Real-Time Kinematic Positioning Using Multi-Frequency Smartphone Measurements ^†

Abstract

Nowadays, several smartphones on the market provide multi-frequency multi-constellations GNSS measurements, including carrier phase ones, allowing the achievement of high-accuracy positioning by exploiting Real Time Kinematic (RTK) or Precise Point Positioning (PPP) techniques. This paper will showcase the effectiveness of using smartphone measurements for RTK under different scenarios and for different applications using baselines of different lengths. The impact of the smartphone’s antenna on the solution will also be analysed. The assessment will be performed by evaluating different key performance indicators, including the time to first fix and the horizontal/vertical accuracy. This paper shows that around a 99% fix position can be achieved even using the smartphones’ antennas for the static case under open sky conditions. Moreover, high percentages of fix solutions can also be achieved in kinematic mode by ad hoc tuning of the RTK algorithm.

1. Introduction

The diffusion of smartphones providing multifrequency GNSS measurements, including carrier phase observations, has opened new opportunities for the achievement of high-accuracy positioning from low-cost devices. A number of research works have assessed the performance of algorithms exploiting smartphone measurements for Real Time Kinematic (RTK) and Precise Point Positioning (PPP) applications (e.g., [1,2,3,4]). This paper showcases the effectiveness of using GNSS observations from the triple-frequency Huawei P40 Mate smartphone (Huawei, Shenzen, China) for RTK processing under different environments and dynamic conditions.

Since the quality of the GNSS measurements is affected by the low-quality smartphone antenna, we conducted a preliminary analysis to quantify the antenna’s impact on the measurements. Survey and patch antennas were connected to the smartphone under open sky/static conditions for comparison purposes. Several key performance indicators were analysed: the signal power, the Time to First Fix (TTFF), the positioning accuracy, and the percentage of fix solutions using bases at different distances (i.e., 5, 10, and 25 km).

The results show that, as expected, the GNSS measurement quality is enhanced using external antennas. However, this paper showcases that, even using the smartphone’s antenna, about 99% of fix solutions and centimetre positioning accuracy can be achieved by ad hoc tuning the RTK algorithm.

Moreover, a dynamic test was conducted by fixing the smartphone to the windshield of a tractor and two automotive tests were performed in a rural environment with the smartphone fixed to the car’s windshield and roof.

This paper will discuss the results obtained by the above tests and the tuning of the RTK algorithm for coping with smartphone measurements. Specifically, this paper is structured as follows: Section 2 reports the methodology and the experimental set-up adopted for assessing the performance achieved using smartphone measurements in RTK mode under both static and dynamic conditions. Section 3 reports the results obtained by connecting the smartphone to external antennas of different grades. Section 4 and Section 5 discuss the results obtained using the smartphone’s internal antenna under both static and dynamic conditions and with different baseline lengths. Finally, Section 6 draws some conclusions.

2. Methodology and Experimental Set-Up

The tests presented in this paper were carried out using the Huawei P40 Mate smartphone. This smartphone was selected because can track three frequencies, specifically, GPS L1/L5, Galileo E1/E5a/E5b, Glonass L1, and Beidou B1/B2a. However, from preliminary tests, it has been observed that the number of Galileo E5b tracked signals is very limited and has very low signal power. Therefore, it was not possible to use the above frequency for positioning. A first analysis has been performed to characterise the antenna’s impact on the quality of smartphone measurements. For this purpose, a survey-grade antenna (PG-F1 [5], Topcon, Livermore, CA, USA) and a patch antenna (ADFGP.60A [6], Taoglas, Dublin, Ireland) have been connected to the smartphone, which was placed in a portable shielded box as shown in the scheme in Figure 1a. Inside the shielded box, a re-radiator was placed to broadcast the signal to the phone. The antennas were placed in the middle of a field under open-sky conditions. At the same time, a high-grade receiver was connected to the same antenna in order to act as a reference. Moreover, a test was performed using the smartphone without any external antenna. As shown in Figure 1b, the smartphone was placed on a tripod in the middle of a field under open-sky conditions. The back of the phone was pointing up, since, after some preliminary tests, it was observed that this position maximised the received signal power. The Google GNSS logger application was used to record files including GNSS measurements that were then processed to obtain a Receiver Independent Exchange Format (Rinex) file using UofC CSV2RINEX; available at https://github.com/FarzanehZangeneh/csv2rinex (accessed on 3 March 2024). The RINEX files were sent as input to the Topcon RTK engine and the binary files were recorded by different base stations. Then, a post-processing kinematic solution was obtained for the different cases. A routine was prepared to test different combinations of settings in the RTK engine and select the best combination, that is, the one leading to the higher percentage of fix solutions, higher accuracy, and lower TTFF. The tests’ set-up description, along with the results, are reported in the following sections. The tests were performed using three bases placed at different distances (CN09: 1 km, CGU2: 5 km, and GUAS: 25 km). The following key performance indicators were evaluated in the signal quality and positioning domains:

