The extension of existing end-user mobility dataset based on generative adversarial networks

Abstract

The end-user mobility patterns play an key role in the process of 5G network design. Massive increase of the RAN infastructure complexity creates additional requirements on precise network planning and overall orchestration of the network as such. The possible solution to enhance the statistics feeding the network planning process is to generate massive dataset of the end-user mobility patterns. Unfortunately, we are still constrained with few number of commercialy available datasets, far from the sample statistics needed for machine learning training purposes. Our solution to overcome this problem is to generate additional end user traffic statistics by using generative adversial networks approach. Given the existing sample-constrained end user mobility datasets, GAN network learns to generate new data with the same statistics as the training set. Thus, by leveraging this approach we are able to generate theoretically unlimited samples of realistic end user mobility trajectories. This artificial data jointly with existing limited datasets have the potential to be used for the training blocks of machine learning within the process of the network planning optimization.