Abstract
There is increasing interest in performing many key networking functions in the optical domain. Optical header recognition is one such key function that may enable rapid reading of optical packets in the future all-optically-switched network core. Many of these optical header recognition functions are enabled through the use of fiber Bragg grating-based optical correlators. A brief background on optical header recognition and FBG correlators is presented, and two fiber Bragg grating-based optical header recognition methods are explained in detail: an optical bypass for an Internet router, and a multi-wavelength header recognition system using sampled fiber Bragg grating optical correlators.

I. Introduction
Future high-performance optical networks may require optical data packets to be rapidly routed by all-optical switches. For efficient and high-throughput switching, the header bits within each packet must be recognized and acted upon quickly. Although electronic techniques are available for reading individual bits at lower bit rates (i.e., <10 Gbit/s), there is a requirement for all-optical methods that can readily “read” a series of header bits on-the-fly and either assist or replace the electronics in making ultra-fast routing decisions.

In conventional Internet routers, packets are steered toward their destinations by interrogating their 32-bit destination addresses and matching them to entries in a large routing table — a time consuming process as modern look-up tables can contain upwards of 100,000 entries. Current optical signal processing technologies, however, do not scale well, making it impractical to build a 100,000-entry all-optical lookup table. However, in the network core, where routers have only four to eight output ports and only a few hops may be necessary to reach a core access point, it may be possible to determine a packet’s destination port by looking at only a subset of the bits in the destination address. Recent research has shown that looking at a subset of a core routing table with as few as 100 entries can successfully route up to 90% of network traffic [1], and by looking at which subsets of those headers are required to identify an output port, a manageable optical lookup table can be constructed.

One common way to perform this optical bit-pattern recognition is through the use of a time-domain optical correlator to match a series of bits to an optical lookup table [2]. The purpose of a correlator is to compare an incoming signal with one that is “stored” in the correlator. At an appropriate sample time, a maximum autocorrelation peak will be produced if the input signal is an exact match to the stored one. Numerous optical correlator designs have been experimentally demonstrated, including those using fiber Bragg gratings (FBGs) [3], FBGs in complement with novel periodically-poled lithium niobate (PPLN) waveguide wavelength shifters [4], erbium-doped fiber [5], free-space holography [6], and deposited metallic fiber mirrors with hardwired optical delays [7]. FBG-based correlators have some specific advantages over other designs, including increased tunability (the reflectivity of FBGs can be tuned via heating [8] or stretching [9] the gratings), ease of manufacture, and relatively low cost. Using FBG-based correlators, a number of innovative optical header recognition systems have been proposed and demonstrated: i) an optical bypass for an internet router using a single-grating topology that can scale to high bit rates, and ii) a reconfigurable WDM header recognition (WDM-HR) system using sampled FBGs.

II. Fiber Bragg Grating-based Optical Correlators
A set of fiber Bragg gratings can be used to construct an effective fiber-based optical correlator [3]. An FBG is fabricated by creating a periodic variation in the fiber’s index of refraction for a few millimeters to a centimeter of length along the fiber core [10]. A conventional FBG acts as a reflective, wavelength-selective filter. A nice feature of FBG filters is that the reflection spectrum can be adjusted by a few nanometers via heating or stretching of the grating. The reflectivity of the grating is nearly 100% at the center of the reflection spectrum and falls off quickly outside the grating bandwidth. FBG correlators can be constructed by writing many gratings in succession with center-to-center spacings equal to of a bit time so that the round-trip time between gratings corresponds to a 1-bit delay. To program a “1” bit in a correlation sequence, a grating is tuned to be partially or wholly reflective, and to program a “0” bit, a grating is tuned away so that it does not reflect. The reflectivity of each grating must be chosen such that the pulses reflecting off each grating have equal power. In practice this can be determined by repeatedly sending a single pulse into the FBG array and adjusting the grating to equalize the powers of the time-delayed output pulses on an oscilloscope (each “1” bit grating will reflect the input pulse at different times, creating multiple pulses at the output). The resulting grating array will produce a correlation output corresponding to the input sequence correlated with the programmed correlation sequence, and a threshold detector can sample the central lobe of this sequence to determine the presence of a pattern match, as shown in Fig. 1 for a correlation sequence of “1101.”

