Optical Fiber Technology 17 (2011) 412–420

Contents lists available at ScienceDirect

Optical Fiber Technology

www.elsevier .com/locate /yof te

Invited Papers

Optical integration and multi-carrier solutions for 100G and beyond

Matthew Mitchell a,⇑, John McNicol b, Vinayak Dangui a, Han Sun b, Kuang-Tsan Wu b, Zhong Pan a,Michael Van Leeuwen c, Jeffrey Rahn a, Stephen Grubb c, Radhakrishnan Nagarajan a, David Welch a

a Infinera Corporation, Sunnyvale, CA, USAb Infinera Corporation, Ottawa, Ontario, Canadac Infinera Corporation, Annapolis Junction, MD, USA

a r t i c l e i n f o a b s t r a c t

Article history:Available online 17 August 2011

Keywords:100 GbEFiber-optic transmission technologyCoherent optical transmissionPhotonic Integrated Circuit

1068-5200/$ - see front matter � 2011 Elsevier Inc. Adoi:10.1016/j.yofte.2011.07.005

⇑ Corresponding author. Address: 169 Java Drive, SE-mail address: mmitchell@infinera.com (M. Mitch

In this paper, we will review techniques to achieve 100 Gb/s and higher optical channels; review theimpact upon system reach; review the benefits of using multiple carriers; and discuss how optical inte-gration is a key enabling technology for future low cost high capacity systems.

� 2011 Elsevier Inc. All rights reserved.

1. Introduction

Optical transmission plays an integral role in all modern com-munication systems, particularly in the long haul back-bone andmetro networks. The need for ever increasing bandwidth requiresboth a continual reduction in the cost per bit as well as an increas-ing single fiber capacity. Transmission systems are currently in themidst of a 20� capacity increase with the transition from on–offkeyed (OOK) to coherent based polarization multiplexed quadra-ture phase-shift keying (PM-QPSK) designs. This will allow for low-er system cost, higher spectral efficiency, longer reach, andreduced cost and complexity in the optical line system. In this pa-per, we will review how the industry has achieved the 100 Gb/smilestone and the means to achieve further increases in reachand capacity; review the benefits of using multiple carriers; anddiscuss how optical integration is a key enabling technology to al-low for low cost high capacity systems.

2. Road to 100 Gb/s PM-QPSK

The development of high spectral efficiency in optical systemshas traditionally lagged behind other communication methodsdue to its ultra-high speed, unique fiber transmission characteris-tics, and the lack of availability of enabling technologies. Fig. 1illustrates the communications evolution ranging from voicebandmodems, microwave radio, wireless (cellular radio) and opticalcommunication systems. In 2000, most commercial optical deploy-

ll rights reserved.

unnyvale, CA 94089, USA.ell).

ments were at a spectral efficiency of 0.1–0.2 b/s/Hz (i.e., 10 Gb/sin 100 or 50 GHz channel bandwidth) and the modulation formatwas on–off keying (OOK), also known as intensity modulation di-rect detection (IMDD).

The first move towards higher spectral efficiencies was theintroduction of 40 Gb/s technologies in the 2000–2008 time frame.These transceiver technologies were typically based on opticalduo-binary, differential phase-shift keying (DPSK), and differen-tially-detected QPSK (DQPSK) transmission formats. The lack of astandardized modulation format and the difficulty of impairmentmitigation hindered the industry and the successful achievementof higher spectral efficiency transmission systems. The first deploy-ments of 40 Gb/s transponders highlighted the difficulty of robustcorrection of transmission impairments in real-world optical net-works. While chromatic dispersion penalties, which increase asthe square of the bit rate, can be corrected with optical techniquesin a straightforward fashion, albeit at an increased cost, PMDproved to be difficult to reliably compensate for and proved to bethe ‘‘Achilles heel’’ of higher bit rate, non-coherent transmissionformats.

The key to moving to higher spectral efficiencies while main-taining reach and tolerance to chromatic dispersion and PMDwas the use of high speed electronics. The first demonstrationmaking use of powerful digital signal processing in optical commu-nications was in 2005 [1] with the introduction of a dispersionpre-compensation transmitter. This was used within a 10G IMDDsystem as a means to remove optical dispersion compensation inthe line amplifiers and had the ability to compensate up to±50,000 ps/nm of dispersion. Although this approach solved thedispersion tolerance issue for non-broadcast designs, it did notimprove upon PMD tolerance.

1975 1980 1985 1990 1995 2000 2005 2010 201510

-1

100

101

year of commercial introduction

spec

tral

effi

cien

cy [b

/s/H

z]

voiceband modemsmicrowave radiowireless (per site)optical fiber -2 -1 0 1 2

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

Fig. 1. Communications evolution.

