![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6469415%2Fd6d88a77837e884f74a1a6ebc658cce6%2F2.png?generation=1717684763258428&alt=media\")
We would like to thank the organizers for hosting an great competition. Additionally, we express our respect to everyone who has participated in this competition over the long period since September last year.
\nThis year, more types of smartphone data were provided compared to last year, and it was an especially challenging endeavor due to the Samsung A-series supporting only L1 single-Frequency.
\nWe are writing this post to share the solution for the entry that won 3rd place in this competition.
\nOur solution are based on the entry from Sejong University, which ranked 7th in the previous Google Smartphone Decimeter Challenge 2022 (GSDC 2022).
\nSejong University proposed several useful positioning improvement strategies considering the limitations of smartphones, such as noise and multipath error reduction using L1/L5 dual-frequency, Doppler-based filtering techniques [1][2], and satellite single-difference-based outlier monitoring (SDOM) technique [3].
\nThe key aspects of this year's approach include the Removal of Outlier Measurements through residual comparison from last year's Doppler-Aided Position/Velocity, and the addition of more accurate Time-Differenced Carrier-phase (TDCP) Velocity.
\n
The main strategies for this competition can be divided into 3 key aspects:
\nUsing the RINEX Converter with SDOM: Convert raw GNSS measurements into pseudo-range, carrier-phase, and doppler measurements using a RINEX converter equipped with the SDOM (Single-Difference Outlier Monitoring) mentioned above. The satellite single-difference technique helps resolve synchronization issues between measurements and effectively removes outlier measurements.
Doppler-based Position and Velocity Update: Using time-synchronized Doppler measurements, it predicts current position and velocity based on previous position and velocity. Doppler measurements tend to be noisier than carrier phase measurements, but have the advantage of being able to be used on a larger number of satellites.
TDCP-based Position and Velocity Update: DGNSS (Differential GNSS) correction is applied to the pseudo-range, compare with updated position using doppler-velocity to calculate residuals and update the position. Similarly, compare TDCP measurements with Doppler-Aided Velocity to calculate residuals and update the velocity.
Before applying the Test Dataset, learning was conducted using the All Training Dataset. It has been confirmed that very different characteristics appear depending on the type of smartphone chipset and whether or not it supports L5 frequency. Smartphones included in the test dataset were categorized by chipset and frequency support.
\nSmartphones that support L5 show very good performance, while smartphones that support L1 single-frequency have very poor performance. I wanted to solve this problem, but the training dataset contained too little Exynos L1 data (1) and Mediatek L1 (2) data. :(
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6469415%2Fdd48d7cf7ddcb6e3055f47e0083df795%2F3.png?generation=1717688233921212&alt=media\")
Through the proposed algorithm, the final leaderboard score was 0.928m for public and 1.342m for private, resulting in 3rd place. Immediately after the competition I discovered a major mistake in my file submission… The late submission score after the competition is 0.890m for public and 1.191m for private.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F6469415%2Fbc984ccf9312c1cf954c176cb23eb806%2F1.png?generation=1717684706536088&alt=media\")
Thank you to everyone who participated in the competition for your hard work again.
\nI would also like to thank my team members (Suyeol Kim, Taejin Youn, Gyeongmin Kim and Wonwoo Park) and my advisor, Byungwoon Park (byungwoon@sejong.ac.kr), for their great help in this competition.
\nFor more detailed information, please refer to the site posted below.
\n[1] J. Yun et al, \"Practical Approaches to Real-Time Position Accuracy Improvement of Android Smartphone Dual-Frequency GNSS,\" Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 2226-2234. https://doi.org/10.33012/2022.18373
\n[2] J. Yun et al, \"Inherent Limitations of Smartphone GNSS Positioning and Effective Methods to Increase the Accuracy Utilizing Dual-Frequency Measurements,\" Sensors 2022, 22, 9879. https://doi.org/10.3390/s22249879
\n[3] J. Yun et al, \"Elevating Android GNSS Raw Measurement Processing: A Universal RINEX Converter for Precise Post-Processing Solutions,\" Proceedings of the 36th International Techincal Meeting of the Satellite Division of The Institue of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023.
", "messages": [ { "id": 2858894, "postDate": "2024-06-06T17:51:26.223Z", "content": "We would like to thank the organizers for hosting an great competition. Additionally, we express our respect to everyone who has participated in this competition over the long period since September last year.
\nThis year, more types of smartphone data were provided compared to last year, and it was an especially challenging endeavor due to the Samsung A-series supporting only L1 single-Frequency.
\nWe are writing this post to share the solution for the entry that won 3rd place in this competition.
