You are on page 1of 5

100 GIGABIT ETHERNET

BEYOND 100G ETHERNET


PETER J. WINZER, BELL LABS, ALCATEL-LUCENT
ABSTRACT
With 100G technologies standardized, in the context of both Ethernet and the optical transport network (OTN), 100G router ports and 100G optical transport interfaces are commercially available. At the same time, heavily data-centric users are starting to ask for higher-rate interfaces. First speculations include 400G as well as 1T as the next possible Ethernet standards. In this article we discuss interface technology options for Ethernet and OTN beyond 100G in light of the current 100G standards, taking into account likely evolution paths of interface technologies over the next 10 years.

100G LAN TECHNOLOGIES


With the current Ethernet standard, IEEE 802.3 has departed from its so far strictly observed 10 bit rate scaling by incorporating 40G Ethernet into the 100G standard. This was done to address the near-term needs of the server and storage community for low-cost Ethernet interfaces in excess of 10G. This move in standardization also reflects the increasing technological gap between desired Ethernet rates and cutting-edge serial transmission bit rates, as illustrated in Fig. 1. The first 100-Gb/s serial research experiments [1, 7] were published not even a year before the pre-standardization efforts of the IEEE Higher-Speed Study Group (HSSG) started, and the first 100-Gb/s integrated receivers [8, 9] and electro-absorption modulated lasers (EMLs) [10] were reported when 100G standardization was already well underway. In comparison, research using 10-Gb/s serial bit rates had been performed more than 10 years prior to the start of 10G Ethernet standardization. Up until ~2006, Ethernet standards and router interface rates fell short of the serial bit rates developed for circuitswitched optical transport networks, and Ethernet could leverage reasonably mature synchronous optical network/ synchronous digital hierarchy (SONET/SDH)/OTN interface technologies. This is evident from Fig. 1, where Ethernet standards are shown by grey circles, and shaded areas denote standardization periods; black and white circles represent router and SONET/SDH/OTN interfaces, respectively. The fact that Ethernet has caught up with the highest serial OTN bit rates not only reflects itself in the 802.3 standard now breaking with its traditional 10 scaling in favor of the 4 scaling traditionally used by SONET/SDH/OTN, but also in the exclusively parallel transmission approach taken by the current physical layer Ethernet interface specifications. None of the newly standardized Ethernet interfaces uses serial 40-Gb/s or 100-Gb/s transmission technologies. In particular, 100G Ethernet specifies 10 parallel lanes, leveraging mature 10-Gb/s technology, or four parallel wavelengths with quickly maturing 25-Gb/s technology on an 800-GHz ITU-T frequency grid around 1300 nm. Related to these choices of parallel transport is the need for non-integer (10:4) multiplexing (gearbox) and subchannel deskew functionality, both embedded in the 100G standard.

100G IS REALITY, WHATS NEXT?


The year 2010 marks the completion of five years of intense research, development, and standardization since the first demonstrations of 100-Gb/s optical transport were reported in 2005 [1]. IEEE 802.3 has just finished the standardization of local area networking (LAN) interfaces for 100G Ethernet. The International Telecommunication Union Telecommunication Standardization Sector (ITU-T), in close collaboration with IEEE, has augmented its G.709 OTN standard to transport 100G traffic over reliable carrier-grade wide area networks1 (WANs). By now, major vendors have announced 100G products, from chip sets to client interfaces to router ports to optical transport gear. AlcatelLucent, for example, has recently announced commercial availability of the industry's first 100G router interfaces and the first single-wavelength 100G transport interfaces. Major carriers have performed end-to-end field trials using prototype 100G equipment [2, 3]. Significant commercial deployment of 100G technology is expected to take place during the remainder of this year. With 100G now commercial reality, the community has recently started to look into technologies beyond 100G [4]. The need for the next tenfold bit rate increase to 1-Tb/s (1T) Ethernet has already been clearly voiced to keep up with the plethora of data services that drive exponential network traffic growth between 40 and 90 percent per year2 [5], most notably by heavily data-centric users [6]. Target dates between 2015 and 2020 have been named for the desired completion of a 1T Ethernet standard, even though the question of whether the customer base will be big enough to start serious standardization efforts is not yet fully settled [4].