Fig. 2. Coherent optical receiver.

M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420 413

The next major advance was the introduction of coherent 40GDP-QPSK in 50G channel bandwidth (0.8 b/s/Hz) in 2008 [2]achieving a spectral efficiency of 0.8 b/s/Hz. The enabling technol-ogy of this design was the use of a coherent receiver, as shown inFig. 2. This receiver makes use of an optical local oscillator, highspeed analog to digital converters (A/D), and digital signal process-ing (DSP) implemented within a dedicated ASIC. Since that time,there has been an explosion of research and development on thecoherent optical modem with the state of the art optical transportsystems now operating at 2 b/s/Hz, e.g., 100 Gb/s in 50 GHz chan-nel spacing.

It has been possible to maintain acceptable optical reach andin some metrics, such as PMD and chromatic dispersiontolerance, improved performance during the recent 20� increasein capacity. This was achieved during the first quadrupling ofcapacity from 10 Gb/s IMDD to 40 Gb/s primarily by the followingmeans:

� The introduction of QPSK, allowing for the transmission of 2 bitsper symbol, versus the single bit per symbol in IMDD.� The introduction of polarization multiplexing (PM), allowing for

2� capacity increase over single polarization IMDD.� The addition of coherent detection, providing 3 dB better OSNR

performance relative to IMDD.

� The combined use of coherent detection, a linear receiver, fastADC, and digital signal processing, allowing for digital compen-sation of chromatic dispersion and PMD penalties.

The evolution from 40 Gb/s to 100 Gb/s required further tech-nology advances to achieve a capacity increase of 2.5�while main-taining similar reach capability. The ability of coherent DSPtechnology to compensate for PMD has been highlighted in recent100G field trials where 100 Gb/s transponders were able to suc-cessfully close extremely high PMD links in which 10G NRZ tran-sponders were not [3]. The use of single wavelength 100 Gb/sPM-QPSK requires a 2.5� increase in baud rate relative to singlewavelength 40 Gb/s, which incurs a 4 dB penalty in OSNR toler-ance. An enabling technology to gain performance is higher per-forming forward error correction (FEC) with soft decoding. Beforesoft FEC was introduced, the most widely used FEC was harddecoding with �7% overhead (OH). By using soft FEC with 15–20% OH, it is possible to gain 2 dB of performance relative to hardFEC. This results in a higher BER threshold of �2% and can allow fora 1 dB higher launch power and 1 dB reduction in other transmis-sion penalties, such as that arising from polarization dependentloss. Some 100 Gb/s design scenarios may require further meansto improve reach to an acceptable distance such as improved lineamplifiers or enhanced DSP techniques.

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

5500

6000

QPSK 8QAM 16QAM

Rea

ch [k

m]

Ideal Reach, EDFA-Only

Ideal Reach, EDFA+Raman

Realistic Reach, EDFA-Only

Realistic Reach, EDFA+Raman

Fig. 4. Reach performance versus modulation formats.

1

1.2

QPSKSP-BPSK

414 M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420

3. Next steps beyond 2 bit/s/Hz with PM-QPSK

3.1. Higher order modulation

A straightforward means of achieving higher system capacity isthe use of higher order modulation formats, such as multi-levelPSK and Quadrature Amplitude Modulation (QAM) [4]. These for-mats improve the spectral efficiency by increasing the number ofbits carried per symbol. As Fig. 3 shows, this benefit comes at a costof requiring larger signal to noise ratios. As an example, using 8-QAM or 16-QAM can increase the spectral efficiency over QPSKby a factor of 1.5 or 2, respectively, but requires increased SNRby approximately 4 dB or 6.7 dB (assuming a required BER thresh-old of 2.3% and ideal conditions). The use of higher order QAMmodulations thus results in reduced reach due to the degradedOSNR sensitivity and degraded nonlinear tolerance resulting fromdecreased phase margins. Accordingly, there is a trade-off betweencapacity and reach when using such higher order modulationformats.

When comparing the reach performance of different modula-tion formats, many parameters must be considered within thetransmitter and receiver design and the optical amplification sys-tem. Examples with respect to transmitter and receiver designare implementation penalties, such as finite extinction ratio inthe Mach–Zehnder modulators, sub-ADC jitter, and transmitternonlinearity. These various impairments tend to exert a larger pen-alty when the OSNR requirements are higher, as is the case forhigher order modulation formats. Consequently, realistic OSNRpenalties when moving to higher order modulation formats are lar-ger than under ideal conditions, barring improvements to thetransceiver used for the higher order modulation. For the line sys-tem, parameters affecting the reach include amplifier noise figureand gain flatness, the use of Raman technology, spectral flatteningcapability, required system OSNR margin, and launch power withthe resulting nonlinear penalties.

