In this article, we utilize the Faster‐Than‐Nyquist (FTN) mechanism combined with the deep neural network (DNN) nonlinear post‐equalization scheme to achieve spectrum compression of the multiband carrierless amplitude phase (CAP) signal in the case of a fixed channel spacing for visible light communication (VLC) system. We found that the number of input layer nodes plays a decisive role in the DNN equalization effect. And a 1.26 Gbps FTN CAP 49QAM VLC system over 1 m free space transmission is successfully demonstrated. We reduced the normalized bandwidth of the CAP 16QAM signal for each sub‐band from 1.05 to 0.9 after quadrature dual‐quaternary (QDQ) coding combined with DNN post‐equalizer, and the total bandwidth of three bands is compressed from 315 MHz to 270 MHz. The spectrum compression ratio is 14.29% while ensuring BER below 7% FEC threshold (3.8e−3), and the spectral efficiency (SE) is 4.67 b/s/Hz. To the best of our knowledge, this is the first time achieving such high compression ratio in Gb/s level DNN based multiband FTN CAP signals in VLC system.

中文翻译：

---

可见光通信系统中基于DNN后均衡的多频带正交双四进制FTN CAP 49QAM信号高频谱压缩方案

在本文中，我们将Fastn-Than-Nyquist（FTN）机制与深度神经网络（DNN）非线性后均衡方案相结合，以在固定情况下实现多频带无载波幅度相位（CAP）信号的频谱压缩可见光通信（VLC）系统的通道间隔。我们发现，输入层节点的数量在DNN均衡效果中起着决定性的作用。成功演示了1.26 Gbps FTN CAP 49QAM VLC系统在1 m自由空间上的传输。在将正交双四进制（QDQ）编码与DNN后均衡器相结合之后，我们将每个子带的CAP 16QAM信号的归一化带宽从1.05降低到0.9，并且三个频段的总带宽从315 MHz压缩到270 MHz 。频谱压缩比为14。同时确保BER低于7％FEC阈值（3.8e-3）时达到29％，频谱效率（SE）为4.67 b / s / Hz。据我们所知，这是VLC系统中首次在基于Gb / s级DNN的多频带FTN CAP信号中实现如此高的压缩率。