RF Detection Localization Using Bayesian Probability and Modified YOLO

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Real-time signal detection, localization, and classification in noisy, cluttered RF environments are challenging due to degradation in low SNR conditions for existing RF machine learning systems using the YOLO algorithm applied to wideband RF spectrograms.

Innovation Solution

A system and method employing Bayesian probability analysis to isolate noise from the magnitude spectrum, utilizing a modified YOLO algorithm for RF detection and localization, and Bayesian Convolutional Neural Networks for signal classification, which enhances accuracy by providing probability information and uncertainty estimation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a modified YOLO algorithm is applied to wideband RF spectrograms for signal detection and localization, then real-time processing capability is improved, but measurement precision deteriorates in low SNR conditions

Engineering Contradiction:

Improvereal-time processing capabilityVSAvoidsignal detection accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent introduces probability analysis as an intermediary layer between the YOLO algorithm and the final detection output. By computing probability maps that represent the likelihood of signal presence at each spectrogram location, the system mediates between fast algorithmic processing and precise detection decisions, allowing real-time operation while improving accuracy through probabilistic reasoning about signal characteristics

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms the detection problem by changing from deterministic threshold-based detection to probabilistic detection. By computing probability values that reflect the confidence of signal presence based on spectral characteristics, the system adapts the detection parameter from a fixed threshold to a dynamic probability metric that performs better in low SNR conditions while maintaining real-time processing

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If probability analysis is employed to isolate noise from the magnitude spectrum, then measurement precision is improved in low SNR conditions, but device complexity increases

Engineering Contradiction:

Improvesignal-to-noise ratio estimation accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex probability analysis task into distinct computational stages: first computing the magnitude spectrum from IQ samples, then calculating probability values based on spectral characteristics, and finally using these probabilities to guide detection decisions. This segmentation allows the system to achieve high measurement precision through detailed probability analysis while managing computational complexity by breaking the problem into manageable, optimized stages

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11562183B1System and method for RF detection, localization, and classification

Publication Date: 2023.01.24 ROCKWELL COLLINS INC

AI summary

A system and method for detecting, localizing, and classifying RF signals via probability analysis in the decision space include receiving a wideband IQ sample stream and performing a probability analysis to isolate noise from the magnitude spectrum. Derived probability information is used for RF detection and localization. The probability analysis is a Bayesian probability analysis and the detection and localization algorithm is a modified “you only look once” (YOLO) algorithm.