In order to compare the reach of various modulation formats,we will present a nearly ideal scenario along with an example ofa more realistic scenario. The following key assumptions apply:

1. 160 Channels at a baud rate of 16 GBaud and channel spac-ing of 17.6 GHz. This represents a channel fill factor of 91%,defined as the ratio of the baud rate to channel spacing.

2. Uncompensated 100 km 22 dB loss spans of SMF fiber.3. FEC threshold set at a BER of 2.3% (Soft decision FEC).4. Differentially encoded formats.5. The use of star constellations for 8 and 16 QAM.6. EDFA noise figure of 6 dB, a Raman gain up to 10 dB when

used.

1.0E-04

1.0E-03

1.0E-02

1.0E-01

0 5 10 15 20 25

Bit

Erro

r Rat

e

SNR (dB) (in symbol-rate BW)

QPSK

8QAM (two QPSK)

16QAM (two 8PSK)

32QAM

64QAM

Fig. 3. BER versus SNR for various modulation formats.

7. Launch powers optimized to provide for maximum reach.8. 2 dB of system OSNR margin.9. An EDFA ripple of 0.8 dB random and 0.6 dB linear for the

realistic scenario, versus flat performance for the ideal.10. For the realistic scenario, a 2 dB implementation penalty for

the transmitter and receiver pair relative to ideal.

Fig. 4 shows the resulting reach comparison for each scenariomaking use of PM QPSK, PM 8 QAM, and PM 16 QAM formats.The ideal scenario with an EDFA only line system shows a reachof 3900 km, 1600 km, and 700 km for QPSK, 8 QAM, and 16 QAMrespectively. The realistic scenario shows the reach degrading to1900 km, 800 km, and 400 km for the three formats. IntroducingRaman amplification can further improve the reach, allowing therealistic scenario to achieve 5200, 2200, and 1000 km for the threeformats and even greater ideal performance. Realistic designscould make use of additional techniques to enhance the perfor-mance of the 8 and 16 QAM scenarios.

The degradation of reach due to increased order of modulationleads to a capacity-reach product versus modulation format. Fig. 5shows this dependence using system parameters similar to therealistic EDFA scenario described above. The capacity-reach prod-uct decreases by approximately 38%, 60% and 75% for 8, 16, and32 QAM, respectively, when compared to QPSK.

0

0.2

0.4

0.6

0.8

0 2 4 6 8 10 12 14

Cap

acity

* R

each

Pro

duct

bits / dual-pol symbol

BPSK

8-QAM

IM-DD

16-QAM

32-QAM

64-QAM

Fig. 5. Reach-capacity product versus modulation formats.

Fig. 6. Illustration of spectral efficiency options. Left figure represents a typical optical system with guard bands (rectangles) between channels for optical add/drop. The rightfigure represents an example configuration with the guard bands removed to allow for increased capacity within a larger band.

M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420 415

Accordingly, the use of higher order modulation increasescapacity but, at the same time, results in an undesirable reductionin reach. Higher cost amplification systems can partially reclaimthe reach loss, but, in order to fully maintain the reach, new tech-nologies are inevitably required.

3.2. Increased utilization efficiency of optical bandwidth

An additional means to increase capacity relative to existingsystems is to achieve a more efficient use of the available opticalbandwidth [5]. Typical optical designs make use of guard bands be-tween wavelengths to enable the adding, dropping, and routing ofeach channel. Systems can be designed to allow for the removal ofsuch guard bands, and use this spectrum to transmit increasedcapacity as illustrated in Fig. 6. As shown in the figure, the use ofa larger band of channels or ‘super channel’ creates a higher capac-ity channel of wider width.

The capacity increase is determined by the amount of guardband removal and the modulation format as shown in Fig. 7. Asan example, within a 4 THz optical band the total capacity ofPM-QPSK modulated channels can be increased from 8 Tb/s with64% fill to upwards of 10 Tb/s via the use of an 80% channel fill.

3.3. Single versus multi-carrier channels

A channel is defined as an amount of optical data that can be ad-dressed (add or dropped) within a system. One can chose to createa fixed capacity channel by making use a single carrier or multiplelower data rate carriers (super channel) as illustrated in Fig. 8. Theselection of the number of carriers is determined by design trade-offs between optical reach performance, cost, power consumption,system density, and time to market. The path of low carrier count(higher data rate) is challenged by the availability and cost of highspeed optoelectronics whereas the path of high carrier count (low-er data rates) is constrained by the cost and size of multiplecomponents.

