Study of 100 Gigabit Ethernet Using Carrierless ... - OSA Publishing

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 31, NO. 9, MAY 1, 2013

1367

Study of 100 Gigabit Ethernet Using Carrierless Amplitude/Phase Modulation and Optical OFDM J. L. Wei, Member, IEEE, D. G. Cunningham, Member, IEEE, R. V. Penty, Senior Member, IEEE, and I. H. White, Fellow, IEEE

Abstract—For the first time, simulations have been performed to evaluate and compare the link power budget and power dissipation of 100 Gb/s carrierless amplitude and phase modulation-16/64 (CAP-16/64) and 16/64-quadrature amplitude modulation-orthogonal frequency division multiplexing (16/64-QAM-OFDM) systems over feedforward error correction (FEC) enhanced single mode fiber (SMF) links using an 18.6 GHz bandwidth directly modulated laser, for both single channel and two coarse wavelength division multiplexing (CWDM) channel cases. It is shown that single channel CAP-16 and 16-QAM-OFDM links can successfully support transmission over 5 km SMF, with a power dissipation of 2 times that of a 4 25 Gb/s NRZ system. Even when the loss of the optical multiplexing/demultiplexing operations is considered, the use of two CWDM channels supports transmission over 5 km SMF with CAP-16 and 16-QAM-OFDM. The CWDM systems do not increase transceiver power dissipation greatly. Index Terms—Carrierless amplitude/phase modulation, Ethernet networks, equalizer, OFDM modulation, wavelength division multiplexing.

I. INTRODUCTION

B

Y using conventional modulation formats, such as non return-to-zero (NRZ), the initial 40/100 Gigabit Ethernet standardization has been completed by IEEE 802.3 [1]. For single mode links 100 Gigabit Ethernet currently uses four wavelengths and wavelength division multiplexing (WDM) optics. However, this may have impact on the system cost given the number of optical components used [2], [3], and in itself WDM does not lead to power savings per se. As a result, higher spectral efficiency modulation formats, such as multilevel modulation, have recently been proposed as an alternative solution to reduce the optics component count and thus the system cost. For example, pulse amplitude modulation (PAM) modulation schemes, together with forward error correction (FEC) using a single laser have been proposed by the IEEE 802.3 Next Generation (NG) 100 GE PMD study group [2], [3] for single mode fibre (SMF) lengths of 500 m to 2 km. Alternatively, Manuscript received November 12, 2012; revised January 05, 2013, February 03, 2013; accepted February 17, 2013. Date of publication February 21, 2013; date of current version March 20, 2013. J. L. Wei, R.V. Penty, and I. H. White are with Centre for Photonic Systems, Electrical Engineering Division, Department of Engineering, University of Cambridge, Cambridge CB3 0FA, U.K. (e-mail: [email protected]; [email protected]; [email protected]; [email protected]). D. G. Cunningham is with Avago Technologies, Framlingham Technology Centre, Station Road, Framlingham, Suffolk, IP13 9EZ, U.K. (e-mail: david. [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JLT.2013.2248122