Fig. 1. An FBG optical correlator is an array of FBG mirrors with tunable reflectivity R. The correlation sequence is programmed as “1101”, with the first, second, and fourth gratings tuned to be reflective, corresponding to “1” bits, and the third grating is tuned away so it does not reflect, corresponding to a “0” bit. The input sequence, also “1101”, reflects off each grating, and the result can be sampled electronically and a pattern match signal generated.

In general, optical correlators cannot uniquely identify a given bit sequence. As gratings representing a “0” bit are tuned away and do not reflect, those bits are actually “don’t-care” bits and a threshold detector will determine a match at the sample time regardless of whether the “don’t-care” bits are a “0” or a “1” and may result in a false positive when a “1” bit is present where a “0” bit is desired. To uniquely recognize any of the 2^N possible N-bit sequences, a second correlator is used, configured in complement to the first one that produces a “match” signal when zero power is present at the sample time. In this “zeros” correlator, a “0” bit is programmed by tuning a grating to be reflective, and a grating corresponding to a “1” bit is tuned away and does not reflect. As only an identical pattern will simultaneously provide a high threshold in the “ones” correlator and zero power in the “zeros” correlator at the sample time, by combining the outputs of the standard “ones” correlator with this additional “zeros” correlator using an AND gate a matching signal is produced that can drive an optical switch.

III. Optical Bypass for an Internet Router
A true all-optical router would need to be capable of 32-bit lookups into 100,000-entry tables at ³40 Gbit/s. Such capabilities are beyond current optical technologies, but some recent developments hint at the feasibility of a partial solution. It may be feasible to build an “optical bypass” to vastly accelerate a conventional router [11]. As much as 90% of the incoming traffic may be routed by this bypass, and remaining traffic, requiring more complicated processing, can be handled by a conventional electronic router. This bypass is made possible through the use of FBG optical correlators as header subset recognition devices. A diagram of this optically-boosted router is shown in Fig. 2. A new correlator design in which an FBG array is constructed from a single uniform FBG using separate, electrically-tunable thin-film micro-heaters is used to construct this optical bypass, and is shown in Fig. 3(a). The grating is fabricated out-of-band and the heaters are used to tune individual parts of the grating in-band. Thus, a multi-bit correlator is constructed from a single long grating (Fig. 3(b)). As these thin-film micro-heaters can be made to lithographic precision, millimeter grating spacings are easily achieved using this technology, making it scalable to ³10 Gbit/s.

Fig. 2. Conceptual diagram of a core router with N (in this case, 4) output ports assisted by a bank of M optical FBG-based correlators that are dynamically configured by a software algorithm which reduces the routing table to a few popular entries. After grouping these entries by output port, each correlator is configured to match to a particular group by recognizing only a subset of the bits in the destination address. Packets that fail to match any of the correlators are routed via conventional electronics.

To configure the “ones” and “zeros” correlators to recognize a bit pattern of “xx1x01x0” (a header where a subset of only 4 out of the 8 bits is required to determine the output port) in a system running at 10 Gbit/s, the third grating in the “ones” correlator is tuned to partially reflect, and the sixth grating is tuned for full reflection. Likewise, the fifth grating in the “zeros” correlator is tuned for partial reflection, and the eighth grating in the “zeros” correlator is tuned for full reflection. Thermal tuning is accomplished by applying a voltage across each micro-heater. A packet-rate timing signal is used to trigger a set of low-speed decision circuits to sample the correlation outputs at the proper time. The two outputs are sent through an AND gate that provides a high signal when there is a match, and a low signal otherwise. The match/no-match signal that results can be used to control an optical switch and route matched packets to an appropriate output port, as shown in Fig. 4.