\nOur solution are based on the entry from Sejong University, which ranked 7th in the previous Google Smartphone Decimeter Challenge 2022 (GSDC 2022).
\nSejong University proposed several useful positioning improvement strategies considering the limitations of smartphones, such as noise and multipath error reduction using L1/L5 dual-frequency, Doppler-based filtering techniques [1][2], and satellite single-difference-based outlier monitoring (SDOM) technique [3].
\nThe key aspects of this year's approach include the Removal of Outlier Measurements through residual comparison from last year's Doppler-Aided Position/Velocity, and the addition of more accurate Time-Differenced Carrier-phase (TDCP) Velocity.
\n
The main strategies for this competition can be divided into 3 key aspects:
\nUsing the RINEX Converter with SDOM: Convert raw GNSS measurements into pseudo-range, carrier-phase, and doppler measurements using a RINEX converter equipped with the SDOM (Single-Difference Outlier Monitoring) mentioned above. The satellite single-difference technique helps resolve synchronization issues between measurements and effectively removes outlier measurements.
Doppler-based Position and Velocity Update: Using time-synchronized Doppler measurements, it predicts current position and velocity based on previous position and velocity. Doppler measurements tend to be noisier than carrier phase measurements, but have the advantage of being able to be used on a larger number of satellites.
TDCP-based Position and Velocity Update: DGNSS (Differential GNSS) correction is applied to the pseudo-range, compare with updated position using doppler-velocity to calculate residuals and update the position. Similarly, compare TDCP measurements with Doppler-Aided Velocity to calculate residuals and update the velocity.
Before applying the Test Dataset, learning was conducted using the All Training Dataset. It has been confirmed that very different characteristics appear depending on the type of smartphone chipset and whether or not it supports L5 frequency. Smartphones included in the test dataset were categorized by chipset and frequency support.
\nSmartphones that support L5 show very good performance, while smartphones that support L1 single-frequency have very poor performance. I wanted to solve this problem, but the training dataset contained too little Exynos L1 data (1) and Mediatek L1 (2) data. :(
\n
Through the proposed algorithm, the final leaderboard score was 0.928m for public and 1.342m for private, resulting in 3rd place. Immediately after the competition I discovered a major mistake in my file submission… The late submission score after the competition is 0.890m for public and 1.191m for private.
\n
Thank you to everyone who participated in the competition for your hard work again.
\nI would also like to thank my team members (Suyeol Kim, Taejin Youn, Gyeongmin Kim and Wonwoo Park) and my advisor, Byungwoon Park (byungwoon@sejong.ac.kr), for their great help in this competition.
\nFor more detailed information, please refer to the site posted below.
\n[1] J. Yun et al, \"Practical Approaches to Real-Time Position Accuracy Improvement of Android Smartphone Dual-Frequency GNSS,\" Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 2226-2234. https://doi.org/10.33012/2022.18373
\n[2] J. Yun et al, \"Inherent Limitations of Smartphone GNSS Positioning and Effective Methods to Increase the Accuracy Utilizing Dual-Frequency Measurements,\" Sensors 2022, 22, 9879. https://doi.org/10.3390/s22249879
\n[3] J. Yun et al, \"Elevating Android GNSS Raw Measurement Processing: A Universal RINEX Converter for Precise Post-Processing Solutions,\" Proceedings of the 36th International Techincal Meeting of the Satellite Division of The Institue of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023.
