The battle of the modulation schemes
Many industry experts are calling data the ‘new oil’ and just as oil once created amazing economic opportunities, it also produced many challenges in drilling for and distributing the oil. The same is true today with data, except the challenge is making the entire pipeline faster, while also ensuring that networks become easier and more cost-effective to deploy.
To meet these bandwidth and network demands, the industry has been undergoing a major transition to higher speed networks that can deliver 100G, 200G and even 400G speeds. This is why we’ve been seeing the rapid success of core components such as coherent 100G optical modules – both CPF2-ACO (analogue coherent optics) and the newer DCO (digital coherent optics) – that have enabled the first 100G network deployments. These pluggable WDM optical modules offer the lower power dissipation, lower install costs, and higher port density needed to deliver these higher data rates, particularly in the long-haul market over distances of more than 80km.
Due to this rapid adoption of coherent systems and components, the coherent WDM market has been growing very quickly. In fact, Andrew Schmitt, of market research firm Cignal AI, is projecting that this market will continue achieving 40 per cent compound annual growth rate between 2017 and 2020.
But, with the onset of more cost-sensitive applications such as data centre interconnect (DCI), and the number of links less than 80km, the industry is starting to think that coherent is too expensive from a ‘dollar per bit-kilometre’ perspective. This is driving the emergence of competing approaches based on different modulations formats such as four-level pulse amplitude modulation (PAM4) where the emphasis is more on the silicon than the optical technology. This article will look at the various modulation schemes (see panel), discuss pros and cons and where each may win or co-exist with one another.
Discrete multi-tone (DMT) is a modulation format – widely used in digital subscriber line (DSL) systems – that splits the available bandwidth on the optical carrier, or wavelength, into a large number of radio-frequency (RF) modulated sub-carriers. For 100Gb/s transmission, 256 individual sub-carriers are generally used, with each sub-carrier occupying about 100MHz of bandwidth. This results in an aggregate signal bandwidth of 25–30GHz on each wavelength.
DMT can transmit the data in such a way that the capacity of every single sub-channel is maximised. Figure 1 shows DMT transmission measurements at 112Gb/s using 256 sub-channels. In 1(a) we see the expected high signal-to-noise ratio (SNR) at low frequency, and lower SNRs as the sub-channel frequency increases. The way in which DMT uses bandwidth efficiently can be seen in Figures 1(b) and 1(c): many bits are packed into the low-frequency sub-channels, such as sub-channel 5 (in this case, five bits per symbol with a 32 quadrature amplitude modulation (32QAM) constellation), whereas the high-frequency sub-channels are still utilised, but with lower bit-packing (two bits per symbol with a quadrature phase-shift keying (QPSK) constellation) due to the lower SNR. This format offers high spectral efficiency, and fits in the QSFP28 form factor.
The downside of this scheme is it requires high optical signal-to-noise ratio (OSNR) while simultaneously meeting the challenging low power consumption required by data centre operators. Data centres are very averse to anything that is not low power because if it gets too hot, they need to add equipment to cool it all down. This requires significant investment, adding to the cost of operation and equipment, while also taking up valuable space in the data centre. Another downside of DMT is it has been engineered with optical dispersion compensation and is only point-to-point, though more advanced versions of DMT such as single-sideband (SSB) DMT may not require optical dispersion compensation.
Figure 1: DMT modulation. The top graph shows measured signal-to-noise ratio (SNR) versus sub-carrier channel measured after DMT demodulation. 112Gb/s is transmitted with n-QAM data on 256 subcarriers spaced by 109MHz. The graph above left shows 32QAM signal, carrying five bits per symbol, transmitted on low frequency subcarrier 5; and above right shows a QPSK signal carrying two bits per symbol transmitted on high frequency subcarrier 250.
Pulse-amplitude modulation (PAM) is a transmission scheme that features multi-level amplitude signalling. We are all familiar with PAM2, typically known as non-return to zero (NRZ) or on-off keying (OOK), which has been used for many years in standard lower-speed optical transmission. Its successor, PAM4, is now being considered as a possible alternative to coherent technology for shorter distances.
PAM4 uses four levels to signal one of four possible symbols, carrying two bits of data per symbol. Figure 2 shows a typical four-level eye, for both 112Gb/s optical PAM4 simulations (a) and measurements (b). The advantage of the PAM4 modulation scheme for optical suppliers is that it can send double the amount of data at the same rate required by NRZ signalling because it transmits two bits per symbol, instead of the one bit in NRZ.
Figure 2: PAM4 modulation. Simulated 56G PAM-4 optical eye diagram (left) and measured 56G PAM-4 optical eye diagram
PAM4 is ideal for shorter distances, such as intra-data centre (from 0.5km to 2km) and inter-data centre (less than 80km) links that require optical transceivers with much lower cost and power consumption than those provided by long-haul coherent optical technologies. Because PAM4 is a relatively simple modulation scheme, all functionality can be implemented in CMOS, with lower power than the more complex DMT modulation scheme. This also enables the PAM4 to be put into smaller form factors. In the case of 400G, for example, only four lasers are required instead of the eight that would be needed by NRZ to achieve the same capacity. This helps reduce cost and power dissipation in the transceiver module. The downside of PAM4 is it places the most stringent requirements on the optics bandwidth, requires an engineered link with dispersion compensation, is only point to point and may have difficulty scaling to single-channel 400G.
