You are on page 1of 24

Accepted Manuscript

An improved method for multi-GNSS baseline processing using single differ-


ence

Hua Chen, Yugang Xiao, Weiping Jiang, Xiaohui Zhou, Hongfei Liu

PII: S0273-1177(17)30642-7
DOI: http://dx.doi.org/10.1016/j.asr.2017.09.009
Reference: JASR 13403

To appear in: Advances in Space Research

Received Date: 18 April 2017


Revised Date: 20 August 2017
Accepted Date: 6 September 2017

Please cite this article as: Chen, H., Xiao, Y., Jiang, W., Zhou, X., Liu, H., An improved method for multi-GNSS
baseline processing using single difference, Advances in Space Research (2017), doi: http://dx.doi.org/10.1016/
j.asr.2017.09.009

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
An improved method for multi-GNSS baseline processing using single difference
Hua Chen1, Yugang Xiao2, Weiping Jiang3,*, Xiaohui Zhou1, Hongfei Liu4

1
School of Geodesy and Geomatics, Wuhan University, Wuhan 430079, China
2
Changjiang Spatial Information Technology Engineering Co. Ltd., Wuhan 430010, China;
3
Research Center of GNSS, Wuhan University, Wuhan 430079, China;
4
Yunnan Province Surveying and Mapping Engineering Institute, Kunming 650033,China

*
Corresponding author: wpjiang@whu.edu.cn
Abstract: The rapid development of multi-GNSS (Global Navigation Satellite System)
constellations has provided numerous benefits for common PNT (Positioning, Navigation and Timing)
services. Generally,DD (Double Difference) observations are used in GNSS baseline processing.
However, in traditional DD data processing methods, DD observations are normally not formed
between different GNSS systems or the inter-system DD ambiguities are rather difficult to be resolved
due to the different wavelength. As a result, DD observations are normally formed in each system or
their inter-system ambiguities are left unresolved, which could not be regarded as the optimal baseline
processing methods for multi-GNSS. This paper presents an improved algorithm for multi-GNSS short
baseline processing where SD (Single Difference) rather than DD observations are formed between two
receivers and this method is found to be equivalent to the ideal DD method with inter-system DD
observations and resolved inter-system DD ambiguities. Furthermore, an average approach and linear
model are proposed to isolate the receiver-dependent Uncalibrated Phase Delays (UPDs) from the SD
ambiguities for CDMA (Code Division Multiple Access) and FDMA (Frequency Division Multiple
Access) GNSS systems, respectively, to address ambiguity resolution. Experiments show that after
removing the UPDs derived by the proposed methods, the fractional parts of the residuals for almost all
ambiguities are less than 0.1 cycles for the GPS, BDS, GLONASS and Galileo systems, which
confirms the validity of our UPD calibration methods. Experiments also show that the mean differences
between the daily solutions derived by GAMIT and the 4-hour solutions derived by the proposed SD
method are approximately -0.12, -0.33, 1.04 and 0.31 cm in the north, east, up and length components,
respectively. We also find that the baseline repeatability of the new SD method outperforms that of the
DD method with only DD observations formed in each GNSS system by 36% and 26% in the north and
up directions, respectively. These improvements are more significant in environments with poor
observation conditions. An improvement of 40% and 30% in the north and up directions are,
respectively, found when only 5 GPS and 4 BDS satellites are available. Therefore, more reliable and
precise baselines can be derived with our new baseline processing method in multi-GNSS cases.

1
Key words: Multi-GNSS; single difference; UPD; baseline processing; MGEX

1. Introduction

With the rapid development of GPS over the last several decades, the importance and merits of
navigation satellite systems have been fully recognized and demonstrated. Several nations and
organizations have begun building their own GNSS (Global Navigation Satellite System) or RNSS
(Regional Navigation Satellite System) to safeguard security and economic benefits. New GNSS
systems, such as Galileo and BDS, are becoming available in addition to the two fully operational
systems GPS and GLONASS (https://www.glonass-iac.ru/en;
http://www.gps.gov/systems/gps/modernization; Al-Shaery et al. 2013). As of Dec. 1, 2015, there are
10 and 20 operational satellites for Galileo and BDS, respectively, and all new satellites transmit
signals over at least three frequency bands (http://ec.europa.eu/growth/sectors/space/galileo/launches;
http://www.beidou.gov.cn/xtjs.html). To improve the performance of existing PNT (Positioning,
Navigation and Timing) services, GPS and GLONASS are slated to be modernized (Urlichich et al.
2011; Revnivykh et al. 2012). For example, a third frequency—L5—is available on new GPS BLOCK
IIF satellites and a new CDMA (Code Division Multiple Access) signal is aligned to the new
GLONASS-K satellites. Moreover, the future GLONASS satellites such as GLONASS-K2,
GLONASS-KM will introduce multi-frequency CDMA signals, which will further improve its
performance (http://www.navipedia.net/index.php/GLONASS_Future_and_Evolutions). In the near
future, there will be 4 GNSS systems and more than 100 navigation satellites for international users (Li
et al. 2015a).
Reports indicate that GNSS performance will be largely degraded due to the blockage of
transmitting signals (Montenbruck et al. 2013; He et al. 2014) in areas with only a few visible satellites,
such as urban canyons, deep open-pit mines and river valleys. Fortunately, this situation can be
improved by using data from multi-GNSS and multi-signals that have become increasingly attractive
and popular in various GNSS applications (Dai 2000; Yamada et al. 2010; Ge et al. 2012; Chen et al.
2014a, b; Li et al. 2015b). Thanks to the development of multi-GNSS experiments (MGEX) initiated
by the International GNSS Service (IGS) in 2012 (Montenbruck et al. 2014b), the benefits of
multi-GNSS in such areas as PPP (Precise Point Positioning; Li et al. 2015a, b; Lou et al. 2015), POD
(Precise Orbit Determination; Lou et al. 2014) and other GNSS applications (Xu 2014) were verified.
In actual data processing, the DD (Double Difference) method (Bock et al. 1986; Dong and Bock
1989; Blewitt 1989) is still very popular for baseline processing due to the elimination of satellite- and
receiver-dependent biases and the mitigation of atmospheric delays by forming DD observations
between two satellites and two stations. However, in multi-constellation cases, DD observations cannot
be formed directly between two different GNSS systems without considering inter-system bias (ISB).

