Deep reinforcement learning-based multi-channel spectrum sharing technology for next generation multi-operator cellular networks

The mobile network operators (MNOs) need to efficiently utilize spectrum resources to meet increasing user demands for massive and ubiquitous connectivity. The licensed spectrum resources are scarce, costly and difficult to acquire. Consequently, the available bandwidth becomes a challenge. In this paper, a deep reinforcement learning (DRL)-based method has been utilized to share the spectrum in a multi-channels multi-operator environment. To intelligently and dynamically assign suitable channels, the proposed DRL model implemented at each MNO takes the load on the gNodeBs (gNBs), such as, the number of packets in the gNB queue and resource requirements of user equipments, such as, achievable data rate of users, into account to estimate the suitable channel selections. The scheduler then utilizes this channel information for efficient channel allocations. The performance of the proposed DRL-based spectrum sharing scheme has been compared with the conventional scheduling-based spectrum allocation scheme using extensive simulations. Results indicate that the dynamicity in network environment and traffic demands can be reasonably handled by the proposed DRL-based multi-channel spectrum sharing scheme, since it adapts feasibly to the varying number of channels, number of UEs, and network traffic conditions, compared to those of the conventional scheme. Furthermore, the proposed scheme shows superior performance gains in terms of throughput, resource utilization, delay, transmission time, and packet drop rates, compared to those of the conventional scheme.