Fig. 3. (a) A set of micro-heaters are deposited on a single long out-of-band FBG, creating small, individually-tunable gratings out of the larger FBG. Each is spaced by 1 cm to correspond to a bit time at 10 Gbit/s. (b) The optical spectrum of the FBG in transmission showing one of the smaller gratings tuned .6 nm away from the central reflection spectrum via thermal tuning.

IV. Reconfigurable Multi-wavelength Optical Correlators Using Sampled Fiber Bragg Gratings
The above method for optical header subset recognition acts on a single wavelength-division-multiplexed (WDM) channel, thus requiring N complete modules in order to recognize the headers on N different WDM channels. A correlation module that can allow for reconfigurable optical correlation of multiple WDM channels simultaneously may use significantly fewer components [12]. Using a set of discrete sampled fiber Bragg gratings, a multi-wavelength FBG correlator can be constructed. A sampled FBG is an FBG that has a superstructure written on top of the grating for which the Fourier transform produces a reflective time delay that is replicated at equal wavelength spacings [13]. When this type of FBG is stretched or heated, the entire reflection spectrum shifts, causing the reflectivity at each wavelength to experience the same variation, as shown in Fig. 4. Thus, the correlation sequence can be reconfigured for all incoming channels simultaneously.

Fig. 4. Experimental routing results showing the successful recognition and switching of packets containing the header subset pattern “xx1x01x0”. The single matched packet is routed to port C, non-matched packets are routed to port D.

While a sampled FBG correlator can be constructed in a manner similar to a standard FBG correlator, the spacing requirements can be problematic. As it takes approximately ~1 cm of fiber to provide a bit time delay at 10 Gbit/s, the center-to-center spacing between gratings must equal this length. Due to their complexity, it can be difficult to manufacture sampled FBGs shorter than 1 cm that have the high reflectivity required to produce good correlation results. This problem can be solved by interleaving the gratings between multiple fiber branches and using a splitter prior to the correlator. This decreases the spacing requirement by a factor equal to the number of branches. While these limitations on sampled FBG systems reduce the scalability of the architecture, a recent report details a new sampled FBG structure that can reduce the length of sampled FBGs while maintaining high reflectivity, enabling the application of this correlation technique to high-bit-rate systems [14].

Fig. 5. The spectrum of a sampled fiber Bragg grating showing multiple reflection peaks. As the grating is tuned via stretching, the spectrum shifts and the reflectivity of each channel decreases. By applying the right amount of stretching, the gratings can be tuned such that there is almost no reflection in each channel, enabling tuning of a correlation sequence.

An interleaved sampled FBG correlator was constructed and configured to match a header of “1010.” In the “ones” correlator, the first grating was tuned via stretching to partially reflect, and the third tuned for full reflection, while in the “zeros” correlator, the second grating was tuned to partially reflect, and the fourth grating tuned for maximum reflectivity. Packets on two WDM channels, each at 10 Gbit/s, were correlated by the gratings and sent to individual packet-rate decision circuits as described in the above section. The resulting match/no-match signals for each channel were used to control an optical switch fabric where header-matched packets were routed to one output port, and all other packets to an alternate output.

V. Conclusion
While optical correlators currently see limited commercial application, the frontier of all-optical networking is rapidly approaching, aided by the ever-increasing demand to transmit more bandwidth over the network core. For applications such as header recognition, technologies that can implement huge arrays or banks of correlators to efficiently test incoming signals against all possible bit sequences may be needed. However, it may also be possible (within the network core) to use small, presently achievable banks of optical correlators to recognize a small subset of the 32 bits in the standard IP header, and thus route a significant percentage of the traffic optically, using electronics only when all available optical lookup techniques fail to produce a matching condition. In addition, optical header recognition techniques that can act simultaneously on multiple wavelength channels in a WDM system may result in significant savings. While reaching these goals presents a significant engineering challenge, progress is being made, and optical header recognition using optical correlators offers great potential in making the all-optical switched network of the future a reality.

