Autoencoder-based Data Compression Model Experiment for Semantic Communication
Jinyoung Oh, Yunkyung Choi, Chanyoung Oh, Woongsoo Na · 2024
In recent years, the convergence of neural network architectures and semantic communication has led to innovative strides in representation learning. This paper explores the application of autoencoders, a subset of neural networks designed for unsupervised learning, in encoding and decoding data to capture semantic nuances. Additionally, we discuss future research directions, proposing a semantic communication model learned from time series data and presenting experimental results. Our experiments involve encoding and decoding time series data using autoencoders, evaluating the feasibility of integrating autoencoder technology into semantic communication. Following the experiment, our proposed model exhibited a loss rate of approximately 0.15-0.2% for the time series data. This outcome represents a notable result, especially when compared to the compression rate of transmission data. It hints at the potential for a future autoencoder-based semantic communication model.