The communications technology journal since 1924 2013 • 11

Overcoming the challenges of very high-speed optical transmission October 14, 2013

Overcoming the challenges of very high-speed optical transmissionOver the past 15 years, the capacity of optical fiber transmission systems has risen by more than three orders of magnitude. At the same time, the maximum achievable distance for a single link is 10 times greater than what it was over a decade ago. Innovative optical technology has resulted in a significant drop in the cost-per-bit for transport. Essentially, operators want this cost to continue to fall, as this will help them to meet the constantly growing demand for bandwidth.

modulation format, but with the arriv-al of 400Gbps and 1Tbps DWDM chan-nels, the race for higher transmission speeds is on. Such rates pose new chal-lenges that are extremely difficult to overcome, so much so that high-speed transmission capabilities are a bench-mark of technology leadership.

Terabit optical transmission: tech-nical challenges and solutionsThe basic requirement for next-gen-eration Tbps systems is to increase spectral efficiency compared with the current 100Gbps systems, for the same link distance.

The need for better spectral efficien-cy originates from two opposing sides of the telecom ecosystem: operators that want to fill the fiber spectrum to delay the deployment of new infra-structure for as long as possible; and component vendors that want to max-imize the sales volumes of 25GHz band-width electronic components, used in 100Gbps, before investing in the next generation.

One of the most commonly deployed solutions designed to improve spectral efficiency is to increase the modulation constellation. This approach is often adopted in satellite and radio links, where very large constellations of up to 2048QAM are common. In optical com-munications, however, amplification noise makes it impractical to use mod-ulation formats that are more compli-cated than 16QAM – even for distances of just a few hundred kilometers.

However, this approach is limited when it comes to achieving channel bitrates over 10Gbps, due to the cost of high-speed electronic devices, and the complexity in mitigating fiber propa-gation impairments. So, in a manner similar to developments made in wire-less communications, intensity modu-lation in optical fiber has been replaced by complex multilevel modulation for-mats. With the addition of inline polar-ization, the symbol rate decreased even further, and coherent detection was added to compensate for propagation impairments in the digital domain.

Current 40Gbps and 100Gbps DWDM systems use DP-QPSK as a

FA BIO CAVA L I E R E , LUCA GIORGI A N D ROBE RTO SA BE L L A

BOX A Terms and abbreviations

ADC analog-to-digital converterBCJR Bahl Cocke Jelinek Raviv CNIT Consorzio Nazionale Interuniversitario per le TelecomunicazioniDP-16QAM dual polarization 16QAMDP-QPSK dual-polarization quadrature phase shift keyingDSP digital signal processingDWDM dense wavelength division multiplexingFEC forward error correctionFFE Feed Forward Equalizer ICI inter-carrier interferenceISI inter-symbolic interferenceLDPC low-density parity checkMAP maximum a posteriori MI mutual information

MIMO multiple input, multiple outputOFDM orthogonal frequency division multiplexingOSNR optical signal-to-noise ratioPAR peak-to-average ratio PDM polarization-division multiplexed PMD polarization mode dispersionQAM quadrature amplitude modulationQPSK quadrature phase shift keyingROADM reconfigurableadd-dropmultiplexerSCM subcarrier multiplexingSDN software-definednetworkingSE spectralefficiencyTFP time-frequency packingWDM wavelength division multiplexingWSS Wavelength Selective Switch

Applying intensity modulation (a relatively simple transmission scheme compared with those of modern radio systems) in spectral direct-detection systems is one method that has been used to increase network capacity. This method uses dense wavelength division multiplexing (DWDM) and optical amplification, to achieve capacities as high as 800Gbps on a single fiber, corresponding to 80x10Gbps DWDM channels, over link distances of up to 3,000km without the need for signal regeneration.

