1. Introduction
With the development of Global Navigation Satellite System (GNSS) in the 21th century, the number of available navigation satellites has increased significantly. It has formed a multi-system global navigation constellation framework mainly composed of American Global Positioning System (GPS), Russian Global Navigation Satellite System (GLONASS), Chinese BeiDou Navigation Satellite System (BDS) and EU’s Galileo system [
1]. Moreover, Japanese Quasi-Zenith Satellite System (QZSS) is widely used to supplement GPS services as a regional navigation satellite system. The development of GNSS also promotes the continuous development of location-based service business based on smartphones, and greatly facilitates both industrial production and daily life.
For a long time, only rough positioning results could be obtained from the smartphones. Launched in 2016, the Android 7.0 operating system supports the access to GNSS raw measurements and navigation messages [
2], making the Android devices function more like a GNSS receiver. In the same year, Google released GNSSLogger, an open-source program that could help retrieve GNSS raw observations from Android smartphones, including the observations of code pseudorange, carrier phase, and Doppler [
3]. In 2017, Geo++ released an application named Geo++ RINEX Logger to provide GNSS raw observation directly in RINEX format [
4]. Both of the above two applications cannot output the information of navigation messages. RinexON, released by Flamingo team at the end of June 2018, is able to output multi-GNSS raw data, including broadcast ephemeris, collected by smartphones in RINEX 3.0.3 format [
5]. The accessibility of GNSS raw data makes it possible to analyze the observation quality and to study the positioning algorithm with Android terminals.
In recent years, GNSS positioning with Android terminals has become one of research focuses. The experiment by Gim et al. [
6] is a Single Point Positioning (SPP) test with code measurements of the Nexus 9 tablet, and the results show that the RMS of positioning errors in horizontal and three-dimensional (3D) are 3.05 m and 3.82 m. In an experiment of double-differenced positioning with single-frequency carrier phases from the tablet and several base stations, the positioning accuracy better than 20 cm can be achieved within 20 min [
7]. The carrier-to-noise ratio (C/N0) value of GNSS raw observations collected by the Nexus 9 tablet is 10 dBHz lower than the representative values obtained from a geodetic-quality antenna and receiver. With time-differenced filtering method, horizontal and vertical accuracies of static positioning can be better than 0.6 and 1.4 m, respectively [
8]. Martin [
9] also used a Nexus 9 tablet for positioning test. The research shows that multipath plays an important role for the expected accuracy of the calculated precise positions, both due to the induced error on the measurements, and due to loss of lock of the GNSS signals, which significantly affects precise positioning from carrier phase measurements. Although these studies have important implications for subsequent experiments using smartphones, the positioning of ordinary smartphones is not comparable to this tablet.
Many scholars have carried out differential GNSS researches on smartphones. Zhang et al. [
10] developed an Android application based on wide area differential location technology, and the static observation results show that the horizontal accuracy of the smartphones can reach about 4 m. The maritime test by Specht et al. [
11] showed that the accuracy of the dynamic positioning with smartphones during vessel maneuvering can reach 10 m, satisfying most of the maritime requirements for navigation accuracy. The Network Real Time Kinematic (NRTK) positioning accuracy with smartphones by Dabove et al. [
12] is about 60 cm. Moreover, Wanninger et al. [
13] performed carrier phase ambiguity fixing for smartphones. With ambiguities successfully fixed, the 3D positioning accuracies (standard deviations) better than 4 cm could be achieved after five minutes of static observation session, and an accuracy of 2 cm is possible for long observation sessions.
Since the launch of Xiaomi 8 in June 2018 [
14], smartphones supporting dual-frequency GPS signals have become the mainstream of the market and motivate the research of smartphones Precise Point Positioning (PPP). The quality analysis results of GNSS raw observations show that the number of visible satellites with Xiaomi 8 is similar to that with geodetic receivers, although the carrier-to-noise ratio and multipath effect with smartphones are worse than the typical values with geodetic receivers [
15]. It is found that carrier phase measurements collected by smartphones might contain gross errors and systematic errors, and that different clocks are used for code and carrier phase observations [
16]. By considering the clock bias between the code and carrier phase measurements, the accuracy of PPP can be better than 1 m [
17]. Shi et al. [
18] conducted static and dynamic observation experiments with Samsung S8, Huawei Mate20 and Xiaomi8. After experiments, he evaluated the GNSS data quality of smartphones in detail. Through proper GNSS data quality control, he initially achieved positioning accuracy within 1 m. Wu et al. [
19] conducted long-term static observation with smartphones. The PPP results show that the positioning accuracy of smartphone with dual frequency data is better than 20 cm, but it takes up to 100 min to converge. These studies confirm the practicability of using GNSS raw data of smartphones to realize PPP, which is of great significance for subsequent PPP studies using smartphones.
