蘑菇直播

Precision Positioning Technology in the Palm of Your Hand

: The Present and Future of Smartphone Based GNSS Precision Positioning

Professor Park Byoung-woon & Dr. Yoon Jeong-hyeon

from the Department of Aerospace Systems Engineering

1. Introduction

Smartphones have evolved far beyond being mere communication devices, and they are now transforming nearly every aspect of our daily lives through Global Navigation Satellite Systems (GNSS), various sensors, and Location-Based Services (LBS). Figure 1 illustrates that LBS, which has traditionally been used in map-making, Geographic Information Systems (GIS), pedestrian and vehicle navigation, object tracking, and traffic monitoring & planning, has recently expanded into a wide range of applications, which include social networking, safety and emergency responses, gaming, and sports [1][2]. Moreover, the demand for ultra-precise positioning information is rapidly increasing in services, such as ride-hailing, Augmented Reality (AR), and smart city infrastructure.

However, the positioning accuracy of smartphones that is commonly used in everyday life still falls far short of meeting these demands. Positioning errors can frequently reach several tens of meters particularly in dense urban environments, which is where accurate location information is critical [3].

Figure 1. Various application areas of location based service using smartphone

Industry experts predict that in order to realize next-generation applications, which include fully autonomous driving, GNSS should achieve horizontal positioning accuracy at the level of 20–30 cm [4]. It is essential to enhance smartphone positioning accuracy through the advancement of GNSS technology and its integration with a wide array of modern sensors in order to meet these expectations and implement next-generation services [5]. However, the current GNSS positioning accuracy in smartphones remains at approximately 5 to 10 meters even in open-sky environments, which can deteriorate to over 100 meters in urban areas that are densely populated with obstacles and tall buildings. This indicates that the current technology has its limitations in regards to applying it to the future applications that need precise positioning [6].

2. Structural Limitations of Smartphone GNSS Technology

GNSS reception performance is largely influenced by hardware-related factors, such as antenna quality, receiver design, measurement frequency, and by external signal environments. However, smartphones inherently possess structural limitations compared to dedicated GNSS equipment, which is due to constraints that are related to device size, manufacturing costs, and battery life. The main limitations are as follows.

First, smartphones use low-cost antennas that are not optimized for high-precision GNSS signal reception. Most smartphones employ linearly polarized (LP) antennas, which have lower reception efficiency compared to the right-hand circularly polarized (RHCP) signals that are transmitted by GNSS satellites. GNSS signals are generally received via a direct line-of-sight (LOS) path. However, the signals can be blocked or reflected by buildings and other obstacles in complex environments, such as urban areas [7], which generates multipath errors. Direct signals are obstructed and only reflected signals are received in non-line-of-sight (NLOS) environments [8][9], and positioning errors are consequently bound to increase. High-end surveying-grade receivers use RHCP antennas to minimize the reception of reflected NLOS signals in order to mitigate these types of errors. They also employ techniques, such as elevation angle-based masking and signal-to-noise ratio (SNR) filtering in order to exclude low-quality signals.

Figure 2. Signal strength variation of survey-grade GNSS receivers and smartphones by elevation angles

Figure 2 shows that receivers that are equipped with RHCP antennas tend to increase the signal-to-noise ratio (SNR) as the satellite elevation angle rises. Smartphones with LP antennas in contrast do not clearly follow this trend, so it is difficult to isolate or filter out low-quality signals. As a result, positioning errors that are caused by reflected or Non-Line-of-Sight (NLOS) signals are exacerbated in urban environments with tall buildings due to the ambiguity in distinguishing received signals.

Second, smartphones often perform GNSS reception in the method of the duty cycle in order to optimize the battery life. For example, the GNSS chipset may operate for only about 200 milliseconds per second, and it remains off for the rest of the time. This intermittent reception method fails to ensure the continuity of the carrier phase data, so it is hard to apply high-precision positioning techniques, such as Real-Time Kinematics (RTK) or Precise Point Positioning (PPP). Moreover, Android smartphones that run version 7.0 or later allow access to raw GNSS data, but many manufacturers either do not provide carrier phase data or include the data with unreliable quality. These limitations pose significant challenges in regards to the practical application of precision positioning.

Third, the internal clock of a smartphone is not fully synchronized with the satellite system time, so timing errors between measurements can occur. Figure 3 shows variations in the pseudorange values that are received by a smartphone from GPS satellite PRN 7, which clearly indicates the presence of drift and jump. This type of a phenomenon is not observed in conventional GNSS receivers. Time errors break the consistency between the Doppler and carrier phase measurements, which thereby hinder accurate position computation.