References
1. J. Bannister, J. Touch, P. Kamath, and A. Patel, “An optical booster for internet routers,” Proc. Eighth Int’l Conf. on High Performance Computing, Dec. 2001

2. N. Wada and K.-I. Kitayama, “Photonic IP routing using optical codes: 10 Gbit/s optical packet transfer experiment,” Conf. on Optical Fiber Communications (OFC) 2000, paper WM51, pp. 362-364, 2000.

3. D. B. Hunter and R. A. Minasian, “Programmable high-speed optical code recognition using fibre Bragg grating arrays,” Electronics Lett., vol. 35, no. 5, 412-414, 1999.

4. D. Gurkan, M. C. Hauer, A. B. Sahin, Z. Pan, A. E. Willner, K. R. Parameswaran, and M. M. Fejer, “Demonstration of multi-wavelength all-optical header recognition using a PPLN and optical correlators,” European Conf. on Optical Communications (ECOC) 2001, paper We.B.25 (Invited), Amsterdam, 2001.

5. J. S. Wey, J. Goldhar, D. L. Butler, and G. L. Burdge, “Investigation of dynamic gratings in erbium-doped fiber for optical bit pattern recognition,” Proc. Conf. on Lasers and Electro-Optics (CLEO) 1997, paper CThW1, 443-444, 1997.

6. J. Widjaja, N. Wada, Y. Ishii, and W. Chijo, “Photonic packet address processor using holographic correlator,” Electronics Lett., vol. 37, no. 11, 703-704, 2001.

7. J.-D. Shin, M.-Y. Jeon, and C.-S. Kang, “Fiber-optic matched filters with metal films deposited on fiber delay-line ends for optical packet address detection,” IEEE Photonic Tech. Lett., vol. 8, no. 7, 941-943, 1996.

8. Y.-W. Song, Z. Pan, D. Starodubov, V. Grubsky, E. Salik, S. A. Havstad, Y. Xie, A. E. Willner, and J. Feinberg, “All-fiber WDM optical crossconnect using ultrastrong widely tunable FBGs,” IEEE Photonic Tech. Lett., vol. 13, no. 10, pp. 1103-1105, 2001.

9. E. R. Lyons and H. P. Lee, “An efficient electrically-tunable etched cladding fiber Bragg grating filter tested under vacuum,” IEEE Photonic Tech. Lett., vol. 13, no. 5, pp. 484-486, 2001.

10. R. Kashyap, Fiber Bragg Gratings, Academic Press, California, 1999.

11. M. C. Hauer, J. McGeehan, J. Touch, P. Kamath, J. Bannister, E. R. Lyons, C. H. Lin, A. A. Au, H. P. Lee, D. S. Starodubov, and A. E. Willner, “Dynamically reconfigurable all-optical correlators to support ultra-fast internet routing,” Conf. on Optical Fiber Communications (OFC) 2002, paper WM7, 268-270, 2002.

12. J. McGeehan, M. C. Hauer, A. B. Sahin, and A. E. Willner, “Reconfigurable multi-wavelength optical correlator for header-based switching and routing,” Conf. on Optical Fiber Communications (OFC) 2002, paper WM4, 264-266, 2002.

13. M. Ibsen, M. K. Durkin, M. J. Cole, and R. I. Laming, “Sinc-sampled fiber Bragg gratings for identical multiple wavelength operation,” IEEE Photonic Tech. Lett., vol. 10, no. 6, pp. 842-844, 1998.

14. N. Yusuki and Y. Shinji, “Realization of various superstructure fiber Bragg gratings for DWDM systems using multiple-phase-shift technique,” Conf. on Optical Fiber Communications (OFC) 2002, paper TuQ3, 110-111, 2002.