For the Long Haul: Maximizing Transmission Distances for 400-Gb/s Signals over the Existing Grid
Internet data traffic continues to increase at a truly impressive pace. The increase is driven on multiple fronts: a surge in cloud computing, increasing use of social media, and the continuing, almost insatiable demand for video. Internet content providers such as YouTube, Netflix, Hulu, Facebook, and many others are pressing the industry for faster speeds so they can send more data to more customers.
In response, the IEEE is looking to adopt a new Ethernet standard to increase the highest data rate from the current 100 Gigabits per second (Gb/s) to 400 Gb/s. This is good news for content providers since it will enable higher-rate interfaces on switches and routers.
Network service providers, responsible for delivering all this data, will need to increase capacity, and it will not be easy. With optical spectrum already being fairly well-utilized, new technologies will be needed to more efficiently use the available spectrum by delivering more bits at a time than is currently possible. But these new, more spectrally efficient technologies come at the cost of greater implementation complexity and reduced reach—the distance light signals can travel before requiring regeneration (see sidebar). Thus almost for the first time, network service providers will have to make tradeoffs among data rate, spectral efficiency, and distance.
As data rates start to extend beyond 100 Gb/s, it’s no longer possible to have both high fiber capacity (total bit rate) and a long reach.
To understand and engineer among these tradeoffs, we at AT&T Research have identified a novel technique that interleaves two high-level modulation formats (e.g., QAM and QPSK) in the time domain, thus allowing for the ability to tune the spectral efficiency for a specific link and distance. The benefit is that modulation techniques with high-spectral efficiency can be combined with modulation techniques with long reach to achieve intermediate spectral efficiencies tuned for a specific link and its distance. Rather than having a fixed spectral efficiency associated with one or two modulation techniques, transmission systems will have a choice of many intermediate spectral efficiencies. The highest spectral efficiency can then be chosen for each link to match the optical reach to the link’s distance.
We have recently demonstrated this method specifically for ultra-long-haul transmissions, in the process setting a distance record (12000 km) for high spectral efficiency (4.125 b/s/Hz) over the current optical wavelength grid.
Increasing need for spectral efficiency
Spectral efficiency was less of an issue in the early days of fiber-optic networks because spectrum was plentiful and data was (relatively) sparse. Fiber-optic systems in the late 1980s carried a single 2.5G signal in a 100-GHz channel, achieving a low spectral efficiency of 0.025 b/s/Hz. As more data started to fill the network, there was a concerted effort to use spectrum more efficiently.
Increases in capacity followed advancements in optical modulation techniques that modify aspects of an optical signal to encode more data. The simplest of these methods, on-off keying (OOK), turns the signal on or off to denote a binary 1 or 0. OOK is an example of amplitude shift keying (ASK), which varies the amplitude (height of the signal) by a certain amount, also to transmit a 0 or 1. A later binary modulation technique, phase shift keying, utilizes phase differences, where a shift in phase indicates a 1 or 0. (See sidebar for more about phase.)
Besides phase and amplitude, light polarization can also be used to convey information. Polarization can be thought of as the property of light to vibrate up and down, or side to side or both. Similar to the amplitude or the phase, the polarization of light can be modulated (i.e., changed from “up and down” to “side to side”) to transmit information. Alternatively, each of the two orthogonal polarizations of light can be used to transmit distinct information streams modulating their amplitude and phase independently. This is called polarization multiplexing. These advanced modulation methods are possible due to the use of coherent detection, which allows amplitude, phase, and polarization to all be detected, measured, and processed. Since coherent detection has become commonplace for these systems, the use of polarization multiplexing has also become de rigueur for long-haul. From this point on, all modulation formats discussed here will be polarization-multiplexed.
Modulation techniques are a big part of how network providers were able to increase network capacity even as the amount of data crossing the network increased exponentially.
But while the demand for communication bandwidth seems unlimited, there is a real limit to the optical bandwidth on one amplified fiber link; this limit is defined by Shannon’s Law (see sidebar). Binary modulation provides spectral efficiency sufficient for the current data rates of 10 to 40 Gb/s, but as data rates reach and extend beyond 100 Gb/s, new higher-order modulation technologies are needed to permit more than a single bit to be transmitted at a time.
One such technique is quadrature phase shift keying (QPSK), which encodes information using four phases of the optical wave, rather than two phases used in simple PSK. Thus where PSK, a binary modulation technique, can transmit only a single bit at a time, QPSK adds two more phases to send an additional bit, and is therefore able to send 2 bits of data at a time for increased spectral efficiency.