coarse WDM (CWDM) with a reduced number of channels is also worth investigation as it also results in cost reduction. Compared with PAM, carrierless amplitude/phase modulation (CAP) [4], [5] and optical orthogonal frequency division multiplexing (OOFDM) [6] can further improve system spectral efficiency and flexibility without greatly increasing the digital signal processing (DSP) required. Additionally, power efficiency has become more important for the communications and networks sector to reduce the carbon footprint of the Internet. It is thus very important to assess the performance of advanced modulation formats from this perspective as well. Motivated by these challenges, for the first time we have investigated the feasibility of 100 Gigabit Ethernet enabled by CAP and OOFDM modulation schemes using directly modulated lasers (DML). Both single channel and two CWDM channel configurations are considered. The system performance for these modulation formats have been evaluated and compared by simulation in terms of both link power budget and power dissipation. The paper is organized as follows: Section II describes the system architecture of various transceivers and presents the simulation parameters. Section III presents the results relating to link power budget for both single channel and two CWDM channel configurations. Section IV analyzes the power dissipation for each transceiver and the paper is summarized in Section V. II. SYSTEM ARCHITECTURES A. CAP-16/64 Transceivers Fig. 1(a) depicts the 100 Gb/s CAP-16 and CAP-64 transceiver architectures. The four 25 Gb/s tributary bit streams are first encoded with FEC and then converted onto two parallel streams. Each bit stream is mapped onto PAM-4/PAM-8 symbols and then pulse shaping is performed using a passband square root raised cosine filter [7]. The two orthogonally shaped signals are combined and converted into an analog signal via a digital to analog convertor (DAC), whose output drives a DML. The optical signal propagates over an SMF link and is detected by a square-law photodetector (PD) in the receiver. The ADC converted electrical signal in the receiver is processed in two parallel lanes using feedforward equalisation (FFE) and decision feedback equalisation (DFE) providing channel equalization and demodulation. The signal processing procedure after the DFE in the receiver is the inverse of that in the transmitter. Though the single channel setup described above brings about the best system cost-effectiveness, it is also worth mentioning another possible configuration with two CWDM

0733-8724/$31.00 © 2013 IEEE

1368


Fig. 1. Diagram of 100 G Ethernet PMD based on (a) CAP and (b) OOFDM. The blocks in dash indicate the cases of using two CWDM channels configurations. (a) CAP-16/CAP-64, (b) OOFDM using 16 QAM or 64 QAM.

channels, as shown in the dashed blocks in Fig. 1(a): two identical baseband square root raised cosine filters are used as shaping filters and the two shaped signals are converted into analog signals by two identical DACs, whose outputs drive two DMLs. The optical signals are multiplexed and transmitted over the SMF link and de-multiplexed and detected by two PDs in the receiver. The single channel configuration represents standard CAP [7], while the second configuration, strictly speaking, consists of two PAM-4/PAM-8 channels with square root raised cosine pulse shaping, which give almost halved signal bandwidth, and thus improved transmission performance, compared to conventional PAM with rectangular shaping [2], [3]. Consequently the CAP signal 3-dBe bandwidth is equal to the symbol rate (half of the symbol rate) in the first (second) configuration. Due to the reduced signal bandwidth, the detected electrical signal in the receiver can be recovered by using either a matched filter [7] or a FFE and DFE equalizer in each parallel lane. B. 16/64-QAM-OFDM Transceivers The 100 Gb/s 16/64-QAM-OFDM transceivers, shown in Fig. 1(b), involve OFDM modems with dense DSP blocks [6]. The transmitter OFDM modem includes variable power loading, symbol mapping using 16/64-QAM, inverse fast Fourier transform (IFFT), cyclic prefix insertion, OFDM symbol serialization and DAC. The two link types described above can be incorporated, the major difference being the IFFT input arrangement: for the single channel case, the IFFT input satisfies Hermitian symmetry so that the IFFT output is real-valued [6], whilst in the CWDM case, all the IFFT inputs carry user information and the IFFT output is complex-valued. Therefore, to achieve the same bit rate, the DAC sampling rate for the second case is only half that in the first case.

The simulation models for CAP and optical OFDM systems have been detailed in [4]. The system performance is obtained by calculating the system symbol error rate (SER) and bit error rate (BER) in the receiver using the complementary error function. All the noise sources are assumed to be additive white Gaussian noise. For optical OFDM systems, the receiver side effective signal to noise ratio (SNR) is obtained by referring to [8] where the signal clipping and DAC/ADC quantization caused SNR degradations are considered in addition to receiver thermal noise and shot noise. C. DML Model A Gaussian filter is used to model the laser response in the spreadsheet link model normally used for estimating the optical link power budget for DML-based Ethernet and Fibre Channel (FC) links that use NRZ format. The Gaussian model has proved sufficient for Gigabit Ethernet, 10 Gigabit Ethernet and the initial versions of 40/100 Gigabit Ethernet [9]. Therefore, for simplicity, the Gaussian model was used to make a first order estimate of the performance of more complex modulation schemes. To estimate the magnitude of the approximation due to using the Gaussian response, a more accurate rate equation based model [6], [10] is also used here. The validity of the rate equation model has been verified rigorously in [6], [10], where the simulated rate equation DML model-based intensity modulation and direct detection (IMDD) optical OFDM system performance agrees very well with the experimental measurements of real-time DML-based IMDD optical OFDM systems. Fig. 2 shows the recovered CAP-16 signals for the two cases of using a Gaussian response model and using a rate equation model, which shows clearly the system performance for the two cases are very similar. Fig. 3 demonstrates the received 16-QAM-OFDM constellation diagrams before and after equalization for the two cases of using a Gaussian response

