Wavelength converters switch the entire bitstream at one wavelength to a different wavelength without affecting its temporal content. Some applications require selective switching of one or more bits to a different port. An example is provided by packet switching in which a packet of tens or hundreds of bits is selected from a bitstream. Another example is provided by the OTDM technique in which a selected bit of a high-speed bit stream is periodically sent to another port. Such applications require time-domain switches that can be turned on for a specific duration using external control.
1. Time-Domain Demultiplexing
As seen in the Optical Time-Division Multiplexing tutorial, an OTDM signal consists of a high-speed bitstream that is composed of several channels, each operating at a lower bitrate and interwoven with others in a periodic fashion. If 10 channels, each operating at 40 Gb/s, are multiplexed in the time domain, every 10th bit of the composite 400-Gb/s bitstream belongs to the same channel. Demultiplexing a channel from such a high-speed OTDM signal requires optical switches that pick all the bits belonging to a specific channel and direct those bits to a different port. Such switches require an optical clock at the single-channel bit rate that is used to switch signal pulses selectively using a nonlinear phenomenon such as XPM or FWM.
In the optical time-division multiplexing tutorial, it shows how XPM inside a NOLM or FWM inside a highly nonlinear fiber can be exploited for time-domain demultiplexing of OTDM channels. The FWM technique was used as early as 1996 to demultiplexed 10-Gb/s channels from a 500-Gb/s bitstream using clock pulses of 1-ps duration. A distinct advantage of using FWM is that the demultiplexed channel is also amplified through parametric gain inside the same fiber. A problem with FWM-based demultiplexer is related to the polarization sensitivity of the FWM process itself because maximum parametric gain occurs only when the pump and signal are co-polarized. If the state of polarization of a signal is not aligned with that of the pump and changes with time in an unpredictable manner, both the signal and idler power levels fluctuate, resulting in poor performance.
A polarization-diversity technique, in which the input signal is separated into two orthogonally polarized parts that are processed separately, can be used, but it adds considerable complexity. A simple scheme for solving the polarization problem was adopted in 2004. It consists of attaching a short piece of polarization-maintaining fiber (PMF) to the input port of the highly nonlinear fiber used for FWM and using an optical phase-locked loop for locking clock pulses to the peak position of incoming signal pulses. As shown in the following figure, the control clock pulses are polarized at 45° with respect to the principal axes of the polarization-maintaining fiber that also splits and separates randomly polarized signal pulses into two orthogonally polarized parts. Since two separate FWM processes take place simultaneously within the same nonlinear fiber, in essence, polarization diversity is realized with this simple experimental arrangement. Such an approach was capable of demultiplexing a 160-Gb/s bitstream into 10-Gb/s individual channels with <0.5 dB polarization sensitivity.
In another approach to solving the polarization problem, the nonlinear fiber in which FWM occurs is itself made birefringent. Moreover, it is divided into two equal sections in which the fast and slow axes are reversed. A single pump in the form of clock pulses, polarized at 45° with respect to the slow axis of the fiber, is launched together with the high-speed signal that needs to be demultiplexed. The orthogonally polarized components of the pump and signal interact through FWM and create the idler containing the demultiplexed channel. Even though the two polarization components separate from each other in the first section, they are brought back together in the second half of the fiber because of the reversal of the slow and fast axes in the second section. An optical filter at the end of the fiber blocks the pumps and signal, resulting in the demultiplexed channel at the idler wavelength.
XPM-based demultiplexing based on NOLMs also suffers from the polarization issue. Several techniques can be employed for the polarization-insensitive operation of a NOLM. One among them is similar to that shown in the figure above. A short piece of PMF is used to split the signal and clock pulses along its slow and fast axes. An optical bandpass filter centered at the signal wavelength is placed at one end of the NOLM so that it blocks the propagation of the clock pulse in one direction. However, this blocking occurs in the other direction only after the clock pulse has passed through the loop and has changed the phase of a specific pulse by π through XPM. As a result, data pulses belonging to the demultiplexed channel appear at the NOLM output, where a second PMF combines the two polarization components.
