Tradeoffs overcome degradation from crossconnect cascades

Jul 1st, 2001
0701feat07 1

Maziar Farhamand and Dogan Atlas

Prohibitively high losses and noise can accumulate from linked switching nodes in an optical network. The allowable number of nodes can be increased by using lower-noise-figure amplifiers or lower-loss fiber spans.

Multiple optical beams in an all-optical network can potentially pass through a large number of wavelength switching nodes (WSNs). A WSN typically integrates an optical crossconnect (OXC), multiplexer-demultiplexer (mux/demux) pair, and input/output optical amplifiers.

The filtering process imposed by cascaded WSN nodes results in implications for filter, switch, and amplifier performance, which must be understood to minimize optical-signal degradation. For example, insufficient filter bandwidth and/or improper filter shape, when combined with laser wavelength drift, can remove important frequency components of the modulated optical spectrum and cause pattern-dependent eye distortion. Therefore, the composite transfer function shape and bandwidth of the WSN should be designed based on an analysis of system links.

To optimize the system design, we have investigated the impact of WSN cascades due to accumulated amplifier noise and bandwidth narrowing. Bandwidth narrowing can cause attenuation and signal spectrum distortion and lead to unacceptable penalties. Bandwidth narrowing combined with laser wavelength drift can impose even more attenuation, spectral distortion, and eye-closure penalty. On top of this, noise accumulation degrades the optical signal-to-noise ratio (OSNR) and can limit the number of cascaded WSNs in a light path.

In a typical WSN, the mux/demux pair enables the WSN to switch at the wavelength level, and the optical amplifier—usually an erbium-doped fiber amplifier (EDFA)—compensates for the losses imposed by the OXC and the mux-demux pair (see Fig. 1). The WSN may also have means of adding or dropping optical signals at the wavelength level for purposes such as wavelength translation.

The transfer function of a WSN is typically dominated by the multiplexer and the demultiplexer filters. The two most common filter types are Gaussian shape and flattop-shape filters. The flattop filter can be modeled using a second-order super-Gaussian transfer function. In our model, the full-width-half-maximum (FWHM) of both Gaussian and flattop filters is 25 Ghz (see Fig. 2). In this case, the cascade bandwidth decreases more rapidly for Gaussian filters than flattop filters. This effect is also confirmed by comparing the FWHM of a cascade of Gaussian and flattop filters vs. the number of filters in the cascade (see Fig. 3).

The wavelength of each optical channel is initially set by adjusting the chip temperature of the corresponding lasers. However, the laser wavelength can still drift because of extreme ambient temperatures and/or the aging process of the laser module. Telecom grade laser modules may have wavelength drift range of approximately ±100 pm from their initially set wavelength. The wavelength drift combined with cascaded WSNs can impose an eye closure penalty in all optical networks.

WDM lasers can be either directly modulated or externally modulated. Directly modulated lasers are typically used for bit rates up to 2.5 Gbit/s (OC-48). At higher bit rates the fiber group velocity dispersion, aggravated by the chirp of the directly modulated laser, limits uncompensated transmission distance to a few kilometers. Therefore, at higher bit rates, such as 10 Gbit/s (OC-192), an externally modulated laser is preferable.

Figure 4 shows the eye closure penalty of a wavelength-drifted light signal after passing through the Gaussian and flattop filters of Figure 2. Each of the two lasers passes through a Gaussian or a flattop filter with FWHM of 25 GHz.

Although the directly modulated laser is modulated at a lower rate, the output spectrum is comparable to that of an externally modulated 10-Gbit/s laser because the chirp associated with direct modulation broadens the spectrum of the output signal. In both the directly modulated 2.5-Gbit/s and externally modulated 10-Gbit/s cases, the eye-closure penalty is initially lower for the flattop filter shape. However, at the extreme of 100-pm drift, the eye-closure penalty becomes unacceptably large for both cases. Therefore, laser wavelength locking is required. Wavelength locking can be accomplished by either adding external lockers or integrated lockers that are copackaged with the laser. Typically the maximum wavelength drift of a locked laser is less than ±20 pm.

