Pushing the Limits and Advancing the Capabilities of Fiber-Optic Networks

Working closely alongside industry giants Nippon Telegraph and Telephone (NTT), NEC Corporation and NEC Laboratories America, Inc., as well as Duke’s Office of Information Technology (OIT), Duke engineers made a major splash at this year’s Optical Fiber Communication Conference and Exhibition (OFC) held in the San Diego Convention Center from during March 24-28.

The premier global event for optical networking and communications research and technology, this year’s OFC attracted 12,500 registrants from 74 countries and showcased more than 630 exhibiting companies. The conference also features a competitive avenue for publishing peer-reviewed research, and Zehao Wang, a third-year Ph.D. student in the laboratory of Tingjun Chen, assistant professor of electrical and computer engineering at Duke, was involved with four accepted papers.

“This conference features the latest advancements and developments of fiber-optic technologies that form the backbone of the global internet, modern data centers and telecom networks that connect the world together,” said Chen. “Our presence at the event this year was particularly special thanks in large part to Zehao’s exceptional work and our collaborators from leading industry and academia.”

Chen’s excitement stems from Wang not just having been involved with the research, but having led one of the papers that was selected as the top-scored paper within its subcommittee, placing it within the top few percent of all submissions. This paper was also selected as finalist for the Corning Outstanding Student Paper Competition and was recently invited to the special issue on top-scored papers from OFC of the IEEE/Optical Journal of Optical Communications and Networking.

In close collaboration with researchers from NEC Labs America, the U.S.-based center for NEC Corporation’s global network of corporate research laboratories, and Trinity College Dublin in Ireland, Wang built an efficient machine learning model to predict the behavior of signal propagation through large multi-span optical networks featuring many amplifiers and parallel wavelength channels carrying diverse data traffic.

Since the power of optical signals degrade as they travel through fiber networks, amplifiers are placed within the network every 10s of miles. However, the behavior of these signals within these amplifiers is not straightforward.

To squeeze as much data into the network as possible, several separate signals in different wavelength channels are multiplexed and combined to be sent over a single fiber strand. But when the data goes through an amplifier, there is not a flat gain across all the wavelength channels, which is a variable that becomes increasingly complex as the data streams goes through multiple amplifiers across large distances and complex networks.

“Optimizing these networks has always required carefully monitoring and measuring the signal power spectrum across the wavelengths and adjusting them based on changes in the system and network configurations,” Wang said. “Our approach is essentially building a ‘digital twin’ of a complex, large-scale fiber-optic network, allowing network operators to predict these effects automatically with largely reduced complexity and overhead.”

A second paper appearing at OFC was noteworthy for highlighting a unique opportunity for corporate/academic research partnerships provided by Duke OIT. Working with Chen and Wang, researchers from Nippon Telegraph and Telephone (NTT) and NEC Labs America (again) sought to add new optical link monitoring functionalities to their commercial transceivers, which convert electrical signals to optical signals and vice versa.

Telecommunication companies typically have to use specialized, expensive devices to investigate how their signals are holding up to travelling long distances by measuring properties such as frequency and polarization state. NTT has been developing techniques to instead get these measurements from the transceivers that are already in the system. [Read NTT’s press release about the research here.]

To put the idea to the test, the researchers needed a dedicated fiber-optic network that wasn’t already in use. That can be an issue, though, as most high-capacity networks are owned by local governments or service providers that aren’t likely to shut down their customers’ internet for hours at a time.

“Duke OIT has deployed and currently operates a fiber network consisting of hundreds of miles of high-capacity fiber-optic cables, and has been very supportive in sharing  this precious resource to facilitate research projects,” Chen said. “This line of research with our industry and academic partners leverages resources from the university, which is a unique aspect of how Duke is fostering collaborations and advancing research in this field.”

This paper was also of special note because it was accepted as a “post-deadline paper,” which is the most competitive collection of entries because they typically involve the latest, most advanced work in the field.

The collaboration also published two additional papers on modeling the input power dependency in transceivers’ bit error rate-optical signal to noise ratio (BER-OSNR) relationship and identifying inline fiber types. All of the work was supported in part by a National Science Foundation EArly-Concept Grants for Exploratory Research (EAGER) award as well as Duke Engineering’s NSF-funded Athena Institute, which focuses on developing edge computing with groundbreaking AI functionality that leverages next-generation communications networks to provide previously impossible services at reasonable cost.

All together, the collection of work supported by close partnerships between Duke Engineering, Duke OIT and commercial colleagues are working to increase the capacity and advance the capability of today’s fiber-optic networks and technologies.

“More than 10 years ago, nobody was paying attention to this sort of work because we were barely able to use up the capacity of the 1 or 10 gigabit Ethernet we had then, especially for wide-area connectivity,” Chen said. “But now with machine learning and AI, we need fiber-optic networks that can carry tens of terabits per second of information. These kinds of projects can help make sure our data networks don’t become a bottleneck for these emerging technologies.”