Area-Selective Deep Reinforcement Learning Scheme for Wireless Localization

Abstract

In this paper, an area-selective deep reinforcement learning scheme is proposed to achieve high-quality wireless localization. The conventional localization schemes based on deep learning face several challenges that are labor-intensive for data collection and labeling, lack of adaptability to large-scale, etc. To address these issues, the proposed scheme incorporates deep reinforcement learning (DRL) with a reward-setting mechanism The localization problem is modeled as a dynamic decision process to leverage the capabilities of DRL. The device location is determined by an iterative decision-making procedure, which is an area-selective process. The proposed scheme consists of two distinct modes, namely train and deployment modes. During the train mode, an agent learns the optimal actions for a given environment. The learned agent estimates the position of device in deployment mode. Simulations were conducted to demonstrate the advantages of the proposed scheme and the results showed that it offers better localization performance, adaptability, and time complexity than conventional schemes.