", "rawMarkdown": "We would like to thank the organizers for hosting an great competition. Additionally, we express our respect to everyone who has participated in this competition over the long period since September last year.\n\nThis year, more types of smartphone data were provided compared to last year, and it was an especially challenging endeavor due to the **Samsung A-series** supporting only L1 single-Frequency.\n\nWe are writing this post to share the solution for the entry that won **3rd place** in this competition.\n\n# Overveiw\nOur solution are based on the entry from Sejong University, which ranked 7th in the previous [Google Smartphone Decimeter Challenge 2022 (GSDC 2022)](https://www.kaggle.com/competitions/smartphone-decimeter-2022/leaderboard). \n\nSejong University proposed several useful positioning improvement strategies considering the limitations of smartphones, such as noise and multipath error reduction using L1/L5 dual-frequency, Doppler-based filtering techniques [1][2], and satellite single-difference-based outlier monitoring (SDOM) technique [3].\n\nThe key aspects of this year's approach include the **Removal of Outlier Measurements** through residual comparison from last year's Doppler-Aided Position/Velocity, and the addition of more accurate **Time-Differenced Carrier-phase (TDCP) Velocity**.\n\n\n\n# Strategy\nThe main strategies for this competition can be divided into 3 key aspects:\n\n1. **Using the RINEX Converter with SDOM:** Convert raw GNSS measurements into pseudo-range, carrier-phase, and doppler measurements using a RINEX converter equipped with the SDOM (Single-Difference Outlier Monitoring) mentioned above. The satellite single-difference technique helps resolve synchronization issues between measurements and effectively removes outlier measurements.\n\n2. **Doppler-based Position and Velocity Update:** Using time-synchronized Doppler measurements, it predicts current position and velocity based on previous position and velocity. Doppler measurements tend to be noisier than carrier phase measurements, but have the advantage of being able to be used on a larger number of satellites.\n\n3. **TDCP-based Position and Velocity Update:** DGNSS (Differential GNSS) correction is applied to the pseudo-range, compare with updated position using doppler-velocity to calculate residuals and update the position. Similarly, compare TDCP measurements with Doppler-Aided Velocity to calculate residuals and update the velocity.\n\n\n# Training Dataset and CV Score\nBefore applying the **Test Dataset**, learning was conducted using the **All Training Dataset**. It has been confirmed that very different characteristics appear depending on the type of smartphone chipset and whether or not it supports L5 frequency. Smartphones included in the test dataset were categorized by chipset and frequency support.\n- **Qualcomm L1/L5:** pixel4, pixel4xl, pixel5\n- **Broadcom L1/L5:** pixel6pro, pixel7pro, sm-g988b\n- **Xiaomi L1/L5:** mi8 (xiaomimi8)\n- **Exynos L1/L5:** sm-s908b\n- **Exynos L1:** sm-a205u, sm-a505u\n- **Mediatek L1:** sm-a325f, samsunga32, samsunga325g\n\nSmartphones that support L5 show very good performance, while smartphones that support L1 single-frequency have very poor performance. I wanted to solve this problem, but the training dataset contained too little Exynos L1 data (1) and Mediatek L1 (2) data. :(\n\n\n\n# Test Dataset and Kaggle Score\nThrough the proposed algorithm, the final leaderboard score was 0.928m for public and 1.342m for private, resulting in 3rd place. Immediately after the competition I discovered a major mistake in my file submission... The late submission score after the competition is 0.890m for public and 1.191m for private.\n\n\n\n# Discussion\nThank you to everyone who participated in the competition for your hard work again.\n- **GNSS/INS integration** have been applied to some smartphones. More details about this solution will be covered in the [ION GNSS+ 2024 presentation](https://www.ion.org/gnss/abstracts.cfm?paperID=13510).\n- More analysis of **L1 single-frequency** is needed. Training Dataset contains too few **Samsung A-series** (sm-a205u, sm-a505u, samsunga32, samsunga325g). The score of Samsung A-series in Training Dataset is only 1.5~1.6m. There are a total of six Samsung A-series phones in the **Test dataset**, and it is expected that better results will be obtained if modeling is conducted on these in the future.\n- (TBC)\n\n# Acknowledgement\nI would also like to thank my team members (**Suyeol Kim**, **Taejin Youn**, **Gyeongmin Kim** and **Wonwoo Park**) and my advisor, **Byungwoon Park (byungwoon@sejong.ac.kr)**, for their great help in this competition.\n\n\n# Reference\nFor more detailed information, please refer to the site posted below.\n\n[1] J. Yun et al, \"Practical Approaches to Real-Time Position Accuracy Improvement of Android Smartphone Dual-Frequency GNSS,\" Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 2226-2234. https://doi.org/10.33012/2022.18373\n\n[2] J. Yun et al, \"Inherent Limitations of Smartphone GNSS Positioning and Effective Methods to Increase the Accuracy Utilizing Dual-Frequency Measurements,\" Sensors 2022, 22, 9879. https://doi.org/10.3390/s22249879\n\n[3] J. Yun et al, \"Elevating Android GNSS Raw Measurement Processing: A Universal RINEX Converter for Precise Post-Processing Solutions,\" Proceedings of the 36th International Techincal Meeting of the Satellite Division of The Institue of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023.