You are on page 1of 20

J Geod (2016) 90:209–228

DOI 10.1007/s00190-015-0867-4

ORIGINAL ARTICLE

Determination of differential code biases with multi-GNSS


observations
Ningbo Wang1,4 · Yunbin Yuan1 · Zishen Li2 · Oliver Montenbruck3 ·
Bingfeng Tan1,4

Received: 26 May 2015 / Accepted: 19 October 2015 / Published online: 11 November 2015
© Springer-Verlag Berlin Heidelberg 2015

Abstract In order to better understand the differential code GPS and GLONASS intra-frequency biases obtained in this
biases (DCBs) of global navigation satellite system, the work show the same precision levels as those estimated by
IGGDCB method is extended to estimate the intra- and inter- DLR (about 0.1 and 0.2–0.4 ns for the two constellations,
frequency biases of the global positioning system (GPS), respectively, with respect to the products of CODE). The
GLONASS, BeiDou navigation satellite system (BDS), and precision levels of IGGDCB-based inter-frequency biases
Galileo based on observations collected by the multi-GNSS estimated over the 24-month period are about 0.29 ns for
experiment (MGEX) of the international GNSS service GPS, 0.56 ns for GLONASS, 0.36 ns for BDS, and 0.24
(IGS). In the approach of IGGDCB, the local ionospheric ns for Galileo, respectively. Here, the accuracies of GPS
total electronic content is modeled with generalized tri- and GLONASS biases are assessed relative to the products
angular series (GTS) function rather than using a global of CODE, while those of BDS and Galileo are compared
ionosphere model or a priori ionospheric information. The with the estimates of DLR. In addition, the monthly stabil-
DCB estimated by the IGGDCB method is compared with ity indices of IGGDCB-based DCBs are 0.11 (GPS), 0.18
the DCB products from the Center for Orbit Determination (GLONASS), 0.17 (BDS), and 0.14 (Galileo) ns for the indi-
in Europe (CODE) and German Aerospace Center (DLR), as vidual constellation.
well as the broadcast timing group delay (TGD) parameters
over a 2-year span (2013 and 2014). The results indicate that Keywords Multi-GNSS experiment (MGEX) · Differ-
ential code bias (DCB) · Timing group delay (TGD) ·
Intra-frequency bias · Inter-frequency bias · IGGDCB · Total
B Ningbo Wang
electronic content (TEC)
wnbigg@asch.whigg.ac.cn
B Yunbin Yuan
yybgps@whigg.ac.cn
1 Background and motivation
Zishen Li
lizishen@aoe.ac.cn
Pseudorange observations of global navigation satellite sys-
Oliver Montenbruck
Oliver.Montenbruck@dlr.de
tem (GNSS) are well known to be affected by the differential
code biases (DCBs), which need to be precisely calibrated
Bingfeng Tan
for pseudorange-based positioning, timing, and ionospheric
bingfengtan@asch.whigg.ac.cn
modeling (Sardón and Zarraoa 1997; Hernández-Pajares
1 State Key Laboratory of Geodesy and Earth’s Dynamics, et al. 2011; Zhang et al. 2012; Montenbruck and Hauschild
Institute of Geodesy and Geophysics, Wuhan, China 2013). Generally, DCBs can be classified into two cate-
2 Academy of Opto-Electronics, Chinese Academy of Sciences, gories: intra-frequency bias and inter-frequency bias. The
Beijing, China first one is the bias between two code observations at the same
3 German Space Operations Center, Deutsches Zentrum für frequency, and the second one is the bias between code obser-
Luft- und Raumfahrt, 82230 Weßling, Germany vations at two different frequencies (Leandro et al. 2007).
4 University of Chinese Academy of Sciences, Beijing, China The satellite DCB-related timing group delay (TGD), broad-

123
210 N. Wang et al.

cast group delay (BGD), and inter-signal correction (ISC) cast TGD values remained in the in-orbit satellites (Sardón
are also the essential parameters in navigation messages of and Zarraoa 1997). Since 1998, more accurate DCB esti-
global positioning system (GPS), BeiDou navigation satel- mates have been routinely provided by the IGS for the
lite system (BDS), and Galileo to compensate the differential post-processing applications of GPS and GLONASS (Fel-
hardware delay for real-time positioning applications (OS- tens 2003a, b; Dow 2009). It is currently common practice
SIS-ICD 2010; CSNO 2013; IS-GPS-200 2012). for network-based DCB determination to estimate one DCB
For intra-frequency bias, only sparse analyses of GPS per station, which is considered identical for all satellites,
P1-C1 and P2-C2 biases are currently available. The satel- and one DCB per satellite, which is considered the same
lite clock offsets in both broadcast and precise ephemeris for all receivers (Hauschild and Montenbruck 2014). The
products are commonly computed by ionosphere-free com- methods for determining satellite DCB based on data from
bination of phase (L1/L2) and code (P1/P2) observations. ground monitoring stations can be mainly separated into two
GPS-related applications using C1/P2 or P1/C2 observa- categories (Li et al. 2014): (1) DCB is determined simultane-
tions differing from the conventional reference signals should ously with global or local ionospheric total electronic content
correct the biases between P1-C1 and P2-C2 observations (TEC) modeling, and (2) DCB is estimated from code dif-
(Montenbruck and Hauschild 2013). GPS P1-C1 products, ferences after accounting for the ionospheric path delays.
obtained as part of satellite clock estimation procedure, have With regard to the first approach, satellite and receiver
been provided by the Center for Orbit Determination in DCBs are included as additional unknowns in mathemati-
Europe (CODE) since 2000. It has been demonstrated that cal models (e.g., polynomials and spherical harmonics) for
the day-to-day repeatability of daily CODE P1-C1 solutions ionospheric TEC determination (Schaer 1999; Arikan et al.
is in the order of 0.05 ns (Schaer 2000). The P1-C1 products 2008; Krankowski et al. 2009; Hernández-Pajares et al.
of CODE are now widely used by the international GNSS 2011). GPS and GLONASS DCBs routinely estimated by
service (IGS) analysis centers and users of IGS clock prod- individual Ionospheric Associate Analysis Center (IAAC,
ucts. The intra-frequency bias can also be directly calculated including CODE, JPL, UPC, and ESA) of the IGS are
from GNSS raw observations, allowing CODE to provide P1- currently considered as the most reliable DCB products
C1 and P2-C2 satellite and receiver bias corrections for GPS (Hernández-Pajares et al. 2009; Dach et al. 2009; Li et al.
and GLONASS measurements (Schaer 2012a). Gao (2001) 2015b). The DCB estimates from different IAACs are pro-
investigated the constant and non-constant components in the duced as part of the global ionosphere maps (GIM). Li et al.
biases between GPS P1 and C1 observations. More recently, (2012) proposed a new approach for GNSS satellite DCB
P1-C1 and P2-C2 code bias estimates based on precise point determination, which was developed at the Institute of Geo-
positioning (PPP) have been analyzed in the studies of Lean- desy and Geophysics (IGG) in Wuhan and is designated
dro et al. (2007) and Li (2015a). The first study achieves as IGGDCB. In contrast to global ionospheric modeling,
an agreement of 0.15 ns between the estimated satellite P1- the ionospheric vertical TEC of each individual station in
C1 biases and monthly CODE solutions, during Jan. 1 to IGGDCB is independently modeled by generalized triangu-
Jan. 10, 2007. The second study reports that the difference lar series (GTS) function. Based on a few tracking stations,
between P1-C1 estimates and CODE products is smaller than it is demonstrated that the accuracy of IGGDCB-based GPS
0.4 ns during a period of 30 days, from mid-July to mid- DCB estimates performs at the level of 0.13 and 0.10 ns for
August 2012. Therefore, the intra-frequency bias estimation the years 2002 and 2009, respectively (Li et al. 2012).
methods can be summarized as follows: (1) estimating the Concerning the second approach, a priori information of
intra-frequency biases in satellite clock analysis procedure; ionosphere (e.g., global or regional ionospheric TEC maps,
(2) directly extracting the biases from two different code experiential ionospheric model, etc.) is generally needed to
observations simultaneously observed at the same frequency; correct the ionospheric TEC along the signal propagation
and (3) determining the biases based on PPP technique. path from GNSS satellite to the receiver. The accuracy of
TGD and DCB parameters are commonly used to describe DCB estimates depends on the degree to which the variation
and compensate satellite/receiver inter-frequency biases. of ionospheric TEC can be captured by empirical ionospheric
GPS broadcast TGDs are initially defined to account for models. The more precise is the priori ionospheric infor-
the effect of space vehicle (SV) group delay differential mation, the more accurate and reliable are the resulting
between the onboard L1P and L2P signals (Wilson et al. DCB estimates. GIMs are commonly employed to mitigate
1999). In addition to TGD parameter, four ISC parameters ionospheric path delays in the studies for DCB determination
are introduced in the new civil navigation (CNAV) message (Keshin 2011; Montenbruck et al. 2014).
to provide DCB corrections for L1C/A, L2C, and L5 signals With the emerging BDS and Galileo, as well as the
w.r.t. L1P(Y) signal (Steigenberger et al. 2015). Although modernization of GPS and GLONASS, multi-GNSS DCB
each individual satellite TGD had been calibrated before estimation has become one of the challenging issues for
launch, a discrepancy between the monitored and broad- the IGS community (Willis et al. 1999; Dach et al. 2009;

