부산대학교 도서관

The purpose of this research is to categorise signal modulation using few-shot learning. The first approach is an intelligent, modular, few-shot learning framework. This system's two components are IQF and 1D-SFP, and it also has a task-driven network design. The 1D-SFP module is responsible for obtaining high-level characteristics from the raw signal, while the IQF module is in charge of retrieving low-level information. The number of 1D-SFP modules required for processing a large amount of data is calculated by studying the correlation in the hidden layer. We evaluate the suggested method using a publicly available dataset and show that it beats standard strategies in terms of accuracy. It can also learn from a small number of instances and is adaptable enough to handle innovative modulation schemes. This enables it to be used in wireless networks for data transmission and reception. The second approach uses a graph convolutional network (GCN) to calculate how many 1D-SFP modules should be integrated into a neural network architecture for a particular job. On benchmark datasets, the presented frameworks outperformed a range of state-of-the-art, conventional methodologies. Both approaches perform preprocessing, representation learning, and classification using their own distinct algorithmic frameworks. The presented strategies have the potential to increase signal modulation classification precision and efficiency in cognitive radios and other kinds of wireless communication. These gains are possible because of the potential of these techniques.