Periodic Reporting for period 1 - ALPI (ALl optical signal recovery by Photonic neural network Integrated in a transceiver module)
Reporting period: 2020-11-01 to 2022-04-30
The always increasing Internet use in our modern society constantly pushes the research toward improvements in telecommunication technologies. Optical fibres play a fundamental role in this field, using light for data transmission at large bandwidths and long distances. The request for increasing capacity in the optical links introduces the need for in-line transceivers for optical power restoration to compensate for higher fibre losses. In conditions of high power transmission, both linear and nonlinear effects alter the shape of the optical pulses travelling in a fiber, which in turn implies the necessity of distortion compensation along the propagation path. Nowadays, digital devices that operate a double conversion between optical and electrical domains accomplish the recovery process. This process introduces latency and large power consumption. Here, at the University of Trento within the ALPI project, we demonstrate an alternative which is based on integrated photonic neural networks to recognize and compensate for distortions in optical signals. The advantages of this technology derive from operating the corrections directly in the optical domain, drastically reducing the power demand and the latency. We demonstrate and validate the use of a 4-channel delayed complex perceptron within IM protocols, while we designed, modelled and fabricated more advanced neural networks, which could operate on multi-symbols or coherent optical protocols. In the validated devices, the input signal is split into 4 channels where the combined actions of delay lines and tunable phase shifters create a desired interference pattern at the output. This working principle has been applied to compensation of distortions induced by linear effects during propagation in fiber. The device has proven effective in its corrective action with the most appreciable results obtained when the distortions in the input signal are more pronounced. Different training methods for the photonic neural network have been explored, with particular attention to the gradient-based Adam algorithm as an alternative to the more widely used Particle Swarm Optimizer. As a result, the delayed complex perceptron and the next-generation ALPI devices represent a viable solution to optical signal recovery. Signal processing is entirely performed on-chip and the training procedure is entirely optical. No external data processing is thus needed, except for the training phase. Moreover, being the photonic neural network of the Feed Forward type, the latency induced in signal processing is maximally reduced. Also, the system is deterministic. The delayed complex perceptron is trained for linear distortions compensation, while nonlinear effects will be treated with the next generation of the ALPI devices.
The POC ALPI proved therefore that the use of a neural network is a practicable approach to mitigate for signal distortions in fiber optic transmission links and can be used to complement more expensive and power-hungry electronic alternatives, such as the use of digital signal processing DSP at the end of the line. The ALPI device brings in several innovations:
■ It is based on a simple NN with few nodes and passive integrated photonic components with advantages on cost, power consumption, footprint reduction and energy saving.
■ It performs real-time correction of both linear dispersions as well as nonlinear self-phase modulation and it does not require intensive electronic and/or computational resources to perform these corrections. The ALPI device can increase the available power budget and relieves the computational efforts of complex DSP in coherent systems.
ALPI’s technology can be applied on an existing network, as an upgrade to improve their performances and as a new transceiver development. Hence, considering, in a conservative manner, a market share of 0.15% in 5 years after the end of the project, we expect to reach a share of the market (Serviceable and obtainable market - SOM) of $26,55 million by applying a disruptive technology such as a neural network that allows ALPI device to perform real time correction of both linear dispersions as well as nonlinear self-phase modulation.
The POC ALPI proved therefore that the use of a neural network is a practicable approach to mitigate for signal distortions in fiber optic transmission links and can be used to complement more expensive and power-hungry electronic alternatives, such as the use of digital signal processing DSP at the end of the line. The ALPI device brings in several innovations:
■ It is based on a simple NN with few nodes and passive integrated photonic components with advantages on cost, power consumption, footprint reduction and energy saving.
■ It performs real-time correction of both linear dispersions as well as nonlinear self-phase modulation and it does not require intensive electronic and/or computational resources to perform these corrections. The ALPI device can increase the available power budget and relieves the computational efforts of complex DSP in coherent systems.
ALPI’s technology can be applied on an existing network, as an upgrade to improve their performances and as a new transceiver development. Hence, considering, in a conservative manner, a market share of 0.15% in 5 years after the end of the project, we expect to reach a share of the market (Serviceable and obtainable market - SOM) of $26,55 million by applying a disruptive technology such as a neural network that allows ALPI device to perform real time correction of both linear dispersions as well as nonlinear self-phase modulation.