UE Orientation Detection Using Network-Trained ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing UE orientation detection methods relying on internal sensors face accuracy issues due to jitter and magnetic interference, especially in indoor environments, and require excessive overhead in RAT-based feedback, necessitating a more reliable and efficient alternative.
Innovation Solution
A secondary method using a machine learning (ML) model trained on UE measurements and orientation data, transferred from a server to the UE via a base station, to determine orientation without continuous RAT feedback, reducing reliance on internal sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If internal sensors are used for orientation detection in UE, then the system can obtain orientation information, but the reliability is insufficient particularly in indoor industrial environments
Solution Approach 1:
The patent introduces an intermediary ML model that bridges the gap between unreliable sensor data and accurate orientation detection. The model acts as a mediator that processes multiple input features (signal strength, channel state information, historical orientation) to produce reliable orientation estimates when sensors fail or provide inaccurate readings in challenging environments like indoor industrial settings
Solution Approach 2:
The patent replaces the mechanical sensor-based orientation detection system with a software-based ML model. Instead of relying on physical sensors that are prone to failure in certain environments, the system substitutes them with an intelligent algorithm that processes wireless communication signals to infer orientation, thereby eliminating the reliability limitations of hardware sensors
2Measurement precision
If ML model training data is collected and processed at the network side, then orientation detection accuracy is improved, but network resource consumption and latency increase
Solution Approach 1:
The patent applies preliminary action by pre-training the ML model offline using collected feature vectors and orientation data. The model is trained in advance during periods when accurate orientation information is available, and then the trained model is transferred to the UE for real-time inference. This separates the time-consuming training phase from the real-time operation phase, eliminating training latency during actual orientation detection
Solution Approach 2:
The patent segments the ML model development process into distinct phases: data collection, offline training, model validation, and deployment. By dividing the process into manageable segments, the system can perform resource-intensive training operations during idle periods or using dedicated computing resources, while the deployed model provides rapid orientation detection at the UE without consuming network resources during inference
3Use of energy by moving object
If ML model is deployed at UE for orientation detection, then power consumption is reduced and latency is lowered, but device complexity increases
Solution Approach 1:
The patent uses copying by transferring the trained ML model from the network side to a copy at the UE. The original model training is performed once at the network using powerful computing resources, and then a copy of the trained model is deployed to the UE. This allows the UE to perform lightweight inference operations locally without repeating the computationally intensive training process, thereby reducing both power consumption and device complexity requirements
Data Source
AI summary
Various techniques are provided for receiving, by a base station (BS) from a user equipment (UE), a communication including a feature vector, storing, by the BS, a dataset including one or more feature vectors associated with the UE, communicating, by the BS to a network device, the dataset associated with the UE, receiving, by the BS from the network device, a machine learning (ML) model, the ML model being trained, using the dataset, to detect UE orientation, and communicating, by the BS to the UE, the trained ML model.