A BRIEF 100G TECHNOLOGY REVIEW


To understand the options in scaling Ethernet beyond 100G, it is important to examine the current status of 100G standards in the light of technology scaling trends for research and commercialization, as visualized in Fig. 1.
The OTN standard specifies network elements for metro/core switching and dense wavelength-division multiplexed (WDM) transport, including features such as survivability, forward error correction (FEC), and operations, administration, and management (OAM) functionalities.
2 Exponential growth rates are conveniently expressed in decibels. For example, 40 percent growth per year corresponds to 10 log10 (1.4) = 1.5 dB/year, and 90 percent corresponds to 2.8 dB/year [5]. These two growth rates are visualized in the upper left corner of Fig. 1. 1

100G WAN TECHNOLOGIES


On the WAN side, ITU-T has not specified a transmission format, apart from a 112-Gb/s line rate and the need to support the standardized ITU wavelength-division multiplexing (WDM) grid. The 50-GHz ITU WDM grid is widely used in todays optically routed mesh networks, with reconfigurable optical add/drop multiplexers (ROADMs) allowing for dynamic provisioning of wavelength services between network nodes. In addition to scaling interface bit rates, WAN systems need to accommodate exponentially rising traffic demands by increasing per-fiber WDM capacities (triangles in Fig. 1). At a fixed optical amplifier bandwidth (typically ~5 THz), this forces WAN transport systems to constantly increase their spectral efficiency compared to the previous transponder generation (i.e., to use higher per-channel bit rates at fixed or decreasing WDM channel spacing). The resulting need for spectral signal compression asks for multilevel modulation

26

IEEE Communications Magazine July 2010

100 GIGABIT ETHERNET

100

Serial interface rates and WDM capacities

Tb/s

2.

10

dB /yr

PDM
dB /yr

/WD

5 1.

M WD
re se a rc

pro

duc

ts

1
DM

100

Se

ter rial in

faces

(rese

arch)

Gb/s

10

Seri
1 1986

al in

a terf

(p ces

rod

ucts

)
ri e nt rfa

ce

s
IEEE ITU-T 2014 2018 2022

u Ro
1994

te

IEEE 2002 2006

IEEE ITU-T 2010

1990

1998

Figure 1. Historic serial bit rate and WDM system capacity scaling in research and products.

and polarization-division multiplexing (PDM). Research over the last several years has concluded that 28-GBaud PDM quadrature phase shift keying (QPSK) with coherent detection [11] is a promising approach for 100G WAN transmission on a 50-GHz WDM grid over ~1500 km of standard fiber including several ROADMs. This format was consequently taken up by the Optical Internetworking Forum (OIF) as the de facto standard for 100G WAN transponders. The first single-carrier 100G PDM-QPSK product has recently been announced by Alcatel-Lucent. Some initial 100G products reused existing 40-Gb/s PDM-QPSK technology by combining two 20-GHz spaced 14.4 GBaud PDM-QPSK optical subcarriers to form an inverse-multiplexed 100G OTN channel [12]. However, in the history of WDM transport, pushing per-wavelength bit rates to the full desired interface rate has generally proven more economical than inverse multiplexing over multiple lower-bit-rate wavelengths.

even 8 50 Gb/s by improving the latter. The 4 100 Gb/s option that would maintain the degree of optical parallelization used by 100G Ethernet seems a stretch today, not only from a component availability perspective but also from the inherently 16 times lower chromatic dispersion tolerance at 100 Gb/s than at 25 Gb/s, and the significantly higher received optical signal power requirements. In short, the degree of optical parallelization is expected to further increase compared to the 100G Ethernet standard, asking for denser photonic integration to make 400G commercially viable.

400G WAN TECHNOLOGY OPTIONS


For WAN applications, the interface line rate likely needs to be upgraded to 448 Gb/s or more, especially if forward error correction (FEC) with an overhead higher than the current 7 percent is used to obtain sufficient transmission reach; overheads of up to 25 percent are currently being discussed [13]. Single-wavelength bit rates as high as 400 Gb/s have not yet been demonstrated in research experiments using purely electronic signal generation and detection, but are clearly imminent. Extrapolating progress in serial bit rate research (green line in Fig. 1), we should expect single-channel research demonstrations of 200 Gb/s (i.e., 400-Gb/s PDM) to be reported very soon. In addition to providing 400-Gb/s interface speeds, 400G WAN transponders must also permit higher spectral efficiencies to allow for increased WDM system capacities. The baseline for 400G WAN technologies is clearly the current generation of 100G PDM-QPSK operating on a 50-GHz WDM grid at a spectral efficiency of 2 b/s/Hz. Ideally, and for maximum compatibility with the existing WAN networking infrastructure, one would want to have a 400G signal occupy a single 50-GHz WDM slot at a spectral efficiency of 8 b/s/Hz. In practice, this would require at least PDM 32-ary quadrature amplitude modulation (32-QAM) at 45 GBaud, and up to PDM 256-QAM at 28 GBaud if non-ideal ROADM characteristics and component frequency offsets are to be supported. Electronically generated orthogonal frequency division multiplexing (OFDM) could also be used

