Multipath mitigation in GNSS receivers with machine learning models
Machine learning techniques are used, in one embodiment, to mitigate multipath in an L5 GNSS receiver. In one embodiment, training data is generated to provide ground truth data for excess path length (EPL) corrections for a set of received GNSS signals. A system extracts features from the set of received GNSS signals and uses the extracted features and the ground truth data to train a set of one or more neural networks that can produce EPL corrections for pseudorange measurements. The trained set of one or more neural networks can be deployed in GNSS receivers and used in the GNSS receivers to correct pseudorange measurements using EPL corrections provided by the trained set of neural networks.
1. A method of operating a global navigation satellite system (GNSS) receiver, the method comprising:
receiving GNSS signals from one or more GNSS satellites (SVs);
extracting a set of features from the received GNSS signals, the set of features being predetermined based on a trained model in the GNSS receiver;
applying the set of features as an input to the trained model;
generating, by the trained model, a set of one or more excess path length (EPL) corrections to correct one or more pseudorange measurements made by the GNSS receiver.
2. The method as in claim 1 , wherein the set of features comprises: (1) correlation vector features from a set of successive correlation vectors from correlators, in the GNSS receiver, for GNSS signals from a GNSS SV; (2) a mutual probability distribution function (pdf) data of the relative delay and relative amplitude from each of the correlators; (3) other features relating to the GNSS receiver or GNSS SV.
3. The method as in claim 2 , wherein the correlation vector features comprise one or more of: (1) relative amplitude of local maximum points in a correlation vector; (2) relative delay of peaks in a correlation vector; (3) a width of a strongest peak in a correlation vector; or (4) a number of strong peaks in a correlation vector.
4. The method as in claim 3 , wherein the mutual pdf is in a 2D (two dimension) matrix format.
5. The method as in claim 4 , wherein the other features comprise one or more of: (1) an elevation of a GNSS SV; (2) a signal to noise ratio (SNR) of a measured pseudorange; (3) a type of antenna used to receive the GNSS signals; (4) a tracking mode of the GNSS receiver; or (5) a tracking loop configuration of the GNSS receiver.
6. The method as in claim 5 , wherein the trained model comprises a set of one or more neural networks, and wherein the set of one or more neural networks comprise: (1) a first convolution neural network (CNN) to receive as inputs the mutual pdf data; (2) a second CNN to receive as inputs the correlation vector features; and (3) a first neural network to receive the other features.
7. The method as in claim 6 , wherein the method further comprises:
pruning data in the set of features;
balancing data in the set of features;
scaling data in the set of features.
8. The method as in claim 1 wherein the trained model comprises a set of one or more neural networks.
9. The method as in claim 1 wherein each of the one or more pseudoranges is corrected by subtracting a corresponding EPL correction from a measured pseudorange from a GNSS SV, and wherein the GNSS signals are wideband GNSS signals that have a chipping rate that exceeds a chipping rate for L1 GPS signals, and wherein the measured pseudorange is measured in a delay locked loop in the GNSS receiver and wherein the one or more EPL corrections are derived based on outputs at the delay locked loop (DLL) which is a pseudorandom noise (PRN) tracking loop to track pseudoranges in the GNSS receiver.
10. The method as in claim 1 , wherein each of the one or more corrected pseudoranges is used in position solution calculations by a position engine, and wherein the set of one or more EPL corrections correct for multipath reflections of GNSS signals in an urban canyon that surrounds the GNSS receiver.
11. A non-transitory machine readable medium storing executable program instructions which when executed by a data processing system cause the data processing system to perform a method, the method comprising:
extracting a set of features from received global navigation satellite system (GNSS) signals, the set of features being predetermined based on a trained model in a GNSS receiver;
applying the set of features as an input to the trained model;
generating, by the trained model, a set of one or more excess path length (EPL) corrections to correct one or more pseudorange measurements made by the GNSS receiver.
12. The non-transitory machine readable medium as in claim 11 , wherein each of the one or more pseudoranges is corrected by subtracting a corresponding EPL correction from a measured pseudorange from a GNSS SV, and wherein the GNSS signals are wideband GNSS signals that have a chipping rate that exceeds a chipping rate for L1 GPS signals, and wherein the measured pseudorange is measured in a pseudorange measurement system in the GNSS receiver and wherein the one or more EPL corrections are derived based on outputs from the pseudorange measurement system in the GNSS receiver, and wherein the set of features comprises: (1) correlation vector features from a set of successive correlation vectors from correlators, in the GNSS receiver, for GNSS signals from a GNSS SV; (2) a mutual probability distribution function (pdf) data of the relative delay and relative amplitude from each of the correlators; (3) other features relating to the GNSS receiver or GNSS SV.
13. The non-transitory machine readable medium as in claim 12 , wherein the correlation vector features comprise one or more of: (1) relative amplitude of local maximum points in a correlation vector; (2) relative delay of peaks in a correlation vector; (3) a width of a strongest peak in a correlation vector; or (4) a number of strong peaks in a correlation vector.
14. The non-transitory machine readable medium as in claim 13 , wherein the mutual pdf is in a 2D (two dimension) matrix format.
15. The non-transitory machine readable medium as in claim 14 , wherein the other features comprise one or more of: (1) an elevation of a GNSS SV; (2) a signal to noise ratio (SNR) of a measured pseudorange; (3) a type of antenna used to receive the GNSS signals; (4) a tracking mode of the GNSS receiver; or (5) a tracking loop configuration of the GNSS receiver.
16. The non-transitory machine readable medium as in claim 15 ,
wherein the trained model comprises a set of one or more neural networks, and wherein the set of one or more neural networks comprise: (1) a first convolution neural network (CNN) to receive as inputs the mutual pdf data; (2) a second CNN to receive as inputs the correlation vector features; and (3) a first neural network to receive the other features.
17. The non-transitory machine readable medium as in claim 16 , wherein the method further comprises:
pruning data in the set of features;
balancing data in the set of features;
scaling data in the set of features.
18. The non-transitory machine readable medium as in claim 11 , wherein the trained model comprises a set of one or more neural networks.