Massive MIMO Power Allocation Using Deep Reinforcement Learning

This paper studies the power allocation strategy in massive multiple-input multiple-output (MIMO) wireless networks to maximize the sum spectral efficiency (SE) in the downlink transmission. Different from the traditional methods capitalizing on nonlinear optimization algorithms which require high computational complexity, we invoke a deep reinforcement learning (DRL) framework to obtain the optimal power allocation. Specifically, from the power allocation design problem for massive MIMO networks, we develop an algorithm leveraging a twin-delayed deep deterministic policy gradient (TD3) framework to seek the optimal policy that maximizes the expected cumulative long-term reward in terms of the SE. The numerical results are provided to demonstrate the achievable SE obtained from the DRL framework in comparison with those from optimization algorithms.