Neural Network Localization via Sub-Area Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CSI-based localization methods require extensive and costly measurement campaigns for high accuracy, are resource-intensive, and lack versatility when environmental changes occur, such as changes in coverage area or wireless channel conditions.
Innovation Solution
The approach involves dividing the coverage area into smaller blocks or sub-areas, training separate neural networks for each sub-area, and activating multiple neural networks for CSI input in the online phase to determine a final location estimation, thereby reducing computational complexity and enhancing resilience to environmental changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is trained on the entire coverage area to achieve high localization accuracy, then the localization precision is improved, but the computational complexity and memory footprint increase significantly
Solution Approach 1:
The coverage area is divided into multiple sub-areas, and separate neural networks are trained for each sub-area instead of training one large network for the entire area. This segmentation reduces the computational complexity and memory footprint of each individual network while maintaining high localization accuracy within each sub-area.
2Measurement precision
If extensive measurement campaigns are conducted to build comprehensive CSI databases for high accuracy localization, then the localization precision is improved, but the time and resources required for data collection increase
Solution Approach 1:
The measurement campaign is divided into multiple smaller campaigns, each collecting CSI data for a specific sub-area. This allows parallel or sequential data collection for different sub-areas, reducing the total time required compared to conducting one comprehensive campaign for the entire coverage area.
Solution Approach 2:
CSI databases are pre-built for each sub-area before deployment. During the online phase, the system can quickly determine location by querying pre-trained neural networks without requiring real-time analysis of raw CSI data, significantly reducing localization time.
3Adaptability or versatility
If a single neural network is trained for the entire coverage area, then the model can handle all locations, but the system lacks adaptability when environmental changes occur in specific areas
Solution Approach 1:
The system uses multiple independent neural networks, each trained for a specific sub-area. When environmental changes occur in one sub-area, only the corresponding neural network needs to be retrained or updated, while other networks remain unaffected. This modular approach enhances adaptability without requiring complete system retraining.
Solution Approach 2:
Each neural network is optimized for its specific sub-area with local characteristics. This allows each network to develop specialized features and parameters tailored to its local environment, improving both accuracy and adaptability to local changes without affecting other sub-areas.
4Adaptability or versatility
If multiple neural networks are trained for different sub-areas, then the system becomes more adaptable to environmental changes, but the number of models to manage increases
Solution Approach 1:
The coverage area is segmented into sub-areas based on geographical or environmental characteristics, creating a natural organizational structure for multiple neural networks. This segmentation provides a clear mapping between physical space and computational models, making the system more manageable despite having multiple networks.
Solution Approach 2:
A coordinate determination function serves as an intermediary that receives CSI data, identifies the relevant sub-area, and routes the data to the appropriate neural network. This mediator simplifies the management of multiple networks by providing a unified interface and automatic routing mechanism.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a computing device and a network device for estimating a location of a terminal device in a wireless network based on channel state information (CSI). In a training phase, the computing device divides a coverage area of the network device into sub-areas, and partitions labelled CSI datasets into two or more CSI data partitions based on the sub-areas. The CSI data partitions are used to train two or more neural network models. Each CSI data partition corresponds to an area that is smaller than the coverage area, so that trainings of the neural network models are eased. Furthermore, the trained neural network models are robust to environment changes. In an online phase, the network device inputs CSI of the terminal device to at least two neural network models to obtain an optimal estimation based on outputs of the at least two neural network models.