ECOC 2008, 21-25 September 2008, Brussels, Belgium

Mo.3.E.3

Optical OFDM, a hype or is it for real?

Sander L. Jansen , Itsuro Morita , Kamyar Forozesh , Sebastian Randel , Dirk van den Borne 2 and Hideaki Tanaka 1: Nokia Siemens Networks, Munich, Germany, email: S.L.Jansen@ieee.org 2: KDDI R&D Laboratories, Saitama, Japan 3: Department of Astronomy and Space Physics, Uppsala University, Uppsala, Sweden 4: Siemens AG, Corporate Technology, Munich, Germany
1 2 3 4 1

Abstract During the last two years, numerous researchers have dedicated their work in optical OFDM and, consequently, the number of papers has grown exponentially. In this paper we evaluate the suitability of OFDM for long-haul 100GbE applications. Introduction Optical OFDM is currently a hot topic in the fiber-optic research community and the number of optical OFDM research papers published in international conferences and journals has grown exponentially over the last couple of years [1-3]. However, the question if this modulation format is really a viable candidate for next generation fiber-optic transmission systems is up till now only partially answered. There exist several different definitions of OFDM in the fiber-optic community. In this paper, we refer to OFDM as a digital multicarrier technique. As the subcarriers are generated in the digital domain, these systems typically consist of many subcarriers (typically more than 50) where channel estimation is realized by periodically inserting training symbols [13]. We therefore exclude in the paper coherent WDM systems that sometimes are referred to as OFDM systems [4]. Coherent WDM systems typically have few subcarriers that are generated in the optical domain. These systems typically do not use training symbols, but rely on blind channel estimation instead. Such systems have more in common with singlecarrier coherent systems and its evaluation is out of scope of this manuscript. In this paper we assess how the performance of OFDM scales with coherently detected single-carrier QPSK [9, 12, 13], another promising modulation formats for long-haul transmission. Even though many different OFDM systems have been proposed, we restrict ourselves to coherent detected OFDM, as this modulation format is most suited for long-haul transmission [1-2]. 100GbE OFDM system design For the transport of 100GbE, overhead is required onto the payload data. Although 100GbE has not been standardized yet, it is foreseen that 7% overhead needs to be allocated for FEC and about 4% for the Ethernet protocol (64B/66B coding). This results in a raw data rate Rraw of 111 Gb/s. Apart from the Ethernet and FEC related overhead, an OFDM system has additional overheads caused by cyclic prefix, training symbols and in some cases pilot subcarriers. Especially at high data rates it is essential to minimize these OFDM related overhead as the system overhead significantly increases the bandwidth requirements at the transmitter and receiver. Training symbols (TS) The overhead for training symbols is dependent on the channel dynamics. In a fiber-optic transmission system channel fluctuations are for instance caused by polarization changes. The more stable a channel is the less training symbols are required. In most coherent transmission systems reported to date, the training symbol overhead is about 2% to 4% [1-3]. Pilot subcarriers (PS) Whether pilot subcarriers for phase noise (PN) compensation are required or not is dependent on the phase noise compensation technique. For fiber-optic transmission system two phase noise compensation schemes have been proposed, namely pilot subcarrier [2] and RF-pilot based PN compensation [5]. With pilot subcarrier PN compensation, typically 10% or more of the subcarriers must be allocated for phase noise compensation [2]. In [5] we proposed RF-pilot based PN compensation, in which a lowpower RF-pilot tone is used to revert the phase noise impairments at the receiver. A major advantage of this technology is that pilot subcarriers are not required, saving about 10% of overhead. Another advantage is that this compensation method is more robust towards phase noise and it has been shown that even PN compensation of conventional DFB lasers can be realized with only a minor penalty [6]. Cyclic prefix (CP) The required cyclic prefix (CP) is dependent on the chromatic dispersion that is to be compensated for. The required cyclic prefix time g of an OFDM signal is dependent on the bandwidth of the OFDM band and can be defined as:

g = DBd c / f 2 ,

(1)

where D represents the chromatic dispersion of the desired transmission distance [s/m], c is the speed of light [m/s], f represents the center frequency of the OFDM band [Hz] and Bd [Hz] is the effective

978-1-4244-2229-6/08/$25.00 2008 IEEE

Vol. 1 - 49

ECOC 2008, 21-25 September 2008, Brussels, Belgium

Mo.3.E.3

bandwidth of the modulated OFDM signal:

R R (1 + TS )(1 + CP )(1 + PS ) , (2) Bd = no min al = raw log 2 ( M ) log 2 ( M )

where Rnominal and R raw are the nominal and raw data rate per polarization [b/s], respectively, M is the constellation size and CP, TS and PS are the cyclic prefix, training symbol and pilot subcarrier overheads, respectively. From (2) it can be concluded that the bandwidth Bd is dependent on the CP overhead, which is defined as [5]:

OFDM bands for a system with 10ns and 100ns symbol length. For a system with 100ns OFDM symbol length, the CP overhead is already reduced from 9.1% to 4.3% by using two OFDM bands. In [1], four OFDM bands were used, reducing the total OFDM related overheads for 100GbE OFDM to ~8%

t = 10ns

Cyclic =

g t g

(3)

t = 100ns

Where t is the total OFDM symbol time (including the cyclic prefix). Combining (1), (2) and (3), results in:

D=

g t log 2 (M ) f2 . c Rraw (1 + TS )(1 + PS ) t

2 g

Fig. 2: Cyclic prefix overhead required for 2,000 km SSMF as a function of the number of bands for a system with 10ns and 100ns OFDM symbol length.

(4)
Transmission performance In this section we will review the linear and nonlinear performance of coherently detected OFDM and compare this to the performance of coherently detected single carrier systems. OSNR sensitivity Both at 40-Gb/s [7, 8] and 100-Gb/s [1, 7] similar BER sensitivities have been reported for single carrier and OFDM transmission systems. The OSNR sensitivity for coherent detected systems is for a fixed nominal data rate independent of the number of subcarriers that is used. It is thus not surprising that the OSNR sensitivity of single carrier and multi-carrier (OFDM) systems are comparable at the same data rate. DAC/ADC sampling/bandwidth requirements In all systems that employ coherent detection with digital equalization ADCs are required at the receiver. The bandwidth requirements of these receivers are dependent only on the data rate and constellation size, not on the number of subcarriers. Therefore, single carrier and OFDM transmission systems have the same bandwidth requirements for the ADCs. In order to prevent high frequency distortion and noise components from being mirrored into the baseband signal oversampling is required [9]. A great advantage of OFDM with respect to a single carrier system is that it is straightforward to reduce the required oversampling to for instance a factor of 1.3. Oversampling in an OFDM system is typically realized by inserting unmodulated OFDM channels at high frequencies [5] and thus unlike single carrier systems no complex re-sampling is required at the receiver. However, the main disadvantage of digital OFDM is that DACs are required at the transmitter for the generation of the signal. The DACs increase the cost and complexity of the system, but at the same time make it also easier to adaptively scale to higher level modulation formats [1, 3].

From Eq. (4) it can be concluded that for a certain data rate and constellation size, the reach of an OFDM transmission system is dependent on the guard time (cyclic prefix), and the OFDM symbol size. For coherently detected OFDM systems the OFDM symbol size is typically limited by the effectiveness of the phase noise compensation scheme as well as the linewidth of the transmitter and local oscillator laser. For a conventional phase noise compensation scheme (using pilot subcarriers), this results in a maximum OFDM symbol length of about 10-15 ns [2] whereas with RF-aided phase noise compensation, an OFDM symbol length of longer than 100 ns can be realized [1]. The dispersion tolerance, expressed in km of SSMF, as a function of the guard time is shown in Fig. 1 for several OFDM symbol rates. This Figure clearly demonstrates the advantage of using long OFDM symbol lengths. For a system with a 10ns OFDM symbol length the reach is far below 1000-km SSMF whereas the OFDM system with 100ns symbol length reaches to more than 5000km.

t = 100ns t = 50ns t = 10ns

Fig. 1: Dispersion tolerance for 100GbE PDM-OFDM in one single band as a function of the guard time for several OFDM symbol lengths.

A method to further increase the dispersion tolerance is by using multi-band OFDM. Basically, the OFDM band is split up into several independently modulated OFDM bands. In [5] we showed that by using multiband OFDM the CP overhead can be significantly reduced. Fig. 2 shows the CP overhead required for 2,000-km SSMF as a function of the number of

