A novel random access (RA) scheme for mixed URLLC-mMTC traffic scenario is proposed using realistic statistical models, with the use mode presenting long-term traffic regularity. The traffic is predicted by a long short-term memory neural network, which enables a traffic-aware resource slicing aided by contention access class barring-based procedure. The method combines a grant-free (GF) RA scheme with an intermediate step to congestion alleviation. The protocol trade-off is a small overhead while enabling a higher number of decoded received packets thanks to the intermediate step. Numerical results evaluate the system performance for each procedure and combined solution. A comparison with GF benchmark reveals substantial improvement in system performance.
@article{arxiv.2303.01159,
title = {LSTM-ACB-Based RA for IoT Mixed Traffic},
author = {Herman L. dos Santos and José Carlos Marinello and Cristiano Magalhaes Panazio and Taufik Abrão},
journal= {arXiv preprint arXiv:2303.01159},
year = {2023}
}