Skip to main content

Signal proCEssing in optical communication NeTworks Using ReservoIr cOmputiNg

Periodic Reporting for period 1 - CENTURION (Signal proCEssing in optical communication NeTworks Using ReservoIr cOmputiNg)

Reporting period: 2016-07-01 to 2018-06-30

The technological advances in optical communications aim at the highest data throughput over the longest distances with energy efficient and low complexity designs. However, transmission impairments put strict limitations on communication speed and distance in fiber-optic communication systems. On the receiver side, high-speed digital signal processing (DSP) algorithms have improved signal recovery by mitigating linear and nonlinear signal distortions, but they are also facing challenges for future trends. They are efficient as long as nonlinear signal distortions do not become too complicated. Lately, machine learning algorithms have been in the spotlight of the optical communications community. They are being considered for optical network monitoring and optimization, optical header recognition and mitigation of transmission effects. Still, the drawback of applying these tools in ultrafast systems is that they are computationally expensive and still far from reaching real-time processing at telecom data rates.

In the CENTURION project, a straightforward objective was defined:
“To train the experienced researcher (ER) in the concepts of reservoir computing and FPGA programming, in order to append intelligence to the decision making processes in optical communication networks and to offer sophisticated solutions for routing and switching in telecom applications and to reduce storing requirements in sensing applications.”
The accomplishment of this objective was in direct dependence of the individual tasks distributed in three (3) technical WPs that covered the 24-month project period:
• WP1: to train the ER in numerical simulations methodologies, FPGA hardware and FPGA programming, and to familiarize with the equipment and facilities of IFISC.
• WP2: to design and implement theoretical investigations and numerical recipes of the advanced concepts of reservoir computing (RC) and to apply them into crucial problems such as the fast optical header recognition in optical telecom networks and data reduction in the readout/storing process of sensing networks.
• WP3: to build and exploit experimental implementations through FPGA-based or photonic RCs that deal with the above applications.
The key objective of the CENTURION project has been successfully accomplished, following the proposed workflow described in the signed agreement. Specifically:
• During WP1, the ER was trained on simulation methods in Matlab, FPGA programming in Quartus and VHDL, and was familiarized with the equipment and facilities of the IFISC and specifically the Nonlinear Photonics Lab.
• During WP2, the ER designed and implemented theoretical investigations and numerical recipes of the advanced concepts of reservoir computing (RC). After investigating all-optical(AO) and opto-electronic(OE) configurations, an AO configuration with single-nonlinear element and time-delayed optical feedback was selected to perform all planned tasks. This configuration guaranteed the fastest processing methodology of optical communication signals. Optical header recognition was easily executed even with an OE configuration. The SNR conditions set by the acquisition system defines the performance of the classification task for multi-bit header recognition. In several of our studies, a significant finding was consolidated. The RC methodology could apply not only to solve an optical header recognition problem for switching and routing, but to generically post-process optical data streams that suffer from distortions as they propagate in nonlinear media. This finding opened a new research path. Therefore, the project focused on post-processing optical communication signals which were nonlinearly distorted by fiber transmission. This is believed to have a great impact on the ways that we process ultra-fast signals, since it directly competes with the up-to-date DSP methodologies used in the latest optical communication protocols.
• During WP3, the ER built and exploited successfully a photonic-based RC to post-process distorted optical communication signals. The FPGA-based implementation for this task – introduced as a back-up plan, for a verstatile but much slower processing solution – was not needed. Our research showed that the PRC methodology to post-process optical communication signals is generic. It has been applied to short-reach (up to 50km) and long-haul (up to 4000km) fiber-optic communication systems, considering even different encoding techniques (NRZ and PAM-4). These systems have been numerically evaluated and experimentally demonstrated by the ER, the PI and their collaborators.

The results obtained for NRZ and PAM-4 encoding schemes with direct detection are fascinating. The photonic RC was able to process the distorted signals without requiring any analytical description of the physical effects – linear and nonlinear – that cause the distortion. For all studied systems, short-reach and long-haul, with encoding rates from 10Gb/s up to 112Gb/s, the RC post-processing has been beneficial, improving the communication distance by up to 75%. For the short-reach transmission systems, the achieved distances are well above the ones described in the newly established IEEE protocol for short-reach fiber transmission, even though the latter refers to the wavelength region of 1.3μm where chromatic dispersion is minimized. The distance range we achieve with photonic RC post-processing is comparable to state-of-the-art DSP solutions.
A number of future challenges for this type of hardware processing with ultrafast optical communication signals remain. The classification of data streams in multiplexed channels (in polarization or wavelength) is of great significance to target 400G short-reach transmission systems at 1550nm. Moreover, complementary multiplexing methods of the to-be-processed signals need to be explored to establish real-time operation.
The results of the project have been disseminated so far in 3 international conferences, 6 international workshops, 1 published and 1 submitted peer-reviewed journal publication and at least other 10 outreach activities (talks, seminars, tutorials, lab demonstrations).
The CENTURION project established the first validation of neuro-inspired information processing based on photonic implementations, addressing critical issues in the field of signal processing for high-speed communications. Techniques, like extreme learning machines (ELM) and reservoir computing (RC), were demonstrated to allow for data recovery of extremely distorted signals. Introducing a simplified RC approach with sequential data processing architecture, we obtained speed- and hardware- efficient implementations. The obtained results of RC-based processing yield very promising performance, that even competes with the state of the art of digital signal processing. The proposed photonic RC implementation represents a revolutionary tool for ultra-fast signal processing of signals with increased complexity. The concept presented in the CENTURION project is expected to influence the next generation of ultra-fast signal processing in optical communications that might benefit from photonic circuits or optoelectronic modules.
Photonic reservoir computing concept for ultra-fast signal post-processing