Abstract: Disclosed are techniques for wireless positioning. In some aspects, a user equipment (UE) may obtain one or more global navigation satellite system (GNSS) measurements. The UE may obtain map data. The UE may obtain a most likely position of the UE based on one or more prediction algorithms and one or more neural network (NN) models applied to the one or more GNSS measurements and the map data.

Abstract: A method of determining a position of a device includes obtaining an initial position of the device without using Global Navigation Satellite System (GNSS) satellites. GNSS measurements are taken of radio frequency (RF) signals transmitted by the GNSS satellites. Initial residuals are determined based, at least in part, on GNSS measured distances determined from the at least a portion of the GNSS measurements and expected distances determined from the initial position. Errors of the GNSS measurements based on the RF signals are estimated. An optimization is performed using some of the estimated errors to produce a modified set of residuals, wherein the optimization is further based on H, wherein H represents a matrix with trigonometric functions of a geometry of the GNSS satellites. A cost minimization method of the modified set of residuals and actual geometry of the GNSS satellites (H) to determine an improved position of the device.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for cache aware dynamic module selection for a computation model. An example method generally includes generating at least one output, in a first inference round, using a first subset of modules of a computational model loaded in a cache memory from another memory, evaluating modules of the computational model to use for a second inference round, using a function that biases evaluation of the first subset of modules of the computational model already in the cache, and performing the second inference round with a second subset of modules of the computational module, based on the evaluation.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for improved machine learning. A sequence of data records is accessed, each data record comprising wireless channel measurements and inertial measurement unit (IMU) data. Known position information corresponding to at least a first data record is accessed. A first sequence of positions is determined by processing the sets of IMU data and known position information using a forward operation. A second sequence of positions is determined by processing the sets of IMU data and known position information using a backward operation. An IMU adjustment parameter is generated using the first and second sequences of positions. A pseudo-label is generated for a second data record using the IMU adjustment parameter and the sets of IMU data. A machine learning model is trained, using the second data record and the pseudo-label, to predict positions using one or more wireless channel measurements.

Abstract: A processor-implemented method for multimodal beam management implemented by a network device includes receiving, by the network device, a stream of inputs from one or more sensors. The network device generates a digital twin modeling an environment of a region observed by the one or more sensors. The digital twin includes one or more objects detected based on the stream of inputs. The network device manages a wireless communication signal beam for communicating with at least one user equipment (UE) in the region observed by the one or more sensors based at least in part on the digital twin.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for biometric authentication using an anti-spoofing protection model refined using online data. The method generally includes receiving a biometric data input for a user. Features for the received biometric data input are extracted through a first machine learning model. It is determined, using the extracted features for the received biometric data input and a second machine learning model, whether the received biometric data input for the user is authentic or inauthentic. It is determined whether to add the extracted features for the received biometric data input, labeled with an indication of whether the received biometric data input is authentic or inauthentic, to a finetuning data set. The second machine learning model is adjusted based on the finetuning data set.

Abstract: Aspects described herein provide a method of performing guided training of a neural network model, including: receiving supplementary domain feature data; providing the supplementary domain feature data to a fully connected layer of a neural network model; receiving from the fully connected layer supplementary domain feature scaling data; providing the supplementary domain feature scaling data to an activation function; receiving from the activation function supplementary domain feature weight data; receiving a set of feature maps from a first convolution layer of the neural network model; fusing the supplementary domain feature weight data with the set of feature maps to form fused feature maps; and providing the fused feature maps to a second convolution layer of the neural network model.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for biometric authentication using an anti-spoofing protection model refined using online data. The method generally includes receiving a biometric data input for a user. Features for the received biometric data input are extracted through a first machine learning model. It is determined, using the extracted features for the received biometric data input and a second machine learning model, whether the received biometric data input for the user is authentic or inauthentic. It is determined whether to add the extracted features for the received biometric data input, labeled with an indication of whether the received biometric data input is authentic or inauthentic, to a finetuning data set. The second machine learning model is adjusted based on the finetuning data set.

Abstract: A processor-implemented method for learning an antenna offset for perception-aided wireless communication includes receiving a stream of inputs from one or more sensors. A dynamic segmentation mask corresponding to an object observed by the one or more sensors is generated based on the stream of inputs. A trajectory for the one or more sensors is determined based on the dynamic segmentation mask. An antenna position for the object is predicted based on the trajectory.

Abstract: A processor-implemented method for beam management using region information and region-specific codebook generation includes receiving a stream of inputs from one or more sensors. A region of a user equipment (UE) is determined using a digital twin that models an environment observed by the network device based on the stream of inputs. The region is determined based on a position of the UE in the environment. A beam estimate is generated based on a codebook selected based on the region.

Abstract: A processor-implemented method for multimodal beam management implemented by a network device includes receiving, by the network device, a stream of inputs from one or more sensors. The network device generates a digital twin modeling an environment of a region observed by the one or more sensors. The digital twin includes one or more objects detected based on the stream of inputs. The network device manages a wireless communication signal beam for communicating with at least one user equipment (UE) in the region observed by the one or more sensors based at least in part on the digital twin.

Abstract: Aspects described herein provide a method of performing guided training of a neural network model, including: receiving supplementary domain feature data; providing the supplementary domain feature data to a fully connected layer of a neural network model; receiving from the fully connected layer supplementary domain feature scaling data; providing the supplementary domain feature scaling data to an activation function; receiving from the activation function supplementary domain feature weight data; receiving a set of feature maps from a first convolution layer of the neural network model; fusing the supplementary domain feature weight data with the set of feature maps to form fused feature maps; and providing the fused feature maps to a second convolution layer of the neural network model.

