You are on page 1of 14

J Geod (2009) 83:953–966

DOI 10.1007/s00190-009-0313-6

ORIGINAL ARTICLE

Improving the GNSS positioning stochastic model in the presence


of ionospheric scintillation
M. Aquino · J. F. G. Monico · A. H. Dodson ·
H. Marques · G. De Franceschi · L. Alfonsi ·
V. Romano · M. Andreotti

Received: 2 October 2008 / Accepted: 9 March 2009 / Published online: 24 March 2009
© Springer-Verlag 2009

Abstract Ionospheric scintillations are caused by time- model used for position computation a more realistic rep-
varying electron density irregularities in the ionosphere, resentation, vis-a-vis the otherwise ‘equal weights’ model.
occurring more often at equatorial and high latitudes. This For pseudorange processing, relative weights were com-
paper focuses exclusively on experiments undertaken in puted, so that a ‘scintillation-mitigated’ solution could be
Europe, at geographic latitudes between ∼50◦ N and ∼80◦ N, performed and compared to the (non-mitigated) ‘equal
where a network of GPS receivers capable of monitoring weights’ solution. An improvement between 17 and 38% in
Total Electron Content and ionospheric scintillation param- height accuracy was achieved when an epoch by epoch dif-
eters was deployed. The widely used ionospheric scintil- ferential solution was computed over baselines ranging from
lation indices S4 and σϕ represent a practical measure of 1 to 750 km. The method was then compared with alterna-
the intensity of amplitude and phase scintillation affecting tive approaches that can be used to improve the least squares
GNSS receivers. However, they do not provide sufficient stochastic model such as weighting according to satellite ele-
information regarding the actual tracking errors that degrade vation angle and by the inverse of the square of the standard
GNSS receiver performance. Suitable receiver tracking mod- deviation of the code/carrier divergence (sigma CCDiv). The
els, sensitive to ionospheric scintillation, allow the compu- influence of multipath effects on the proposed mitigation
tation of the variance of the output error of the receiver PLL approach is also discussed. With the use of high rate scin-
(Phase Locked Loop) and DLL (Delay Locked Loop), which tillation data in addition to the scintillation indices a carrier
expresses the quality of the range measurements used by the phase based mitigated solution was also implemented and
receiver to calculate user position. The ability of such models compared with the conventional solution. During a period of
of incorporating phase and amplitude scintillation effects into occurrence of high phase scintillation it was observed that
the variance of these tracking errors underpins our proposed problems related to ambiguity resolution can be reduced by
method of applying relative weights to measurements from the use of the proposed mitigated solution.
different satellites. That gives the least squares stochastic
Keywords Global navigation satellites system · Global
M. Aquino (B) · A. H. Dodson positioning system · Ionospheric scintillation · Receiver
Institute of Engineering Surveying and Space Geodesy (IESSG),
tracking models · Mitigation · Stochastic model
University of Nottingham, Nottingham, UK
e-mail: marcio.aquino@nottingham.ac.uk
1 Introduction
J. F. G. Monico · H. Marques
Department of Cartography, Sao Paulo State University,
Pres. Prudente, São Paulo, SP, Brazil
At high latitudes, the effects of ionospheric scintillations on
GNSS are exacerbated during solar maximum, when geo-
G. De Franceschi · L. Alfonsi · V. Romano magnetic storms may occur in connection with enhanced
Instituto Nazionale di Geofisica e Vulcanologia (INGV), solar activity. The polar ionosphere may become highly tur-
Rome, Italy
bulent leading to scintillation scenarios, as for instance in the
M. Andreotti northern hemisphere during the severe storms of October and
Geospatial Research Center Ltd., Christchurch, New Zealand November 2003 (e.g De Franceschi et al. 2008). The adverse

123
954 M. Aquino et al.