4

6

8

10

12

14

16

18

20

22

PM-QPSK PM-8QAM PM-16QAMTota

l Cap

acity

[Tb/

s], o

ver a

4 T

Hz

spec

trum

64% Channel Fill

80% Channel Fill

Fig. 7. Total capacity of 4 THz C band versus format and fill ratio.

One of the inputs to determining the reach performance is thetolerance to nonlinear penalties exhibited by the optical channel.The variation of this penalty versus carrier count depends upondeployment details such as the dispersion map and the mix ofchannels within the transmission fiber. As coherent implementa-tions become the dominant transmission technology carriers willmake use of uncompensated fiber spans due to cost savings andadditional reach enhancements this provides. In Fig. 9 [6], we showthe variation of the nonlinear threshold (NLT) defined as the launchpower at which the infinitesimal improvement in Rx OSNR is bal-anced by the infinitesimal increase in nonlinear penalties, versuschannel baud rate for PM-QPSK, PM-8 QAM and PM-16 QAM chan-nels. The calculation assumes square root raised-cosine filteringwith a roll-off factor of 0.1, a carrier spacing equal to 1.1 timesthe baud rate, transmitted over 20 spans of uncompensatedG.652. The NLT is plotted in units of dBm/THz, so as to removeany baud rate dependency. A flat NLT would indicate that the sys-tem reach is independent of the per-carrier baud rate. We can ob-serve that lower baud rates exhibit a slightly better NLT and thus ahigher reach potential for the three modulation formats of QPSK, 8QAM, and 16 QAM.

A significant factor in the design choice of single carrier versusmulti-carrier can be differentiated by the difficulties encounteredduring implementation such as the availability of high speed A/Dconverters, RF components, complexities and cost in high speedpackaging technologies, and heat dissipation for the DSP ASIC. Suit-able A/D converters in CMOS technology are available only recentlysuch as a 6 bit 24 GS/s device in 90 nm CMOS introduced in 2008[7], and in 2010 a 56 GS/s device in 65 nm CMOS [8] targeted forsingle channel 100 Gb/s systems using 7% overhead. A 100 Gb/ssystem using higher overhead soft FEC may require A/D samplingrates at approximately 64 GS/s. Such a device is not yet commer-cially available, but is expected by the end of 2011.

When moving beyond 100 Gb/s channels the use of multiplecarriers has significant advantages as shown when consideringthe creation of 400 Gb/s channels. Making use of a single wave-length would require CMOS ADCs running at 256 GS/s or 128 GS/s, assuming T/2 sampling, for QPSK or 16 QAM respectively. Theseare beyond existing technology and will require at least one CMOSnode shrink to be achievable. Making use of two carriers reducesthese rates to 128 GS/s and 64 GS/s for QPSK or 16 QAM respec-tively. Thus, two carriers operating in 16 QAM could enable400 Gb/s channels making use of technology that will be availablein the near term.

Fig. 8. Illustration of equivalent capacity channel designs using one to four carrierswithin a fixed optical bandwidth.

Fig. 9. Nonlinear threshold versus carrier baud rate and modulation format.

416 M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420

The use of multiple carriers helps with the heat and thermalchallenges inherent in DSP design. Power reduction within an ASICis important as it can become the most heat challenged componentwithin a line card. Due to the sheer volume of data that needs to beprocessed inside the DSP ASIC, the ASIC heat dissipation is a prob-lem that must be handled from the beginning of algorithm designthrough to layout. The high date rates will continually force theDSP ASIC into the latest CMOS processing technology in order tolimit its power consumption. A rough size estimate of a singlechannel DSP ASIC with integrated soft FEC at 100 Gb/s would likelybe in excess of 50 million gates, with soft-FEC accounting for half ofit. While ADC sample rates of the above systems are in the 30–60 GS/s range, power efficient realization in CMOS necessitatesthe parallel processing of approximately one hundred samples atsystem clock rates of hundreds of MHz. Some processing stagessuch as chromatic dispersion compensation are most efficientlyhandled in the frequency domain using fast Fourier transforms.Accordingly, for a given capacity in multi-carrier optical modem,the choice of modulation bandwidth and the number of sub-carri-ers has a modest influence on DSP power consumption. It is esti-mated that DSP power decreases some 5–10% for each factor oftwo reduction in the modulation bandwidth (baud rate) and in-crease in the number of sub-carriers.

