Radar Autofocus Using Neural Network Denoiser for Position Perturbations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed radar imaging systems face challenges in generating high-resolution radar images due to inaccurate antenna position calibration and position perturbations, leading to out-of-focus images.
Innovation Solution
The system employs a neural network denoiser-based approach to autofocus distributed antennas with unknown position perturbations, reformulating the radar autofocus problem to account for position ambiguities as spatial shift kernels, and using alternating optimization with regularizers and neural network denoisers to solve the problem efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If standard reconstruction techniques are applied without accounting for position perturbations, then the radar imaging process is simple and fast, but the radar images become out-of-focus due to position ambiguities
Solution Approach 1:
The patent applies preliminary action by pre-computing a library of possible shift kernels corresponding to different position perturbations before the imaging process. During imaging, the system selects the appropriate shift kernel based on the actual antenna positions, avoiding the need for complex real-time optimization while ensuring focus quality is maintained across different perturbation scenarios.
Solution Approach 2:
The patent uses copying by creating a library of synthetic shift kernels that replicate the effect of various position perturbations. Instead of directly solving the complex autofocus problem with unknown positions, the system copies the perturbation effects into pre-computed shift kernels and applies the appropriate one to correct the radar image, transforming an ill-posed inverse problem into a lookup and application problem.
2Device complexity
If additional constraints are imposed on the radar autofocus problem to make it tractable, then the computational complexity is reduced, but the solution may not be desirable or accurate
Solution Approach 1:
The patent resolves this contradiction by performing the complex computation of shift kernels in advance during system setup, when the radar system has full knowledge of its geometry. This preliminary computation creates a lookup table of possible corrections, so that during actual imaging with unknown position perturbations, the system only needs to search and apply pre-computed solutions rather than solving complex optimization problems in real-time.
Solution Approach 2:
The patent introduces shift kernels as intermediary objects that mediate between the known radar geometry and the unknown antenna positions. These shift kernels serve as a bridge, allowing the system to work with simplified models of position errors rather than directly handling the complex autofocus problem, thus reducing computational complexity while maintaining accuracy.
3Difficulty of detecting and measuring
If the radar autofocus problem is formulated as recovering a correct radar image from incorrect measurements, then the problem can be addressed, but the number of unknowns increases significantly making the problem ill-posed
Solution Approach 1:
The patent extracts the position error information from the measurement process and separates it into distinct shift kernels. By taking out the unknown position perturbations and representing them as separate, pre-computed shift operators, the system reduces the coupled complexity of simultaneously solving for both the radar image and position errors, transforming the ill-posed problem into a more manageable form.
Solution Approach 2:
The patent performs the extraction and parameterization of position errors in advance by pre-computing shift kernels for all possible position perturbations. This preliminary action removes the unknowns from the real-time imaging process, allowing the system to simply search and apply the appropriate pre-computed shift rather than solving for position errors during imaging.
Data Source
AI summary
The present disclosure provides a system and a method for generating a radar image of a scene. The method comprises receiving radar measurements of a scene collected from a set of antennas, wherein the set of antennas are under uncertainties caused by one or a combination of position ambiguities and clock ambiguities of each of the antennas. The method further comprises generating the radar image of the scene by solving a sparse recovery problem. The sparse recovery problem determines, until a termination condition is met, a set of image shifts of the radar image corresponding to different uncertainties of the antennas and updates an estimate of the radar image, based on the determined set of image shifts of the radar image. The sparse recovery problem is solved with a neural network denoiser that denoises a filtering of the estimate of the radar image.