effects of events of this type on GNSS positioning and on indication also that the models of Conker et al. (2003) can
the performance of augmentation services (such as WAAS, successfully predict the receiver tracking errors. For a gen-
the US Wide Area Augmentation System and EGNOS, the eral use of these tracking models and restrictions on their
European Geostationary Navigation Overlay System) have validity the reader is referred to Conker et al. (2003).
been reported in the literature by several authors (e.g. Moore In Sect. 2, we discuss our approach, first introducing the
et al. 2002; Skone et al. 2004; Grejner-Brzezinska et al. 2006; receiver tracking models of Conker et al. (2003) and then
Aquino et al. 2005a,b, 2007a; Alfonsi et al. 2006). presenting our strategy to improve the least squares stochas-
Ionospheric irregularities causing scintillations tend to tic model. In Sect. 3, we give a brief description of the
form mostly within the irregularity oval, defined as a mod- experimental setup. In Sect. 4, we show the results of our
ification of the auroral oval, which expands and contracts experiments for pseudorange processing, while discussing
under varied geomagnetic conditions (Aarons 1997). The the influence of multipath against actual scintillation occur-
irregularity oval is approximately the auroral oval and dur- rence. Section 5 refers to initial findings when the carrier
ing periods of low solar and magnetic activity is also less phase is introduced. Section 6 contains our conclusions.
intense and contracts to higher latitudes, whereas during a
geomagnetic storm it moves equatorward. Scintillation is
expected to occur during instances when a signal from a 2 Method
GNSS satellite traverses the ionosphere inside the region of
the irregularity oval. The impact of scintillations on GPS 2.1 Receiver tracking models
positioning may be severe, including loss of signal track-
ing and positioning accuracy degradation (e.g. Aquino et al. We consider that, during the occurrence of scintillation, range
2005a,b). The latter occurs mostly due to some of the satel- measurements (of both pseudorange and carrier phase) made
lites being affected by scintillation as their links traverse the by a GPS receiver to different satellites are degraded differ-
irregularities. User position computation is accomplished by ently, depending on how each link is affected by the irregu-
the receiver through a least squares process where the mea- larities in the ionosphere. We suggest that the variances of the
surements from all satellites in view are usually considered output error of the receiver’s DLL (Delay Locked Loop) and
of the same quality. Ideally the satellite-receiver links most PLL (Phase Locked Loop), can be used to express, respec-
affected should be isolated and left out of the solution; how- tively, the degree of degradation of the pseudorange and the
ever this would be detrimental to the quality of the resulting carrier phase measurements. The DLL maintains the exact
geometry. alignment between the received and the locally generated
It is beyond the scope of this paper to enter into the codes, its output being directly related to the accuracy of the
physical aspects of the ionospheric irregularities causing signal travel time and therefore to the pseudorange measure-
scintillations. Its aim is rather to present an original math- ments. The PLL minimizes the error between the input phase
ematical approach possibly useful for future application to and its estimated phase output, which feeds the receiver pro-
GNSS receivers. We present a strategy to calculate and apply cessor. The loop will remain in lock or not depending on the
weights to measurements from different satellite links, which magnitude of this error. For the pseudoranges, the models
could be advantageous to use under scintillation conditions. of Conker et al. (2003) calculate the variance of the output
The widely used scintillation indices (S4 and σϕ —see Van code tracking errors from a first order DLL for L1 GPS code
Dierendonck et al. 1993 and Van Dierendonck 2001 for and from a first order DLL for L2 semicodeless aided by L1.
details on these indices and their measurement) and high For the variance of the output phase tracking error, Conker
rate data were collected by specially modified GPS receivers et al. (2003) presents models for two common types of PLL:
able to record the phase and amplitude of the L1 signals (Van a third-order L1 carrier PLL and a second-order L1-aided L2
Dierendonck et al. 1993). Such data, acquired during peri- carrier PLL, the latter also referred to as L2 semicodeless. We
ods of high geomagnetic activity, were used to calculate the used these models to compute the variance of the code and
code and carrier phase tracking error variance according to phase tracking errors (or tracking jitters), which are functions
the models described by Conker et al. (2003). The capability of the receiver hardware parameters and, respectively of the
of translating phase and amplitude scintillation effects into S4 amplitude scintillation index and of the spectral parame-
the variance of the tracking errors underpins our proposed ters (the slope p, and the strength at 1 Hz, T ) of the Power
method of applying relative weights to measurements from Spectral Density (PSD) of phase scintillation (obtained from
different satellites, giving a more realistic representation to high rate data, see below). Our proposed technique is based
the least squares stochastic model used in position compu- on the establishment of weights given by the inverse of the
tation. Under the experimental circumstances described in variance of these tracking errors. These weights form the
this paper, results show that the proposed method leads to basis of a modified least squares stochastic model that is
correctly assigning relative weights to the observations, an then used for position computation.