Figure 3. Clock asynchrony in smartphone GNSS measurements

Smartphones compared to dedicated receivers are more likely to receive abnormal GNSS measurements even under standard reception conditions, which directly lead to decreased accuracy, due to structural limitations. This issue becomes even more prominent in urban areas where the likelihood of multipath and Non-Line-of-Sight (NLOS) signals is high. Figure 4 illustrates a smartphone’s reception environment in a city where LOS and NLOS signals are mixed, which shows how the inclusion of abnormal signals can significantly impact the estimation of a user’s position. Techniques, such as elevation angle masking and SNR filtering can mitigate the impact of these types of signals to some extent in conventional GNSS equipment. However, the effectiveness of the conventional technique applied to survey-grade GNSS receivers is limited due to structural constraints of smartphones. A new method is therefore needed to detect and exclude abnormal measurements without relying on elevation angle or SNR criteria. This research team proposes and implements a method in regards to improving the reliability and accuracy of smartphone positioning through the GNSS outliers detection technique based on the measurement consistency and the integrated analysis with MEMS sensors.

Figure 4. Smartphone GNSS signal reception in urban environments: Building blockage and reflection

3. Precision Positioning Service Improvement: A Method for Detecting Abnormal GNSS Measurements in Smartphones

Smartphone GNSS measurements can still achieve meaningful accuracy improvements simply by precisely analyzing the signals and eliminating abnormal values despite various limitations. This section introduces three key techniques that enhance the positioning accuracy using only the GNSS measurement data that is embedded in smartphones, which includes the utilization of L5 signals, Doppler-based filtering, and the Single-Difference based Outlier Monitoring (SDOM).

L5 signals (GPS L5 and Galileo E5) have a chip rate that is ten times faster than traditional L1 signals, and they include a pilot channel, which allows for a long integration time. As a result, they are more resistant to multipath effects and show significantly lower measurement noise. L5 signals recorded up to 60–70% lower positioning error compared to L1 signals under the same satellite conditions in actual experiments. However, most current Differential GNSS (DGNSS) infrastructures support only the L1 frequency, so it is necessary to establish additional infrastructure in order to directly apply L5 signals to precision positioning. This research team accordingly proposed a method that indirectly generates L5 corrections by applying ionospheric correction data, which are offered by Satellite-Based Augmentation Systems (SBAS), to L1 corrections. This approach offers a practical alternative for enhancing the utilization of L5 signals without the need for additional hardware.

Doppler data can be used in situations where GNSS carrier phase measurements on smartphones are unstable or unavailable in order to filter code measurements [14]. Doppler, which is based on the relative velocity between the receiver and the satellite, is noisier than the carrier phase, but it offers superior continuity and is reliably received on most smartphones. The horizontal positioning error was improved to an average of 1.2 meters and the vertical error to approximately 2.3 meters in an experiment where a Doppler-based Kalman filter was applied, and 95% of the total measurements achieved accuracy within 5 meters.

The most critical technique is the Single-Difference based Outlier Monitoring (SDOM) that is tailored for smartphones. Smartphone measurements, such as pseudorange, carrier phase, and Doppler are recorded based on different clock sources, which is unlike conventional GNSS receivers where all the observation data is recorded based on a single receiver clock. This study uses the difference between one reference satellite and other satellites in order to eliminate inconsistencies that are caused by the receiver clock sources and effectively detect only the actual outliers. Temporary clock errors at the receiver can be eliminated in the satellite-to-satellite difference by applying single-differencing between satellites (∇), and only abnormal measurements can be isolated as a result.

Figure 5. Outlier detection before and after applying the Single-Difference based Outlier Monitoring (SDOM) method

Figure 5 illustrates the results of outlier detection before, which are in blue, and after, which are in orange, the SDOM method for GPS PRN 7 satellite was applied. Outlier detection was performed based on the Doppler-based predicted values and the difference between time-differenced pseudorange and carrier phase measurements. The thresholds were set to 5 m/s and 5 cm/s [10][15]. 99.83% of the pseudorange monitoring values were detected as outliers before the application of SDOM, but 65.6% of these were false detections that were caused by time inconsistencies among the measurements as opposed to being actual anomalies. The lack of valid measurements made it impossible to compute a reliable position due to the excessive false detections. However, the influence of clock term discrepancies was mitigated after the application of SDOM, which significantly reduced the false detection rate, preserved valid measurements, and thereby enabled accurate position estimation. 58.96% were initially classified as outliers before SDOM was applied in the carrier phase measurements, which is shown in the figure on the right. Only 10.34% exceeded the 5 cm/s threshold after the application of the method, which indicated that the method effectively filtered out only the actual errors. As such, SDOM is an outlier detection approach that is designed to address the structural limitations of smartphones, and it is evaluated to be a key technology in regards to remarkably enhancing the reliability and accuracy of smartphone GNSS measurements.

