AI Thrust Seminar | Feedback Communication via Deep Learning?

In modern communication systems with feedback, there are an increasing number of scenarios where the transmitter has much less power than the receiver, such as medical implant devices. This type of channel is referred to as a noise-asymmetric channel and it has a higher-quality feedback link than the forward link. However, conventional feedback schemes for cellular communications, such as hybrid ARQ, do not fully take advantage of the high-quality feedback link. The design of channel coding for feedback channels has been an open problem since Shannon's work in the field of communication theory.

In this talk, we will discuss the adaptation of research papers in the real world product. We will begin by discussing conventional feedback codes and deep learning-based channel codes and demonstrate that neither of these approaches is suitable for real-world products. To address this challenge, we will introduce the Compressed Error Hybrid ARQ, a solution that harmoniously combines feedback and hybrid ARQ for noise-asymmetric channels. Our proposed method significantly improves reliability, latency, and spectral efficiency compared to conventional hybrid ARQ in various practical scenarios where the transmitter is resource-constrained. Although inspired by deep learning, the solution is not necessarily based on deep learning.