Subcarrier Allocation for Multiuser OFDM Systems by Using Deep Neural Networks

Previously, we proposed schemes in [1] and [2] for the classical subcarrier, bit, and power allocation problem [3] to minimize the total transmit power for multiuser orthogonal frequency division multiplexing systems in downlink transmission. In this paper, we propose a deep neural network (DNN) structure to speed up solving this complex problem. We propose a deep learning frame structure in which each group of allocation is termed as a batch; after some numbers of iterations and epochs, the loss will tend to converge to a constant value. The simulation results reveal that the proposed DNN-based schemes offer competitive performance and reduce computing time tremendously compared with those of the existing approaches.