4. GNSS Error Detection and Reliability Enhancement Based on a Multi-Sensor Integrated System in Smartphones

Most modern smartphones are equipped with GNSS as well as also with various sensors that are capable of measuring motion, orientation, and environmental conditions [11]. Common examples include accelerometers, gyroscopes, magnetometers, and temperature/humidity sensors, which are all widely utilized across various applications. For instance, gaming apps use gravity sensors and accelerometers in order to interpret complex user gestures, such as tilting, shaking, rotating, and swinging. Travel apps use magnetometers and accelerometers in order to calibrate compass directions. Figure 6 illustrates the configuration of major Micro Electro Mechanical Systems (MEMS) sensors that are installed into smartphones [12].

Figure 6. Structure of a smartphone MEMS sensor [12]

Motion sensors that include accelerometers and gyroscopes play a crucial role among these in regards to measuring device movement and orientation. Accelerometers measure linear acceleration in consideration of gravity, and gyroscopes measure angular velocity. Inertial Navigation System (INS) positioning, which uses the combination of these sensors, can estimate 3D position and orientation independently of external signals, so it becomes complementary to GNSS. It in particular has the advantage of allowing continuous position estimation even in environments where signals are weak or blocked, such as indoors, underwater, urban canyons, and in areas with severe signal interference. INS estimates relative position by integrating acceleration and angular velocity over time based on an initial position. However, it is common to combine INS with GNSS and use the absolute position data from GNSS in order to correct the relative estimates from INS as accumulated errors can grow over time.

Figure 7. Method for eliminating abnormal GNSS measurements using INS

*1. INS based position update

*2. Elimination of abnormal GNSS measurements

*3. Position re-estimation through selected satellites

The research team adopted a tightly-coupled method in the measurement domain between GNSS and INS in this study as opposed to employing a loosely-coupled approach in the positioning domain. Figure 7 illustrates that the INS navigation system predicts the position, and the predicted position is compared with the GNSS-derived position. Abnormal measurements from satellites are then identified and removed based on this comparison. The magnetometer and altimeter data were also in particular utilized in order to reduce the risk of diverging accumulated errors in the INS over time, and sensor fusion was adaptively performed by incorporating the user's motion characteristics.

The algorithm begins by estimating the previous position using satellite navigation data and a tightly-coupled based Extended Kalman Filter (EKF). INS mechanization was applied next in order to predict the pseudorange and Doppler values based on the previously estimated position. These predicted values are then compared with the actual GNSS measurements. If the difference exceeds a certain threshold, the corresponding satellite measurement is classified as abnormal and excluded. Finally, the position is re-estimated with the use of the selected satellite measurements only after the outlier satellites are removed. This enables more reliable positioning estimation compared to the use of GNSS alone.

Figure 8. Results of LOS and NLOS satellite classification using the tightly-coupled GNSS/INS system

Figure 8 shows the results of applying the proposed tightly-coupled GNSS/INS algorithm in a real urban environment. Each satellite was classified as being either LOS (green) or NLOS (red) based on the GNSS data that was collected from a smartphone mounted on a vehicle. It was observed that satellites R13 (GLONASS 13) and R24 (GLONASS 24) were classified as NLOS at 1350 seconds, which corresponds to the presence of a tall building to the east of the test vehicle, when projected onto the actual building layout in the sky plot. This result is highly consistent and supports the validity of the algorithm considering the actual building configuration. Four satellites (G05, G09, R04, and R15) located in the southwest direction were also classified as NLOS at the 1491-second mark, and this too corresponded to an area that was obstructed by buildings.

As a result, the approach of removing abnormal GNSS measurements using INS based predicted position information and re-determining the position with only selected high-quality satellites proved to be a highly effective strategy in regard to providing more stable and reliable location-based services to smartphone users.

5. Experiment and Validation: 3rd Place in the Smartphone Decimeter Challenge

This research team evaluated the algorithm performance using the open dataset from the Google Smartphone Decimeter Challenge (GSDC) in order to validate the effectiveness of the various GNSS accuracy enhancement techniques that were previously introduced. GSDC is a large-scale international competition, which is jointly hosted by Google, the Institute of Navigation (ION), and the data analytics platform Kaggle. Its goal is to develop technologies that are capable of achieving decimeter-level (tens of centimeters) positioning accuracy using only smartphones. The competition has been held annually since 2021. The participants were required to develop precise positioning algorithms using GNSS/IMU data that was collected from various smartphone models in real-world urban environments.

