High-Level Review of New Modulation Formats for High-Capacity Optical Networking This technical tutorial provides a high-level description of various modulation formats used to address high-capacity needs. This paper covers coherent detection approach, from 50G to 400G+ channel rates, and describes an example of direct detection scheme as well. Insatiable Need for Higher Capacity The exponential demand for bandwidth is fueled not only by increases in the number of users, the access methods and rates, but also the increased importance of cloud, cloud Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide services, and data centers and the number of services (such as mobile, social media, and video in general). On any given night Netflix and YouTube make up greater than 50% of www.xtera.com 1 traffic in the US; 62% of consumers watched online videos in 2015. A great influence on network evolution is likely to be the bandwidth requirements for 4K video, which is hitting the market, to be followed by 8K video. This evolution is illustrated by the advent of new players whose bandwidth needs are strongly increasing and add to those from traditional telecoms service providers. E.g. traffic delivered through content delivery network players progressively increased, and now accounts for about 50% of global Internet traffic. In a traditional telecom network traffic flows from the user to the network central office and back – this is commonly referred to as northsouth traffic. The proliferation of data centers built by OTT providers and communication service providers is having a drastic effect on traffic. Most traffic in a data center network is generated by machine-to-machine traffic between data centers – termed east-west traffic. This east-west DCI traffic represents now a very predominant part of all traffic compared with the north-south traffic. The requirements of the optical backbone networks supporting east-west traffic significantly differ from those for traditional telecom operators with higher capacity per fiber, longer optical data paths and fewer sites where access to the traffic is required. Introduction of coherent approach was driven by higher spectral efficiency needs. And at the Beginning Was Direct Detection Optical data transport started with the simplest (and therefore cheapest) digital coding schemes: On/Off-Keying (OOK), where the physical parameter that is modulated is amplitude (power) of the optical wave. The detection is carried out by a simple photodetector that detects power level hitting the receiver and converts it into a photocurrent reflecting the optical amplitude modulation in the so-called direct detection implementation. Each transmitted symbol takes two values: low optical power level (digital 0) or high optical power level (digital 1). Therefore each transmitted symbol can be encoded with one bit (binary digit): 0 for low optical power level 1 for high optical power level Figure 1: Traditional two-level power modulation for direct detection, with one symbol corresponding to one bit. The basic role of the receiver is to make a decision for each received symbol, which may be corrupted by optical noise and symbol distortion following fiber propagation. In the case of a two-level power modulation like OOK format, the symbol rate is equal to the bit rate. For a 10G carrier modulated by an Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide OOK signal, 10 giga symbols are transmitted per second, corresponding to the same number of bits transmitted per second, i.e. a bit rate of 10 Gbit/s. In telecommunication theory, baud is the unit of the symbol rate; 10 giga symbols per second is then equivalent to 10 Gbaud. Current commercially-available www.xtera.com 2 optoelectronics technologies are limited to about 35 Gbaud (in this document, we will assume a figure of 30 Gbaud for the sake of simplicity). In order to transmit, e.g., 100 Gbit/s per wavelength, more bits per symbol are required. As this has been done for many decades in wireless communications, this can be simply achieved by using multilevel modulation formats where symbols can take more than two values (levels) of one or several physical parameters. Two simple examples of multi-level modulation formats are provided here: Four-level modulation format o Symbols can take 4 different values/levels (e.g. 0, 1, 2 and 3) o Two bits are encoded in one symbol to enable four different values/levels (00, 01, 10 and 11) Sixteen-level modulation format o Symbols can take 16 different values/levels (e.g. 0 to 15). o Four bits are encoded in one symbol to enable 16 different values/levels (0000 to 1111). Here Comes (Optical) Coherent Detection Continuous capacity growth and lower cost per transported bit are key challenges any longhaul optical transmission infrastructure needs to cope with. The advent at the beginning of this decade of 100G channel rate in commercial networks, associated with digital coherent detection (again, not a new thing in radio communications), offered a 10-fold capacity increase compared to the previous transmission technology based on 10G waves. Unlike the 40G developments where multiple formats were assessed and proposed (NRZ, ODB/PSBT, CSRZ, ADPSK, DBPSK, DQPSK...), 100G development avoided the multiple modulation formats “mistake” of 40G. A general consensus was quickly reached around PM (or DP)-QPSK modulation format with coherent detection (PM: Polarization Multiplexing, DP: Dual Polarization, QPSK: Quadrature Phase Shift Keying). Although the net channel rate is close to 100 Gbit/s, the gross line rate is in the range of 120 Gbit/s, or even larger, to account for overhead (e.g. for forward error correction). The basic principle of PM-QPSK modulation format is to trade speed for parallelism at the expense of added complexity, as illustrated in Figure 2. The first parallelism (or multiplexing) step is to encode the data to be transported not into two states like in any binary modulation format (e.g. OOK format), but into four phase states. Consequently one phase symbol transports two bits of data. The quadrature phase shift keying halves the symbol rate compared to the binary phase shift keying (or, equivalently, doubles the number of bits per second, from 30 Gbaud to 60 Gbit/s). QPSK modulation format also halves the spectral width (the symbol duration is longer than the “equivalent” bit duration) but doubles the count of components (e.g. two Mach Zehnder modulators in parallel are required in today’s practical implementation). Figure 2: PM-QPSK modulation format leading to a gross line rate of 120 Gbit/s, assuming 30 Gbaud symbol rate (net channel rate of 100G). Each yellow circle in the constellation plot represents one of the four possible phase states/levels/symbols. Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide www.xtera.com 3 The next parallelism step is to use the optical polarization dimension. Instead of modulating a single optical carrier at 100G using 100 Gbit/s binary modulation format, two independent orthogonal states of optical polarization, at the very same optical frequency (because delivered by the same single laser source), are modulated. Each state of optical polarization carries a 60 Gbit/s stream, leading to an optical signal carrying a gross rate of 120 Gbit/s. can be achieved using the same technologies and techniques (30 Gbaud optoelectronics and two multiplexing dimensions – modulation level and polarization) as for 100G PM-QPSK solution. Going Above (and Below) 100G The second step of multiplexing (multiplexing two orthogonal states of optical polarization) doubles the number of bits per second, bringing the gross PM-16QAM carrier rate from 120 to 240 Gbit/s. This 200G solution, depicted in Figure 3, is the one that has been deployed in the field since 2015, starting with terrestrial networks where high capacities are required (typically for interconnecting data centers). Continuous capacity growth forecast shows a clear need for increasing the capacity per channel beyond 100G for further maximizing the line capacity and lowering the cost of optical transport. Increase in the channel rate For the sake of practicality, the number of states of optical polarization is kept to two, while the number of levels in the modulation format can be increased from 4 to 16, using 16-level Quadrature Amplitude Modulation (16QAM) format. Four bits are encoded in one of the 16 symbols as shown in Figure 3. Compared to the symbol rate, this quadruples the number of bits per second from 30 Gbaud to 120 Gbit/s. Figure 3: PM-16QAM modulation format leading to a gross line rate of 240 Gbit/s (net channel rate of 200G). The 16QAM modulation format requires higher Optical Signal-to-Noise Ratio (OSNR) than QPSK due to the shortest distance between the dots in the symbol constellation plot. If we normalize the average signal energy by the QPSK constellation corner to 1, the normalized average signal energy is reduced to 5/9 for 16QAM constellation. Since signal vectors no longer have the same amplitude over the constellation, 16QAM signals are also less robust to fiber nonlinearities than their QPSK counterparts. Both OSNR and fiber nonlinearities significantly reduce the reach of Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide 200G carriers in comparison with 100G channel performance (typically by a factor of four). Some applications may require fiber capacity higher than what could be enabled by 100G carriers (typically 88 x 100G in the C band of erbium-doped fiber amplifier) together with reach longer than what is achieved with 200G carriers. To respond to this need, optical PM-8QAM modulation was developed (Figure 4). www.xtera.com 4 Figure 4: PM-8QAM modulation format leading to a gross line rate of 180 Gbit/s (net channel rate of 150G). In the other direction, carrier rate lower than 100G may be required for very specific applications, like upgrades