Assessment of Affordable Real-Time PPP Solutions for Transportation Applications

Abstract

With the availability of multi-frequency, multi-constellation global navigation satellite system (GNSS) modules, precise transportation applications have become attainable. For transportation applications, GNSS geodetic-grade receivers can achieve an accuracy of a few centimeters to a few decimeters through differential, precise point positioning (PPP), real-time kinematic (RTK), and PPP-RTK solutions in both post-processing and real-time modes; however, these receivers are costly. Therefore, this research aims to assess the accuracy of a cost-effective multi-GNSS real-time PPP solution for transportation applications. For this purpose, the U-blox ZED-F9P module is utilized to collect dual-frequency multi-GNSS observations through a moving vehicle in a suburban area in New Aswan City, Egypt; thereafter, datasets involving different multi-GNSS combination scenarios are processed, including GPS, GPS/GLONASS, GPS/Galileo, and GPS/GLONASS/Galileo, using both RT-PPP and RTK solutions. For the RT-PPP solution, the satellite clock and orbit correction products from Bundesamt für Kartographie und Geodäsie (BKG), Centre National d’Etudes Spatiales (CNES), and the GNSS research center of Wuhan University (WHU) are applied to account for the real-time mode. Moreover, GNSS datasets from two geodetic-grade Trimble R4s receivers are collected; hence, the datasets are processed using the traditional kinematic differential solution to provide a reference solution. The results indicate that this cost-effective multi-GNSS RT-PPP solution can attain positioning accuracy within 1–3 dm, and is thus suitable for a variety of transportation applications, including intelligent transportation system (ITS), self-driving cars, and automobile navigation applications.

1. Introduction

The accurate positioning, velocity, and time (PVT) estimation of a vehicle is essential for intelligent transportation system (ITS) applications. PVT can be effectively determined using global navigation satellite system (GNSS) techniques, which include differential, real-time kinematic (RTK), precise point positioning (PPP), and PPP-RTK solutions. GNSS has some key applications in the vehicle transportation domain, including navigation, vehicle tracking, optimal route determination, and traffic monitoring, and can be used to track road networks and enhance their safety and efficiency. GNSS is commonly used in self-driving cars, parking automation, and advanced driver assistance systems (ADAS), such as pedestrian detection and lane departure warning applications; moreover, GNSS can be integrated with internet of things (IoT) and 5G technologies to foster connectivity, which significantly helps in real-time transportation and monitoring applications.

On the other hand, GNSS has some limitations in urban areas, such as signal blocks and signal losses in tunnels or under bridges; as a result, GNSS accuracy may be degraded. To address this problem, the inertial navigation system (INS) can be combined with GNSS to bridge the gaps during signal outages by tracking a vehicle’s motion and orientation; furthermore, the sensor fusion technique can be utilized by integrating GNSS with an additional sensor, such as an odometer or light detection and ranging (LiDAR). Another possible solution is to apply the visual–inertial odometry (VIO) technique, which combines a high-end camera with the INS to determine vehicle positions and directions when GNSS signals are lost. This technology is widely used in autonomous vehicle navigation domains.

In recent years, numerous studies have investigated the performance of GNSS and other integrating system strategies in both open and urban areas; see, e.g., [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17]. An et al. [1] developed a PPP-RTK model using real-time state space representation (SSR) corrections, which exhibited instantaneous ambiguity resolution and rapid convergence time; their model has been examined for vehicular transportation applications. Their PPP-RTK solution had a precision of 0.1 m, except in the case of signal outages due to bridge passages; additionally, about 92% and 94% of the PPP-RTK estimates converged within 5 s in the east and north components, respectively. Furthermore, Li et al. [10] proposed a multi-sensor fusion system for vehicle navigation based on PPP-RTK, INS, and an odometer along with vehicle motion constraints (VMCs); hence, the collected datasets were tightly coupled into a Kalman filter. Their developed system has been explored in complex urban scenarios, and the positioning errors were at the centimeter level. In addition, Li et al. [11] created an RTK/LiDAR/INS integrated system that focused on enhancing the RTK in GNSS-denied environments; thereafter, their developed system was examined in a road test. The proposed system enhanced RTK performance in GNSS-denied areas; however, LiDAR data processing and the use of a particle filter increased the system’s computation time, resulting in RTK output latencies.