2
Even if the ISB is corrected, the inter-system DD ambiguity resolution (AR) is problematic due to the
different frequency bands of different GNSS systems (Tian et al., 2017). This situation can be
exacerbated if GLONASS is adopted because of the disturbance of inter-frequency bias (IFB).
Consequently, DD observations are typically formed in each system or inter-system DD ambiguities are
always unresolved, thus making it difficult to fully utilize multi-GNSS in PNT services (Pratt et al.
1998; Al-Shaery et al. 2013). Therefore, the normal DD method could not be regarded as the optimal
baseline processing method in multi-GNSS cases. Nowadays, the traditional undifferenced method,
such as PPP, is popular in data processing for multi-GNSS (Cai and Gao 2007, 2013; Tu et al. 2013; Li
et al. 2014). However, its performance depends heavily on the quality of orbit products and satellite
clock products provided by the IGS or others, which hinders the application of GNSS in certain
domains with high-precision and high temporal requirements, such as deformation monitoring.
To overcome the shortcomings of the existing multi-GNSS data processing methods mentioned
above, an improved short baseline processing method is proposed where SD instead of DD
observations between different stations are used, and this new method is found to be theoretically equal
to the ideal double difference method where DD observations between different GNSS systems are
formed and inter-system DD ambiguities are also resolved. And the equivalency would be discussed
and proved later. Furthermore, in order to resolve the SD ambiguities, methods are also proposed to
isolate the receiver-dependent UPDs (Uncalibrated Phase Delays) from ambiguities for CDMA and
FDMA GNSS systems, respectively. It should be noted that, the UPDs in this paper is different from
those for PPP in Ge et al. (2008), Collins (2008) and Li et al. (2011). For PPP, UPDs normally
represent satellite-dependent UPDs which need to be estimated in advance from a global or regional
network, and the receiver-dependent UPDs are normally removed by difference between satellites (Ge
et al. 2008) or absorbed by receiver clocks (Colins 2008). In this paper, only receiver-dependent UPDs
are discussed due to the satellite-dependent UPDs are removed by difference between stations. This
new method is validated by experiments with data from the MGEX network using all 4 GNSS systems.
The features of the Differential Code Bias (DCBs), Inter-system Bias (ISBs), Inter-frequency Bias
(IFBs) and UPDs derived from this new method are also investigated to further improve the quality of
products for multi-GNSS and multi-signals.
A brief overview of the traditional DD algorithm is presented in the next section. The improved
model for multi-GNSS baseline data processing and its benefits are introduced and discussed in the
third section. The experimental validation and conclusions are shown in fourth and fifth sections,
respectively.

2. DD method for baseline data processing

Double difference observations between receivers and satellites have been widely utilized in data

3
processing since the onset of GPS applications (Dong and Bock 1989; Rothacher et al. 1993), and they
have been widely applied to ground positioning and crustal deformation studies (Dong et al. 1989). The
corresponding model is rather simple and its computation is very efficient because receiver and satellite
clock biases, as well as UPDs, are eliminated. Moreover, DD ambiguities have integer values in this
case and can be estimated and fixed directly, although it is slightly different in the multi-GNSS case.
It is well known that the basic equations for GNSS code and phase observations can be generally
expressed as

Pi ks  is  cdti  cdt s  misTi  nis I i  uiks  hiks  iks


(1)
 ksiks  is  cdti  cdt s  misTi  nis I i   ks biks   iks   iks   iks

where P and  are the code and phase measurements, respectively;  is the wavelength; i , k

and s are the indices for receivers, frequencies and satellites, respectively;  is the geometric

distance between the satellites and receiver antennas; c is the speed of light in a vacuum; dti and

dt s are the receiver and satellite clock biases, respectively; m and n are the mapping functions

for tropospheric and ionospheric delays, respectively; T and I are the tropospheric and ionospheric
zenith delays, respectively; b is the integer ambiguity in cycles;  and  are the receiver- and

satellite-dependent UPDs (Ge et al. 2008; Li et al. 2011), respectively; u and h are the code bias
for receiver and satellite, respectively; and  and  are the measurement noise for the code and

phase, respectively.
The receiver and satellite clock biases are totally eliminated by forming the DD observations. For
simplicity, we assume that the baseline is very short (e.g., within a few km) (Al-Shaery et al. 2013) so
that the ionospheric and tropospheric delays can be reduced to a negligible level. Based on Eq. (1), the

DD equations between stations i and j and satellites s and q can be expressed as

Pijksq  ijsq   uijks  uijkq   ijksq


(2)
 ksijks   kqijkq  ijsq   ks bijks   kq bijkq   ijks   ijkq    ijksq

where Pijksq is the DD code observation; ijsq is the DD geometric distance; uijks and uijkq are the

j for satellites s and q , respectively; ij and


ks
SD receiver code biases between stations i and

ijkq are the SD phase observations; bijks and bijkq are the SD ambiguities;  ijks and  ijkq are the

j for satellites s and q , respectively; and ij and


ksq
SD receiver UPDs between stations i and

4
 ijksq are the DD noise for code and phase, respectively. For a multi-GNSS case, ISB should be

included in the equation (1), however, due to DD could eliminate the ISB, no special bias need to be
considered in equation (2).
It should be mentioned that when the DD operation is carried out in one GNSS system, both

u ks
ij  uijkq  and  ijks   ijkq  in Eq. (2) are zero and  ks is equal to  kq . In this case, the

ambiguities can be directly estimated and fixed. However, when the DD operation is implemented
between two different systems, the situations above become rather complicated. The inter-system
biases must be estimated or corrected at the inter-system DD observations. And the frequencies of
different GNSS systems are typically different, which indicates that the combined DD ambiguity would
have no integer nature anymore. In this case, the ambiguity resolution can be rather complicated,
especially when the GLONASS data are included. These problems hinder the application of the DD
method in multi-GNSS. Because of this, a new algorithm must be developed to utilize the full benefits
of multi-GNSS.

