All-in-One: VQ-VAE for End-to-End Joint Source-Channel Coding

Vector Quantized Variational Autoencoder (VQ-VAE) has been regarded as a promising representation of diverse and complex data distributions in deep learning ecosystem. However, its use in a systematic way leveraging existing wireless communications has not been well addressed. In this paper, we explore the VQ-VAE characteristics in a point-to-point wireless communication and modify its training process to design a joint source-channel coding that is robust against noisy wireless channels. With all due respect to the source-channel separation theorem, various factors prevent error-free transmissions of conventional coding schemes in reality. Likewise, the proposed model is not error-free, but it compromises the reliability and complexity of the system. Thus, the proposed model makes the physical/link layer lighter while preserving reliability. It is considered an alternative for further data compression compared to the conventional separated source-channel coding schemes. Our system has been evaluated with extensive simulations, providing insightful observations and findings.

This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. Upon acceptance in IEEE: Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to pubs-permissions@ieee.org. (Copyright (c) 2015 IEEE.)