Real-Time Navigation of LEO Satellite Using

QZSS MADOCA-PPP Signal Based on RTKLIB

Hideki Yamada, Toshitaka Sasaki, Kyohei Akiyama, Saya Matsushita, Masato Okeya, Keito Yoshida, Wani Zahid Mushtaq,
and Satoshi Kogure.
Japan Aerospace Exploration Agency

Biographies

Hideki Yamada is a researcher of Japan Aerospace Exploration Agency (JAXA), Japan. He received his Dr. Eng. in Marine
Engineering from Tokyo University of Marine Science and Technology (TUMST) for his work of GPS/GLONASS RTK
positioning. He worked in Electronic Navigation Research Institute (ENRI) and TUMST from 2011 to 2014 and in Hitachi
Zosen corporation from 2014 to 2019. He joined JAXA since 2019, and he has been currently working for precise point
positioning (PPP) and PPP/INS integration.

Toshitaka Sasaki is an Associate Senior Engineer at Japan Aerospace Exploration Agency (JAXA). He received a B.S. in
aerospace engineering from Nihon University in 1999. His specialties are flight dynamics, satellite attitude dynamics, and
satellite navigation technology. He is currently engaged in improving JAXA’s precise orbit determination software for GNSS
and developing on-board components to improve the performance of the next-generation Quasi-Zenith Satellite System.

Kyohei Akiyama is a researcher of Japan Aerospace Exploration Agency (JAXA), Japan. He received his M.S. in Mechanical
and Aerospace Engineering at Tokyo Institute of Technology. He has several years of experience on GNSS as an engineer of
flight dynamics, specifically for the precise orbit determination using GNSS and SLR measurements. He is currently
contributing to improve POD software for GNSS satellites and the conceptual design of Quasi-Zenith Satellite System for
future constellation.

Saya Matsushita is a researcher of Japan Aerospace Exploration Agency (JAXA), Japan. She received the B.S. 2018 degree
in engineering from Keio University. She joined JAXA from April 2018, as a researcher. She has been working for precise
orbit and clock determination of GNSS and PPP since 2018.

Masato Okeya is a researcher of Japan Aerospace Exploration Agency (JAXA), Japan. He has a master's degree from Muroran
Institute of Technology, and belonged to Space Engineering Development (SED(Japan)) from 2009 to 2020. He has been
seconded to JAXA since 2020, and has been working for precise orbit and clock determination of GNSS and PPP.

Keito Yoshida is a researcher of Japan Aerospace Exploration Agency (JAXA), Japan.He received the B.S. 2022 degree in
engineering from National Institute of Technology (KOSEN), Ibaraki College. He joined JAXA from April 2022, as a
researcher. He has been working for engaged in work such as demonstration experiments for social implementation of
MADOCA-PPP.

Wani Zahid Mushtaq is a researcher at Japan Aerospace Exploration Agency (JAXA), Japan. He received his Ph.D. in 2021,
at the Asian Institute of Technology (AIT) for his work on GNSS positioning in urban areas. He worked at Asia Technology
Industry (ATI) from 2019 – 2021 on navigation algorithms for autonomous driving and in REAPLE INC. Japan in 2022, He
joined JAXA in 2023 and has been working on algorithms for MADOCA-PPP, and on improving the PPP-AR accuracy and
convergence time.

Satoshi Kogure is Senior Chief Officer of Satellite Navigation Technology, Japan Aerospace Exploration Agency (JAXA).
He received an MS in aeronautical engineering from Nagoya University in 1993 and an another MS in aerospace engineering
from University of Colorado at Boulder in 2001. He started his career at former JAXA in 1993. He has been working for the
development of QZSS since 2001 and he worked for Cabinet Office, Government of Japan to lead QZSS deployment and its
service provision.
Abstract
Quasi-zenith satellite system (QZSS) multi-GNSS advanced orbit and clock augmentation - precise point positioning
(MADOCA-PPP) is a Japanese precise positioning service by way of QZSS satellite communication links, whose PPP/PPP-
ambiguity resolution (AR) reference software based the well-known open-source RTKLIB is developed. MADOCA-PPP has
the potential to perform the precise real-time navigation of a low earth orbit (LEO) satellite. The purpose of this study is to
show and evaluate the precise orbit determination (POD) algorithm using the MADOCA-PPP approach combined with
RTKLIB modifications suitable for the LEO satellite POD such as satellite dynamic models and tuning techniques in PPP filter.
In several data samples of Sentinel-6A, the reduced dynamics improved the 3D root mean square (3D-RMS) position error by
9–16 % and the convergence time by 2 % compared to the kinematic PPP/PPP-AR solutions. Additionally, the fixed solution
with PPP-AR reduced the position accuracy by 3–11 % and the convergence time by 5 % compared to the float solution and
the average fixed ratio was 84.9 %. Overall, the position solution combining the satellite dynamic model and PPP-AR with the
tuning techniques provided the best results, with an average 3D-RMS position error of 11.4 cm and convergence time of 30.7
minutes confirmed in several samples. From these results, the proposed QZSS MADOCA-PPP based on RTKLIB was found
to achieve LEO POD with decimeter-level position accuracy. The availability of QZSS MADOCA-PPP would be limited in
comparison with global PPP services, however, this is expected to work as the more continuous and precise real-time navigation
of the LEO satellite by considering further algorithm improvement and additional GNSS equipment.

