A history of beating optical limits is our best hope for the future

The demand for bandwidth has been a constant drumbeat for decades, and the rise of AI is putting it front and center once again. The meteoric rise of Nvidia stock and the proposed re-furbishing of Three Mile Island’s nuclear reactors highlight the unprecedented need for both AI processing and electricity, but network capacity is also a constraint. Optical technologies will play a key role in data center interconnect for tasks like training machine learning models, as well as interconnect to edge clouds and on-premises servers that for latency reasons, need to do inferencing locally.

This puts the focus on scaling optical networks, an area where Nokia Bell Labs has a storied history. The Labs has been a pioneer of communications technologies generally and optical networking in particular, establishing a research-driven approach that has profoundly influenced industrial labs worldwide. It has pushed the boundaries of what is possible and continues to produce innovations that will be key to the future of optics and AI.

Ensuring optical capacity for the future

Although the capacity of optical fibers was at one time understood as practically limitless, we have had to face the limits set first set forth by Bell Lab’s Claude Shannon. Today, with the extreme demands of AI upon us, we are exhausting the capacity of the C-band erbium amplifier, which has been the mainstay of optical communications for 30 years. But there are other vectors to explore, beyond the C band. According to Shannon, there are three vectors which determine the data capacity of any information channel: spectrum, spectral efficiency and spatial (in this case, fiber utilization). It’s good to remind ourselves quickly of how these vectors have progressed over the years.

Spatial 

In 1970, Corning Glass Works researchers reported the development of an optical fiber with a loss of 20 dB/km. This was good enough to transmit an optical signal and, in the same year, Morton Panish and Izuo Hayashi at Bell Labs, along with a group at the Ioffe Physical Institute in Leningrad, followed this up with an actual transmission using a semiconductor laser. Within seven years, live telephone traffic was being transmitted by AT&T, GTE and others.

The 1970s was in many ways, the decade for glass development, with the Japanese achieving attenuation of almost 0.2 dB/km using infrared near 1.55 μm, close to the Rayleigh scattering limit. The slower work on regeneration would take the next couple of decades leading eventually to wave-division multiplexing (WDM) systems. But even these systems crucially depended on glass advances as well. The most important of these was probably Andrew Chraplyvy and Robert Tkach’s 1994 discovery at Bell Labs of non-zero dispersion fiber, which overcame the non-linearities in low chromatic dispersion glass, which had proven fatal to WDM scenarios. TrueWave fiber would prove to be essential to the rise of WDM in the 1990s.

Although most of the advances in the following decades would be in terms of spectrum and spectral efficiency, we are seeing a renewed interest in spatial for solving some of our current limitations, which we will discuss below.

Spectral efficiency

Modulation is one of the keys to achieving greater spectral efficiency. The earliest optical transmission systems progressed through simple on-off keying (OOK) to phase-shift keying (PSK) and its many variants. Bell Labs later pioneered the use of polarization rotations and mode dispersion, which enabled polarization-division multiplexing (PDM). This is the basis for today’s highest coherent modulation formats such as PDM-QPSK in long-haul and PDM 64-QAM in metro optical networks.

Even more critical than modulation formats was the development of the digital signal processor (DSP), which recovers the original information bits buried in the modulated wave form. In 1979, Bell Labs developed the first single-chip DSP, the DSP-1, a digital filter critical to enabling today’s high-speed transmissions at 100G and above. Many of the advances in carrying capacity relate to algorithms used by the DSP to filter the received signal.

In 2009, Sethumadhavan Chandrasekhar and Xiang Liu at Bell Labs introduced superchannel transmissions using CO-OFDM and flexible grid WDM systems for better spectral efficiency. James E. Mazo’s work at the Labs led to his discovery of the faster-than-Nyquist (FTN) algorithm in 1975, now used in 200G systems to improve tolerance to bandwidth limitations caused by optical filtering. And in 2018, Bell Labs researchers, Junho Cho and Peter Winzer, advanced the use of probabilistic constellation shaping (PCS) to further enhanced coherent optical transceiver performance, enabling today's 400G and 800G systems.

Spectrum

The initial fiber optic systems operated in the infrared range of 1230-1360 nm, as this band exhibited minimal chromatic dispersion and signal loss. Modern fibers, however, show the lowest losses in the C-Band (1530-1565 nm). When the C-Band cannot meet bandwidth requirements, the L-Band is utilized, with occasional use of the S-Band. Future advancements in fiber technology may further enable the use of the S- and L-Bands, as well as other bands for specific applications.

Bell Lab’s researchers, Emmanuel Desurvire, Jay Simpson and Phillippe Becker, in 1987, announced their findings on their erbium-doped fiber amplifier, which was a key development allowing for the amplification of multiple wavelengths simultaneously and enabling WDM. It revolutionized optical transport systems especially sub-sea cable systems, which arguably have been the critical innovation enabling the global internet. Ongoing innovations in amplifier technology will be crucial for expanding the use of other spectrum bands.

Pushing the limits

As we look to the future, we can see that spectral efficiency is now reaching the Shannon Limit. The evolution of DSPs and other components such as analog-to-digital converters (ADCs) will play a key role in improving the spectral efficiency of optical fiber networks, but the spectrum and spatial vectors promise more significant gains. As we have seen, expansion into the C- and L-bands, where Nokia is a leader, has opened up new capacity — but innovations in amplifiers are key. Recent work by Bell Labs and others are showing promise in the S-band using semiconductor optical amplifiers (SOA). Bell Labs researchers recently (2020) demonstrated transmissions of 115 Tb/s across 100nm of spectrum in the 1510–1610nm range. This technology will allow ~60nm of gain in the L- and S-bands.

In terms of spatial, what we see in subsea cables is that capacity has been growing by shrinking the diameter of fibers and thereby increasing the number of fibers per cable without changing the diameter of the cable, which is operationally constrained. Cable diameters for terrestrial cable are not so constrained. Here we have seen 288 fibers per cable and more. Cable manufacturers are also using high-density ribbon fibers to add as many as 3,400 fibers to a 33 mm cable, with 7,800 anticipated. The most promising innovations are in multi-core and hollow core fibers, which offer further scaling using spatial division multiplexing (SDM).

Bell Labs' commitment to pushing the boundaries of optical communication extends beyond its own research labs, fostering collaborations with leading institutions worldwide. One such partnership, with the National Institute of Information and Communications Technology (NICT) in Japan, exemplifies this collaborative spirit. This joint effort focuses on exploring the potential of SDM fiber, particularly multi-core and multi-mode fibers.

As we’ve seen, Bell Labs has been at the center of the optics revolution since its inception playing key roles in the exponential growth of optical network capacity. As we shift our focus from spectral efficiency to the spectrum and spatial vectors of Shannon’s formula, we expect that our researchers will continue to push the limits as we race to meet the needs of AI and other bandwidth hungry technologies.