123
Determination of differential code biases with multi-GNSS observations 211

Dow 2009). The IGS Bias and Calibration Working Group 2 Introduction to multi-GNSS biases
(BCWG) was established for GNSS bias retrieval and moni-
toring (Schaer 2012b). However, only limited information is With the development of the modernized GPS and the emerg-
presently available for satellite and receiver DCBs related to ing BDS and Galileo, the IGS has initiated the MGEX project,
the modernized GPS, BDS, as well as Galileo signals, par- to familiarize IGS participants and users with the new satel-
ticularly for the diverse observation types supported by the lite navigation systems (Montenbruck et al. 2013; Rizos et al.
Receiver Independent Exchange Format Version 3 (RINEX 2013). The MGEX network comprises almost 120 stations
3, IGS 2013). The IGS multi-GNSS experiment (MGEX) by the end of 2014 (see Fig. 1), and each station is able
multi-constellation and multi-frequency observations offer a to support at least one of the new navigation systems (e.g.,
basis for an independent determination of ionosphere and BDS, Galileo, and QZSS) in addition to the legacy GPS
DCB parameters (Montenbruck et al. 2013; Rizos et al. and GLONASS systems. The current set of MGEX stations
2013). Currently, satellite and receiver DCB estimates for shows an almost global coverage. A summary of the con-
all the available GNSS constellations are determined by the tributed receivers and the corresponding observations within
German Aerospace Center (DLR) based on MGEX observa- MGEX is provided in Table 1. Observation types for GPS,
tions and GIMs. A day-to-day repeatability of 0.05–0.3 ns GLONASS (GLO), as well as BDS and Galileo (GAL) are
of the estimated satellite bias is reported (Montenbruck et al. based on the RINEX 3 format (IGS 2013). It can be found that
2014), and the accuracy of BDS satellite DCB estimates of five basic receiver types (i.e., Trimble, Javad, Leica, Septen-
DLR has also been assessed in the study of Guo (2015). trio, and NovAtel) are currently employed within the MGEX
In the present study, the IGGDCB method is extended to network. GPS and GLONASS are supported by virtually all
estimate the intra- and inter-frequency biases of currently receivers, and Galileo satellites are also monitored by the
available GNSSs, including GPS and GLONASS, as well majority of MGEX sites. BDS B1, B2, and optionally, B3
as BDS and Galileo based on MGEX observations. A soft- signals can be tracked by three types of receivers, including
ware package has been developed for daily DCB estimation Trimble NETR9, Leica GR25, and Septentrio PolaRx4/4TR.
in an automatic mode, and a comprehensive assessment and However, most of these receivers are located in Europe and
a comparison of the MGEX-based DCB estimates are pre- Americas (see Fig. 1). Therefore, the regional BDS constella-
sented. This article is organized as follows: Sect. 2 presents tion [i.e., satellites in geostationary orbit (GEO) and inclined
an overview of the MGEX tracking network and an introduc- geosynchronous orbit (IGSO)] is not well covered by current
tion to multi-GNSS biases. Section 3 describes the IGGDCB network.
algorithms, including intra-frequency bias estimation and The DCB products of P1-P2, P1-C1, and P2-C2 released
extraction of ionospheric observable from GNSS data, as well by CODE are produced based on the legacy GPS and
as the modeling of local ionospheric TEC and determination GLONASS RINEX 2 observations. Within the MGEX net-
of satellite- and receiver-specific DCBs. Section 4 performs work, more signals and observations in RINEX 3 format are
an assessment of the accuracy and stability of GNSS satel- monitored with the modernized GPS, as well as the new BDS
lite DCB estimates based on IGGDCB. Section 5, finally, and Galileo systems, meaning that the concept of traditional
presents conclusions and future work. DCB needs to be extended to multi-GNSS biases. It should be

Fig. 1 MGEX station distribution and receiver type (as of Dec. 2014)

123
212 N. Wang et al.

mentioned that no common standard choice of multi-GNSS This means that a non-redundant or minimal set of DCB
biases has yet been recommended by the IGS BCWG. The- would be sufficient. However, it is difficult to derive such a set
oretically, if there are N types of code observations, only of DCB in practice due to two reasons: (1) up to 15 different
(N−1) types of intra- and inter-frequency biases are needed. signals are defined in RINEX 3 standard for each individual
navigation constellation, which may form a large number
of possible signal pairs and the associated differential code
Table 1 Receiver and observation types within MGEX for differential biases; (2) current monitoring receivers may not support the
code bias estimation (as of Dec. 2014) tracking of the corresponding signals which have been used
Receiver type GNSS observation Sites to form the pre-defined DCB. For the available MGEX obser-
vations, we divide the code observations by different types
Trimble NETR9 GPS: 1C, 2W, 2X, 5X 44 of receivers and satellite constellations and provide our own
GLO: 1C, 1P, 2C, 2P set of multi-GNSS differential code biases (see Table 2).
GAL: 1X, 5X, 7X, 8X The adopted set of biases in Table 2 does not contain
BDS: 2I, 7I, 6I redundant bias information, and the DCB estimates would
Javad TR_G2T, GPS: 1C, 1W, 2W, 2X, 5X 32 directly meet the demands of most multi-GNSS applications.
Javad TRE_G3TH GLO: 1C, 1P, 2C, 2P Some bias types are missing in comparison with that of DLR;
GAL: 1X, 5X however, the DCB of interest may be reconstructed from the
Leica GR10/ GR25 GPS: 1C, 2S, 2W, 5Q 23 linearly independent set of DCB shown in Table 2. It should
GRX1200+GNSS GLO: 1C, 2C, 2P be mentioned, though, that the chaining of multiple DCB
GAL: 1C, 5Q, 7Q, 8Q may differ from direct DCB estimates in practice, due to the
BDS: 2I (1I), 7I different sets of stations/receivers used in the determination
Septentrio GPS: 1C, 1W, 2W, 2L, 5Q 15 of DCB and the fact that satellite and receiver biases are not
PolaRx4/4TR, GLO: 1C, 2C, 2P fully decoupled (Hauschild and Montenbruck 2014).
AsteRx3 GAL: 1C, 5Q, 7Q, 8Q
BDS: 2I (1I), 7I
NovAtel OEM6 GPS: 1C, 2W, 5Q 1 3 Multi-GNSS bias estimation algorithms
GLO: 1C, 2P
GAL: 1C, 5Q The IGGDCB proposed by Li et al. (2012) was improved
to estimate the multi-GNSS biases. Generally, there are four

Table 2 Selected set of intra-


GNSS system Code bias type Receiver type
and inter-frequency biases for
current MGEX code Trimble NETR9 Javad Leica Septentrio
observations
√ √
GPS Intra-frequency bias C1C-C1W
√ √
C2W-C2X

C2W-C2S

C2W-C2L
√ √ √ √
Inter-frequency bias C1W-C2W
√ √
C1C-C5X
√ √
C1C-C5Q
√ √
GLO Intra-frequency bias C1C-C1P
√ √ √ √
C2C-C2P
√ √
Inter-frequency bias C1P-C2P
√ √
GAL Inter-frequency bias C1C-C5Q
√ √
C1C-C7Q