Assuming one A/D at 64 GS/s has the same heat as two A/Ds.each at half the rate, and also assuming feed-forward carrier recov-ery are equalized between the two cases, the increase in ASIC com-plexity for single channel compared to two channel case isestimated to be between 20% and 30%. In the limit where manychannels are used to implement 100 Gb/s the baud rate is so lowthat no inter-symbol interference (ISI) compensation is needed,and all the DSP functions can be reduced to a 1 tap polarization

1x 3

2Gba

ud

2x 1

6Gba

ud

4x 8

Gba

ud

8 x 4

Gba

ud

Increasing Carriers

Incr

easi

ng D

SP C

ompl

exity

0

1

0.5

Fig. 10. DSP complexity comparison versus carrier count.

demux filter and a carrier recovery circuit. Fig. 10 is an estimateof the DSP complexity savings in reducing the baud rate from32 Gbaud to 4 Gbaud.

3.4. Benefits of optical integration

The economics of optical transport demand continual reduc-tions in the cost per Gb/s. The use of large scale optical integrationis a powerful means to this end by reducing cost, packaging com-plexity, size, and power consumption within an optical system. Asdiscussed previously, the use of multiple carriers will be requiredto allow for single channel and total capacity advances to beachieved. While the use of digital signal processing and coherentreceivers allows for significant increases in spectral efficiency, itcomes at the expense of increased complexity within the opticalmodem. The number of optical components increases nearly 4�relative to IMDD systems due to the need for dual polarizationnested Mach–Zehnder modulators, dual lasers, receiver optical hy-brids and photodiode arrays. Secondly, as we have shown, increas-ing the total system capacity can result in a capacity versus reachtradeoff. If a system operator must have increased capacity on aroute, there is a risk that additional regeneration will be requiredwhich leads to significant costs. Both of these effects require theability to provide low cost OEO functionality at high systemdensities.

Optical transport systems based on Photonic Integrated Circuitshave been shipping since 2005 and have greatly influenced thearchitecture of WDM systems [9]. The reliability of Photonic Inte-grated Circuits (PICs) have also been proven, with greater than400 million operational field hours of 100 Gb/s PICs achieved with-out failure. The Infinera 100 Gb/s PIC design consisted of ten 10 Gb/s channels monolithically integrated into two photonic circuits,and with multiple PIC line cards, supporting up to 1.6 Tb/s totalcapacity within a 4 THz C band. To support higher system capaci-ties Infinera has developed a next generation PIC providing fivechannels of 100 Gb/s coherent PM-QPSK for a 5� increase in perPIC and system capacity [10]. Fig 11 shows the schematic diagramof such an InP 500 Gb/s PM-QPSK transmitter PIC. Each channelconsists of a DFB laser with a backside power monitor and a TE/TM splitter sending light to a nested pair of Mach–Zehnder modu-lators. The channels from each polarization are then combined viatwo separate AWGs and exit the module for external combining.An InP 500 Gb/s PM-QPSK receiver PIC architecture is shown inFig. 12 [11]. Two separate inputs, each containing 10 carriers sendTE/TM light through an AWG. The different carrier outputs are thencombined separately with a unique local oscillator via a 90� optical

Fig. 11. Schematic diagram of a 10 channel PM-QPSK transmitter PIC.

Fig. 12. Coherent receiver PIC architecture for dual polarization QPSK modulation format.

M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420 417

hybrid, and then detected using balanced photodetector pairs. Fig13 shows measured constellation data of a 500 Gb/s PIC pair (Txinto Rx) operating at 14.25 Gbaud.

Infinera recently completed a field demonstration of the nextgeneration 500 Gb/s PIC pair [12] showing the reach capability ofthe integrated PIC and the ability to support both 10G and 100Gchannels simultaneously upon the same fiber.

Fig. 13. Measured constellation diagram from Infinera 5 � 100 Gb/s coherent PM-QPSK integrated photonic circuit.

3.5. Reach and capacity optimized network designs

As the industry advances to higher spectral efficiencies beyond2 bits/s/Hz the challenge of the capacity reach tradeoff will requireintelligent planning of the optical network. The modulation formatused for each traffic demand can be optimized based upon the re-quired reach as illustrated in Fig. 14. For example, a long route maybe required to use lower spectral efficiency channels such as QPSKwhereas a shorter route could be closed while making use of ahigher efficiency format such as 16 QAM. This spectral efficiencyoptimization process would allow for highly efficient networksbut will add operational complexity due to the need to plan andmanage multiple channels of potentially different optical band-

widths. To support this capability, optical modems should havethe ability to change the modulation formats via software in areconfigurable manner, in a fashion similar to software-definedradios used in wireless communications.