Coherent and Coherent-Lite
As noted, coherent optical transmission is a highly efficient technology for per-wavelength data rates above 100Gb/s and distances above 80km. Coherent transmission simplifies the optical network by using chromatic dispersion (CD) and polarisation mode dispersion (PMD) compensation in the digital signal processor (DSP) chip, rather than optical dispersion compensation as has been traditional for 10Gb/s and lower rate networks. Flexibility in modulation format selection allows spectrum allocation and reach to be optimised; modulation formats from binary phase-shift keying (BPSK) to 64QAM can be chosen depending on the specific link requirements. Coherent also provides the best OSNR sensitivity among transmission technologies, easing network design constraints.
FlexCoherent is a coherent system with flexible symbol rate and grid spacing that can be adjusted to maximise capacity. This technology enables network operators to optimise transmission performance via a range of advanced modulation formats available on a single line card. FlexCoherent technology supports an increasing variety of advanced modulation formats, and is ideal for multi-haul, simple line systems, and scales to higher data rates. The downside of this scheme is that it uses the most complex DSP, consumes the highest power and is not the most cost-effective approach for shorter reaches.
Currently under standardisation in the OIF, ‘Coherent-Lite’ is based upon standard coherent technology, but the silicon chip does not have as many features as those for today’s long-haul markets. While retaining excellent OSNR sensitivity and simplicity of network design, Coherent-Lite offers a reduced feature set comprised of single-rate/single modulation format operation, and DSP processing and forward error correction (FEC) for compensating CD and PMD up to 120km. This leads to lower power consumption as the chip does not need as much processing circuitry and will enable a solution for 80km.
Like other coherent approaches, this scheme works with a simple line system, offers the lowest power per bit and can achieve single-wavelength 400G. The disadvantage is it will require an expensive 7nm DSP chip that represents significant technical risk. Advanced CMOS technology will be required to reduce power consumption and shrink the die area as these chips need to fit in high-density packages such as Double Density QSFP (QSFP-DD) and Octal Small Form Factor Pluggable (OSFP) at 400G. State-of-the-art 7nm CMOS processing is new and this process node must achieve very aggressive power dissipation targets if it is to meet the requirements of the DCI market.
Oclaro's CFP2-ACO coherent module
Weighing up pros and cons
Considering all the pros and cons of the different schemes, coherent transmission will likely prevail for link distances that are 100km or greater coupled with its ability to scale as data rates increase from 200G to 400G and beyond. For short reach DCI applications of between 40 and 80km at 100G, the jury is still out on the preferred modulation scheme. History has shown in the fibre-optic component and transceiver market that the solution that succeeds will be the one offering the fastest time to deployment (fit for purpose), lowest power consumption and lowest cost for the total end-to-end solution. This enables customers to deliver efficiencies in capex and opex.
As a developer of optical components and pluggable transceivers in the coherent transmission world, Oclaro is evaluating all modulation schemes to provide solutions to help solve customers’ most complex problems. For example, it recently announced it is sampling a CFP8 module using PAM4 in an 8x50GBd configuration for 10km transmission over singlemode fibre. This availability of both PAM4 and coherent modulation schemes in optical products is a great step towards delivering speed and bandwidth the industry needs to handle the ongoing data explosion. If data is the world’s ‘new oil’, it will be innovative optics, based on these modulation schemes, driving the rewards.
• Adam Carter is chief commercial officer and Dan Tauber is director, systems engineering, for Oclaro
High-speed modulation formats compared
Pros: High spectral efficiency; Adaptive; Less sensitive to bandwidth of optics; Fits in QSFP28 form factor; Proven technology in low-speed ASDL
Cons: Required OSNR; Power consumption; Chip complexity; Sensitive to system nonlinearities; Engineered link with dispersion compensation; Only point to point operation
Pros: Simplest higher-order modulation format; Lowest complexity chip; Best required OSNR for a direct detect scheme; Lowest power consumption; Fits in QSFP28 form factor
Cons: Most stringent requirements on optics bandwidth; Engineered link with dispersion compensation; Only point to point operation; Difficult to scale to 400G
Pros: Under standardisation in OIF; Simple line system; Lowest power per bit; 400G single wavelength
Cons: Expensive 7nm DSP; Significant technical risk; Very application specific (sub 120km distances only)
Pros: Flexible modulation format-baud rate grid spacing to maximise capacity on any system; Multi-haul application, from DCI to submarine system; Simple line systems; 100-600G under development and scales to even higher data rates
Cons: Most complex DSP; Highest power consumption; Designed for flexibility not for short reach cost-effectiveness