Notably, these techniques are costly because they depend on geodetic-grade receivers and high-end sensors; hence, several researchers have used low-cost GNSS modules and sensors in GNSS transportation applications [18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37]. For example, Vana and Bisnath [20] developed a navigation system that consisted of a triple-frequency low-cost GNSS receiver, a micro-electromechanical (MEMS)-based inertial measurement unit (IMU), and a patch antenna; their system was assessed in urban and suburban areas. Their developed system demonstrated a positioning accuracy at the decimeter level, as well as an overall root mean square error (RMSE) that was 10-40% superior to that obtained with the dual-frequency PPP/IMU solution. Amalfitano et al. [26] introduced a low-cost Internet of Things (IoT) GNSS device that delivered commercial PPP-RTK corrections; their configuration was explored in the moving vehicular domain. Their device obtained horizontal and vertical positioning accuracy at the decimeter level. Additionally, Elmezayen and El-Rabbany [31] presented a low-cost triple-constellation GNSS/INS system using a tightly coupled scheme for precise land vehicular applications. Their system was examined through three integration systems, and an adaptive Kalman filter was applied to remove the measurement outliers. Thereafter, the system was examined in three kinematic trials, showing that sub-meter- to meter-level positioning accuracy was achieved in the three components.

Furthermore, Yang et al. [33] developed a new multiple-constraint-fused Kalman-based navigation (MCKN) method for low-cost GNSS/INS integration in urban areas to overcome accumulated errors due to GNSS outages. Their method showed high-level performance in denied areas. Barbosa et al. [37] proposed an affordable GNSS solution for tracking vehicles and identifying areas of the road where speed limits are exceeded. For this purpose, a system was developed using two low-cost GNSS modules to determine GNSS speed; furthermore, a smartphone was utilized to record the speedometer on video. Their proposed system showed great potential in automatic speed control and monitoring applications.

These studies demonstrate that both RTK and PPP-RTK GNSS techniques have been used widely; nevertheless, these solutions are costly because they require two-receiver configurations or commercial corrections from a GNSS reference station network. As a result, the key innovation of our study is the proposal of a cost-effective real-time precise point positioning (RT-PPP) solution for transportation applications involving an affordable GNSS module and instantaneous satellite orbit and clock correction products. The main reason for choosing the RT-PPP scenario in our study is the fact that it employs only a one-receiver configuration; in addition, real-time satellite orbit and clock products are freely available from different international GNSS service (IGS) analysis centers, such as the Bundesamt für Kartographie und Geodäsie (BKG), Centre National d’Etudes Spatiales (CNES), and the GNSS research center of Wuhan University (WHU). Furthermore, these products are available in the instantaneous SSR correction format.

At present, instantaneous clock and orbit products are widely utilized in various applications, including positioning, maritime, surveying, tropospheric, and ionospheric modeling applications. Recently, several studies have examined the accuracy of these products in different scenarios [38,39,40,41,42,43,44,45,46,47,48,49]. For example, Yao et al. [45] explored the accuracy of estimated real-time zenith tropospheric delay using GPS, Galileo, and BeiDou RT-PPP solutions based on SSR products from various centers. They revealed that the derived zenith tropospheric delay values using WHU satellite orbit and clock products had the highest accuracy. Vidal et al. [46] utilized a cost-effective real-time GNSS system for geodynamic applications, demonstrating its efficiency in improving positioning accuracy.

To achieve the objective of our study, we acquired multi-constellation GNSS observations utilizing an affordable ZED-F9P GNSS device within a moving car in a suburban environment; then, the datasets were processed using the RT-PPP processing model under various scenarios: GPS, GPS/GLONASS, GPS/Galileo, and GPS/GLONASS/Galileo RT-PPP. Instantaneous SSR corrections from BKG, CNES, and WHU ACs were used to adapt the real-time mode. Additionally, a conventional RTK solution was applied with the aforementioned GNSS satellite system combination. To explore the accuracy of our proposed solutions, position estimates obtained using both the RT-PPP and RTK techniques were validated against those obtained with the differential GNSS kinematic method. Finally, statistical analysis of the obtained position differences was performed.