1 INTRODUCTION
In recent years, earth observation satellites in LEO have been actively using high-resolution imagers such as optical sensors
and synthetic aperture radar (JAXA EORC.,2023). As small satellites equipped with such high-resolution imagers become
more common, the number of such satellites is increasing. To obtain high-resolution images, it is necessary to determine the
orbital position of the LEO satellite at the cm level. In general, the orbital positions are calculated using ground-based POD,
however, it takes several hours. To overcome this time-intensive processing, a methodology is proposed to determine the real-
time position of the LEO satellite at the cm level. The previous study (V. B. Damon et al.,2021) evaluated a kinematic POD
using GNSS broadcast ephemerides and smoothing method with real-time navigation as the target, but in order to obtain more
precision in real-time, a PPP/PPP-AR technique with a highly accurate ephemeris and LEO satellite dynamic model are required
(Wang, Z et al.,2022). The goal of our research is to perform PPP/PPP-AR, namely MADOCA-PPP with a civilian GNSS
receiver that can receive the L6E orbit and clock augmentation signal from the QZSS in orbit (including space environment
resistance) (CAO,2022). A computing device is developed and installed on the LEO satellite to perform PPP processing. In
order to develop a PPP processing receiver that can be applied to LEO constellations and various other missions (ionospheric
occultation observation, etc.) the RTKLIB, an open-source program package for GNSS positioning, is used. The RTKLIB has
a proven track record as software for a low-cost receiver (Takasu T et al.,2009), which needs to be modified so that it can be
applied in orbit. The purpose of the research presents the results of evaluating the POD algorithm and feasibility using RTKLIB,
in which we have implemented the archive data decoding function of QZSS L6E and the on-orbit navigation function
(Kinematic and Reduced- dynamic models). In the remaining part of the Introduction, a detailed explanation is presented about
the differences and novelties of our onboard POD algorithm and related work.

The novelty of the research is that the LEO POD was evaluated using the QZSS L6E archived data released in 2023. In addition
to MADOCA-PPP using QZSS L6E, other PPP services around the world include Galileo high accuracy service (HAS)
(GSA,2020) and BeiDou-3 PPP-B2b (CSNO,2021). In the literature (Hauschild, A et al.,2022), LEO POD using the GPS and
Galileo augmentation information by Galileo HAS, was evaluated and achieved 3D-RMS position errors of less than 10 cm. In
contrast, the LEO POD using the GNSSs that can be augmented by QZSS L6E including GPS, QZSS, Galileo, and GLONASS
is expected to achieve the same or higher accuracy while shortening the initial convergence time. However, an additional GNSS
antenna to the along-track direction is required at the onboard LEO satellite that can receive QZSS L6E signals behind the
earth's horizon. The research on the evaluation of LEO POD using L6E distributed from satellites has not been performed and
the only literature (Allahvirdi-Zadeh, A, et al.,2021) available is about LEO POD using MADOCA orbit/clock products for
Internet distribution. In other words, to be precise, the evaluation results of LEO POD using QZSS L6E have not yet been
reported.

Furthermore, the distinguishing point of this research is that not only kinematic but also reduced-dynamic models are
implemented in RTKLIB to evaluate LEO POD using the L6E augmentation signal. The RTKLIB modified for POD has been
used in the literature (Giordano, P, et al.,2017), but the results were limited only to kinematics using IGS real-time products.
However, we implemented an on-orbit navigation model (kinematic and reduced-dynamical models) in a RTKLIB-derived
software called MADOCALIB (CAO,2023) which can decode augmentation information in the compact state space
representation (CSSR) format of L6E published in May 2023. This made it possible to evaluate LEO POD using RTKLIB and
L6E archive data. In addition, RTKLIB's PPP not only uses ionospheric-free linear combinations to reduce the effects of
ionospheric delay but also has the ability to estimate ionospheric delay. The latter method has the advantage that even if only
one frequency of the two-frequency observables is available during the estimation process at successive epochs, the remaining
observables can be used for PPP analysis. Compared to the group and phase ionosphere correction (GRAPHIC) method, which
is well known as a single frequency-based POD (Conrad, A.V, et al.,2023), it is possible to estimate the ionospheric delay
component and monitor its magnitude.

Based on the above background, in the next section, we describe the LEO POD algorithm using RTKLIB, L6E, and the reduced-
dynamic model, and demonstrate its position accuracy and convergence performance.

