Fig. 7From: Dynamic handoff policy for RAN slicing by exploiting deep reinforcement learningNumber of handoff. In this figure, we compare the handoff probability and the cumulative handoff times of different algorithms with the progress of the handover decision. As shown in the figure, at the beginning of decision-making, the handoff number of DQN algorithm is slightly higher than greedy, RSS and fixed algorithm. However, as the decision progresses, the number of handoffs required by the DQN algorithm is gradually reduced. It is worth noting that after 30 decision times, the DQN algorithm only performs two more handoffs than the RSS algorithm, but the throughput performance is improved by nearly 30% compared to the RSS algorithmBack to article page