WEI et al.: 100 GIGABIT ETHERNET USING CARRIERLESS AMPLITUDE/PHASE MODULATION

1369

TABLE I OPTICAL TRANSCEIVER AND FIBRE LINK PARAMETERS

Fig. 2. CAP-16 receiver In-phase eye diagram after equalization using (a) Gaussian response DML model and (b) rate equation DML model. The SMF length is set to 2 km.

Gb/s NRZ and PAM systems, the rate equation model has to be used due to the relatively high symbol rates that lead to a strong sensitivity of the system performance to the DML nonlinearity [11]. D. Simulation Parameters

Fig. 3. 16-QAM-OFDM received constellation diagrams before and after equalization. (a) and (b) are constellation diagrams before equalization and (c) and (d) are constellation diagrams after equalization. The SMF length is set to 2 km. (a) Gaussian response model, (b) Rate equation model, (c) Gaussian response model, (d) Rate equation model.

model and using a rate equation model. Though the use of rate equations brings about worse constellation diagrams compared with those using a Gaussian response, the performances of the two cases are similar at a BER of (the FEC threshold is assumed to be at that BER). Simulation results show that, by using the two different DML models, the required optical power to achieve a BER of for an optical back to back 16-QAM-OFDM system have a difference of roughly 0.2 dBo, which is negligible, although the optical power difference can be very large to achieve a lower BER. Moreover, we also observe from simulation results that the achievable link power margins for the two DML model cases are very similar, with the difference being less than 1 dBo. Similar results are also observed for CAP-64 and 64-QAMOFDM. Therefore, based on the above results and discussions, the conclusion is made that the Gaussian model based DML model is sufficiently accurate for 100 Gb/s CAP and optical OFDM systems. However, it should be pointed out that for 100

Both the CAP and OOFDM links use the optical transceiver parameters shown in Table I. The reference receiver has a sensitivity of dBm at a BER of (corresponding to dBm at a BER of ) for a reference 28 Gb/s NRZ system. Thus, the total link power budgets for a launch power of 0 dBm is 13.7 dB using FEC —i.e., FEC which generates a BER of for an input BER of . FEC based on transcoding of 64B/66B is assumed, which gives rise to a FEC overhead of 3.13%. For the CAP link, two parallel decorrelated PRBS streams are used and the square root raised cosine filter has a roll-off coefficient of 0.5. The CAP receiver equalizer is composed of a 20 tap, T/4 (T is the symbol period) spaced FFE and a 3 tap DFE filter. OOFDM adopts an IFFT/FFT size of 64 with 31 or 64 parallel decorrelated PRBS stream sources and a cyclic prefix of 6.25% or 12.5% for the single channel or the CWDM configuration, respectively. The channel estimation used in OOFDM simulation is performed by transmitting pilot tones which leads to an overhead of 5%. For 16(64)QAM-OFDM, to minimize the clipping and quantization induced noise, an optimum clipping ratio of 7 dB (9 dB) and an optimum number of DAC/ADC quantization bits of 6 (8) [4], [8], [12] are used, which gives rise to a maximized effective SNR in the receiver. Based on the above parameters, the required DAC/ADC sampling rates for each transceiver, to achieve an aggregate bit rate of 100 Gb/s, are listed in Table II. Note that the per lane electrical signaling rate for current and next generation data communication standards is based on the Optical Internetworking Forum, Common Electrical Interface for Very Short Reach: OIF CEI-28G-VSR. Therefore, the CEI-28G-VSR interface is the basis for the per lane rates of 100 Gigabit Ethernet (4 lanes at 25.78125 Gb/s), 32GFC (28.025 Gb/s) and 128GFC (4 lanes at 28.025 Gb/s). Since the highest signaling rate is 28.025 Gb/s, 28 Gb/s was selected as the signaling rate for the reference receiver.