123
Improving the GNSS positioning stochastic model 955

For details of these receiver tracking models the reader To use the above PLL model in practice it was necessary
is referred to Conker et al. (2003). Here we provide only a to retrieve the spectral parameters T and p (Eq. 5) from the
brief description. The model of the L1 C/A code (C1) DLL PSD of fluctuations in the carrier phase output after detr-
tracking jitter variance is given by (in code chips): ending (i.e. correction for all other effects such as integrated
  Doppler due to satellite motion, satellite clocks, user clocks
Bn d 1 + η(c/n ) 1  and tropospheric delay). The PSD was obtained from high
0 L1−C/A 1−2S4 (L1)
2
στ2C1 = (1) rate (50 Hz) phase data, available for all stations from the IS-
2(c/n 0 )L1−C/A (1 − S42 (L1)) ACCO (Ionospheric Scintillations Arctic Coordinated Cam-
where Bn is the one-sided noise bandwidth; d is the paign Observations) project (De Franceschi et al. 2006). This
correlator spacing; (c/n 0 )L1−C/A is the fractional form of 50 Hz raw scintillation data allowed the direct computation
signal-to-noise density ratio, equal to 100.1C/N 0 ; η is the of the spectral parameters T and p through FFT (Fast Fou-
predetection integration time, equal to 0.02 s; model valid rier Transform) spectral lines (Van Dierendonck 2001). For
only for S4 (L1) < 0.707; loss of lock is assumed for greater practical future use of our method it is envisaged that the strat-
values of S4. egy proposed in Aquino et al. (2007b) and further developed
Similarly the model of the L2 semicodeless P code (P2) in Strangeways (2008) can be applied and the σϕ scintillation
DLL tracking jitter variance is represented by: index can be used (without the direct need of high rate phase
  data) to compute the tracking error variance.
Bn 1 + 2η(c/n ) (1−S 2 (L1))
1
0 L1P
στ2P2 =
4
(2) 2.2 Modified stochastic model
2(c/n 0 )L2P (1 − S42 (L2))
Mitigation of GNSS systematic errors may be achieved by
where in this case η = 1.96 × 10−6 s.
augmenting the least squares functional model with the intro-
The model for the L1 carrier PLL accounts for the effects
duction of additional parameters to account for unmodeled
of scintillation on the input phase and computes the tracking
errors. Alves and Monico (2007) successfully experimented
error variance at the output of the PLL (σφ2 ) as:
with a semi-parametric model and a penalized least squares
σφ2 = σφ2s + σφ2T + σφ2osc (3) technique to mitigate multipath effects, while Souza and
Monico (2007) used the wavelet method to reduce iono-
where σφ2s , σφ2T and σφ2osc are, respectively, the error vari- spheric effects. Another approach to mitigate systematic
ance components relating to the phase scintillation, the ther- errors is to directly modify the least squares stochastic model
mal noise (amplitude scintillation) and the oscillator noise (i.e. the observations weight matrix) in order to ensure that
(assumed as 0.01 radians in the receiver used in our experi- it more accurately represents the statistical properties of the
ments). data. Alves and Monico (2007) describe the equivalence of
In Eq. 3 above, amplitude scintillation is modeled as an both approaches. We chose the latter approach, as it can be
increase in the thermal noise, related to the decrease in the implemented through a much simpler algorithm.
received signal power: We take as an example the typical GPS pseudorange point
  positioning case, for which the least squares model, for a
Bn 1 + 2η(c/n ) 1 receiver at station A observing satellites i, j, k and l, nor-
0 L1−C/A (1−2S4 (L1))
2
σφ2T = (4) mally considers a weight matrix of the observations W, of
(c/n 0 )L1−C/A (1 − S42 (L1)) the form:
⎡ 2 ⎤
where, again Bn is the L1 third-order PLL one-sided band- 1/σ Ai 0 0 0
width, equal to 10 Hz; (c/n 0 )L1−C/A is the fractional form ⎢ j2 ⎥
⎢0 1/σ A 0 0 ⎥
of signal-to-noise density ratio, equal to 100.1C/N 0 ; η is the W =⎢ ⎥ (6)
⎣0 0 k 2
1/σ A 0 ⎦
predetection integration time, equal to 0.02 s; model valid 2
only for S4 (L1) < 0.707; loss of lock is assumed for greater 0 0 0 1/σ Al
values of S4. where it is assumed that there is no correlation between obser-
The phase scintillation component is modeled by vations and that their precisions are the same, i.e.
πT j
σ Ai = σ A = σ Ak = σ Al = σ
σφ2s =   (5) (7)
p−1 [2k+1− p]π
k f n sin 2k This ultimately simplifies the weight matrix to
where T is the spectral strength of the phase noise at 1 Hz;
W = 1/σ 2 (I ) (8)
p is the spectral slope of the phase PSD; k is the order of the
PLL; f n is the loop natural frequency. where I is the identity matrix.

123
956 M. Aquino et al.

Clearly, this weight matrix will become more complex resulted in accuracies compatible with geodetic grade
when different observables, combining different satellites receivers (e.g. Aquino et al. 2005a,b, 2007a). As explained
and stations, are considered. For instance, in a double differ- in 2.1, the spectral parameters T and p could be retrieved
ence pseudorange solution, as the observable involves two from 50 Hz data also recorded by the GISTMs. As in a real
satellites and two stations, correlation between observations world situation one would not expect all these data to be
must be accounted for. However the assumption that all mea- readily available, this paper is primarily intended as a proof
surements are of the same precision is still usual practice, of concept.
which simplifies the stochastic model. For instance, when The GISTM is a dual-frequency OEM4 NovAtel card
four satellites are observed, the double difference weight modified by AJ Systems to measure and log phase and ampli-
matrix Wdd assumes the form: tude scintillation data based on L1 GPS frequency data sam-
⎡ ⎤−1 pled at 50 Hz (see Van Dierendonck 2001). The INGV has
422 permanently deployed GISTMs at Svalbard (Norway), in the
Wdd = 1/σ 2 ⎣ 2 4 2 ⎦ (9) frame of the ISACCO project (De Franceschi et al. 2006)
224 and has stored the corresponding data in a publicly accessi-
ble database (http://www.eswua.ingv.it, Romano et al. 2007).
The IESSG had deployed 4 GISTMs in UK and Norway dur-
3 Experimental setup ing part of the latest solar maximum (June 2001 to December
2003). Recently another GISTM was installed by the IESSG
We tested our method initially on a relative solution. The in Dourbes, Belgium, in collaboration with the Royal Meteo-
experiments were undertaken during periods ionospherically rological Institute. The analyses presented in this paper refer
disturbed, so that it was not expected that errors related to to data collected by these receivers over a region ranging
the ionosphere, in particular scintillation, would cancel out from ∼50◦ N to ∼80◦ N in geographic latitude, as shown in
by differentiation. We used UNESP (São Paulo State Uni- Fig. 1.
versity)’s in-house GPS double difference software GPSeq After a selection of days of disturbed geomagnetic con-
to compare the accuracy on baseline computation based on ditions, when the planetary geomagnetic index Kp is ≥ 4
a conventional, ‘non-mitigated’, or ‘equal weights’ solution and ionospheric scintillations are more likely to occur in
against a ‘mitigated’ solution using the approach described the region being studied (see statistical analyses in Rodri-
in Sect. 2. This software uses double differences of C1 (L1 gues et al. 2004), different baselines were processed. Table 1
C/A code) and P2 (L2 P code) pseudoranges, but also allows describes these baselines and corresponding geomagnetic
the inclusion of L1 and L2 carrier phase data, on an epoch conditions.
by epoch solution. The user can choose between a pseudor-
ange-only solution or a pseudorange/carrier phase combined
solution. It is important to note that all the available observ- 4 Pseudorange processing
ables at each epoch, in our case the C/A and P2 pseudoranges
(and the L1 and L2 carrier phase ranges, if these are to be 4.1 Overall assessment
included), are used and that the algorithm does not perform
any linear combination of them (i.e. for instance no iono- At initial experiments we applied the technique on a C1
free linear combination is attempted). In its original version, and P2 pseudorange-only solution, as previously reported in
the software assigns relative weights between the different Aquino et al. (2006). For an overall assessment of the method,
observables, which however are the same for all satellites each baseline (Table 1) was processed on a 1 min epoch by
at all epochs. The proposed mitigated solution is obtained epoch solution using the GPSeq software, for a period of 24 h
by modifying the stochastic model with the introduction of during which elevated values of the scintillation indices were
satellite and epoch specific weights based on the inverse of observed. No modeling of error sources such as the tropo-
the variances of the output error of the GPS receiver DLL sphere and the background ionosphere was attempted, nei-
and PLL (of L1 and L2), calculated by the models of Conker ther for the mitigated nor the non-mitigated solution, so that
et al. (2003). GPSeq was modified to include this option. they could be compared. Each baseline solution is analysed
The receivers used in the experiments were the GISTMs individually. Results were compared with accurately deter-
(GPS Ionospheric Scintillations and TEC Monitor) mined ground truth coordinates of the stations. Table 2 gives
(Van Dierendonck et al. 1993), so the S4 and σφ indices a summary of the results for a non-mitigated solution against
required for the computations were readily available. the proposed mitigated solution, showing the improvement
Although these receivers have not been designed for use in on the horizontal and vertical components when the pro-
geodetic applications, previous experiments involving posed approach is applied. In the last column of Table 2, we
pseudorange and carrier phase processing with GISTM data show the percentage improvement on the error in height, the