The optical line system must likewise support programmableoptical filtering and routing capability due to the use of multiple

Fig. 15. Illustration of 100 Gb/s channels and 1 Tb/s super channels with differing modulation formats and optical bandwidths.

Super-Channel Routing

Fig. 14. Illustration of network design where the modulation format used for a circuit is dictated by the reach requirement. The longer route makes use of QPSK while theshorter route is 16 QAM.

418 M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420

modulation formats within the same fiber. The optimum opticalbandwidth for higher order modulated channels and super chan-nels may vary as illustrated in Fig. 15. The system operator willneed the ability to assign different modulation formats, and result-ing optical channel widths, to spectral regions within the fiber andmultiplex, de-multiplex, and route the signals.

4. Paths to capacity and reach improvements

4.1. Limits of FEC improvement

Although forward error correction is certainly one of the meth-ods to improve the performance of future optical transport systemsit is important to consider how much can be gained in practice.Fig. 16 calculates the Shannon limits [13] for both hard decoding(HD) and soft decoding (SD) for BPSK/QPSK at an output BER of10�15, expressed as coding gain versus FEC overhead (OH). The

9

10

11

12

13

14

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6

Cod

ing

Gai

n (d

B)

FEC OH (n/k-1)

Soft Decoding Shannon LimitHard Decoding Shannon LimitSD implementationHD implementation

Fig. 16. Coding gain (dB) at a BER = 1e�15 versus FEC technology and overhead.

OH is defined as n/k � 1, where n is the code word length in bitsand k is the number of the information bits. The SD limit is ob-served to be better than the HD limit by 1.2–1.5 dB in the OH rangeof 10–35%. The realizable coding gain achievable in practice is alsoillustrated within the figure. As an example, it is shown that a prac-tical coding gain of �9.2 dB occurs with a 7% HD FEC which repre-sents the typical performance in existing optical transport systemsbefore soft FEC is introduced. When SD FEC is implemented, typicalcoding gains are expected to be approximately 11 dB for 15% and11.3 dB for 20% OH, respectively, a value expected for productcodes and LDPC. Although more coding gain can be obtained withmore iterations with SD, heat and latency will introduce con-straints limiting the practical number of iterations possible. It isinteresting to observe from the figure that the performance ofpractical SD implementation approaches the HD Shannon limit,but still remains more than 1 dB away from the SD Shannon limit.It is also noted that using more OH has a diminishing return in cod-ing gain in practice. For example, going from 25% to 50%, the Shan-non HD limit shows a gain of only �1 dB. More OH means higherbandwidth is required for optical and electrical components andthus a design tradeoff has to be evaluated against the impairmentassociated with the increase of bandwidth expansion versus thecoding gain. In summary, after applying SD FEC with reasonableOH, further gain is very modest in using FEC to improve theperformance.

4.2. Future digital compensation techniques

It is well known that a coherent receiver is very effective incombating linear distortions such as chromatic dispersion (CD),PMD, and linear filtering impairments. However, in fiber transmis-sion, the nonlinear propagation is one of the major sources of pen-alty to the system performance. One way to reduce the nonlinearpenalty is to reduce the launch power, but this reduces the reachdue to a corresponding OSNR reduction. After employing soft FEC

M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420 419

to improve the reach, one may consider the addition of digital non-linear penalty compensation (NLC). NLC will prove to be challeng-ing due to the requirement of massive signal processing, butpresents a logical next step to gain performance. For example, amost elegant way for NLC is via backward propagation in the Rx[14,15], in which one has to implement the split-step Fouriermethod (SSFM) for each span using some reasonable step sizesfor SSFM. This however requires significant signal processingpower, which may be still beyond the capability of a CMOS ASICin the foreseeable future. Note that back propagation also requiresthe knowledge of the launch power profile at the receiver, or atleast some methods of self-discovering it at the receiver, and thatit can compensate for only the nonlinearity generated in the fiberfrom the channel that is received (the effect is known as self-phasemodulation, SPM). Some simplifications will be needed, and vari-ous research groups are exploring alternatives. As an example, anonlinear polarization crosstalk canceller is proposed in [16] thatis simple enough to be implemented and has significant benefitin dispersion-managed links. The SPM in a 3200 km un-compen-sated SMF link can be compensated using only four steps in theSSFM [17], the complexity reduction comes from low pass filteringof the intensity signal in the SSFM calculation. MAP (maximum apriori probability) equalizers that compensate for nonlinear ISIhave gained attention and have been demonstrated to have signif-icant benefit in both linear and nonlinear channel impairments[18]. A review of the research in this area can be found in [14,19].

