Hierarchical Neural Network Location Classifier
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing location determination methods using fingerprinting face challenges with high memory and computation requirements, impracticality for large areas, and limited scalability, especially with flat neural network classifiers, which struggle with convergence and accuracy in outdoor scenarios due to factors of variation and multipath interference.
Innovation Solution
A hierarchical neural network location classifier is designed and trained using a hierarchical architecture, reducing the number of parameters and improving learning efficiency by exploiting environmental hierarchies, characterizing clutter elements, and using ray-tracing for environment characterization, thereby improving location determination accuracy and reducing computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a flat neural network location classifier is used for location determination, then the system can process location data, but it suffers from high memory requirements, high computation overhead, and poor convergence accuracy especially in outdoor scenarios
Solution Approach 1:
The patent segments the location determination task into two hierarchical stages: first predicting a coarse location region, then predicting a fine location within that region. This segmentation divides the originally complex single-step classification into manageable sub-tasks, reducing the computational burden and memory requirements while improving overall accuracy, particularly in outdoor environments with multipath interference.
Solution Approach 2:
The patent introduces a hierarchical dimension to the location determination process by organizing location predictions into multiple levels (coarse and fine). This dimensional transformation allows the system to handle large-scale outdoor areas more effectively by breaking down the vast search space into smaller, more manageable regions, thereby reducing computational complexity without sacrificing precision.
2Reliability
If a flat neural network classifier is used, then the implementation is straightforward, but it struggles with convergence and accuracy in outdoor scenarios due to factors of variation and multipath interference
Solution Approach 1:
By segmenting the classification task into hierarchical stages, the system improves reliability in outdoor scenarios where flat classifiers fail due to multipath interference and environmental variations. The two-stage approach allows the model to first identify broad regions with more stable characteristics, then refine predictions within those regions, thereby achieving better convergence and accuracy despite increased implementation complexity.
3Area of stationary object
If a traditional fingerprinting method is used for location determination, then the method can determine location, but it becomes impractical for large areas due to high memory requirements and limited scalability
Solution Approach 1:
The hierarchical neural network classifier segments the large coverage area into coarse and fine location regions, allowing the system to scale to large areas without proportionally increasing memory requirements. Instead of storing and processing fingerprints for every possible location point, the system stores hierarchical representations that summarize location characteristics at different scales, dramatically improving scalability.
Solution Approach 2:
The patent transforms the traditional flat location space into a hierarchical dimensional structure with multiple levels of granularity. This dimensional change enables the system to efficiently represent and process large coverage areas by organizing location information hierarchically, reducing the memory burden while expanding the practical coverage area for location determination.
Data Source
AI summary
Some embodiments of the present disclosure provide a neural network location classifier that is designed and trained in accordance with a hierarchical architecture, thereby producing a hierarchical neural network location classifier. Further embodiments relate to obtaining, through use of the hierarchical neural network location classifier, an inferred hierarchical label for a user equipment location. The inferred hierarchical label may then be decoded to obtain a location.