Wavelength-selective switches (WSS) are crucial elements in a 40G network, but degrade signal quality. Measurements show crosstalk can increase the BER by an order of magnitude, which is equivalent to additional four WSS. However, when the WSS are replaced with a tunable alternative, more than 1.5-dB/0.1-nm improvement in OSNR results.
By M. Jordan, E. Granot, M. Caspi, Y. Stav, N. Narkiss, M. Roelens, S. Frisken, S. Poole, J. Leuthold, and S. Ben-Ezra
The continuing growth of IP traffic challenges the modern network. On one hand, there is a need to make the communication channels as spectrally efficient as possible, and on the other hand the network has to be flexible and versatile.
The spectrally efficient requirement is expressed in adopting DWDM technology, which means that the spectral distance between adjacent channels is approximately 50 GHz. The flexibility requirement suggests that every such channel can be added or dropped without considerably affecting the signal or its adjacent channels. When the data rate is ~43 Gbps this is a real challenge, since part of the spectrum penetrates the neighboring channels. [1-3]
To enable transmission of 43-Gbps signals through cascaded wavelength-selective switches (WSS) designed for 50-GHz spaced DWDM networks, a careful engineering of the WSS as well as the 43G differential phase-shift keying (DPSK) signal is required both at the transmitting and the receiving ends. 
In this paper we experimentally investigate the compliance of the 43-Gbps transponder to the DWDM technology by measuring the bit-error-rate (BER) degradation of transmitted signals due to cascaded WSS and crosstalk from adjacent channels.
Experimental setup and results
Figure 1 shows the experimental loop schematic. The loop consists of Finisar's 43-Gbps DPSK transponder (transmitter and receiver), WSS, and fibers. The loop allows us to emulate a real link with three EDFAs, filters, 80 km of SMF28 fiber including 12 km of dispersion compensating fibers (DCFs) and two variable optical attenuators (VOAs) for loop power balancing.
Figure 1. Loop system schematic.
In the experiment, the modulation spectral bandwidth at FWHM was 27 GHz while the free spectral range (FSR) of the delay interferometer (DI) was 50 GHz.
When analyzing the effective distance we ignored the DCFs' length, since it is usually concentrated in specific places.
Figure 2 presents the experimental results. From the figure, one can see that only after 800 km and 20 WSS the BER exceeds 10-9, and only after additional 800 km and 40 WSS the BER reaches the forward error correction (FEC) limit at 2x10-3. This indicates that the relatively narrow modulator's spectral bandwidth makes the signal less susceptible against marginal dispersion effects. This result is considerably better than Ref.5.
Figure 2. The BER deterioration with length and number of WSS.
Crosstalk and BER
Another challenge of the existing DPSK and WSS technologies in complying with DWDM technology requirements is the reduction of crosstalk influence. To measure crosstalk, we sent a stream of data at 43 Gbps in a 50-GHz channel. We then populated its two adjacent channels and took the BER measurements for the middle channel only. Since the spectrum of the signal is spread over a spectral range that is larger than the filter's bandpass, crosstalk between the channels is inevitable.
In Figure 3 we plot the BER measurements of several scenarios. As illustrated, 100-GHz spacing always exhibits higher performance than 50 GHz. This is not a surprise but it shows that there is a crosstalk effect, which is relatively minor at less than 1dB/0.1nm OSNR penalty.
Figure 3. BER measurement as a function of OSNR for several scenarios: Back-to-Back (filled star); 100-GHz spacing, with three neighboring channels with 43 Gbps (empty circle); 100-GHz spacing, with three neighboring channels with 43 Gbps but with a WaveShaper instead of WSS (filled circle); 50-GHz spacing, with three neighboring channels with 43 Gbps (small circle); 50-GHz spacing, with three neighboring channels with 43 Gbps but with a WaveShaper instead of WSS (half-filled diamond); 50-GHz spacing, with two neighboring channels populated with 10-Gbps NRZ (triangle); 50-GHz spacing, with two neighboring channels with 10 Gbps but with a WaveShaper instead of a WSS (half-filled rectangle).
On the other hand, a somewhat surprising result is that there is a negligible difference between neighboring channels that are populated with 43 Gbps and 10 Gbps. (For a numerical study of mixed duobinary WDM channels see Ref.6; for the cross-phase modulation penalty of 10-Gbps NRZ signal on a 40-Gbps DPSK signal, see Ref.7 It should be stressed that in this paper ”43 Gbps” indicates DPSK signals, while ”10 Gbps” indicates NRZ signals. This result is important since it demonstrates the compliance of the transponder to a mix of 10-Gbps and 40-Gbps networks.) Perhaps this minor difference is due to the fact that 50-GHz spacing is the fifth harmonic of the 10 Gbps, and in the 43-Gbps case, the channel spectral spacing falls approximately on the valley of the spectrum. This result is consistent with the tendency of Ref.5 although our results are substantially better.
The main improvement in BER occured when the WSS was replaced with a WaveShaper . More than 1-dB/0.1-nm improvement in OSNR was gained -- in some cases even 1.5 dB/0.1 nm.
