3D-Mapping-Aided GNSS Exploiting Galileo for
Better Accuracy in Dense Urban Environments
Mounir Adjrad and Paul Groves
University College London
London, United Kingdom
Abstract—Conventional single-epoch GNSS positioning in
dense urban areas can exhibit errors of tens of meters due to
blockage and reflection of signals by the surrounding buildings.
Here, we present the first implementation of 3D-mapping-aided
(3DMA) GNSS to use Galileo signals as well as GPS and
GLONASS. Our intelligent urban positioning (IUP) concept
combines conventional ranging-based GNSS positioning
enhanced by 3D mapping with the GNSS shadow-matching
technique. Shadow matching (SM) determines position by
comparing the measured signal availability with that predicted
over a grid of candidate positions using 3D mapping. Thus, IUP
uses both pseudo-range and signal-to-noise measurements to
determine position. All algorithms incorporate terrain-height
aiding and use measurements from a single epoch in time.
The 3DMA ranging algorithms presented in this work are
based on computing the likelihoods of a grid of candidate
position hypotheses. The likelihood-based ranging algorithm
(LB-3DMA) uses the same candidate position hypotheses as
shadow matching and makes different assumptions about which
signals are direct line-of-sight (LOS) and non-line-of-sight
(NLOS) at each candidate position. A strategy for integrating
LB- 3DMA with shadow matching is presented. It uses a
hypothesis-domain approach where candidate position scores
from the LB-3DMA and shadow matching algorithms are
combined prior to extracting a joint position solution.
With the increase of the number of operational Galileo
satellites in orbit, it becomes interesting to assess the
performance of the proposed approach using Galileo signals.
With more precise satellite clocks, faster satellite orbit
determination and improved signal modulation and larger
bandwidth, the Galileo system promises a better performance
when it is fully operational, compared to existing fully
operational GNSS (i.e., GPS and GLONASS), in particular in
terms of the effect of multipath interference.
GPS, GLONASS and Galileo test data were recorded using a
u-blox EVK M8T consumer-grade GNSS receiver at 18 locations
in the City of London area. The single-epoch RMS horizontal
(i.e., 2D) error across all locations using all these constellations
was 3.4 m using 3DMA GNSS, compared to 24.4 m for
conventional positioning, a factor of 7.2 improvement. The
corresponding accuracies using GPS and GLONASS only were
3.6m using 3DMA GNSS and 25.9m using conventional
positioning.
Keywords—Intelligent Urban positioning; shadow matching;
3D mapping aided GNSS
I. INTRODUCTION
A host of potential applications would benefit from a sub-
5m real-time position accuracy using low-cost equipment.
These applications include location-based services; augmented
reality; mobile mapping; tracking vulnerable people and
valuable assets; intelligent mobility, situation awareness of
emergency, security and military personnel and vehicles;
emergency caller location, navigation for the visually
impaired; lane-level road positioning for intelligent
transportation systems; aerial surveillance for law
enforcement, emergency management, building management;
and advanced rail signaling.
In dense urban areas, buildings and other obstacles degrade
GNSS positioning. This happens in three ways. Firstly, where
signals are completely blocked, they are simply unavailable
for positioning, degrading the signal geometry. Secondly,
where the direct signal is blocked (or severely attenuated), but
the signal is received via a (much stronger) reflected path, this
is known as non-line-of-sight (NLOS) reception. NLOS
signals exhibit positive ranging errors corresponding to the
path delay (the difference between the reflected and direct
paths). These are typically a few tens of meters in dense urban
areas, but can be much larger if a signal is reflected by a
distant building. Thirdly, where both direct line-of-sight
(LOS) and reflected signals are received, multipath
interference occurs. This can lead to both positive and
negative ranging errors, the magnitude of which depends on
the signal and receiver designs. NLOS reception and multipath
interference are often grouped together and referred to simply
as “multipath”. However, to do so is highly misleading as the
two phenomena have different characteristics and can require
different mitigation techniques [1].
