Position-grid based machine learning for GNSS warm-start position accuracy improvement
Aspects presented herein may improve the performance and accuracy of GNSS-based positioning, where a position-grid based ML may be implemented by a UE or a location server to improve the accuracy of identifying a warm-start position of the UE. In one aspect, a UE or a location server determines, for each grid point within a range of an initial position of a UE, a set of PR residuals based on PRs for each SV of a set of SVs. The UE or the location server determines an estimated position of the UE based on the sets of determined PR residuals.
1. An apparatus for position estimation, comprising:
at least one memory; and
at least one processor coupled to the at least one memory and, based at least in part on information stored in the at least one memory, the at least one processor is configured to:
determine, for each grid point within a range of an initial position of a user equipment (UE), a set of pseudorange (PR) residuals based on PRs for each space vehicle (SV) of a set of SVs;
determine, based on a machine learning (ML) classifier, a likelihood of whether the UE is approximate to a grid point based on a distribution pattern of PR residuals in a corresponding set of PR residuals; and
determine an estimated position of the UE based on the likelihood of whether the UE is approximate to the grid point.
2. The apparatus of claim 1 , wherein the PRs comprise a measured PR and a predicted PR, and each PR residual in the set of PR residuals is a measure of a difference between the measured PR and the predicted PR.
3. The apparatus of claim 1 , wherein the at least one processor is further configured to measure, for each SV of the set of SVs, the PR between the UE and the SV.
4. The apparatus of claim 1 , wherein the at least one processor is further configured to predict, for each grid point and for each SV of the set of SVs, the PR between the UE and the SV at the grid point.
5. The apparatus of claim 4 , wherein to determine the set of PR residuals, the at least one processor is configured to determine each PR residual in the set of PR residuals based on a difference between a measured PR and a predicted PR.
6. The apparatus of claim 1 , wherein to determine, based on the ML classifier, the likelihood of whether the UE is approximate to the grid point based on the distribution pattern of PR residuals in the corresponding set of PR residuals, the at least one processor is configured to:
input, to the ML classifier, the distribution pattern of PR residuals in the corresponding set of PR residuals; and
receive, from the ML classifier as inference output, the likelihood of whether the UE is approximate to the grid point.
7. The apparatus of claim 1 , wherein the distribution pattern of PR residuals in the corresponding set of PR residuals is different when the grid point is closer to an actual position of the UE compared to the grid point is further away from the actual position of the UE.
8. The apparatus of claim 1 , wherein the at least one processor is further configured to:
determine for each SV of the set of SVs at least a geometric orientation with respect to the UE; and
determine, based on the ML classifier or a second ML classifier and the determined geometric orientation with respect to the UE for each SV of at least one subset of the set of SVs, a relative PR weight for each SV of the set of SVs,
wherein to determine the estimated position of the UE, the at least one processor is further configured to determine the estimated position of the UE based on the relative PR weight for each SV of the set of SVs.
9. The apparatus of claim 8 , wherein the geometric orientation is a function of at least an azimuth angle and a zenith angle between the UE and a corresponding SV of the set of SVs.
10. The apparatus of claim 8 , wherein the relative PR weight for one SV of the set of SVs is based on a spherical distance between the one SV and each SV of the at least one subset of the set of SVs, the spherical distance between the one SV and an other SV being based on the geometric orientation of the one SV compared to the geometric orientation of the other SV.
11. The apparatus of claim 10 , wherein the relative PR weight for one SV of the set of SVs is further based on one or more of a carrier-to-noise ratio, an auto-correlation function, a code-carrier phase consistency, measurement status or error flags, consistency between different band measurements of the one SV, weighted least squares a-posteriori residuals, or signal integration information.
12. The apparatus of claim 8 , wherein the relative PR weight is based on a predicted relative PR error by the ML classifier.
13. A method of position estimation, comprising:
determining, for each grid point within a range of an initial position of a user equipment (UE), a set of pseudorange (PR) residuals based on PRs for each space vehicle (SV) of a set of SVs;
determining, based on a machine learning (ML) classifier, a likelihood of whether the UE is approximate to a grid point based on a distribution pattern of PR residuals in a corresponding set of PR residuals; and
determining an estimated position of the UE based on the likelihood of whether the UE is approximate to the grid point.
14. The method of claim 13 , wherein the PRs comprise a measured PR and a predicted PR, and each PR residual in the set of PR residuals is a measure of a difference between the measured PR and the predicted PR.
