Multiuser Data Dissemination in OFDMA System Based on Deep Q-Network

In this paper, a multiuser data dissemination problem is analyzed in an orthogonal frequency division multiple access(OFDMA) downlink system. By dynamically allocating subchannels and power to the mobile users, the system aims to minimize the time consumption in order to successfully deliver data to multiple mobile users under the restriction of total energy consumption. Both the objective and the constraint of the optimization are related to the real-time resource allocation strategies. Moreover, neither the statistics nor the full channel state information is known to the base station. In order to solve the global optimization problem with partial channel information, a Deep Q-Network algorithm is adopted. The numerical results show that compared with the other algorithms, Deep Q-Network can learn the optimal resource allocation strategies and achieve very good system performance.