2 METHODS
2.1 QZSS MADOCA-PPP
QZSS is a Japanese satellite navigation system with a regional service coverage, and a total of four satellites are in operation
under the control of the cabinet office, government of Japan (CAO). The QZSS provides signals for ranging and GNSS
augmentation to enable navigation service to Asia Oceania region as well as Japanese territory, and among the augmentation
signals, the four QZSS satellites transmit L6E signals which include correction messages for PPP service. The augmentation
information is generated by four systems on ground: two systems from the Hitachi-Ota station and two systems from the Kobe
station, and then is uploaded from the ground station to the QZSS satellites. Figure 1 shows the structure of L6E message. Each
L6E message is transmitted at 2,000 bits/s and is composed of a 49-bit header, a 1695-bit data section, and a 256-bit Reed-
Solomon code. The data part of MADOCA-PPP includes RTCM message type 4073 “Compact SSR” and is given in state
space representation (SSR) form, as standardized in the applicable document (RTCM,2016). The types of L6E correction
messages are the SSR mask which defines the satellites and signals to be augmented, the satellite orbit and the satellite clock
corrections for broadcast ephemeris, the differential code bias (DCB), the fractional cycle bias (FCB), and the user range
accuracy (URA). The update interval is 5 s for the satellite clock correction and 30 s for other ones. Currently the GNSS to
which L6E correction messages are applied and distributed are GPS, QZSS, GLONASS, and Galileo.

Figure 1: L6 Message Structure (CAO,2022)

To achieve the development of MADOCA-PPP software, CAO has released MADOCALIB (RTKLIB v2.4.3b34-based) 1.0b
internationally, including the source code since June 5th, 2023, and also has published L6E message archived data since April
in 2023 (CAO,2023). MADOCALIB enables to read the CSSR message of MADOCA-PPP and process PPP/PPP-AR with a
utility called "RNX2RTKP" for post-process positioning. In this study, we used L6E archive data and a modified version of
MADOCALIB to evaluate the real-time POD of a LEO satellite using the QZSS MADOCA-PPP signal. However, since the
archive data is properly aligned from the beginning of the preamble in every epoch, it is difficult to simulate the
transmission/reception latency that may occur in an actual L6E reception environment or the case where data is received from
the middle of the data. Therefore, evaluation results equivalent to real-time except for these cases are mentioned in this study.

Next, we discuss the application of MADOCA-PPP to a LEO satellite. As mentioned above, MADOCA-PPP signals are
intended for use in regional areas on Earth, so when these signals are received by the LEO satellite, there are periods when
QZSS satellites are not visible. Figure 2(a) shows the invisible zone of QZSS satellites (CASE1) when one GNSS antenna
(capable of receiving L6E signals) is mounted on the anti-earth plane of the LEO satellite (altitude: 700 km), and when two
GNSS antennas are mounted on the front and rear side of the satellite's direction of travel. The invisible zone of QZSS satellites
(CASE2) is shown in Figure 2(b). It was confirmed that the QZSS invisible periods of CASE1 is up to about 40 minutes (one
orbital period), while that of CASE2 can be reduced to about 20 minutes (one orbital period). Possible ways to deal with the
QZSS invisible zone include " uploading ultra-rapid precise ephemerides from ground stations", "orbit propagation using a
high-precision dynamic model", and so on. A better method for evaluating the effectiveness of MADOCA-PPP for the LEO
satellite includes the consideration regarding the invisible zone; however, this evaluation will be clarified in a future study.
Therefore, in this study, as the first evaluation, we discuss the position accuracy of MADOCA-PPP in the LEO satellite,
assuming that the QZSS satellite is always visible on the LEO satellite.

(a) One antenna mounted on anti-earth plane

(b) Two antennas mounted on the front and rear side of the satellite's direction of travel

Figure 2: QZSS invisible zone

2.2 PPP/PPP-AR observation model

Since MADOCALIB can originally be applied to PPP/PPP-AR on terrestrial static stations or moving platforms, the terms of
tides and tropospheric delays of the observation models are not affected and should be excluded for space navigation, such as
for LEO satellites. And also, in the receiver phase center offset (PCO), the attitude information e.g., quaternion, of the LEO
satellite is required. Therefore, the PPP/PPP-AR observation models are shown in Equations (1) – (2).

𝑓
𝜆 Φ, = 𝑟 + 𝑐 ∙ (𝑑𝑡 − 𝑑𝑇 ) + 𝑐 ∙ 𝐼𝑆𝐵 − 𝐼 + 𝐵 + 𝑝𝑐𝑜 + 𝑝𝑐𝑜 + 𝑝𝑐𝑣 + 𝑝𝑐𝑣 + 𝑑𝜙
𝑓
+ 𝜀(Φ) (1)
 𝑓
𝑃, = 𝑟 + 𝑐 ∙ (𝑑𝑡 − 𝑑𝑇 ) + 𝑐 ∙ 𝐼𝑆𝐵 + 𝐼 +𝑏, +𝑏 ,, + 𝑝𝑐𝑜 + 𝑝𝑐𝑣 + +𝜀(P) (2)
𝑓

,where the subscripts i, SYS, and s denote the first or the second frequency band of the GNSS signals, symbol of the GNSS,
and a number representing the GNSS satellite, respectively. Additionally, Φ and 𝑃 are the carrier phase and code observations,
respectively, and 𝜆 is the wavelength. Furthermore, f is the carrier phase frequency, and 𝑟 is the geometric distance from the
GNSS satellite s to the receiver. Moreover, dt and dTs are the receiver and satellite clock offsets, respectively, ISB is the inter-
system bias between GPS and other GNSS, I is the ionospheric delay, 𝑏 , 𝑏 are the satellite and receiver code bias including
DCB, B is the carrier phase ambiguity, pco and pcv are the corrections for the phase center offset and phase center variations,
and 𝑑𝜙 is the phase-wind up effect, and 𝜀 is the multipath and the receiver noise.