Digital data transmitted by QPSK can be represented by points on a circle, each corresponding to four different phases, here 45, 135, 225, and 315 degrees, and enabling 2 bits of information to be sent at a time.
QPSK is just now starting to be used commercially in 100G and some 40G systems.
A higher-order modulation technique of more complexity is quadrature amplitude modulation (QAM), which combines amplitude modulation and phase modulation. Used in the wireless domain since the 1970s, QAM modulates the amplitude of two separate waves that are out of phase by 90 degrees, increasing the number of combinations of amplitude and phase from 2 in the binary formats and 4 in QPSK to 8, 16, or more.
QAM sends a symbol chosen from N combinations of amplitude and phase. Shown here is 16-QAM.
The combination of states (on-off, or any combination of amplitude, phase, or polarization) used to convey information is referred to as a symbol. Different modulation techniques are able to transmit a different number of bits per symbol. In OOK, which has only two states (on/off), a symbol transmits a single bit. Higher-modulation techniques such as QPSK and QAM, which combine various states of amplitude/phase/polarization, can transmit multiple bits per symbol, thus increasing the overall bit rate for the same symbol rate, which is called baud.
Combining polarization multiplexing with high-level modulation techniques further increases the amount of transmitted information, since each polarization can carry many bits at a time (via QAM). Where 8QAM by itself has a theoretical spectral efficiency of 3 bits/symbol, polarization-multiplexed 8QAM (or PM-8QAM) has a spectral efficiency of 6 bits/symbol.
Higher-level modulation techniques such as QPSK and QAM, which send multiple bits per symbol, achieve a high spectral efficiency. There is a cost, however, in the form of greater implementation complexity and reduced optical reach.
Compared to QAM, QPSK has longer reach, but QAM has better spectral efficiency. The following reach estimates are based on real-world conditions and take into account aging, the varying quality of fiber and devices such as amplifiers, and other constraints.
*For a single polarization. Coherent systems generally use polarization multiplexing, which would increase the bits/symbol numbers here by a factor of 2.
Because of their more limited reach, formats of 8QAM and higher are not ideal for the variety of transport distances that occur in real networks. For long-haul distances such as those between major US cities, providers must thus settle for less spectral efficiency, while reserving 8QAM and higher spectral efficiency formats for short-distance links.
A novel method for maximizing spectral efficiency for each link
To provide more flexibility than is currently possible with the fixed spectral efficiency of a single QAM method, we are proposing a technique that interleaves two QAMs in the time domain to obtain much finer granularities of spectral efficiency and thus produce the maximum spectral efficiency for any given transmission distance. And we do so within a fixed grid of optical channels (e.g., 100 GHz was used for the long-distance record).
Instead of encoding exactly one higher-order format at a time—QPSK or 8-level QAM—this new hybrid QAM technique assigns two regular QAMs with different spectral efficiency to different time slots within each time-division-multiplexed frame.
The recent experiment demonstrated not only that the hybrid-QAM approach can transmit 400-Gb/s signals over the conventional 100 GHz ITU-T grid, but that it can do so for distances of up to 12000 km.
By appropriately subdividing the time frame and selecting a ratio of the two QAMs, it’s possible to select any spectral efficiency that falls between the spectral efficiency of the two regular QAMs. The time frame can be subdivided into any number of slots, from 2 up to 100 or more.
In a simple example, the first time slot (or symbol period) might be one of four QPSK symbols, while the second might be one of eight QAM symbols. Thus the signal is a mix of the two formats (QPSK, 8QAM, QPSK, 8QAM...) and achieves a transmission performance halfway between the performance of QPSK alone and 8QAM alone.
The following illustration shows a more sophisticated configuration which results in a spectral efficiency of 5.25 bits/symbol. The time frame is subdivided into four slots, with three assigned 32 QAM (5 bits in each symbol) and the fourth with 64 QAM (6 bits in each symbol), thus [(1x6 + 3x5 bits)/4 symbol slots] for 5.25 bits/symbol.
The spectral efficiency is tuned for a specific link by subdividing the time frame into the appropriate number of slots and choosing between two high-level modulation formats.
For a slightly higher spectral efficiency—5.33 bits/symbol for example—the time frame could be subdivided into three slots, with 64 QAM assigned to the first slot (for 6 binary bits/symbol) and 32 QAM assigned to the second and third (for 5 binary bits in each symbol), thus [(1x6 + 2x5 bits)/3 symbol slots]. These formulas are greatly simplified to illustrate the concept of combining two modulation techniques to achieve an intermediate spectral efficiency; they do not take into account polarization multiplexing and other complexities inherent in an actual network deployment that impact the net spectral efficiency.
