Binary Pruning for Intrapulse Modulation Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current classification techniques for intentional modulation of radio signals, such as radar signals, lack accuracy due to the numerous types and classes of signals of interest, especially when the signal types are unknown, leading to low signal-to-noise ratios and high false positive rates.
Innovation Solution
A machine learning-based approach using support vector machine algorithms and binary tree pruning to generate two-dimensional maps of extracted features from radio signals, allowing for the recognition of multiple modulation types with high accuracy by determining discriminating vectors that can classify fourteen common modulation types with reduced false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional classification techniques are used for modulation recognition, then the system is simpler to implement, but the classification accuracy is insufficient due to numerous signal types and unknown signal classes
Solution Approach 1:
The patent segments the modulation recognition task into multiple stages: signal preprocessing, feature extraction (generating 2D maps from signal characteristics), and classification using a binary tree of discriminating vectors. This segmentation allows each stage to be optimized independently, achieving high accuracy through systematic feature analysis while managing complexity through modular processing steps
Solution Approach 2:
The patent transforms 1D signal data into 2D feature maps by extracting multiple characteristics (amplitude, frequency, phase features) and visualizing them in two-dimensional space. This dimensional transformation enables more effective pattern recognition and classification by revealing structures and relationships that are not apparent in the original signal domain
2Adaptability or versatility
If the system attempts to recognize all possible modulation types, then the adaptability increases, but the false positive rate increases due to low signal-to-noise ratios
Solution Approach 1:
The patent performs preliminary feature extraction and 2D map generation before classification, creating a robust feature representation that enhances signal-to-noise ratio. The binary tree structure is pre-trained with discriminating vectors that encode knowledge about multiple modulation types, allowing the system to reliably distinguish between them even under low SNR conditions
Solution Approach 2:
The system incorporates iterative optimization where classification performance is evaluated and used to refine the discriminating vectors and feature extraction parameters. This feedback mechanism continuously improves reliability by learning from classification outcomes and adjusting the decision boundaries to reduce false positives while maintaining high adaptability
3Measurement precision
If more features are extracted from the signals, then the classification accuracy improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent extracts only the most discriminative features for classification by using 2D maps that highlight key signal characteristics. The feature extraction process identifies and isolates the most informative aspects of the signal (amplitude modulation patterns, frequency modulation patterns, phase relationships) while discarding redundant information, achieving high accuracy with computationally efficient feature sets
Solution Approach 2:
The system extracts a focused set of critical features rather than analyzing all possible signal characteristics. By concentrating computational resources on the most discriminative features (represented in 2D maps), the system achieves sufficient classification accuracy without the excessive processing time that would result from comprehensive feature analysis
Data Source
AI summary
Radio signals including modulated radar signals of an unknown modulation type selected from among a predetermined group of modulation types are received, and a plurality of features are extracted for the received radio signals. A plurality of two dimensional (2D) maps are generated for pairs of the extracted features from the received radio signals. The 2D maps of extracted feature pairs for the received radio signals are processed using a binary tree of discriminating vectors, each of the discriminating vectors corresponding to recognition of at least one of the predetermined modulation types based on 2D feature maps and each of the discriminating vectors determined by processing 2D maps for pairs of features extracted from training samples using a support vector machine learning algorithm. The binary tree is derived by pruning permutations of sequences for applying the discriminating vectors according to iterative testing of modulation type recognition accuracy.