Modulation Signal Recognition Based on Information Entropy and Ensemble Learning.

Area of Science:

• Digital Signal Processing
• Machine Learning
• Information Theory

Background:

• Signal recognition is crucial in various communication systems.
• Traditional methods often struggle with complex modulated signals and noise.
• Entropy-based features offer a novel approach for signal characterization.

Purpose of the Study:

• To propose novel signal recognition algorithms using information entropy and ensemble learning.
• To evaluate the effectiveness of different feature selection methods on entropy features.
• To compare the performance of various classifiers for modulated signal recognition.

Main Methods:

• Extraction of 16 entropy features (Rényi entropy, energy entropy) from 9 modulated signal types.
• Application of Sequence Forward Selection (SFS), Sequence Forward Floating Selection (SFFS), and RELIEF-F for feature selection.
• Classification using k-Nearest Neighbor (KNN), Support Vector Machine (SVM), Adaboost, Gradient Boosting Decision Tree (GBDT), and eXtreme Gradient Boosting (XGBoost).

Main Results:

• SFS and SFFS feature subsets yielded significant improvements: 48% with KNN and 34% with SVM.
• For Adaboost, GBDT, and XGBoost, the original feature set performed best.
• XGBoost achieved the highest overall recognition rate (97.74%), with 82% accuracy at -10 dB SNR.

Conclusions:

• Feature selection, particularly SFS and SFFS, enhances performance for certain classifiers like KNN and SVM.
• Ensemble methods, especially XGBoost, demonstrate robust performance across various conditions.
• The proposed entropy-based approach offers a promising direction for advanced signal recognition.