of old-generation repeatered subsea cable systems where coherent 100G carriers do not work properly due to the line fiber specifications and chromatic dispersion management scheme. For such demanding applications from an optical transport perspective, two-level Binary Phase Shift Keying (BPSK) modulation format has been implemented in conjunction with polarization multiplexing as shown in Figure 5, for building 50G net channel rate. Figure 5: PM-BPSK modulation format leading to a gross line rate of 60 Gbit/s (net channel rate of 50G). In this case, one symbol is equivalent to one bit. The distance between the two dots of the constellation plot is larger than what is found in the QPSK, 8QAM and 16QAM plots, leading to higher OSNR tolerance and higher robustness to fiber nonlinearities. Coherent 100G channel is the new 10G with excellent performance – cost tradeoff for both terrestrial and subsea applications. Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide Beyond 200G with Today’s Technologies Using the currently-available 30 Gbaud optoelectronics components, 400G channel rate can be achieved by introducing a new multiplexing dimension: multiplexing two 200G carriers in the wavelength domain in order to form a Dual-Carrier (DC) 400G channel. This approach is depicted in Figure 6 where the bit rate figures correspond to gross rates accounting for overhead on the top of the service-carrying traffic. www.xtera.com 5 Figure 6: DC-PM-16QAM modulation format leading to a gross line rate of 480 Gbit/s (net channel rate of 400G). In the practical implementation enabled by the technology available and deployable in 2016, 400G channels are actually mini super channels made of two 200G carriers (hence the full name of today’s solution for 400G channels: DC-PM-16QAM). Many field trials with, e.g., 1 Tbit/s channels claimed in 20152016 were actually carried out by using super channels made of multiple 200G Carriers (like 5 x 200G for a so-called 1 Tbit/s channel). The spacing between the two 200G carriers is an adjustable parameter allowing to increase the spectral efficiency and the resulting fiber capacity. Effective signal processing and spectral shaping optimization at the transmit end can enable carrier spacing as low as 37.5 GHz with acceptable transmission penalty. Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide The figure on next page summarizes the different channel rates that can be obtained with current 30 Gbaud optoelectronics components. The left-hand column indicates the net rate (available for the optical layer payload) while the three right-hand columns give the gross line rate taking into account overhead for forward error correction. The top row includes the three multiplexing dimensions that can be combined to get a net channel rate ranging from 50G to 400G, being understood that the wavelength multiplexing step can be extended to combine more than two 200G carriers in order to build a channel rate larger than 400G. www.xtera.com 6 Figure 7: Table showing different ways to use the various multiplexing dimensions in order to get a net channel rate ranging from 50G to 400G, assuming 30 Gbaud optoelectronics (enabling a symbol rate of 30 Gbaud). Evolutions of Coherent Modulation Formats 60 Gbaud optoelectronics components are expected to become available in the midterm. This will change the game with the possibility, e.g., to build a single-carrier 400G channel based on PM-16QAM modulation format as shown in Figure 8. 60 Gbaud technology will also enable to build 200G carriers with PMQPSK modulation format instead of the PM16QAM that is used today: this means that tomorrow’s 200G PM-QPSK carriers will offer better reach performance than today’s 200G PM-16QAM carriers. Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide Another way to increase the channel rate is to play with the number of levels in the modulation format. Some experimental works have been already reported with 32QAM, 64QAM and even 128QAM modulation format, but they have not turned into actual products yet. In addition to the number of modulation levels and the speed of the optoelectronics components, a lot of activity at component and equipment levels is spent to improve the spectral shaping of the optical signals, increase the spectral efficiency and make more powerful digital signal processing at both transmit and receive ends. www.xtera.com 7 Figure 8: New [Modulation format – Channel rate] combinations enabled by 60 Gbaud optoelectronics (100G PM-QPSK with 30 Gbaud optoelectronics is shown for reference). The carrier multiplexing dimension is available for higher channel rate. Return of Direct Detection: PAM4 Modulation Format Multi-level modulation formats and coherent detection were introduced into optical backbone networks with the primary objectives to increase the fiber capacity and lower the cost per transported bit over fiber lengths (1,000+ km). One clear beneficiary is longhaul