Sébastien Bigo
Alcatel Research and Innovation, Route de Nozay, 91460, Marcoussis, France
Ph.: +33 (0)1 69631478, fax: 0169631865, email: Sebastien.bigo@alcatel.fr
Abstract: In the history of WDM, experiments at 40Gbit/s channel rate have outperformed that at lower channel rates in term of capacity but not yet in term of capacity times distance. Ultra-narrow optical filtering may lead them one step further.
Out of the history of hero WDM experiments, we have selected the laboratory demonstrations with 10, 20 and 40Gbit/s channel rates complying with the basic requirements of existing terrestrial networks, i.e. incorporating fibers of a single type within repeaters and relatively large repeater spacings (in excess of 65km from 1996 on). In addition, we made a selection of the most eventful single-channel experiments reported since 1993, with bit-rates equal or larger than 80Gbit/s, whether they emulate terrestrial systems or not. Figure 1 represents the demonstrated capacities in both WDM and single-channel versus time. In both cases, two lines materialize the overall best capacities at a given date. Whereas the record single-channel rates grow exponentially at a somewhat regular pace, doubling every 24 months, WDM capacities have evolved in a more erratic manner, but clearly faster than single-channel capacities. At the beginning of the year 2001, they lay more than a decade beyond. Though, the single-channel terabit/s barrier was broken by the turn of the century by M. Nakazawa et al [1].
|
Figure 1: Evolution of capacities in WDM terrestrial laboratory experiments, and hero singe-channel experiments. |
In the early years of WDM, up to the year 1996, most laboratory experiments with record-breaking capacities were conducted at 20Gbit/s channel bit-rate. Terabit/s capacities were reported for the first time by three different teams in 1996 not only with 20Gbit/s [2, 3] but also 100Gbit/s channel rate [4]. However, all these experiments had been conducted over limited distances. The compliance with the basic requirements of terrestrial links in terms of distance was the focus of the next two years. In this consolidation phase, a renewed interest towards systems with 10Gbit/s rate was clearly observed. Their capacity caught up with that of 20Gbit/s systems, as the latter turned handicapped by their incompatibility with SDH transmission standards. Later on, from the year 1998 to now, the improvement of electronics made possible the rapid development of laboratory demonstrations based on 40Gbit/s rate, driving total capacity upwards to 10Tbit/s [5,6], but over limited distances. It can be predicted that a consolidation phase similar to that which followed the pioneer terabit/s transmission experiments will follow within the next years, with likely focus on the improvement of the transmission distance. In ’mary, the results in Figure 1 tend to illustrate that higher bit-rates per channel are more amenable to larger capacities. Whether this trend will continue faces the stumbling block of the fundamental limits of spectral efficiency. Increasing the channel rate from 10Gbit/s to 40Gbit/s has been beneficial in terms of capacity only because researchers have managed to pack any two channels at 40Gbit/s into less optical bandwidth than eight channels at 10Gbit/s. Nonetheless, spectral efficiencies equal or higher than 1bit/s/Hz have already been demonstrated [5], which leaves a limited margin for further improvement in this field.
Besides, systems with a high spectral efficiency are more sensitive to linear and non-linear cross-talk, which ultimately limits the maximum achievable error-free distance. This remark must be weighted in the prospects of ultra-long haul terrestrial systems spanning over several thousands of kilometers. Such systems have recently been requested by operators in order to foresee the future implementation of all-optical networks. Not surprisingly, at a given date, the overall capacities achieved in these ultra-long haul hero experiments are at best equal to that achieved over long-haul distances. Hence, a conventional way to rank transmission experiments, whether they focus on distance or on capacity, has been to use the capacity times distance product CxD. We have compiled the CxDs and depicted them in Figure 2, varying symbols whenever the bit-rate per channel is changed.
 |
Figure 2: Evolution of capacity times distance product of terrestrial WDM lab experiments. White symbols correspond to distances larger than 1000km. |
The evolution of the CxD in Fig. 2 clearly contrasts with that of the capacity in Figure 1. While hero results obtained with channel rates of 80Gbit/s and above, regardless their compatibility with a terrestrial environment still lay behind, the CxD of laboratory experiments at 10Gbit/s and 20Gbit/s rates have grown exponentially in the past years, at the same, regular, pace for both rates. This clearly illustrates the back-and-forth shift of focus from capacity to distance, which has paved the history of WDM. The hero demonstrations at 40Gbit/s rate came out later, but rapidly caught up with that of 10 and 20Gbit/s rates, doubling their CxD every 7 months. As of now (2001), they have outperformed the others in terms of capacity but are only just matching in terms of CxD.
Whether they will also ultimately exhibit significantly larger CxDs than 10Gbit/s and 20Gbit/s solutions may rely on the possibilities offered by reshaping optical filters. Indeed, spectral reshaping is facilitated at 40Gbit/s by the larger channel bandwidth, within access of existing filter technologies. In particular, we recently proposed a new wavelength allocation scheme in combination with ultra-narrow, off-center, optical filtering at the receiver to demultiplex WDM channels at 40Gbit/s rate. With this scheme and regular NRZ format, we demonstrated the transmission of 128 channels at 40Gbit/s (5.1Tbit/s total capacity) over 3x100km of TeraLightTM fibre [7]. Next, we theoretically investigate theoretically (remark: to have a destingtion to the VSB paper) the tradeoffs involved in the filtering technique.
In our experiment, we used a non-periodical frequency grid for our 128 WDM channels at 40Gbit/s, i.e. they are alternatively spaced by a GHz and b GHz, as schematized in the inset of Fig. 3. To achieve 5Tbit/s capacity over C and L bands (8THz total bandwidth), the condition a+b=125GHz has to be met. We consider next three channel spacing ratios, i.e. (a/b)=(40GHz/85GHz), (62.5GHz/62.5GHz) and the ratio used in the 5Tbit/s experiment (50GHz/75GHz). Consider a Gaussian filter with variable bandwidth from 0.2 to 0.8nm. This filter is detuned off the carrier of the channel under test, always towards the farther of the two neighboring channels, realizing a so-called vestigial side band (VSB) filtering. Thus, the side-band experiencing the smallest overlap with adjacent channels is isolated while the impact of the cross-talk affecting the other side-band is minimized. In Figure 3, we show the eye-opening Q’ in back-to-back. The best performance is a trade-off between crosstalk and filter bandwidth limitations. Whatever (a/b), a centered filter never yields the best performance. A narrow filter strongly attenuates high frequencies components of both side-bands apart the carrier, whereas a larger one does not provide sufficient cross-talk suppression. Conversely, detuning the filter off the carrier enhances the higher frequencies of the favored side-band, while making very narrow filtering possible, and thereby efficient reduction of cross-talk. This conclusion is evident when the channel spacing ratio (a/b) differs from unity (62.5GHz/62.5GHz), down to configurations where the main limitation comes from the cross-talk with the closer of the two neighboring channels (e.g. 40/85GHz). The optimal Q’ eye-opening as a function of the channel spacing ratio, as derived from Fig. 3. The channel allocation plan of [5,7], i.e. (a/b)=(50GHz/75GHz), appears as reasonable trade-off. In this configuration, the optimal filter bandwidth nears 30GHz, as actually used in [7].
 |
Figure 3: Interest of using an optical filter tuned off the channel carrier frequency to demultiplex 40Gbit/s |
In summary, we have shown that, to reach 5Tbit/s capacity over C+L bands, an asymmetric (50GHz/75GHz) wavelength allocation scheme provides optimal performance when associated with a narrow detuned filter of 30GHz bandwidth, as demonstrated over 3x100km.