioning solution degradation due to the influence of multipath effects is highly important. This goal can be achieved by implementing a robust, preferably real-time multipath detector. The output of the multipath detector can either be used to warn the user directly or can be included into the receiver’s post processing algorithms to exclude pseudorange measurements from multipath-affected satellites from the positioning solution.
In the past, different methods for multipath detection have been proposed and investigated. These can be classified by their operation domain(s), i.e. the signal properties which are used to distinguish between the line-of-sight (LOS) and multipath propagation. In the time domain, signal degradation in general also leads to a deformation of the correlation function between the received signal and its local replica. Therefore, the shape of the correlation function is the main feature, which is analyzed by classical signal quality monitoring techniques [1, 2] in the time domain. This analysis can be extended to cover multiple GNSS services, comparing the signal characteristics at different frequencies [3]. Array antenna based approaches employ spatial diversity by trying to estimate the number of signals in the space domain [4]. By using dual-polarization antennas, the changed polarization of the reflected signal can be additionally exploited [5].
Even though different methods have been implemented for demonstration purposes or in operational receivers and tested in simulated or realistic conditions, most of them rely only on single-domain detection. This leads to performance degradation of the multipath detector in cases where the LOS and multipath signals are highly correlated in the method’s operation domain.
To overcome this problem, a combination of different methods in different domains for mutual compensation in adverse conditions is desirable for a robust detection. As a solution approach, we propose a framework that allows combining different detection methods from different signal domains to one detection result. As a first step, we use the time and the space domain.
In the time domain, we use an approach of feature-based binary classification known from information retrieval and machine learning applications. We assume that the signal properties observed in the multipath-free and multipath-affected conditions differ. This allows allocating both cases into corresponding classes. As observable features in the time domain, we use the outputs of qualitative and quantitative metrics. These either analyze the asymmetries in the sampled cross-correlation function of the received signal or its deviations from the ideal expectation. To combine different features in the time domain, we discuss a fusion on decision-level by combining single-feature decisions or on feature-level by combining all features into an overall metric to be used for detection.
In the space domain, we employ multiple model order estimation techniques, such as the Akaike Information Criterion (AIC) or Minimum Description Length (MDL) to estimate the number of signals present in the received signal. Since all model order estimation algorithms show different behavior in advantageous and disadvantageous situations, we use an evidence-based intra-domain fusion approach often referred to as the Dempster-Shafer Theory of Evidence [6] to combine the estimations into a non-binary signal-number estimate. The estimate is then transformed into a binary detection result, which is used as multipath indicator in the space domain.
Due to the weaknesses of time domain detection methods in temporally correlated and space domain detection methods in spatially correlated conditions, we provide multiple approaches for an inter-domain combination by either exploiting properties of array antennas to enhance the time domain solution or by directly combining binary decisions from different domains by logical operators. This allows for an optimization of either detection or false-alarm rates.
Since the respective detection methods in one domain and the combination of features or decisions within each domain do not depend on other domains, the proposed approach offers the flexibility to also incorporate detection methods from the frequency or polarization domain into the overall decision. Additionally, it enables the investigation of alternative fusion algorithms. The methodology is designed and tested to be suitable for implementation and operation in state-of-the-art real-time software-defined GNSS receivers.
The performance of the proposed methods has been evaluated by utilizing software- and hardware-generated GNSS signals. To assess the theoretical performance limits, we process signals generated by a simplified baseband model using an in-house software-defined GNSS receiver. In the time domain, we can observe an increase in detection rate and a decrease in the false-alarm rate. This performance gain arises from combined classification decisions compared to single-feature classifications. In the space domain, the evidence-based fusion allows to significantly decrease the false-alarm probability without a major loss in the detection rate. The inter-domain combination finally allows for reliable multipath detection even in temporally or spatially correlated situations.
To study the performance in practically more relevant conditions, we process different GNSS multipath scenarios, which we generated with a hardware GNSS simulator. In the time domain, we confirm that combining different features leads to a reasonable detection rate, which is not dependent on the individual detection quality of each feature. In the space domain, we in general observe larger false-alarm rates, in comparison to the simple baseband model. However, these are reduced if the proposed evidence-based intra-domain fusion method is applied.
The obtained results clearly indicate that the proposed approach outperforms conventional single-domain single-feature multipath detection methods.

[1] Phelts, E. “Multicorrelator Techniques for Robust Mitigation of Threats to GPS Signal Quality”, PhD thesis, Stanford University, 2001.
[2] Irsigler, M. et al. “Multipath performance analysis for future GNSS signals”, Proceedings of the ION ITM, 2005
[3] Mubarak, O.M. et al. “Analysis of early late phase in single-and dual-frequency GPS receivers for multipath detection”, In GPS Solutions, Vol. 14, No 4., 2010
[4] Wax, M. et al. “Detection of signals by information theoretic criteria”, IEEE Transactions on Acoustics, Speech, and Signal Processing, 1985
[5] Brennemann, M. et al, “Mitigation of GPS multipath using polarization and spatial diversities”, Proceedings of the ION ITM, 2007
[6] Shafer, G. et al. ”A mathematical theory of evidence”, Princeton university press, Princeton, 1976.



Previous Abstract Return to Session B3 Next Abstract