Much of the literature on multipath mitigation is
dominated by receiver-based signal-processing techniques [2].
However, because they work by separating out the direct and
reflected signals within the receiver, they can only be used to
mitigate multipath; they have no effect on NLOS reception at
all. Consistency checking selects the most consistent subset of
the signals received to compute a position solution from. This
is based on the principle that measurements from “clean”
direct LOS signals produce a more consistent navigation
solution than those from NLOS and severely multipath-
contaminated signals. In dense urban areas, a subset
comparison approach is more robust that conventional
sequential testing [3].
Over the past six years, there has been a lot of interest in
3D-mapping-aided (3DMA) GNSS, a range of different
techniques that use 3D mapping data to improve GNSS
positioning accuracy in dense urban areas. The simplest form
of 3DMA GNSS is terrain height aiding. For most land
applications, the antenna is at a known height above the
terrain. By using a digital terrain model (DTM), also known as
a digital elevation model (DEM), the position solution may be
constrained to a surface. In conventional least-squares
positioning, this is done by generating a virtual ranging
measurement [4]. By effectively removing a dimension from
the position solution, this improves the accuracy of the
remaining dimensions. In open areas, terrain height aiding
only improves the vertical position solution (as one might
expect). However, in dense urban areas where the signal
geometry is poor, it can improve the horizontal accuracy by
almost a factor of two [5].
3D models of the buildings can be used to predict which
signals are blocked and which are directly visible at any
location [6][7]. This can be computationally intensive.
However, the real-time computational load can be reduced
dramatically by using building boundaries [8]. These describe
the minimum elevation above which satellite signals can be
received at a series of azimuths and are precomputed for each
candidate position. A signal can then be classified as LOS or
NLOS simply by comparing the satellite elevation with that of
the building boundary at the corresponding azimuth.
The shadow-matching technique [9] determines position
by comparing the measured signal availability and strength
with predictions made using a 3D city model over a range of
candidate positions. Several research groups have
demonstrated this experimentally, using both single and
multiple epochs of GNSS data
[10][11][12][13][14][15][16][17]. Cross-street position
accuracies of a few meters have been achieved in dense urban
areas, enabling users to determine which side of the street they
are on. This complements GNSS ranging, which is more
accurate in the along-street direction in these environments
because more direct LOS signals are received along the street
than across it. shadow matching has also been demonstrated in
real time on an Android smartphone [18]. A review of shadow
matching, including its error sources and how it could be
developed further may be found in [19].
3D models of the buildings can also be used to aid
conventional ranging-based GNSS positioning. Where the user
position is already approximately known, it is straightforward
to use a 3D city model to predict the NLOS signals and
eliminate them from the position solution [20][21][22].
However, for most urban positioning applications there is
significant position uncertainty. One solution is to define a
search area centered on the conventional GNSS position
solution and compute the proportion of candidate positions at
which each signal is receivable via direct LOS. This can then
be used to re-weight a least-squares position solution and aid
consistency checking [5]. More sophisticated approaches
which score position hypotheses using the GNSS pseudo-
range measurements and satellite visibility predictions at each
candidate position are presented in [23] and in Section 2.2 of
this paper.
Several groups have extended 3D-mapping-aided GNSS
ranging by using the 3D city model to predict the path delay of
the NLOS signals across an array of candidate positions
[24][25][26][27]. A single-epoch positioning accuracy of 4m
has been reported [26]. However, unless the search area is
small, this approach is very computationally intensive as the
path delay cannot easily be pre-computed. The urban trench
approach presented in [28] enables the path delays of NLOS
signals to be computed very efficiently, but only if the
building layout is highly symmetric, so it can only be used in
suitable environments. Therefore, NLOS path delay
predictions are not used in the work presented here. 3DMA
GNSS ranging has also been combined with ‘direct
positioning’ which uses the receiver correlator outputs to score
an array of position hypothesis [29]
Figure 1: Intelligent urban positioning
Clearly, to get the best performance out of GNSS aided by
3D mapping, as much information as possible should be used.
