Synthetic LiFi Channel Model Using Generative Adversarial Networks

In this paper, we present our research on modeling a synthetic light fidelity (LiFi) channel model that uses a deep learning architecture called generative adversarial networks (GAN). A research in LiFi that requires the generation of many multipath channel impulse responses (CIRs) can benefit from our proposed model. For example, future developments of autonomous (deep learning-based) network management systems that use LiFi as one of its high-speed wireless access technologies might require a dataset of many CIRs. In this paper, we use TimeGAN, which is a GAN architecture for time-series data. We will show that modifications are necessary to adopt TimeGAN in our use case. Consequently, synthetic CIRs generated by our model can track long-term dependency of LiFi multipath CIRs. The Kullback-Leibler divergence (KLD) is used in this paper to measure the small difference between samples of synthetic CIRs and real CIRs. Lastly, we also show a simple demonstration of our model that can run on a small virtual machine hosted over the Internet.