Modulation and signal class labelling using active learning and classification using machine learning

TL;DR

This paper introduces an active learning framework for real-time wireless modulation and signal classification, achieving high accuracy with minimal labeled data using machine learning algorithms like KNN and SVM.

Contribution

It presents a novel application of active learning to wireless signal classification, reducing the amount of labeled data needed for accurate real-time classification.

Findings

Active learning achieves 86% accuracy at 18 dB SNR.

KNN model reaches 99.8% accuracy at 18 dB SNR.

Active learning reduces labeling effort in wireless signal classification.

Abstract

Supervised learning in machine learning (ML) requires labelled data set. Further real-time data classification requires an easily available methodology for labelling. Wireless modulation and signal classification find their application in plenty of areas such as military, commercial and electronic reconaissance and cognitive radio. This paper mainly aims to solve the problem of real-time wireless modulation and signal class labelling with an active learning framework. Further modulation and signal classification is performed with machine learning algorithms such as KNN, SVM, Naive bayes. Active learning helps in labelling the data points belonging to different classes with the least amount of data samples trained. An accuracy of 86 percent is obtained by the active learning algorithm for the signal with SNR 18 dB. Further, KNN based model for modulation and signal classification performs well over range of SNR, and an accuracy of 99.8 percent is obtained for 18 dB signal. The novelty of this work exists in applying active learning for wireless modulation and signal class labelling. Both modulation and signal classes are labelled at a given time with help of couplet formation from the data samples.