Power Control in a modified Strict Frequency Reuse Algorithm Utilizing Q-Learning

In this paper, Strict Frequency Reuse (FR) algorithm is studied for a single tier indoor Spatial Point Poisson cellular network under a 3GPP multi-slope path loss model. While the regular Strict FR algorithm assumes that the Cell-Edge Sub-band Group (CEG) is only reused in a group of the adjacent cells, this modified algorithm of this work allows all BSs share the CEG to improve the spectral efficiency. The machine learning, particularly Deep Q Network, is utilized to optimize the transmission power on Cell-Center Sub-band Group (CCG) and CEG. Instead of assuming that the ratio of the transmission power on CCG and CEG is a fixed number, the transmission ratio is determined by DQN. The simulation results indicate that the modified Strict FR algorithm can achieve significantly higher performance compared to Full FR.