contact button
September 28, 2017

Infinera Makes (Terabit) Waves at ECOC 2017

By Geoff Bennett
Director, Solutions and Technology

Last week the Infinera team, along with our mobile demonstration unit, the Infinera Express, attended the European Conference on Optical Communication (ECOC) 2017 exhibition and conference in Gothenburg, Sweden.  The exhibition attracted 5,234 visitors, with 313 exhibiting companies from 27 countries and a record 3,189 people visiting the Market Focus sessions across the three days.

At the conference, Infinera achieved two industry-first milestones around advanced coherent technologies:

  • A demonstration of a 1024QAM (quadrature amplitude modulation) signal including constellation shaping.
  • A 100 gigabaud (GBaud), 32QAM single wavelength transmission.

As you can see from the figure below, Baud rate and modulation order are two of three possible axes of scaling for individual transponder capacity, and if 1024QAM was combined with 100 GBaud transmission the result would be a 1.5 terabit per second (Tb/s) data rate!

Three Axes of Scaling for Individual Transponder Capacity
Figure 1: Three Axes of Scaling for Individual Transponder Capacity

At this point I will come clean.  In an Optical Society of America webinar a few years ago, I explained that it would be “really hard” to implement a single wavelength with terabit data rates, and at the same time I explained that increasing the modulation order results in dramatically shorter optical reach.  This is one reason why the parallel approach of super-channels for scaling transponder or appliance capacity has been phenomenally successful in the data center interconnect (DCI), long-haul and subsea markets.  Super-channels allow us to combine lower-data-rate waves into a higher-data-rate optical channel that’s treated as a single entity from an operational point of view.  Service providers can run bigger networks with smaller teams of engineers, and can respond to more dynamic changes in terabit-scale demand using unique architectural innovations such as Instant Network.

So why was increasing Baud rate so difficult a few years ago?  Let’s delve a bit deeper into the technology options for scaling line card capacity:

  • Baud rate. The Baud rate of a wavelength is the rate at which modulation symbols are sent.  The most common long-haul data rate today is 100 gigabits per second (Gb/s), using polarization-multiplexed quadrature phase-shift keying (PM-QPSK) modulation.  The typical Baud rate is 32 GBaud, and PM-QPSK carries 4 bits per Baud.  The additional data over 100 Gb/s is consumed by Optical Transport Network forward error correction.  In general, the Baud rate is limited by the opto-electronics technology that is available.  For several years 32 GBaud has been the most cost-effective Baud rate, but more recently this limit has been exceeded thanks to smaller feature sizes on the application-specific integrated circuits used in modern coherent systems.  In other words, after reaching a plateau for several years, opto-electronics has moved to a higher rate, with 48 to 66 GBaud products on the horizon.
  • Modulation order. Another way to increase data rate is to carry more bits in each symbol.  polarization-multiplexed binary phase-shift keying (PM-BPSK) carries 2 bits per symbol, PM-QPSK carries 4 bits, PM-8QAM carries 6 bits, and PM-16QAM carries 8 bits.  Thus, a 32 GBaud, 16QAM wavelength will have a data rate of 200 Gb/s.  There is a drawback to increasing modulation order, however, because the result is dramatically reduced reach.  While PM-16QAM increases the data rate by a factor of two compared to PM-QPSK, it reduces reach by about 80 percent.  However, PM-16QAM is a very important modulation technique where longer reach isn’t needed – such as the metro and DCI markets.
  • Infinera uniquely has access to a third axis of scalability: implementing multiple, parallel wavelengths on the same line card in order to create a coherent super-channel. We make this practical using our large-scale photonic integrated circuit (PIC) technology, but it also gives us a distinct advantage because we can focus development on one, two, or all three of these dimensions, depending on the current state of optical and electronic technologies.

Let’s look at our two post-deadline papers in more detail. In the first paper, Dr. Ryan Going of Infinera’s PIC team presented lab results for what we believe is the first reported demonstration of single-wavelength, 1 Tb/s data rates using 32QAM modulation at an astonishing 100 GBaud symbol rate.  While opto-electronic power has definitely moved on since my OSA webinar a few years ago, a key capability that makes this very high Baud rate possible is the use of photonic integration.  In simple terms, as Baud rates increase there is a very high value in being able to locate optical components as close together as possible, and the ultimate limit is for those components (i.e. laser, modulator, waveguides) to be integrated onto the same chip, using the same material – in this case indium phosphide.  Silicon germanium drivers were also integrated into the package.