2. Materials and Methods

This section discusses the development of the mathematical formulas for the RT-PPP processing model; additionally, it discusses the implementation of the kinematic survey campaign.

2.1. Mathematical Model of Multi-GNSS Kinematic RT-PPP

The Basic GNSS observation formulas for both the code range and carrier phase can be expressed mathematically [50]:

$$P_i^J={\rho }_r^J+cd{t}_{r,J}-c{dt}^J+{\displaystyle \frac{f_i^2}{f_0^2}}{I}_i^J+T_r^J+{\epsilon }_{p,i}^J$$

(1)

$${\mathsf{\Phi }}_i^J={\rho }_r^J+cd{t}_{r,J}-c{dt}^J-{\displaystyle \frac{f_i^2}{f_0^2}}{I}_i^J+T_r^J+{\lambda }_i{N}_i+{\epsilon }_{\mathsf{\Phi },i}^J$$

(2)

where $i$, $J$, and $r$ denote the signal frequency, GNSS satellite, and receiver, respectively; $P$ and $\mathsf{\Phi }$ are the code range and carrier phase measurements, respectively; $\rho$ defines the satellite-receiver geometric range; $c$ represents the speed of light in a vacuum; $d{t}_{r,J}$ is the receiver clock error; ${dt}^J$ represents the satellite clock errors; $f_i$ represents the carrier phase frequency; $f_0$ is the basic GNSS frequency ($f_0$ = 10.23 MHz); $I_i^J$ represents the ionospheric delay; $T$ is the tropospheric delay; ${\lambda }_i$ is the carrier phase’s wavelength; $N_i$ is the real-value ambiguity term; ${\epsilon }_{p,i}$ and ${\epsilon }_{\mathsf{\Phi },\mathrm{i}}$ are the noise in the code and carrier.

It is well known that ionospheric delay is the dominant error source for GNSS measurements; therefore, ionosphere-free linear combinations with both the code and the carrier are used to remove the ionospheric delay. Furthermore, a precise point positioning processing model is applied depending on the ionosphere-free linear combination, in addition to other corrections, such as using accurate satellite orbit and clock parameters.

The fundamental multi-GNSS ionosphere-free dual-frequency real-time precise point positioning mathematical model can be defined as follows [50]:

$$P_3^G={\rho }_r^G+c\left(d{t}_{r,G}-{dt}^G\right)+c\left(b_{r,G,3}-b_3^G\right)+T_r^G+{\epsilon }_{p,3}^G$$

(3)

$${\mathsf{\Phi }}_3^G={\rho }_r^G+c\left(d{t}_{r,G}-{dt}^G\right)+c\left({\delta }_{r,G,3}-{\delta }_3^G\right)+T_r^G+{\lambda N}_3^G+{\epsilon }_{\mathsf{\Phi },3}^G$$

(4)

$$P_3^J={\rho }_r^J+c\left(d{t}_{r,G}-{dt}^J\right)+c\left(b_{r,G,3}-b_3^J\right)+T_r^J+{ISB}^J+{\epsilon }_{p,3}^J$$

(5)

$${\mathsf{\Phi }}_3^J={\rho }_r^J+c\left(d{t}_{r,G}-{dt}^J\right)+c\left({\delta }_{r,G,3}-{\delta }_3^J\right)+T_r^J+{ISB}^J+{\lambda N}_3^J+{\epsilon }_{\mathsf{\Phi },3}^J$$

(6)

where $P_3$ and ${\mathsf{\Phi }}_3$ indicate the ionosphere-free pseudorange and phase linear combinations, respectively; $G$ is the GPS satellite system; $J$ refers to other navigation systems (i.e., GLONASS or Galileo); $b_{r,G,3}$ denotes the ionosphere-free receiver differential code bias (DCB); ${\delta }_{r,G,3}$ represents the ionosphere-free receiver differential phase bias (DPB); $b_3^G$ and $b_3^J$ are the ionosphere-free satellite differential code biases for GPS and other satellite systems, respectively; ${\delta }_3^G$ and ${\delta }_3^J$ refer to the ionosphere-free satellite differential phase biases for GPS and other navigation systems, respectively; ${ISB}^J$ corresponds to the inter-system bias (ISB), which includes the difference in receiver DCB between the GPS and other satellite systems; ${\lambda N}_3^G$ and ${\lambda N}_3^J$ represent the ionosphere-free real-value ambiguity terms for GPS and GNSS, respectively.