• TTFF: number of epochs needed to achieve the first fix position solution in RTK mode.
• Percentage of fixed solutions.
• Horizontal/vertical accuracy (computed with respect to the position surveyed exploiting the NET-G5 receiver used as a reference).

In addition, two kinematic tests were performed using a car in a rural/semi-urban environment. As shown in Figure 2, some sections of the path were characterised by the presence of tall trees with rich foliage rendering the signal reception very challenging. Moreover, in other sections of the path, buildings and houses also made the environment difficult in terms of signal reception.

For the automotive test, the Topcon survey-grade PG-F1 antenna was mounted on the roof of a car and connected to a survey-grade HyperVR receiver (Topcon, Livermore, CA, USA) [7] that was used as a reference. During the first test, the smartphone was placed inside the car, specifically, attached to the front car’s glass with the back of the phone pointing outside. The setup is shown in Figure 3a. For a second test, the smartphone was mounted on the car’s roof so as to have better visibility. Finally, a further kinematic test was conducted by fixing the smartphone on the windshield of a tractor, as shown in Figure 3b. Also, in this case, the above survey-grade antenna/receivers were used for reference purposes.

3. Static Tests

To understand the impact of the antenna on the quality of the smartphone measurements, we compared the mean Carrier-to-Noise Ratio (C/N₀) versus the satellite elevation for the GPS L1 case using the external antennas and the smartphone’s internal antenna. The mean was computed over five-degree elevation bins and considering all satellites jointly. When the survey grade PG-F1 antenna was used (Figure 4), the maximum C/N₀ was around 50 dB-Hz, while, when a patch antenna was used (Figure 5), the maximum C/N₀ was around 45 dB-Hz and the signal fluctuations due to multipath were more evident. As expected, in both cases, the signal power increases for higher satellite elevations.

The same analysis was repeated using the smartphone without any external antenna (Figure 6). In this case, the signal fluctuations due to multipath are evident even for higher satellite elevations. Moreover, the increase in C/N₀ with respect to the satellite elevation was less significant than in the previous two cases.

In the following, the statistics are reported for the different static tests to showcase the results that can be achieved with the GNSS measurements from a smartphone and RTK positioning with different baseline lengths. Specifically, in

Table 1. PG-F1 antenna (CN09: 1 km, CGU2: 5 km, GUAS: 25 km).

Base	Samples	TTFF [s]	Fix %	Horizontal Accuracy [m]	Vertical Accuracy [m]	Horizontal Precision [m]	Vertical Precision [m]
CN09	28,571	12	99.9	0.010	0.011	0.006	0.006
CGU2	28,571	14	99.9	0.013	0.014	0.009	0.010
GUAS	28,571	64	99.7	0.040	0.034	0.016	0.034

, the results obtained with the PG-F1 antenna are summarised. The test lasted about eight hours and samples were collected at 1 Hz. With the increase in the baseline length, an increase in the TTFF can be observed. However, the percentage of fix solutions is above 99% for all cases. As expected, a slight degradation in the root mean squared (rms) of the position accuracy and precision can be observed with the increase in the baseline length. In

Table 2. Patch antenna (CN09: 1 km, CGU2: 5 km, GUAS:25 km).

Base	Samples	TTFF [s]	Fix %	Horizontal Accuracy [m]	Vertical Accuracy [m]	Horizontal Precision [m]	Vertical Precision [m]
CN09	27,741	28	99.5	0.011	0.016	0.008	0.015
CGU2	27,741	29	95.9	0.013	0.020	0.012	0.020
GUAS	27,741	205	91.1	0.051	0.043	0.026	0.052

, the results obtained with the patch antenna are also summarised. Also, in this case, the increase in the baseline length affects the TTFF. Moreover, for the baseline of 25 km, a decrease in the percentage of fix solutions can also be observed. Finally, in

Table 3. Internal antenna (CN09: 1 km, CGU2: 5 km, GUAS: 25 km).

