We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon
Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Over the past few weeks, Infinera has had the opportunity to perform a submarine network field trial over the MAREA trans-Atlantic cable with our ICE6 optical engine, which already leads the industry in terrestrial reach at its headline data rate of 800 Gb/s over terrestrial network routes.
MAREA is essentially the benchmark for trans-Atlantic cable systems, as it was specifically designed for high-performance coherent transmission – it is the cable system on which every submarine network vendor’s implementations will be measured.
The ICE6 results were truly amazing, achieving two incredible industry firsts. As with all trials like this, we need to be clear on the difference between the “hero result,” which has no additional deployment margin, and the “deployable result,” where there is allocated margin to allow real services to be deployed at this data rate or with this total capacity. In summary:
MAREA hero results over 6,640 km:
30 Tb/s of total capacity on a single fiber pair
700 Gb/s data rate per wavelength
MAREA deployable results over 6,640 km:
28 Tb/s of total capacity on a single fiber pair
Up to 650 Gb/s data rate per wavelength
I’m fortunate to have had the chance to discuss the ICE6 trial with two of the key people who made it happen: Dr. Steve Grubb, Global Network Optical Architect at Facebook, which operates several of the MAREA fiber pairs, and Dr. Pierre Mertz, Infinera Fellow and a world-class submarine networking engineer when it comes to optimizing subsea transponder performance.
The MAREA Cable SystemGeoff Bennett: Steve, you were involved in the design of MAREA – perhaps you could give a quick summary of why it’s so iconic in terms of performance metrics.
Steve Grubb: Of course. MAREA is a trans-Atlantic cable that stretches from Bilbao, Spain to Virginia Beach, VA, U.S. – a cable distance of about 6,640 km. I think it’s fair to say that MAREA is the highest-performing cable system in the world because it was specifically optimized for maximum capacity per fiber pair. It utilizes a large-area, low-loss optical fiber type based on a pure silica core and has excellent performance. The cable system also has short amplifier spacing of 56 km, allowing an excellent optical signal-to-noise ratio. When MAREA became ready for service in February 2018, it had a design capacity that rivaled all of the other contemporary operational trans-Atlantic cables combined.
As a result, every vendor of submarine transponder equipment has trialed its transponder technology on this cable (both in single-pass and the 13,200 km loopback in order to assess trans-Pacific performance) in a bid to set the highest records in total fiber capacity and spectral efficiency. Pierre and I were involved in a similar trial of Infinera’s ICE4 technology in September of 2018, and that deployed capacity record still stands. Now it’s time for the next generation to show us what it can do.
Geoff Bennett: Steve, I remember seeing a conference presentation from you, I think it might have been back in 2013, at the time when the MAREA cable was being designed. You said then that despite the incredible capacity boost that the first generation of coherent transponders had delivered, you were pretty confident that there was “a lot more gas in the tank” in terms of subsea fiber capacity. Did you ever think that would translate into the ability to operate error-free 700 Gb/s wavelengths over an Atlantic cable?
Steve Grubb: Geoff, of course I did – I never had a moment of doubt! No, I’m kidding…I think back then even the idea of 8QAM or 16QAM trans-Atlantic was considered to be a real stretch. But we’ve had 16QAM technology deployed and delivering over 24 Tb/s of capacity on MAREA for over a year. To give readers some context, with the results of the latest trial, we see that a single MAREA fiber pair can deliver almost as much capacity as the entire SpaceX satellite system will when it’s fully deployed (Ed: 32 Tb/s indicated on page 5 of the FCC filing). The total MAREA cable capacity should be capable of 224 Tb/s given these latest trial results.
Geoff Bennett: According to analyst reports, the hyperscale ICPs like Facebook already account for over two-thirds of international submarine network capacity. What do you use it all for?
Steve Grubb: 3.1 billion people worldwide use one of the four major Facebook platforms at least one time per month, and the vast majority of these users are outside of the U.S. We also have an increasing number of data centers outside of the U.S. Hence the need for large amounts of international submarine traffic.
Geoff Bennett: Pierre, if we look at the increase in transponder capacity that happened around 2012 with the move to the first generation of coherent, that delivered a 10-fold increase on older submarine cable types. MAREA and ICE4 together deliver another threefold increase on top of that. What would you say were the biggest factors that gave that improvement?
Pierre Mertz: As a matter of fact, Steve and I gave a webinar on those factors after the ICE4 trial. But if I had to choose one feature it would be the move to Nyquist subcarriers with ICE4. As we move to ICE6, there’s a really fantastic interaction between subcarriers, our long-codeword PCS, and new nonlinear algorithms and mitigation techniques (e.g., super-Gaussian distributions), as well as dynamic bandwidth allocation that provides us a lot of granularity in the way we can tune the performance of an individual wavelength.