C1C-C8Q
√ √
C1X-C5X

C1X-C7X

C1X-C8X
√ √
BDS Inter-frequency bias C2I-C7I

C2I-C6I

123
Determination of differential code biases with multi-GNSS observations 213

Fig. 2 Flowchart of daily DCB


process with multi-GNSS
observations

j j j j
parts in IGGDCB algorithms for multi-GNSS bias estima- Pi = ρi + c · (ti − t j ) + Ti + Ii + c · (bi + b j ) (1)
tion: (1) determination of intra-frequency bias directly from
raw multi-GNSS observations; (2) extraction of ionospheric j
based on the geometric range (ρi ), the satellite and receiver
observables from multi-GNSS observations; (3) establish-
clock offsets (t j and ti ), the speed of light c, tropospheric and
ment of local ionospheric model and satellite-plus-receiver j j
ionospheric range delays (Ti and Ii ), as well as the hardware
(SPR) inter-frequency bias estimation over each individual
delays in the satellite (b j ) and receiver (bi ). It is obvious
station; and (4) calculation of satellite- and receiver-specific
that the pseudorange difference for noise- and multipath-free
inter-frequency bias based on SPR estimates. A flowchart of
observations directly provides the corresponding SPR intra-
the daily DCB estimation method is illustrated in Fig. 2. The
frequency bias.
estimation of intra-frequency biases is discussed in Sect. 3.1,
In practice, the satellite and receiver DCB parameters are
while the remaining steps (2)–(4) related to the estimation of
considered to be constant through the data arc (e.g. 1-day).
inter-frequency biases are jointly discussed in Sect. 3.2. The
Assuming that the observation noise and multipath errors
intra-frequency biases are estimated first, because various
have a zero mean over a single day, the combined SPR intra-
GPS receivers provide only C1C, but no C1W observations.
frequency biases (SPRs1 −s2 ) for each signal pair, station, and
Making use of the C1CC1W bias from step (1), C1WC2W
satellite can be determined from a mean value over the avail-
inter-frequency bias can subsequently be determined using
able data arc:
all stations in the network.

1 
N
SPRs1 −s2 = (Ps1,k − Ps2,k ), (2)
3.1 Intra-frequency bias determination N
k=1

The intra-frequency bias is directly extracted from raw multi- where Ps1,k and Ps2,k are the pseudorange observations simul-
GNSS code observations in this study. Ignoring multipath and taneously obtained by tracking two distinct signals s1 and s2
j
noise, the measurement of code delay (Pi ) between the jth at the same frequency for epoch k, and N is total epoch
satellite and ith receiver can be expressed as number. With a zero-mean reference for the satellites, the

123
214 N. Wang et al.

intra-frequency bias in satellite and receiver can be sepa- where φ and φ0 are the geographic latitude of the ionospheric
rately estimated. Details of this processing can refer to that IPP and station, respectively; h is the local time at the
described in Sect. 3.2. ionospheric IPP; n max and m max are the maximum degree
of the polynomials; kmax is the maximum degree of the finite
3.2 Inter-frequency bias estimation Fourier series; E nm , Ck , and Sk are the model coefficients to
be estimated; Hion is the altitude of ionospheric single-layer
The inter-frequency bias estimation comprises the extrac- shell; and RE is the mean radius of the earth. The degree
tion of ionospheric observables from raw multi-GNSS data, of GTS coefficients can be set differently for different sites
the local ionospheric modeling, and finally, the satellite- and during different levels of solar activities (Yuan and Ou
and receiver-specific DCB determination. The ionospheric 2004; Li et al. 2012). In this study, the maximum degrees
observables can be simply obtained by the geometry-free of the polynomials and the finite Fourier series were set
combination of dual-frequency code observations, but they as 2 (n max = m max = 2) and 4 (kmax = 4), respectively,
are seriously affected by the noise and multipath effects. and the height of the assumed single layer was selected as
In order to improve the accuracy of ionospheric observ- Hion = 400 km.
ables, the approach of “pseudorange-leveled carrier-phase” Assuming that the pseudorange and carrier phase mea-
is used (Mannucci et al. 1998; Schaer 1999). The processed surements are uncorrelated, the variance of the smoothed
ionospheric TEC and DCB measurement can be expressed ionospheric measurement σ 2 can be estimated as (Schaer
P̃4,k
as 1999)

P̃4,k = v · ST ECk + c · SPR + ε 1 2 N −1 2
(3) σ P̃2 = σ + σ L 4,k , (5)
SPR = Bs + Br , 4,k N P4,k N

where σ P24,k and σ L24,k are the variances of TEC measure-


where P̃4,k is the smoothed ionospheric TEC and DCB
ments derived from geometry-free combination of code and
measurement after applying the technique of “pseudorange-
carrier phase observations, respectively, and N is the smooth-
leveled carrier-phase” at epoch k; v = 40.28 × 1016 ×
ing length. σ P4,k and σ L 4,k are set as 3 TECu and 0.1 TECu
( f 1−2 − f 2−2 ), where f 1 and f 2 are the two frequencies of the
in this study, which corresponds to an accuracy of 0.3 m for
carrier phase; STECk is the line-of-sight (LOS) ionospheric
pseudorange and 0.05 cycle for carrier phase measurements,
TEC along the signal propagation path from satellite to
respectively.
receiver; SPR is the satellite-plus-receiver bias for the signals
It can be seen that the accuracy of the smoothed
used, which can be divided into the sum of a satellite-specific
ionospheric measurement is related to the smoothing length.
part (B s ) and a receiver-specific contribution (Br ); c is the
For a given smoothing length N = 60 (corresponding to 30
speed of light; and ε is the noise of ionospheric measurement.
mins for GNSS measurements with a sampling rate of 30 s),
In order to derive SPR DCBs from the processed
the noise level of the smoothed ionospheric measurement is
ionospheric TEC and DCB measurement, the correspond-
less than 0.4 TECu. As shown in Eq. (6), the weight of the
ing ionospheric delay should be precisely corrected. Within
ionospheric TEC measurement also depends on the elevation
IGGDCB, the ionospheric delay is individually modeled by
angle (e) of the satellite.
the GTS function over each station based on a thin-layer
approximation (Yuan and Ou 2004), in which the slant total
electron content STEC(z, φ, h) along LOS is described as p = 1/(1 + cos2 e). (6)
the product of vertical electron content VTEC(φ, h) at the
ionospheric pierce point (IPP) and a simplified mapping func- Based on the ionospheric observables obtained from a
tion M(z) depending on the satellite’s zenith angle z at the single site, and the assumption that SPR DCB is con-
corresponding station. The VTEC is itself described as the stant throughout a single day (Schaer 1999), the GTS
sum of a two-dimensional polynomial of geographic latitude ionospheric coefficients and the combined SPR DCB of each
and local time and a finite Fourier series of local time: satellite/receiver can be determined by least squares (LS)
⎧ estimation. The covariance matrix of the estimated parame-
⎪ STEC(z, φ, h) = VTEC(φ, h) · M(z) ters can be expressed as

⎪ n
max m


max

⎨ VTEC(φ, h) = { E nm (φ − φ0 )n · h m } ⎡ ⎤
n=0 m=0 D X̂
(4) GTS X̂ GTS
⎪ k 0

⎪ +
max
{Ck cos(k · h) + Sk sin(k · h)} ⎢ σ̂02 ⎥
⎪ D X̂ X̂ = σ̂02 · ⎣ ⎦, (7)

⎪ D X̂
⎩ k=0 0 SPR X̂ SPR
M(z) = [1 − sin2 z · (1 + Hion /R E /)]−1/2 σ̂02

123
Determination of differential code biases with multi-GNSS observations 215

where σ̂02 is a posteriori variance of unit weight of the LS solu- where H is the constraint vector, and the elements of H
tion; D X̂ GTS X̂ GTS and D X̂ SPR X̂ SPR are the covariance matrices corresponding to X̂ sat,dcb are one but zero for others.
corresponding to GTS coefficients and SPR DCB parame- The quality of SPR DCB “observables” obtained at each
ters, respectively. site is different, which is caused by the noise of the smoothed
By repeating above procedures station by station, the com- ionospheric measurement and the varying performance of the
bined SPR DCBs can be determined for all contributing sites GTS ionospheric model at different stations. A weight matrix
and all observed satellites. Assuming that the numbers of P, as shown in Eq. (10), is used to reflect the varying quality
satellite and receiver DCBs for estimation are u 1 and u 2 , of SPR DCBs:
respectively, the SPR DCBs for all satellites and receivers
can be written in the matrix form as P = D −1 , (10)
Ẑ SPR Ẑ SPR


⎪ Z SPR = F · X̂ dcb

⎪  T where D Ẑ SPR Ẑ SPR is the covariance matrix corresponding to

⎨ T T SPR DCBs, obtained from the LS solution of the local GTS
X̂ dcb = X̂ sat,dcb X̂ rec,dcb
(u 1 +u 2 )×1
(8) ionospheric modeling. Combining Eqs. (8)–(10), satellite

