Compressed chaotic signal reconstruction based on deep learning

Abstract

Chaotic signals are often compressed on account of limited hardware resources in the Internet of Things (IoT). In this paper, we show that using 1 or 2 bits/symbol in the front-end quantizer suffices for subsequent high-quality recovery at the back-end. Specifically, we introduce a general chaotic signal recovery scheme, which is conceived to be deployed at the level of IoT edge to reconstruct received compressed signals. The scheme relies on the simplified convolutional denoising auto-encoder (SCDAE), which has advantages in terms of network architecture and parameter amount. Compared with U-Net, SCDAE has 50 times smaller parameters of storage, and 26 times smaller computational complexity.