Vol. 1 - 50

ECOC 2008, 21-25 September 2008, Brussels, Belgium

Mo.3.E.3

The use of multi-band OFDM as discussed in the previous section does not only provide a reduction of the CP overhead, in [5, 8] we have shown that the required bandwidth for both DACs and ADCs can be relaxed as well. This comes at the cost however of a more complex transmitter and receiver layout. Chromatic dispersion and PMD tolerance From (4) it can be concluded that the CD tolerance of an OFDM transmission system is dependent on the bandwidth of the signal (which scales with the nominal data rate) and the allocated CP. In an OFDM transmission system practically arbitrary amounts of CD can be compensated for as long as the OFDM symbol length is not restricted and sufficient CP is allocated. Basically, the increase in CP causes a rise of the CP overhead, which can be mitigated by choosing a long enough OFDM symbol length. In single carrier systems, the amount of dispersion tolerance is dependent on the number of taps that is used in the frequency domain compensation [7]. This value scales with the dispersion that is to be compensated for. Thus the receiver complexity scales with dispersion for single carrier systems, whereas the overhead increases in an OFDM system. This holds for the PMD tolerance as well, although PMD compensation is less stringent for a coherent receiver as it generally has a much shorter impulse response. In coherent detected transmission systems, PMD and CD provide a combined penalty [7, 8]. PMD however, causes depolarization as well, thus in order to realize a large PMD tolerance a polarization diverse receiver is required. It has been shown for that both single carrier [7] and multi-carrier (OFDM) [8] systems can offer a practically unlimited tolerance towards PMD. Narrowband filtering tolerance OFDM is known to have a well defined optical spectrum. Because of this narrow optical spectrum, it is often stated that the obtainable spectral efficiency of an OFDM signal is higher than that of its single carrier counterpart [4]. But from information theory we know that this is not per definition true [10]. In [11] we have investigated the narrowband filtering tolerance of 120-Gb/s polarization division multiplexed (PDM) OFDM. The simulation results are shown in Fig. 3. In this simulation the optical signal-to-noise ratio (OSNR) is fixed to 10.6-dB, resulting in a BER of -3 1x10 . Because of the confined spectrum of the OFDM signal a negligible BER penalty is observed as long as the optical filter is broader than the OFDM band (30-GHz for 120-Gb/s PDM-OFDM). At narrower bandwidths the subcarriers located on the sides of the OFDM signal are attenuated and thereby the SNRs of the subcarriers located at these frequencies are severely degraded. As a result, a steep increase in BER is observed when the filter bandwidth is lower than 30 GHz. The 30 GHz filter

tolerance is comparable to that observed for single carrier 100GbE [12]. It can thus be concluded that even though the unfiltered spectrum of PDM-OFDM is significantly smaller, it has a similar tolerance towards narrowband optical filtering compared to single carrier PDM-modulated signals with coherent detection (for the same constellation size M).
Spectral width of 120Gb/s PDM-OFDM with QPSK

Fig. 3: BER as a function of optical filtering bandwidth of 120-Gb/s PDM-OFDM.

Nonlinear tolerance Probably the main disadvantage of OFDM is that compared to single carrier systems the peak to average power ratio (PAPR) is significantly higher. One would therefore expect the nonlinear tolerance of OFDM to be lower resulting in a limited reach. However, when comparing the reach on a transmission link without DCF of 40 Gb/s PDM-OFDM [8] with that of single carrier 40-Gb/s QPSK [13], a similar reach is observed. The main reason for this is that although single carrier systems have lower PAPR in back-to-back configuration, the uncompensated chromatic dispersion along the transmission link causes the PAPR of single carrier systems to rise to similar values of OFDM systems, resulting in a similar nonlinear tolerance. Recently, we have investigated the impact of nonlinear impairments in a periodic compensated dispersion map [14]. These simulations showed that the nonlinear tolerance of OFDM is significantly reduced in periodically compensated dispersion maps. Fig. 4 shows the required OSNR after -3 1,200-km transmission for a BER of 10 as a function of the launch power. Simulated is one polarization of the 40-Gb/s PDM-OFDM experiment with the OFDM configuration as reported in [8]. In order to evaluate the influence of SPM and XPM both single channel and DWDM (50-GHz channel spacing) are simulated. Three dispersion maps are compared, namely without dispersion compensation, with fully periodic compensation and with a dispersion map that is optimal for 10-Gb/s OOK systems. The highest nonlinear tolerance is observed for the dispersion map without inline dispersion compensation. For the single channel simulations, the maximum tolerable launch power for a 1-dB OSNR penalty is -3.1 dBm, -5.7 dBm and -5.8 dBm for the dispersion map without DCF, the fully periodic and the 10G optimized dispersion map (10G OPT), respectively. Thus for both periodically compensated dispersion

978-1-4244-2229-6/08/$25.00 2008 IEEE

Vol. 1 - 51

ECOC 2008, 21-25 September 2008, Brussels, Belgium

Mo.3.E.3

maps an SPM induced launch power penalty of about 2.6 dB is observed with respect to the dispersion map without inline DCF. In the WDM simulations, a similar behavior with respect to the individual dispersion maps is observed. For the dispersion map without inline DCF, the XPMpenalty is only 0.8 dB, whereas for the fully periodic and 10G Opt dispersion maps the XPM penalties are 2.9 dB and 1.9-dB, respectively. It can thus be concluded that both SPM and XPM penalties are significantly higher for periodic compensated dispersion maps
Dispersion maps No inline DCF Fully Periodic 10G OPT No inline DCF Fully Periodic 10G OPT