⎪ 
1×u 1 1×u 2


⎪ and receiver DCBs for the signals used can be separately

⎩ F = n×uA B ,
n×(u 1 +u 2 ) 1 n×u 2 determined by the constrained LS estimation. The satellite
and receiver DCB estimates are shown as follows:
where the vector Z SPR , with a total of n rows, is the SPR DCB ⎧ −1
⎨ X̂ dcb = (N + H H ) W
T
estimates obtained from individual site, considered as the ⎪
“observable” in LS adjustment; X̂ dcb is the DCB estimates, D X̂ X̂ = σ̂ 2 · (N + H T H )−1 (11)
which consists of the vector of satellite ( X̂ sat,dcb ) and receiver ⎪

N = F T P F, W = F T P Z SPR
( X̂ rec,dcb ) DCBs; F is the design matrix that consists of the
matrix A and B, where just a single element in each row of
where X̂ dcb is the IGGDCB-based satellite and receiver DCB
matrix A and B is one for the corresponding satellite and
estimates; D X̂ X̂ is the covariance matrix of the IGGDCB-
receiver, but zero for others.
based DCB estimates; σ̂ 2 is a posteriori variance of unit
The resulting normal equation matrix F T F exhibits a rank
weight and P is the weight matrix.
deficiency since the bias common to all satellites cannot be
IGGDCB method takes advantage of local ionospheric
distinguished from the corresponding bias common to all
information generated with all available observations of the
receivers. Therefore, Eq. (8) cannot be estimated if no a priori
LOS GNSS satellites over each station, which overcomes
satellite or receiver DCB reference value is given. Generally,
the shortage of modern GNSS tracking stations (site number
the rank deficiency can be eliminated through (Montenbruck
and distribution) for global ionospheric modeling. While a
et al. 2014) (1) fixing a single receiver bias; (2) imposing a
few stations would be suitable for DCB estimation within
zero-mean condition for all satellites; or (3) selecting those
IGGDCB (Li et al. 2012), it is expected that more globally
satellites with relatively stable DCB as the zero reference.
well-distributed receivers would increase the accuracy and
The first approach is suitable for GNSS control segment with
stability of the DCB solution. In this study, all available track-
calibrated reference receivers, and the second approach is
ing stations within the MGEX network were used to estimate
commonly used for DCB separation within the IGS. The
the multi-GNSS DCBs as defined in Table 2. In addition, the
zero-mean satellite reference does not consider the differ-
cutoff elevation was set as 20◦ to reduce the multipath and
ent levels of stability in individual satellite DCB, which may
ionospheric mapping function error.
affect the final DCB estimates. To reduce the impact of the
satellite with poor stability, some activities have been prac-
ticed to select those satellites with relatively stable DCBs as
the satellite reference (IGS 2012; Li et al. 2012). In accor- 4 Results and discussion
dance with the satellite and receiver DCB separation strategy
within the IGS, the zero-mean satellite reference, as shown Satellite and receiver intra- and inter-frequency bias esti-
in Eq. (9), is also applied in this study: mates from January 1, 2013 to December 31, 2014 have been
generated with all MGEX tracking stations and observations
⎧ at an interval of 30 s. The performances of the IGGDCB-

⎪ H · X̂ dcb = 0

⎪  T based satellite DCB estimates (denoted by IGG) are assessed


⎨ T T
X̂ dcb = X̂ sat,dcb X̂ rec,dcb by comparing with those released by CODE, DLR as well
(u 1 +u 2 )×1 1×u 1 1×u 2 (9) as the broadcast TGD or BGD parameters. The LS residuals

⎪  

⎪  
⎪ e within IGGDCB are analyzed in Sect. 4.1. The accuracy of
⎩ 1×(uH+u ) = 1×u 1 1×u 2 , e = 1 . . . 1 ,
⎪ 0
1 2 1×u 1 GPS and GLONASS satellite DCB estimates (including both

123
216 N. Wang et al.

the intra- and inter-frequency biases) are validated in Sect. at both sites, the residuals of WUH2 are more scattered than
4.2, and the performances of BDS and Galileo satellite DCB that of CUT0 [3.45 TECu vs. 1.79 TECu in terms of stan-
estimates are evaluated in Sects. 4.3 and 4.4, respectively. dard deviation (STD)]. This indicates that GTS function can
capture the subtle variation of local ionosphere; however, the
4.1 Validation of IGGDCB algorithms performance of GTS function may be different at different
stations.
Since SPR DCBs were estimated together with local In order to validate the above assumption, a posteriori
ionospheric modeling within IGGDCB, the established local variance of unit weight in Eq. (7), which indicates the consis-
ionospheric model should be capable of representing the tency between the established GTS function and ionospheric
structure and variation of the corresponding local ionosphere. TEC measurements in LS solution, is shown in Fig. 4 for the
In order to demonstrate the performance of the local GTS same day, with the stations aligned by geomagnetic latitudes.
model, the ionospheric residuals in GTS and SPR DCB solu- It can be seen that the differences between the ionospheric
tion of a particular day (July 24, 2014) at WUH2 and CUT0 TEC obtained from local ionospheric model and multi-GNSS
sites are illustrated in Fig. 3. WUH2 is located at low lat- observations are less than 3 TECu with the exception of two
itudes of China, which supports to track GPS, GLONASS, stations at low latitudes of southern hemisphere. In addi-
and Galileo signals; and CUT0 is located at mid-latitudes tion, the accuracy of local GTS ionospheric model is lower
of Australia, which also supports to track BeiDou signals at low latitudes than that obtained at high and mid-latitudes,
in addition to the above three constellations. As shown in which presents a similar trend with the levels of ionospheric
Fig. 3, while the ionospheric residuals show a near zero bias activity at different latitudes. To mitigate the impact of
(−0.11 and −0.08 TECu for WUH2 and CUT0, respectively) ionospheric modeling errors on the final satellite and receiver

Fig. 3 Residual distribution of


ionospheric delay in GTS and
SPR DCB solution at WUH2
and CUT0 sites (July 24, 2014)

Fig. 4 A posteriori sigma of


the IGGDCB-established local
GTS ionospheric model at each
station. The stations are aligned
by geomagnetic latitudes at
horizontal axis

123
Determination of differential code biases with multi-GNSS observations 217

Fig. 5 Residual distribution of


satellite- and receiver-specific
DCB solution for GPS,
GLONASS, BDS and Galileo
(July 24, 2014)

DCB estimates, a weighting matrix as shown in Eq. (10) file (denoted by CODE), which is produced as part of
was used to reflect the varying quality of SPR DCBs in LS CODE’s satellite clock estimation procedure and only con-
estimation. tains the DCB in satellites. The other one is provided in
The residual distribution of satellite- and receiver-specific P1C1yymm_RINEX.DCB or P2C2yymm_RINEX.DCB file
DCB solution for GPS, GLONASS, as well as BDS and (denoted as CODERNX), which is directly generated from
Galileo for the same day (July 24, 2014) is also shown in GPS or GLONASS RINEX pseudorange measurements and
Fig. 5. It can be found that SPR residuals of the four con- contains the DCB in both satellites and receivers. Further-
stellations fit well with normal distribution but seem to be more, the daily intra-frequency bias estimates of DLR are
slightly asymmetric about zero. The STDs of SPR residu- also introduced for comparison.
als for GPS, BDS, and Galileo are 0.62, 0.58, and 0.33 ns, The monthly GPS satellite C1CC1W estimates of CODE,
respectively, while that of GLONASS is 1.27 ns. The SPR DLR, and IGG for December 2014 are presented in Fig.
residuals of GLONASS are more scattered than those of GPS, 6. For unambiguous identification, individual satellites are
BDS, and Galileo, which would affect the final satellite and labeled by their pseudo-random noise (PRN) number fol-
receiver DCB estimates of GLONASS. lowed by their space vehicle number (SVN) given in
brackets. It is shown that GPS C1CC1W biases mainly
vary between -2.0 and 3.0 ns, and C1CC1W estimates
4.2 GPS and GLONASS DCBs
of CODE, DLR, and IGG are very similar. Differences
between the monthly C1CC1W biases calculated by CODE
GPS and GLONASS intra-frequency biases
and other organizations are illustrated in Fig. 7. It can be
seen that the differences vary from −0.6 to 0.5 ns, and
The monthly P1-C1 (corresponding to C1WC1C for GPS G29 (G057) exhibits the largest difference during the period
and C1PC1C for GLONASS) and P2-C2 (C2PC2C for (0.59, 0.53, and 0.52 ns for DLR, IGG, and CODERNX,
GLONASS) DCB products of CODE are used for val- respectively). The root-mean-square (RMS) values of the dif-
idating the performance of the estimated intra-frequency ferences between CODE and DLR/IGG/CODERNX agree
biases in GPS and GLONASS satellites. Two types of well with each other (0.26, 0.23, and 0.23 ns for indi-
intra-frequency bias released by CODE are included in vidual estimate), indicating that C1CC1W biases directly
the analysis. The first one is provided in P1C1yymm.DCB

123
218 N. Wang et al.

