Application recognition based on bit-stream shape in mobile networks

Abstract

For mobile network operators, it is important to know the service used by their clients for multiple reasons such as Quality of Service guarantee, throttling, prioritization of traffic, or differentiated pricing. The first step to achieve this is to recognize what application is used by a customer on this smartdevice connected to mobile network. Classical, port-based, or deep packet inspection methods face difficulties due to widely used encryption. In this study, we investigate the feasibility of application recognition only from the bit-stream shape on mobile communication channel. For this purpose, a convolutional neural network for time-series classification was used on the dataset collected by the authors. Training of the convolutional neural network was performed on download and upload bit-streams using a sample duration of 60, 300 and 600 s.