Regarding the RT-PPP processing model, the satellite and receiver DCBs are grouped into corresponding clock parameters; in addition, the satellite clock parameter is corrected using real-time satellite clock products. Hence, the reformulated RT-PPP model can be written as follows:

$$P_3^G={\rho }_r^G+c\widetilde{d{t}_{r,G}}-c{dt}^{G,corr}+T_r^G+{\epsilon }_{p,3}^G$$

(7)

$${\mathsf{\Phi }}_3^G={\rho }_r^G+c\widetilde{d{t}_{r,G}}+T_r^G+\overline{{\lambda N}_3^G}+{\epsilon }_{\mathsf{\Phi },3}^G$$

(8)

$$P_3^J={\rho }_r^J+c\widetilde{d{t}_{r,G}}-c{dt}^{J,corr}+T_r^J+{ISB}^J+{\epsilon }_{p,3}^J$$

(9)

$${\mathsf{\Phi }}_3^J={\rho }_r^J+c\widetilde{d{t}_{r,G}}+T_r^J+{ISB}^J+\overline{{\lambda N}_3^J}+{\epsilon }_{\mathsf{\Phi },3}^J$$

(10)

where $\widetilde{d{t}_{r,G}}$ refers to the GPS receiver clock error plus the receiver DCB (i.e., $d{t}_{r,G}+b_{r,G,3}$); ${dt}^{G,corr}$ and ${dt}^{J,corr}$ represent the corrected satellite clock parameter; and $\overline{{\lambda N}_3^G}$ and $\overline{{\lambda N}_3^J}$ are the ambiguity terms for GPS and other satellite navigation systems, respectively, taking the following formula:

$$\overline{{\lambda N}_3^G}={\lambda N}_3^G+\left(b_{r,G,3}+{\delta }_{r,G,3}\right)-\left(b_3^G+{\delta }_3^G\right)$$

(11)

$$\overline{{\lambda N}_3^J}={\lambda N}_3^J+\left(b_{r,J,3}+{\delta }_{r,J,3}\right)-\left(b_3^J+{\delta }_3^J\right)$$

(12)

The tropospheric delay factor is divided into two parts: hydrostatic delay and wet delay. The hydrostatic tropospheric component is determined using an empirical tropospheric model, but the wet tropospheric component is estimated as an unknown parameter; thus, the vector of the estimated variables ($\overrightarrow{\mathit{X}}$) can be expressed as follows:

$$\overrightarrow{\mathit{X}}={\left[∆x, ∆y, ∆z, c{dt}_r^{~}, T_w,{ ISB}^J, \overline{N_3^G}, \overline{N_3^j}\right]}^T$$

(13)

where $∆x, ∆y,$ and $∆z$ indicate the receiver’s coordinate corrections, and $T_w$ denotes the wet tropospheric delay parameter.

2.2. Kinematic RT-PPP Campaign Setup

To explore the accuracy of our proposed affordable multi-GNSS RT-PPP scenario for vehicular applications, a kinematic survey campaign was executed in a suburban environment in New Aswan City, Egypt, on day of the year (DOY) 60 in 2024. In our kinematic trial, we used three Trimble R4s geodetic receivers; two were mounted on the vehicle with the cost-effective ZED-F9P device (Figure 1a), while the other was located at the base station (Figure 1b). Figure 2 illustrates the vehicular trajectory layout in which the multi-GNSS datasets were collected using the ZED-F9P module over a period of 1 h with 1 s observation intervals. The number of tracked GNSS satellites and the position dilution of precision (PDOP) values are plotted in Figure 3. The tracked satellite number significantly increases when the combined GPS/GLONASS/Galileo is used, reaching about 20 satellites on average; meanwhile, the PDOP is dramatically improved.