3. SD method for baseline data processing

3.1 Mathematical model

For short baselines, the atmospheric delays can usually be mitigated to a negligible level; then the SD
observation equations can be derived from Eq. (1) as

Pijks  ijs  cdtij  uijks  ijks


(3)
 ksijks  ijs  cdtij   ks bijks   ijks   ijks

where Pijks is the differenced code between two receivers; ijs is the differenced geometric distance;

dtij is the differenced receiver clock bias; uijks is the differenced receiver code bias; ijks and bijks

are the SD phase and ambiguity, respectively;  ijks is the differenced receiver UPD; and ijks and

 ijks are the SD noises for code and phase, respectively.

The stochastic model for SD method can be expressed as

CSD  DCUD DT (4)

where CSD is the covariance matrix for SD observations; CUD is the covariance matrix for

undifferenced (UD) observations; and D is the operator matrix to transform all of the observations
from UD to SD. Traditionally, the a priori variances for UD phase and code observations can be set as

5
0.0052 and 1 m2, respectively. CUD is usually a diagonal matrix with the variances of observations as

the diagonal elements.


As shown in Eq. (3), the satellite clock biases, satellite-dependent code biases and UPDs,
consisting of hardware delays and initial phase biases, can be eliminated by the single difference

approach. Due to the poor stability of receiver clock biases, dtij should be estimated at each epoch,

whereas uijks and  ijks can be estimated as determined parameters simultaneously under the

assumption that the receiver-dependent code biases are stable in one day (Choi et al. 2013;
Montenbruck et al. 2014a). Furthermore, for systems based on CDMA technique (e.g., GPS, Galileo,

BDS), uijks is identical for observations at the same frequency band from any satellite of this system,

as is  ijks . As for the FDMA GLONASS, uijks and  ijks become different for different satellites,

which results in different UPDs and code biases for different satellites.
To avoid the singularity of the final normal equations (NEQ), a datum for the code bias must be
provided. A simple choice is to set the code bias of any type of observations to zero. As a result, the
code bias which is selected to provide the datum appears in the receiver clock biases. Because the
phase and code observations share the same receiver clock parameters, the code bias is also introduced
into the phase observation equations. Then, Eq. (3) can be rewritten as

Pijks  ijs  cdtij'  ijks


 ksijks  ijs  cdtij'   ks bijks   ijks  uijks    ijks
Pijlt  ijt  cdtij'  uijltks  ijlt
 ltijlt  ijt  cdtij'   lt bijlt   ijlt  uijks    ijlt
(5)

dtij'  dtij  uijks / c


uijltks  uijlt  uijks

where k and s are the indices for the referenced frequency and satellite and l and t represent
the non-referenced ones. dtij' is the combined receiver clock bias.  ks
ij  uijks  represents the

combined UPDs along with  lt


ij  uijks  . If the satellites indicated by s and t are from the same

GNSS system, then uijltks denotes the differenced DCB for frequency k and l ; otherwise, uijltks

denotes the differenced ISB between these two receivers.


However, Eq. (5) still cannot be solved directly due to the high correlation between ambiguities
and UPDs. Therefore, combined ambiguity parameters are considered. Eq. (5) is rewritten as

6
Pijks  ijs  cdtij'  ijks
 ksijks  ijs  cdtij'   ks bijks '   ijks
Pijlt  ijt  cdtij'  uijltks  ijlt
 ltijlt  ijt  cdtij'   lt bijlt '   ijlt
(6)
dtij'  dtij  uijks / c
uijltks  uijlt  uijks
bijks '  bijks    ijks  uijks  /  ks
bijlt '  biltj    ijlt  uijks  /  lt

where bijks ' and bijlt ' are the combined ambiguity parameters.  ks
ij  uijks  /  ks and

 lt
ij  uijks  /  lt are the combined UPDs in cycle. In this way, Eq. (6) can be resolved, although the

ambiguities no longer present as integers due to their combination with UPDs.

3.2 Ambiguity resolution

As shown in Eq. (6), the estimated float ambiguities are mixtures of the ambiguities and the combined
UPDs, which must be removed for ambiguity resolution. For systems based on CDMA technique, all
UPDs for the same system and same frequency are typically identical. As a result, the fractional parts
of the combined UPDs can be approximated with an average approach (Ge et al. 2008) directly from
the original estimated float ambiguities, which can be expressed as

 ijks '  bij 


ks

 (7)
 ks n

where  ijks ' is the estimated fractional part of the combined UPDs in meter; bijks is the float

ambiguity from Eq. (6); n is the number of ambiguities of the same type.  indicates taking the

fractional parts of the encompassed values. For clarity,  ijks ' is the UPD discussed below in this paper.

Then the ambiguities with integer nature can be acquired as

 ijks '
ks 
N  b  ks
ks
(8)
ij
 ij

where N ijks are the ambiguities with integer nature.

The ambiguity resolution procedure can be explained as follows: first, a simple round calculation
is carried out for the float ambiguities to obtain the fractional parts of the UPDs; all fractional parts are

then averaged for each system at each frequency band where a possible 1 cycle is considered;
7
finally, the averaged UPDs are utilized to correct the float ambiguities and recover the integer nature of
the ambiguities, and a LAMBDA (Teunissen 1993, 1995) or decision function (Dong et al. 1989)
method is employed for ambiguity resolution. It is worth noting that an iteration procedure is necessary
when calculating the averaged UPDs to eliminate the effects of outliers that typically arise from poor
observations. The threshold for detecting the outliers is usually set to 3 , where  is the STD
value of the UPDs from last iteration.
However, FDMA systems, such as GLONASS, are more complicated due to the existence of IFBs.
Fortunately, Pratt et al. (1998) and Wanninger and Wallstab-Freitag (2007) noted that GLONASS
receiver phase IFBs are nearly linearly correlated to the frequency numbers of satellites, which can be
expressed as

 it  ai  bi  K t (9)

where i and t are the indices of the receiver and satellite, respectively;  it is the receiver phase

IFB in meter; ai and bi are two constants which are generally stable and vary with the receiver type;

K t is the frequency number of satellite t .


