Low-Complexity Recurrent Neural Network Based Equalizer With Embedded Parallelization for 100-Gbit/s/λ PON

• DOI: 10.1109/jlt.2021.3128579
• Publication Date: 2021-11-17T23:03:31Z
• Authors: Xiaoan Huang; Dongxu Zhang; Xiaofeng Hu; Chenhui Ye; Kaibin Zhang
• Keywords: Gigabit; Modulation (music)

To meet the demand of emerging applications, such as fixed-mobile convergence for fifth generation mobile networks and beyond, a 100-Gbit/s/λ access network becomes next priority passive optical roadmap. We experimentally demonstrate transmission intensity modulation direct detection based on four-level pulsed amplitude in O-band by using 25G-class optics. mitigate severe distortions caused inter-symbol interference fiber nonlinearity, low-complexity recurrent neural equalizer with parallel outputs is proposed. Experimental results show that proposed can consistently outperform fully-connected same input/output size number training parameters. The equalizer's sensitivity against quantization also evaluated. further understand complexity actual hardware resource consumption parallel-output equalizers, we implement an 8bits-integer-quantized model FPGA, benefits challenges validated discussed.