1370


Fig. 4. Link power budget for CAP and OFDM for the single channel configuration using DFE and FEC.

TABLE II DAC/ADC SAMPLING RATE REQUIRED FOR VARIOUS 100 GB/S SYSTEMS

III. RESULTS FOR LINK POWER BUDGET Fig. 4 shows the link power penalty for the single channel configuration for SMF lengths ranging from 500 m to 5 km. A total link power budget of 13.7 dB is used. It should be pointed out that a 100 Gb/s NRZ system under this link configuration does not work, even if FEC and equalization are incorporated. The failure for NRZ is mainly because of its relatively high signal symbol rate, leading to strong susceptibility to DML nonlinearities [11]. The link power penalty comprises contributions from the relative receiver sensitivity [4], dispersion penalty, relative intensity noise (RIN) penalty, the optical link loss (which includes fiber attenuation and 2 dBo connector loss), implementation penalty due to jitter and other effects [2], and unallocated penalty if available. The relative receiver sensitivity is defined as the ratio of the receiver sensitivity of the proposed 100 Gb/s system to that of the 28 Gb/s reference NRZ system for the optical back-to-back case. The dispersion penalty/RIN penalty is defined as the ratio of the relative receiver sensitivity taking into account fibre dispersion/laser RIN noise to that without. The jitter penalty associated with implementation is assumed to be fixed according to [2]. Therefore, except for the link loss, all other penalty components are independent of fibre length. The

unallocated penalty is a direct reflection of the system power margin: the more, the better. If there is a negative unallocated penalty, then the link fails. Thus, Fig. 4 clearly shows that the CAP-16 and 16-QAM-OFDM modulation schemes can support transmission up to 5 km SMF which is almost twice the target 2 km SMF length [2], [3]. In addition, CAP-64 can support transmission only for 500 m of SMF and 64-QAM-OFDM fails, due to the high multilevel penalty and clipping and quantization induced strong penalty. The success of CAP-16 and 16-QAM-OFDM or the failure of CAP-64 and 64-QAM-OFDM rely on the relative receiver sensitivity which is dominated by multilevel penalty and noise enhancement for CAP-64 and multilevel penalty as well as clipping and quantization induced penalty for 64-QAM-OFDM. Both CAP-16 and 16-QAM-OFDM have much smaller multilevel penalties as fewer levels are used to represent symbols compared to the other two schemes. For a fixed number of symbol levels, OFDM exhibits a slightly larger relative receiver sensitivity compared to its CAP counterpart, as shown in Fig. 6. This is because the clipping and quantization penalty for OOFDM systems is stronger than the equalization related noise enhancement penalty of CAP. Other penalty constituents are quite similar for both CAP and OOFDM. The fiber dispersion is negligible since the SMF length is only a few km. The RIN penalty increases with the number of symbol levels used, e.g., RIN penalty for CAP-16 is less than that for CAP-64. The implementation penalty from jitter and other effects of 1.1 dB has been included [2]. Note that in Fig. 4, the DAC/ADC quantization penalty is negligible for CAP systems as the optimum number of quantization bits of 6 has been adopted. In other words, it has the same performance to that using an ideal DAC/ADC. Therefore, the link power budget of CAP systems shown in Fig. 4 also represents that of CAP systems without using DAC/ADC. Such systems may be implemented using analog components [5], [11]. These conclusions also apply to Fig. 5. Fig. 5 shows the link power budgets for various modulation schemes using two CWDM channels configuration. The loss