Singel channel WDM

dispersion will cause the PAPR to rise after transmission. However, in a periodic dispersion map, the chromatic dispersion is compensated for after each span and thus the waveform is returned to its initial form in the high power regions. In this case, reducing the PAPR at the transmitter might improve the nonlinear tolerance significantly. For many vendors and network providers, backward compatibility is crucial. All OFDM transmission experiments reported so far are realized without optical dispersion compensation as the dispersion can be compensated for in the electrical domain by choosing an appropriate cyclic prefix. For green field deployments this would result in a benefit, as the system configuration is simplified. However, existing 10 Gb/s DWDM networks usually employ periodic inline dispersion compensation and when a 40-Gb/s or 100-Gb/s OFDM channel would be used to upgrade such a link, a periodic dispersion map is inevitable for the OFDM signal. Conclusions In this paper we have discussed the design of a 100GbE PDM-OFDM system and compared the performance of PDM-OFDM to that of PDM-QPSK. PDM-OFDM can be easily scaled to higher constellation sizes and allows for flexible oversampling rates, but because of its complex transmitter and limited performance on periodically compensated dispersion maps it is unlikely that OFDM will be a viable alternative for 100GbE systems in the near future. Acknowledgment The authors would like to thank Dr. T.C.W. Schenk and Mr. M.S. Alfiad for the many fruitful discussions and Dr. S. Akiba, Dr. M. Suzuki for their support. This work was partly supported by a project of the National Institute of Information and Communications Technology of Japan. References
1 S.L. Jansen, et al., in proc. OFC 2008, PDP 2. 2 Q.Yang, et al., in proc. OFC 2008, PDP 7. 3 B.J.C. Schmidt, et al., in proc. OFC 2007 PDP 18. 4 E. Yamada, et al., in proc. OFC 2008, PDP 8. 5 S.L. Jansen, et al., JLT, Vol. 26, pp. 6-15, 2008. 6 S.L. Jansen, et al., in proc. ECOC 2007, Tu 5.2.2. 7 D. van den Borne, et al., in proc OFC 8 S.L. Jansen, et al., OSA-JON, Vol. 7, pp. 173-182, 2008 9 E. Ip, et al., JLT, Vol 25, pp. 2033-2043, 2008. 10 C.E. Shannon, Bell Syst. Tech. J. 27, 379, 1948 11 S.L. Jansen, et al. in proc. Sum. Top., 2008, WD2.3. 12 G. Charlet, et al., in proc. OFC 2008, PDP 3 13 H. Sun et al. Optics Express, vol. 16, pp. 873-879, 2008 14 K. Forozesh, et al. in proc. Sum. Top., 2008, WC2.4. 15 A.J. Lowery, Opt. Expr., vol. 15, pp. 12965-12970, 2007. 16 B. Goebel, et al., in proc. OFC 2008, JWA58.

Fig. 4: OSNR penalties (BER of 10 ) for single channel and WDM transmission over 1,200 km.

-3

By fully compensating for the chromatic dispersion after each span, all first-order nonlinear perturbations are identical. In the 10G Optimized case, the undercompensation is relatively small and thus the perturbations are still highly correlated. However, when all DCF is removed from the transmission line, all perturbations are completely uncorrelated and average out. Because of this averaging an improvement in nonlinear tolerance is observed. In principle this holds for single carrier systems as well, but in this case the periodic dispersion map is preferred as the peak-to-average power ratio (PAPR) of most single carrier systems is significantly lower for low dispersion values. Thus, by periodically compensating for the chromatic dispersion, the nonlinear regions are mapped to low dispersion values where the PAPR is low. Additionally, in a periodic dispersion map there is some amount of nonlinear impairments that is compensated for by chromatic dispersion (a soliton-like behaviour). A possible method to improve the nonlinear tolerance of OFDM for periodic dispersion maps is to apply preand post compensation as proposed in [15]. It has been observed that for dispersion maps without dispersion compensation the effect of pre- and post compensation is limited, especially for high dispersive fibers. However, as the nonlinear regions in a periodic dispersion map fall all in the same dispersion window the efficiency of this method is greatly improved. Alternatively the nonlinear performance can be improved by reducing the PAPR at the transmitter [16]. Similar to pre- and post compensation, the effect of PAPR reduction is limited in a dispersion map without dispersion compensation, as chromatic

Vol. 1 - 52