123
Improving the GNSS positioning stochastic model 957

Fig. 1 Network of receivers


used in the experiments

Table 1 Description of baselines processed with the exception of very few epochs when the mitigated
Baseline Length Date Maximum solution worsens the results (epoch 562 the most notable
(km) Kp one). A possible reason is that the S4 index of a satellite pre-
sented sudden jumps at these epochs, which may relate to
Nya0/Nya1 1 30/10/2004 6
receiver firmware miscomputations that occasionally occur.
Lyb0/Nya1 125 10/12/2006 4
These jumps lead to assigning a poor precision to the range
Lyb0/Nya1 125 11/12/2006 5 of that satellite, and to the wrong stochastic representation
Nott/Dour 511 28/07/2006 6 of the data.
Bron/Hamm 752 10/06/2002 5

4.2 Multipath considerations

component usually most affected by the ionosphere. It is In order to investigate the influence of multipath and to dis-
evident from Table 2 that the proposed method leads to an criminate between potential multipath fades and actual occur-
improved least squares solution by correctly assigning rela- rence of amplitude scintillation (measured by S4) we used the
tive weights to the observations. approach suggested by Van Dierendonck (2001). The basis
These results indicate that the improved stochastic model of this approach is the fact that in the presence of multi-
can successfully mitigate both horizontal and vertical errors. path the divergence of code/carrier phase is faster than the
However, for the remainder of this paper we concentrate on divergence due to the ionosphere. The GISTM receiver cal-
the errors in height, assuming these to represent, in most culates the code/carrier divergence every second, estimat-
instances, the worst case scenario. More detailed analyses on ing their average and standard deviation every minute on
horizontal effects will be the subject of follow-on research. the minute. The standard deviation of the code/carrier diver-
Results for baseline Lyb0/Nya1 on 10/12/2006 were fur- gence can then be used to distinguish between S4 due to
ther analysed. Figure 2 shows the state of the ionosphere multipath and S4 due to scintillation, as there is no diver-
by the 15 s TEC rate for all satellites observed at station gence due to scintillation (GPS Silicon Valley 2004). If the
Nya1. The 15 s TEC rate is the change in TEC for a par- values of the standard deviation of code/carrier divergence
ticular satellite/receiver link during a 15-s period, calculated are plotted against S4 values, true amplitude scintillation
based on the carrier phase measurements at L1 and L2 to that will be associated with low standard deviations, while multi-
satellite. path will be associated with high standard deviations. Based
Figures 3 and 4 are the time series of the height errors on this plot, the method uses a dividing line to distinguish
respectively for the non-mitigated solution and the solution between multipath and amplitude scintillation. This line
based on our approach. The improvement is quite significant, is only an approximation to filter out multipath and

123
958 M. Aquino et al.