Figure 9. Research Team from Sejong University Navigation Systems Research Laboratory Won 3rd Place in the Smartphone Decimeter Challenge (Source: Left – ION official website, Right – ION Newsletter)

The GSDC, which was held three times in total in 2021, 2022, and the 2023–2024 seasons, garnered a significant amount of attention from both academia and the industry. A total of 1,662 teams and 1,991 participants from around the world took part in the challenge over the course of three years in order to push the limits of smartphone-based technologies [13].  Figure 9 illustrates that the research team from Sejong University's Navigation Systems Research Laboratory won 3rd place out of 322 teams in the most recent 2023–2024 season, which showed the potential for achieving decimeter-level accuracy using only smartphones.

The research team implemented and applied a hybrid positioning algorithm that incorporated Doppler-based velocity prediction and position updates, Time-Differenced Carrier Phase (TDCP)-based velocity correction, Satellite Differenced Outlier Mitigation (SDOM) using single-differenced measurements between satellites, and INS assistance based NLOS satellite filtering based on raw GNSS/IMU data that was received from smartphones. As a result, the final Kaggle scores, which were based on the average of 50% and 95% positioning errors, were 0.89 meters for the public set and 1.19 meters for the private set. The algorithm notably had a horizontal error of approximately 0.3 meters and a vertical error of around 0.4 meters, which achieved an exceptional level of accuracy, with high-end smartphones that support dual-frequency GNSS (L1/L5). The positioning accuracy reached approximately 1.5 meters even on smartphones that use single-frequency L1 GNSS, which proved both the versatility and practicality of the algorithm. The first and second place teams utilized post-processing techniques by leveraging predefined driving routes and the provided file-based datasets, but the Sejong University research team implemented their algorithm in a forward-only or in the way of approaching real-time processing conditions. The decimeter-level accuracy therefore achieved by the research team is expected to be similarly attainable in real-time services.

Figure 10 illustrates a comparison of smartphone GNSS positioning results along the actual driving route, which is shown in orange. It was verified that the errors in the position information offered by Google, which are shown in blue, were significantly reduced by the algorithm that was developed by the research team from Sejong University's Navigation Systems Research Laboratory (green), which resulted in a trajectory that closely matches the true driving path.

Figure 10. Comparison of smartphone GNSS positioning results. The research team from Sejong University's Navigation Systems Research Laboratory (green), Google’s offered position information (blue), and the actual driving path (orange)

6. Conclusion

Smartphone GNSS technology has established itself as a core component of today’s diverse location-based services, and its scope of application continues to expand. The potential of smartphone-based positioning technology has recently been garnering a significant amount of attention even in advanced application areas that demand decimeter-level accuracy, such as autonomous vehicles, smart city infrastructure, precision agriculture, hybrid indoor-outdoor navigation, and Collaborative Intelligent Transport Systems (C-ITS). However, the current positioning accuracy of smartphone GNSS remains limited to approximately 5 to 10 meters even in open-sky environments and in dense urban areas, which errors can exceed 100 meters.

The primary cause of these types of errors lie in the structural limitations of smartphones, which receive GNSS measurements under various constraints, such as low-quality antennas, discontinuous measurement conditions, low signal sensitivity, and asynchronous clock timing. These limitations frequently lead to abnormal GNSS measurements. It is impossible to make accurate positioning without the identification and elimination of these types of anomalies.

This study experimentally proved that decimeter-level precise positioning is achievable even with smartphones by applying GNSS measurements based outlier removal algorithm, fusion with Inertial Navigation System (INS) based on IMU, and satellite signal quality classification and filtering techniques in order to overcome these structural limitations.

The research team of the Navigation Systems Research Laboratory at Sejong University notably showed the global competitiveness of smartphone based GNSS positioning technology by winning 3rd place in the Google Smartphone Decimeter Challenge (GSDC), which was co-hosted by Google and the Institute of Navigation (ION). The algorithm that was developed by the research team was developed for real-time processing, and it has strong potential for future practical applications.

These research results reveal that precision positioning using smartphones has moved beyond experimental possibility and reached a level where it can be practically applied in next-generation spatial information domains, such as autonomous driving, drone navigation, augmented reality (AR), and high-precision location sharing services. The positioning performance achieved by the research team in particular surpasses that of the widely used Qualcomm Snapdragon platform in the global market. It is therefore expected to significantly contribute to technological self-reliance and enhanced international competitiveness for domestic smartphone manufacturers and the GNSS industry.