Similar to the case of wavelength conversion, it is not necessary to employ a NOLM for making use of XPM. In a 2001 experiment, a scheme is similar to that shown in the following figure was used for time-domain demultiplexing. The only difference was that the role of the probe at the wavelength λ1 was played by the OTDM data signal, while intense clock pulses at the wavelength λ2 played the role of the pump. The clock pulses shifted the spectrum through XPM of only those data pulses that overlap with them in the time domain. An optical filter was then used to select these pulses, resulting in a demultiplexed channel at the clock wavelength. This experiment used a 5-km-long fiber with its zero-dispersion wavelength at 1543 nm. The 14-ps control pulses at a repetition rate of 10 GHz had a wavelength of 1534 nm and were propagated with the 80-Gb/s OTDM signal at 1538.5 nm.
As seen in the optical time-division multiplexing tutorial, the group-velocity mismatch between the signal and control pulses plays a major role in XPM-based optical switching. This mismatch can be reduced by locating the control and signal pulses on opposite sides of the zero-dispersion wavelength of the fiber. In addition, the use of highly nonlinear fiber not only reduces the required average power of control pulses but also helps with the problem of group-velocity mismatch as much shorter lengths are needed. An added benefit of this technique is that it can be used to demultiplex multiple channels simultaneously by simply employing multiple control pulses at different wavelengths. The following figure shows such a scheme schematically. It was implemented in a 2002 experiment to demultiplex four 10-Gb/s channels from a 40-Gb/s composite bitstream through XPM inside a 500-m-long highly nonlinear fiber. Only a 100-m-long fiber was employed in another experiment to demultiplex 10-Gb/s channels from a 160-Gb/s bitstream.
Much smaller fiber lengths can be employed by using microstructured fibers or a non-silica fiber made of a material with high values of n2. Only a 1-m-long piece of bismuth-oxide fiber was needed in a 2005 experiment because this fiber exhibited a value ~1100 W-1/km for the nonlinear parameter γ. The train of 3.5-ps control pulses at a 10-GHz repetition rate was amplified to an average power level close to 0.4 W to ensure a high peak power (P0 > 10 W) so that the value of γP0L exceeded 10 even for the 1-m-long fiber. This experiment employed the fiber as a Kerr shutter and made use of the XPM-induced nonlinear birefringence that changed the state of polarization of selected signal pulses such that only they were transmitted through the polarizer placed at the output end of the fiber. Because the walk-off effects were negligible for the short fiber, the measured switching window was narrow enough (only 2.6-ps wide) to demultiplexed a 160-Gb/s bitstream.
Polarization-independent operation can be realized by using a linearly birefringent PMF or a spun fiber exhibiting circular birefringence. A 30-m-long photonic crystal fiber exhibiting linear birefringence was employed in a 2006 experiment. Clock pulses were polarized at 45° to the slow axis of the fiber so that their energy was divided equally between the slow and fast axes. The SOP of both the data and clock pulses evolved periodically with different beat lengths because of their different wavelengths. As a result, their relative SOP varied in a nearly random fashion. This feature resulted in an averaging of the XPM effect and produced an output that was independent of the signal polarization. In a later experiment, a high-speed polarization scrambler was employed to randomize the SOP of 160-Gb.s data pulses but the SOP of the 10-Gb/s clock pulses was kept fixed. The XPM-induced spectral broadening occurred inside a 2-m-long bismuth-oxide fiber. Because of polarization scrambling, the performance of such a demultiplexer exhibited little sensitivity to the input SOP of the data bitstream.
The main limitation of a fiber-based demultiplexer stems from the weak fiber nonlinearity requiring long lengths. Although, the required fiber length can be reduced by using highly nonlinear fibers, the use of an SOA provides an alternative. Both the XPM and FWM schemes have been shown to work with SOAs. An electro-absorption modulator has also been used for demultiplexing purposes. In the case of a NOLM, an SOA is inserted within the fiber loop. The XPM-induced phase shift occurs because of changes in the refractive index induced by the clock pulses as they saturate the SOA gain. As the phase shift occurs selectively only for the data bits belonging to a specific channel, that channel is demultiplexed. The refractive-index change induced by the SOA is large enough that a relative phase shift of π can be induced at moderate power levels by an SOA of <1-mm length.
SOAs suffer from a relatively slow temporal response governed by the carrier lifetime (~100 ps). A faster response can be realized by using a gating scheme. For example, by placing an SOA asymmetrically within the NOLM such that the counter-propagating signals enter the SOA at different times, the device can be made to respond at a time scale ~1 ps. Such a device is referred to as the terahertz optical asymmetrical demultiplexer (TOAD). Its operation at bit rates as high as 250 Gb/s was demonstrated by 1994. An MZ interferometer with two SOAs in its two branches can also demultiplex an OTDM signal at high speeds and can be fabricated in the form of an integrated compact chip using the InP technology. The silica-on-silicon technology has also been used to make a compact MZ demultiplexer in a symmetric configuration that was capable of demultiplexing a 168-Gb/s signal. If the SOAs are placed in an asymmetric fashion, the device operates similar to a TOAD device. Figure (a) below shows such a MZ device fabricated with the InGaAsP/InP technology. The offset between the two SOAs plays a critical role in this device and is typically <1 mm.