Furthermore, a unique feature of the observation models in this study is to use a method that estimates the ionospheric delays
in the slant direction with an extended Kalman filter (EKF) instead of doing ionospheric-free linear combinations, although
both methods have already been implemented in MADOCALIB, for use in missions to observe ionospheric occultations. The
ionospheric delays can be estimated assuming time correlation with the random walk model. The initial values are calculated
by the dual-frequency code observations, expressed using Equation. (3), so the receiver DCB needs to be separated from the
estimated value to monitor the true ionospheric delay. On the contrary, the position solution might not be affected even if the
term is not done since this term that is common to the equations pertaining to observations is absorbed to the receiver clock
term. The approach to estimate ionospheric delays can be used to continue PPP processing as long as single-frequency
observation at each satellite is obtained after the initial epoch 𝑡 to use each satellite which requires dual-frequency one in
Equation. (3).

𝑃 −𝑃 −𝑏 , −𝑏 , ,
𝐼 (𝑡 ) = (3)
𝑓
1−
𝑓

Next, in PPP-AR, the estimated float ambiguities from the single differences between the GNSS satellites are used in the form
of zero-difference since the temporal variation of the receiver FCB is generally not stable. These float ambiguities and the
matrix of their covariances are input to the AR process, which uses the LAMBDA method reported by Teunissen, PJG (1995).
MADOCALIB was originally an implementation that could only input GPS ambiguities to AR, but it was modified so that the
input ambiguities for GNSS satellites other than GLONASS can be used to enhance increased accuracy using the LAMBDA
method. Additionally, the nominal yaw attitude models of GNSS satellites are implemented in the phase wind-up model to
obtain fixed solutions for PPP-AR, however, the case where the beta angle of the GNSS satellite becomes small is not
considered.

As other modifications to the observation model in MADOCALIB, some methods were also added to effectively process multi-
GNSS signals. The first method was added to allow users to tune the noise model in observation data for each GNSS and
frequency, based on the residual obtained from the EKF. The observation error covariance matrix is expressed using Equation.
(4), and the ratio in Equation (4) is the weight that specifies the number of times the code observation noises are multiplied by
those of the carrier phase. And also, the standard deviation σ of observation errors representing each element of the matrix is
expressed using Equation (5). “ The PPP filter residual approach” is a more accurate and more efficient way to tune the standard
deviation of unknown observation errors rather than an empirical approach which can be used to identify a combination of a
and b. The paper (Yamada, H, et al.,2022) is referred for the details of this approach.

𝑹 = 𝒅𝒊𝒂𝒈((𝜎 ) , (𝜎 ) , … , (𝜎 ) , 𝑟𝑎𝑡𝑖𝑜 ∙ (𝜎 ) , 𝑟𝑎𝑡𝑖𝑜 ∙ (𝜎 ) , … , 𝑟𝑎𝑡𝑖𝑜 ∙ (𝜎 ) ) (4)

𝑏
(𝜎 ) = (𝑎 ) + (5)
sin(𝐸𝑙𝑒𝑣𝑎𝑡𝑖𝑜𝑛)

Finally, the added method is to estimate the ISB as a constant considering time correlation under the condition of not changing
the variation of the magnitude of the ISB over time. This method reduces the estimated parameter by one in the estimation
process using the EKF, so it is considered to improve the position accuracy of PPP/PPP-AR in situations where the number of
observations is fewer, compared to the epoch-by-epoch estimation one.
2.3 LEO satellite dynamics model
The differential equation for a LEO satellite motion is described in the framework of Newtonian physics (Montenbruck, O, et
al., 2000). By modeling the perturbations that affect the satellite's acceleration, the current position and velocity can be obtained
by the numerical integration by feeding the initial values of its position and velocity at the previous period into the equation of
motion. It has been confirmed that by using the reduced-dynamic model (Yunck, T.P, et al.,1990; Montenbruck, O, et al., 2005)
which applies the output of the numerical integration of the satellite's equation of motion to the state transition matrix of the
PPP filter, the position accuracy is improved compared to the kinematic solution that performs POD using only GNSS data
(Wang, Z et al.,2022) . In this study, the reduced-dynamic mode was also incorporated into MADOCALIB to enable LEO POD
more accurately. For each acceleration model, we considered Earth's gravitational potential, solid tides, the gravity of the moon
and the sun, the atmospheric drag, the solar radiation pressure, and empirical acceleration that can compensate for unmodeled
forces. On the other hand, the relativistic effects, the earth radiation pressure, and the ocean tides are not considered. Details of
the model used are described in the next section. Regarding the equation of motion, it can be solved in ITRF (ECEF coordinate
system) by using the fourth-order Runge-Kutta method, which is a type of numerical integration, and adding terms related to
centrifugal acceleration and Coriolis force shown in the literature (Montenbruck, O, et al.,2008;Yang,Y, et al.,2015) to the
motion model. In the reduced-dynamic model, Equation (6) expresses their unknown vector. Our algorithm estimates the LEO
satellite position (x, y, z) and velocity (vx, vy, vz), a receiver clock bias, scaling factors CR, CD of the solar radiation pressure
and the atmospheric drag, radial, along, track direction components 𝑎 , 𝑎 , 𝑎 for empirical acceleration, ISB, the L1 and L2
carrier phase float ambiguities, and slant ionospheric delays, using the EKF. The superscript n denotes the total number of
GNSS satellites used for PPP. More details of the estimation conditions are described in the next section.

