December 1, 1997 / Vol. 22, No. 23 / OPTICS LETTERS 1757 Optical bit-pattern recognition by use of dynamic gratings in erbium-doped fiber Jun Shan Wey, Douglas L. Butler, Nicholas W. Rush, Geoffrey L. Burdge, and Julius Goldhar Laboratory for Physical Sciences and Department of Electrical Engineering, University of Maryland, 8050 Greenmead Drive, College Park, Maryland 20740 Received June 26, 1997 We present a novel optical bit-pattern-recognition technique that uses erbium-doped fiber at room temperature. Counterpropagating beams write a patterned gain-depletion grating in pumped erbium-doped fiber. This grating, recorded in the erbium gain medium, can be used for correlation with other optical bit patterns. We have demonstrated correlation of arbitrary return-to-zero bit patterns of as many as 8 bits at 1 Gbitys. Theory suggests that scaling to higher bit rates is feasible. 1997 Optical Society of America Optical correlation techniques can be used to identify specific optical bit patterns for header recognition or for code-division multiplexing in high-speed op- tical communication systems. Several research groups have demonstrated all-optical correlation by using inhomogeneously broadened absorption in low-temperature crystals, 1,2 high-temperature vapor, 3 and spectral holography. 4 We are investigating a technique that uses erbium-doped fiber (EDF) at room temperature as the medium for recording a desired pattern and correlating 5 it with an optical signal. Our technique, in which correlation is performed directly in the time domain, is mathematically equivalent to techniques that operate in the spectral domain. 1–4 The principle of operation is illustrated in Fig. 1. First we record the desired patterns, which are stored as distributed spatial gain gratings in the EDF. During the writing cycle, counterpropagating waves from the same optical source set up a standing- wave pattern that causes a spatially inhomogeneous saturation of the gain. This gain modulation creates a Bragg grating that potentially can have strong ref lectance. 5 In principle, the EDF can be operated in either the absorption or the gain regime; however, we observe much stronger reflection when the EDF is in the gain regime. These depletion gratings have a lifetime of the order of 1 ms, which is related to the excited-state lifetime of the erbium ions. After the gratings are written, the stored pattern can either be read out or used for correlation. When a single reading pulse is sent into the EDF as shown in Fig. 1, the ref lection from the stored gratings recon- structs a time-reversed version of the test pattern. For an unknown pattern, for example, an address in a fiber- optic transmission, that is being sent into the EDF, the reflection corresponds to a correlation between the input and the stored test patterns. If the unknown pattern is the same as the test pattern, a large correla- tion peak appears, which indicates a match. Reading or correlation must be performed before the gratings decay. Otherwise, the writing process has to be re- peated to refresh the gratings. The experimental configuration, shown in Fig. 2, can be used to write, read out, or correlate optical bit sequences. The output of a cw 1535-nm laser with a long coherence length s,1 kmd is first divided into two paths. The length difference between the top and the bottom paths is much less than the coherence length of the laser. Independently controlled Mach– Zehnder modulators and optical amplifiers define the bit patterns for each beam. Modulator #1 (see Fig. 2) generates the reference beam, and modulators #2 and #3 create the write and read patterns. The bit patterns are programmed by use of the subrate outputs of a 12-GHz bit-pattern generator such that all patterns are synchronized. Interference gratings are written in the 1.5-m-long EDF with an absorption coefficient of 34 dBym at 1535 nm. The EDF is pumped by a 980-nm laser diode with a maximum pump power of 90 mW. During the writing time, the top path generates a single reference pulse, while the bottom path produces the test pattern. The single reference pulse and the counterpropagating test pattern collide in the EDF and write a set of gratings that corresponds to the test pat- tern. This process is repeated every 16 ns, which is the round-trip time through the EDF. In this experi- ment we wrote and refreshed the gratings 100 times s1.6 msd before reading or correlation was performed. Fig. 1. Principle of operation for writing, reading, and correlation with dynamic gratings in erbium-doped fiber. Counterpropagating reference and test patterns create a grating pattern in the EDF. After an interference grating is stored, it can be read out with a single pulse or correlated with a specif ic pattern. 0146-9592/97/231757-03$10.00/0 1997 Optical Society of America
