Few-shot modulation recognition using Recurrence Plot Algorithm

Woojin Yun, Jiyeon Park, Hyeongyun Kim, Haewoon Nam

Abstract: Deep learning techniques have shown high performance for Automatic Modulation Classification (AMC) tasks. However, the tasks need a burden of collecting large-scale annotated data, where the trained network model should be re-trained if new classes are given. In this paper, Few-shot learning (FSL) based AMC is introduced to handle this problem. Also imaging algorithm using recurrence plot (RP) is considered to make the input data more suitable for few shot learning. The results demonstrate that the proposed approach is able to classify images of new classes with high accuracy without further updating the network.