400G OR 1T, THAT IS THE QUESTION


The simultaneous standardization of 40G and 100G in IEEE 802.3 gives reason to speculate that the next Ethernet rate and, with it, the next OTN rate could be 400G, either as a free-standing next step or in combination with a 1T Ethernet standard. In this section we describe possible technology options in light of historic scaling trends (Fig. 1), for both 400Gb/s and 1-Tb/s LAN and WAN interfaces.

400G LAN TECHNOLOGY OPTIONS


Due to the widening gap between serial transmission research and commercialization, discussed above in the context of 100G LAN technologies, 400G LAN interfaces will likely continue the parallel transmission approach that has been exclusively taken by 100G Ethernet. Interfaces for 400-Gb/s LAN applications could be based on, say, 25 Gb/s over 16 parallel lanes/wavelengths [4], but could also use 10 40 Gb/s by exploiting commercially available 40-Gb/s EML technology, or

IEEE Communications Magazine July 2010

27

100 GIGABIT ETHERNET


Electrical spectrum Optical spectrum f Laser (a) Mod. f Laser f Laser (b) Mod. f (c) f Comb Mod. Mod. f Mod. f 10 dB 60 GHz Mod. f f + f

Figure 2. Optical transmitter options.

instead of single-carrier modulation, as shown in Figs. 2a and 2b, but OFDM is typically somewhat less spectrally efficient than the corresponding single-carrier format due to cyclic prefix, pilot, and training symbol overheads [14]. The digital signal processing complexity of the two options is about the same [15]. Considering recently reported single-carrier high-speed QAM demonstrations (cf. signal constellations shown in Fig. 3), near-term commercialization of 32-QAM or 256-QAM at 448 Gb/s seems out of reach, for both single-carrier modulation and electronically processed OFDM. One solution would be to relax the high spectral efficiency requirement by abandoning the rigid 50-GHz WDM grid and allowing, for example, 56 GBaud PDM 16-QAM on a flexible ~70- to ~80-GHz WDM grid at a spectral efficiency between 6 and 5 b/s/Hz, the latter with adequate ROADM support. A preference for flexible solutions of this kind is typically voiced by heavily data-centric users operating their own high-capacity data pipes between their data centers. In contrast, telecommunication carriers with large mesh networks supporting diverse services tend to prefer sticking to the established 50GHz grid. To accommodate the 50-GHz boundary condition, one could inverse multiplex a 448-Gb/s channel onto two 224-Gb/s wavelengths on a 50-GHz grid using, say, 28GBaud PDM 16-QAM for a net spectral efficiency of 4 b/s/Hz [19] and an associated doubling in per-fiber WDM capacity compared to 100G PDM-QPSK. Another solution would be to substitute a single-carrier 448-Gb/s signal by an orthogonal 3 multiplex of lower-speed optical subcarriers modulated at 32-QAM or more. This solution, referred to as coherent WDM or coherent optical OFDM, is qualitatively different from dense WDM inverse multiplexing in that it can preserve the spectral efficiency of single-carrier modulation for a given modulation format [21]. As can be seen from the associated transmitter architecture in Fig. 2c, electrical multiplexing and digital signal processing complexity are traded for electro-optic parallelism. Note that the degree of parallelism can differ between transmitter and receiver. For example, and as shown in the inset to Fig. 2c [22], a 448Gb/s transmitter could use 10 individually modulated orthogonal optical subcarriers, while the receiver could detect them in two groups of five, an approach taken in a recent 448Gb/s demonstration [22]. As discussed above, striving for the highest possible subcarrier bit rates that can still be handled
3

