Design and Analysis of Deep Learning Models for Massive Machine-Type Communications

Significant research efforts have recently been concentrated on developing new and efficient techniques for the future 6G standard. Massive Machine-Type Communications (mMTC) refers to the communication model that supports large numbers of devices transmitting small amounts of data over wireless networks. This principle is fundamental as it underlies the capabilities of the Internet of Things (IoT), which enables a range of applications in areas such as smart cities, industrial automation and healthcare. The growing number of connected devices poses significant challenges, particularly regarding their ability to efficiently access the radio medium. To address this challenge, grant-free random access schemes have emerged as a viable solution. These schemes allow devices to transmit data without waiting for permission from the network, thereby reducing delays and improving overall throughput. The integration of deep learning techniques is a promising avenue for improving the performance of grant-free access schemes. This thesis proposes a Deep Learning (DL)-enhanced receiver capable of detecting twin packets, i.e., replicas of the same packet, and combining them to improve decoding probability during the BS processing. The contents of this thesis are organized as follows: 1. In Chapter 1, the asynchronous Massive Multiple Access (MMA) scenario is described, with focus on the asynchronous coded random access protocol and its characteristics. 2. Chapter 2 provides a brief overview of key deep learning techniques, followed by the proposed Convolutional Neural Network (CNN) designed to enhance base station processing. 3. Chapter 3 discusses the processing at the base station, detailing all stages, including the integration of CNN into the decoding process. 4. Chapter 4 evaluates the performance of the proposed DL-based system, with simulations comparing the results to state-of-the-art approaches.