Find in Library

Localization problem is a significant component of the Internet of Things (IoT) and interference source localization is of great importance in the context of spectrum monitoring and management. However, it remains challenging to quickly but accurately locate an interference source from the distance, especially when little is known about the interference source. To handle this problem, a single learning algorithm can be exploited to search and locate the interference source. However, it is varying dynamics in varying environments that can make the design of such a learning algorithm intractable. In our study, we employ an unmanned aerial vehicle (UAV) to realize the localization. Moreover, a novel multimodal Q-learning framework along with its algorithm is proposed, and the framework combines pattern recognition with Q-learning. The proposed learning architecture can adjust the parameters of Q-learning algorithm dynamically based on the changing environments so as to achieve better detection precision, longer localization distance and shorter searching time. The simulation verifies multimodal Q-learning algorithm's performance on interference source localization along with its capability of adapting to environmental change. The simulation results confirm the proposed concept of multimodal Q-learning. It is shown that the multimodal Q-learning based localization algorithms can outperform various baselines in terms of both accuracy and detection distance. The searching time consumed by the UAV is also largely reduced. This observation indicates that the capability of environmental adaption introduced by the proposed multimodal framework can benefit the Q-learning algorithms.