electronically while keeping optical parallelization to a minimum will generally lead to the most practical and likely most cost-effective solution. Apart from the technological difficulties associated with high spectral efficiency 400G WAN transport, the achievable transmission reach is of great concern. Minimizing the number of generally expensive and power-hungry opto-electronic regenerators by increasing the optically transparent reach of a system is key to its economic success. From a modulation point of view, the system reach is primarily determined by the format-specific signal-to-noise ratio (SNR) required at the receiver to guarantee error-free performance. As shown in Fig. 4, Shannons theory dictates a trade-off between spectral efficiency and the required SNR per bit. Squares represent recent experimental results for multi-Gbaud QAM, and circles denote the corresponding theoretical limits, assuming state-ofthe-art 7 percent overhead hard-decision FEC. Note that for low constellation sizes, increasing the constellation results in modestly higher SNR requirements. For example, going from QPSK to 16-QAM doubles the spectral efficiency (and with it the WDM capacity within a given optical amplification band) at an SNR penalty of 3.8 dB. However, doubling the spectral efficiency again by going from 16-QAM to 256-QAM results in a further SNR penalty of 8.8 dB, all practical implementation penalties aside. In addition, increasing the per-channel symbol rate and/or the size of the symbol constellation may affect the optical signal power that can be launched into the fiber without generating excessive nonlinear signal distortions [24]. A possible launch power reduction enters the link budget the same way as an SNR penalty from using a richer symbol constellation. A quantification of this nonlinear penalty depends on a host of system parameters and cannot be given generally. In order to maintain the reach of a typical 100G WAN transport system (~1500 km), several system improvements have to be made to recover the lost SNR, including the use of low-loss low-nonlinearity optical fiber, Raman amplification, and stronger FEC.

1T TECHNOLOGY OPTIONS
As became evident in our discussion of 400G interface options, 1-Tb/s interface rates seem distant today from a technology point of view, unless massively parallel transport in the LAN and moderate spectral efficiencies in the WAN are being considered. Extrapolating the progress of serial bit rate research (green line in Fig. 1), 1-Tb/s serial interface experiments could be expected around 2020. The exponential scaling of serial bit rates at about 12 percent (0.5 dB) per year follows the speed increase of semiconductor devices [25],

Two subcarriers are orthogonal if they are spaced in frequency by the symbol rate and if the symbols on each subcarrier are temporally aligned.

28

IEEE Communications Magazine July 2010

100 GIGABIT ETHERNET

PDM 256-QAM 4 GBaud (64 Gb/s) [16]

PDM 64-QAM 9.3 GBaud (112 Gb/s) [17]

PDM 32-QAM 10 GBaud (93 Gb/s) [18]

PDM 16-QAM 28 GBaud (112 Gb/s) [19]

PDM QPSK 56 GBaud (224 Gb/s) [20]

Figure 3. Recently reported high-speed multi-level constellations for single-carrier modulation [1620]. Reproduced with permission. which are currently able to provide 172-Gb/s logic circuits [26] that could be used to implement serial bit rates close to 300 Gb/s. Note that the use of higher-order modulation (M-QAM) is unlikely to change this scaling, since richer constellations require a higher effective number of bits (ENoB) in digital-toanalog and analog-to-digital conversion (DAC, ADC). Comparing the increased ADC requirements for M-QAM [27] with Waldens observations on the scaling of ADCs [28], which improve at about 1/3 ENoB per year at constant converter speed, we find little room for an extra increase in serial bit rates by going to higher-order modulation in addition to adopting higher symbol rates. As a consequence, and in contrast to 400G, 1T Ethernet and OTN will almost certainly need significantly parallel transport interfaces, in the LAN and in the WAN. Experimental demonstrations of such 1.2Tb/s optical superchannels have been reported [29] using the scheme shown in Fig. 2c. Solutions of this kind require considerable optical parallelization and hence ask for massive photonic integration for viable commercialization. Recent examples of monolithic photonic integration include a 10 45.6 Gb/s differential PDM-QPSK direct detection receiver [30] and a 4 43 Gb/s PDM-QPSK coherent receiver [31] implemented on InP, and a 112-Gb/s coherent receiver implemented on Si [32]. In the WAN a further increase beyond the 8-b/s/Hz spectral efficiency mentioned above for 400G to, say, 16 b/s/Hz is certainly very challenging (in particular for commercialization by 2020), but is not yet fundamentally impossible. Recent studies on the Shannon capacity of nonlinear fiber channels estimate the spectral efficiency limit of a polarization-multiplexed 1000-km link of standard fiber to be at ~16 b/s/Hz [24]. Nevertheless, reaching such high values at high interface rates will be increasingly difficult. The fact that current research has approached fundamental Shannon limits to within a strikingly close factor of ~3 [24] is also reflected in Fig. 1 (triangles). The slope in WDM research capacity growth (and in PDM/WDM capacity growth after 2000) has changed from ~2.5 dB/year (~78 percent/year) to ~0.8 dB/year (~20 percent/year), with a similar trend seen in WDM products. It is very well possible that WAN transport will eventually have to resort to spatially parallel transmission technologies, in the form of either multiple parallel fibers or multimode optical fiber structures in order to guarantee the scaling of network capacities well into the future [33]. able to use polarization-multiplexed multilevel modulation on a single optical carrier in the WAN. Single-wavelength optical 100G WAN interfaces as well as 100G router ports are commercially available today. Interfaces at 400G seem to be a technologically viable next step for standardization, in both the LAN and the WAN. With 400G, some optical parallelization may be entering the WAN, from both the interface bit rate and spectral efficiency points of view. Significant optical parallelization will be needed for 1T interfaces in LAN and WAN, even when taking into account the anticipated progress in serial bit rate research until 2020. The need for optical parallelization will require substantial improvements in photonic integration technologies to meet some of the explicit requirements for Ethernet standardization, which include technical feasibility, economic feasibility, and a broad market potential.