2

E R I C S S O N R E V I E W • O CTO B ER 14, 2013

Spectrally efficient Tbps fiber

One way around this problem in a radio environment is to use frequen-cy multiplexing because this approach lowers the constellation size and sym-bol rate, and can support 400Gbps and 1Tbps transmission rates. However, applying the same techniques success-fully used in radio to optical commu-nication is not always straightforward. OFDM, for example, is an encoding method that is widely used in radio sys-tems, but its application to optical com-munications is much more complex.

For optical to reach the desired trans-mission rates, a number of things need to be in place: accurate high-speed analog-to-digital converters (ADCs) are required; the optical carriers need to be phase synchronized; a long cyclic prefix overhead is needed to compen-sate for chromatic dispersion; fiber sensitivity to phase noise needs to be accounted for, as do the nonlinear effects that result from high peak-to-average ratio (PAR).

The solution usually adopted in fiber systems is Nyquist-WDM, subcarrier multiplexing (SCM) that is spectrally efficient, where modulated carriers are first filtered close to the Nyquist band-width – half the symbol rate – and then densely frequency multiplexed with the help of steep electrical or optical filters to prevent inter-carrier interfer-ence (ICI) from intensifying.

As an example, 2.24Tbps can be transmitted on 375GHz bandwidth using 10×DP-16QAM-modulated carri-ers, giving an SE of 5.97bps/Hz – almost three times the 2bps/Hz provided by 100Gbps systems. However, there are at least two reasons why Nyquist-WDM is not the ultimate solution for high-speed transmission:

it is incompatible with installed ROADMs that support 50GHz-spaced optical channels (in accordance with the ITU-T standard); and it is incompatible with many of the installed DWDM links on account of its low tolerance-to-noise ratio and non-linear effects.

Is it possible to overcome these issues and satisfy operator requirements to carry out capacity upgrades on their networks in a seamless way? This is the challenge the Ericsson Research team in Pisa, Italy has taken on together

with its local academic partners Scuola Superiore Sant’Anna and CNIT.

The Ericsson approach: time- frequency-packing modulation Orthogonal signaling is the usual method to maximize spectrum usage for both radio and optical commu-nications, as it removes both ISI and ICI. The maximum spectral efficiency possible with orthogonal signaling is SE = log2M, where M is the number of constellation symbols. So, it appears that good spectral efficiency is only achievable with large constellations – and these do not work in optical links. But is this really the case?

Consider a scenario where multiple carriers are spaced in frequency by F, each employing an M-ary modulation format with a symbol interval of T. By relaxing the orthogonality condition (FT < 1), SE can be increased – without increasing M, and by acting on F and T instead. From information theory, we get SE = MI /(FT), where MI is the mutual information between transmitter and receiver and is a decreasing function of FT. This guarantees that a maximum value exists for SE, which is illustrated in Figure 1.

Faster-than-Nyquist signaling1 is an example of non-orthogonal trans-mission and involves sending time and frequency overlapping pulse trains. The optimal values of F and T are the smallest values that guaran-tee the minimum distance between the constellation symbols is equal to the Nyquist case. However, the

complexity of the receiver quickly becomes unmanageable when this approach is used.

In the Ericsson approach SE is maximized once receiver complex-ity has been fixed through design. Implementing the TFP concept in prac-tice can be achieved by applying nar-row filtering, with a bandwidth that is much lower than the Nyquist one, and an LDPC-encoded QPSK signal – which is easier to generate and more resilient to noise than 16QAM. The lower band-width of this approach allows F to be reduced, which ultimately increases spectral efficiency.

Frequency overlapping was not used in the practical implementation of TFP in an Ericsson-Telstra field trial (which is described later in this article). Instead, each carrier was individually sampled and processed.

While such an implementation is a sub-optimal one, an additional increase in SE is possible using chan-nel shortening and by applying multi-channel receiver techniques. As illustrated in Figure 2 (see over) the distortion introduced by the transmis-sion filter at the receiver is recovered by digital signal processing (DSP) – an adaptive equalizer processes the sig-nals received on two orthogonal states of polarization to compensate for fiber-chromatic and polarization-mode dis-persion; the equalizer output is then input to a BCJR detector, which itera-tively exchanges information with an LDPC decoder until the correct code word is detected.

