Neural Network UE Positioning Fusing Sensor Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional RAT-assisted UE positioning techniques face challenges in achieving high accuracy due to excessive complexity, resource consumption, and the lack of integration with UE sensor data, leading to inefficient positioning in urban and indoor environments.
Innovation Solution
The implementation of jointly trained neural networks that fuse reference signal measurements with UE local sensor data to generate more accurate and meaningful position estimates, utilizing a set of neural networks trained for end-to-end processing in UE positioning, including reference signal transmission, measurement, and sensor reporting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional RAT-assisted UE positioning techniques are used, then positioning can be achieved, but accuracy is insufficient due to excessive complexity and lack of sensor data integration
Solution Approach 1:
The patent merges reference signal measurements with UE sensor data (accelerometer, gyroscope, magnetometer) by feeding both types of data into the neural network. This combination allows the system to achieve higher positioning accuracy by leveraging multiple data sources while the neural network handles the complexity of integrating these diverse data types.
Solution Approach 2:
The patent replaces conventional signal processing mechanisms with a neural network-based system. Instead of using traditional algorithms to process reference signals and calculate position, the system uses a neural network that automatically learns patterns from both reference signals and sensor data, thereby improving accuracy while abstracting away the complexity through the learning model.
2Productivity
If conventional positioning methods are used, then basic positioning functionality is provided, but resource consumption is excessive
Solution Approach 1:
The neural network is pre-trained offline using extensive simulation data and labeled positioning information. This preliminary training allows the network to make efficient predictions during actual operation without requiring heavy real-time computation or energy consumption. The complex pattern recognition is performed once during training, then executed efficiently during inference.
Solution Approach 2:
The patent substitutes energy-intensive conventional signal processing algorithms with a neural network model that, once trained, requires significantly fewer computational resources. The neural network's optimized weight matrices and activation functions enable faster, more energy-efficient positioning calculations compared to traditional multi-step signal processing pipelines.
3Reliability
If GNSS is used for positioning, then high accuracy can be achieved in open environments, but it fails in urban and indoor environments due to interference and multi-path losses
Solution Approach 1:
The patent creates a universal positioning system that can operate across multiple environments (outdoor, urban, indoor) by integrating multiple data sources: reference signals from the cellular network, sensor data from the UE, and optionally GNSS data. The neural network learns to weigh and combine these sources appropriately, providing reliable positioning regardless of which sources are available or reliable in a given environment.
Solution Approach 2:
The neural network acts as an intermediary that processes and fuses multiple input signals (reference signals and sensor data) to produce robust position estimates. This intermediary layer filters out the harmful effects of signal interference and multi-path losses by learning patterns that compensate for these distortions, enabling reliable positioning even when traditional signal-based methods fail.
Data Source
AI summary
A wireless communication system employs DNNs or other neural networks to provide for RAT-assisted positioning of UEs. A TX DNN at the BS generates and provides for wireless transmission of a reference signal to the UE. An RX DNN at the UE receives the reference signal and local UE sensor data as input, and from this input generates a UE measurement and sensor report. A TX DNN at the UE receives the report as an input, and from this input generates an RF signal representing the UE measurement and sensor report for transmission to the BS. An RX DNN at the BS receives the report from the RF signal as input, and from this input generates a position estimate of the UE.