1371

Fig. 5. Link power budget for CAP and OFDM for the two CWDM channels configuration using FEC. U: unequalized; D: DFE. The link loss for CAP and optical OFDM do not include WDM loss, while the link loss for the 4 25 Gb/s NRZ Ethernet system includes WDM loss and connectors’ loss.

caused by optical multiplexing/demultiplexing operations is ignored in order to identify the change of each penalty constituent in comparison to the single channel case. It can be clearly seen that all the modulation schemes can support transmissions up to 2 km SMF which is the target SMF length [2]. Apart from 64-QAM-OFDM, the other schemes can further support transmission up to 5 km and indeed for even longer distances for CAP-16 and 16-QAM-OFDM. It should be noted that CAP schemes without FFE/DFE also work well with the two CWDM channels configuration, although the relative receiver sensitivity is improved considerably if the FFE/DFE equalizer is incorporated. However, for CAP-64, the relative receiver sensitivity improvement caused by FFE/DFE is very small. This is because CAP-64 has a smaller bandwidth compared with CAP-16, which means it has much less spectral/waveform distortion, or in other words inter-symbol interference, than CAP-16. Thus, the FFE/DFE for CAP-64 receiver functions simply as a matched filter rather than an equalizer. It is interesting to compare the link power budget shown in Fig. 5 with that in Fig. 4. Relative to the single channel case, the use of two CWDM channels improves the system link power margin shown in Fig. 4 by 1.5 dB and 3.5 dB for 16-QAM-OFDM and CAP-16 systems, respectively, though it doubles the number of DACs/ADCs, DMLs and PDs. For CAP, the power margin enhancement is mainly attributed to the negative noise enhancement penalty from DFE equalization, where the equalizer simply works as a matched filter since inter-symbol interference is relatively small due to the halved CAP signal bandwidth. For OOFDM, the enhancement is because the receiver noise bandwidth is halved. When the loss from the optical multiplexing/demultiplexing is considered for the 2 CWDM channel configuration, the improvement in power margin mentioned above is partially offset by the WDM loss, which by referring to 100 GBASE-LR4 [1], is assumed to be 4 dBo in total. It is assumed that even with the WDM multiplexer a per-lane transmit power of 0 dBm can be achieved. As a result, by subtracting the 2 dBo loss for the WDM

demultiplexer from the power margin of CAP and OOFDM schemes shown in Fig. 5, only 64-QAM-OFDM scheme fails to support links length up to 2 km. Whilst CAP-16 and 16-QAMOFDM can support transmission up to a link length of 5 km of SMF. It is interesting to point out that, as a reference link performance, the link power budget of the IEEE 802.3ba 4 25 Gb/s NRZ Ethernet system has also been investigated for an optical back to back case and results are shown in Fig. 5. This shows that the 4 25 Gb/s NRZ Ethernet system has roughly the same link power margin when compared with that of the unequalized CAP-16 using 2 CWDM lanes when the WDM loss in the CAP-16 receiver is accounted for. IV. SYSTEM POWER DISSIPATION ESTIMATES COMPARISONS

AND

The power dissipation of the various systems are estimated based on 65 nm CMOS technology as referred to in recently published reports [2], [3], [12]–[16]. For comparison, the power consumptions for a single lane 25 Gb/s NRZ system and a 100 Gb/s NRZ system consisting of 4 parallel CWDM lanes are also presented. The single lane 25 Gb/s NRZ system is taken as the reference NRZ system, whose transceiver power dissipation is about 1 watt [4]. The transceiver power consumptions for other modulation schemes are normalized to that of the reference NRZ system. The normalized power dissipations are plotted in Fig. 6, where the power dissipation for major transceiver constituent components is also given. It can be seen that the CAP systems consume similar power to their OOFDM counterparts. For the single channel case, the power consumption for single channel CAP-16 and 16-QAM-OFDM is 8 times and 2 times that of a single channel 25 Gb/s NRZ system and a 4 25 Gb/s NRZ 100 Gigabit Ethernet system, respectively. It should be pointed out that the power dissipation of the 4 25 Gb/s NRZ 100 Gigabit Ethernet system did not take into account power dissipation of the laser cooling devices. Compared with