ACKNOWLEDGMENT
The author is grateful for many enlightening discussions with Bob Tkach, Andy Chraplyvy, Rene-Jean Essiambre, Xiang Liu, S. Chandrasekhar, Alan Gnauck, Jeff Sinsky, Chris Doerr, Steve Trowbridge, Bernd Teichmann, Nils Weimann, Y. K. Chen, Nick Sauer, and Andras Kalmar.

REFERENCES
[1] P. J. Winzer, G. Raybon, and M. Duelk, 107-Gb/s Optical ETDM Transmitter for 100G Ethernet Transport, Proc. ECOC, paper Th4.1.1, 2005.

10 Spectral efficiency (b/s/Hz) 256 64 16 QAM 128-QAM 32-QAM

no Sh an

n
4 16-QAM 4-QAM (QPSK) 1 0 5 10 15 Required SNR per bit (dB) 20 25

CONCLUSION
With the new 100G standard, Ethernet has caught up with OTN in employing the highest serial bit rates that are technologically feasible today. As a consequence, 100G has to resort to parallel transport technologies in the LAN, while still being

Figure 4. Trade-off between spectral efficiency and SNR as dictated by Shannon; spectral efficiency values refer to a single polarization. (Circles: Theoretical performance; squares: experimental results.) From [23], reproduced with permission.

