Toward the Realization of Encoder and Decoder Using Deep Neural Networks

Abstract

Deep learning continues to offer groundbreaking technologies in communications systems just as it has in a variety of other fields. Despite the notable advancements of technologies based on DNN-based technologies in recent years, their high computational complexity has been a major obstacle to the application of DNN in practical communications systems where real-time operation is required. In this sense, challenges regarding its practical implementation must be addressed before the proliferation of DNN-based intelligent communications becomes a reality. To the best of the authors' knowledge, the present article is the first to present an efficient learning architecture and related design strategies including link-level verification through the implementation of digital circuits using hardware description language (HDL) to mitigate this challenge and to deduce the feasibility and potential of DNNs in communications systems. In particular, a DNN is applied to an encoder and a decoder to enable flexible adaptation with respect to their system environments, without the need for any domain-specific information. Extensive investigations and interdisciplinary design considerations, including a DNN-based autoencoder structure, learning framework, and implementation of a low-complexity digital circuit for real-time operation, are taken into account, all of which support the use of DNN-based communications in practice.