800G Generation Coherent and the New Normal

Sept. 15, 2021
Now that vaccine rollouts are gathering speed in many, though not all, parts of the world, there is light at the end of the tunnel. But what will the new normal look like?

As the saying goes, “It’s hard to predict, especially the future.” Few people foresaw the huge impact COVID-19 would have on our health or the global economy, let alone our networks – though in many respects, COVID just accelerated trends like the shift to online shopping, streaming entertainment, virtual education, and remote work. Traffic patterns shifted, in terms of both geography and peak time of day, as large numbers of people were required to work or learn from home. Overall internet traffic growth increased from around 25% in a typical year to around 35%.

On the other hand, some 5G projects were delayed and incremental capacity additions on existing platforms were prioritized over network refreshes with new platforms and technologies. Chip shortages became widespread, affecting multiple sectors, from automotive to consumer electronics to networking equipment. Meanwhile, customers and governments became even less tolerant of any network downtime. Now that vaccine rollouts are gathering speed in many, though not all, parts of the world, there is light at the end of the tunnel. But what will the new normal look like?

Things just won’t be the same

The shifts to online shopping and entertainment are likely to remain. The shift to working from home is unlikely to completely reverse. However, many new home workers will be reluctant to begin commuting to the office full time again, and for many, hybrid models that mix in-person office work and remote work will become the new normal. Business travel will eventually resume, though the cost and time efficiency of online meetings and conferences means they will likely continue. Education is the most likely to return to the old normal (i.e., the classroom or lecture hall), though many of the online tools used during the pandemic will continue to be used to complement in-person learning. Internet traffic growth is forecasted to drop back to a still-healthy pre-pandemic mid-20s percentage. The pandemic-induced chip shortages won’t ease for a couple of years, until new fabrication capacity comes online. Post-pandemic, we are likely to see the resumption of delayed 5G projects, together with stimulus-backed broadband infrastructure buildouts to address the digital divide and to better connect rural and underserved communities.

In terms of optical networks, operators continue to face a number of preexisting challenges as they seek to address traffic growth in both a cost-effective and timely manner. They need to reduce the cost per bit per kilometer. They need to scale bandwidth without also scaling power consumption or the number of required rack units. Many also want to grow revenues with new services such as 400 Gigabit Ethernet (GbE) leased lines. And they need to maximize fiber capacity due to the high cost of acquiring and lighting new fiber, especially given the reduced availability of wholesale dark fiber in several markets.

Advances in coherent technology

So how can coherent optical technology help address these challenges? Well, the evolution of the silicon technology used to build ASICs that provide digital signal processing (DSP) and other digital functions to a 7-nm CMOS process node, together with additional improvements in photonic technology that support greater than 90 Gbaud symbol rates, have enabled the evolution to 800-Gbps coherent optical engines. Unlike coherent engines that are constrained by power consumption limits and the size of pluggable form factors like QSFP-DD and CFP2, embedded optical engines prioritize optical performance and the most advanced optical features with larger and more powerful ASICs/DSPs. For example, one such ASIC/DSP has more than 5 billion transistors, compared to around 1.5 billion in the ASIC/DSP of a 400G coherent pluggable.

To address the cost-per-bit-per-km challenge, 800G generation optical engines dramatically increase the outcome of the wavelength capacity-reach dynamic compared to previous generations, as shown in Figure 1. For example, a 600G generation based on 16-nm CMOS and before that a 400G generation based on 28-nm CMOS could deliver their headline data rates to distances of a few hundred kilometers. The new 800G generation can deliver 600 Gbps to more than 2,500 km and 400 Gbps to 7,500 km in terrestrial networks, with even longer distances possible in submarine networks. That 400-Gbps reach of up to 7,500 km makes this new generation ideal for cost-effectively delivering 400GbE services anywhere. With dual-wavelength optical engines, network operators can also deliver three 400GbE client services over two 600G line-side wavelengths, reducing the cost per 400GbE by a third compared to single-wavelength engines that would not be able to utilize the excess 200 Gbps of capacity in each 600-Gbps wavelength. Even the headline data rate of 800 Gbps can extend beyond 800 km with production levels of margin.

