Sparsity-Aware Multi-Agent RL-Random Access Protocol using Grant Free for Massive MIMO Networks in Cellular IoT

Cellular Internet of Things (IoT) devices greatly benefit from the incorporation of large Multiple Input Multiple Output (MIMO) networks in terms of connectivity and information processing. Effective and reliable random access techniques face challenges due to the increased density of IoT devices and the irregularity of their transmissions. Conventional grant-based access mechanisms are ineffective in high-density IoT environments due to higher signaling costs and delays. While they require less communication, grant-free random access techniques struggle with managing interference and collisions, especially in large MIMO settings. This research aims to develop a sparsity-aware, grant-free randomization technique using Multi-Agent Reinforcement Learning (MARL) to improve the efficiency of large MIMO systems in cellular IoT applications. The main objectives are to enhance access efficiency, reduce collision rates, and effectively manage interference. In the proposed technique, each IoT device acts as an agent within a MARL framework. Computational findings show that the sparsity-aware, MARL-based, grant-free random access protocols significantly outperform conventional approaches in terms of connection success probability, delay, and overall network performance. In dense IoT environments, the proposed protocol effectively handles conflicts and interference, resulting in more reliable and efficient communications. The implementation of a sparsity-aware, grant-free randomization method based on MARL in cellular IoT improves the efficiency of large MIMO networks. This approach addresses critical issues of interference control and connectivity efficiency, paving the way for further advancements in cellular IoT communications.