Note that the number of bits will differ, and therefore the net bit rate, depending on the mix of different QAM formats. The system adjusts the bit rate for each link to maximize the spectral efficiency for link distance. Very long links would be tuned to use mostly or entirely QPSK to take advantage of that modulation format’s longer reach. Medium-length links might be half QPSK and half 8QAM, while short links would be adjusted to use 8QAM in a larger fraction of slots in a frame.
A hybrid QAM method enables the spectral efficiency to be optimized based on actual distance.
The estimates presented in the following graph, which shows the trade-off between reach and spectral efficiency, are based on optimal conditions using the best available devices (e.g., amplifiers) and fiber, and assume the following: polarization multiplexing, WDM with a 50-GHz channel width with no ROADMs, a modulation rate of 40Gbaud, an FEC overhead of 20%, all-Raman amplification and large-area, single-mode fiber. In an actual deployment the reach values would be reduced by a factor of ~5 to allow for various real-world issues.
Hybrid QAM’s ability to tune the spectral efficiency allows for the selection of intermediate spectral efficiencies between the fixed spectral efficiencies of the single QAM methods.
Testing the hybrid approach for long-haul transmission
We have already demonstrated the efficacy of a hybrid format (specifically 32-QAM/64-QAM), testing it for a distance of 1200 km with 50 GHz spacing, achieving a very high spectral efficiency (8.25 b/s/Hz). Results of this experiment were presented at OFC/NFOEC2012 in paper OM2A.2. (X. Zhou, L. E. Nelson, P. Magill, R. Issac, B. Zhu, D. W. Peckham, P. Borel, and K. Carlson, “1200km Transmission of 50GHz spaced, 5x504-Gb/s PDM-32-64 hybrid QAM using Electrical and Optical Spectral Shaping.”)
To test the hybrid technique for long-haul distances, we assembled a new experiment with a time-domain-hybrid modulation format using a mix of 48 QPSK and 77 8QAM symbols to maximize reach while staying within a 100-GHz channel and were able to achieve a reach of 12000 km (7,500 miles) with high spectral efficiency (4.125 bit/s/Hz). Detailed results of the experiment were presented at OFC/NFOEC in March 2013 as paper OTu2B.4.
The following paragraphs give a high-level summary of the experiment setup, which emulates real-world network conditions with two exceptions: we used a special low-loss, ultra-large-effective area (ULA) fiber from OFS Labs, and we used specially modified transmitters and receivers that adjusted the bit rate for the distance. The 12000-km distance was implemented with a loop apparatus transmitting through a total of 120 fiber spans, each consisting of 100km of ultra-large area fiber. The 400-Gb/s signals mentioned here actually totaled 495-Gb/s: 400 Gb/s of simulated client data and 95 Gb/s of overhead, mostly for forward-error-correcting (FEC) coding.
In the experiment, we generate eight 100-GHz-spaced 400-Gb/s polarization-multiplexed time-domain hybrid QAM signals from sixteen 50-GHz-spaced external-cavity-lasers each with a linewidth of about 100 kHz. Each 400-Gb/s signal is created by using two adjacent wavelengths: we first use each wavelength to generate five frequency-locked 9.9-GHz-spaced subcarriers through a novel five-subcarrier generation method, and then modulate each subcarrier with a spectrally shaped 9.7-Gbaud time-domain hybrid QPSK-8QAM baseband electrical signal. Using digital methods, we achieve close-to-ideal Nyquist pulse shaping. Finally we pass the modulated optical signals through a polarization-multiplexing emulator to generate polarization-multiplexed hybrid QAM optical signals.
The WDM transmission is carried out over a re-circulating loop-based transmission test platform consisting of four 100-km fiber spans. To reduce the fiber nonlinear impairments, we use the previously mentioned ultra-large area fiber having, at 1550nm, 150um2 average effective area and 0.177dB/km average attenuation. To reduce noise from the optical amplifiers, low-noise bidirectional all-Raman amplification is used in every span.
The experimental setup for time-domain hybrid QAM and standard 100GHz channel spacing to demonstrate ultra-long-haul transmission. (MZM = Mach-Zehnder modulator, ILF = interleaving filter, pre-DSP = pre-transmission digital signal processing, IQ = in-phase and quadrature-phase, MOD = modulator, DAC = digital-to-analog converter, POL MUX = polarization multiplexer, OA = optical amplifier, Pol. scrambler = polarization scrambler, pumps = Raman pump lasers, O/E = optical-to-electronic converter, ADC = analog-to-digital converter, post-DSP = digital signal processing following transmission.)
