Deep Learning for Replica Detection and Combining in Asynchronous Grant-Free mMTC

The Internet of Things has fostered the rapid growth of new services such as massive machine-type communications (mMTC), featuring transmission of short packets from a massive number of low duty-cycle devices. Grant-free (GF) access is gaining an increasing interest to support the mMTC traffic and tackle the related challenges. This letter proposes a deep learning-based solution for GF asynchronous mMTC systems where packet diversity is exploited to enhance the probability of successful uplink transmission. The proposed processing aims at detecting twin packets, i.e., replicas of the same packet, among the received signal samples, which allows combining them to achieve a higher successful decoding probability. Numerical results highlight remarkable gains over more conventional approaches.