Thus, both pseudo-range and SNR measurements from a
multi-constellation GNSS receiver should be used, together
with both LOS/NLOS predictions and terrain height from 3D
mapping. This concept is known as intelligent urban
positioning (IUP) [30] and is illustrated in Figure 1.
A preliminary implementation of the IUP concept is
presented in [31] This integrates shadow matching with a
3DMA least-squares GNSS ranging algorithm incorporating
terrain height aiding, consistency checking, and weighting of
the pseudo-ranges according to the average predicted satellite
visibility over a search area. Position-domain integration is
used with two different weighting approaches. Error
covariance-based weighting was found to perform slightly
better than weighting using the street azimuth. The overall
root mean square (RMS) horizontal (i.e., 2D) single-epoch
position accuracy obtained using a u-blox EVK M8T receiver
was 6.1 m, compared to 25.9 m using conventional GNSS
positioning, a factor of four improvement.
This work has been extended by
Deriving a 3DMA GNSS ranging algorithm based on
computing the likelihood of an array of candidate
position hypotheses based on the satellite visibility
predictions at each position (the least-squares algorithm
is retained for initialization);
Performing a hypothesis-domain integration of 3DMA
ranging with shadow matching;
Preliminary results using GPS and GLONASS were
presented in [32]. This paper extends the work further by also
using Galileo signals, taking advantage of Galileo’s recent
initial operational capability (IOC). With the increase of the
number of operational Galileo satellites in orbit, it becomes
interesting to assess the performance of the proposed approach
using Galileo signals. With more precise satellite clocks, faster
satellite orbit determination and improved signal modulation
and larger bandwidth, the Galileo system promises a better
performance when it is fully operational, compared to existing
fully operational GNSS (i.e., GPS and GLONASS), in
particular in terms of the effect of multipath interference.
All results presented here are based on a single epoch of
GNSS measurements, which suits many location-based service
(LBS) applications that require a quick one-time fix. 3DMA
GNSS is particularly important for single-epoch positioning
because other augmentations, such as carrier-smoothing,
carrier-phase positioning and integration with inertial sensors,
only work with multiple epochs of GNSS data [33].
An alternative implementation of the intelligent urban
positioning concept is presented in [23]. The shadow-
matching algorithm is simpler than that used here. A different
likelihood-based 3DMA GNSS ranging algorithm is also
implemented which uses only the signals predicted to be direct
LOS at each candidate position. The experimental tests
demonstrate that the method works well. However, as the
results presented combine measurements from multiple
epochs, they are not directly comparable with the single-epoch
results presented here.
Extending the IUP implementation presented here to
multiple epochs for navigation and tracking applications is a
subject for future work. Better performance can be expected as
several researchers have already demonstrated that filtering
can improve 3DMA GNSS performance [16][17][23].
Conventional GNSS positioning also works much better
with multiple epochs of data. With an extended Kalman filter
(within which carrier-smoothing is normally inherent), it is
much easier to detect outliers due to NLOS reception and
severe multipath interference than it is using single-epoch
least-squares positioning. However, 3DMA GNSS also has an
important role to play in multi-epoch positioning as it will
enable carrier-smoothed, inertially aided and potentially even
real-time kinematic (RTK) carrier-phase positioning to be
accurately initialized and re-initialized in challenging urban
environments.
Section II summarizes the 3DMA GNSS positioning
algorithms, including the least-squares and likelihood-based
3DMA ranging algorithms, the shadow matching algorithm
and the integration algorithms. Section III presents
experimental test results from data collected using a u-blox
EVK M8T consumer-grade GNSS receiver at 18 locations in
the City of London area. Finally, Sections IV summarizes the
conclusions and plans for future work.
II. PROPOSED IUP APPROACH
The proposed IUP system is illustrated in Figure 2. The
conventional GNSS solution and 3D mapping information are
exploited by the least-squares 3DMA GNSS ranging (LS-
3DMA) algorithm as explained in [5]. LS-3DMA is then used
to initialize the likelihood-based 3DMA GNSS ranging (LB-
3DMA) algorithm and the shadow-matching algorithm,
enabling them to use a much smaller search area than if the
conventional GNSS position was used for initialization.
