Silicon photonic-electronic neural network for fibre nonlinearity\n compensation

Abstract

In optical communication systems, fibre nonlinearity is the major obstacle in\nincreasing the transmission capacity. Typically, digital signal processing\ntechniques and hardware are used to deal with optical communication signals,\nbut increasing speed and computational complexity create challenges for such\napproaches. Highly parallel, ultrafast neural networks using photonic devices\nhave the potential to ease the requirements placed on the digital signal\nprocessing circuits by processing the optical signals in the analogue domain.\nHere we report a silicon photonice-lectronic neural network for solving fibre\nnonlinearity compensation of submarine optical fibre transmission systems. Our\napproach uses a photonic neural network based on wavelength-division\nmultiplexing built on a CMOS-compatible silicon photonic platform. We show that\nthe platform can be used to compensate optical fibre nonlinearities and improve\nthe signal quality (Q)-factor in a 10,080 km submarine fibre communication\nsystem. The Q-factor improvement is comparable to that of a software-based\nneural network implemented on a 32-bit graphic processing unit-assisted\nworkstation. Our reconfigurable photonic-electronic integrated neural network\npromises to address pressing challenges in high-speed intelligent signal\nprocessing.\n