optical transmission infrastructure transporting huge amount of date over long distances, as found in content delivery networks. Over transport distances shorter than 200 km, there is a gap between lower- cost optical interfaces supporting up to 10 km reach and higher-cost coherent channel cards. Short optical links result in large output OSNR, making solutions with noise tolerance lower than that of coherent detection suitable and attractive for these applications. Several advanced direct detection solutions, offering different channel capacities, have been examined to fill that optical reach gap, while offering cost, footprint and power consumption advantages over coherent detection approach. This section focuses on the 4-level Pulse Amplitude Modulation (PAM4) format, which is one of the direct detection options identified for achieving optical reach in the range of 100 km. Figure 9: Pulse pattern and eye diagram of a PAM4 signal (Source: Tektronix). Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide www.xtera.com 8 Since PAM4 is a four-level modulation format (Figure 9), each symbol can take 4 different values (e.g. 0, 1, 2 and 3), and two bits are encoded in one symbol (00, 01, 10 and 11). Therefore a 100G carrier (supporting actually a line rate close to 120 Gbit/s to account for FEC overhead) could be achieved using 60 Gbaud optoelectronics components. Due to current limitations in the component bandwidth, the short-term practical approach is to build PAM4 50G signals at 30 Gbaud symbol rate (for an actual line rate of 60 Gbit/s). The polarization multiplexing dimension, as used in 100G PM-QPSK modulation format, is not available here because the game is to use simple direct detection instead of a coherent approach. A 100G channel is then enabled by two PAM4 50G signals, while a 400G signal relies on 8 x PAM4 50G signals. Direct detection is used at the receive end of the link. The line fiber chromatic dispersion must be compensated for by optical dispersion compensator but feed forward equalizer can be used for improving the chromatic dispersion tolerance. An example of feed forward equalization is represented in Figure 10 where the voltage levels of symbols prior to transitions are boosted by a constant factor relative to the voltage levels of non-transition bits. This technique boosts the high frequency components of the waveform relative to its low frequency components in an effort to invert the impairments caused by the transmission channel (backplane or line fiber). Figure 10: PAM4 eye diagrams prior to and after propagating through a backplane, top without transmitter Feed Forward Equalization (FFE) and, bottom, with transmitter FFE (Source: Tektronix). In addition to various types of equalization, adjusting the levels of the signal to walk away from equally-spaced levels is another approach to increase the transmission performance. March 2016 was announced a reference design for a DWDM QSFP28 transceiver that uses PAM4 for 100 Gbit/s (2 x 50 Gbit/s) transmission for data center interconnect requirements at reaches up to 80 km. In March 2016, 400 Gbit/s transmission over 100 km via PAM4 optical modulation format was reported; the transmission comprised eight channels of 50 Gbit/s apiece. Also in Short-reach 400G PAM4 solutions are currently under standardization at IEEE 802.3bs. Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide www.xtera.com 9 Conclusion The objective of this technical tutorial was not to go in the details of the technology of how to design and build high-speed optical transceivers. Instead, the goal was to explain in simple terms the past and future evolution of the optical signals delivered by coherent line interfaces for high-capacity optical transport applications. After the successful introduction of 100G channels delivering excellent performance – cost tradeoff, additional coherent channel rates were developed to cope with different needs (either very long optical reach over non-optimized line fiber, or very high fiber capacity over shorter distances). Coherent approach responds to the exponential rise in bandwidth requirements in backbone networks but is often over engineered from a cost, form factor and power consumption perspectives for shorter optical reach applications. Several players in the industry have determined that a direct detect approach would be a better fit for data center interconnect over distances shorter than 100 km. There has been recently renewed interest in direct detection scheme, with the advent of new formats like PAM4, to meet more modest optical reach requirements, while offering cost, footprint and power consumption advantages over coherent detection approach. Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide www.xtera.com 10 Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide Edition Date: June 2016 Version: 1.0 Maximizing Network Capacity, Reach and Value Over land, under sea, worldwide www.xtera.com 11