According to the above published researches, the difficulties of precise positioning with smartphones were found. Compared with geodetic receivers, smartphones are prone to suffer from frequent losses of lock, unstable clock and poor quality of measurements, due to the relatively low-cost GNSS chips and antennas [
15,
16,
17,
18]. Therefore, the GNSS raw observations collected by smartphones are likely to contain a large number of cycle slips, and to be seriously affected by multipath effects and low carrier-to-noise ratio, jeopardizing the PPP accuracies with smartphones. To overcome the disadvantages of smartphones in GNSS data collection, it is necessary to quantitatively analyze the quality of observations of smartphones and to study the method of cycle slip detection.
Although some types of smartphones are designed with the nominal capabilities of tracking dual-frequency GNSS signals, the real measurements are usually far from enough for dual-frequency PPP processing all through the day with the traditional method, due to various reasons such as the frequent interruptions of dual-frequency data. Thus, the single-frequency PPP is still a common processing mode for smartphones, and the ionospheric delay are corrected by the Global Ionospheric Maps (GIM), the products of which can correct about 80% of the ionospheric delay [
20]. Although the remaining ionospheric delay can still reach decimeter level, it is sufficient for the single-frequency PPP with smartphones, considering that the precision of code pseudorange measurements from smartphones is nearly 10 m [
18]. In theory, GIM can be used to build constraint equations, to improve the reliability of single-frequency PPP processing of smartphones and shorten the convergence time [
21]. However, there is a lack of research on the impact of ionosphere constraints on the smartphones PPP.
This paper contributes to GNSS data quality analysis and PPP with Android smartphones.
Section 2 represents the principles and methodologies of single-frequency PPP, smoothing code pseudorange and cycle slip detection, in special consideration of the characteristics of data collected by smartphones. In
Section 3, the quality of GNSS measurements from two different Android devices and one geodetic receiver are analyzed and compared from the aspects of satellite tracking performance, carrier-to-noise ratio and multipath effects, followed by PPP experiment with GNSS Analysis Software for Multi-constellation and Multi-frequency Precise Positioning (GAMP) [
22]. Conclusions are given in
Section 5.
3. GNSS Data Quality Analysis
The main device used in this experiment is a Huawei Mate30 smartphone (hereinafter referred to as Mate30). For comparison analysis, a Huawei honorV20 smartphone (hereinafter referred to as V20) and a geodetic receiver (Trimble R8) were also used. The Mate30 smartphone is a dual-frequency GNSS smartphone that collects the first frequency signals of GPS, GLONASS, BDS, Galileo and QZSS, and the second frequency signals of GPS, Galileo and QZSS. Although the V20 smartphone is cheaper than the Mate30 smartphone, it also supports dual-frequency observations and all of the five systems. Listed in
Table 1 are the GNSS related characteristics of the three devices used in this paper.
In the experiment, Trimble R8 is used as the reference, and its antenna is only about 10 cm away from the two smartphones. To compare the observing positioning performances of the two smartphones with those of the geodetic receiver, synchronous observations were conducted with the three devices. Data was collected on the evening of 13 November, the afternoon and the evening of 18 November 2019, and each of the three observing periods lasted for 2–3 h. Since the devices were equipped at almost the same place during the three observing periods, the difference of the surroundings could be neglected.
3.1. Satellite Tracking
Shown in
Figure 4 are the numbers of different GNSS satellites with the signal of the first frequency tracked by the geodetic receiver and the two smartphones. The Mate30 smartphone is able to track about 38 satellites in total, while the V20 smartphone tracks only 30 ones. Since the GNSS chip of the Mate30 is nominally supporting BDS-3 signals, it can track more BDS satellites than the V20, which can only track BDS-2 satellites. Compared with the geodetic receiver, both of the two smartphones suffer from frequent fluctuations in the numbers of visible satellites, although they can track as many satellites as, if not more than, the geodetic receivers.
Shown in
Figure 5 are the numbers of satellites with dual-frequency signals tracked by the Mate30, the V20 and the Trimble R8. The number of satellites with dual-frequency signals tracked by the Mate30 smartphone fluctuates between 3 and 10, while the number of satellites with dual-frequency signals tracked by the V20 smartphone fluctuates between 1 and 9. The dual-frequency measurements collected by either the Mate30 or the V20 are far from sufficient for continuous PPP experiment with the ionosphere-free combination of L1/L5 or E1/E5A observables, and the case might be worse in real scenarios with urban canyons.