Although the combined UPDs of GLONASS may exceed one cycle, the fractional parts of
UPDs can still be assumed to be linearly correlated to the frequency numbers in this paper due to the
slight differences of the wavelengths for different satellites. Combining Eq. (6) and Eq. (9), the
fractional part of GLONASS combined UPD can be approximated as

 ijks '  bijks    ks  aij  bij  K t (10)

where  ijks ' is the fractional part of the GLONASS combined UPD in meter; Then the values of aij

and bij can be acquired with the linear fitting method directly from all the fractional part of combined

UPDs of the same type, and the integer nature of GOLNASS SD ambiguities can be retrieved as

 ijks '
ks 
N  b  ks
ks
ij ij
 (11)
 ij  aij  bij  K t
ks '

where N ijks are the GLONASS ambiguities with integer nature. bijks is the original float ambiguity;

 ks is the wavelength; aij and bij are the estimated constants from Eq. (10).

Similarly, ambiguity resolution can be carried out following the above procedure for CDMA

GNSS, except that a least square approach is used to determine aij and bij to calculate the

8
fractional parts of UPDs instead of an average approach.
Theoretically, the UPD calibration method above can always apply if the ambiguities of the same
type in a session are more than one. When there is only one ambiguity, which rarely happens, it will
have to be left float.
It also should be mentioned that this SD method displayed above is only applicable for short
baselines (e.g., within a few km). For long baseline data processing where the atmospheric delays must
be considered, the algorithm will become much more complicated.

3.3 Advantages of the new method

The new method is theoretically equivalent to the optimal DD method; not only is DD applied between
observations from different systems, but inter-system DD ambiguities are also resolved, which is
clearly better than the current DD methods with only DD observations formed in each system or with
inter-system DD ambiguities are left float. We offer the following example for a better understanding of
this method.
Let us assume that there are 2 receivers observing 2 GPS satellites and 2 BDS satellites. If DD is
only used in each system independently, then only 2 DD observations can be derived. When DD is
applied for both GPS and BDS satellites, 3 DD observations can be obtained. However, as previously
noted, the inter-system DD ambiguity is difficult to resolve due to the different wavelengths of GPS
and BDS signals. In the SD method case, 4 SD observations can be formed and only 1 receiver clock
parameter is introduced, which is equal to the case of DD being applied to all GPS and BDS satellites.
Moreover, all SD ambiguities can be resolved with no wavelength problem. Therefore, the new SD
method should theoretically outperform the current DD method with only DD observations formed in
each system or with float inter-system ambiguities.

4. Experiments and results

To investigate the validity and reliability of the new SD method and demonstrate the benefits of
multi-GNSS, particularly in areas with poor observation conditions, experiments are carried out based
on a self-developed software package. Data from DOY 274 to 303 in 2015 of two MGEX stations,
KIR8 and KIRU, are utilized in these experiments. The MGEX project was initiated by IGS to enable
an early familiarization with the new systems and to prepare their incorporation into high-precision
GNSS modeling and analysis (Montenbruck et al. 2014b). As of October 2015, the MGEX tracking
network included 116 active stations. KIR8 and KIRU are two stations located in Kiruna, Sweden at a
distance of approximately 4.5 km. Both are capable of tracking all GNSS observations, i.e., GPS,
GLONASS, Galileo and BDS. The data processing strategy for multi-GNSS adopted in this experiment
is summarized in Table 1.

9
Table 1 Data processing strategies and models for multi-GNSS

Items Strategies & Models

All parameters, such as station coordinates, clock biases,


Estimator tropospheric delays, DCBs and ISBs, are estimated
through stacking the NEQs
Dual-frequency phase and code, GPS: L1/L2; GLONASS:
Observations
L1/L2; Galileo: E1/E5a; BDS: B1/B2
Sampling interval 30s
Elevation cutoff 10°
Elevation dependent weight. The initial standard deviation
Weight strategy (STD) values for code and phase of all systems are set to 1
and 0.005 m, respectively (Cai and Gao 2013)
Satellite orbit Broadcast ephemeris
Receiver clock Estimated epoch-by-epoch
Corrected with “igs08.atx” and site information from
ftp://cddis.gsfc.nasa.gov/gnss/data/campaign/mgex/log/.
Receiver antenna PCO and PCV
The same corrections as GPS are applied for Galileo and
BDS
Earth rotation error Corrected
Ionospheric delay Neglected for the short baselines
Saastamoinen model with Global Mapping Function. The
Tropospheric delay
residual effects are estimated as constants
Relativistic effects Corrected
DCB, ISB and IFB Estimated as constants

4.1 Satellite visibility

The number of visible GNSS satellites and geometric dilution of precision (GDOP) values are of
interest for navigation and positioning users. GDOP is an important indicator for measuring the
positioning accuracy of GNSS, which represents the magnification of the range error between receiver
and satellite (Yang et al. 2011). An increase in the number of visible satellites usually leads to a
substantial reduction of GDOP values and can mitigate the impact of poor environments, improve the
performance of GNSS PNT services.

10
Fig. 1 Variations of the numbers of visible satellites and GDOP values for GPS-only and multi-GNSS
constellations with the cutoff elevation angles of 10° (top) and 30° (bottom) at station KIR8 on October
1, 2015

Fig. 1 shows the numbers of visible satellites and GDOP values for different constellations and
different cutoff elevation angles at KIR8 on Oct. 1, 2015. When the cutoff angle is 10°, which is a
typical case for GNSS PNT users, the results indicate that the GDOP values for multi-GNSS are more
stable than that for GPS-only constellation. Furthermore, the GPS-only GDOP values show strong
correlations with the number of visible satellites; this is not so obvious for the multi-GNSS. The
number of visible satellites has increased from an average of 10 to 21 through the additional GNSS
satellites and subsequently leads to a reduction in the average GDOP value from 2.6 to 1.6, even
though the full constellations for Galileo and BDS have not been fully achieved. The 62.5% increase of
GDOP values compared to GPS-only will certainly play a significant role in improving the accuracy
and reliability of GNSS applications.
When the cutoff elevation angle is 30° (which is also common in the constrained environments
discussed in the first section), it can be seen that the GDOP values for the GPS-only case show
significant fluctuations at many epochs while the multi-GNSS values remain quite stable. Moreover,
there are 178 epochs (6.2% of the total investigated epochs) when less than 4 GPS satellites are
observed, whereas this significantly improves in the multi-GNSS case. The average numbers of visible
satellites for GPS-only and multi-GNSS are 4 and 11, respectively. Meanwhile, the GDOP values fall
dramatically from an average of 19.5 to 4.3 on account of the additional GNSS satellites. As discussed
above, the GDOP values are directly correlated with the quality of the GNSS services. The substantial
drop in GDOP values shows the potential for improving the performance of GNSS by mitigating the
impact of signal blockage when working in areas with poor observation conditions.

