Reinforcement learning for energy efficient wireless transmission: Green communications for a green hand transmitter

Reinforcement learning is applied to the green communication of a green hand wireless transmitter which has no a priori information about the environment. Via reinforcement learning, radio emission power, transmission rate and transmitter hardware mode (transmit, idle and sleep) are controlled to optimize the energy consumption. To alleviate the problem of prohibitively many states (the curse of dimensions), an approach of state merge and split is proposed to adapt the state space to the number of observations and achieve the tradeoff between learning speed and final performance. On using a typical configuration of wireless transmitter, numerical results show that the learning procedure converges and significantly improves the energy efficiency. They also demonstrate the effectiveness of the proposed technique of state merge and split.