The data continuities of GPS, Galileo and QZSS are also compared and the results are shown in
Figure 5. The number of GPS, Galileo and QZSS dual-frequency satellites of two smartphones fluctuates seriously. And in a certain period of time, the dual-frequency satellite of Galileo cannot be observed by the two smartphones. Considering that QZSS system is a regional navigation satellite amplification system developed by Japan, adopting the Inclined Geosynchronous Orbit (IGSO), the precision of positioning service provided is limited [
22]. Therefore, the reliable dual-frequency measurements collected by the smartphones are still mainly from GPS satellites at present.
3.2. Carrier-to-Noise Ratio
The carrier-to-noise ratio refers to the ratio of the average power of the carrier signal received at the receiver end to the average power of the noise when the signal is interfered in the process of propagation. The carrier-to-noise ratio reflects the noise level of the measurement [
28]. The higher the carrier-to-noise ratio is, the better the observation quality is.
Shown in
Figure 6 are the mean values of carrier-to-noise ratio in three experiments. Generally, the mean carrier-to-noise ratio with the geodetic receiver is the highest among all of the three devices, although for several BDS satellites the mean carrier-to-noise ratio with the Mate30 smartphone is the highest. The carrier-to-noise ratios with the Mate30 smartphone are around 40 dBHz, obviously better than those with the V20 smartphone. This may be due to better antenna and GNSS chip of the Mate30 smartphone.
Figure 7 shows the typical relationships between the carrier-to-noise ratio and the elevation angle for the three devices. The carrier-to-noise ratio with the geodetic receiver increases as the elevation angle increases, while for the two smartphones, the correlation between carrier-to-noise ratio and elevation angle is not obvious. This could be explained by the different polarization modes. Instead of the right-handed circular polarization, the linear polarization is adopted by the built-in GNSS antennas of the smartphones. Therefore, the smartphones are more vulnerable to signal interferences.
In addition, through the comparison between the carrier-to-noise ratios of L1 and L5 observations collected by the smartphones, it is found that the carrier-to-noise ratios of the L5 observations are equivalent to or even better than those of the L1 observations in the cases of low elevations. It can also be inferred that theL5 signal outperforms the L2 signal in anti-jamming in the cases of low elevations. However, the average carrier-to-noise ratios of the L5 signals received by the smartphones are around 4 dBHz lower than those of the L1 signals. This might be due to the imperfect multi-frequency antenna design of the smartphones, and further study is needed.
3.3. Multipath Effect
The propagation direction, amplitude and phase of GNSS signal are prone to be affected by the reflections of the surrounding at the antenna, and the reflected signals can cause multipath effects [
29]. The observing environments for smartphone users are complicated, and it is inconvenient for the antenna of a smartphone to suppress the multipath error through hardware like the choke ring. Therefore, the multipath error usually plays a dominant role in the code measurement collected by smartphone [
9]. Since the multipath error of code measurement is as 200 times large as that of carrier phase [
30,
31], we focus on the multipath error of the code measurement in the study.
When dual-frequency observations are available, the code multipath errors can be studied with the multipath combination, which can be expressed as
where the subscripts
and
denote different frequency bands, and
is the frequency value. The combination expressed by Equation (5) mainly contains the multipath error of corresponding code measurement and the linear combination of the ambiguities. If no cycle slip occurs, the ambiguities are considered constants and can be removed through averaging over epochs [
32]. The subscripts for receiver and satellite have been omitted here for simplicity.
Shown in
Figure 8 are the standard deviations of L1 frequency code multipath error of the three devices in the three experiments. The standard deviations of code multipath errors of the Mate30 and the V20 are as ten times large as that of the Trimble R8. This phenomenon shows that the two smartphones have disadvantages in suppressing the multipath effects.
Figure 9 and
Figure 10 show the typical standard deviations values and the multipath errors time series of L1, L5 or L2 code measurements with the three devices. It can be seen that the code multipath errors of the two smartphones vary from −4 and 4 m, and that the code measurements with the geodetic receiver are less affected by the multipath error. It suggests that the multipath error should be one of the main factors limiting the accuracy of PPP with the smartphones. Meanwhile, the standard deviationsof the code multipath errors of L1 and L5 code measurements are about 2 m for smartphones, and L5 code measurements are generally less affected by multipath errors than L1 code measurements.