\n\n\n\n\n\n\n\n\n\n\n\n\n", "votes": 16 } ], "comments": [], "raw_markdown_by_id": { "2858894": "We would like to thank the organizers for hosting an great competition. Additionally, we express our respect to everyone who has participated in this competition over the long period since September last year.\n\nThis year, more types of smartphone data were provided compared to last year, and it was an especially challenging endeavor due to the **Samsung A-series** supporting only L1 single-Frequency.\n\nWe are writing this post to share the solution for the entry that won **3rd place** in this competition.\n\n# Overveiw\nOur solution are based on the entry from Sejong University, which ranked 7th in the previous [Google Smartphone Decimeter Challenge 2022 (GSDC 2022)](https://www.kaggle.com/competitions/smartphone-decimeter-2022/leaderboard). \n\nSejong University proposed several useful positioning improvement strategies considering the limitations of smartphones, such as noise and multipath error reduction using L1/L5 dual-frequency, Doppler-based filtering techniques [1][2], and satellite single-difference-based outlier monitoring (SDOM) technique [3].\n\nThe key aspects of this year's approach include the **Removal of Outlier Measurements** through residual comparison from last year's Doppler-Aided Position/Velocity, and the addition of more accurate **Time-Differenced Carrier-phase (TDCP) Velocity**.\n\n\n\n# Strategy\nThe main strategies for this competition can be divided into 3 key aspects:\n\n1. **Using the RINEX Converter with SDOM:** Convert raw GNSS measurements into pseudo-range, carrier-phase, and doppler measurements using a RINEX converter equipped with the SDOM (Single-Difference Outlier Monitoring) mentioned above. The satellite single-difference technique helps resolve synchronization issues between measurements and effectively removes outlier measurements.\n\n2. **Doppler-based Position and Velocity Update:** Using time-synchronized Doppler measurements, it predicts current position and velocity based on previous position and velocity. Doppler measurements tend to be noisier than carrier phase measurements, but have the advantage of being able to be used on a larger number of satellites.\n\n3. **TDCP-based Position and Velocity Update:** DGNSS (Differential GNSS) correction is applied to the pseudo-range, compare with updated position using doppler-velocity to calculate residuals and update the position. Similarly, compare TDCP measurements with Doppler-Aided Velocity to calculate residuals and update the velocity.\n\n\n# Training Dataset and CV Score\nBefore applying the **Test Dataset**, learning was conducted using the **All Training Dataset**. It has been confirmed that very different characteristics appear depending on the type of smartphone chipset and whether or not it supports L5 frequency. Smartphones included in the test dataset were categorized by chipset and frequency support.\n- **Qualcomm L1/L5:** pixel4, pixel4xl, pixel5\n- **Broadcom L1/L5:** pixel6pro, pixel7pro, sm-g988b\n- **Xiaomi L1/L5:** mi8 (xiaomimi8)\n- **Exynos L1/L5:** sm-s908b\n- **Exynos L1:** sm-a205u, sm-a505u\n- **Mediatek L1:** sm-a325f, samsunga32, samsunga325g\n\nSmartphones that support L5 show very good performance, while smartphones that support L1 single-frequency have very poor performance. I wanted to solve this problem, but the training dataset contained too little Exynos L1 data (1) and Mediatek L1 (2) data. :(\n\n\n\n# Test Dataset and Kaggle Score\nThrough the proposed algorithm, the final leaderboard score was 0.928m for public and 1.342m for private, resulting in 3rd place. Immediately after the competition I discovered a major mistake in my file submission... The late submission score after the competition is 0.890m for public and 1.191m for private.\n\n\n\n# Discussion\nThank you to everyone who participated in the competition for your hard work again.\n- **GNSS/INS integration** have been applied to some smartphones. More details about this solution will be covered in the [ION GNSS+ 2024 presentation](https://www.ion.org/gnss/abstracts.cfm?paperID=13510).\n- More analysis of **L1 single-frequency** is needed. Training Dataset contains too few **Samsung A-series** (sm-a205u, sm-a505u, samsunga32, samsunga325g). The score of Samsung A-series in Training Dataset is only 1.5~1.6m. There are a total of six Samsung A-series phones in the **Test dataset**, and it is expected that better results will be obtained if modeling is conducted on these in the future.\n- (TBC)\n\n# Acknowledgement\nI would also like to thank my team members (**Suyeol Kim**, **Taejin Youn**, **Gyeongmin Kim** and **Wonwoo Park**) and my advisor, **Byungwoon Park (byungwoon@sejong.ac.kr)**, for their great help in this competition.\n\n\n# Reference\nFor more detailed information, please refer to the site posted below.\n\n[1] J. Yun et al, \"Practical Approaches to Real-Time Position Accuracy Improvement of Android Smartphone Dual-Frequency GNSS,\" Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 2226-2234. https://doi.org/10.33012/2022.18373\n\n[2] J. Yun et al, \"Inherent Limitations of Smartphone GNSS Positioning and Effective Methods to Increase the Accuracy Utilizing Dual-Frequency Measurements,\" Sensors 2022, 22, 9879. https://doi.org/10.3390/s22249879\n\n[3] J. Yun et al, \"Elevating Android GNSS Raw Measurement Processing: A Universal RINEX Converter for Precise Post-Processing Solutions,\" Proceedings of the 36th International Techincal Meeting of the Satellite Division of The Institue of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023.\n\n\n\n\n\n\n\n\n\n\n\n\n" } }