A time will come when anyone can enjoy centimeter-level location based services that surpass even the decimeter-level accuracy that was achieved in this study. This research will hopefully serve as a major milestone in regards to turning that future into reality.

References

[1]    F. van Diggelen, W. Roy, and W. Wang, “How to achieve 1-meter accuracy in Android,” GPS World, July 3, 2018, Jul. 2018.

[2]    J. Paziewski, “Recent advances and perspectives for positioning and applications with smartphone GNSS observations,” Meas Sci Technol, vol. 31, no. 9, p. 091001, Sep. 2020, doi: 10.1088/1361-6501/ab8a7d.

[3]    M. M. Farhad, M. Kurum, and A. C. Gurbuz, “A Ubiquitous GNSS-R Methodology to Estimate Surface Reflectivity Using Spinning Smartphone Onboard a Small UAS,” IEEE J Sel Top Appl Earth Obs Remote Sens, vol. 16, pp. 6568–6578, 2023, doi: 10.1109/JSTARS.2023.3294833.

[4]    F. Zangenehnejad and Y. Gao, “GNSS smartphones positioning: advances, challenges, opportunities, and future perspectives,” Satellite Navigation, vol. 2, no. 1, p. 24, Nov. 2021, doi: 10.1186/s43020-021-00054-y.

[5]    P. Dabove and V. Di Pietra, “Towards high accuracy GNSS real-time positioning with smartphones,” Advances in Space Research, vol. 63, no. 1, pp. 94–102, Jan. 2019, doi: 10.1016/j.asr.2018.08.025.

[6]    G. (Michael) Fu, M. Khider, and F. van Diggelen, “Android Raw GNSS Measurement Datasets for Precise Positioning,” in 33rd International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2020), Oct. 2020, pp. 1925–1937. doi: 10.33012/2020.17628.

[7]    G. A. McGraw, P. D. Groves, and B. W. Ashman, “Robust Positioning in the Presence of Multipath and NLOS GNSS Signals,” in Position, Navigation, and Timing Technologies in the 21st Century, Wiley, 2020, pp. 551–589. doi: 10.1002/9781119458449.ch22.

[8]    Y. Lee and B. Park, “Nonlinear Regression-Based GNSS Multipath Modelling in Deep Urban Area,” Mathematics, vol. 10, no. 3, p. 412, Jan. 2022, doi: 10.3390/math10030412.

[9]    Y. Lee, P. Wang, and B. Park, “Nonlinear Regression-Based GNSS Multipath Dynamic Map Construction and Its Application in Deep Urban Areas,” IEEE Transactions on Intelligent Transportation Systems, pp. 1–12, 2023, doi: 10.1109/TITS.2023.3246493.

[10]    J. Yun, C. Lim, and B. Park, “Inherent Limitations of Smartphone GNSS Positioning and Effective Methods to Increase the Accuracy Utilizing Dual-Frequency Measurements,” Sensors, vol. 22, no. 24, p. 9879, Dec. 2022, doi: 10.3390/s22249879.

[11]    J. Yun and B. Park, “A GNSS/Barometric Altimeter Tightly Coupled Integration for Three-Dimensional Semi-Indoor Mapping With Android Smartphones,” IEEE Geoscience and Remote Sensing Letters, vol. 21, pp. 1–5, 2024, doi: 10.1109/LGRS.2024.3365610.

[12]    Y. Wang, N. Li, X. Chen, and M. Liu, “Design and Implementation of an AHRS Based on MEMS Sensors and Complementary Filtering,” Advances in Mechanical Engineering, vol. 6, p. 214726, Jan. 2014, doi: 10.1155/2014/214726.

[13]    A. Chow, D. Orendorff, M. Fu, M. Khider, S. Dane, and V. Gulati, “Google Smartphone Decimeter Challenge 2023-2024,” Kaggle. https://kaggle.com/competitions/smartphone-decimeter-2023, 2023. 

[14]    B. Park, C. Lim, Y. Yun, E. Kim, and C. Kee, “Optimal Divergence-Free Hatch Filter for GNSS Single-Frequency Measurement,” Sensors, vol. 17, no. 3, p. 448, Feb. 2017, doi: 10.3390/s17030448.

[15]    J. Yun, C. Lim, B. Park, and J. Morton, “Detecting Ionospheric Irregularity Based on ROT Variation Using Android Devices Cloud System,” in 33rd International Technical Meeting of the Satellite Division of the Institute of Navigation, ION GNSS+ 2020, Oct. 2020, pp. 1850–1872. doi: 10.33012/2020.17624.