Figure 2: Proposed IUP algorithm
The integration algorithms then compute a joint position
solution from likelihood-based 3DMA ranging and shadow
matching. A hypothesis-domain integration algorithm (HI-LB)
is used in this work.
Figure 3: 3DMA GNSS approach
Figure 3 illustrates the 3DMA GNSS approach which
includes LS-3DMA, LB-3DMA and shadow matching. The
three algorithms share the same overall methodology: Position
information from a GNSS receiver (or from a pre-processing
step of GNSS position information) is used to define a search
area of possible solutions. The same information along with
3D mapping data are exploited to predict satellite visibility.
This latter is then combined with GNSS position information
(or its pre-processing byproduct) to define a scoring strategy
and then determine the position solution. The following
subsection summarizes each algorithm.
A. Least-Squares 3DMA GNSS Ranging
The LS-3DMA algorithm comprises six steps:
1) A search area is determined using the conventional
GNSS position solution on the first iteration and the previous
solution on subsequent iterations, together with an appropriate
confidence interval.
2) Using 3D mapping converted to precomputed building
boundaries, the proportion of the search area within which
each satellite is directly visible is computed, giving the
probability that the signal is direct LOS.
3) A consistency-checking process is applied to the
ranging measurements, using the direct LOS probabilities
from the 3D mapping.
4) The set of signals resulting from the consistency
checking process is subjected to a weighting strategy based on
the previously determined LOS probabilities and carrier-
power-to-noise-density ratio, C/N
0
.
5) Terrain height is extracted from the 3D mapping and a
virtual range measurement is generated using the position at
the centre of the search area.
6) Finally, a position solution is derived from the pseudo-
ranges and virtual range measurement using weighted least-
squares estimation.
The algorithm is then iterated several times to improve the
position solution. Full details are presented in [5] (final
version) and [34] (preliminary version).
B. Likelihood-based 3DMA GNSS Ranging
In LB-3DMA, an array of candidate position hypotheses
are scored according to the correspondence between the
predicted and measured pseudo-ranges. This enables different
error distributions to be assumed for a given GNSS signal at
different candidate positions. Thus, at positions where a signal
is predicted from the 3D mapping (via precomputed building
boundaries), to be NLOS, a skew normal (Gaussian)
distribution is assumed, biased towards positive ranging
errors. Elsewhere, a conventional symmetric normal
distribution is assumed. Figure 4 illustrates the scoring
strategy used by LB-3DMA algorithm.
Figure 4: LB-3DMA algorithm scoring strategy
Terrain height aiding is inherent in generating the position
hypotheses, enabling a single height to be associated with each
horizontal position and thus avoiding the computational load
of a 3D search area. The receiver clock bias is eliminated by
differencing all pseudo-range measurements across satellites.
Other LB-3DMA algorithms based on candidate position
hypothesis scoring have been described in the literature.
However, they differ from the approach proposed here. In [25]
and [26], pseudo-ranges predicted to be NLOS are corrected
using path delays predicted from the 3D mapping. This is
potentially more accurate, but the path delay computation is
highly computationally intensive. In [23], a least-squares
position solution is computed using only those signals
predicted to be direct LOS and the candidate position is then
scored according to its Mahalanobis distance from the least-
squares position solution.
The LB-3DMA algorithm comprises the following six steps:
1) A circular search area of radius 40m is defined with its
centre at the least-squares 3DMA ranging position solution.
Within this search area, a grid of candidate positions is set up
with a spacing of 1m.
2) For each candidate position, the satellite visibility is
predicted using the building boundaries precomputed from the
3D city model. At each candidate position, the highest
elevation satellite predicted to be direct LOS is selected as the
reference satellite.
3) At each candidate position, the direct LOS range to
each satellite is computed. Measurement innovations are then
computed by subtracting the computed ranges from the
measured pseudo-ranges and then differencing with respect to
the reference satellite.
