Radar DoA Estimation via ML Covariance Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Radar systems face performance degradation in direction of arrival (DoA) estimation due to correlation between range-Doppler-map data across channels, especially when additional virtual channels are introduced, leading to reduced accuracy in resolving nearby targets.
Innovation Solution
A radar system that uses a machine learning model, specifically a neural network, to generate additional spatial covariance matrices for different chirp center frequencies, which are then combined with the original spatial covariance matrix to increase the rank and improve DoA estimation without requiring additional bandwidth or physical antenna extensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If additional virtual channels are introduced to improve DoA estimation accuracy, then the ability to resolve nearby targets is enhanced, but signal correlation between channels increases leading to performance degradation
Solution Approach 1:
The patent introduces a signal processing intermediary (spatial covariance matrix transformation) between the received channel signals and the DoA estimation process. By transforming the spatial covariance matrix to decorrelate the channel signals, the system mediates the conflict between utilizing multiple virtual channels for improved accuracy and avoiding their strong correlation that degrades performance.
2Measurement precision
If the number of receive antennas is increased to improve DoA resolution, then the aperture is increased and nearby targets can be resolved better, but the device complexity and cost increase
Solution Approach 1:
The patent creates virtual copies of antenna channels through signal processing techniques. By generating virtual channels from existing physical antenna signals and applying spatial covariance matrix transformation, the system achieves the effect of having more antenna elements without physically adding more antennas, thus copying the functionality of a larger aperture system.
Solution Approach 2:
The patent transforms the problem from the spatial domain to the covariance matrix domain. By operating on the spatial covariance matrix rather than directly on the antenna signals, the system introduces a mathematical dimension that allows aperture extension and target resolution improvement without physically increasing the antenna array dimensions.
3Measurement precision
If bandwidth is increased to improve DoA estimation, then the resolution and accuracy are enhanced, but the use of energy and system complexity increase
Solution Approach 1:
The patent changes the processing parameters from time-domain signal processing to covariance matrix domain processing. By transforming and processing the spatial covariance matrix instead of increasing bandwidth, the system achieves improved DoA estimation accuracy through parameter transformation rather than increasing the energy-consuming bandwidth parameter.
Data Source
AI summary
According to various embodiments, a radar system is described including a direction of arrival pre-processor configured to, for a detected peak, obtain a Doppler Fourier transform result vector, generate a spatial covariance matrix for the Doppler Fourier transform result vector, and generate an additional spatial covariance matrix by inputting the spatial covariance matrix to a machine learning model trained to predict, from an input spatial covariance matrix, an output spatial covariance matrix such that the output spatial covariance matrix corresponds to a different chirp center frequency than the input covariance spatial covariance matrix and including a direction of arrival estimator configured to perform direction-of-arrival estimation using the additional spatial covariance matrix.