Optical bit-pattern recognition by use of dynamicgratings in erbium-doped fiber
Jun Shan Wey, Douglas L. Butler, Nicholas W. Rush, Geoffrey L. Burdge, and Julius Goldhar
Laboratory for Physical Sciences and Department of Electrical Engineering, University of Maryland, 8050 Greenmead Drive,College Park, Maryland 20740
Received June 26, 1997
We present a novel optical bit-pattern-recognition technique that uses erbium-doped fiber at room temperature.Counterpropagating beams write a patterned gain-depletion grating in pumped erbium-doped fiber. Thisgrating, recorded in the erbium gain medium, can be used for correlation with other optical bit patterns. Wehave demonstrated correlation of arbitrary return-to-zero bit patterns of as many as 8 bits at 1 Gbitys. Theorysuggests that scaling to higher bit rates is feasible. 1997 Optical Society of America
Optical correlation techniques can be used to identifyspecific optical bit patterns for header recognitionor for code-division multiplexing in high-speed op-tical communication systems. Several researchgroups have demonstrated all-optical correlationby using inhomogeneously broadened absorption inlow-temperature crystals,1,2 high-temperature vapor,3
and spectral holography.4 We are investigating atechnique that uses erbium-doped fiber (EDF) at roomtemperature as the medium for recording a desiredpattern and correlating5 it with an optical signal. Ourtechnique, in which correlation is performed directlyin the time domain, is mathematically equivalent totechniques that operate in the spectral domain.1 – 4
The principle of operation is illustrated in Fig. 1.First we record the desired patterns, which arestored as distributed spatial gain gratings in theEDF. During the writing cycle, counterpropagatingwaves from the same optical source set up a standing-wave pattern that causes a spatially inhomogeneoussaturation of the gain. This gain modulation createsa Bragg grating that potentially can have strongref lectance.5 In principle, the EDF can be operatedin either the absorption or the gain regime; however,we observe much stronger ref lection when the EDF isin the gain regime. These depletion gratings have alifetime of the order of 1 ms, which is related to theexcited-state lifetime of the erbium ions.
After the gratings are written, the stored patterncan either be read out or used for correlation. Whena single reading pulse is sent into the EDF as shownin Fig. 1, the ref lection from the stored gratings recon-structs a time-reversed version of the test pattern. Foran unknown pattern, for example, an address in a fiber-optic transmission, that is being sent into the EDF,the ref lection corresponds to a correlation between theinput and the stored test patterns. If the unknownpattern is the same as the test pattern, a large correla-tion peak appears, which indicates a match. Readingor correlation must be performed before the gratingsdecay. Otherwise, the writing process has to be re-peated to refresh the gratings.
The experimental configuration, shown in Fig. 2,can be used to write, read out, or correlate optical bitsequences. The output of a cw 1535-nm laser witha long coherence length s,1 kmd is f irst divided into
0146-9592/97/231757-03$10.00/0
two paths. The length difference between the top andthe bottom paths is much less than the coherencelength of the laser. Independently controlled Mach–Zehnder modulators and optical amplifiers define thebit patterns for each beam. Modulator #1 (see Fig. 2)generates the reference beam, and modulators #2and #3 create the write and read patterns. The bitpatterns are programmed by use of the subrate outputsof a 12-GHz bit-pattern generator such that all patternsare synchronized. Interference gratings are writtenin the 1.5-m-long EDF with an absorption coeff icientof 34 dBym at 1535 nm. The EDF is pumped by a980-nm laser diode with a maximum pump power of90 mW.
During the writing time, the top path generates asingle reference pulse, while the bottom path producesthe test pattern. The single reference pulse and thecounterpropagating test pattern collide in the EDF andwrite a set of gratings that corresponds to the test pat-tern. This process is repeated every 16 ns, which isthe round-trip time through the EDF. In this experi-ment we wrote and refreshed the gratings 100 timess1.6 msd before reading or correlation was performed.