In the second paper. Dr. Robert Maher discussed the practicalities and consequences of extremely high order modulation – in this case up to 1024QAM at 66 GBaud, including advanced constellation-shaping algorithms.  To understand the value of constellation shaping we need to imagine what a “symbol” actually means, and in this explanation, I will focus on a single polarization state – remember that in a real implementation there would be an X and Y version of each symbol, resulting in twice as many bits carried.

Comparing 16QAM, 1024QAM, and a Shaped 1024QAM Constellation
Figure 2: Comparing 16QAM, 1024QAM, and a Shaped 1024QAM Constellation

In Figure 2, we see a 16QAM constellation on the left, which consists of 16 symbols, with each symbol transporting four bits of information. Typically, each symbol has an equal chance of being transmitted through the system, so on average the amount of information being sent is 4 bits per symbol. For 1024QAM, shown in the center of Figure 2, there are 1024 independent symbols, each carrying 10 bits per symbol, assuming each symbol is given an equal chance of being transmitted through the system.

However, constellation shaping constrains the chances of certain symbols being transmitted in order to apply a form of “natural selection” in order to optimize the carrying capacity of the wavelength. This can be done because the lower-power symbols, closer to the origin, have a greater chance of being transmitted than the higher-power symbols. What this ultimately achieves is increased spacing between the symbols, thus making the shaped format less sensitive to noise. The trade-off is reduced transmission capacity, but the big advantage is that the noise tolerance can be adaptively tailored to match the signal-to-noise ratio of any given transmission link.  In other words, Constellation Shaping allows us to maximize the data rate for the optical budget of a given fiber path (i.e. a given reach).

On the right of Figure 2 is the actual experimental result of constellation-shaped 66 GBaud 1024QAM presented by Dr. Maher at ECOC.

Note that the optical reach for 1024QAM will be relatively short – but it could find a sweet spot for the distances needed by DCI customers as they consume ever-increasing amounts of capacity.

From Demonstrations to Production Networks

The two papers I’ve referred to are technology demonstrations – their purpose is to show what kind of technologies will become important in future products.  We know, for example, that 400 Gb/s products are getting closer to shipping, and we also saw at ECOC technology demonstrations of 600 Gb/s per wavelengths taking place and participated in discussions of 800 Gb/s per wavelength. Infinera’s fifth-generation Infinite Capacity Engine (ICE5) is targeting 600 Gb/s per wavelength and ICE6, our sixth generation, is targeting at least 800 Gb/s per wavelength.

The optical reach for these higher data rates and for modulations like 1024QAM will be relatively short in terms of distance – but it will likely find a sweet spot for short-haul DCI customers as they consume ever-increasing amounts of capacity.

Infinera has also recently demonstrated some leadership at the other end of the distance spectrum – subsea. Recently Seaborn put their Seabras-1 subsea cable into production using Infinera’s commercially available XTS-3300 based on ICE4, the fourth generation of our Infinite Capacity Engine. On Sept 20, Infinera and Seaborn jointly announced that a new industry benchmark was set with 18.2 Tb/s over the 10,500+ km link using 8QAM and exceptionally tight channel spacing.


The innovative terabit wave capability test at ECOC is an indication of where single-wavelength access and metro solutions, as well as multi-wavelength sliceable super-channel technology for DCI, metro core, long-haul and subsea, could be heading in years to come. For Infinera this is a vital proof point of our ability to push the boundaries of the single-wavelength optical performance curve, but to also have the unique ability to combine multiple wavelengths in a single PIC-based module to deliver the benefits of super-channel scalability, namely service responsiveness and ease of operations, at the optimum price point for our customers. Infinera is commercially shipping ICE4 products today that are leading the market in terms of performance and ease of use.  We are already demonstrating ICE5 and ICE6 test results, as well as showcasing technologies, like those shown at ECOC, that highlight the tremendous future scalability of coherent digital signal processors combined with photonic integrated circuits.

For more information, click here to contact us.

Related links:


Legal Disclaimer