𝑥, 𝑦, 𝑧, 𝑣𝑥, 𝑣𝑦, 𝑣𝑧, 𝑑𝑡, 𝐶𝑅, 𝐶𝐷, 𝑎 , 𝑎 , 𝑎 , 𝐼𝑆𝐵,

𝑿= (6)
𝐵 ,𝐵 ,…,𝐵 ,𝐵 ,𝐵 ,…,𝐵 ,𝐼 ,𝐼 ,…,𝐼

Next, the state transition matrix is expressed by Equation (7). By performing numerical integration using the variational
equation (Montenbruck, O, et al., 2000; Montenbruck, O, et al.,2008; Yang,Y, et al.,2015), the differential coefficients of the
satellite's acceleration were obtained, and these coefficients were set as Φ(6×6) in the terms of the position/velocity and S(6×2)
in the terms of the scaling factors of the solar radiation pressure and the atmospheric drag. Note that the diagonal term of the
empirical acceleration in equation (7) is written as d・I(3×3) in related papers (Montenbruck, O, et al., 2008; Yang,Y, et
al.,2015), where d is a damping factor expressed as an exponential function. However, as a result of tuning, since d=1, it was
simply set to I(3×3).

𝚽( × ) 𝐎( × ) 𝐒( × ) 𝐒( × ) 𝚶( × ) 𝚶( × ) 𝚶( × )

⎛ 𝚶( × ) 𝐎( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) ⎞
⎜ 𝚶( × ) 𝐎( × ) 𝐈( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) ⎟
⎜ ⎟
𝚽 =⎜ 𝚶( 𝐎( 𝚶( × ) d ∙ 𝐈( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) (7)
× ) × ) ⎟
⎜ 𝚶( × ) 𝐎( × ) 𝚶( × ) 𝚶( × ) 𝐈( × ) 𝚶( × ) 𝚶( × ) ⎟
⎜ 𝚶( 𝐎( 𝚶( 𝚶( 𝚶( 𝐈( × ) 𝚶( × )
⎟
× ) × ) × ) × ) × )

⎝ 𝚶( × ) 𝐎( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) 𝚶( × ) 𝐈( × ) ⎠

Regarding S(6×3), we used the transformation matrix E from the ECEF coordinate system to the RAC (radial, along-track and
cross-track) coordinate system , expressing in Equation (8) and set to S(6×3) using Equation (10) since the empirical acceleration
was described shown in Equation (9).

𝑬( × ) = (𝒆 ( × ) 𝒆 ( × ) 𝒆 ( × )) (8)
𝑎
𝒂 =𝑬 𝑎 (9)
𝑎
𝑑𝒂
𝑺( × ) = = 𝑬 (1: 3,1)( × ) 𝑬 (1: 3,2)( × ) 𝑬 (1: 3,3)( × ) (10)
𝑑𝒑

Finally, the single point positioning (SPP) solution was used as the initial position and velocity values in the numerical
integration. The velocity vector in SPP is obtained by adding the observation equation of the pseudorange rate. The velocity
observation model uses equations described in the literature (P.D. Groves, 2015). However, although the range rate observables
are not included in the PPP filter process, the position and velocity terms of the Kalman gain matrix are updated by the results
of the numerical integration, so the position and velocity terms are updated and corrected more accurately in the measurement
update (in the estimation of the float solution).

3 EVALUATION
To verify the effect of QZSS MADOCA-PPP/PPP-AR based modified RTKLIB, we evaluated the position accuracy and
convergence time (CT) using observation data in the LEO satellite.

