A Hybrid-Policy MADQN-Based Resource Allocation Solution for Power-Domain NOMA-Enhanced RAN Slicing

Although both service capacity (i.e., the number of users whose quality of service requirements are satisfied) and security (as impacted by resource isolation) are critical for radio access network (RAN) slicing, there often exists a trade-off between them. Enhancing security by reducing spectrum sharing among slices across different types inevitably compromises the service capacity. Moreover, traditional orthogonal resource utilization significantly limits service capacity, making it difficult to accommodate the rapid growth in the number of diverse users. To address this, we propose a non-orthogonal multiple access (NOMA)-enhanced RAN slicing scheme by introducing power-domain NOMA. We then focus on the subchannel set optimization problem in a multi-base-station uplink RAN slicing system. To cope with the network dynamics caused by user mobility and stochastic packet arrivals, we design a hybrid-policy multi-agent deep Q-network-based solution. Numerical results show that the proposed scheme achieves good performance, significantly enhancing service capacity while ensuring security of slices.