Photonic neural networks for signal equalization in optical fiber transmission

Optical fibers represent the backbone of the telecommunication infrastructure, providing for long-distance transmission at large bandwidths. The massive use of the Internet in modern society, however, generates an unprecedented traffic demand, pushing research towards new solutions that adapt to different scenarios and costs. This involves an evolution in optical transceivers, which are modules responsible for the transmission and reception of optical signals in fiber. The implementation of higher modulation frequencies, complex modulation formats, and Dense Wavelength Division Multiplexing increases the transmitted bandwidth, but simultaneously worsens linear and nonlinear effects accumulated during fiber propagation. These create distortions in the propagated sequence, introducing the need for recovering its integrity after the transmission, operating an equalization process. Nowadays, this is mainly provided at the transceiver level via Digital Signal Processing, requiring huge computational effort, power consumption, and latency. In this work, I explore the use of an integrated photonic neural network for efficient signal equalization in Intensity Modulation/Direct Detection links. The device under test implements an 8-tap time-delayed complex perceptron equipped with tunable amplitude and phase weights on each tap. These determine the optical response of the device and are optimized to provide equalization in various transmission scenarios. Contrary to electronics-based solutions, signal processing is operated entirely in the optical domain, thus minimizing latency and power consumption. The device is realized in a Silicon photonic platform, thus compatible with standard CMOS processes for reduced unit cost in mass production. Its small footprint makes it suitable for scalability and integration in the next-generation transceivers. This design is experimentally tested against Chromatic Dispersion and Self-Phase modulation in 10 Gbaud Pulse amplitude-modulated signals. The training is performed offline by maximizing the eye diagram aperture, verifying the equalization in single- and multi-span links. Simulations allowed studying the scalability of the device to 100G applications, adapting the existing layouts and testing new architectures for future implementations. A comparison with the current state-of-the-art reveals the promising benefits introduced by the proposed all-optical approach, encouraging further steps towards photonic implementations for telecom.

Grantor dc:publisher

Università degli studi di Trento

Year dc:date

2026

Author dc:creator

• Staffoli, Emiliano

Contributors dc:contributor

• Pavesi, Lorenzo

• Photonic Neural Network
• Signal Equalization
• Chromatic Dispersion
• Self-Phase Modulation
• Silicon Photonics

Statement dc:rights

• info:eu-repo/semantics/openAccess
• license:Tutti i diritti riservati (All rights reserved)
• license uri:iris.PRI01

Language dc:language

eng

Identifier

http://dx.doi.org/10.15168/11572_473471
10.15168/11572_473471

OAI identifier oai:identifier

oai:iris.unitn.it:11572/473471