Table 2 Results of
pseudorange-only solution Baseline RMS error (non-mitigated) RMS error (mitigated) % improvement
on the error in height
Horizontal Vertical Horizontal Vertical
(m) (m) (m) (m)

Nya0/Nya1 (30/10/2004) 0.26 0.54 0.18 0.43 21


Lyb0/Nya1 (10/12/06) 0.81 1.44 0.61 0.90 38
Lyb0/Nya1 (11/12/06) 0.78 1.28 0.55 0.85 34
Nott/Dour (28/07/2006) 1.71 1.52 1.31 1.26 17
Bron/Hamm (10/6/2002) 4.81 6.43 4.20 5.32 17

Fig. 2 TEC rate for all 4


15 s TEC Rate at Nya1, 10 dec 2006
satellites observed at station
Nya1
3

2
dTEC (TEC units)

-1

-2

-3

3:00 06:00 09:00 12:00 15:00 18:00 21:00


-4 I I I I I I I
Time (UT)

Fig. 3 Height error from 15


pseudorange-only solution Lyb0/Nya1 baseline, 10 Dec 2006, C1 and P2 only
without mitigation non-mitigated solution
10

5
height error (m)

-5

-10
rms = 1.44 m
Minutes on the Day (24 hours - 10Dec06)
-15
1 181 361 541 721 901 1081 1261

its choice is a compromise between the knowledge of the which was estimated from experience for the case of a chal-
environmental multipath at the station and the probability lenging multipath environment (Van Dierendonck, personal
of scintillation occurrence. For illustration, in this paper we communication, 2009). It is important to note that the def-
adopted a conservative approach and used a dividing line inition of this dividing line does not play any direct role in

123
Improving the GNSS positioning stochastic model 959

Fig. 4 Height error from 15


pseudorange-only solution with Lyb0/Nya1 baseline, 10 Dec 2006, C1 and P2 only
mitigation mitigated solution (Conker models)
10

height error (m)


0

-5

-10

rms = 0.90 m
Minutes on the Day (24 hours - 10Dec06)
-15
1 181 361 541 721 901 1081 1261

our proposed mitigation strategy. The aim here is solely to Figure 7 shows the difference in the height error between
illustrate a possible method to identify the influence of mul- the two days when the solutions are not mitigated. Results of
tipath on the S4 index. the second day were shifted by 4 min so that the geometry
Following this approach, in Fig. 5a and b, the standard of the arriving signals is approximately the same, and mul-
deviation of the L1 code/carrier divergence (Sigma CCDiv), tipath is mostly removed. In Fig. 8, the same was done with
measured by the GISTM, respectively, at stations Lyb0 and the mitigated solutions. Clearly, on an epoch by epoch basis,
Nya1, is plotted against S4, which was corrected for ambi- the proposed mitigation approach leads to the (non-multi-
ent noise (Corrected S4). In the plots, the multipath filter- path related) errors overall being reduced, with the RMS of
ing dividing line (dotted line), chosen as explained above, these differences decreasing from 1.17 to 0.84 m, between
discriminates between S4 predominantly due to multipath the non-mitigated and the mitigate solutions, as in Figs. 7
fading (points above the line) and possible amplitude scintil- and 8, respectively.
lation (points below the line). While for station Lyb0 the plot
shows a few possible occurrences of significant amplitude 4.3 Amplitude scintillation
scintillation (S4 > 0.5), for station Nya1 almost all occur-
rences of S4 below the line are <0.5 characterising weak to To ascertain how the method handles actual occurrence of
moderate amplitude scintillation. Indeed long term statistics scintillation, despite the low values of S4 observed in our
by Rodrigues et al. (2004) confirm that phase scintillation is data set, we selected a 1-h session (22:00 to 23:00 UT on
dominant and amplitude scintillation is small at GPS frequen- 10/12/2006), when above average values of S4 occurred for
cies in this region. Furthermore in Fig. 6 it can be seen that the satellites observed at high elevation from station Lyb0. The
higher values of (corrected) S4 occur mostly at low elevation baseline is Lyb0/Nya1. Figure 9 shows the corrected S4 for
angles, in particular for station Nya1, where nearly all values station Lyb0, exclusively for satellite elevations >30◦ .
> 0.3 refer to angles below 15◦ (data was processed with a Although high S4 values are not observed, an enhancement
5◦ elevation cut-off). Although station Lyb0 presents higher against the overall background is seen early in the session,
levels of S4, the overall picture suggests that these higher val- which could in all likelihood relate to (albeit weak) ampli-
ues in reality predominantly relate to multipath, rather than tude scintillation (and not multipath). At station Nya1 lev-
to actual amplitude fades due to the ionosphere (scintilla- els of S4 are negligible (<0.07). Figure 10 shows the height
tion). What is interesting however is that, with the input of errors when data is processed without applying our technique
the corrected S4 in the DLL tracking model, our approach (non-mitigated solution). Slightly larger errors roughly cor-
provides an effective way to improve the solution, clearly respond with the period when higher S4 values occurred.
also mitigating any multipath effects that may be degrading Figure 11 then shows the results when the mitigation tech-
the DLL tracking performance. Despite this finding, further nique is applied. Improvement at the start of the session is
analyses were carried out aiming to assess how much of non- apparent, with the session’s RMS height error dropping from
multipath related error the method is able to mitigate. This is 1.38 to 0.89 m. Despite the encouraging results, it is clear
shown in Figs. 7 and 8, where we compare the errors on two however that further tests must be carried out on data sets
consecutive days, 10 and 11 December 2006. with stronger levels of amplitude scintillation.

123
960 M. Aquino et al.

