DUSE: A Data Expansion Framework for Low-resource Automatic Modulation Recognition based on Active Learning

Yao Lu

doi:10.1016/j.dsp.2026.106155

DUSE: A Data Expansion Framework for Low-resource Automatic Modulation Recognition based on Active Learning

2026-04-15

0

PRE

AI

DOI：10.1016/j.dsp.2026.106155

Original

Original request for help

Save

Abstract

En 中文

• We propose and formally define the concept of data expansion in the AMR field. • We introduce DUSE, a novel uncertainty-driven framework based on active learning. • DUSE outperforms 8 coreset strategies across three benchmark AMR datasets. • Our method shows superior performance in both balanced and imbalanced settings. • The expanded datasets demonstrate strong cross-architecture generalization.

Keywords:

data expansion

active learning

AMR

uncertainty-driven

coreset strategies

DUSE: A Data Expansion Framework for Low-resource Automatic Modulation Recognition based on Active Learning

Abstract

Journal

Organization