11
4.2 Ambiguity resolution

Integer ambiguity resolution (AR) is crucial for precise GNSS applications. Correct, reliable and fast
ambiguity resolution benefits significantly from a greater number of available multi-GNSS satellites.
However, the ambiguities obtained from the SD method cannot be resolved directly due to the
existence of UPDs. Thus, AR must eliminate the impact of UPDs. The method outlined in the second
section is used to determine the UPDs for GPS, BDS and Galileo satellites; the results are shown in Fig.
2.

Fig. 2 Estimated UPDs of all ambiguity arcs for GPS (top), BDS (middle) and Galileo (bottom) on
October 1, 2015. The average values for each series are shown as equations

It can be seen that most of the UPDs at the same frequency band concur with each other, except
for a few outliers. Further investigation finds that all of the outliers are either ambiguities with few
observations or observations against low elevation angles. The RMS values of the residuals after
removing the UPDs are 0.05 and 0.04 cycles for GPS L1and L2 frequency bands, respectively, which is
0.04 and 0.03 cycles for BDS B1 and B2, and 0.12 and 0.03 cycles for Galileo E1 and E5a frequency
bands, respectively. The fractional parts of the residuals for ambiguities after removing the UPDs are
shown in Fig. 3.

12
Fig. 3 Fractional parts of float ambiguities after eliminating the impact of UPDs for GPS (top), BDS
(middle) and Galileo (bottom) on October 1, 2015

Fig. 3 shows that most of the fractional parts of residuals for ambiguities fluctuate near zero;
approximately 95% of the residuals are less than 0.1 cycles. It can also be seen that the residuals of two
ambiguities at the E1 frequency for Galileo are greater than 0.15 cycles; this is because these two
ambiguities have short observation arcs. Overall, the UPDs and the ambiguities show good
consistencies with each other, which implies that good ambiguity resolution can be carried out after
removing the UPDs.

Fig. 4 Fractional parts of float ambiguities before (top) and after (middle and bottom, in meters and
cycles, respectively) removing the UPDs for GLONASS on October 1, 2015. The two equations shown
in the top sub-figure are the linear fitting results for the GLONASS L1 and L2 ambiguities

As discussed in the second section, the process becomes a bit more complicated for GLONASS.
The top sub-figure in Figure 4 shows the fractional parts of the GLONASS ambiguities, where a clear

13
linear relationship can be found. The linear slopes are calculated with the method discussed in the
second section; we found the slopes for UPDs at both L1 and L2 frequency bands to be -0.0074 m/FN,
whereas the offsets are -0.0012 m and 0.0629 m at the L1 and L2 frequency bands, respectively. These
results coincide with those of Pratt et al. (1998), Wanninger and Wallstab-Freitag (2007) and others.
The middle sub-figure in Figure 4 shows the residuals of ambiguity fractional parts after removing the
UPDs calculated by the linear slopes and offsets. The RMS values are 0.6 and 1.3 cm for the L1 and L2
frequency bands, respectively, which confirms the availability of calculating UPDs with linear slopes
and offsets. The bottom sub-figure shows that approximately 97% of the ambiguities fluctuate near
integers within 0.1 cycles after removing the UPDs derived from the linear equations, which also
implies that most of ambiguities can be resolved.
After the integer nature of the ambiguities are retrieved, all of the frequently used AR approaches,
such as LAMBDA (Teunissen 1993, 1995) and the decision function (Dong and Bock 1989), can be
utilized to achieve ambiguity-fixed solutions. This paper adopts the LAMBDA method to fix the
ambiguities, with the threshold for ratio values setting to 3 (Leick 2003). All of the ambiguities will
remain as floats if the ratio values do not reach the threshold.

4.3 DCB & ISB

Fig. 5 Stability of DCBs (top) and ISBs (bottom) for GPS, Galileo and BDS. The average values for
each parameter are shown as equations

Apart from the baseline results, the DCBs between dual frequencies for each system and the ISBs
between two GNSS systems can also be obtained in this SD algorithm. To analyze their stability, 30
daily solutions were acquired from Oct. 1 to Oct. 30 in 2015 (DOY 274-303). The code bias at the GPS
L1 frequency band was selected as the datum so that all of the ISBs are relative to GPS. Fig. 5 shows
the stability of DCB and ISB over the 30-day time span. It shows that all of the DCB and ISB

14
parameters are very stable in the experimental period. The STDs of DCBs from GPS, Galileo and BDS
are 0.05, 0.05 and 0.02 m, respectively. The ISBs for Galileo and BDS exhibit similar performance to
DCBs, with STDs of 0.03 and 0.04 m, respectively. Based on the results above, it can be concluded that
the DCBs and ISBs of CDMA GNSS systems from different epochs can be estimated as one common
parameter due to their significant stability over long time spans.

Fig. 6 Code IFBs for satellites with different GLONASS frequency numbers on October 1, 2015. The
equations shown in the figure are their linear fitting results

The code IFBs are also estimated for GLONASS to craft a harmonious modeling of code
observations. As an example, the code IFBs for each GLONASS satellite on October 1, 2015 are shown
in Figure 6. Although the code bias from the GPS L1 code is in fact included in the estimated code
IFBs, it is assumed to be a constant and no detrimental influence is introduced in this analysis.
It can be seen that the code IFBs on both frequency bands are also approximately linearly
correlated to the frequency numbers. The RMS values of the linear fitting results are 0.33 and 0.17 m
for the L1 and L2 frequency bands, respectively, and the linear slopes are -0.22 and -0.13 m/FN for the
L1 and L2 frequency bands, respectively. Therefore, it is possible to express the code IFBs for
GLONASS receivers with a linear slope and an offset with respect to the frequency numbers.

