Reinforcement learning for PHY layer communications