4.3. New fiber technology

While improvements to the electronic processing represent onepath to enable increased capacities, another alternative lies in thedirection of better fiber parameters. These benefits can include: (1)reduction in the loss per length metric (2) reduction in nonlinearpenalties due to large effective area fibers and (3) the use of highlydispersive fibers. Ideally these improvements can be combined toenable significant reach benefits.

The realization of lower loss transmission is now possiblethrough the use of new fibers such as Ultra Low Loss Fiber (ULLF)[20]. Typical loss numbers of 0.17 dB/km are possible, as opposedto a typical value of 0.20 dB/km. This allows for a 3 dB loss reduc-tion for 100 km spans resulting in a net �15% reach improvementdepending upon the link design as compared to G.652.

The reduction of nonlinear penalties is possible through the useof another new fiber design as is the case in Ultra Large Area Fiber(ULAF), or in a more radical fashion, by using air-core photonic-bandgap fibers, although air-core fibers’ propagation losses are stillfar from the standards required for long-haul transmission. Theeffective area of ULAF is 112–150 lm2 as opposed to 80 lm2 instandard G.652 fiber allowing for higher launch powers into thetransmission fiber for a fixed nonlinear penalty. This is a directwin for spans making use of EDFA amplification as long as theamplifiers have sufficient output power available. The benefit isdiminished significantly when using distributed Raman amplifica-tion due to the reduced Raman gain that occurs due to the largercore area. The use of higher power pumps would be needed toovercome this reduction leading to added costs to the amplifica-tion system. Multi-order Raman pumping is another techniquethat has been shown to effectively reduce the line system noise fig-ure, albeit at an increased cost and complexity to the line system.

The last method of improving transmission performance via fi-ber advances would be to make use of higher dispersion fibers. Ithas been shown [21] that the higher local dispersion further re-duces the nonlinear penalties due to a larger walk-off betweenneighboring channels. This increased channel walk-off in turnleads to a reduction in the strength of XPM-generated phase noise,

thus improving nonlinear tolerance, and resulting in better systemreach.

5. Summary and conclusions

Over the past decade, the spectral efficiency of optical transmis-sion systems has increased some 20-fold while system reach haslargely been maintained and tolerance to the fiber linear effects en-hanced. The enabling technologies include coherent transmission,forward error correction which approaches the Shannon limit,and CMOS-based digital signal processing. Further spectral effi-ciency gains are anticipated by reducing the gaps between carrierswhile assembling optically-routable blocks of carriers (super-channels) and through the application of higher order modulationformats resulting in an improvement by a factor of 1.25� and 1.5–2� respectively. For a given amplified line configuration, increasedspectral efficiency will involve a trade-off with system reach.

An important factor in the design of multi-carrier optical mod-ems – such as those having capacities of 0.5 or 1 Tb/s – is the selec-tion of the number of sub-carriers and the modulation bandwidthper sub-carrier. Studies of pulse propagation given fiber’s disper-sive and non-linear characteristics show some performance edgefor systems with lower baud rates and more carriers. Accordinglycost-effective super-channels are anticipated to be comprised ofmultiple optical sub-carriers. Photonic integration is critical tothe practical, cost effective implementation of higher spectrallyefficient transmission systems. PIC technology has already madea significant impact in 10 Gb/s WDM networks over the past5 years, and is expected to play an ever increasing role in the futuredue to the need for flexible format based line cards making use ofmultiple carrier super channel transmission formats.

References

[1] J. McNicol, M. O’Sullivan, K. Roberts, A. Comeau, D. McGhan, L. Strawczynski,Electrical domain compensation of optical dispersion (optical fibrecommunication applications), OFC, 2005.

[2] H. Sun, K.-T. Wu, K. Roberts, Real-time measurements of a 40 Gb/s coherentsystem, Optics Express 16 (2008) 873–879.

[3] T.J. Xia, G. Wellbrock, M. Pollock, W. Lee, D. Peterson, D. Doucet, J. Sitch, K.Ghazian, P. Bryan, P. Rochon, 92-Gb/s Field Trial with Ultra-High PMDTolerance of 107-ps DGD, OFC/NFOEC 2009, NThB3.

[4] W. Winzer, R.J. Essiambre, Advanced optical modulation formats, Proc. IEEE 94(5) (2006) 952–985.

[5] S. Chandrasekhar, X. Liu, Terabit superchannels for high spectral efficiencytransmission, ECOC (2010) Tu.3.C.5.