Wavelength capacity-reach matters because if an optical engine can deliver 50% more wavelength capacity, it will reduce the cost per bit by 33%, while doubling the wavelength capacity will halve the cost per bit. Wavelength capacity-reach is also closely correlated to power consumption in terms of watts per bit per km as well as footprint in terms of transponder rack units per bit per km.

Three main performance factors

So, what enables 800G generation optical engines to deliver such a leap in wavelength capacity-reach? The answer is simple: baud rates, advanced features, and modem signal-to-noise ratio (SNR), as shown in Figure 2. Increasing the baud rate enables the use of lower-order modulation for the same data rate. Lower-order modulation has a greater Euclidean distance between the constellation points, making them easier to distinguish correctly in the presence of noise. The reduced reach of the higher baud rate itself is largely offset by the ability to increase the wavelength’s power by the same power spectral density, as increasing the baud rate proportionally increases the spectrum of the wavelength. So, as a rule of thumb, doubling the baud rate lets you double the wavelength capacity for less than a 10% decrease in reach. Plus, at 800 Gbps, even small increases in baud rate can dramatically increase reach. For example, increasing the baud rate by 15% can deliver four times the reach of 800 Gbps with probabilistic constellation shaping (PCS) compared to the minimum baud rate required for 800 Gbps at the full PM-64QAM.

A second factor is advanced features. The 7-nm CMOS technology used in 800G generation coherent DSPs has enabled the number of transistors to approximately double compared to the previous 600G generation based on 16-nm CMOS. This extra processing power enables the DSPs to deliver new features such as long-codeword and second-generation Nyquist digital subcarriers.

Compared to conventional QAM modulation, PCS delivers a number of benefits. It provides much smoother granularity compared to the rigid bandwidth steps of conventional modulation and enables the optimal baud rate to be used for a given data rate rather than one that matches the bits per symbol of conventional modulation. In addition, for the same wavelength power and bits per symbol, PCS can provide greater Euclidean distance between the constellation points, thus reducing sensitivity to noise and increasing capacity-reach, with a long codeword taking this gain close to its theoretical maximum.

Meanwhile, Nyquist digital subcarriers reduce the chromatic dispersion (CD) and therefore the noise generated inside the optical engine when compensating for CD. More subcarriers are better than fewer due to the squared relationship between baud rate and CD. For example, eight subcarriers reduce the effect of CD by a factor of 64, compared to a factor of 16 with four subcarriers. Another innovation, dynamic bandwidth allocation (DBA), combines these two features, with PCS used to optimize the data rate of each individual subcarrier based on its penalties, typically with slightly higher data rates on the inner subcarriers and slightly lower data rates on the outer subcarriers.

The third factor is modem SNR, driven by the amount of noise generated inside the optical engine. Keeping this noise low and therefore the modem SNR high is critical, especially at high (i.e., 800 Gbps) data rates, when sensitivity to any noise is very high. High modem SNR requires high performance for all the critical components of the optical engine, digital ASIC/DSP, analog electronics (drivers and transimpedance amplifiers), and photonics (laser, modulator, photodetectors, etc.), as well as the radio frequency and optical interconnects between these components. It also requires these components to be holistically co-designed, which in turn requires deep vendor vertical integration.

800G generation embedded optical engines also leverage these same advanced features (PCS, Nyquist subcarriers, DBA, etc.) and high modem SNR together to maximize spectral efficiency, with C-Band capacity in excess of 40 Tbps possible. Optional L-Band support can further increase total fiber capacity to more than 80 Tbps (C+L). For submarine networks, the modulation and baud-rate flexibility of 800G generation optical engines enable different parts of the fiber’s spectrum to be optimized with different modes for maximum spectral efficiency.

800G generation embedded optical engines can therefore address many of the challenges faced by operators, especially in long-haul and submarine optical networks. They can also play an important role in specific data center interconnect (DCI) and metro scenarios, such as fiber-constrained DCI or metros with high ROADM cascades. So, while it is hard to predict, especially the future, I predict that 2021 will be a great year for the adoption of 800G generation coherent across a broad range of network operators and applications.

PAUL MOMTAHAN is director, solutions marketing, at Infinera.