3.1 Observation data from Sentinel-6A satellite

We used the observation data for DOY 60 to 66 in 2023 from the Sentinel-6A satellite, which is a radar altimeter satellite
developed by the European Space Agency, for algorithm verification. Firstly, the magnitudes of a,b, and ratio in Equation (4)-
(5) were determined using the fine-tuning weighting method (Yamada, H, et al.,2022). Table 1 lists the estimated a,b, and ratio
for each GPS and Galileo carrier observation errors. In this fine-tuning, we used 24-hour RINEX data from DOY60 to estimate
only a, which is a non-elevation dependent term, out of a and b that make up carrier-phase noise. In order to strictly determine
b, it is necessary to verify this using more data, so in this study, we simply approximated b to be the same size as a to avoid the
estimation error of b As shown in Table 1, these results show the magnitude of the effect due to not corrected by QZSS L6E
messages, and the positioning of its error factors e.g., estimation errors of ionospheric delays in observation data. Additionally,
Table 1 indicates the magnitude of the corrected residual error using QZSS MADOCA-PPP during this period is generally not
largely compared to the observation errors (carrier-phases: 10 mm, code: 1 m) corrected with Galileo HAS (Hauschild, A et
al.,2022).

Table 1: Values of observation noise standard deviation for QZSS MADOCA-PPP in Sentinel-6A
Carrier-phase (a,b) Code/carrier ratio (Ratio)
Note: a,b are in Equation (5)
GPS L1: a=b=4.1 mm, L1: Ratio(a)=88, Ratio(b)=36
L2: a=b=2.6 mm L2: Ratio(a)=152, Ratio(b)=80
Galileo E1: E1:
L1 GPS a(=b) multiplied by 0.87 L1 GPS Ratio(a,b) multiplied by 0.54
E5a: E5a:
L2 GPS a(=b) multiplied by 0.77 L2 GPS Ratio(a,b) multiplied by 0.59

Next, Table 2 shows the PPP/PPP-AR model and processing strategy in this section and Table 3 shows an example of the effect
of applying this fine-tuning to PPP-AR. These results revealed that by using this method, the number of fixed solutions
increased by 14% and the convergence time decreased by 40%. Therefore, we confirmed this tuning technique is also effective
in the RTKLIB used in this study as well as in the literature (Yamada, H, et al.,2022).
 Table 2: PPP/PPP-AR model and processing strategy
Parameter estimation EKF
GNSS GPS, Galileo
Measurements Dual-frequency carrier phase and pseudorange measurements
(L1,L2,E1,E5a) noise model (refer to Table 1.)
Elevation mask 0°
Ephemerides Broadcast ephemerides corrected by QZSS L6E archived data
DCB and FCB corrections Corrected by QZSS L6E archived data
Ionospheric delay Estimation assuming time correlation with random walk
model (process noise: 0.01 m, initial standard deviation: 0.03
m)

Receiver clock bias Estimated epoch by epoch (initial standard deviation: 60 m)

ISB Estimated ISB between GPS and Galileo as a constant (initial

standard deviation: 1 m)

PCO/PCV corrections igs14.atx for GNSS satellite PCO/PCVs

Corrected the lever arm between antenna phase center and
center of mass in Sentinel-6A.
No correction for receiver PCV
Phase wind-up Wu, J.T, et al.,1993

Ambiguity resolution LAMBDA with Ratio test (threshold:2.5)

Numerical Integration Fourth-order Runge-Kutta (integration step: 10 s)

Initial values: SPP position and velocity and their initial
standard deviation: 1 m,1m/s
Earth's gravitational potential EGM 2008 (70×70)
Solid tides IERS 2010 (Petit, G, et al., 2010)
Gravity of the moon and the sun NASA JPL DE405 (Standish, E.M.,1998)
Atmospheric drag Cannon-ball and NRLMSISE-00 (Picone,J. M, et al.,2003)
CD: estimated as a constant.( initial standard deviation: 0.4)
Solar radiation pressure Cannon-ball (Montenbruck, O, et al., 2000)
CR: estimated as a constant. (initial standard deviation:0.2)
Empirical acceleration Estimated by transitioning states using numerical integration.
Process noise (Montenbruck, O, et al.,2008)
𝜎𝑒𝑚 𝑝𝑎𝑐𝑐 = (500 𝑛𝑚/𝑠 2 , 600 𝑛𝑚/𝑠 2 , 500 𝑛𝑚/𝑠 2 )
τ= 50 s,
Initial standard deviation: (R,A,C) = 0.01 m/s2

Table 3: Effect of the fine-tuning weighting method for the QZSS MADOCA-PPP-AR with reduced-dynamics model and
GPS/Galileo observations (Sentinel-6A onboard GNSS receiver) on 1st March (DOY60), 2023
Weighting method Fixed ratio with PPP-AR Convergence Time (CT*)
Popular used σ 77.5 % 23 min
(a=b=3 mm, ratio=100)
Fine-tuning σ (Ref: Tabel 1) 88.6 % 14 min

*: CT is defined as the first time within 20 cm of a three-dimensional (3D) position error from the start time of the filtering
process.