4. PPP Results
In this paper, a single-frequency PPP processing strategy based on ionosphere constraints is employed. Shown in
Table 2 are the setting details of the single-frequency PPP for smartphones. The products of satellite orbit, satellite clock and Earth rotation are downloaded from MGEX data center (
http://www.cddis.gsfc.nasa.gov/). Ionospheric delay products are downloaded from CODE. And ionospheric delay is constrained by GIM products of CODE.
In order to compare the positioning results with the different strategies and to validate the abovementioned method, three experiments were conducted with different settings. The details of these projects are shown in
Table 3.
The standard deviations of positioning errors of the Mate30 using different projects are shown in
Table 4. Considering that the horizontal position of smartphones is more widely used [
10,
11,
12,
13], the horizontal accuracy of smartphones was only recorded and analyzed. The standard deviations of positioning errors in the East and North direction of the Mate30 using the SPP and traditional PPP projects are about 2–4 m. And that of the improved PPP project is less than 1 m.
Table 4 show that the positioning accuracy of the Mate30 using the improved PPP project is obviously improved compare with the other two projects.
The time series of horizontal positioning errors are shown in
Figure 11. When the improved PPP model is employed, the positioning errors in E and N directions during all of the three time periods can converge to less than 1 m and stay relatively stable. With the SPP and traditional PPP model, the positioning errors in E and N directions exceed 2 m, and the positioning results are unstable. In general, the performance of SPP is highly related to the data quality of code pseudorange. The positioning accuracies with the traditional PPP and SPP models are at the same level, although precise orbit and clock products are adopted in the PPP model. The poor performance of the traditional PPP in the above experiments could be explained from two aspects. On one hand, the quality of the code measurements collected by the Mate30 is poor. One the other hand, frequent loss of lock for the satellites make it difficult for the Kalman filter to work with the smartphones. Compared with the traditional PPP model, the improved PPP model is able to smooth code pseudorange with Doppler and to detect cycle slip more effectively, so the positioning accuracy is significantly improved.
5. Conclusions and Discussion
In this paper, the GNSS raw observations of two smartphones are analyzed through synchronous observations with a geodetic receiver. To study the positioning performances with the smartphones, experiments are conducted with a V20 smartphone and a Mate30 smartphone, as well as a Trimble R8 receiver. With the ionosphere-constrained single-frequency PPP strategy, an improvement in smartphone positioning is achieved. The horizontal positioning accuracy better than 1 m can be reached with the Mate30 smartphone.
Compared with a Trimble R8 geodetic receiver and a V20 smartphone, the GNSS performance of the Mate30 is evaluated. Firstly, there is little difference between the number of observable satellites in the first frequency of the Mate30 and the Trimble R8 receiver. The number of observable satellites with dual-frequency measurements available of the Mate30 and the V20 is not enough for dual-frequency PPP. The satellites lock loss of the Mate30 and the V20 is more frequent than the Trimble R8. The satellite tracking of smartphones is related to its chip. Secondly, the carrier-to-noise ratio of the smartphones is worse than that of the Trimble R8. And the carrier-to-noise ratio of the Mate30 is better than that of the V20. The carrier-to-noise ratio of smartphones has no significant relationship with elevation angle of satellites. Meanwhile, the multipath effect of smartphones is more serious than the Trimble R8. The standard deviations of code multipath errors of the Mate30 and the V20 are about ten times that of the Trimble R8. The performances of the smartphones against multipath effects of are worse than that of the geodetic receiver.
Since the number of navigation satellite with dual-frequency signals observed by smartphones is not enough for dual-frequency PPP, ionosphere-constrained single-frequency PPP model is used for smartphones. In this study, the standard deviations of horizontal positioning errors with the Mate30 are less than 1 m and the Mate30 achieves a relatively stable positioning result. Comparing the different positioning projects, we believe that there are two main reasons for the poor performance using the traditional PPP project of smartphones. For the current smartphones, the data qualities of the code measurements are relatively poor, and the losses of lock for satellites occur frequently. The ionosphere-constrained single-frequency PPP model along with the cycle slip detection method proposed in this work, to a certain extent, circumvents the above disadvantages and has considerable applicability on the smartphones.
The GNSS chip and antenna performance of smartphones is worse than that of geodetic receivers because the design of smartphones needs to consider the beauty and portability. This leads to frequent satellite lock loss, poor code pseudorange quality and serious multipath effects. Therefore, improving hardware quality and algorithm research are effective methods to improve smartphones positioning accuracy. Although the dual-frequency PPP model is unrealistic for smartphones at present due to their incompetence in collecting dual-frequency measurements, it is expected that the performance of PPP with smartphones will be significantly improved in the future with both the rapid development of BDS and the iterative upgrades of smartphones.