24 April 2025
14:00 Master's Defense Room 85 of IC2
Topic
Intelligent mMTC Random Access Collision Detection and Prioritization in IoT Cellular Networks
Student
Giancarlo Maldonado Cárdenas
Advisor / Teacher
Nelson Luis Saldanha da Fonseca - Co-advisor: Carlos Alberto Astudillo Trujillo
Brief summary
The exponential growth of Internet of Things (IoT) devices in cellular networks, driven by massive machine type communications (mMTC), presents significant resource management challenges, especially during Random Access Procedure (RAP). High traffic volumes increase preamble collisions, leading to inefficiencies and repeated Random Access (RA) attempts, which deteriorate network performance. This dissertation presents a novel solution for RA collision detection for mMTC, using machine learning techniques to improve detection accuracy and optimize resource allocation in cellular networks. Three proprietary datasets were generated and analyzed using MATLAB LTE System Toolbox to simulate realistic mMTC environments, including various channel conditions, such as Extended Pedestrian A (EPA) and Extended Typical Urban (ETU) models. Several machine learning classifiers were employed in the study, including Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), Random Forest (RF), Naive Bayes, K-Nearest Neighbor (KNN), LightGBM, XG-Boost, and Neural Networks (NN). The classifiers were trained and evaluated for their ability to detect collisions. Among them, Neural Networks achieved the highest balanced accuracy, with 98%, demonstrating superior performance in both training and testing scenarios. In addition, advanced model optimization techniques, such as Dynamic Range Quantization and Full Integer Quantization, were applied to reduce inference time and memory usage. These techniques made the solution applicable in resource-constrained and real-time environments, achieving a reduction of up to 99% in inference time. In optimizing the AR procedure, the concept of Collision-Aware Random Access was introduced. The machine learning model was integrated into the LTE-Sim simulator and two prioritization algorithms were introduced, which use the inferred collision information to optimize resource allocation, prioritizing collision-free preambles and ensuring fast conflict resolution. This adaptive strategy reduces latency and improves network access efficiency, being especially effective in critical scenarios with high device density. The proposed approach improves on traditional correlation-based methods, offering a scalable and adaptive solution for dynamic and high-density IoT networks. The results indicate that machine learning, combined with intelligent resource allocation strategies, can significantly improve collision detection efficiency, reduce network latency, and optimize device connectivity. This thesis contributes to more robust AR procedures, promoting efficient resource management in next-generation cellular IoT networks.
Examination Board
Headlines:
| Nelson Luis Saldanha da Fonseca | IC / UNICAMP |
| Diana Cristina González González | CEATEC/PUC-Campinas |
| Judy Carolina Guevara Amaya | IC / UNICAMP |
Substitutes:
| Allan Mariano de Souza | IC / UNICAMP |
| Juliano Araujo Wickboldt | INF / UFRGS |