FIGURE 1 Time frequency packing – working principle

1/FT

System working pointMutual information (MI)between transmitterand receiver

SE=MI/FT

Frequency spacing x symbol time (FT)

3

E R I C S S O N R E V I E W • O CTO B ER 14, 2013

The total aggregate bitrate of 1.12Tbps was obtained by eight optical carriers, each modulated by a 140Gbps narrow filtered DP-QPSK signal, corre-sponding to 35Gbaud. The baseband fil-tering bandwidth was 10GHz – which is much lower than the Nyquist-WDM value.

Although it is not required by the TFP modulation format, the carriers were unequally spaced so that a pair of them could be allocated to each ROADM frequency slot (as illustrated in Figure 3B). The system exploited different LDPC codes to balance net SE with error-correction capability, pro-viding a way to implement an adap-tive optical interface without having to change modulation format – the more traditional solution.

To finely adjust the transmitted capacity to the propagation conditions, the system could be configured with six different code rates (9/10, 8/9, 5/6, 4/5, 3/4 and 2/3), depending on accu-mulated OSNR and propagation pen-alties. This technique is more accurate and less hardware demanding than 16QAM-to-QPSK modulation switch-ing, which always requires a halving of the transmission capacity.

Receiver structureFor each carrier, the receiver uses a con-ventional polarization-diversity opti-cal front end, in which the incoming signal and local oscillator are first split into two orthogonal polarization states and then combined separately. Local oscillator frequency and phase are not locked to the frequency and phase of the signal, and any difference is com-pensated for by DSP, which can recover any practical value of frequency offset in the order of 1GHz.

The output of the front end, the pho-to-detected signals of four balanced photodiodes (two for each polariza-tion), is sampled and then digitally processed. No chromatic dispersion compensation is performed along the link, so the accumulated dispersion, of about 17,000ps/nm/km, is entirely com-pensated for by a frequency domain

Ericsson-Telstra field trial The theory of TFP and the benefits it brings were put to the test in a field tri-al carried out by Ericsson on a Telstra optical network, transmitting a 1Tbps channel on a 995km fiber link between Sydney and Melbourne. To provide experimental evidence that TFP is a viable solution for seamless capac-ity upgrades of operator networks, the 1Tbps channel was transmitted through a ROADM node placed in Canberra together with one 100Gbps DP-QPSK and three 40Gbps co-propa-gating channels.

The link scheme, including fiber span distances, is shown in Figure 3A. In the ROADM node, optical filters are present with 50GHz spaced central fre-quencies and 41GHz bandwidth.

FIGURE 2 Time-frequency packing receiver architecture

Inputsignal

Localoscillator

Optical frontend

LDPCdecoderEqualizer

BCJRdetector

FIGURE 3A Field trial link and transmitted spectrum

71 75 55 53 40 67 53 50 80 79 54 75 90 54 100

Frequency (THz)192.85 192.90 192.95 193.00 193.05

Distances in km

Transmitter 1Tb/s Receiver 1Tb/s

Melbourne

MelbourneCanberra

Sydney

Sydney

f)

g)

Canberra

AUSTRALIA

FIGURE 3B Field trial link ROADM frequency curves

4

E R I C S S O N R E V I E W • O CTO B ER 14, 2013

Spectrally efficient Tbps fiber

equalizer, and then by an adaptive FFE – which accounts for other linear impairments, such as polarization rotation, residual dispersion and polar-ization mode dispersion. The equaliz-er output feeds four parallel 4-state BCJR detectors2 followed by four LDPC decoders. The BCJR and LDPC blocks iteratively exchange information, for a maximum of 20 iterations, to achieve MAP detection according to the turbo-equalization principle3.

Field trial resultsFigure 4A illustrates the measured ROADM amplitude transfer function and SE within each ROADM frequency slot for three cases: 1. 1Tbps channel alone; 2. 1Tbps channel with 800GHz; and 3. 100GHz spaced 40Gbps and 100Gbps

co-propagating channels.