After transmission, a coherent receiver with offline DSP is used to receive and process the transmitted optical signals. Two new DSP algorithms have been implemented at the receiver to improve the transmission performance. First, a new cascaded, two-stage adaptive equalization method is used for polarization demultiplexing and mitigation of some transmission impairments. The first-stage equalizer, implemented before the carrier phase recovery, has a relatively shorter filter length (21 taps used in this experiment), while the second-stage equalizer, used after the phase recovery, has a longer filter length (201 taps were used in this experiment). Such a two-stage equalization method effectively mitigates the low-frequency distortions from the transmitter and the receiver. Secondly, a new training-assisted method is used for carrier-phase recovery, free of cyclic-slips and error propagation. This not only improves the receiver sensitivity by about 1dB by eliminating the need for differential coding and decoding, but also enables the use of the above described two-stage equalization method.
Nyquist-shaped, 100-GHz-spaced 8x495Gb/s WDM signals. Insets show the constellation diagrams of one of the 495Gb/s signals before and after transmission.
What’s required to make it work
The recent experiment demonstrated not only that the hybrid-QAM approach can transmit 400-Gb/s signals over the conventional 100 GHz ITU-T grid, but that it can do so for distances of up to 12000 km even while maintaining high spectral efficiency (4.125 b/s/Hz).
While others have reported 400-Gb/s WDM transmission, they have done so using nonconventional channel spacings (e.g., 75, 85, 125, and 135 GHz) that are inconsistent with the existing grid and not currently used in any operating network. A method incorporating unconventional spacing, if deployed, would require new types of ROADMs. It would also require modifying transponders to simultaneously generate a range of data speeds, a potentially wasteful scenario when a device capable of generating 200-GHz-wide signals is frequently used to generate signals of only 50 GHz width on the shorter links. Another significant challenge would be managing the spectrum and its use on all routes in a mesh network. Because the light-paths generated at different locations and with different widths will, at times, ride over the same link (as they split off from one another or combine with others), some blocks of spectrum could easily become “stranded” with no node able to use them.
Our proposed method will also require some changes to network devices, but we believe these changes will add little cost. Transponders will need to adjust the bit rate based on the optimal spectral efficiency for a specific distance of a link, and will need to do this separately for each channel, while receivers will need to detect which hybrid QAM the transmitter is sending. More significantly, our proposed method will require a tighter coupling between an aggregation/grooming layer (e.g., OTN switching) and the transport layer.
But few other changes would be needed, allowing most of the standardization already in place to remain. No changes would have to be made to the current grid spacing (whether 50 or 100 GHz), as long as the number of bits per second is allowed to vary. Nor would changes to existing ROADM technologies be needed, since each channel in the hybrid method is the same width. Likewise, other parts of the common infrastructure, including the fiber and optical amplifiers, require no modification. And except for the tighter coupling between OTN switching and transport layer, no new demands are placed on managing or operating the transport network such as wavelength planning.
Perhaps the most significant adjustment will be thinking in terms of flexible, not fixed, bit rates while considering the tradeoffs needed as the amount of network traffic on some fibers starts to approach the Shannon limit of the available spectrum. For two decades the industry has been able to count on rapid fiber capacity growth. As data rates start to extend beyond 100 Gb/s, it’s no longer possible to have both high fiber capacity (total bit rate) and a long reach. Something has to give.
But it is possible to work within these constraints by making the system more flexible and by using creative means to support intermediate spectral efficiencies and intermediate reaches, while adjusting the bit rate appropriately, as we have demonstrated with this time-domain hybrid QAM method.
It’s all about spectral efficiency
Spectral efficiency (SE) measures how efficiently the available spectrum is being utilized. It is stated as the number of bits transmitted per second for each Hertz of optical spectrum (b/s/Hz).
A single typical metro 10G signal sent over a channel of 200 GHz using binary modulation has an SE of 0.05 b/s/Hz, very poor. Today’s 100G long-haul packs more information into a single 50 GHz channel (via QPSK and polarization multiplexing) resulting in a much better SE (2 b/s/Hz). Even better SE (up to 8 b/s/Hz) is possible through higher-level modulations such as QAM.
With constant improvements in SE over the years as well as other advancements, the fiber-optic industry has been able to increase network capacity even while the amount of network traffic has skyrocketed.
Reach vs. spectral efficiency
Reach refers to the distance a signal can go before requiring regeneration.
