Multi-Channel AoA Estimation Using ML Power Spectra
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the angle of arrival (AoA) of radio frequency signals face challenges such as poor estimation in multipath propagation, environment-specific disturbances, and antenna array imperfections, leading to inaccurate AoA determination.
Innovation Solution
A method utilizing a pre-trained machine learning algorithm to process power spectra from multiple frequency channels, incorporating quality indicators, to enhance AoA estimation robustness and precision, minimizing the impact of multipath effects and antenna array imperfections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional signal processing algorithms are used for determining AoA, then the system structure remains simple, but the measurement precision deteriorates in multipath propagation and NLoS/OLoS environments
Solution Approach 1:
The patent replaces conventional signal processing algorithms (mechanical/systematic approach) with a machine learning-based approach. The machine learning model is trained on measurement data from antenna arrays to directly predict AoA, substituting traditional signal processing methods with a data-driven model that achieves superior accuracy in multipath and NLoS/OLoS environments.
Solution Approach 2:
The patent transforms the AoA determination problem by changing the approach from direct signal processing to using power spectra as intermediate features. The machine learning model processes power spectra across multiple frequency channels to determine AoA, representing a parameter transformation from raw measurement data to spectral features to final angle estimation.
2Reliability
If conventional signal processing methods are used, then the computational requirements remain low, but the reliability deteriorates under environmental disturbances and antenna array imperfections
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model on extensive measurement data before actual AoA determination. The pre-training phase captures environmental characteristics and antenna array imperfections, enabling the model to reliably handle NLoS/OLoS conditions during operation without requiring additional computational resources during the actual measurement phase.
Solution Approach 2:
The patent creates a virtual model of the antenna array and environment through the machine learning model. By training on measurement data that includes various environmental conditions and antenna imperfections, the model learns to replicate accurate AoA determination even when physical conditions deviate from ideal scenarios.
3Adaptability or versatility
If different antenna array geometries are used, then the adaptability to different deployment scenarios improves, but the measurement precision deteriorates due to geometry-specific variations
Solution Approach 1:
The patent creates a universal machine learning model that can handle different antenna array geometries (uniform linear arrays, uniform circular arrays, non-uniform arrays) with the same input feature dimension. The model is trained on data from various array configurations, enabling it to universally process measurements from different geometries while maintaining consistent AoA estimation accuracy across all configurations.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The present disclosure relates to a method (100) for determining an angle of arrival, AoA, of received radio frequency, RF, measurement signals. The method comprises obtaining (101) measurement data based on the received RF measurement signals from an antenna array, wherein the RF measurement signals are representative of multiple frequency channels. The method further comprises determining (102) power spectra, comprising determining at least one power spectrum for each of the multiple frequency channels by using the measurement data. The method further comprises providing (105) a machine learning algorithm, which is pre-trained to determine an AoA based on power spectra of multiple frequency channels. The method further comprises determining (106) the AoA of the received RF measurement signals by using the machine learning algorithm and the determined power spectra.