Load and Location Aware Resource Allocation in GF-NOMA IoT Networks

In this paper, we propose a load and location-aware resource allocation algorithm for the internet of things networks with grant-free non-orthogonal multiple access (GF-NOMA). The networks with GF-NOMA have randomly activating users, and different locations have different loads and requirements. If a large number of users simultaneously try to access may overwhelm the random access channel. As a result, resource allocation in such a scenario is challenging and leads to transmission failure and access latency. Therefore, resource allocation relying on locations while considering load in each region is a feasible solution. To this end, we divided the cell area into different regions and set up a resource pool for each tier. In response to the estimated load, the base station determines the number of sub-channels for each region, and the G F users find the optimal power level for transmission using multi-agent deep reinforcement learning (MA-DRL). The specific design of our algorithm ensures independent complexity for each region, overcoming the high computational costs associated with a huge state space. The numerical results demonstrate that the proposed scheme is more efficient at achieving convergence than the conventional GF-NOMA algorithm and provides more system throughput. Moreover, the proposed framework significantly reduces the action and state spaces, which is a common problem in MA-DRL algorithms.