Fig. 6 GPS C1CC1W DCB


estimates of CODE, DLR, and
IGG for December 2014

Fig. 7 Differences of
DLR/IGG/CODERNX GPS
C1CC1W DCB estimates with
respect to the monthly P1-C1
product of CODE for December
2014

extracted from raw RINEX observation may perform at the length) is desired for DCB estimates (Montenbruck et al.
same level if a sufficient number of contributing stations is 2014).
used. The differential code biases between GLONASS C1C
The performance of GPS C2WC2L, C2WC2S, and and C1P, as well as C2C and C2P observations are rarely
C2WC2X biases have earlier been assessed in the study of discussed. At present, estimates for both of these biases
Montenbruck et al. (2014) over a 3-month period in 2013, are directly extracted from raw RINEX measurements by
and a standard deviation of 0.3 ns between the differences of CODE and DLR. Figure 9 gives an example of GLONASS
three biases was reported. Up to December 2014, the mod- C1CC1P and C2CC2P biases for December 2014. It can
ernized civil L2 signal (L2C) has been transmitted by a total be seen that C1CC1P estimates vary from −3.0 to 4.0 ns,
of 15 GPS satellites. The differential code biases between and C2CC2P vary from −1.0 to 2.0 ns. The biases esti-
C2W and C2L/C2S/C2X signals are presented in Fig. 8. It mated from CODERNX and that from IGG agree well.
shows that the three biases vary between −0.8 and 1.5 ns, The precision levels of GPS and GLONASS intra-frequency
and the estimates of C2WC2S and C2WC2X are quite simi- bias estimates within MGEX network are summarized in
lar. STD of the differences between C2WC2L and C2WC2S Table 3. The DCB estimates of IGG and DLR are compared
is 0.31 ns, and that of C2WC2L and C2WC2X is 0.25 ns. The to direct (CODERNX) and indirect (CODE) intra-frequency
distinction of L2C tracking mode is impossible within the old bias products of CODE. It can be seen that the accuracies of
RINEX 2 format, which inhibits a more detailed investiga- the bias estimates of DLR and IGG are in similar magnitude.
tion of the characteristics of the three biases with a large set However, it should be noted that GLONASS C1CC1P and
of IGS legacy tracking stations. However, it has been sug- C2CC2P bias errors (about 0.2 and 0.4 ns) relative to CODE
gested to carefully distinguish the three L2C tracking modes are much higher than that of GPS C1CC1W estimates (about
if an accuracies of 0.7 ns (i.e., below a carrier phase wave- 0.1 ns).

123
Determination of differential code biases with multi-GNSS observations 219

Fig. 8 Monthly variation of


GPS C2WC2L, C2WC2S, and
C2WC2X biases for December
2014

Fig. 9 GLONASS C1CC1P


and C2CC2P DCB estimates
directly extracted from RINEX
observations for December
2014. Satellites are identified by
the GLONASS slot number
followed by the frequency
channel number in brackets

Table 3 Consistency of GPS


System Bias type IGG-CODE DLR-CODE Notes
and GLONASS satellite
intra-frequency biases of GPS C1CC1W 0.20 0.23 Relative to CODE
DLR/IGG with CODE for the
year 2014 (unit: ns) GPS C1CC1W 0.08 0.12 Relative to CODERNX
GLO C1CC1P 0.21 0.18 Relative to CODERNX
GLO C2CC2P 0.39 0.30 Relative to CODERNX
The table provides the RMS for the differences of two DCB products after a zero-mean alignment

GPS and GLONASS inter-frequency biases validation. Figure 10 shows the mean value and STD of GPS
C1WC2W and GLONASS C1PC2P estimates for each indi-
GPS and GLONASS P1-P2 (corresponding to C1WC2W vidual satellite during the period 2013–2014. It is obvious
for GPS and C1PC2P for GLONASS) DCB products of that the consistency of GPS inter-frequency bias estimates
CODE provided in the daily GIM product in IONosphere with CODE is much higher than that of GLONASS biases.
map Exchange (IONEX) format are used as reference for C1WC2W estimates of DLR and IGG have a mean differ-

123
220 N. Wang et al.

Fig. 10 Bias and STD of the


MGEX-based C1WC2W and
C1PC2P DCB estimates relative
to CODE for the period
2013–2014. Dots denote the
mean of DLR/IGG-CODE
differences of the corresponding
satellite, and the length of the
vertical lines denotes the
respective standard deviations

Table 4 Consistency of the


System Bias type IGG-CODE DLR-CODE System Bias Type IGG-DLR TGD-DLR
MGEX-based satellite
inter-frequency bias estimates GPS C1WC1W 0.29 0.24 GPS C1CC5X 0.22
with CODE/DLR for the period
2013–2014 (unit: ns) GLO C1PC2P 0.56 0.84 C1CC5Q 0.29

System Bias Type IGG-DLR BGD-DLR System Bias Type IGG-DLR TGD-DLR

GAL C1XC5X 0.22 0.53 BDS C2IC7I 0.33 1.27


C1XC7X 0.21 0.56 C2IC6I 0.39 1.38
C1XC8X 0.22
C1CC5Q 0.26
C1CC7Q 0.26
C1CC8Q 0.27
The table provides the RMS for the differences of two DCB products after zero-mean alignment

ence of about −0.5–0.7 ns relative to those of CODE, while and IGG bias estimates (which are based on the MGEX net-
the difference varies between −1.2 and 2.0 ns for the esti- work) and the CODE products (which are based on the legacy
mated C1PC2P bias. The standard deviation of C1WC2W IGS network). To quantify the impact of the network on the
and C1PC2P bias differences is approximately 0.07–0.50 and present results, the IGGDCB method has also been applied
0.15–0.65 ns, respectively. RMS values for the differences for the determination of GPS and GLONASS inter-frequency
between MGEX-based GPS and GLONASS bias estimates biases with about 180 sites of the IGS for July, 2014. While
and those of CODE are presented in Table 4. The RMS val- the RMS of the IGG-CODE difference of C1WC2W and
ues for DLR-CODE differences of C1WC2W and C1PC2P C1PC2P biases amounts to 0.26 and 0.52 ns for the MGEX-
are 0.24 and 0.84 ns, while they are 0.29 (C1WC2W) and based solution in this period, it decreases to 0.12 and 0.33
0.56 ns (C1PC2P), respectively, for the IGG solution. ns, respectively, when using the IGS network. Since the IGG-
It has earlier been indicated that bias estimates may be CODE bias differences for individual satellites well exceed
systematically affected by the choice of the tracking network the standard deviation of the bias determination when using
and the employed receivers (Montenbruck et al. 2014). This the MGEX network, the improved consistency of the IGS-
concern is likewise applicable for the comparison of DLR based solution is attributed to the different set of receivers

