Circular Convolutional Network for DOA Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing radar systems struggle to accurately determine the direction of arrival (DOA) of multiple closely spaced objects within their field of view, particularly in real-time applications with constraints on decision speed and power consumption.
Innovation Solution
The implementation of a radar system with an array of antennas, a radar transceiver, and a signal processing unit that includes multiple vector processing units. These units perform circular convolutions and apply non-linear activation functions to refine angular spectrum vectors, enabling precise DOA estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional radar signal processing methods are used, then the system is simpler and faster, but the DOA estimation precision is insufficient to distinguish closely spaced objects
Solution Approach 1:
The signal processing is divided into multiple iterative stages (L iterations) with each stage containing separate convolution operations for the antenna input vector and angular spectrum vector. This segmentation allows complex DOA estimation to be broken down into manageable processing steps that can be executed efficiently on vector processing units.
Solution Approach 2:
The patent transforms the traditional 1D signal processing approach into a 2D dimensional framework by simultaneously processing the antenna input vector across multiple antenna elements and the angular spectrum vector across frequency bins. This dimensional expansion enables the system to resolve closely spaced objects through enhanced spatial and spectral analysis.
2Measurement precision
If high-resolution DOA estimation algorithms are implemented, then closely spaced objects can be distinguished, but the processing time increases
Solution Approach 1:
Weighting vectors are pre-computed and stored in memory before the actual DOA estimation process. These weighting vectors contain pre-calculated complex exponential terms that would otherwise require computationally intensive real-time calculations. By preparing these weights in advance, the system achieves high angular resolution without the corresponding computational time penalty during live operation.
Solution Approach 2:
The patent replaces traditional iterative optimization algorithms with a direct computational approach using pre-computed weighting vectors and vector processing units. This substitution eliminates the need for time-consuming iterative searches and convergence checks, achieving high-resolution DOA estimation through direct mathematical computation that can be parallelized efficiently.
3Measurement precision
If complex signal processing operations are performed in real-time, then accurate DOA estimation is achieved, but power consumption increases
Solution Approach 1:
The computationally intensive portions of the signal processing are extracted and offloaded to dedicated vector processing units with specialized hardware accelerators. The main processor only needs to manage data flow and coordinate operations, while the heavy lifting of complex convolutions and calculations is performed by energy-efficient vector processing hardware designed specifically for this type of computation.
Solution Approach 2:
The system changes the computational parameters by using fixed pre-computed weighting vectors instead of dynamically calculated ones. This parameter fixation allows the vector processing units to use optimized instruction sequences and hardware configurations that minimize power consumption while maintaining calculation accuracy. The trade-off of storing more pre-computed data in memory is favorable compared to the continuous power consumption of real-time complex calculations.
Data Source
AI summary
Embodiments of systems and methods for estimating direction of arrival are disclosed. A device includes a signal processing unit that includes processing circuitry and memory coupled to the processing circuitry, where the processing circuitry includes multiple vector processing units, each vector processing unit configured to receive an antenna input vector, receive an angular spectrum vector, retrieve a first and second weighting vectors from the memory, generate a processed antenna input vector by performing a circular convolution of the antenna input vector with the first weighting vector, generate a processed angular spectrum vector by performing a circular convolution of the angular spectrum vector with the second weighting vector, and generate a refined angular spectrum vector, which indicates angular position of one or more radar targets, by applying a non-linear activation function to a sum of the processed antenna input vector and the processed angular spectrum vector.