4) At each candidate position, the measurement innovation
for each satellite predicted to be NLOS is re-mapped to a skew
normal distribution.
5) Likelihood score for each candidate position is
computed using the vector of measurement innovations and
the measurement error covariance matrix.
6) A position solution is derived by using the likelihood
scores to weight the candidate positions.
Further details are presented in [32], while full details of
the algorithm will be presented in a forthcoming journal
submission, currently under preparation.
C. Shadow Matching
The shadow matching algorithm is a modified version of
that presented in [15]; further details are presented in [32]. The
scoring strategy is illustrated on Figure 5.
Figure 5: Shadow matching scoring strategy
The shadow matching algorithm comprises the following five
steps:
1) A circular search area of radius 40m is defined with its
centre at the least-squares 3DMA ranging position solution.
Within this search area, a grid of candidate positions is set up
is set up with a spacing of 1m.
2) For each candidate position, the satellite visibility is
predicted using the building boundaries precomputed from the
3D city model. If the satellite elevation is above the building
boundary at the relevant azimuth, the LOS probability
predicted from the building boundary, p(LOS|BB), is set to
0.85. Otherwise, it is set to 0.2. These values allow for
diffraction and 3D model errors.
3) The observed satellite visibility is determined from the
GNSS receiver’s C/N
0
or signal to noise ratio (SNR)
measurements. From these, a probability that each received
signal is direct LOS is estimated.
4) Each candidate position is scored according to the
match between the predicted and measured satellite visibility.
The overall likelihood score for each position is then the
product of the individual satellite probabilities.
5) A position solution is derived by using the likelihood
scores to weight the candidate positions.
D. Hypothesis-Domain Integration
Both shadow matching and LB-3DMA can produce
multimodal position distributions where there is a good match
between predictions and measurements in more than one part
of the search area. These will typically comprise the true
position hypothesis and one or more false hypotheses. In
general, the true position hypothesis will be consistent across
the two positioning methods whereas the false hypotheses will
not be. Hypothesis-domain integration therefore helps to
eliminate false position hypotheses by computing a joint
ranging and shadow matching likelihood surface by
multiplying the ranging and shadow matching likelihoods for
each candidate position, then computing a position solution by
using the joint likelihood scores to weight the candidate
positions. Further details are presented in [32].
III. IUP EXPERIMENTAL RESULTS
GNSS measurements were collected in August 2016 using
a u-blox EVK M8T GNSS receiver running the latest
firmware (release 3.01) supporting GPS, GLONASS and
Galileo. Raw data was collected for those three constellations
by interfacing the u-blox receiver to a Raspberry Pi (via USB)
and where this latter was powered by a battery pack and
configured as a WiFi hotspot to which a smartphone was
connected (using the mobile SSH App) to configure the
system and enable data logging. Figure 6 illustrates the u-
blox-based hardware.
Figure 6. u-blox EVK M8T-based data logging hardware
Two rounds of data collection were performed at 18
locations in the City of London. The City of London area is a
hybrid European city in the sense there exist narrow and wide
streets and buildings made of brick/stones or glass/steel as
illustrated on Figure 7. The sites were paired with data
collected on opposite sides of the street on the edge of the
footpath next to the road (Figure 8). The truth was established
to decimeter-level accuracy using a 3D city model to identify
landmarks and tape measure to measure the relative position
of the user from those identified landmarks. The two rounds of
data at each site were separated by approximately 2 hours,
ensuring that the satellite positions in the two datasets were
independent.
The first dataset was used for calibrating the shadow-
matching algorithm. The second dataset was then used for
testing the positioning algorithms. 4 minutes of data were
collected at each site on each round. A 3D city model of the
area, from Ordnance Survey (OS), was used to generate the
building boundary data used for the subsequent analysis. The
model is stored in the Virtual Reality Modelling Language
(VRML) format. Figures 9 illustrate the 3D model used in this
study.