123
Determination of differential code biases with multi-GNSS observations 221

in this network rather than just the larger overall number the area where ionospheric activities are relatively quiet for
of stations. This is in proper accord with Hauschild and IGGDCB-based DCB estimation (Li et al. 2012). In addition,
Montenbruck (2014) who identified a notable impact of the the ionospheric mismodeling induced by the simple constant
receiver type on the estimation of satellite DCB. For the next height thin-layer model should be noted. This simple assump-
step, a classification of receivers into consistent groups and tion was demonstrated to be responsible for a significant and
an estimation of biases for each group will be further verified. systematic ionospheric modeling error, compared with two-
The bias and STD of C1PC2P DCBs relative to CODE shell voxel models (Hernández-Pajares et al. 1999), and such
are sorted by GLONASS frequency channel number (Fig. 10, error is likewise mapped in terms of DCB determination error
bottom). It can be seen that the variations of DLR-CODE and (Juan et al. 1997). This will be verified for further improve-
IGG-CODE are highly correlated, even though the range of ment of the IGGDCB approach.
DLR-CODE differences appears higher than that of IGG- Figure 12 shows the accuracy and monthly stability of the
CODE. There is also a notable dependence on frequency GPS C1CC5X and C1CC5Q estimates for the period 2013–
channel number for both DLR-CODE and IGG-CODE dif- 2014. Bias and STD of the two DCBs are calculated with
ferences, which may be attributed to inter-channel biases. respected to the estimates of DLR. The ranges of the differ-
Different from the DCB estimation of the GPS constellation, ences between the estimates of IGG and DLR are −0.17–0.12
the receiver bias of GLONASS also depends on the frequency and −0.26–0.15 ns for C1CC5X and C1CC5Q, respectively.
channel since GLONASS satellites transmit their signals on The STDs of C1CC5X and C1CC5Q differences between
slightly different frequencies. IGG and DLR amount to 0.13–0.45 and 0.18–0.45 ns, respec-
The average monthly stability of GPS and GLONASS tively. For the monthly stability, the estimates of DLR are still
DCB estimates of each individual satellite obtained by more stable than those of IGG. The mean values of 0.09 and
CODE, DLR and IGG during 2013–2014 is illustrated in 0.12 ns are obtained for C1CC5X and C1CC5Q estimates of
Fig. 11. The monthly stability index is calculated once DLR, respectively, whereas higher scatters are obtained for
every calendar month (Wilson and Mannucci 1993), which the estimates of IGG. It is obvious that the IGGDCB-based
is shown as GPS C1CC5X and C1CC5Q estimates are reliable compared
 with the products of DLR.
D
d=1 (b̂ − b̄ j )2
d, j
S =
j
, (12) 4.3 BDS DCBs
D−1
where S j is the stability index of satellite j in one month; The current BDS transmits B1, B2, and B3 signals and is
b̂d, j is the daily DCB estimates for satellite jobtained on the thus the first full triple-frequency navigation system. Dif-
day of d; b̄ j is the monthly mean DCB estimates of satellite ferent from the other GNSS constellations, BDS satellite
j; D is the number of days in the corresponding month. clock corrections are referred to the B3 signal rather than the
GPS C1WC2W monthly stability indices for individual dual-frequency ionospheric-free linear combination. To keep
satellites exhibit a scatter in the range of (0.03, 0.07), (0.06, signal coherence, two timing group delay parameters (TGD1
0.10), and (0.09, 0.14) ns for CODE, DLR, and IGG, respec- and TGD2) are defined in BDS navigation message, to pro-
tively. For GLONASS C1PC2P biases, stabilities of (0.05, vide corrections for B1 and B2 signals w.r.t B3 signal (CSNO
0.12), (0.09, 0.23), and (0.12, 0.26) ns are obtained, respec- 2013). TGD1 and TGD2 can be explained as B1–B3 and B2–
tively. The results show that (1) the bias estimates of GPS B3 differential code biases, respectively. The inter-frequency
satellites are more stable than those of GLONASS satellites bias estimates (including C2IC7I, C2IC6I, and C7IC6I) pro-
for the three solutions (CODE, DLR, and IGG); (2) the stabil- vided by DLR exhibit the most reliable BDS bias products
ity index of GPS and GLONASS DCB of CODE estimated at present. In the following analysis, BDS broadcast TGD
with legacy IGS tracking stations is much better than those and DCB estimates of IGG are compared with that of DLR.
of the MGEX-based bias solutions of DLR and IGG (pre- All TGD parameters are normalized to a zero constellation
sumably as a result of the notably larger number of stations average before validation. The satellite C13 (or the so-called
and also the three-day solution of CODE DCB products); BeiDou M3 satellite) has malfunctioned since the beginning
and (3) the bias estimates of IGG present the worst monthly of the year 2014 and no longer serves as part of the current
stability; however, it can still achieve a mean value of 0.11 BDS. Thus, the performance of the DCB estimates of C13 is
and 0.18 ns for C1WC2W and C1PC2P biases, respectively. not included during the corresponding period.
The instability of IGG solution may be caused by the inac- BDS C2IC7I and C2IC6I DCB products of DLR for a full
curate modeling of local ionospheric model at low-latitude span of two years (2013–2014) are shown in Fig. 13. It can
or equatorial sites in the procedure of IGGDCB (as shown be found that the resulting biases appear to be fairly stable,
in Fig. 4). It has been recommended to select the tracking which are confined to a range of ±8 ns with an exception of
stations that can provide high-quality data and are located in GEO satellite C01 (the value of C2IC7I DCB of C01 is about

123
222 N. Wang et al.

Fig. 11 Monthly stability of


C1WC2W and C1PC2P DCB
estimates determined by CODE,
DLR, and IGG for the period
2013–2014

Fig. 12 Statistical results of the


estimated GPS C1CC5X and
C1CC5Q DCBs, including the
bias and STD of the two
IGGDCB-based DCBs relative
to DLR (upper) and the monthly
stability of the two biases
determined by DLR and IGG
(bottom) during the period
2013–2014

15 ns). If look closely at the bias variation of the MEO (C11– available until mid-2014 for DLR, although its navigation
C14) satellites, one may note a small oscillation with a period message terminated earlier.
of approximately one week. This variation may be related Bias and STD of C2IC7I and C2IC6I for each individual
to the one-week repeat cycle of the MEO ground track and BeiDou satellite during the period 2013–2014 are shown in
indicates a systematic impact of the tracking network. It also Fig. 14. It can be seen that the differences between the DCBs
should be mentioned that DCB estimates of C13 were still estimated by IGG and those by DLR mainly vary from −0.15

123
Determination of differential code biases with multi-GNSS observations 223

Fig. 13 Time series of BDS


C2IC7I and C2IC6I DCBs
during the period 2013–2014.
The results have been
normalized by a zero-mean
constraint across the 13
continuously operating satellites
with the exception of C13

Fig. 14 Difference of IGG


DCB estimates and broadcast
TDGs with respect to DLR
DCBs of BDS during the period
2013–2014. Dots and bars
indicate the mean values and the
standard deviations of the daily
bias differences, respectively

to 0.25 ns except for the satellite C03, whereas the broadcast bias products. This is partly related to the different sets of
TGD shows a mean bias of −2.5 to 2.6 ns relative to DLR. the tracking network and receivers used by IGG/DLR and
The standard deviation of IGSO (C06–C10) satellites is sig- BeiDou control segment. In fact, Guo (2015) has shown that
nificantly smaller than that of GEO (C01–C05) and MEO the improved single-point positioning results are achieved in
(C11–C14) satellites. Compared to BDS broadcast TGDs, the BeiDou-only positioning with modern BDS receivers when
IGG results exhibit a notably better consistency with DLR using MGEX-based DCB rather than the broadcast TGD. The

123
224 N. Wang et al.

differences between the broadcast TGDs and the estimates of concluded that the precision of BDS DCB derived by IGG is
DLR may amount to a few nanoseconds, for individual satel- generally better than 0.4 ns with respect to DLR, and that of
lite, which has raised concerns that TGDs may include the broadcast TGD is about 1.4 ns (Table 4).
satellite-induced group delay contributions (Montenbruck The monthly stability of BDS DCB estimates of DLR and
et al. 2014). IGG is illustrated in Fig. 15 for the example of C2IC7I bias.
In terms of the DCB estimates of IGG, the precision levels The monthly stability values of DLR bias solutions are about
of 0.41, 0.16, and 0.32 ns are achieved for C2IC7I biases of 0.24, 0.12, and 0.19 ns for GEO, IGSO, and MEO satellites,
GEO, IGSO, and MEO satellites, respectively, and those for respectively, which are similar for the years 2013 and 2014.
C2IC6I estimates are 0.50, 0.28, and 0.39 ns. With regard The stability of the IGG solutions in 2014 is more stable
to the broadcast TGDs, the consistencies between broadcast than that in 2013 (0.19 vs. 0.26 ns on average). As shown
TGD1 and DLR C2IC7I bias are 1.69, 0.73, and 1.43 ns in Fig. 16, more contributing sites are used in the processing
for GEO, IGSO, and MEO satellites, respectively, and the of IGG for the year 2014 than that for 2013, indicating that
same levels of precision are achieved for TGD1-TGD2 (cor- the DCB estimates may become more stable with more con-
responding to C2IC6I bias), which are 1.64, 0.83, and 1.75 ns. tributing stations and observations. There appears a notable
DCB estimates of IGSO satellites show a better performance drop in the station numbers contributing for DCB estimation
than those of GEO and MEO satellites. This can be attributed of DLR since Sep. 2014, but this issue has been fixed in the
to the longer observation period and larger number of obser- MGEX bias product for the first quarter of 2015. Consider-
vations as compared to the MEO satellites [this phenomenon ing individual types of BDS satellites, the stability indices
has also been noted and discussed in the study of Zhang again confirm an improved performance of the IGSO DCB
et al. 2014)] as well as the adverse multipath characteristics estimates as compared to that of GEO and MEO satellites.
of the GEO observations (Wang et al. 2014). Furthermore,
the weekly bias variation of the MEO satellites, as shown in 4.4 Galileo DCBs
Fig. 13, would affect the stability index, which may be one
of the reasons why MEO satellite DCBs appear to have an The European Galileo comprises four in-orbit-validation
inferior performance than those of IGSO satellites. It can be (IOV) satellites by the end of 2014, offering a total of three

