Abstract: Machine learning techniques are used to perform RANSAC like processing in a GNSS receiver. A model (e.g., one or more neural networks) is trained to perform this processing to generate a selection of a subset of GNSS SVs. In one embodiment, the trained model is used during inferencing in a GNSS receiver.

Abstract: Machine learning techniques are used to compute predicted range rate errors in a GNSS receiver. In one embodiment, training data is computed to provide true range rate error data 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 true range rate error data to train a model (e.g., a set of one or more neural networks) that can produce predicted range rate errors for use in correcting measurements. The trained set of one or more neural networks can be deployed in GNSS receivers and used in the GNSS receivers to correct Doppler measurements using the predicted range rate errors provided by the trained set of one or more neural networks.

Abstract: 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.

Abstract: Machine learning techniques can be used to mitigate multipath in a GNSS receiver that includes a first trained model that provides extra path length (EPL) corrections in the GNSS receiver. The first trained model can be updated using an updated and trained model from one or more assistance servers that are in communication with the GNSS receiver. The GNSS receiver can provide, for a particular computed position and time, extracted features from received GNSS signals to the one or more assistance servers. The assistance servers can then use the extracted features and a source of true EPL corrections (e.g., from a 3D building map database for the particular computed position and time) to train a server model. The server model, once trained to a desired level of accuracy, can be transmitted to the GNSS receiver to replace the first trained model. The server model can be compared to the first trained model to verify it can provide more accurate EPL corrections than the first trained model.

Abstract: This disclosure describes methods, systems and machine readable media that can provide position solutions using, for example, pattern matching with GNSS signals in urban canyons. In one method, based upon an approximate location in an urban canyon and a set of 3D data about building structures in the urban canyon, an expected signal reception data can be generated for both line of sight and non-line of sight GNSS signals from GNSS satellites, or other sources of GNSS signals, at each point in a set of points in a grid (or other model) in the vicinity of the approximate location). This expected signal reception data can be matched to a received set of GNSS signals that have been received by a GNSS receiver, and the result of the matching can produce an adjustment to the approximate location that is used in the position solution of the GNSS receiver.

Abstract: A vehicular simultaneous localization and mapping may fuse lidar data and pseudoranges extracted from ambient cellular LTE towers. An ICP algorithm may be used to extract odometry measurements from successive lidar scans. A robust and computationally efficient feature extraction method may be used to detect edge lines and feature points from the lidars point cloud. Then, a point registration technique using a maximum likelihood approach allows the estimation of the covariance of the odometry error, which is used in EKF propagation. The proposed approach consists of a mapping mode when GNSS signals are available and subsequently a SLAM mode when GNSS signals become unavailable. The cellular transmitters states, namely position and clock bias and clock drift, are continuously estimated in both modes. Simulation and experimental results validate the accuracy of these systems and methods, and provides lane-level localization without GNSS signals.