15
Fig. 7 Stability of code IFBs from PRN01 and PRN05 (with the same frequency number of +1) of
GLONASS at L1 and L2 frequency bands from Oct. 1 to Oct. 30 in 2015 (DOY 274-303)

The stability of code IFBs for GLONASS is further analyzed with a similar approach. As an
example, PRN01 and PRN05 are used in the analysis due to their same frequency number of +1 in the
experimental span. Figure 7 shows the code IFBs for these two satellites over a 30-day time span. The
figure reveals that all of the code IFBs show good stability in the experimental period, as the IFBs are
almost the same for both satellites. The average values for IFBs at the L1 and L2 frequency bands are
-6.52 and -5.23 m, with STDs of 0.07 and 0.06 m, respectively. However, there is a small jump on
DOY 288. After inspecting the RINEX observation file from KIRU, it seems that the tracking for all of
the visible satellites was interrupted for approximately 1 minute on that day. We suspect that some
event occurred at that time to cause this small jump.
All of these results show the potential for calibrating all of the above DCB, ISB and IFB
parameters in advance and then providing them as products for users to improve GNSS services.

4.4 Baseline results

In this section, the precision and accuracy of baselines calculated with the SD method are assessed by
comparing the repeatability of the baseline results. The benefits of multi-GNSS are also investigated.
Considering that only several BDS and Galileo satellites can be observed in the selected stations, three
strategies are designed to process the data mentioned above for comparison: in the first, only GPS
observations are used; in the second, only GPS and GLONASS data are adopted; in the third, all of the
data from four GNSS systems are processed together. In addition, because daily GPS-only solutions
can also achieve very high accuracy and can inhibit the confirmation of the improvements caused by
multi-GNSS, 4-hour data segments for each day are selected for these experiments.
Table 2 Repeatability of baseline components for float and fixed solutions and the relevant

16
improvements as a percentage

GPS-only GPS and GLONASS Four systems


Constellation
N E U N E U N E U

Ambiguity-float (cm) 0.32 0.48 0.85 0.28 0.26 0.57 0.26 0.26 0.58
Ambiguity-fixed (cm) 0.25 0.16 0.70 0.22 0.12 0.52 0.21 0.17 0.50
Improvement (%) 22 67 18 21 54 9 19 35 14

Table 2 shows the repeatability of baselines from the three experiments before and after the
ambiguity-fixing operation. It is evident that the repeatability for all components is improved
significantly after the ambiguity resolution, with the improvement in the east direction reaching 67%
for the GPS-only results. These results also validate the ambiguity resolution method proposed in this
paper. In addition, it can also be seen that when the data for GPS and GLONASS or all four systems
are used, the results are superior to those of the GPS-only solution, particularly in the vertical
component, which suggests that the quality of baselines can be improved by using multi-GNSS data
together. We also find that the repeatability becomes slightly inferior when the data of all four systems
is used compared with those only utilizing GPS and GLONASS; this is probably caused by the
unavailable phase center offsets of BDS and Galileo (phase center offsets of GPS are used for BDS and
Galileo in the experiments).
To further investigate the accuracy of the new SD method, the daily baseline solutions are derived
with GAMIT software using only GPS data, and the results are taken as “ground truth.”

Fig. 8 Biases of the results from GPS-only solution at 4-hour intervals against the “ground truth” in the
north, east, up and length components from Oct. 1 to Oct. 30, 2015

Fig. 8 shows the biases of the baseline components from the GPS-only solution using SD method
against the “ground truth.” It can be seen most of the biases are within ±5 mm for the horizontal

17
components. The average biases in the north, east, up and length components are -0.12, -0.33, 1.04 and
0.31 cm, respectively. Although the magnitude of average biases is small, there appears to be
systematic biases in the east and up components. This is probably due to unmodeled errors, such as
multipaths, that are averaged in the daily solutions but degrade the accuracy of the 4-hour solutions.
To confirm the benefits of multi-GNSS, three experiments using 4-hour data from different GNSS
systems mentioned above are carried out with different cutoff angles. Table 3 shows the repeatability of
the results for cases with cutoff angles of 10, 20, 30 and 40 degrees; the numbers in the brackets are the
improvements compared with the results of the GPS-only solution. It can be seen that with the increase
of cutoff angle, more significant improvements can be made by using multi-GNSS.
Table 3 Repeatability of baseline components in centimeters from different constellations and different
cutoff elevation angles at 4-hour intervals. Brackets show the improvements of multi-GNSS relative to
GPS-only results as in percentage

Cutoff GPS-only GPS and GLONASS Four systems


angle
N E U N E U N E U
(degree)

10 0.25 0.16 0.70 0.22 (12) 0.12 (25) 0.52 (26) 0.21 (16) 0.17 (-6) 0.50 (29)
20 0.30 0.18 1.27 0.30 (0) 0.17 (6) 1.05 (17) 0.22 (27) 0.18 (0) 1.21 (5)
30 1.14 0.28 5.29 0.27 (76) 0.14 (50) 1.92 (64) 0.26 (77) 0.15 (46) 1.71 (68)
40 3.66 2.96 31.47 1.34 (63) 2.40 (19) 23.68 (25) 1.02 (72) 1.93 (35) 11.12 (65)

We find that the improvements are less than 30% with cutoff angles of 10°and 20°. However,
improvements reach 70% as the cutoff angle increases to 40°. This is because when the cutoff angles
increase, the number of visible GPS satellites decreases, which significantly degrades the GPS-only
results. However, in the multi-GNSS cases, more GNSS satellites are available, even if a high cutoff
angle is set. It is interesting to note that the repeatability of results for the four GNSS systems is
inferior to that of GPS-only in the east component when the cutoff angle is set to 10°, whereas it
outperforms the results of GPS-only when the cutoff angle increases to 30°. As previously mentioned,
this is because the phase center offsets for BDS and Galileo are still unavailable. The consistency
between GPS, BDS and Galileo needs to be further improved, which leads to a slight degradation in the
four GNSS systems case when the cutoff angle is set to 10°. However, when the cutoff angle increases,
the observed GPS satellites are insufficient for obtaining a precise result; thus, the advantage of
multi-GNSS becomes apparent.
To further investigate the benefits of multi-GNSS, 1-hour solutions instead of 4-hour solutions are
analyzed with different cutoff angles. The results are shown in Table 4. Shorter processing sessions
suggest fewer observed satellites. As expected, the improvements from using multi-GNSS are quite