Fig. 15 Monthly stability of


BDS C2IC7I DCB estimates
determined by DLR and IGG for
the year 2013 and 2014

Fig. 16 The number of stations


contributing to BDS satellite
DCB estimation at DLR and
IGG for the period of
2013–2014

123
Determination of differential code biases with multi-GNSS observations 225

signals (E5a, E5b, and E5a+b) in the combined E5ab band fined to a range of ±5 ns, which is obviously smaller than the
in addition to the E1 open service. Two distinct satellite range of B1–B2 and B1–B3 code biases encountered in Bei-
clock offsets are defined in Galileo Interface Control Doc- Dou satellites. Furthermore, differences between C1XC5X,
ument (ICD), which are calculated from the conventional C1XC7X, and C1XC8X DCBs of the individual satellite are
E1/E5a and E1/E5b dual-frequency ionospheric-free linear less than 0.5 ns.
combination. Hence, DCB or BGD parameters accounting Bias and STD of the estimated Galileo satellite DCBs
for satellite hardware delay should be applied for the Galileo relative to that of DLR during the period 2013–2014 are
E1, E5a, or E5b single-frequency users when using the cor- shown in Fig. 18. For each individual satellite, the mean
responding satellite clock corrections (OS-SIS-ICD 2010). difference of both the IGG-DLR and BGD-DLR bias differ-
The Galileo DCB estimates of DLR are regarded as ref- ences is almost zero, but the STD of broadcast BGD is much
erence in the following comparison. In addition, all BGD larger than the estimates of IGG, in particular for satellite
parameters are converted to DCBs and also normalized to E20. This result indicates that the DCB estimates of IGG are
a zero constellation average before validation. More details more stable than the broadcast BGD parameters, which may
about the relationship between Galileo DCB and BGD para- be attributed to the notably smaller network of the current
meters can be found in Montenbruck and Hauschild (2013). Galileo sensor stations as well as different processing strate-
It should be noted that the satellite E20 (IOV_4) had a fail- gies between IGG and the Galileo control segment (Lucas
ure since mid-2014 and is only transmitting the E1 signal Rodriguez 2013). The STD of satellite E20 is significantly
since then. The performance of E20 is not included for that larger than those of other satellites for both broadcast BGDs
period. and DCB estimates of IGG. While a larger variation of the
Time series of Galileo C1XC5X, C1XC7X, and C1XC8X onboard group delays might be suspected for this satellite
products of DLR are shown in Fig. 17. Several gaps are in view of the later E5 transmission failure, it is presently
noted during the time span, which can be attributed to the unclear why the three DCB estimation methods (IGG, DLR,
data discontinuity of the tracking stations. Obvious jumps in and Galileo control segment) would respond different to such
the DCB time series of satellites E19 and E20 can be found variations.
for the first quarter of 2013; however, all DCBs of the four The consistency of the Galileo BGD parameters with post-
Galileo IOV satellites appear to be quite stable since then. processed MGEX DCB products is higher than that of BDS
As shown in the plot, most of the observed DCBs are con- broadcast TGDs (see Figs. 14 and 18). As shown in Galileo

Fig. 17 Time series of Galileo


C1XC5X, C1XC7X, and
C1XC8X DCBs during the
period 2013–2014. The results
have been normalized by a
zero-mean constraint for the
satellites E11, E12, and E19

123
226 N. Wang et al.

Fig. 18 IGG-DLR and


BGD-DLR bias differences
during the period 2013–2014.
Dots and bars indicate the mean
values and the standard
deviations of the daily bias
differences, respectively

broadcast ephemeris, the BGD parameters are updated with GLONASS DCBs is approximately 0.2–0.3 and 0.5–0.8 ns
a time interval of one day, which is more frequent than that of for the estimates of IGG and DLR relative to the products of
BDS TGDs. Furthermore, the 12 Galileo sensor stations are CODE; (3) the precision of BDS and Galileo DCB estimates
geographically distributed on a global scale, while the con- of IGG is 0.36 and 0.24 ns compared to the products of DLR;
tributing stations of BDS are almost limited in the mainland and (4) the monthly stability indices of the IGGDCB-based
and nearby regions of China. This implies that the number biases are approximately 0.11, 0.18, 0.17, and 0.14 ns for
of tracking stations and their distributions would affect the the DCB estimates of GPS, GLONASS, BDS, and Galileo,
accuracy and stability of satellite DCB estimates. Overall, it respectively.
can be concluded that the precision of Galileo satellite DCB Since mid October 2015, we start to contribute this multi-
estimates of IGG is at the level of 0.23 ns, and a mean preci- GNSS DCB product to the MGEX, which is now made
sion of 0.55 ns is also achieved by broadcast BGDs compared available at the IGS CDDIS (ftp://cddis.gsfc.nasa.gov/pub/
with DLR solutions (Table 4). gps/products/mgex/dcb) and IGN (ftp://igs.ign.fr/pub/igs/
products/mgex/dcb) repositories, as well as our own ftp
archive (ftp://ftp.gipp.org.cn/).
5 Conclusions and future work Despite the highly promising results, various issues will
deserve attention to further improve the IGGDCB method.
In this study, the IGGDCB approach is extended to esti- These include the direct estimation of DCB correction for all
mate the DCB in multi-GNSS satellites based on the MGEX possible signals pairs, which is expected to better represent
network. The satellite-plus-receiver DCB is individually GNSS pseudorange observations than chaining of multiple
determined by local GTS-based ionospheric TEC model DCB (Montenbruck and Hauschild 2013). Concerning BDS,
instead of global ionospheric modeling or using predeter- further investigation will be required to understand and miti-
mined TEC maps to mitigate the LOS ionospheric delay. gate the different performance of bias estimates for GEO and
Therefore, the local GTS ionospheric model, which is inde- MEO satellites as opposed to the IGSO satellites. Finally,
pendently applied for each individual station, may efficiently it shall be examined, whether a more accurate and reliable
capture the subtle variations in the ionospheric TEC and also satellite DCB solution can be obtained by selecting tracking
avoid the requirement for a large number of globally distrib- stations in mid-latitudes rather than low latitudes on the basis
uted tracking stations if only a few ground tracking sites are of the accuracy of the local ionospheric modeling.
available.
Acknowledgments The authors are grateful to the editor-in-chief
The DCB products of CODE and DLR are used to validate (Roland Klees), the associate editor (Pascal Willis) and the three
the performance of the IGGDCB-based DCB estimates over anonymous reviewers for the editorial feedback and valuable sugges-
a 24-month period from 2013 to 2014. The validation results tions. We would like to acknowledge the IGS Multi-GNSS Experiment
show that (1) the intra-frequency bias estimates of IGG show (MGEX), International GNSS Monitoring and Assessment System
(iGMAS), Center for Orbit Determination in Europe (CODE) and Ger-
the same precision as that of DLR (about 0.1 ns for GPS man Aerospace Center (DLR) for providing access to GNSS data and
and 0.2–0.4 ns for GLONASS relative to the CODE’s direct differential code bias (DCB) products. We also acknowledge Dr. Ying
bias products, respectively); (2) the precision of GPS and Li for helping to revise this paper, Bruno Garayt and Carey Noll for