Fig. 1. Principle of operation for writing, reading, andcorrelation with dynamic gratings in erbium-doped fiber.Counterpropagating reference and test patterns create agrating pattern in the EDF. After an interference gratingis stored, it can be read out with a single pulse or correlatedwith a specif ic pattern.
Fig. 2. Experimental conf iguration for writing, reading,and correlating bit sequences. The top path generates thereference pulses, and the bottom path creates the writeand read patterns. The gratings are stored in and readout from the EDF. EDFA, erbium-doped f iber amplif ier;WDM, wavelength-division multiplexing.
The average powers of the reference beam and thewrite–read beam entering the EDF are approximately0.5 and 0.1 mW, respectively.
During reading, one turns off the reference pulses inthe top path by programming the pattern generator.A single reading pulse or an arbitrary pattern isgenerated in the bottom path (see Fig. 2) to read out orto correlate with the stored pattern. For enhancementof the ref lected signal, the peak power of a readingpulse is 50 to 100 times larger than that of a writingpulse. The ref lection output is monitored by a high-speed sampling oscilloscope.
We first characterized the grating ref lectance asa function of reading-pulse duration. To accomplishthis, we recorded cw gratings in the entire EDF,and a single reading pulse with a pulse width of83 ps to 25 ns was sent into the EDF. When thereading-pulse duration was longer than the time off light through the EDF s,8 nsd, the ref lected pulseshape was similar to the shape of reading pulse.When the reading-pulse width was much shorter than8 ns, the ref lected pulse showed distinct features ofthe gratings. Some of these features are caused byfiber birefringence and can be eliminated by useof polarization-maintaining erbium fiber. The gaindistribution inside the pumped EDF results in strongersignals’ being ref lected from gratings farther fromthe input. We observed that ref lections from part ofthe fiber had a relatively constant amplitude. Thisregion, corresponding to a 5-ns ref lected signal, is thebest location for pattern storage.
Figure 3 shows both the experimental and the cal-culated results for the power ref lectance as a functionof the reading-pulse width. The ref lectance can ex-ceed 1 because of gain in the EDF. The numerical cal-culation was performed with a steady-state model ofsaturated EDF,6 where the gain modulation was deter-mined by the interference between the counterpropa-gating waves. We assume that the refractive-indexgrating is negligible and the ref lection is caused bythe gain modulation only. Our experimental resultsconfirm this assumption. Polarization variations inthe EDF cause the experimental output to be lower in
power, because our calculations assume a single polar-ization for all beams.
For reading-pulse widths greater than 8 ns (timeof f light through the EDF), the ref lectance remainsconstant. At less than 8 ns, the ref lectance decreasesas the reading-pulse width decreases, because the re-f lected field is the coherent addition of the waves re-f lected from a spatial region that corresponds to halfthe optical pulse length in f iber. As the pulse widthchanges, the ref lected electric field varies proportion-ally. This behavior is clearly observed in Fig. 3. Onemay conclude from the data that the ref lectance is toolow for this technique to be useful at high bit rates.However, it is important to point out that this is trueonly in reading. As the bit rate increases, we can useproportionally longer bit patterns. Given a constantgrating strength and the same spatial extent of bit pat-terns, and assuming that there are equal numbers of1’s and 0’s in a pattern, the amplitude of the correla-tion peak will be independent of the bit rate.
Figure 4 shows the readout of a short grating writ-ten at different locations within the EDF. The refer-ence and the writing pulses are both 2 ns in duration.When the grating is written near the end where thereading pulse enters the EDF, the ref lected pulse ex-periences little gain, and its amplitude is low. As thegrating is moved further into the EDF, the read signalsexperience more gain, and their amplitudes increase.This result is important in determining the grating lo-cation for the largest ref lected signal.
With the optimal grating location identif ied, we per-formed experiments to demonstrate writing, reading,and correlation of arbitrary bit patterns in the pumpedEDF. Figure 5 shows reading and correlation of 4-bitreturn-to-zero (RZ) patterns at 500 Mbitsys. Here wemeasure the electric-field amplitudes because of theinterference of the weak ref lected pulses with the cwleakage in the reference beam. As expected, the ex-tinction ratio between the correlation peak and thesidelobes is 2 :1.