15. The method of claim 13 , further comprising measuring, for each SV of the set of SVs, the PR between the UE and the SV.
16. The method of claim 13 , further comprising predicting, for each grid point and for each SV of the set of SVs, the PR between the UE and the SV at the grid point.
17. The method of claim 16 , wherein determining the set of PR residuals comprises determining each PR residual in the set of PR residuals based on a difference between a measured PR and a predicted PR.
18. The method of claim 13 , wherein determining, based on the ML classifier, the likelihood of whether the UE is approximate to the grid point based on the distribution pattern of PR residuals in the corresponding set of PR residuals comprises:
inputting, to the ML classifier, the distribution pattern of PR residuals in the corresponding set of PR residuals; and
receiving, from the ML classifier as inference output, the likelihood of whether the UE is approximate to the grid point.
19. The method of claim 13 , wherein the distribution pattern of PR residuals in the corresponding set of PR residuals is different when the grid point is closer to an actual position of the UE compared to the grid point is further away from the actual position of the UE.
20. The method of claim 13 , further comprising:
determining for each SV of the set of SVs at least a geometric orientation with respect to the UE; and
determining, based on the ML classifier or a second ML classifier and the determined geometric orientation with respect to the UE for each SV of at least one subset of the set of SVs, a relative PR weight for each SV of the set of SVs,
wherein determining the estimated position of the UE comprises determining the estimated position of the UE based on the relative PR weight for each SV of the set of SVs.
21. The method of claim 20 , wherein the geometric orientation is a function of at least an azimuth angle and a zenith angle between the UE and a corresponding SV of the set of SVs.
22. The method of claim 20 , wherein the relative PR weight for one SV of the set of SVs is based on a spherical distance between the one SV and each SV of the at least one subset of the set of SVs, the spherical distance between the one SV and an other SV being based on the geometric orientation of the one SV compared to the geometric orientation of the other SV.
23. The method of claim 22 , wherein the relative PR weight for one SV of the set of SVs is further based on one or more of a carrier-to-noise ratio, an auto-correlation function, a code-carrier phase consistency, measurement status or error flags, consistency between different band measurements of the one SV, weighted least squares a-posteriori residuals, or signal integration information.
24. The method of claim 20 , wherein the relative PR weight is based on a predicted relative PR error by the ML classifier.
25. An apparatus for position estimation, comprising:
means for determining, for each grid point within a range of an initial position of a user equipment (UE), a set of pseudorange (PR) residuals based on PRs for each space vehicle (SV) of a set of SVs;
means for determining, based on a machine learning (ML) classifier, a likelihood of whether the UE is approximate to a grid point based on a distribution pattern of PR residuals in a corresponding set of PR residuals; and
means for determining an estimated position of the UE based on the likelihood of whether the UE is approximate to the grid point.
26. The apparatus of claim 25 , wherein the PRs comprise a measured PR and a predicted PR, and each PR residual in the set of PR residuals is a measure of an excess delay based on a comparison of the measured PR and the predicted PR.
27. The apparatus of claim 25 , further comprising means for measuring, for each SV of the set of SVs, the PR between the UE and the SV.
28. The apparatus of claim 25 , wherein the means for determining, based on the ML classifier, the likelihood of whether the UE is approximate to the grid point based on the distribution pattern of PR residuals in the corresponding set of PR residuals include configuring the apparatus to:
input, to the ML classifier, the distribution pattern of PR residuals in the corresponding set of PR residuals; and
receive, from the ML classifier as inference output, the likelihood of whether the UE is approximate to the grid point.
29. The apparatus of claim 25 , wherein the distribution pattern of PR residuals in the corresponding set of PR residuals is different when the grid point is closer to an actual position of the UE compared to the grid point is further away from the actual position of the UE.
30. A non-transitory computer-readable medium storing computer executable code, the code when executed by a processor causes the processor to:
determine, for each grid point within a range of an initial position of a user equipment (UE), a set of pseudorange (PR) residuals based on PRs for each space vehicle (SV) of a set of SVs;
determine, based on a machine learning (ML) classifier, a likelihood of whether the UE is approximate to a grid point based on a distribution pattern of PR residuals in a corresponding set of PR residuals; and
determine an estimated position of the UE based on the likelihood of whether the UE is approximate to the grid point.