The operating principle behind the MZ-TOAD device can be understood from the figure above. The clock signal (control) enters from port 3 of the MZ interferometer and is split into two branches. It enters the SOA1 first, saturates its gain, and opens the MZ switch through XPM-induced phase shift. A few picoseconds later, the SOA2 is saturated by the clock signal. The resulting phase shift closes the MZ switch. The duration of the switching window can be precisely controlled by the relative location of the two SOAs as shown in figure (b) above. Such a device is not limited by the carrier lifetime an can operate at high bit rates when designed properly.
Several other SOA-based schemes have been implemented in recent years. In a 2006 experiment, transient XPM, discussed earlier in the context of wavelength conversion, was used to demultiplex 40-Gb/s channels from a 320-Gb/s OTDM bitstream. This scheme employs an optical filter that is shifted from the clock wavelength by a suitable amount and works in a way identical to that of a wavelength converter. By 2007, it was extended to operate on a 640-Gb/s OTDM signal consisting of 0.8-ps-wide optical pulses. In a 2009 experiment, a symmetric MZ configuration, shown schematically in the following figure, was used for demultiplexing a 640-Gb/s bitstream. The clock pulses at a repetition rate of 40 Gb/s were injected in both arms with a relative delay of about 1.4 ps. As seen in the figure above, such a device can act as a gate that opens only for the duration of this relative delay despite of slow response of two SOAs.
2. Data-Format Conversion
Both the RZ and NRZ formats can be employed when used for data transmission. The NRZ format is often employed in WDM networks as it is most efficient spectrally. The use of RZ format, or one of its variants such as the carrier-suppressed RZ (CSRZ) format, becomes necessary at high bit rates, and it is the format of choice for OTDM systems. In a network environment, the conversion among these formats may become necessary. Several techniques for conversion between the NRZ and RZ formats make use of the nonlinear effects occurring inside fibers or SOAs.
The following figure shows how XPM inside a NOLM can be used for conversion between the NRZ and RZ formats. In the case of NRZ-to-RZ conversion, the phase of NRZ pulses is shifted inside the loop by launching an optical clock (a regular train of pulses at the bit rate) such that it propagates in one direction only. In the case of RZ-to-NRZ conversion, the phase of a CW beam is shifted by the RZ data pulses propagating in one direction only. The main limitation is set by the walk-off effects that govern the switching window of the NOLM. An SOA-based NOLM has also been used to convert the RZ or NRZ bitstream into one with the CSRZ format.
Several other schemes have been developed in recent years for fiber-based format conversion. In a 2005 experiment, the XPM-induced wavelength shift inside a nonlinear fiber was used for RZ-to-NRZ conversion. The scheme is similar to that shown in the figure below (in the context of wavelength conversion), the only difference being that the optical filter is centered exactly at the wavelength of the CW probe. The RZ signal acts as the pump and modulates the phase of the CW probe. The resulting chirp shifts the wavelength of pulses representing 1 bits. The filter blocks these pulses but lets pass the 0 bits. The resulting bitstream is a polarity-reversed NRZ version of the original RZ signal.
A similar scheme can be adopted for NRZ-to-RZ conversion. In this case, an optical clock acting as the pump is sent through the fiber together with the NRZ signal. The XPM interaction between the two broadens the signal spectrum. The optical filter is offset from the signal wavelength similar to the case of wavelength conversion. The output is an RZ version of the signal at the same wavelength. Such a scheme suffers from polarization sensitivity, because the nonlinear process of XPM itself is polarization-dependent. It can be made polarization-insensitive by employing a polarization-diversity loop. The polarization of the clock (control) is oriented at 45° with respect to the principal axes of the polarization beam splitter (PBS) so that its power is divided equally in two counterpropagating directions. The NRZ signal with a random SOP is also divided into two orthogonally polarized parts. The same PBS combines the two parts. An optical circulator directs the output toward an optical filter whose passband is offset properly.