Signals weaken over distance due to loss in the fiber and must be amplified every 80 to 100 km. Amplification involves passing the signal through optical amplifiers (specifically EFDAs) placed along the link. Optical amplifiers are very effective, inexpensive, and fast, but they do introduce noise that degenerates the signal.
Once the degradation reaches a point just shy of introducing errors in the transmission (at the optical reach of that signal on that system), the signal must be regenerated, an involved process requiring expensive devices to perform both optical-to-electrical and electrical-to-optical conversions to reconstitute the original signal.
Signals with complex modulation (e.g., QAM) are more susceptible to noise and thus require regeneration at shorter distances than simpler techniques. That is, they have a shorter reach. And it’s because of the short reach of QAM and other complex modulation that network providers generally use the less spectrally efficient QPSK for long-haul distances and reserve consideration of the higher spectral efficiency of QAM for only short distances.
Shifting the phase to convey information
Phase describes the position of the waveform relative to some reference point (which may be another wave). It is measured as an angle in degrees.
Here are three examples of a phase shift relative to the “zero” on the time axis: (a) a full wavelength cycle having a zero or, equivalently, 360 degrees, (b) 90 degrees (1/4 of a wavelength cycle), and (c) 180 degrees (half cycle).
Binary phase shift keying (BPSK) and quadrature phase shift keying (QPSK) as PSK techniques both exploit phase to encode and transmit data. BPSK uses a simple shift in phase to send a single bit (0 or 1); QPSK uses four phases to send 2 bits at a time.
Bumping up against Shannon’s Law
In information theory, Shannon’s Law (or Shannon—Hartley’s Theorem) defines the limits of how much usable data can be transmitted over a communications channel. Specifically it states that the maximum possible data speed achievable in a single channel is a function of the bandwidth and the signal-to-noise ratio.
For the last two decades, Shannon’s Law has been more of a theoretical concern for the fiber-optic industry. Optical fiber has enormous capacity to transport information, and so much spectrum was available, the industry was able to simultaneously increase capacity, reach, and data rates.
But as data rates approach 400 Gb/s, the limits defined by Shannon’s Law start to come into view, meaning future gains in capacity will entail tradeoffs among spectral efficiency, the data rate, or the reach.
About the authors
Xiang Zhou received his PhD degree in electrical engineering from Beijing University of Posts and Telecommunications in 1999. From 1999 to 2001, he was a research fellow at Nanyang Technological University, Singapore where he conducted research on optical CDMA and wideband Raman amplification. Since October 2001, he has been a senior member of the technical staff at AT&T Labs - Research, working on aspects of long-haul optical transmission and photonic networking technologies. His areas of interest include Raman amplification, polarization-related impairments, optical power transient control, advanced modulation formats, and DSP at 100 Gb/s and beyond. He has authored or co-authorized more than 100 peer-reviewed journals and conference publications. He holds 34 patents in the US and is the associate editor of Optics Express. He is member of the OSA and a senior member of IEEE.
Lynn Nelson received an Sc.B. degree in engineering from Brown University, and MS and PhD degrees in electrical engineering from the Massachusetts Institute of Technology. Before coming to AT&T Research in 2007, she was with Bell Laboratories, Lucent Technologies, working on fiber nonlinearities, wavelength division multiplexing, and polarization mode dispersion; and she was technical manager of the Fiber Systems Testing Group for Optical Fiber Solutions, which became OFS Labs. Now at AT&T Labs - Research, she focuses on high-capacity transmission at 100Gb/s and beyond, modulation formats, and polarization issues for AT&T’s long-haul network.
She has authored or co-authored over 150 peer-reviewed technical publications and four book chapters, holds 15 U.S. patents, and is a fellow of the OSA.
Peter Magill received his BS in Physics from the University of Dayton, Ohio in 1979 and his PhD in Physics from the Massachusetts Institute of Technology in 1987. He then joined AT&T Bell Labs at the Crawford Hill Lab on the characterization of advanced lasers, optical access networks and data-over-cable access protocols. He transferred with Lucent Technologies as it was spun out of AT&T in 1996, to head their access research department. He managed the R&D of passive optical network (PON) systems and cable modem headend equipment. In 2000 he returned to AT&T, focusing on advancing fiber communication technologies for the entire network (inter-city, metro and access) including high-speed transmission systems (100 Gb/s and beyond) and dynamic networks at wavelength and sub-wavelength rates. In 2012 he became Assistant Vice President of Communications Technology in AT&T Labs supporting research on wireless technology and networks, network design and optimization as well as optical systems.