User-Centric Association in Ultra-Dense mmWave Networks via Deep Reinforcement Learning

For ultra-dense networks, user-centric architecture is regarded as a promising candidate to offer mobile users better quality of service. One of the main challenges of user-centric architecture is exploring efficient scheme for user association in the ultra-dense network. In this letter, we study dynamic user-centric association (UCA) problem for ultra-dense millimeter wave (mmWave) networks to provide reliable connectivity and high achievable data rate. We consider time-varying network environments and propose a deep Q-network based UCA scheme to find the optimal association policy based on the historical experience. Simulation results are presented to verify the performance gain of our proposed scheme.