Abstract Visible light communications (VLC) has been regarded as a promising technology for high-speed indoor wireless accessing since it can offer both lighting and network. However, the spectral efficiency of the VLC system based on orthogonal frequency division multiplexing (OFDM) is always smaller than RF-OFDM because light-emitting diodes (LED) require real-value signals. Pilots occupy the spectrum in proportion for channel estimation(CE) to improve communication quality. Based on this consideration, we firstly present the idea of introducing deep learning technology into the CE scheme in the VLC system and propose a CE scheme based on Deep Neural Networks(DNN) perform as well as conventional CE schemes with fewer pilots. The result of experiments validates the feasibility of DNN-based CE.

中文翻译：

---

可见光通信信道估计的深度神经网络方法

摘要 可见光通信（VLC）由于可以提供照明和网络，被认为是一种有前途的高速室内无线接入技术。然而，基于正交频分复用 (OFDM) 的 VLC 系统的频谱效率总是小于 RF-OFDM，因为发光二极管 (LED) 需要实值信号。导频按比例占用频谱用于信道估计（CE）以提高通信质量。基于这种考虑，我们首先提出了将深度学习技术引入到 VLC 系统中的 CE 方案的想法，并提出了一种基于深度神经网络（DNN）的 CE 方案，其性能与传统的 CE 方案一样少，导频数更少。实验结果验证了基于 DNN 的 CE 的可行性。