Fig. 5 Multipath and possible 1


(a)
amplitude scintillation for
0.9
stations Lyb0 (a) and Nya1 (b) Lyb0 station, 10 Dec 2006
0.8

0.7

Sigma CCDiv (meters)


Potential multipath (Sigma CCDiv > 0.625 S4corr - 0.125)
0.6

0.5

0.4

0.3

Possible
0.2
amplitude scintillation
0.1 (weak for S4 < 0.4)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Corrected S4

1
(b)
0.9
Nya1 station, 10 Dec 2006
0.8
Sigma CCDiv (meters)

0.7
Potential multipath (Sigma CCDiv > 0.625 S4corr - 0.125)
0.6

0.5

0.4

0.3

0.2 Possible
amplitude scintillation
0.1
(weak for S4 < 0.4)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Corrected S4

4.4 Alternative approaches the pseudorange tracking error, suitable to improve the sto-
chastic model. Results presented in Figs. 12 and 13 are com-
Finally, Figs. 12 and 13 show another two potential approa- patible with the results of our approach and imply that, at least
ches to mitigate positional errors. Baseline and date are the under weak amplitude scintillation, all three may perform
same as in Figs. 3 and 4. Figure 12 shows the height errors equally well. We expect, however, that under stronger ampli-
for a DGPS solution (differential GPS, based only on the tude scintillation our approach will be more robust. This must
C1 pseudoranges) where observations are weighted by the be tested in future, possibly using data from equatorial lati-
sine squared of the elevation angle of individual satellites. tudes, where higher values of S4 are observed.
Figure 13 shows a mitigated solution where we used the
inverse of the square of the standard deviation of the L1
code/carrier divergence to define individual satellites and 5 Carrier phase processing
epochwise weights for the C1 measurements, and the inverse
of the DLL variance (from Conker et al. 2003) to similarly Kim and Tinin (2007) used perturbation theory to investi-
define weights for the P2 observables. In the absence of a gate the residual errors in dual-frequency GPS measurements
cycle slip, the error in the code/carrier divergence combines and concluded that random irregularities in the ionosphere
multipath and ionospheric errors of both the pseudorange can significantly contribute to the phase range error. Aquino
and the carrier (plus noise of both). As these are small for et al. (2005a,b, 2007a) carried out real data experiments
the carrier phase when compared with the pseudorange, this where errors in carrier phase based GPS positioning were
parameter (the CCDiv error, or Sigma CCDiv) is essentially found to be correlated with scintillation events caused, in all

123
Improving the GNSS positioning stochastic model 961

Fig. 6 Corrected S4 against 90


elevation angle for stations Lyb0 (a)
(a) and Nya1 (b) 80 Lyb0 station, 10 Dec 2006
70

60

Elevation angle
50

40

30
S4 > 0.3 mostly for low elevation angles
20

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Corrected S4

90
(b)
80 Nya1 station, 10 Dec 2006
70

60
Elevation angle

50

40

30
S4 > 0.3 mostly for low elevation angles
20

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Corrected S4

Fig. 7 Height error difference 15


between 2 days without 11 Dec - 10 Dec, non-mitigated solutions
mitigation
10
height error difference (m)

-5

-10

rms = 1.17 m
Time - minutes on 24 hours
-15
1 181 361 541 721 901 1081 1261

123
962 M. Aquino et al.

Fig. 8 Height error difference 15


between 2 days with mitigation 11 Dec - 10 Dec, mitigated solutions

10

height error difference (m)


5

-5

-10
rms = 0.84m
Time - minutes on 24 hours
-15
1 181 361 541 721 901 1081 1261

Fig. 9 Corrected S4 for S4 corrected for ambient noise at station LYB0


satellites observed at station
all satellites above 30o, 22-23UT 10 Dec 2006
Lyb0 with elevation >30◦ 0.5

0.45

0.4

0.35

0.3

0.25

0.2

0.15

0.1

0.05

0
22:15 22:30 22:45

Fig. 10 Height error from 15


pseudorange-only solution LYB0-NYA1 baseline, 22-23UT, 10 Dec 2006, C1 and P2 only
without mitigation non-mitigated solution
10

5
height error (m)

-5

-10

rms = 1.38 m 22-23UT, 10Dec06

-15
22:15 22:30 22:45

123
Improving the GNSS positioning stochastic model 963

Fig. 11 Height error from 15


pseudorange-only solution with LYB0-NYA1 baseline, 22-23UT, 10 Dec 2006, C1 and P2 only
mitigation mitigated solution
10

height error (m)


0

-5

-10

rms = 0.89 m 22-23UT, 10Dec06

-15
22:15 22:30 22:45

Fig. 12 Height errors from 15


DGPS Solution and weighting Lyb0-Nya1, 10 Dec 2006, DGPS solution (C1 only)
by sin (elevation angle) sin(elev) weighting
10

5
height error (m)

-5

-10
rms = 0.88 m
Minutes on the day (24 hours - 10Dec06)
-15
1 181 361 541 721 901 1081 1261

Fig. 13 Height errors from 15


mitigated solution based on Lyb0-Nya1 baseline, 10 Dec 2006, C1 and P2 only
weighting C1 measurements by sigma CCdiv weighting (on C1)
the standard deviation of the 10
code/carrier divergence

5
height error (m)

-5

-10

rms = 0.96 m
Minutes on the Hour (24 hours - 10Dec06)
-15
1 181 361 541 721 901 1081 1261

123
964 M. Aquino et al.

