Localization with Deep Neural Networks using mmWave Ray Tracing Simulations
Localization with Deep Neural Networks using mmWave Ray Tracing Simulations
复制标题
DOI：
10.1109/southeastcon44009.2020.9249699
复制
发表时间：
2020-02
期刊：
影响因子：
--
通讯作者：
Udita Bhattacherjee;C. K. Anjinappa;L. Smith;Ender Ozturk;I. Guvenc
中科院分区：
文献类型：
--
作者：
Udita Bhattacherjee;C. K. Anjinappa;L. Smith;Ender Ozturk;I. Guvenc
世界正在朝着更快的数据转换发展，更有效的用户本地化是首要要求。这项工作研究了使用深度学习技术进行无线定位，同时考虑毫米波（mmWave）和低于6 GHz的频率。学习新的神经网络模型的能力使得定位过程更容易和更快。在这项研究中，深度神经网络（DNN）被用于在两个静态场景中定位用户设备（UE）。我们提出了两种不同的方法来训练神经网络，一个使用信道参数（功能），另一个使用信道响应向量，并使用初步的计算机模拟比较它们的性能。我们观察到，前一种方法产生高的定位精度：考虑到所有的用户有一个固定数量的多径分量（MPC），这种方法是依赖于MPC的数量。另一方面，后一种方法独立于MPC，但与第一种方法相比，它的性能相对较差。
The world is moving towards faster data transformation with more efficient localization of a user being the preliminary requirement. This work investigates the use of a deep learning technique for wireless localization, considering both millimeter-wave (mmWave) and sub-6 GHz frequencies. The capability of learning a new neural network model makes the localization process easier and faster. In this study, a Deep Neural Network (DNN) was used to localize User Equipment (UE) in two static scenarios. We propose two different methods to train a neural network, one using channel parameters (features) and another using a channel response vector, and compare their performances using preliminary computer simulations. We observe that the former approach produces high localization accuracy: considering that all of the users have a fixed number of multipath components (MPCs), this method is reliant on the number of MPCs. On the other hand, the latter approach is independent of the MPCs, but it performs relatively poorly compared to the first approach.