Radar Signal Processing Using Deep Neural Network

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for detecting objects in monitored areas using radar signals face challenges such as inadequate performance due to insufficient filtering and mixing signals from multiple transmitters, which complicates training and increases computational effort and time.

Innovation Solution

A computer-implemented method that processes raw radar signal data without filtering, dividing it into sets for each receiver-transmitter pair and inputting these into a deep neural network with separate receiver and transmitter layers for feature extraction, enabling improved accuracy and efficiency in object detection and classification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If filtering is applied to radar data before inputting to neural network, then noise is reduced, but detection accuracy deteriorates due to loss of useful information

Engineering Contradiction:
ImprovenoiseVSAvoiddetection accuracy
Core Design Contradiction:
Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The patent extracts and removes only the harmful mixing signals from multi-transmitters through signal separation techniques, while preserving the useful reflected radar wave information. This selective extraction approach eliminates noise without the information loss that occurs with conventional filtering methods that remove entire frequency bands.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the approach from frequency-domain filtering to a signal separation methodology that operates on the raw radar data parameters. By using techniques like orthogonal matching pursuit or compressed sensing to separate signals from different transmitters, the system maintains the original signal characteristics while removing interference.

Inventive Principle:
Principle #35Parameter changes

2Ease of operation

If distance map is created using multiple Fourier transforms, then object detection becomes easier and performance measurement is simplified, but computational effort and time increase significantly

Engineering Contradiction:
Improvedetection convenienceVSAvoidprocessing time
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The patent performs signal separation and feature extraction directly on the raw radar data before any mapping operations. By preprocessing the data to separate transmitter signals and extract relevant features in advance, the system reduces the computational burden of subsequent processing steps and eliminates the need for multiple Fourier transforms to create distance maps.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent replaces the traditional mechanical process of creating distance maps through multiple Fourier transforms with a direct neural network-based processing approach. The neural network operates on the separated signal data to directly produce detection results, substituting the multi-step transform process with a more efficient computational model.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Ease of manufacture

If signals from multiple transmitters are mixed together, then data collection is simplified, but training difficulty increases and performance deteriorates

Engineering Contradiction:
Improvedata collection simplicityVSAvoidtraining complexity
Core Design Contradiction:
Ease of manufactureVSDevice complexity

Solution Approach 1:

The patent segments the mixed signal data into separate transmitter components using signal separation techniques. By dividing the combined signals from multiple transmitters into individual transmitter contributions, the system maintains the simplicity of collecting mixed signals while enabling effective training through separated signal processing in the neural network.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary signal separation process that acts as a mediator between the mixed raw data and the neural network training. This intermediary step separates the transmitter signals while preserving the original data collection simplicity, making the training process more effective without requiring complex separate signal collection setups.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If convolutional layers are added to neural network to compensate for insufficient filtering, then detection performance may improve, but device complexity and computational resources increase

Engineering Contradiction:
Improvedetection accuracyVSAvoidneural network complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs signal separation and noise removal as a preliminary action before data enters the neural network. By preprocessing the radar signals to separate transmitter components and remove interference beforehand, the neural network receives cleaner, more structured data that requires fewer complex convolutional layers to achieve the same detection accuracy.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4418000A1Radar data processing by dnn
Publication Date: 2024.08.21 IMRA EURO
  • EP4418000A1 patent drawingFigure 1~2
  • EP4418000A1 patent drawingFigure 3~4
  • EP4418000A1 patent drawingFigure 5~6

AI summary

A computer-implemented method for detecting an object (1) in a monitored area (200) of a monitoring device (100) having a first predetermined number of receivers (21) and a second predetermined number of transmitters (22), the method comprising the steps of: - dividing raw signal data into divided signal data for each receiver (21) linked to each transmitter (22), so as to have a divided set of a third predetermined number of divided signal data, the third predetermined number being equal to the first predetermined number multiplied by the second predetermined number, - inputting the divided set in a deep neural network (DNN) having a receiver layer having a number of feature extraction branches equal to the first predetermined number, and a transmitter layer having a number of feature extraction branches equal to the second predetermined number, - outputting a detection of the at least one object (1).