Next, we confirmed the effect of the ambiguity resolution on the real-time LEO POD, Figure 3 shows an example of this effect
and Figure 4 shows the temporal variation of the number of used satellites (the green and the brown of plots represent the status
of each fixed solution and float solution). In most epochs, the PPP-AR positioning solution has a smaller variation in positioning
error than the PPP, and during this period, the PPP-AR has a 3D-RMS error of the PPP positioning solution of about 10%
reduction.
 (a) PPP

(b) PPP-AR

Figure 3: Comparison between PPP (up) and PPP-AR (down) for the reduced-dynamics model with GPS and Galileo
(the same data as those in Table 3)
 Figure 4: The number of satellites used in PPP/PPP-AR

Additionally, in Figures 3 and 4, it can be seen that the several positioning errors on the spikes tend to correspond with decreases
in the number of satellites used for positioning (For example, around 0-15,000 seconds or 40,000-45,000 seconds). Also, the
average number of satellites used during the epochs with positioning errors (absolute value) of 15 cm or more was an average
of 9, whereas in the case where the positioning errors (absolute value) was 15 cm or less, the average 12 satellites were used in
PPP/PPP-AR. Therefore, there is room for improvement in the position accuracy of real-time navigation using QZSS
MADOCA-PPP in the future by using GNSS receivers onboard LEO satellites that can use not only GPS and Galileo but also
other GNSSs.

Finally, Table 4 shows the effect of the QZSS MADOCA-PPP/PPP-AR on the position accuracy and convergence time in
several data samples. The daily 3DRMS position error (3D-RMSE) at Sentinel-6A was calculated by excluding the epochs of
1 h from the start of the initial convergence process and of the positioning errors of 1 m or higher. The 3D-RMSE values in
Table 4 are the average daily 3D-RMSE values for 7 days (DOY60-66). Table 4 reveals that generally, the reduced-dynamics
approach improves the position accuracy by 9 –16 % and the convergence time by 2 % in comparison with the kinematic one.
Among these positioning modes, we confirmed that PPP-AR with a reduced-dynamics model provides the best results in terms
of position accuracy and convergence time, and can obtain decimeter-level position accuracy.

Table 4: 3D-RMSE, Fix ratio, and CT of PPP/PPP-AR in the Sentinel-6A onboard data for DOY 60-66 in 2023
3D-RMSE( Avg. /Mdn.) Fixed ratio (Avg.) CT (Avg.)
Kinematic (PPP) 13.9 cm / 11.1 cm － 32.9 min
Kinematic (PPP-AR) 13.4 cm / 10.5 cm 84.7% 31.2 min
Reduced-dynamics (PPP) 11.7 cm / 10.1 cm － 32.3 min
Reduced-dynamics (PPP-AR) 11.4 cm / 9.4 cm 84.9% 30.7 min
4 CONCLUSIONS
The research implemented a LEO POD algorithm using QZSS MADOCA-PPP in RTKLIB and evaluated the position accuracy
and convergence time. The proposed algorithm was found to achieve LEO POD with decimeter-level position accuracy by
using the reduced-dynamic model and ambiguity resolution, with the help of the fine-tuning of the observation error model in
the PPP filter. Therefore, we confirmed that the implemented satellite dynamic model and ambiguity resolution work effectively
on the real-time POD of LEO satellites. On the other hand, there is a correlation between epochs when there are fewer satellites
and accuracy degradation, so there is still room for improvement in QZSS MADOCA-PPP by using more GNSS and improving
other algorithms. Future work will explore methods to complement QZSS MADOCA-PPP in the invisible zone of QZSS, and
will also consider ways to maintain better accuracy even with a small number of satellites. The prototype version and algorithm
developed in this research are expected to serve as reference information for designing future LEO PODs.

ACKNOWLEDGMENTS
The authors thank CAO for providing MADOCALIB and TOSHIBA ELECTRONIC TECHNOLOGIES CORPORATION for
supporting its development..

REFERENCES
JAXA EORC. (2023). ALOS Research and Application Project: https://www.eorc.jaxa.jp/ALOS/en/index_e.htm

V.B. Damon., Axelrad, P., & Palo, S. (2021). A low complexity smoothing algorithm for improved GPS point solutions on
board LEO spacecraft. NAVIGATION, 68(1), 185-198. https://doi.org/10.1002/navi.410

Wang, Z., Li, Z., Wang, N., Hoque, M., Wang, L., Li, R., Zhang, Y & Yuan, H. (2022). Real-Time Precise Orbit
Determination for LEO between Kinematic and Reduced-Dynamic with Ambiguity Resolution. Aerospace, 9(1), 25;
https://doi.org/10.3390/aerospace9010025

CAO. (2022). IS-QZSS-MDC-001, Quasi-Zenith Satellite System Interface Specification Multi-GNSS Advanced Orbit and
Clock Augmentation-Precise Point Positioning.

Takasu T & Yasuda A. (2009). Development of the low-cost RTK-GPS receiver with an open source program package
RTKLIB, International Symposium on GPS/GNSS, International Convention Center Jeju, Korea, November 4–6.

