Radar CA-CFAR Processing Using Separable 1D Convolutions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional 2D CFAR algorithms require significant processing operations and 2D convolution capabilities, which are not feasible on systems with limited processing power, such as internet of things and automotive applications.

Innovation Solution

A method to perform 2D CFAR radar processing using one-dimensional convolutions by spatially separating 2D kernels into pairs of one-dimensional vectors and applying superposition principles, reducing the number of operations required.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional 2D CFAR algorithms are used, then accurate constant false alarm rate processing is achieved, but the processing operations and computational complexity increase significantly

Engineering Contradiction:
Improveconstant false alarm rate processing accuracyVSAvoidprocessing operations
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the 2D CFAR processing into multiple 1D convolution operations. Instead of performing a single complex 2D convolution, the algorithm divides the 2D kernel into multiple 1D vectors and performs sequential 1D convolutions along different dimensions. This segmentation reduces the computational complexity from O(M×N) for 2D convolution to O(M+N) for 1D convolutions, while maintaining the same filtering effect and constant false alarm rate accuracy.

Inventive Principle:
Principle #1Segmentation

2Reliability

If conventional 2D CFAR algorithms are used, then accurate constant false alarm rate processing is achieved, but the processing power requirements become prohibitive for IoT and automotive applications

Engineering Contradiction:
Improveconstant false alarm rate processing accuracyVSAvoidprocessing power
Core Design Contradiction:
ReliabilityVSPower

Solution Approach 1:

The patent segments the 2D CFAR processing into multiple 1D convolution operations. Instead of performing a single complex 2D convolution, the algorithm divides the 2D kernel into multiple 1D vectors and performs sequential 1D convolutions along different dimensions. This segmentation reduces the computational complexity from O(M×N) for 2D convolution to O(M+N) for 1D convolutions, while maintaining the same filtering effect and constant false alarm rate accuracy.

Inventive Principle:
Principle #1Segmentation

3Reliability

If 2D convolution capabilities are implemented, then complete 2D CFAR processing is achieved, but the device complexity and resource requirements increase

Engineering Contradiction:
Improve2D CFAR processing capabilityVSAvoid2D convolution capabilities
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent transforms the 2D convolution operation into a sequence of 1D convolution operations by changing the dimensionality of the processing approach. Instead of applying a 2D kernel directly to the radar data matrix, the algorithm applies 1D kernels sequentially along rows and columns. This dimensionality change allows the system to achieve equivalent 2D filtering results using only 1D convolution capabilities, reducing the complexity requirements for the processing device.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4711806A1A radar processor
Publication Date: 2026.03.18 NXP BV
  • EP4711806A1 patent drawingFigure 1
  • EP4711806A1 patent drawingFigure 2
  • EP4711806A1 patent drawingFigure 3A~3B

AI summary

A cell-averaging constant false alarm rate, CA-CFAR, radar processor configured to: receive a two-dimensional radar data array (202); for each of a plurality of pairs of one-dimensional vectors (204-1, 204-2): perform a convolution between the two-dimensional radar data array (202) and a first one-dimensional vector (204-1) of the pair of one-dimensional vectors to obtain an intermediate convolution result (210); and perform a convolution between the intermediate convolution result (210) and a second one-dimensional vector (204-2) of the pair of one-dimensional vectors to obtain an intermediate CA-CFAR result (208); and combine the intermediate CA-CFAR results to provide a CA-CFAR output.