We study short-packet communications, subject to latency and reliability constraints, under the premises of limited frequency diversity and no time diversity. The question addressed is whether, and when, hybrid automatic repeat request (HARQ) outperforms fixed-blocklength schemes with no feedback (FBL-NF) in such a setting. We derive an achievability bound for HARQ, under the assumption of a limited number of transmissions. The bound relies on pilot-assisted transmission to estimate the fading channel and scaled nearest-neighbor decoding at the receiver. We compare our achievability bound for HARQ to stateof-the-art achievability bounds for FBL-NF communications and show that for a given latency, reliability, number of information bits, and number of diversity branches, HARQ may significantly outperform FBL-NF. For example, for an average latency of 1 ms, a target error probability of 10^-3, 30 information bits, and 3 diversity branches, the gain in energy per bit is about 4 dB.
@article{arxiv.1809.06560,
title = {Low-Latency Short-Packet Transmissions: Fixed Length or HARQ?},
author = {Johan Östman and Rahul Devassy and Guido C. Ferrante and Giuseppe Durisi},
journal= {arXiv preprint arXiv:1809.06560},
year = {2018}
}