We investigate a power control problem for overlay device-to-device (D2D) communication networks relying on a deep deterministic policy gradient (DDPG), which is a model-free off-policy algorithm for learning continuous actions such as transmitting power levels. We propose a DDPG-based self-regulating power control scheme whereby each D2D transmitter can autonomously determine its transmission power level with only local channel gains that can be measured from the sounding symbols transmitted by D2D receivers. The performance of the proposed scheme is analyzed in terms of average sum-rate and energy efficiency and compared to several conventional schemes. Our numerical results show that the proposed scheme increases the average sum-rate compared to the conventional schemes, even with severe interference caused by increasing the number of D2D pairs or high transmission power, and the proposed scheme has the highest energy efficiency.
Keywords: deep deterministic policy gradient (DDPG); deep reinforcement learning (DRL); device to device (D2D); power control.