The datasets were then processed employing a variety of processing strategies, including RT-PPP and RTK using GPS, GPS/GLONASS, GPS/Galileo, and GPS/GLONASS/Galileo observations. Although our ZED-F9P is a dual-frequency multi-GNSS device, only single-frequency BeiDou observations were gathered; as a result, we exclude BeiDou observations from our proposed solutions. To consider the orbit and clock errors, the instantaneous BKG, CNES, and WHU SSR correction products were applied [51]; conversely, quad constellation broadcast ephemeris (i.e., BRDM) products [52] were adopted for the RTK processing model.

To develop a reference solution scenario, GNSS measurements were collected through two geodetic-grade Trimble R4s rover devices synchronous with one base receiver located at a reference station; hence, the datasets are processed using the traditional differential solution. The attained RT-PPP and RTK positioning solutions from the ZED-F9P module (i.e., E_u and N_u) were compared with the differential kinematic GNSS position estimates from the two geodetic receivers utilizing the configuration and formula provided in Figure 4. The RT-PPP, RTK, and differential processing solutions were obtained through the Net-Diff GNSS V. 1.13 software package [53]. Tropospheric delay was estimated using the Saastamoinen tropospheric model [54] and dry/wet Vienna mapping functions (VMF1) [55]. In addition, a Kalman filter was adopted for the parameter estimation procedure. It should be pointed out that the ambiguity float solution was adopted in our processing strategies. A summary of each processing strategy is provided in

Table 1. Parameters of data processing strategy.

Parameter	Processing Strategy
	RT-PPP	RTK	Differential
System	GPS, GLONASS, Galileo
Frequency	GPS: L1/L2; GLONASS: G1/G2; Galileo: E1/E5b
Mathematical model	Undifferenced	Differenced	Differenced
Observation interval	1 HZ
Elevation angle	10 degrees
Orbits and clocks	BKG, CNE, WHU	BRDM	IGS-Final
Tropospheric modeling	Saastamoinen model + VMF
Parameter estimation	Kalman filter

; in addition, a flowchart illustrates the processing and evaluation scheme in Figure 5. It should be mentioned that the computed coordinates are part of the projected WGS84/UTM36N system.

3. Results

To analyze the accuracy of our proposed multi-GNSS RT-PPP solutions for transportation applications, different processing scenarios were applied: GPS (G), GPS/GLONASS (GR), GPS/Galileo (GE), and GPS/GLONASS/Galileo (GRE). Furthermore, three instantaneous SSR corrections were applied, including BKG, WHU, and CNE products. The real-time SSR products were used to correct the satellite orbit and clock values. Furthermore, multi-GNSS RTK solutions were explored for transportation applications, which also included G, GR, GE, and GRE; consequently, the accuracy of around 16 solutions is under investigation in our study, as indicated in

Table 2. Proposed processing solutions and their symbols.

GNSS	RT-PPP			RTK
	BKG	CNE	WHU
G	RT-BKG-G	RT-CNE-G	RT-WHU-G	RTK-G
GR	RT-BKG-GR	RT-CNE-GR	RT-WHU-GR	RTK-GR
GE	RT-BKG-GE	RT-CNE-GE	RT-WHU-GE	RTK-GE
GRE	RT-BKG-GRE	RT-CNE-GRE	RT-WHU-GRE	RTK-GRE

. The derived UTM coordinates from all solutions were validated against the kinematic differential solution.

For vehicular transportation applications, horizontal positioning accuracy is more important than vertical accuracy; thus, 2D positioning accuracy is primarily explored in our work. Figure 6 illustrates the horizontal positioning errors for the RT-PPP solutions using different SSR corrections and RTK solutions; the RT-PPP models converge to a few decimeters, while the RTK models converge to less than one decimeter. For the RT-BKG processing scenarios, the RT-BKG-GE converges faster than the other solutions, reaching less than 0.1 m; on the other hand, the other three solutions converge to 0.2 m. It can also be seen that the RT-BKG-G has a slightly longer convergence time.

Likewise, compared to the other RT-CNE scenarios, the RT-CNE-GE technique has a faster convergence time. In addition, there are some variations in the 2D positioning error values for the RT-CNE-G model; there is also an obvious increment in the positioning errors for the three models at the end of the observation window, particularly the RT-CNE-GR model. To the contrary, the RT-WHU-GR solution’s performance improves along with the RT-WHU-GE technique, which also demonstrates rapid convergence. Moreover, the RT-WHU-G technique has a longer convergence, approaching 0.2 m. Thus, the RTK solutions are clearly superior because the positioning errors dramatically decrease to less than 0.02 m, with some fluctuations reaching 0.04 m.

