Building Scenarios on Mobile Network Testbed with a Transmission Characteristics Similarity Model

Optimizing the Quality of Experience (QoE) of mobile video streaming service has emerged as a hot research topic. Simulation has been the dominant research methodology of validation. However, it is becoming clear that simulation does not sufficiently capture the network capacity irregularity of a real mobile network, especially with mobility. The researchers tend to evaluate their optimizations using prototype implementations by real world experiments. Instead of directly running time and cost intensive field test, it is more efficient to evaluate the optimizations on the mobile network testbed. Only when the transmission characteristics of the testbed is similar to those of the actual mobile network, will the results be meaningful to guide further works. In this paper, we emulate experimental scenarios on the mobile network testbed by tuning the testbed configuration parameters. In these scenarios, the testbed has similar transmission characteristics to the actual mobile network. We build a transmission characteristics data set from a real world mobile network, and train the similarity model through metric learning to guide building experimental scenarios on the testbed. The results show that the proposed similarity model outperforms state-of-The-Art similarity models. We also verify that the experimental scenarios really reflect real world scenarios by evaluating the QoE of HTTP Adaptive Streaming (HAS) on the testbed and the actual mobile network.