Deep Learning Based Detection of Sleeping Cells in Next Generation Cellular Networks
Deep Learning Based Detection of Sleeping Cells in Next Generation Cellular Networks
复制标题
DOI：
10.1109/glocom.2018.8647689
复制
发表时间：
2018-12
期刊：
影响因子：
--
通讯作者：
Usama Masood;Ahmad Asghar;A. Imran;A. Mian
中科院分区：
文献类型：
--
作者：
Usama Masood;Ahmad Asghar;A. Imran;A. Mian
不断增长的用户体验质量需求对移动的蜂窝网络运营商提出了重大挑战。一个这样的挑战是蜂窝网络中休眠小区的自主检测。休眠小区（SC）是小区降级问题，并且是小区中断检测（COD）中的特殊情况，因为它不会由于BS中的硬件或软件问题而触发任何警报。为了最大限度地减少此类中断的影响，研究人员提出了自主中断检测和补偿解决方案。现有技术的SC检测依赖于路测和用户投诉来识别受影响的小区。然而，由于业务费用不断增加，这种做法很快变得不可持续。为了解决这个特殊问题，我们采用了基于深度学习的框架，该框架使用LTE网络中引入的最小化路测（MDT）功能。在我们提出的框架中，MDT测量用于训练深度学习模型。然后可以快速检测和定位网络中的异常或小区中断，从而显著降低SON中自愈过程的占空比。在我们的模拟设置中，我们还定量地比较和展示了我们提出的方法与最先进的机器学习算法（如使用多个性能指标的一类SVM）的上级性能。
The growing subscriber Quality of Experience demands are posing significant challenges to the mobile cellular network operators. One such challenge is the autonomic detection of sleeping cells in cellular networks. Sleeping Cell (SC) is a cell degradation problem, and a special case in Cell Outage Detection (COD) because it does not trigger any alarm due to hardware or software problems in the BS. To minimize the effect of such outages, researchers have proposed autonomous outage detection and compensation solutions. State-of-the-art SC detection depends on drive tests and subscriber complaints to identify the effected cells. However, this approach is quickly becoming unsustainable due to rising operational expenses. To address this particular issue, we employ a Deep Learning based framework which uses Minimization of Drive Tests (MDT) functionality introduced in LTE networks. In our proposed framework, MDT measurements are used to train the deep learning model. Anomalies or cell outages in the network can be then quickly detected and localized, thus significantly reducing the duty cycle of self-healing process in SON. In our simulation setup, we also quantitatively compare and demonstrate superior performance of our proposed approach with state of the art machine learning algorithm such as One Class SVM using multiple performance metrics.