18
apparent. All of these results confirm the benefits of multi-GNSS.
Table 4 Repeatability of baseline components in centimeters from different constellations and different
cutoff angles at 1-hour intervals. Brackets show the improvements of multi-GNSS relative to the
GPS-only results as a percentage

Cutoff GPS-only GPS and GLONASS Four systems


angle
N E U N E U N E U
(degree)

10 1.17 1.12 2.67 1.19 (-2) 0.18 (84) 1.12 (58) 0.38 (68) 0.30 (73) 1.39 (48)
20 2.82 1.86 11.13 3.14 (-11) 0.90 (52) 2.72 (76) 0.52 (82) 0.30 (84) 3.23 (71)
30 4.30 6.72 25.38 3.67 (15) 4.21 (37) 22.18 (13) 1.78 (59) 2.88 (57) 15.99 (37)

The advantage of the SD method over the traditional DD method is also assessed. Two
experiments are carried out where only GPS and BDS satellites are used because only these two
systems are supported by Dam Deformation Monitoring Software (DDMS) (Xiao et al. 2016). DDMS
is a software package for short baseline processing using the DD method where DD observations are
only formed in a single GNSS system. In the first experiment (Expt. 1), all of the visible GPS and BDS
satellites are included, whereas only 5 GPS and 4 BDS satellites are chosen in the second experiment
(Expt. 2) to simulate areas with poor observation conditions. The repeatability for both experiments is
calculated and shown in Table 5.
Table 5 Repeatability of baseline components for DD and SD methods

Expt. 1 Expt. 2
Algorithm
N E U N E U

DD (cm) 0.33 0.20 0.77 0.86 0.27 2.49


SD (cm) 0.21 0.20 0.57 0.52 0.32 1.73

It can be seen that the proposed SD method can improve the repeatability of baselines compared
with the traditional DD method without DD observations formed between different systems.
Improvements of 36% and 26% are found for the north and up directions by the SD method. In
Experiment 2, improvements of almost 40% and 30% are achieved in the north and up directions,
respectively. These results suggest that the SD method has the potential to improve the availability of
GNSS in constrained areas. In addition, a slight degradation is shown in the east direction in
Experiment 2; this is probably caused by the inconsistent modeling between GPS and BDS, which may
need to be further confirmed.

19
5. Conclusions

There is increased interest in processing multi-GNSS data at the observation level, where more reliable
and precise products are expected with additional satellites. However, the traditional DD method for
multi-GNSS becomes more complicated as a result of ISBs and different frequencies between systems.
Therefore, in the traditional DD method, DD observations are usually only formed in each GNSS
system; otherwise, the inter-system ambiguities are hard to resolve. This paper presents a new method
for multi-GNSS data processing where SD instead of DD observations between two receivers are
formed and receiver-dependent uncalibrated phase delays are calibrated for ambiguity resolution. This
method is theoretically equivalent to the optimal DD method, where not only DD observations are
formed between all available satellites but inter-system ambiguities are also resolved.
In the new SD method, UPDs are critical for recovering the integer nature of the SD ambiguities.
Therefore, an average approach and a linear model are proposed to calibrate the UPDs for CDMA and
FDMA GNSS systems. Experiments show that after removing the UPDs derived by the methods we
proposed, the fractional parts of residuals for almost all ambiguities are less than 0.1 cycles, which
verifies the validity of our UPD calibration methods. Meanwhile, the stability of DCBs, ISBs and IFBs
over the selected two receivers is also assessed over a period of one month. The STD values for DCBs
and ISBs are less than 0.05 m for the GPS, Galileo and BDS systems. A similar linear trend feature is
also found for the code IFBs of GLONASS. Overall, these results show the potential for calibrating
DCB, ISB and IFB in advance and providing them to users as products.
Experiments also show that the mean differences between daily solutions derived by GAMIT and
the 4-hour solutions derived by the proposed SD method are approximately -0.12, -0.33, 1.04 and 0.31
cm in the north, east, up and length components, respectively. The benefits brought by multi-GNSS are
also investigated. Experiments show that more reliable results can be derived when multi-GNSS
satellites are used, even if the cutoff angle is 40 degrees.
It is also shown that the baseline repeatability of the new SD method outperforms that of the DD
method, with DD observations only formed in each GNSS system by 36% and 26% in the north and up
directions, respectively. To further confirm these improvements, we simulate an environment with poor
observation conditions where only 5 GPS and 4 BDS satellites are available. The results of the SD
method show improvements of 40% and 30% in the north and up directions, respectively.
More reliable and precise baseline results can be derived with the SD baseline processing method
when multi-GNSS data are utilized. In fact, this method is also available in triple-frequency cases and
long baseline data processing. However, for long baselines, the ambiguity resolution for the
ionosphere-free combination of GLONASS remains an open question and continued study is required.

20
Acknowledgements

We are very grateful to IGS MGEX for providing multi-GNSS data. This work is supported by the
National Natural Science Foundation of China (NO. 41525014 and 41704030), and the Program for
Changjiang Scholars of the Ministry of Education of China, together with the Surveying and Mapping
Basic Research Program of National Administration of Surveying, Mapping and Geoinformation (No.
15-02-01) and China Postdoctoral Science Foundation (NO. 2016M590714).

References

Al-Shaery A, Zhang S, Rizos C (2013) An enhanced calibration method of GLONASS inter-channel