The WaveShaper is an amplitude- and phase-tunable add/drop filter. Every 50-GHz spectral band can be manipulated. Every channel consists of seven spectral segments. The amplitude and phase of each one of these segments can be tuned. Thus, the phase and amplitude spectra of the filter can be determined almost arbitrarily with a spectral resolution of approximately 7 GHz.
When the WaveShaper is placed inside the loop (instead of ordinary WSS) its parameters can be tuned to improve the loop's performance. There are many spectral deformations in the loop that the filter can fix; however, it seems that the main deformations are the marginal effects of dispersion, which were not fixed by the DCF.
Figure 4 presents the BER performance as a function of the number of WSS.
Figure 4. BER as a function of WSS number for different scenarios: 50-GHz spacing, with three adjacent channels with 43 Gbps (half-filled circle); 50-GHz spacing, with three adjacent channels with 43 Gbps but when the WSS is replaced with a WaveShaper (full circle); 100-GHz spacing, with three adjacent channels with 43 Gbps (filled triangle); 50-GHz spacing, with three adjacent channels with 43 Gbps but with a WaveShaper instead of a WSS (full rectangle); 50-GHz spacing, when the neighboring channels carry a 10-Gbps NRZ signal (half-filled diamond); 50-GHz spacing, when the neighboring channels carry a 10-Gbps NRZ signal, but with a WaveShaper instead of a WSS (filled star); 50-GHz spacing, with four populated neighbors: 43-10-43-10-43 Gbps (half-filled rectangle); 50-GHz spacing, with four populated neighbors: 43-10-43-10-43 Gbps with a WSS replaced with a WaveShaper (full diamond).
In all cases we see that when using fewer than 18 WSS the BER is better than . When the spacing changes from 100 GHz to 50 GHz the BER is degraded by an order of magnitude, which is equivalent to four additional WSS.
We also learn that only neighboring channels have an influence on the overall performance. That is, the following two scenarios yield the same BER: a) when three neighboring channels are populated with 10 Gbps, 43 Gbps and 10 Gbps signals, respectively, and b) when five neighboring channels are populated with 43 Gbps, 10 Gbps, 43 Gbps, 10 Gbps and 43 Gbps signals, respectively. In both cases the central channels were measured, and the fact that the two scenarios demonstrated similar BER indicates that only the closest neighbors have a substantial effect on the crosstalk.
As can be learned from Figure 3, there isn't a substantial difference between the cases where the neighboring channels are 43 Gbps or 10 Gbps. Nevertheless, as was explained above, the 43-Gbps neighbors show lower crosstalk and smaller, albeit less than half an order of magnitude, BER penalty.
 Tongyu Song, Hanyi Zhang, Yili Guo, Xiaoping Zheng, Optics Commun., "Statistical study of crosstalk accumulation in WDM optical network using different RWA" 202, 131-128 (2002).
 Mary R. Phillips and Daniel M. Ott, "Crosstalk Due to Optical Fiber Nonlinearities in WDM CATV Lightwave Systems," J. Lightwave Technol. 17, 1782- (1999).
 M. Kalyvas, C. Bintjas, H. Avramopoulos,and A. Boskovic, "Experimental and Theoretical Investigation of Nonlinear-Crosstalk in SCM-WDM CATV Systems," Optical and Quantum Electronics 36, 413-430 (2004).
 M. Zaacks, U. Mahlab, P. Mamyshev, C. Rasmussen, J. Calvitti, and K. Falta, "Demonstration of 1000km 43 Gbps RZ-DPSK Transmission through a 50GHz Channel Spaced WSS," OFC/NFOEC (2007).
 A. Tan, and E. Pincemin, "Performance Comparison of Duobinary Formats for 40-Gbps and Mixed 10/40-Gbps Long-Haul WDM Transmission on SSMF and LEAF Fibers," J. LightWave Technol. 27, 396 (2009).
 X.Liu and S. Chandrasekhar, "Suppression of XPM Penalty on 40-Gbps DQPSK Resulting from 10-Gbps OOK Channels by Dispersion Management," paper OMQ6, OFC/NFOFC 2008.
 M. Jordan (1,2), E. Granot (1,3), M. Caspi (1), Y. Stav (1), N. Narkiss (1), M. Roelens(4),, S. Frisken(4), S. Poole(4), J.Leuthold (2), and S. Ben-Ezra, "Enhancement of 43 Gbps DPSK Transmission Though 66 Wavelength Selective Switches Using Adaptive Channel Shape Optimization," to be published in ECOC 2009.
M. Jordan works for Finisar Israel, Nez Ziona, Israel, and teaches at the Institute of Photonics and Quantum Electronics, University of Karlsruhe, Germany.
E. Granot works for Finisar Israel, Nez Ziona, Israel, and teaches at the Department of Electrical and Electronics Engineering, Ariel University Center of Samaria, Ariel, Israel.
M. Caspi, Y. Stav, N. Narkiss, and S. Ben-Ezra work for Finisar Israel, Nez Ziona, Israel.
M. Roelens, S. Frisken, and S. Poole work for Finisar Australia, Sydney, Australia.
J. Leuthold teaches at the Institute of Photonics and Quantum Electronics, University of Karlsruhe, Germany.