Abstract : These days, mobile devices usually have multiple network interfaces and there are many usable access networks around the devices. To utilize a wide range of network options properly and make decisions more intelligently, the mobile devices should be able to understand networks’ situations autonomously. The current mobile devices have powerful computing power and are able to collect various network information, and people nowadays almost always carry their mobile devices. Thus, the mobile devices can be utilized to figure out practical quality of service/experience and infer the network situation/context. However, networks have become not only larger but also more complex and dynamic than in the past, so it is hard to devise models, algorithms, or system platforms for mobile devices to understand such complex and diverse networks. To overcome this limitation, we leverage deep learning to devise a mobile device-centric approach to identifying problem spot having the most likely cause of network quality degradation, MoNPI. By using MoNPI, mobile devices are able to identify the network problem spot, which is like a black box to end nodes heretofore. Mobile devices with MoNPI are able to understand networks’ situations and thus take a more proper action.
Index terms : Deep learning, mobile, network, problem spot identification, transmission control protocol