Two-Stage UWB LOS/NLOS Classification Across Multiple Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LOS/NLOS classification models in UWB environments are not robust across different environments, requiring new models for each distinct environment, leading to performance degradation and inefficiency in applications like automotive key tracking and assistance robots.
Innovation Solution
A two-stage classification model is employed, where a pre-classification circuit determines the environment class, and a subsequent AI model classifies LOS/NLOS based on the environment class and channel impulse response (CIR) characteristics, using a convolutional neural network (CNN) for environment classification and a multilayer perceptron (MLP) for LOS/NLOS determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If specialized classification models are used for each environment, then LOS/NLOS identification accuracy is improved, but device complexity and memory requirements increase
Solution Approach 1:
The classification process is segmented into two stages: first classifying the environment type (LOS/NLOS), then using this classification to select or adjust the appropriate measurement model. This segmentation allows the system to use simpler, specialized models for each environment type rather than requiring a single complex multi-environment model, thereby maintaining high accuracy while reducing overall device complexity and memory requirements.
2Measurement precision
If new classification models are trained for each distinct environment, then classification performance is improved, but loss of time and productivity decrease
Solution Approach 1:
Environment classification is performed as a preliminary action before executing the main LOS/NLOS measurement task. By pre-classifying the environment type, the system can immediately select the appropriate pre-trained model without requiring retraining during operation. This preliminary classification step enables the system to maintain high classification performance across different environments while avoiding the time loss associated with retraining models in real-time.
3Device complexity
If a single multi-environment classification model is used, then device complexity is reduced, but measurement precision and reliability deteriorate
Solution Approach 1:
The system dynamically adapts its classification approach by first determining the environment type and then selecting or adjusting the appropriate measurement model accordingly. This dynamic adaptation allows the system to maintain high reliability and measurement precision for each specific environment type while keeping the overall device complexity manageable through modular model selection rather than requiring a single complex multi-environment model.
Data Source
AI summary
A multi-environment LOS/NLOS classifier for a UWB ranging device includes an environment classifier and an LOS/NLOS classifier. The environment classifier uses a convolutional neural network (CNN) fed by channel impulse response (CIR) of a received radio signal, cascaded with a multilayer perceptron (MLP) fed by a set of statistical characteristics extracted from the CIR. The environment classifier provides an environment class to the LOS/NLOS classifier, typically an MLP also receiving as input the output of the CNN and another set of statistical and physical characteristics extracted from the CIR. The environment class may form a model input to the MLP operating as an LOS/NLOS classifier, or it may be a modulation parameter of the MLP, typically a bias or weight modification factor of the MLP.