Radar Range-Doppler Learning for Low-SNR Clutter Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radar signal processing techniques are less effective due to short pulse width and low signal-to-noise ratio (SNR) in radar return signals, leading to reduced detection range and increased noise interference.
Innovation Solution
A physics enhanced dynamic reservoir computer system that includes an online learning layer with gradient descent algorithm, L2 and L1 norm penalty terms, and a neural combiner to improve SNR and eliminate noise, enabling real-time radar range-doppler map learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Length of moving object
If matched filtering with long coherent/incoherent pulse integration is used, then detection range is improved, but signal processing time and system complexity increase
Solution Approach 1:
The patent replaces traditional mechanical signal processing approaches (matched filtering, coherent/incoherent integration) with a reservoir computing system that uses recurrent neural networks to process radar signals. This substitution enables real-time processing of short pulse width signals while maintaining detection performance, thereby reducing processing time and system complexity compared to conventional methods.
Solution Approach 2:
The invention changes the fundamental processing parameters by using very short pulse widths (microsecond or sub-microsecond range) compared to traditional long pulse integration methods. This parameter change, combined with reservoir computing, allows achieving detection range improvement without the associated increase in processing time and complexity.
2Measurement precision
If short pulse width is used, then resolution is improved, but signal-to-noise ratio deteriorates
Solution Approach 1:
The patent substitutes traditional signal integration methods with reservoir computing to overcome the low SNR inherent in short pulse width signals. The recurrent neural network structure of the reservoir computer effectively integrates signal energy over time while maintaining the short pulse width advantage for resolution, thereby improving SNR without sacrificing measurement precision.
Solution Approach 2:
The reservoir computing system acts as an intermediary between the short pulse width signal and the detection process. It receives the low SNR signal, processes it through its recurrent neural network structure, and outputs enhanced signals with improved SNR while preserving the fine range resolution achieved by short pulse width transmission.
3Reliability
If long coherent/incoherent pulse integration is applied, then signal-to-noise ratio is improved, but detection time increases
Solution Approach 1:
The patent replaces sequential integration methods with parallel reservoir computing processing. The recurrent neural network structure enables simultaneous processing of multiple signal components, achieving effective integration gain without the time penalty of sequential coherent or incoherent integration, thereby reducing detection time while maintaining improved SNR.
Solution Approach 2:
The reservoir computing system performs preliminary signal enhancement and feature extraction in real-time as signals arrive, rather than requiring complete signal accumulation before processing. This preliminary action within the neural network structure accelerates the detection process while maintaining the SNR improvement benefits of integration.
Data Source
AI summary
Described is a system for real-time radar range-doppler map learning. The system includes a physics enhanced dynamic reservoir core operable for receiving as an input signal a time delayed copy of a transmitted complex in-phase and quadrature (I/Q) radar waveform and mapping the input signal onto a neural network to determine complex-valued output weights of neural states of the neural network. An online learning layer is included that adapts the complex-valued output weights of the neural states to predict a most likely next value of the input signal. A neural combiner is used to combine a set of delayed neural state vectors with the complex-valued output weights of the online learning layer to compute a denoised and decluttered output signal.