European GNSS Agency (GSA). (2020). Galileo high accuracy service HAS info note; European Global Navigation Satellite
Systems Agency. https://www.gsc-europa.eu/sites/default/files/sites/all/files/Galileo_HAS_Info_Note.pdf

China Satellite Navigation Office (CSNO). (2021). BeiDou Navigation Satellite System Open Service Performance Standard
(Version 3.0).

Hauschild, A., Montenbruck, O., Steigenberger, P., Martini, I & Fernandez‑Hernandez, I. (2022). Orbit determination of
Sentinel-6A using the Galileo high accuracy service test signal. GPS Solutions, pp. 1–13. https://doi.org/10.1007/s10291-
022-01312-5

Allahvirdi-Zadeh, A., Wang, K & El-Mowafy, A. (2021). POD of small LEO satellites based on precise real-time MADOCA
and SBAS-aided PPP corrections. GPS Solutions, pp. 1–14. https://doi.org/10.1007/s10291-020-01078-8.

Giordano, P., Zoccarato, P., Otten, M & Crisci, M. (2017). P2OD: Real-time Precise Onboard Orbit Determination for LEO
satellites, Proceedings of the 30th International Technical Meeting of the Satellite Division of The Institute of Navigation
(ION GNSS+ 2017), Portland, Oregon, September 2017, pp. 1754-1771. https://doi.org/10.33012/2017.15190

CAO. (2023). MADOCA Test Library: https://qzss.go.jp/en/technical/dod/madoca/madoca_test-library.html

Conrad, A.V., Axelrad, P., Haines, B., Zuffada, C & O’B, Andrew. (2023). Improved GPS-Based Single-Frequency Orbit
Determination for the CYGNSS Spacecraft Using GipsyX. NAVIGATION, 70 (1). https://doi.org/10.33012/navi.565
Radio Technical Commission for Maritime Services (RTCM). (2016). RTCM STANDARD 10403.3 DIFFRENTIAL GNSS
(GLOBAL NAVIGATION SATELLITE SYSTEMS) SERVICE –VERSION 3, RTCM SPECIAL COMMITTEE NO.104,
7-OCT-2016.

CAO. (2023). L6E archive data: https://sys.qzss.go.jp/dod/en/archives/agree_madoca.html

Teunissen PJG. (1995). The least-square ambiguity decorrelation adjustment: a method for fast GPS ambiguity estimation,
Journal of Geodesy, pp. 65–82. http://doi.org/10.1007/BF00863419

Yamada, H., Matsushita, S., Okeya, M., Kogure, S & Takasu, T. (2022). Weighting Strategy of Observations for Kinematic
Multi-GNSS PPP/PPP-AR using Residual Approach. Proceedings of the 35th International Technical Meeting of the Satellite
Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 2810-2821.
https://doi.org/10.33012/2022.18523.

Montenbruck, O & Gill, E. “Satellite Orbits Models Method Applications”. 1st edition, 2000.

Yunck, T.P., Wu, S-C., WU, J-T & Thornton, C.L (1990). Precise Tracking of Remote Sensing Satellites With the Global
Positioning System, IEEE Transactions on Geoscience and Remote Sensing, Volume: 28, Issue: 1. 10.1109/36.45753.

Montenbruck, O., Helleputte, T.V., Kroes, R & Gill, E. (2005). Reduced dynamic orbit determination using GPS code and
carrier measurements, Aerospace Science and Technology 9 (2005) pp.261–271. 10.1016/j.ast.2005.01.003.

Montenbruck, O & R-B, Pere. (2008). Precision real-time navigation of LEO satellites using global positioning system
measurements. GPS Solutions, pp. 187–198 10.1007/s10291-007-0080-x.

Yang,Y., Yue, X & Yuan, J. (2015). GPS Based Reduced-Dynamic Orbit Determination for Low Earth Orbiters with
Ambiguity Fixing. Hindawi Publishing Corporation International Journal of Aerospace Engineering, Volume 2015, pp. 1-
11. http://dx.doi.org/10.1155/2015/723414.

P.D. Groves, “Principles of GNSS, inertial, and 12multisensory integrated navigation systems”, 2nd edition, 2015.

Wu, J.T., Wu, S.C., Hajj, G.A., Bertiger, W.I., & Lichten, S.M. (1993). Effects of antenna orientation on GPS carrier phase,
Manuscripta Geodaetica, Vol 18, No. 2, pp. 91-98.

Earth Gravitational Model 2008 (EGM2008). http://earth-info.nga.mil/GandG/wgs84/gravitymod/egm2008/index.html

Petit, G & Luzum, B (eds.) (2010) IERS Conventions (2010).

Standish, E.M. (1998). JPL Planetary and Lunar Ephemerides, DE/LE405, JPL IOM312.F-98-048.

Picone,J. M., Hedin,A. E., Drob,D. P. & Aikin, A. C. (2003). NRLMSISE-00 Empirical Model of the Atmosphere:
Statistical Comparisons and Scientific Issues, J.Geophys.Res, 2003. https://doi.org/10.1029/2002JA009430.