A Hybrid Collaborative Learning for Age of Information Minimization in Massive Access

Supporting massive access from Internet of Things (IoT) devices plays a pivotal role in the design of 6G networks. Nonetheless, concurrent massive access to the network deteriorates the quality of 6G communications. To overcome the challenge of massive access, our focus shifts to optimizing the age-critical frameless ALOHA (ACFA) random access protocol. The conventional ACFA suffers from high latency and unreliability when massive devices attempt to access the network. Consequently, we introduce an adaptive algorithm to address the transmission issues. This paper optimizes the random access channel (RACH) procedure by maximizing a long-term multi-objective function, which consists of the average age of information (AoI), normalized throughput, traffic load and the average number of successfully accessed machine-type communication devices. To achieve the optimal objective in ACFA, we apply deep reinforcement learning (DRL) algorithms. In our algorithms, agents take action in both distributed and centralized manners. In the distributed approach, each device learns the choice of the access probability and the slot, guided by feedback from the base station (BS). Simultaneously, in the centralized approach, the BS restricts a specific number of devices from accessing the network and dynamically adjusts the frame length based on the transmission results of devices. Our simulation results demonstrate that the proposed scheme surpasses benchmark schemes and exhibits significant potential to minimize AoI performance.