To visualize the positioning performance of the proposed RT-PPP and RTK scenarios, the cumulative distribution function (CDF) of the two-dimensional position discrepancies for each scenario is plotted (Figure 7). It can be seen that, for the RT-BKG scenarios, about 90% of the 2D position discrepancies are less than 0.22 m for the RT-BKG-GR, RT-BKG-GE, and RT-BKG-GRE solutions, with slight superiority for the RT-BKG-GRE solution. However, they dramatically increase to 0.4 m for the RT-BKG-G solution. On the other hand, the two-dimensional errors are within 0.2 m, 0.3 m, 0.42 m, and 0.5 m for 90% of the time using the RT-CNE-GE, RT-CNE-GRE, RT-CNE-G, and RT-CNE-GR solutions, respectively, with obvious superiority in the RT-CNE-GE scenario. The position differences in the horizontal component for the RT-WHU processing strategy are within 0.3 when the GR, GE, and GRE solutions are used; nevertheless, they are approximately 0.38 m in the GPS solution. The outperformance of the RTK scenarios is notable because 90% of the horizontal positioning errors are smaller than 0.02 m.

As another analysis perspective, it can be seen that a four-decimeter-level horizontal positioning accuracy (i.e., 0.4 m) can be accomplished about 90% of the time when using the BKG GPS instantaneous satellite orbit and clock correction products. Conversely, this accuracy is achieved about 95% of the time when using the BKG GPS/GLONASS SRR corrections. Moreover, this percentage increases to 98% for both the GPS/Galileo and GPS/GLONASS/Galileo scenarios. For the GPS CNE orbit and clock products, 85% of the RT-PPP solution outputs achieve a 0.4 m positioning accuracy; however, this percentage drastically decreases to 65% when the GPS and GLONASS correction products are both employed. Meanwhile, the CNE Galileo SSR products significantly increased the percentage to about 98% of the position estimates with 0.4 m horizontal accuracy. Likewise, the WHU Galileo orbit and clock corrections fostered two-dimensional positioning accuracy when they were applied to the GPS and GPS/GLONASS RT-PPP solutions, as about 98% of the 2D estimates were less than 0.4 m. Furthermore, both the WHU GPS and GLONASS SSR corrections are accurate enough that about 95% of the estimates have four-decimeter horizontal accuracy.

In general, the Galileo SSR corrections from the three analysis centers are more accurate than the GPS and GLONASS SSR corrections; furthermore, GLONASS real-time orbit and clock products have some limitations regarding their accuracy.

A histogram of the two-dimensional positioning differences is presented in Figure 8 to evaluate the accuracy of the proposed RT-PPP and RTK approaches. When the RT-BKG solution is applied, the 2D error range is less than 0.6 m for the RT-BKG-G solution and less than 0.45 m for the RT-BKG-GR solution; furthermore, it is lower than 0.35 m and 0.3 m for the RT-BKG-GE and RT-BKG-GRE solutions, respectively. Furthermore, it can be seen that the 2D error is under 0.3 m in 66%, 96%, 94%, and 98% of the epochs for the RT-BKG-G, RT-BKG-GR, RT-BKG-GE, and RT-BKG-GRE solutions, respectively.

For RT-CNE, the extent of the two-dimensional discrepancy is no more than 0.6 m for both the RT-CNE-G and RT-CNE-GR solutions; on the other hand, it substantially decreases to under 0.25 m and 0.3 m for the RT-CNE-GE and RT-CNE-GRE scenarios, respectively. In the same context, around 67%, 48%, 99%, and 91% of the 2D discrepancies are less than 0.3 m when the RT-CNE-G, RT-CNE-GR, RT-CNE-GE, and RT-CNE-GRE scenarios are adopted, respectively.

For the RT-WHU solution, the horizontal positioning error distribution is below 0.4 m for all scenarios, and 2D errors are less than 0.3 m approximately 63% and 86% of the time using the RT-WHU-G and RT-WHU-GR strategies, respectively. However, these percentages increase to 91% and 92% when the RT-WHU-GE and RT-WHU-GRE strategies are employed, respectively.