Base	Samples	TTFF [s]	Fix %	Horizontal Precision [m]	Vertical Precision [m]
CN09	9560	28	97.6	0.020	0.037
CGU2	9560	29	95.9	0.046	0.048
GUAS	9560	175	91.3	0.090	0.134

, the results obtained with the smartphone’s antenna are reported. In this case, the test was shorter, since it lasted around 2.5 h. In this case, the percentage of fix solutions is above 91% for all cases, also when the baseline length is 25 km. For this test, we report the horizontal and vertical precision instead of the accuracy since it was not possible to survey the exact position of the smartphone in an accurate way. A degradation in the solution precision can be observed with the increase in the baseline length.

4. Kinematic Tests

As described in Section 2, two automotive tests were conducted, first fixing the smartphone to the windshield of the car and then placing the smartphone on the roof of the car. In addition, a further kinematic test was carried out by fixing the smartphone on the windshield of a tractor. The full path travelled by the car during the automotive tests is shown in Figure 7a. During the automotive tests, the base was at a distance below 5 km. As mentioned in Section 4, the path was characterised by the presence of houses and trees with foliage, rendering signal reception quite difficult. In Figure 8a,b, sections of the path are also reported. The solution provided by the smartphone (for the case of the smartphone placed inside the car and fixed on the windshield) is reported in blue while the solution given by the reference is represented in green. The pictures show that the smartphone solution follows the reference one quite closely, even in the correspondence of the curves. The statistics of the tests are reported in

Table 4. Statistic of the kinematic test.

	TTFF [s]	FIX %	TTFF Reference [s]	Fix % (Reference)
Car (inside)	49	75.34	0	97.76
Car (outside)	0	78.5	3	98.34
Tractor	41	81.95	1	99.9

, along with those obtained for the reference ones. As expected, when the smartphone is placed on the roof of the car, better performance can be achieved in terms of TTFF, thanks to better satellite visibility. However, in both cases, the percentage of fix solutions is above 75%. In Figure 9, the full path travelled with the tractor is shown. Also, in this case, the smartphone solution (in blue) closely follows the reference one (in green). However, in correspondence with the turns, some outliers are present. Indeed, the turns are more challenging due to the dynamic changes. The statistics for this test are also reported in Table 4. In this case, the performance is slightly better than the automotive test cases thanks to the most favourable scenarios since the field where the test was performed was under quasi-open sky conditions.

5. RTK Tuning Discussion

The smartphone measurements are noisier than the ones of standard GNSS receivers. Therefore, the results presented in this paper were obtained by ad hoc tuning the Topcon RTK engine to take into account the peculiarities of the smartphone observations and the specific test scenarios. As mentioned above, a routine was developed to find the best setting parameters for the specific test. The best parameters were considered the ones which maximised the horizontal/vertical accuracy and percentage of fix solutions and minimised the TTFF. The set of parameters included the following:

• Satellite elevation mask;
• C/N₀ cut-off values for the different signals;
• Observation weighting approach;
• Code and phase noise.

For all tests, the code and phase noise were assumed to be ten times larger than the ones set for the standard GNSS receiver cases. Moreover, for the different test cases, the selected elevation cut-off values ranged between 15 and 20 degrees for both static and kinematic tests. These high values helped to reject the measurement of bad quality, which decreased the final performance. Finally, for all cases, an observation weighting scheme depending on the signal C/N₀ was selected. Indeed, smartphone measurement can also be deteriorated for high satellite elevations. Therefore, it is necessary to take into account the signal power to give less weight to the lower-quality measurements. It should be considered that, for the automotive tests, the environment was quite variable, ranging from an open sky section to a section with tall trees and buildings. However, a unique set of best parameters was also selected in this case. An a priori identification of the scenario type could assist in further optimising the selection of the RTK setting parameters.

6. Conclusions

This paper showcased the effectiveness of using smartphone measurements for RTK applications. The results showed that, even with the internal smartphone antenna, around 99% of fix solutions can be achieved under static and open sky conditions. Moreover, for all kinematic tests, the percentage of fix solutions was above 75%, even in challenging scenarios where the signal reception was difficult due to the presence of tall trees with rich foliage. In future work, the GNSS measurements will be combined with the ones from inertial sensors in order to enhance the performance in the presence of signal obstructions to match the ones encountered during the performed kinematic tests. Moreover, the identification of different types of environments will be performed in order to allow the tuning of RTK parameters on the fly.