Radar Perception Processor Using AI Neural Networks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional radar systems require extensive redesign and calibration for new applications, leading to increased manufacturing costs and downtime, and struggle with high-resolution imaging due to computational complexity and data handling limitations.

Innovation Solution

A radar system utilizing a hybrid architecture with AI engines, trainable signal processing blocks, and a flexible processing pipeline that generates radar perception data directly from digital samples, reducing the need for specific signal processing blocks and allowing easy retraining and scaling for different configurations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional radar systems use traditional signal processing blocks for high-resolution imaging, then imaging resolution is improved, but computational complexity and data handling requirements increase significantly

Engineering Contradiction:

Improveimaging resolutionVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical signal processing blocks with an AI-based neural network system. The neural network directly processes raw radar signals to generate perception data, eliminating the need for complex intermediate signal processing stages while maintaining or improving imaging resolution. This substitution reduces computational complexity by using learned patterns instead of exhaustive algorithmic processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The AI engine serves multiple functions simultaneously: it performs signal processing, feature extraction, target detection, and classification in a unified architecture. This multi-functional approach replaces multiple specialized signal processing blocks, reducing overall system complexity while maintaining high-resolution imaging capabilities across different radar configurations.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Reliability

If conventional radar systems are redesigned for new applications, then application-specific performance is improved, but manufacturing costs and downtime increase

Engineering Contradiction:

Improveapplication-specific performanceVSAvoidmanufacturing cost and downtime

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The radar system employs a dynamic AI model that can be retrained and adapted to different applications through software updates rather than hardware redesign. The neural network architecture remains fixed, but its parameters and training data can be dynamically adjusted to meet specific application requirements, enabling rapid deployment to new applications without manufacturing changes.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system achieves application-specific optimization by changing the training parameters and input data of the AI model rather than modifying physical components. By adjusting the training dataset, loss functions, and hyperparameters, the same hardware platform can be optimized for different radar configurations and applications, significantly reducing manufacturing costs and deployment time.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If conventional radar systems process high-resolution data through multiple processing stages, then data processing capability is improved, but latency increases due to buffering intermediate outputs

Engineering Contradiction:

Improvedata processing capabilityVSAvoidprocessing latency

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent extracts and eliminates the intermediate buffering stages from the traditional multi-stage processing pipeline. The AI engine takes raw radar signals directly as input and produces perception data as output in a single integrated operation, removing the need to store and manage intermediate processing results. This extraction of intermediate steps significantly reduces processing latency while maintaining high data processing capability.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent merges multiple sequential signal processing functions into a single unified AI processing stage. Instead of separately performing range processing, Doppler processing, and target detection in distinct stages with intermediate buffering, the neural network performs all these functions simultaneously in one operation, eliminating latency introduced by stage transitions and memory access.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11614514B2Apparatus, system and method of generating radar perception data

Publication Date: 2023.03.28 INTEL CORP

AI summary

For example, an apparatus may include a radar, the radar may include a reconfigurable radio configured, based on a plurality of reconfigurable radio parameters, to transmit a plurality of Transmit (Tx) radar signals via a plurality of Tx antennas, to receive via a plurality of Receive (Rx) antennas a plurality of Rx radar signals based on the plurality of Tx radar signals, and to provide digital radar samples based on the Rx radar signals; a radar perception processor configured to generate radar perception data based on the digital radar samples, the radar perception data representing semantic information of an environment of the radar; and a feedback controller to configure the plurality of reconfigurable radio parameters based on the radar perception data, and to feedback the reconfigurable radio parameters to the reconfigurable radio.