Figure 6 shows the correlation outputs of 8-bit RZpatterns at 1 Gbitys. Because the cw leakage is re-duced here, the output displays the ref lected power.
Fig. 3. Experimental and theoretical results of power re-f lectance as a function of reading-pulse width. Theory andexperimental data do not coincide because of polarizationeffects.
Fig. 4. Readout of a short grating written in differentlocations in the EDF. The shaded region (round-trip timethrough the EDF) shows the relative location of the gratingwith respect to the EDF. Pulses ref lected from the farend of the EDF experience double-pass gain and hence arelarger in amplitude.
Fig. 5. Reading and correlation of 4-bit RZ patterns at500 Mbitsys. The top trace shows the single-pulse readoutfrom the recorded pattern. The middle (bottom) trace is acorrelation between the recorded and the identical (inverse)read patterns.
We observe an extinction ratio that is in good agree-ment with the theoretical prediction of 16:9 for a coher-ent correlation. In Figs. 5 and 6 one can see a strongcorrelation peak when the correct pattern is input.When the input patterns do not match the dynamicallystored pattern, we obtain lower power outputs with nodistinct central peak.
Several major challenges remain. In the writingprocess the wavelengths of the reference and the writebeams need to be identical. When they interact inthe EDF, they also need to be coherent and of thesame polarization. One can satisfy this requirementby generating the reference and the write beams fromthe same long-coherence-length laser. In the read-ing–correlation process the wavelength of the reading–correlating beam must also be within 0.001 nm of thewrite beam to meet the Bragg condition. The polar-ization state of the reading–correlating beam does not
Fig. 6. Correlation of the stored pattern with the identical(top) and the inverse (bottom) 8-bit RZ pattern at 1 Gbitys.
have to match the write beam. In a more realistic situ-ation, when the incoming pattern is generated from alaser source that is different from the laser used torecord the gratings, a special data-regeneration pro-cedure may be necessary for production of correlatingpatterns that are suitable for use with this technique.
Another challenge is scaling to high bit rates. Al-though the correlation process is independent of bitrates, as we mentioned above, the grating strengthreduces as the bit rate increases (pulse width de-creases). Other techniques, now under developmentin our laboratory, can be used to generate gratings inthe EDF. With the current experimental configura-tion and a better writing technique, we should be ableto process 100-RZ data bits at 10 Gbitsys in 1 m of EDF.
In summary, we have demonstrated a novel all-optical-fiber, room-temperature technique for opticalbit-pattern recognition. This technique operates inthe standard telecommunication wavelength regimeand can process bit patterns at gigabit-per-second rateswith a storage lifetime of the order of 1 ms.
We thank Daniel Mahgerefteh, Dave Rush, Lance Jo-neckis, John Jacob, Curt Lowry, and Kunal Kishore fortheir valuable discussions and experimental support.
References
1. K. D. Merkel and W. R. Babbitt, Opt. Lett. 21, 71 (1996).2. X. A. Shen and R. Kachru, Opt. Lett. 20, 2508 (1995).3. Y. S. Bai, W. R. Babbit, N. W. Carlson, and T. W.
Mossberg, Appl. Phys. Lett. 45, 714 (1984).4. H. P. Sardesai and A. M. Weiner, in Conference on Lasers
and Electro-Optics, Vol. 11 of 1997 OSA Technical DigestSeries (Optical Society of America, Washington, D.C.,1997), paper CTHW3.
5. B. Fischer, J. L. Zyskind, J. W. Sulhoff, and D. J.DiGiovanni, Opt. Lett. 18, 2108 (1993); Electron. Lett.29, 1858 (1993).
6. J. S. Wey, ‘‘Experimental and theoretical studies of aharmonically modelocked erbium fiber ring laser,’’ Ph.D.dissertation (University of Maryland, College Park, Md.,1995).