1372


Fig. 6. Estimated transceiver power dissipation for various transceivers. U: unequalized; D: 20 tap T/4 spaced FFE and 3 tap DFE; Q: QAM.

the single channel cases, the transceiver power dissipation of its corresponding two CWDM channels system does not increase greatly. Note that DAC/ADC power dissipation [15], [16], dominates the transceiver power consumption. For e.g., CAP-16 (16-QAM-OFDM) with single channel configuration, the DAC/ADC power consumption accounts for 62% 50% of that for the overall transceiver consumption. The DAC/ADC power dissipation almost linearly depends on its sampling rate. Therefore, by considering the DAC/ADC sampling rate required by each modulation scheme shown in Table II, it is easy to explain why CAP-16 (16-QAM-OFDM) consumes more power than CAP-64 (64-QAM-OFDM). For a fixed modulation scheme, the DAC/ADC power dissipation for single channel and the two CWDM channels cases are quite similar by considering the number of DAC/ADC used and their sampling rates. For OOFDM, the DSP in optical OFDM systems also account for a significant percentage of the overall transceiver power consumption [12]. The “Other” item shown in Fig. 6 refers to the normalized power dissipation of all the components that not listed in Fig. 6 but appear in Fig. 1. Taking CAP-16 as an example, it refers to the normalized power dissipation of 0.44 for , 0.22 for S/P and P/S, 0.3 for transmitter shaping filter, 0.45 for mapping and 0.17 for photo-diode and TIA amplifier. Therefore, great power consumption savings are possible if DAC/ADC power consumption can be reduced by using more advanced CMOS technologies such as 40 nm or even 20 nm CMOS gate size [11]. Alternatively, a system implemented without ADC/DAC brings about significantly high power efficiency. We have recently experimentally demonstrated a proof-of-concept 50 Gb/s CAP system by using analog transversal filters in the transmitter and receiver [11]. Note that the ‘CAP-16 Demo’ shown in Fig. 6 represents the demonstrated CAP-16 system without requiring the use of DAC/ADC. Since the analog transmitter and receiver will dissipate much less power than the DAC and ADC a lower bound on the power dissipation of an analog implementation for the CAP

schemes can be obtained by subtracting the power dissipation of the DAC/ADC’s shown in Fig. 4. Due to the absence of DAC/ADC, the demonstrated CAP system has transceiver power consumption even lower than that for a 4 25 Gb/s system, indicating great potential for both cost-effectiveness and power efficiency. Compared with a 4 25 Gb/s NRZ system, the other advantages of complex modulation include: first, the minimization of the number of optoelectronic front end components and the reduction in the number of critical optical alignments. This should greatly increase reliability and manufacturability. Second, there is the opportunity to take advantage of Moore’s law for electronics, which should mean that the cost of the module will dramatically reduce as electronic speed and complexity increase with time. Third, the use of complex modulation is progressive in that to achieve 400 Gigabit Ethernet (the next target rate) the use of 25 Gb/s lanes would be very inefficient. So schemes like CAP will be required to reduce the number of lanes. In addition, considering the required DAC/ADC sampling speed, it might be challenging to implement CAP and optical OFDM using state-of-the-art techniques. To address this, the CAP may be implemented by using analog components [5], [11] and using a QAM receiver to decrease its timing jitter sensitivity [17]. Also optical OFDM may be implemented by using multiple bands in the RF domain to parallelize the high speed signal into a few low speed channel signals [18]. V. CONCLUSION We have successfully demonstrated the feasibility of 100 Gigabit Ethernet PMD enabled by CAP-16/CAP-64 and 16-QAM-OFDM/64-QAM-OFDM. Simulation results show that, incorporating FEC and DFE, single channel CAP-16 and 16-QAM-OFDM can support transmission over km of SMF, with transceiver power dissipation of about 2 times that of the 4 25 Gb/s NRZ version of 100 Gigabit Ethernet. The two CWDM channel systems also support transmission over 5 km of SMF via the CAP-16 and 16-QAM-OFDM


schemes. The two CWDM channel schemes exhibit a modest increase in transceiver power dissipation compared to the single channel case. Moreover, CAP schemes which do not require DAC/ADCs have great potential for cost-effectiveness and power efficiency. ACKNOWLEDGMENT