123
Determination of differential code biases with multi-GNSS observations 227

coordinating and helping with the delivery of our MGEX DCB prod- Leandro RF, Langley RB, Santos MC (2007) Estimation of P2-C2 biases
ucts to the IGS IGN and CDDIS ftp archive, and the funding supports by means of Precise Point Positioning. In: Proceedings of the 63rd
by National 973 (No. 2012CB825604), China Natural Science Funds annual meeting of the Institute of Navigation, Cambridge, 23-25
(No. 41231064, 41304034, 41104012 and 41621063), Beijing Nat- Apr, pp 225–231
ural Science Foundation (No. 4144094), the CAS/SAFEA International Li H, Xu T, Li B, Huang S, Wang J (2015a) A new differential code
Partnership Program for Creative Research Teams (KZZD-EW-TZ-05) bias (C1–P1) estimation method and its performance evaluation.
and the State Key Laboratory of Geodesy and Earth’s Dynamics (Insti- GPS Solut. doi:10.1007/s10291-015-0438-4
tute of Geodesy and Geophysics, CAS) (SKLGED2014-3-1-E). Li Z, Yuan Y, Li H, Ou J, Huo X (2012) Two-step method for the deter-
mination of the differential code biases of COMPASS satellites. J
Geod 86(11):1059–1076
Li Z, Yuan Y, Fan L, Huo X, Hsu H (2014) Determination of the Differ-
References ential Code Bias for Current BDS Satellites. IEEE Trans Geosci
Remote Sens 52(7):3968–3979
Arikan F, Nayir H, Sezen U, Arikan O (2008) Estimation of single Li Z, Yuan Y, Wang N, Hernandez-Pajares M, Huo X (2015b) SHPTS:
station interfrequency receiver bias using GPS-TEC. Radio Sci towards a new method for generating precise global ionospheric
43(4). doi:10.1029/2007rs003785 TEC map based on spherical harmonic and generalized trigono-
CSNO (2013) BeiDou navigation satellite system signal in space inter- metric series functions. J Geod 89(4):331–345
face control document-open service signal (Version 2.0). China Lucas Rodriguez R (2013) Galileo IOV status and results. In: Proceed-
Satellite Navigation Office, Dec 2013 ings of ION GNSS 2013, Nashville, 16-20 Sep, pp 3065–3093
Dach R, Brockmann E, Schaer S, Beutler G, Meindl M, Prange L, Bock Mannucci A, Wilson B, Yuan D, Ho C, Lindqwister U, Runge T (1998)
H, Jäggi A, Ostini L (2009) GNSS processing at CODE: status A global mapping technique for GPS-derived ionospheric total
report. J Geod 83(3–4):353–365 electron content measurements. Radio Sci 33(3):565–582
Dow JM, Neilan RE, Rizos C (2009) The International GNSS Service Montenbruck O, Hauschild A (2013) Code biases in multi-GNSS point
in a changing landscape of Global Navigation Satellite Systems. J positioning. In: Proceedings of ION ITM 2013, San Diego
Geod 83(3–4):191–198 Montenbruck O, Steigenberger P, Khachikyan R, Weber G, Lang-
Feltens J (2003a) The international GPS service (IGS) ionosphere work- ley R, Mervart L, Hugentobler U (2013) IGS-MGEX: preparing
ing group. Adv Space Res 31(3):635–644 the ground for multi-constellation GNSS science. Inside GNSS
Feltens J (2003b) The activities of the ionosphere working group of the 9(1):42–49
International GPS Service (IGS). GPS Solut 7(1):41–46 Montenbruck O, Hauschild A, Steigenberger P (2014) Differential
Gao Y, Lahaye F, Heroux P, Liao X, Beck N, Olynik M (2001) Modeling Code Bias Estimation using Multi-GNSS Observations and Global
and estimation of C1–P1 bias in GPS receivers. J Geod 74(9):621– Ionosphere Maps. Navigation 61(3):191–201
626 OS-SIS-ICD (2010) European GNSS open service signal in space inter-
Guo F, Zhang X, Wang J (2015) Timing group delay and differential face control document (OS-SIS-ICD, Issue 1.1). European Union,
code bias corrections for BeiDou positioning. J Geod 89(5):427– Sep 2010
445 Rizos C, Montenbruck O, Weber R, Weber G, Neilan R, Hugentobler U
Hauschild A, Montenbruck O (2014) A study on the dependency (2013) The IGS MGEX experiment as a milestone for a compre-
of GNSS pseudorange biases on correlator spacing. GPS Solut. hensive multi-GNSS service. In: Proceedings of ION PNT 2013,
doi:10.1007/s10291-014-0426-0 Honolulu
Hernández-Pajares M, Juan JM, Sanz J (1999) New approaches in global Sardón E, Zarraoa N (1997) Estimation of total electron content using
ionospheric determination using ground GPS data. J Atmos Sol- GPS data: how stable are the differential satellite and receiver
Terr Phys 61(16):1237–1247 instrumental biases? Radio Sci 32(5):1899–1910
Hernández-Pajares M, Juan JM, Sanz J, Orus R, Garcia-Rigo A, Feltens Schaer S (1999) Mapping and predicting the earth’s ionosphere using
J, Komjathy A, Schaer SC, Krankowski A (2009) The IGS VTEC the global positioning system. Ph.D Dissertation, University of
maps: a reliable source of ionospheric information since 1998. J Berne, Switzerland
Geod 83(3–4):263–275 Schaer S (2000) Monitoring (P1-C1) code biases. IGS Mail No. 2827,
Hernández-Pajares M, Juan JM, Sanz J, Aragón-Àngel À, García-Rigo 9 May
A, Salazar D, Escudero M (2011) The ionosphere: effects, GPS Schaer S (2012a) Overview of GNSS biases. In: Proceedings of IGS
modeling and the benefits for space geodetic techniques. J Geod Workshop on GNSS Biases, University of Bern, Switzerland
85(12):887–907 Schaer S (2012b) Activities of IGS Bias and Calibration Working
IGS (2012) IGS Technical Report 2011. University of Berne, Astro- Group. In: Meindl M, Dach R, Jean Y (eds) IGS Technical Report
nomical Institute, Switzerland, July 2012 2011. University of Berne, Astronomical Institute, Switzerland, pp
I GS (2013) RINEX: the receiver independent exchange format (Version 139–154
3.02). IGS Central 2013 Steigenberger P, Montenbruck O, Hessel U (2015) Performance Evalu-
IS-GPS-200 (2012) Navistar GPS Space Segment/Navigation User Seg- ation of the Early CNAV Navigation Message. In: Proceedings of
ment Interfaces (IS-GPS-200, Revision G). Glob Position Syst ION ITM 2015, 26–28 Jan. Dana Point, USA, pp 155–163
Direct Wang G, de Jong K, Zhao Q, Hu Z, Guo J (2014) Multipath analysis
Juan JM, Rius A, Hernandez-Pajares M, Sanz J (1997) A two-layer of code measurements for BeiDou geostationary satellites. GPS
model of the ionosphere using global positioning system data. Solut 19(1):129–139. doi:10.1007/s10291-014-0374-8
Geophys Res Lett 24(4):393–396 Willis P, Beutler G, Gurtner W, Hein G, Neilan R, Noll C, Slater J (1999)
Keshin M (2011) A new algorithm for single receiver DCB estimation IGEX: International GLONASS Experiment-scientific objectives
using IGS TEC maps. GPS Solut 16(3):283–292. doi:10.1007/ and preparation. Adv Space Res 23(4):659–663
s10291-011-0230-z Wilson BD, Mannucci AJ (1993) Instrumental Biases in Ionospheric
Krankowski A, Shagimuratov II, Ephishov II, Krypiak-Gregorczyk A, Measurements Derived from GPS Data. In: Proceedings of the
Yakimova G (2009) The occurrence of the mid-latitude ionospheric ION GPS-93, Salt Lake City, USA, 22–24 Sep, pp 1343–1351
trough in GPS-TEC measurements. Adv Space Res 43(11):1721– Wilson B, Yinger C, Feess W, Shank C (1999) New and improved: the
1731 broadcast interfrequency biases. GPS World 10(9):56–66

123
228 N. Wang et al.

Yuan Y, Ou J (2004) A generalized trigonometric series function model Zhang R, Song W, Yao Y, Shi C, Lou Y, Yi W (2014) Modeling regional
for determining ionospheric delay. Prog Nat Sci 14(11):1010–1014 ionospheric delay with ground-based BeiDou and GPS observa-
Zhang B, Ou J, Yuan Y, Li Z (2012) Extraction of line-of-sight tions in China. GPS Solut. doi:10.1007/s10291-014-0419-z
ionospheric observables from GPS data using precise point posi-
tioning. Sci China Earth Sciences 55(11):1919–1928

123

You might also like