Abstract: In some implementations, a global navigation satellite system (GNSS) device may determine its approximate location, and, for each pseudorange measurement of a plurality of pseudorange measurements performed by the GNSS device: determine a location of a respective satellite vehicle (SV) that transmits a respective GNSS signal of which the pseudorange measurement is performed, and determine a respective residual grid, where the respective residual grid is based on respective information from the pseudorange measurement and the location of the respective SV, and the respective residual grid is indicative of possible locations of the GNSS device within a geographical region including the approximate location of the GNSS device. The GNSS device may aggregate the residual grids corresponding to at least a portion of the plurality of pseudorange measurements and may determine a location estimate of the GNSS device based on the aggregation of the residual grids.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for improved machine learning. A sequence of data records is accessed, each data record comprising wireless channel measurements and inertial measurement unit (IMU) data. Known position information corresponding to at least a first data record is accessed. A first sequence of positions is determined by processing the sets of IMU data and known position information using a forward operation. A second sequence of positions is determined by processing the sets of IMU data and known position information using a backward operation. An IMU adjustment parameter is generated using the first and second sequences of positions. A pseudo-label is generated for a second data record using the IMU adjustment parameter and the sets of IMU data. A machine learning model is trained, using the second data record and the pseudo-label, to predict positions using one or more wireless channel measurements.

Abstract: A method of determining a position of a device includes obtaining an initial position of the device without using Global Navigation Satellite System (GNSS) satellites. GNSS measurements are taken of radio frequency (RF) signals transmitted by the GNSS satellites. Initial residuals are determined based, at least in part, on GNSS measured distances determined from the at least a portion of the GNSS measurements and expected distances determined from the initial position. Errors of the GNSS measurements based on the RF signals are estimated. An optimization is performed using some of the estimated errors to produce a modified set of residuals, wherein the optimization is further based on H, wherein H represents a matrix with trigonometric functions of a geometry of the GNSS satellites. A cost minimization method of the modified set of residuals and actual geometry of the GNSS satellites (H) to determine an improved position of the device.

Abstract: Some disclosed methods involve obtaining current A-line data corresponding to reflections of ultrasonic waves from a target object detected by a single receiver pixel, obtaining current ultrasonic fingerprint image data corresponding to reflections of ultrasonic waves from a target object surface, obtaining previously-obtained A-line data that was previously obtained from an authorized user, and obtaining previously-obtained ultrasonic fingerprint image data that was previously obtained from the authorized user. Some disclosed methods involve estimating, based at least in part on the current A-line data, the previously-obtained A-line data, the current ultrasonic fingerprint image data and the previously-obtained ultrasonic fingerprint image data, whether the target object is a finger of the authorized user. The estimation may involve an anti-spoofing process based at least in part on the current A-line data and the previously-obtained A-line data.

Abstract: Disclosed are techniques for fusing camera and radar frames to perform object detection in one or more spatial domains. In an aspect, an on-board computer of a host vehicle receives, from a camera sensor of the host vehicle, a plurality of camera frames, receives, from a radar sensor of the host vehicle, a plurality of radar frames, performs a camera feature extraction process on a first camera frame of the plurality of camera frames to generate a first camera feature map, performs a radar feature extraction process on a first radar frame of the plurality of radar frames to generate a first radar feature map, converts the first camera feature map and/or the first radar feature map to a common spatial domain, and concatenates the first radar feature map and the first camera feature map to generate a first concatenated feature map in the common spatial domain.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for beam selection using machine learning. A plurality of data samples corresponding to a plurality of data modalities is accessed. A plurality of features is generated by, for each respective data sample of the plurality of data samples, performing feature extraction based at least in part on a respective modality of the respective data sample. The plurality of features is fused using one or more attention-based models, and a wireless communication configuration is generated based on processing the fused plurality of features using a machine learning model.

Abstract: Disclosed are techniques for employing deep learning to analyze radar signals. In an aspect, an on-board computer of a host vehicle receives, from a radar sensor of the vehicle, a plurality of radar frames, executes a neural network on a subset of the plurality of radar frames, and detects one or more objects in the subset of the plurality of radar frames based on execution of the neural network on the subset of the plurality of radar frames. Further, techniques for transforming polar coordinates to Cartesian coordinates in a neural network are disclosed. In an aspect, a neural network receives a plurality of radar frames in polar coordinate space, a polar-to-Cartesian transformation layer of the neural network transforms the plurality of radar frames to Cartesian coordinate space, and the neural network outputs the plurality of radar frames in the Cartesian coordinate space.

Abstract: Disclosed are techniques for fusing camera and radar frames to perform object detection in one or more spatial domains. In an aspect, an on-board computer of a host vehicle receives, from a camera sensor of the host vehicle, a plurality of camera frames, receives, from a radar sensor of the host vehicle, a plurality of radar frames, performs a camera feature extraction process on a first camera frame of the plurality of camera frames to generate a first camera feature map, performs a radar feature extraction process on a first radar frame of the plurality of radar frames to generate a first radar feature map, converts the first camera feature map and/or the first radar feature map to a common spatial domain, and concatenates the first radar feature map and the first camera feature map to generate a first concatenated feature map in the common spatial domain.