IEEE Communications Magazine July 2010

29

100 GIGABIT ETHERNET


[2] M. Birk et al. , Field Trial of Real-Time, Single-Wavelength, Coherent 100-Gb/s PM-QPSK Channel U p grade of an Installed 1800km Link, Proc. OFC, paper PDPD1, 2010. [3] T. J. Xia et al., End-to-End Native IP Data 100G Single Carrier Real Time DSP Coherent Detection Transport over 1520-km Field Deployed Fiber, Proc. OFC, paper PDPD4, 2010. [4] J. DAmbrosia, 100 Gigabit Ethernet and Beyond, IEEE Commun. Mag., vol. 48, no. 3, 2010, pp. S6S13. [5] R. W. Tkach, Scaling Optical Communications for the Next Decade and Beyond, Bell Labs Tech. J., vol. 14, no. 4, 2010, pp. 310. [6] C. Matsumoto, The 400-Gig Vision, LightReading, Sept. 21, 2009; S. Lawson, Facebook Sees Need for Terabit Ethernet, LightReading, Feb. 2, 2010; C. Matsumoto, Dare We Aim for Terabit Ethernet?, LightReading, Mar. 5, 2010. [7] P. J. Winzer et al., 107-Gb/s Optical Signal Generation Using Electronic Time-Division Multiplexing, J. Lightwave Tech., vol. 24, no. 8, 2006, pp. 310713. [8] R. Derksen et al., Integrated 100-Gb/s ETDM Receiver in a Transmission Experiment over 480 km DMF, Proc. OFC, paper PDP37, 2006. [9] J. H. Sinsky et al., 107-Gb/s Opto-Electronic Receiver with Hybrid Integrated Photodetector and Demulti p lexer, Proc. OFC , p a p er PDP30, 2007. [10] C. Kazmierski et al., 100-Gb/s Operation of an AlGaInAs Semiconductor Insulating Buried Heterojunction EML, Proc. OFC , p a p er OThT7, 2009. [11] C. R.S. Fludger et al., Coherent Equalization and POLMUX-RZ-DQPSK for Robust 100-GE Transmission, J. Lightwave Technol., vol. 26, no. 1, 2008, pp. 6472. [12] F. Ruhl et al. , 2038km and Four 50GHz OADM/Filters Transmission Field Trial of 115.2-Gb/s Coherent CoFDM Modem in the Telstra Network, Proc. OFC, paper NTuC4, 2010. [13] F. Chang, K. Onohara, and T. Mizuochi, Forward Error Correction for 100G Transport Networks, IEEE Commun. Mag., vol. 48, no. 3, 2010, pp. S48S55. [14] S. L. Jansen et al. , O p tical OFDM, a hy p e or is it for real?, Proc. ECOC, Mo.3.E.3, 2008. [15] B. Spinnler, Complexity of Algorithms for Digital Coherent Receivers, Proc. ECOC, paper 7.3.6, 2009. [16] M. Nakazawa et al., 256 QAM (64 Gb/s) Coherent Optical Transmission over 160 km with an Optical Bandwidth of 5.4 GHz, Proc. OFC, paper OMJ5, 2010. [17] Y. Yu et al., 112.8-Gb/s PM-RZ-64QAM Optical Signal Generation and Transmission on a 12.5GHz WDM Grid, Proc. OFC, paper OThM1, 2010. [18] Y. Mori et al., 200-km Transmission of 100-Gb/s 32-QAM Dual-Polarization Signals Using a Digital Coherent Receiver, Proc. ECOC, paper 8.4.6, 2009. [19] A. H. Gnauck et al., 10 224-Gb/s WDM Transmission of 28-GBaud PDM 16-QAM on a 50-GHz Grid over 1,200 km of Fiber, Proc. OFC , paper PDPB8, 2010. [20] A. H. Gnauck et al., 10 224-Gb/s WDM Transmission of 56-Gbaud PDM-QPSK Signals over 1,890 km of Fiber, Photon. Technol. Lett., vol. 22, no. 13, 2010, pp. 95456. [21] A. Sano et al. , No-Guard-Interval Coherent O p tical OFDM for 100Gb/s Long-Haul WDM Transmission, J. Lightwave Tech. , vol. 27, no. 16, 2010, pp. 370513. [22] X. Liu et al., Transmission of a 448-Gb/s Reduced-Guard-Interval COOFDM Signal with a 60-GHz Optical Bandwidth over 2000 km of ULAF and Five 80-GHz-Grid ROADMs, Proc. OFC, paper PDPC2 (2010). [23] P. J. Winzer et al., Spectrally Efficient Long-Haul Optical Networking Using 112-Gb/s PDM 16-QAM, J. Lightwave Tech., vol. 28, 2009, p 547. [24] R.-J. Essiambre et al., Capacity Limits of Optical Fiber Networks, J. Lightwave Tech., vol. 28, no. 4, 2010, pp. 662701. [25] International Technology Roadmap for Semiconductors, www.itrs.net [26] C. Monier et al., 172 GHz Divide-by-Two Circuit Using a 0.25-m InP HBT Technology, Proc. IPRM, 2009, pp. 2427. [27] T. Pfau , S. Hoffmann, and R. Noe, Hardware-Efficient Coherent Digital Receiver Conce p t with Feedforward Carrier Recovery for M-QAM Constellations, J. Lightwave Tech., vol. 27, no. 8, 2009, pp. 98999. [28] R. H. Walden, Analog-to-Digital Converters and Associated IC Technologies, Proc. CSIC (2008); R. H. Walden, Analog-to-Digital Converter Survey and Analysis, IEEE JSAC, vol. 17, no. 4, 1999, pp. 53950. [29] S. Chandrasekhar et al. , Transmission of a 1.2-Tb/s 24-Carrier NoGuard-Interval Coherent OFDM Su perchannel over 7200-km of UltraLarge-Area Fiber, Proc. ECOC, paper PD2.6 (2009). [30] R. Nagarajan et al. , 10 Channel, 45Gb/s p er Channel, Polarization Multiplexed DQPSK InP Receiver Photonic Integrated Circuit, Proc. OFC, paper PDPB2 (2010). [31] C. R. Doerr, L. Zhang, and P. J. Winzer, Monolithic InP Multi-Wavelength Coherent Receiver, Proc. OFC, paper PDPB1 (2010). [32] C. R. Doerr et al., Monolithic Silicon Coherent Receiver, Proc. OFC, paper PDPB2 (2009). [33] A. R. Chraplyvy, The Coming Capacity Crunch, ECOC Plenary Talk, 2009.

30
IENYCM2694.indd 1 6/17/10 8:26:24 PM

IEEE Communications Magazine July 2010

You might also like