Geoff Bennett: Steve, I listed two different data rates in the summary – 700 Gb/s and 650 Gb/s. Could you clarify the difference between a hero value and a deployable value?
Steve Grubb: Submarine networks have a long tradition of pushing the boundaries of optical performance, which then spills over to terrestrial network performance. So, it’s normal to include data rate or capacity results that are right on the edge of the forward error correction (FEC) limit – in other words, it’s error-free transmission but an operator would not normally feel comfortable operating services under those conditions. So, the capacity of 30 Tb/s and the per-wavelength data rate of 700 Gb/s are both hero numbers, representing the upper bound of this innovative technology.
In contrast, what Facebook relies on is how much capacity we can expect under real operational conditions with an optical signal-to-noise ratio (OSNR) safety margin that takes into account natural fluctuations along the amplifier chain or losses we might incur if the cable suffered a break and repair, for example. We believe those are the deployable results of 28 Tb/s fiber capacity and up to 650 Gb/s per wavelength.
Geoff Bennett: Pierre – how did you achieve 28 Tb/s over a single fiber pair with commercial margins? I have to say I thought the ICE4 capacity of 24.0 Tb/s must be running close to the Shannon limit…are we breaking the laws of physics here?
Pierre Mertz: Not quite…but you’re right that we really are close to the theoretical limits with ICE6. Just to be clear, every fiber pair in the world has its own unique Shannon limit, which is to say that every fiber pair in the world has a unique bandwidth number, and a unique signal-to-noise ratio. What we’re able to do with ICE6 is dial in the right set of performance parameters to allow our compensation algorithms – whether it’s the FEC, or the DBA, or our nonlinear algorithms – all of that flexibility is what allows us to push right up against the Shannon limit.
To do that, we found that operating at a lower baud rate provides us better spectral efficiency – and the result is an optimal 450 Gb/s per wave in order to achieve 28 Tb/s in total. This is a really interesting result because it means that as a submarine network operator starts to fill the cable, they could operate at high data rates, like 650 Gb/s, that are less spectrally efficient, but more cost effective on an interface basis…and then as the cable utilization increases, they can switch toward a lower data rate and higher spectral efficiency and add transponders to provide the capacity they need.
Geoff Bennett: And to be clear – did you just test one wavelength on the entire fiber, so there was no chance of inter-channel interference?
Pierre Mertz: No, that wouldn’t be a realistic result. We always try to test as close to reality as we can because we use amplified spontaneous emission (ASE) utilizing ASE noise generators – these are like an erbium-doped fiber amplifier on steroids – to generate optical load across the entire spectrum, just like if you fill the fiber with data channels.
So, we’re not actually breaking the laws of physics…however, we’re pushing them harder and harder. In this diagram you can see a simplified Shannon equation. It really only has two terms on the right-hand side – the bandwidth term, B, and the log term based on the signal (S)-to-noise (N) ratio. When MAREA was first brought into service, the amplifier bandwidth was exceptional because they have a very flat gain over about 4.5 THz. Increasing that amplifier bandwidth versus older submarine amplifier technology increased the capacity in a linear fashion. Inside the log term is where ICE6 is doing the work, and on the graph you can see that if we increase the capability of the transponder, it gets harder and harder to translate that into more capacity.
Geoff Bennett: So, Steve, I assume Facebook still needs additional submarine network capacity growth year on year – where do you go next on the technology roadmap?
Steve Grubb:We still have capacity on MAREA, with additional fiber pairs to light up. We are also continuing to build and plan new submarine cables. The great thing about ICE6 is that, as Pierre said, the higher data rate per wave, the better the cost per bit – and that includes the fact that we need fewer transponders that consume less rack space and less electrical power. The 28 Tb/s option with ICE6 reduces the network element count by 60% vs. the boxes previously required for 24 Tb/s…that’s a huge improvement, especially in cable landing stations where real estate and power are often at a premium.
Where do we go after that? Long-term, the trend for new submarine cables is space-division multiplexing, or SDM, a topic that Pierre and I also gave a webinar presentation on along with Dr. Sergei Makovejs of Corning. In simple terms, SDM is where we would design a cable system based on less capacity per fiber pair, but the ability to support more fiber pairs per cable. In fact, there is a viable roadmap toward a 1 petabit trans-Atlantic cable in the next five to 10 years, should we need to stay on the same growth trajectory.
Geoff Bennett: That’s an amazing prediction – however, I know submarine network demand is forecast to grow to the point where we may need that petabit cable even sooner! Steve and Pierre –