1373

[17] J. L. Wei, L. Geng, D. G. Cunningham, R. V. Penty, and I. H. White, “100 Gigabit Ethernet transmission enabled by carrierless amplitude and phase modulation using QAM receivers,” in Proc. OFC/NFOEC13, 2013, Paper OW4A.5. [18] R. P. Giddings, E. Hugues-Salas, and J. M. Tang, “Experimental demonstration of record high 19.125 Gb/s real-time end-to-end dual-band optical OFDM transmission over 25 km SMF in a simple EML-based IMDD system,” Opt. Exp., vol. 20, no. 18, pp. 20666–20679, Aug. 2012.

This work was supported by EPSRC via the INTERNET project. REFERENCES [1] IEEE 802.3ba—40 Gb/s and 100 Gb/s Ethernet, 2010. [2] S. Bhoja, “Study of PAM modulation for 100 GE over a single laser,” IEEE Next Gen 100 G Optical Ethernet Study Group Mar. 2012 [Online]. Available: http://www.ieee802.org/3/100GNGOPTX/public/jan12/bhoja_01_0112_NG100GOPTX.pdf [3] V. Bhatt, B. Dama, and G. Nicholl, “Update on advanced modulation for a low cost 100 G single mode fiber PMD,” IEEE Next Gen 100 G Optical Ethernet Study Group Jul. 2012 [Online]. Available: http://www.ieee802.org/3/100GNGOPTX/public/jul12/nicholl_01_0712_optx.pdf [4] J. L. Wei, J. D. Ingham, D. G. Cunningham, R. V. Penty, and I. H. White, “Performance and power dissipation comparisons between 28 Gb/s NRZ, PAM, CAP and optical OFDM systems for datacommunication applications,” J. Lightw. Technol., vol. 30, no. 20, pp. 3273–3280, Oct. 2012. [5] J. D. Ingham, R. V. Penty, and I. H. White, “40 Gb/s carrierless amplitude and phase modulation for low-cost optical datacommunication links,” presented at the OFC/NFOEC11, 2011, Paper OThZ3. [6] J. L. Wei, C. Sánches, R. P. Giddings, E. Hugues-Salas, and J. M. Tang, “Significant improvements in optical power budgets of real-time optical OFDM PON systems,” Opt. Exp., vol. 18, no. 20, pp. 20732–20745, Sep. 2010. [7] J. J. Werner, “Tutorial on carrierless AM/PM,” Part I and II, ANSI X3T9.5 TP/PMD Working Group 1992&1993. [8] E. Vanin, “Performance evaluation of intensity modulated optical OFDM system with digital baseband distortion,” Opt. Exp., vol. 19, no. 5, pp. 4280–4293, Feb. 2011. [9] Fibre Channel—Methodologies for Signal Quality Specification (FCMSQS), InterNational Committee for Information Technology Standards (INCITS)/TR-46-2011, INCITS, Dec. 2011. [10] X. Zheng, X. Q. Jin, R. P. Giddings, J. L. Wei, E. Hugues-Salas, Y. H. Hong, and J. M. Tang, “Negative power penalties of optical OFDM signal transmissions in directly modulated DFB laser-based IMDD systems incorporating negative dispersion fibres,” IEEE Photonics J., vol. 2, no. 4, pp. 532–542, Jun. 2010. [11] J. L. Wei, J. D. Ingham, R. V. Penty, and I. H. White, “Performance studies of 100 Gigabit Ethernet enabled by advanced modulation formats,” IEEE Next Gen 100 G Optical Ethernet Study Group May 2012 [Online]. Available: http://www.ieee802.org/3/100GNGOPTX/public/ may12/ingham_01_0512_optx.pdf [12] P. Milder, R. Bouziane, R. Koutsoyannis, C. R. Berger, Y. Benlachtar, R. I. Killey, M. Glick, and J. C. Hoe, “Design and simulation of 25 Gb/s optical OFDM transceiver ASICs,” presented at the ECOC 2011 Geneva, Switzerland, 2011, Paper We.9.A.5. [13] R. S. Tucker, “Green optical communication—Part I: Energy limitations in transport,” IEEE J. Sel. Topics Quantum Electron., vol. 17, no. 2, pp. 245–260, Mar./Apr. 2011. [14] M. Möller, “High-speed electronic circuits for 100 Gb/s transport networks,” in Proc. OFC/NFOEC10, 2010, Paper OThC6. [15] 55–65 GSa/s 8-bit DAC [Online]. Available: http://www.chais.info [16] 56GSa/s 8-BIT ADC [Online]. Available: http://www.chais.info Fujitsu factsheet, preliminary data.