bias for GNSS RTK. GPS Solut 17(2):165-173
Blewitt G (1989) Carrier phase ambiguity resolution for the Global Positioning System applied to
geodetic baselines up to 2000 km. J Geophys Res 94(B8):10187-10203
Bock Y, Gourevitch SA, Councelman III CC, King RW, Abbot RI (1986) Interferometric analysis of
GPS phase observations. Manuscripta Geodaetica 11(4):282-288
Cai C, Gao Y (2007) Precise point positioning using combined GPS and GLONASS observations. J
Glob Position Syst 6(1):13-22
Cai C, Gao Y (2013) Modeling and assessment of combined GPS/GLONASS precise point positioning.
GPS Solut 17(2):223-236
Chen H, Jiang W, Ge M, Wickert J, Schuh H (2014a) An enhanced strategy for GNSS data processing
of massive networks. J Geod 88(9):857-867
Chen H, Jiang W, Ge M, Wickert J, Schuh H (2014b) Efficient high-rate satellite clock estimation for
PPP ambiguity resolution using carrier-ranges. Sensors 14(12):22300-22312
Choi B, Park J, Min Roh K, Lee S (2013) Comparison of GPS receiver DCB estimation methods using
a GPS network. Earth Planet Sp 65(7):707-711
Collins P (2008) Isolating and estimating undifferenced GPS integer ambiguities. In: Proceedings of
National Technical Meeting, San Diego, USA, pp 720-732
Dai L (2000) Dual-frequency GPS/GLONASS real-time ambiguity resolution for medium-range
kinematic positioning. In: Proceedings of the ION GNSS 2000. The institute of Navigation, UT,
Salt Lake City, pp 1071-1080
Dong D, Bock Y (1989) Global positioning system network analysis with phase ambiguity resolution
applied to crustal deformation studies in California. J Geophys Res 94(B4):3949-3966
Ge M, Gendt G, Rothacher M, Shi C, Liu J (2008) Resolution of GPS carrier-phase ambiguities in
precise point positioning (PPP) with daily observations. J Geod 82(7):389-399
Ge M, Zhang H, Jia X, Song S, Wickert J (2012) What is achievable with the current COMPASS

21
constellation? In: Proceedings of the ION GNSS 2012. The institute of Navigation, TN, Nashville,
pp 331-339
He H, Li J, Yang Y, Xu J, Guo H, Wang A (2014) Performance assessment of single- and
dual-frequency BeiDou/GPS single-epoch kinematic positioning. GPS Solut 18(3):393-403
Leick A (2003) GPS satellite surveying, 3rd edn. Wiley, New York
Li M, Qu L, Zhao Q, Guo J, Su X, Li X (2014) Precise point positioning with the BeiDou navigation
satellite system. Sensors 14(1):927-943
Li X, Zhang X, Ge M (2011) Regional reference network augmented precise point positioning for
instantaneous ambiguity resolution. J Geod 85(3):151-158
Li X, Ge M, Dai X, Ren X, Fritsche M, Wickert J, Schuh H (2015a) Accuracy and reliability of
multi-GNSS real-time precise positioning: GPS, GLONASS, BeiDou and Galileo. J Geod
89(6):607-635
Li X, Zhang X, Ren X, Fritsche M, Wickert J, Schuh H (2015b) Precise positioning with current
multi-constellation Global Navigation Satellite Systems: GPS, GLONASS, Galileo and BeiDou.
Sci Rep 5:8328
Lou Y, Liu Y, Shi C, Wang B, Yao X, Zheng F (2014) Precise orbit determination of BeiDou
constellation: method comparison. GPS Solut. 2016, 20(2):259-268
Lou Y, Zheng F, Gu S, Wang C, Guo H, Feng Y (2015) Multi-GNSS precise point positioning with
raw single-frequency and dual-frequency measurement models. GPS Solut. 2016, 20(4):849-862
Montenbruck O, Hauschild A, Steigenberger P, Hugentobler U, Teunissen P, Nakamura S (2013)
Initial assessment of the COMPASS/BeiDou-2 regional navigation satellite system. GPS Solut
17(2):211-222
Montenbruck O, Steigenberger P, Hauschild A (2014a) Differential code bias estimation using
multi-GNSS observations and global ionosphere maps. Navigation 61(3):191-201
Montenbruck O, Steigenberger P, Khachikyan R, Weber G, Langley R, Mervart L, Hugentobler U
(2014b) IGS-MGEX: preparing the ground for multi-constellation GNSS science. Inside GNSS
9(1):42-49
Pratt M, Burke B, Misra P (1998) Single-epoch integer ambiguity resolution with GPS-GLONASS
L1-L2 Data. In: Proceedings of the ION GNSS 1998. The institute of Navigation, TN, Nashville,
pp 389-398
Revnivykh S (2012) GLONASS status and modernization. In: Proceedings of the ION GNSS 2012.
The institute of Navigation, TN, Nashville, pp 538-552
Rothacher M, Beutler G, Gurtner W, Brockmann E, Mervart L (1993) The Bernese GPS Software
Version 3.4. Astronomical Institute, University of Bern
Teunissen PJG (1993) Least-squares estimation of the integer GPS ambiguities. Invited lecture, section

22
IV theory and methodology. IAG general meeting, Beijing, China
Teunissen PJG (1995) The least-squares ambiguity decorrelation adjustment: a method for fast GPS
integer ambiguity estimation. J Geod 70(1):65-82
Tian Y, Ge M, Neitzel F, Zhu J. (2017). Particle filter-based estimation of inter-system phase bias for
real-time integer ambiguity resolution. GPS Solut. 2017, 21(3):949-961
Tu R, Ge M, Zhang H, Huang G (2013) The realization and convergence analysis of combined PPP
based on raw observation. Adv Space Res 52(1):211-221
Urlichich Y, Subbotin V, Stupak G, Dvorkin V, Povalyaev A, Karutin S (2011) GLONASS
modernization. In: Proceedings of the ION GNSS 2011. The Institute of Navigation, Portland, OR,
pp 3125-3128
Wanninger L, Wallstab-Freitag S (2007) Combined processing of GPS, GLONASS, and SBAS code
phase and carrier phase measurements. In: Proceedings of the ION GNSS 2007. The institute of
Navigation, TX, Fort Worth, pp 866-875
Xiao Y, Jiang W, Chen H, Yuan P, Xi R (2016) Research and realization of deformation monitoring
algorithm with millimeter level precision based on BeiDou Navigation Satellite System (in
Chinese). Acta Geod Cartogr Sci 45(1):16-21
Xu T, Yu S, Li J (2014) Earth rotation parameters estimation using BDS and GPS data based on
MGEX network. In: Proceedings of lecture notes in electrical engineering, China Satellite
Navigation Conference (CSNC), vol 305, pp 289-299
Yamada H, Takasu T, Kubo N, Yasuda A (2010) Evaluation and calibration of receiver inter-channel
biases for RTK-GPS/GLONASS. In: Proceedings of the ION GNSS 2010. The Institute of
Navigation, Portland, OR, pp 1580-1587
Yang Y, Li J, Xu J, Tang J, Guo H, He H (2011) Contribution of the Compass satellite navigation
system to global PNT users. Chinese Sci Bull 56(26):2813-2819

23

You might also like