In all-optical networks a light signal can undergo many filtering processes imposed by WSNs. This will cause even greater eye-closure penalty, to the extent that even locked lasers may not be enough to compensate, and a redesign of the WSN passband and its transfer function shape may be necessary.

Figure 5 shows the eye-closure penalty for locked directly modulated and externally modulated lasers passing through a cascade of WSNs. Since the lasers are locked, a worst-case wavelength drift of 20 pm is assumed.

Even with locked lasers the penalty caused by a cascade of Gaussian-shape filters is excessive. This problem can be avoided by proper design of the shape of the transfer function. As shown in the figure, a cascade of flattop-shape filters imposes a considerably lower eye-closure penalty than that imposed by a cascade of Gaussian-shape filters. If this penalty is still more than the tolerable penalty set by specific design rules for the system, then a broader passband should be employed.

Design analyses of all-optical networks should take into account the OSNR degradation caused by the noise introduced by the amplifiers associated with a cascade of WSNs. Optical amplifiers such as EDFAs compensate for the loss of the mux/demux pair and the OXC and reduce the overall insertion loss of the WSN. However, they also introduce noise and therefore degrade the OSNR.

The amount of OSNR degradation depends on the noise figure, gain, and the number of cascaded EDFAs. As illustration, we considered a typical light path in an all-optical network. The light path is assumed to contain a number, N, of equally spaced WSNs separated by optical fiber links. We further assumed that the net insertion loss of the WSN is zero, the EDFA associated with each WSN has a 20-dB gain, each two WSNs are connected by a fiber link, and the loss of each fiber link is completely compensated by an associated EDFA.

Figure 6, left, shows the calculated OSNR vs. the number of WSNs in the cascade for this system. We assumed that all the EDFAs are equivalent and have the same noise figure. We also assumed that the link span amounts to 20 dB of loss. The three curves in this figure correspond to cases where the EDFA noise figure is 5, 6, or 7 dB.

As expected, the OSNR degrades as the number of the WSNs increase. We also observed that larger numbers of WSNs can be traversed if lower-noise amplifiers are used. Knowing the minimum tolerable OSNR, the designer can calculate the number of WSNs that can be cascaded to form an optical-path—in other words, the number of WSNs that can be traversed before electrical signal regeneration is required.

The OSNR also depends on the link span between OXCs and improves for shorter spans. Figure 6, right, shows the OSNR versus number of WSNs for three different spans of 10, 15, and 20 dB. For each curve, the span loss is constant and completely compensated by an associated EDFA, the gain of which is equal to the span loss. Both the EDFAs associated with the links and the EDFAs associated with the WSNs are assumed to have a noise figure of 6 dB.

The noise accumulates more slowly in lower-loss spans and therefore, as shown in Figure 6, right, OSNR is higher in the case of lower-loss spans. Consequently, a larger number of cascaded WSNs can be allowed in the light path if lower-loss spans are used.

The impact of WSN cascading is an important consideration in the design of all-optical networks. In particular, WSN cascading results in passband narrowing and OSNR degradation.

As we have shown, laser wavelength drift can increase the penalties caused by bandwidth narrowing. To avoid such penalties, the drift of the laser wavelength should be minimized using wavelength-locking methods. In addition, the passband and the shape of the WSN composite transfer function should be properly designed such that cascaded WSNs can accommodate the allowable laser wavelength drift.

The OSNR degradation is caused by noise accumulation and increases with the number of WSNs in the cascade. The minimum acceptable OSNR sets an upper limit on the number of allowable WSNs in a light path. Consequently, the allowable number of WSNs can be increased by using lower-noise-figure amplifiers and/or lower-loss spans.

Maziar Farhamand is a senior engineer and Dogan Atlas is the vice president of optical transport engineering at Movaz Networks, One Technology Parkway South, Norcross, GA 30092. They can be contacted at 678-728-8600 or at and

More in DWDM & ROADM