Jinlong Wei (S’09–M’11) received the B.S. degree in computer communication from the University of Electronic Science and Technology of China, Chengdu, China in 2005, and the Ph.D. degree in electronic engineering from the University of Wales, Bangor, Bangor, U.K., in 2010 and worked there as a post-doc researcher afterwards. He joined the Department of Engineering, University of Cambridge, Cambridge, UK as a research associate in Aug. 2011, and is working on energyefficient datacommunication systems. He has been involved in a few EU/national funded research projects such as ALPHA and INTERNET. He has been contributing to the next generation 100 Gigabit Ethernet study within IEEE 802.3. His research interest includes high-speed optical communication systems using advanced modulation formats such as optical OFDM, carrierless amplitude/phase modulation and partial response modulations, passive optical networks, semiconductor optical amplifiers, single-mode fiber, multi-mode fiber and plastic optical fiber data links. He has authored or co-authored over 50 publications in leading peer-reviewed journals and conferences and contributed to 2 U.K./European/U.S. patents. Dr. Wei has served as a reviewer for a few prestigious journals including IEEE JOURNAL OF SELECTED AREAS OF COMMUNICATION, IEEE JOURNAL OF SELECTED TOPICS OF QUANTUM ELECTRONICS, Laser Physics Letters, IEEE PHOTONICS TECHNOLOGY LETTERS, IEEE/OSA Journal of Optical Communications Networking, and IET Optoelectronics Journal. He also served as an organizer and chair for international conference sessions.

Richard V. Penty (M’00–SM’10) received the Ph.D. degree in engineering for his research on optical fiber devices for signal processing applications from the University of Cambridge, Cambridge, U.K., in 1989. He was a Science and Engineering Research Council Information Technology Fellow with the University of Cambridge, where he worked on all-optical nonlinearities in waveguide devices. He is currently the Professor of Photonics at the University of Cambridge, having previously held academic posts at the University of Bath, Bath, U.K., and the University of Bristol, Bristol, U.K. He has been the author of more than 750 refereed journal and conference papers. His research interests include high-speed optical communications systems, photonic integration and green photonics. Prof. Penty is the Editor-in-Chief of the IET Optoelectronics Journal and a Fellow of the Royal Academy of Engineering.

Ian H. White (S’82-M’83-SM’00-F’05) received the B.A. and Ph.D. degrees from the University of Cambridge, UK, in 1980 and 1984, respectively. He was appointed as a Research Fellow and an Assistant Lecturer with the University of Cambridge before he became a Professor of Physics with the University of Bath, UK, in 1990. In 1996, he moved to the University of Bristol, UK, where he was a Professor of Optical Communications and Head of the Department of Electrical and Electronic Engineering in 1998, and the Deputy Director of the Centre for Communications Research. He returned to the University of Cambridge in October 2001 as a van Eck Professor of Engineering. He is currently the Head of Photonics Research and Master of Jesus College in the University of Cambridge. He is the author of more than 800 publications and the holder of 28 patents. Prof. White was a Member of the Board of Governors of the IEEE Photonics Society (2008–2010) and is currently Editor-in-Chief of Electronics Letters.