SE is defined as the ratio between the maximum net bitrate (not including FEC overhead) that ensures error-free operation and the ROADM bandwidth. Figure 4B shows the resulting SE val-ues, obtained by optimizing the code rate individually for each carrier.

There is no appreciable difference in performance among the three cas-es, which implies that the interference between the 1Tbps channel and neigh-boring channels, with lower bitrates, is negligible. Similarly, no penalty was measured on either the 40Gbps or 100Gbps channel due to the presence of the 1Tbps channel.

To put pressure on the system, the polarization mode dispersion (PMD) was increased by means of a PMD emulator. No SE variation was detect-ed up to 170ps of additional differen-tial group delay, and a 5 percent drop was observed with a delay of 200ps. These are excellent results given that the maximum group delay expected in a 3,000km link is about 50ps.

To further test the stability of the solution, measurements were taken every 15 minutes during a 24-hour peri-od, and no difference was detected in the overall performance.

By varying the carrier power (using a code rate of 5/6 for all the sub-chan-nels), the resilience of fiber to non-lin-ear propagation was also tested during the field trial. An optimal sub-channel

power of about -1dBm was found to minimize the non-linear propaga-tion penalty, and a maximum OSNR penalty of 1.3dB was obtained with a power range between -3dBm and 1dBm – demonstrating that TFP can cope with typical system tolerances, such as those due to gain flatness of the optical amplifiers.

Taking the optimized carrier power as -1dBm, the received OSNR with 0.1nm resolution bandwidth varies between 15dB and 1dB along the carriers – a value that is compat-ible with the majority of installed DWDM links.

FIGURE 4A Maximum spectral efficiency

7.0

6.5

6.0

5.5

5.0

3.

1Tbps channel with 800GHz spaced adjacent channel

1Tbps channel with 100GHz spaced adjacent channel

1Tbps channel alone

192.80 192.85 192.90 192.95 193.00

192.8 193.2193.0 193.6193.4 194.0193.8 194.4194.2

Frequency (THz)

Frequency (THz)

Spectral efficiency

7.0

1.

2.

WSS curve

ConclusionsThe field trial demonstrates the suit-ability of the TFP approach to meet operator demand for high- capacity upgrade of DWDM networks. The proof points provided by the field tri-al show that the long-haul distances of the 1Tbps system are comparable with that of a 100Gbps system. The 1Tbps system, however, provides three times the spectral efficiency, is compatible with 50GHz ITU-T frequency grid and installed ROADMs, it can coexist with installed 40Gbps and 100Gbps chan-nels, it provides stable operation over time, and is robust with respect to sys-tem and fiber tolerances.

FIGURE 4B Received spectra

5

E R I C S S O N R E V I E W • O CTO B ER 14, 2013

Although the focus of the field was on 1Tbps transmission, the mod-ularity of the TFP solution makes it adaptable for 400Gbps. A straightfor-ward method of four carriers instead of eight can be used together with channel shortening and multichan-nel detection techniques to improve the spectral efficiency dramatical-ly. This adaptation is important giv-en that 400Gbps optical interfaces for metro and regional distances, based on 16QAM, are the current focus of stan-dardization work.

But a solution for long haul, with link distances greater than 500km, has yet to be found, and TFP can play a key role in providing one: 400Gbps capabilities are being introduced in SSR and SPO product families, enabling high-speed connectivity between routers in next-generation IP over WDM and ultra-high capacity packet optical transport.

In addition, the capability of TFP to finely adjust the throughput on a per carrier basis works well with concept of SDN – where bandwidth resources need to be rerouted easily according to new service demands or dynam-ic changes of the traffic load – as is required by data center virtualization.

Beyond the technical result, the field trial was a good example of an agile and informal working environment, where researchers from industry and aca-demia came together, exchanged ideas, and encouraged unconventional think-ing to create an innovative solution.

