Predictive Resource Allocation with Interference Coordination by Deep Learning
Zhaoqi Xu, Jia Guo, Chenyang Yang · 2019
Predictive resource allocation with interference coordination can provide high throughput for densely deployed cellular network by harnessing future average channel gains of mobile users in a minute-long time window. However, except for the necessity of learning the future information, finding the optimal policy incurs prohibitive computational complexity. In this paper, we resort to deep neural networks (DNNs)to learn the optimal policy for heterogeneous networks. To show the potential of the DNN-based solution, we first assume that future information in the prediction window is known perfectly. For the considered problem of coordinating interference by muting base stations and user association, we find that there exists dataset bias issue, where the optimal policy contains much more “0” than “1”. To deal with this issue, we design a DNN to learn the intermediate result of the original optimization problem. Then, we design a user association algorithm. We proceed to extend the designed DNN, such that the DNN can learn from the historical data in an end-to-end manner. Simulation results show that the proposed DNN-based solution performs closely to the numerically-obtained optimal solution with much lower computational complexity, even when the input of the DNN is the historical data.