Fig. 14 Values of Phi60 Phi60 measured at station Nya1, 10 Dec 2006, 22-23UT
between 22:00 and 23:00 UT at 1.4
station Nya1 on 10 December
2006
1.2

Phi60 (radians)
0.8

0.6

0.4

0.2

0
Time - 22-23UT of 10 December 2006

Fig. 15 Height errors from Lyb0-Nya1 baseline, 22-23UT, 10 Dec 2006


pseudoranges and carrier phase phase and code solution
solution without mitigation 20
no weighting
15

10
height error (m)

-5

-10

-15
rms = 1.87 m
-20
22-23UT, 10Dec06

Fig. 16 Height errors from Lyb0-Nya1 baseline, 22-23UT, 10 Dec 2006


pseudoranges and carrier phase phase and code solution
solution with mitigation 20
mitigated solution
15

10
height error (m)

-5

-10

-15
rms = 0.88 m
-20
22-23UT, 10Dec06

123
Improving the GNSS positioning stochastic model 965

likelihood, by small scale ionospheric irregularities. In this and on suitable receiver tracking models. Results indicate
paper, initial experiments aiming to test the proposed mitiga- potential for the development and implementation of algo-
tion approach were carried out involving carrier phase data. rithms that could be embedded in a scintillation monitoring
GPSeq was used in a 1 min epoch by epoch solution based on network with the aim to mitigate these effects. The exper-
both pseudoranges and phase measurements on the L1 and iments presented in this paper and corresponding results,
L2 GPS frequencies. We analysed the 1-h session from base- although encouraging, are considered of a preliminary nature
line Lyb0/Nya1, between 22:00 and 23:00 UT on 10/12/2006. and further research is necessary for future usability of our
During that session strong GPS ionospheric phase scintilla- method. This paper is intended as a proof of concept, as
tion was observed by the GISTMs in the region. Figure 14 clearly, normal receivers do not and will not in principle
shows the values of Phi60 (the 60 s version of σφ ) observed at measure scintillation parameters. However, as scintillation
station Nya1. For each epoch the 1 min 50 Hz data was used prediction models are improved it is legitimate to imagine
to calculate the spectral parameters T and p from the corre- that the method can be used in conjunction with one such
sponding PSD, as explained in Sect. 2.1. The height errors, model, e.g. an improved WBMOD (Wide Band MODel)
again calculated from comparison against ground truth coor- Ionospheric Scintillation Model (http://www.nwra-az.com/
dinates, for the non-mitigated and mitigated solutions are ionoscint/wbmod.html) or the GISM (Global Ionosphere
shown in Figs. 15 and 16, respectively. Figures 14 and 15 Scintillation Model—Béniguel 2002). Furthermore the
show the correlation of larger height errors with high val- method may also be used in post-processing, in particular
ues of Phi60. GPSeq uses the LAMBDA method (Teunissen in applications involving the carrier phase observable, as
1996) to resolve the carrier phase ambiguities. By analysing results in Sect. 5 suggest. Plans are in place to undertake
the results of the method’s ratio test it was possible to ver- further investigations in this area which will take advantage
ify that GPSeq fixed the ambiguities only at certain epochs, of a common database that is being developed. Also, data
exactly in coincidence with the periods of high phase scintil- from equatorial regions will be taken into account.
lation occurrence, and that they were however fixed wrongly.
Clearly, given the length of the baseline, fixing the ambigu- Acknowledgments Data used in this research was obtained through
funding from the Engineering and Physical Sciences Research Council
ities to their correct values would be very unlikely, however (EPSRC) in the UK and from the ISACCO Project, supported by the
the occurrence of scintillation somehow led to this outcome Italian Program of Antarctic Researches (PNRA). Thanks are due to
and consequently to unreliable epochwise solutions (Fig. 15). the Royal Society for two International Joint Project grants involving
Interestingly, when applying our mitigation approach, the IESSG, INGV and UNESP. This collaboration has also been motivated
by the work carried out within the COST296 MIERS (Mitigation of
float ambiguities solution prevails at all epochs, providing Ionospheric Effects on Radio Systems). Thanks are also due to CAPES
more accurate solutions (Fig. 16). (Brazilian Research Council) for providing funds to UNESP based co-
For practical future use of the method in carrier phase posi- authors.
tioning it is envisaged that the strategy proposed in Aquino
et al. (2007b) and further developed in Strangeways (2008)
can be applied, where the scintillation indices can be used
(without the direct need of high rate phase data) to compute References
the PLL tracking error variance.
Aarons J (1997) Global positioning system phase fluctuations at auroral
latitudes. J Geophys Res 102(A8):17219–17232
Alves DBM, Monico JFG (2007) Modifying the stochastic model to
6 Conclusions mitigate GPS systematic errors in relative positioning. IAG Sym-
posia (Springer), vol 130, Chapter 26, pp 166–171
Ionospheric scintillations are caused by time-varying den- Alfonsi L, De Franceschi G, Romano V, Aquino M and Dodson A (2006)
sity irregularities in the ionosphere, occurring more often at Positioning errors during the space weather event of October 2003.
Location, ISSN 0973-4627, 1, issue 5
equatorial and high latitudes under external disturbed condi- Aquino M, Rodrigues FS, Souter J, Moore T, Dodson A, Waugh
tions. They can degrade navigation and positioning accuracy S (2005) Ionospheric scintillation and impact on GNSS users
and may cause loss of signal tracking, disrupting safety-criti- in Northern Europe: results of a 3 year study. Space Commun
cal applications, such as marine navigation and civil aviation. 20(1/2):17–30
Aquino M, Moore T, Dodson A, Waugh S, Souter J, Rodrigues
Degradation of accuracy due to the occurrence of scintillation FS (2005) Implications of ionospheric scintillation for GNSS
can also affect carrier phase based applications and in partic- users in Northern Europe. J Navig 58(2):241–256
ular disrupt the resolution of the ambiguities inherent to the Aquino M, Monico JFG, Dodson A, Marques H (2006) Mitigating the
carrier phase observable. This paper describes a technique effect of ionospheric scintillations on position estimates. In: online
Proceedings of ESA 3rd European space weather week, Brussels,
that was successfully used to counter the effects of scintilla- Belgium. http://sidc.oma.be/esww3
tions on GNSS positioning. The proposed approach is based Aquino M, Dodson A, Souter J, Moore T (2007a) Ionospheric scintil-
on data obtained from GPS scintillation monitor receivers lation effects on GPS carrier phase positioning accuracy: analysis

