Joint Pilot Spacing and Power Optimization Scheme for Nonstationary Wireless Channel: A Deep Reinforcement Learning Approach

Pilot-assisted channel estimation techniques are essential for wireless communication systems. Most studies focus on the estimation algorithms and interpolation techniques. However, the design of pilot pattern is often neglected. In this letter, we propose a joint pilot spacing and power optimization scheme based on deep reinforcement learning (DRL) to address the mismatch problem of pilot configuration for nonstationary wireless channel. First, we model the adaptive pilot design decision-making process as a Markov Decision Process (MDP) to reduce pilot overhead and power loss. Then a deep Q-network (DQN) based learning algorithm is proposed to optimize the spacing and power of pilots so as to maximize estimation performance while reducing system cost. Simulation results show that the performance of the proposed approach is better than conventional pilot configuration algorithms. Moreover, we analyze the key factors that affect the performance of the proposed scheme.