In the RTK scenario, the spread of two-dimension positioning differences is less than 0.035 m and 0.03 m for the RTK-G and RTK-GR solutions, respectively, while it is no more than 0.02 m for both the RTK-GE and RTK-GRE solutions. In addition, only the RTK solutions follow a normal distribution pattern, particularly the RTK-GE and RTK-GRE strategies.

To further examine the performance of the multi-GNSS RT-PPP processing models applying various SSR corrections, a horizontal error box plot is illustrated in Figure 9 for each scenario. For the RT-BKG processing solution, the 2D position errors are widely distributed for RT-BKG-G; by contrast, they are narrowly distributed for RT-BKG-GR, with a medium to narrow distribution for both RT-BKG-GE and RT-BKG-GRE. To provide a robust estimate of the positioning differences and their central tendencies, the interquartile range (IQR) is computed for each processing solution; the IQR estimates are 0.255 m and 0.043 m for the RT-BKG-G and RT-BKG-GR solutions, respectively, whereas they are 0.101 m and 0.063 m for the RT-BKG-GE and RT-BKG-GRE models, respectively. Therefore, it can be inferred that the real-time BKG GPS SSR corrections are less accurate than GLONASS and Galileo corrections, which is reflected in the resultant positioning accuracy.

In terms of the RT-CNE solution, the RT-CNE-GR strategy demonstrates a large distribution of two-dimensional errors; in contrast, the other three processing strategies have somewhat equal two-dimensional positioning error distributions. The IQR estimates are 0.149 m, 0.269 m, 0.146 m, and 0.130 m for the RT-CNE-G, RT-CNE-GR, RT-CNE-GE, and RT-CNE-GRE processing strategies, respectively; as a result, it can be concluded that the RT-CNE-GR has the worst performance among the RT-CNE processing strategies. This can be attributed to the accuracy of the instantaneous GLONASS orbit and clock products provided by the CNE analysis center.

The box plot indicates that RT-WHU-GE has a slightly wide horizontal positioning difference distribution compared with the other RT-WHU processing scenarios, while RT-WHU-GR has the narrowest error distribution. The IQR estimates are 0.133 m, 0.103 m, 0.166 m, and 0.088 m for RT-WHU-G, RT-WHU-GR, RT-WHU-GE, and RT-WHU-GRE, respectively. The horizontal positioning errors provided by the RTK scenarios are significantly smaller than those provided by the RT-PPP solutions, with ultra-narrow error distributions and IQR values equal 0.007 m, 0.005 m, 0.004 m, and 0.004 m for RTK-G, RTK-GR, RTK-GE, and RTK-GRE, respectively.

4. Discussion

At present, multi-GNSS positioning techniques are extensively used in the position, navigation, and timing (PNT) domain, being crucial for intelligent transportation system applications. A number of GNSS techniques, including RTK, PPP, and PPP-RTK, are employed for this purpose; however, both RTK and PPP-RTK approaches have the drawback of being expensive due to their dependence on two receivers or one receiver with network-based corrections. Furthermore, the PPP technique is usually applied in the post-processed mode. Consequently, we propose using ultra-low-cost RT-PPP scenarios in transportation applications, utilizing multi-GNSS ZED-F9P observations and SSR corrections from BKG, CNE, and WHU streams.

Our proposed RT-PPP solution position estimates were validated against traditional differential kinematic counterparts, allowing for an overall statistical analysis of the two-dimensional positioning performance, including the mean bias error (MBE), RMSE, and standard deviation (STD). These statistical parameters were computed for the horizontal positioning components; they are summarized in Figure 10 for the RT-PPP and RTK processing strategies.

