Learning-Based Rich Feedback HARQ for Energy-Efficient Uplink Short Packet Transmission

TL;DR

This work tackles energy efficiency in uplink short-packet transmission under reliability and latency constraints by relocating computational burden from a power-constrained transmitter to a capable receiver. It introduces Reinforcement-based Adaptive Feedback (RAF), a rich feedback HARQ scheme where the receiver learns a puncturing pattern for the next round based on the decoder's a posteriori entropy, using Deep Q-Learning within a finite-horizon MDP. The method leverages the state $\mathbf{h}_t$ (entropy vector) and actions that specify how many symbols to retransmit, optimizing the discounted objective $\mathbb{E}[E_b(T)\gamma^T]$ to balance energy, latency, and reliability. Across simulations with non-binary GF($q$) codes and various modulation schemes, RAF achieves Pareto-dominant performance over conventional HARQ baselines, including under fading channels, demonstrating robust adaptability and practical impact for energy-limited uplink devices.

Abstract

The trade-off between reliability, latency, and energy efficiency is a central problem in communication systems. Advanced hybrid automated repeat request (HARQ) techniques reduce retransmissions required for reliable communication but incur high computational costs. Strict energy constraints apply mainly to devices, while the access point receiving their packets is usually connected to the electrical grid. Therefore, moving the computational complexity from the transmitter to the receiver may provide a way to improve this trade-off. We propose the reinforcement-based adaptive feedback (RAF) scheme, a departure from traditional single-bit feedback HARQ, introducing adaptive rich feedback where the receiver requests the coded retransmission of specific symbols. Simulation results show that RAF achieves a better trade-off between energy efficiency, reliability, and latency, compared to existing HARQ solutions. Our RAF scheme can easily adapt to different modulation schemes and can also generalize to different channel statistics.

Figures (8)

• Figure 1: Diagram of the proposed signaling protocol between the energy-constrained transmitter and the receiver.
• Figure 2: Communication protocol for the $t$-th round.
• Figure 3: Schematic of feedback included in common retransmission schemes.
• Figure 4: Block diagram of the proposed rich feedback communication protocol, .
• Figure 5: Performance of the five policies , , , , and in terms of the average number of successfully transmitted bits per mJ and average latency when $\alpha=0.5$ and $L_p=1$ as a function of their respective parameters: $L_{\text{static}}$ for and , the entropy threshold $H_{\text{th}}$ for and , and the discount factor $\gamma_{_{\text{DQN}}}$ for .