A scheme for RZ-to-NRZ conversion makes use of only SPM-induced spectral broadening inside a normally dispersive optical fiber. The RZ pulses are chirped through SPM and undergo considerable broadening inside the fiber. If the fiber length is chosen such that this pulse broadening is large enough to fill the entire bit slot, the output is an NRZ version of the original bitstream.
Several schemes make use of the nonlinear effects inside SOAs for format conversion. A MZ interferometer with SOAs in its two arms was employed in a 2003 experiment. The following figure shows the underlying idea schematically. In the case of NRZ-to-RZ conversion, the input RZ signal is injected into the control port, while an RZ clock at the same bit rate is fed into the interferometer designed to block clock pulses in the absence of the control signal. The phase shift induced by the NRZ signal converts clock pulses into an RZ signal. In the case of RZ-to-NRZ conversion, a pulse duplicator is employed to make multiple shifted copies of the input RZ signal (within one bit period) before it is injected into the control port. Multiple copies maintain the XPM-induced phase shift over the entire bit duration converting a CW beam into an NRZ signal. An SOA inside a Sagnac loop placed with a fixed offset from the loop center can be used to create a fast switch in a way similar to that shown before. Such a loop was used in 2004 for converting the NRZ and RZ formats into the CSRZ format at a bit rate of 10 Gb/s. It can also be used for NRZ-to-RZ conversion.
Similar to the case of wavelength conversion, it is not necessary to use an interferometer. Format conversion can be realized using FWM, XPM, or cross-gain saturation inside a single SOA. These nonlinear processes can also be used to realize the conversion of RZ and NRZ formats to BPSK or DPSK formats. In some cases, SOAs can be operated at a bit rate of up to 40 Gb/s. For example, XPM inside a single SOA was used in a 2007 experiment to convert a 42.6-Gb/s NRZ signal to the RZ format. In another experiment, an optical filter was placed after the SOA to convert a 40-Gb/s NRZ signal such that its passband was offset from the signal wavelength by an optimum amount. A 40-Gb/s RZ signal can also be converted to the NRZ format using cross-gain saturation. The main point to note is that SOAs are quite useful for optical signal processing.
3. Packet Switching
Optical packet switching is a complicated process requiring many components for buffering, header processing, and switching. Packet-switched networks route information in the form of packets consisting of hundreds of bits. Each packet begins with a header that contains the destination information. When a packet arrives at a node, a router reads the header and sends it toward its destination. Considerable progress has been made in recent years in realizing such an all-optical router by using optical flip-flops and other time-domain switches.
The basic element of an optical router is a packet switch that can direct each incoming packet to different output ports depending on the information in the header. The following figure shows one implementation of such a packet switch. The optical power of the input packet is split into two branches using a directional coupler. One branch processes the header, while the other delivers the payload dan simply contains a fiber-delay line to compensate for the latency in the header branch. Between the header and the payload, a few 0 bits are inserted that serve as the guard time. The entire switch is composed of three units. The header-processing unit is a time-domain switch (e.g., a NOLM). The flip-flop memory unit is implemented using two coupled lasers that switch the output between two wavelengths, say, λ1 and λ2. The third unit is simply a wavelength converter; it converts the wavelength of the incoming data packet to the output wavelength of the flip-flop. With the use of a demultiplexer, the switch directs output at different wavelengths to different ports depending on the header information. By 2008, such a packet switch could be operated at a bit rate of 160 Gb/s using a flip-flop design similar to that shown in the previous tutorial. Both the flip-flop and the wavelength converter employed SOAs in this experiment.
In another packet-switching scheme, a single DFB laser was used as an optical flip-flop capable of switching between its low- and high-power states by injecting the set and reset pulses. In this case, multiple DFB lasers operating at different wavelengths can be used simultaneously such that a set pulse from the header processor turns on a specific flip-flop, resulting in a header output at that wavelength. As before, a wavelength converter and demultiplexer can then direct the packet to different output ports.
A monolithic tunable optical router was realized in 2010 by integrating more than 200 devices on a single InP chip. The 8 x 8 packet switch was capable of operating at 40-Gb/s, corresponding to a throughput of 320 Gb/s. The 1.45-cm-long router (with a width of 4.25 mm) integrated 8 wavelength converters (using SOAs in an MZ configuration) with a passive 8 x 8 arrayed-waveguide grating router. Such photonic integrated circuits show that packet switching is reaching a stage where a single chip will be able to route packets optically.