123
966 M. Aquino et al.

at Auroral and Sub-Auroral latitudes. IAG Symposia (Springer), of the satellite division of the Institute of Navigation: ION GPS
vol 130, chapter 121, pp 859–866 2002, Portland, Oregon, USA, pp 1297–1306
Aquino M, Andreotti M, Dodson A, Strangeways H (2007) On the use Rodrigues FS, Aquino M, Dodson A, Moore T, Waugh S (2004) Statis-
of ionospheric scintillation indices as input to receiver tracking tical analysis of GPS ionospheric scintillation and short-time TEC
models. J Adv Space Res 40(3):426–435 variations over Northern Europe. J Inst Navig 51(1):59–75
Béniguel Y (2002) GISM, A Global Ionospheric Propagation Model Romano V, Pau S, Pezzopane M, Zuccheretti E, Zolesi B, De Franceschi
for scintillations of transmitted signals. Radio Sci 37(3). doi:10. G, Locatelli S (2007) The electronic space weather upper atmo-
1029/2000RS002393 sphere (eSWua) project at INGV: advancements and state of the
Conker RS, El-Arini MB, Hegarty CJ, Hsiao T (2003) Modeling the art. Ann Geophys 25:1–7 (in press)
effects of ionospheric scintillation on GPS/satellite-based aug- Skone S, Yousuf R, Coster A (2004) Performance evaluation of the
mentation system availability. Radio Sci 38(1):1001. doi:10.1029/ Wide Area Augmentation System for ionospheric storm events.
2000RS002604 J Glob Position Syst 3(1–2):251–258
De Franceschi G, Alfonsi L, Romano V (2006) ISACCO: an Italian Souza EM, Monico JFG (2007) The wavelet method as an alternative
project to monitor the high latitudes ionosphere by means of GPS for reducing ionospheric effects from L1 GPS receivers. J Geod
receivers. GPS Solut 10(4):263–267 81(12):799–804
De Franceschi G, Alfonsi L, Romano V, Aquino M, Dodson A, Mitch- Strangeways HJ (2008) Determining scintillation effects on GPS receiv-
ell C N, Wernik AW (2008) Dynamics of high latitude patches ers. In: Proceedings of the 12th international ionospheric effects
and associated small scale irregularities. J Atmos Solar-Terrestrial symposium IES2008, Alexandria, Washington DC, USA, 13–15
Phys 70:879–888. doi:10.1016/j.jastp.2007.05.018 May 2008, pp 550–557
GPS Silicon Valley (2004) GSV4004/GSV4004A GPS Ionospheric Teunissen PJG (1996) GPS carrier phase ambiguity fixing concepts.
Scintillation and TEC Monitor (GISTM) User’s Manual In: Kleusberg A and Teunissen P (eds) GPS for Geodesy. Verlag,
Grejner-Brzezinska D, Wielgosz P, Kashani I, Smith DA, Robertson DS, Berlin, pp 263–336
Mader GL, Komjathy A (2006) The impact of severe ionospheric Van Dierendonck AJ, Klobuchar J, Hua Q (1993) Ionospheric scintil-
conditions on the accuracy of kinematic position estimation: per- lation monitoring using commercial single frequency C/A code
formance analysis of various ionospheric modeling techniques. receivers. In: Proceedings ION GPS-93: sixth international tech-
Navigation 53(3):203–217 nical meeting of the satellite division of the Institute of Navigation,
Kim BC, Tinin MV (2007) Contribution of ionospheric irregularities to Salt Lake City, Utah, pp 1333–1342
the error of dual-frequency GNSS positioning. J Geod 81(3):189– Van Dierendonck AJ (2001) Measuring ionospheric scintillation effects
199 from GPS signals. In: Proceedings of 57th annual meeting of the
Moore T, Aquino M, Waugh S, Dodson A, Hill C (2002) Evaluation Institute of Navigation, Albuquerque, New Mexico, USA, pp 391–
of the EGNOS ionospheric correction model under scintillation in 396
Northern Europe. In: Proceedings of the 15th technical meeting

123

You might also like