Figure 10 shows that adding another satellite system (i.e., GLONASS, Galileo, or both) to the GPS solution improves the positioning accuracy, except for the RT-CNE-GR solution. This is due to the accuracy of the GLONASS SSR products obtained through the CNE stream. Additionally, the RT-PPP positioning accuracy is within 1–3 dm with three products; nevertheless, it is within a few centimeters for all RTK solutions, and notable enhancements in the horizontal components are achieved when the GE solution is used in all RT-PPP processing strategies. This is due to the Galileo SSR product’s accuracy, which comes from the three analysis centers. For the RT-BKG processing model, the two-dimensional positioning accuracy is enhanced from 0.262 m to 0.216 m, 0.145 m, and 0.172 m when the G, GR, GE, and GRE solutions are utilized, respectively; in addition, using the GE and GRE solutions improves the horizontal positioning accuracy for the RT-CNE-G processing model from 0.317 m to 0.137 m and 0.227 m, respectively. Furthermore, the RMSE values of the RT-WHU-G solution are reduced from 0.311 m to 0.220 m, 0.173 m, and 0.192 m by RT-WHU-GR, RT-WHU-GE, and RT-WHU-GRE, respectively. For the RTK scenarios, the RMSE values are within the one-centimeter level, superior to the Galileo-based solution. Meanwhile, for the precision investigation, the STD values for all RT-PPP solutions are less than 0.177 m, with outperformance in the GRE scenarios; moreover, the STD values for the RTK models are less than 0.007 m.

The following conclusions can be drawn from the findings for the RT-PPP and RTK solutions:

• A one-to-three-decimeter positioning accuracy level can be obtained using the RT-PPP solution, utilizing both the cost-effective ZED-F9P module and satellite orbit and clock correction products from the BKG, CNE, or WHU streams.
• Incorporating GLONASS, Galileo, or both into the GPS RT-PPP solution improves positioning accuracy.
• The BKG and WHU SSR products increase the GPS/GLONASS RT-PPP two-dimensional positioning accuracy by about 18% and 29%, respectively, compared with their counterparts in the GPS RT-PPP solution.
• The enhancement percentages of the GPS/Galileo RT-PPP’s 2D accuracy using orbit and clock products from BKG, CNE, and WHU are around 18%, 57%, and 44%, respectively, compared with their counterparts in the GPS RT-PPP solutions.
• Compared with their counterparts from the GPS RT-PPP solutions, satellite orbit and clock products from the BKG, CNE, and WHU centers enhance the GPS/GLONASS/Galileo RT-PPP positioning accuracy by about 34%, 28%, and 38%, respectively.
• BKG orbit and clock products offer the best solutions for the GPS and GPS/GLONASS/Galileo RT-PPP solutions, while CNE SSR products are superior for the GPS/Galileo RT-PPP solutions, and both BKG and WHU demonstrate similar performance for the GPS/GLONASS RT-PPP solution.
• The GPS-based RT-PPP solutions from the three BKG, CNE, and WHU SSR products require a long time to converge compared with the other solutions, which could be a limitation of our suggested solutions. For this reason, the RT-PPP ambiguity resolution solution will be considered in our future work to improve the convergence time.
• Based on the attained positioning accuracy and compared with the common PPP, RTK, and PPP-RTK solutions, our proposed RT-PPP approach is a cost-effective solution, particularly when compared with the geodetic-grade PPP solution; in addition, its configuration is less complicated than the RTK and PPP-RTK solutions.
• Finally, our processing scenario is based on a suburban context, and continuous SSR corrections are necessary for a successful solution.

5. Conclusions

We evaluated the accuracy of a multi-GNSS real-time precise point positioning solution employing both a cost-effective ZED-F9P device and instantaneous BKG, CNE, and WHU orbit and clock products for transportation applications. A vehicular trajectory was implemented in a suburban environment with a setup using rover receivers including the ZED-F9P device and two geodetic-grade Trimble R4s receivers, as well as another Trimble R4s receiver statically located at a reference station. The collected datasets from the ZED-F9P module were processed using GPS, GPS/GLONASS, GPS/Galileo, and GPS/GLONASS/Galileo RT-PPP processing scenarios as well as RTK. The differential GNSS kinematic solutions formed by the two rover geodetic devices and the base one were used as references. The obtained RT-PPP and RTK processing solutions were validated against their counterparts in the differential kinematic solutions. The low-cost RT-PPP solutions achieved better decimeter-level positioning accuracy than the BKG SSR-based RT-PPP solutions. The attained positioning accuracy is appropriate for cost-effective autonomous vehicles, intelligent transportation systems, and urban transportation applications.