1. IEEE/OSA Optics Express, December 2011, Faster-than-Nyquist and beyond: howtoimprovespectralefficiencybyacceptinginterference,availableat:http://www.opticsinfobase.org/oe/abstract.cfm?uri=oe-19-27-26600

2. IEEE/OSA, Journal of Lightwave Technology, 2007, Using LDPC-Coded Modulation and Coherent Detection for Ultra Highspeed Optical Transmission, available at: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=4357908

3. IEEE,GlobalCommunicationsConference,2012,EfficientconcatenatedcodingschemesforerrorfloorreductionofLDPCandturboproductcodes,availableat:http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6503469

References

6

E R I C S S O N R E V I E W • O CTO B ER 14, 2013

Spectrally efficient Tbps fiber

Fabio Cavaliere

joined Ericsson Research in 2005. He is an expert in photonic systems and technologies

focusing on WDM metro solutions for aggregation and backhauling networks and ultra-high-speed optical transmission. He is theauthor of several publications, and is responsible for various patents and standardization contributions in the area of optical communications systems. He holds a D.Eng. in telecommunications engineering from the University of Pisa, Italy.

Luca Giorgi

joined Ericsson Research in 2007. He manages the Ericsson Research Optical

Laboratory in Italy. His research focusesonstudy,prototypingandfieldtrials of innovative optical systems for radio access, backhaul and transport networks. He has authored several publications and holds various international patents. He has a D.Eng. in telecommunications engineering from the University of Pisa, Italy.

Roberto Sabella

joined Ericsson in 1988. He is manager of the Italian branch of Ericsson Research, and of

Research and Innovation in Italy. His areas of expertise are packet-optical networksandtechnologies,trafficengineering and routing, and telecommunications networks. He has authored more than 100 papers for international journals, magazines and conferences, as well as two books on optical communications. He holds more than 20 patents and has been adjunct professor at the Sapienza University of Rome. He is a senior member of IEEE and has guest edited many special issues in several journals and magazines. He holds a D.Eng. in electronic engineering from the Sapienza University of Rome, Italy.

Acknowledgements

The authors gratefully acknowledge Telstra and everyone involved in the fieldtrialfortheircontribution.

7

E R I C S S O N R E V I E W • O CTO B ER 14, 2013

To bring the best of Ericsson’s research world to you, our employees have been writing articles for Ericsson Review – our communications technology journal – since 1924. Today, Ericsson Review articles have a two-to-five year perspective and our objective is to provide our readers – who include ICT players, decision makers, telecom operators, service providers, manufacturers, students, people interested in telecoms, and players in all other industries that depend on networks and telecommunications – with up to date insights on how things are shaping up for the Networked Society.

Address :Telefonaktiebolaget LM EricssonSE-164 83 Stockholm, SwedenPhone: +46 8 719 00 00Fax: +46 8 522 915 99

Web:www ericsson.com/review

Ericsson Technology InsightsAll Ericsson Review articles are available on the Ericsson Technology Insights app available for Android and iOS devices.Download from Google PlayDownload from the App Store

Publisher: U f Ewaldsson

Editorial board: Håkan Andersson, Hans Antvik, Ulrika Bergström, Joakim Cerwall, Deirdre P. Doyle, Dan Fahrman, Anita Frisell, Magnus Frodigh, Jonas Högberg, U f Jönsson, Magnus Karlson, Cenk Kirbas, Sara Kullman, Kristin Lindqvist, U f Olsson, Patrik Regårdh and Patrik Roséen

Editor: Deirdre P. Doyledeirdre.doyle@jgcommunication se

Chief subeditor: Birgitte van den Muyzenberg

Subeditor: Ian Nicholson

Art director: Carola Pilarz

Illustrations: Claes-Göran Andersson

Telefonaktiebolaget LM EricssonSE-164 83 Stockholm, SwedenPhone: + 46 10 719 0000Fax: +46 8 522 915 99

284 23 -3215 | Uen ISSN 0014-0171

© Ericsson AB 2013