[6] J. McNicol, V. Dangui, H. Sun, D.J. Krause, K.T. Wu, M.L. Mitchell, D.F., Welch,Single-Carrier Versus Sub-Carrier Bandwidth Considerations for CoherentOptical Systems, Invited Paper, SPIE, 2010.

[7] P. Schvan, et al., A 24 GS/s 6b ADC in 90 nm CMOS, in: Proc. Digest Tech.Papers. IEEE Int. Solid-State Circuits Conf., February 2008 (ISSCC), pp. 544–634.

[8] I. Dedic, 56 GS/s ADC: Enabling 100GbE, OFC/NFOEC, 2010, p. OThT6.[9] F. Kish, D. Welch, R. Nagarajan, J. Pleumeekers, V. Lal, M. Ziari, A. Nilsson, M.

Kato, S. Murthy, P. Evans, S. Corzine, M. Mitchell, P. Samra, M. Missey, S.Demars, R. Schneider, M. Reffle, T. Butrie, J. Rahn, M. VanLeeuwen J. Stewart, D.Lambert, R. Muthiah, H.S. Tsai, J. Bostak, A. Dentai, K.T. Wu, H. Sun, D. Pavinski,J. Zhang, J. Tang, J. McNicol, M. Kuntz, V. Dominic, B. Taylor, R. Salvatore, M.Fisher, A. Spannagel, E. Strzelecka, P. Studenkov, M. Raburn, W. Williams, D.Christini, J. Thomson, S. Agashe, R. Malendevich, G. Goldfarb, S. Melle, C. Joyner,M. Kaufman, S.G. Grubb, Current status of large-scale InP photonic integratedcircuits (invited), JSTQE, in press.

[10] S. Corzine, P. Evans, M. Fisher, J. Gheorma, M. Kato, V. Dominic, P. Samra, A.Nilsson, J. Rahn, I. Lyubomirsky, A. Dentai, P. Studenkov, M. Missey, D. Lambert,A. Spannagel, R. Muthiah, R. Salvatore, S. Murthy, E. Strzelecka, J. Pleumeekers,A. Chen, R. Schneider Jr., R. Nagarajan, M. Ziari, J. Stewart, C. Joyner, F. Kish, D.Welch, Large-scale InP transmitter PICs for PM-DQPSK fiber transmissionsystems, IEEE Photon. Technol. Lett. 22 (14) (2010) 1015–1017.

[11] R. Nagarajan, D. Lambert, M. Kato, V. Lal, G. Goldfarb, J. Rahn, M. Kuntz, J.Pleumeekers, A. Dentai, H. Tsai, R. Malendevich, M. Missey, K. Wu, H. Sun, J.McNicol, J. Tang, J. Zhang, T. Butrie, A. Nilsson, M. Reffle, F. Kish, D. Welch, 10Channel, 100 Gbit/s per Channel, Dual Polarization, Coherent QPSK, MonolithicInP Receiver Photonic Integrated Circuit, OFC/NFOEC 2011, Paper OML7, LosAngeles, USA, March 2011.

[12] http://www.infinera.com/j7/servlet/NewsItem?newsItemID=272.

420 M. Mitchell et al. / Optical Fiber Technology 17 (2011) 412–420

[13] Proakis, Digital Communications, 5th ed. or any previous edition, McGraw Hill,2008.

[14] E. Ip, J.M. Kahn, Compensation of dispersion and nonlinear impairments usingdigital backpropagation, J. Lightwave. Technol. 26 (20) (2008) 3416–3425.

[15] E. Ip, J. Kahn, Fiber impairment compensation using coherent detection anddigital signal processing, J. Lightwave Technol. (2010).

[16] L. Li et al., Nonlinear Polarization Crosstalk Canceller for Dual-PolarizationDigital Coherent Receivers, OFC 2010, OWE3.

[17] L. Du, A. Lowery, Improved single channel backpropagation for intra-channelfiber nonlinearity compensation in long-haul optical communication systems,Optics Express 18 (16) (2010).

[18] Y. Cai, MAP detection for linear and nonlinear ISI mitigation in long-haulcoherent detection systems, in: IEEE Photonics Society Summer TopicalMeeting, July 2010, pp. 42–43.

[19] T. Hoshida et al., Recent Progress on Nonlinear Compensation Technique inDigital Coherent Receiver, OFC/NFOEC 2010, OTuE5.

[20] http://www.corning.com/WorkArea/showcontent.aspx?id=14357.[21] A. Carena, G. Bosco, V. Curri, Coherent polarization-multiplexed formats:

receiver requirements and mitigation of fiber non-linear effects, ECOC (2001)Mo2C1.