Optical data packet synchronization and multiplexing using a tunable optical delay based on wavelength conversion and inter-channel chromatic dispersion

Abstract: 10 Gb/s non-return-to-zero (NRZ) on-off keyed (OOK) optical data packets are synchronized and time-multiplexed using a 26-ns tunable all-optical delay line. The delay element is based on wavelength conversion in periodically poled lithium niobate (PPLN) waveguides, inter-channel chromatic dispersion in dispersion compensating fiber (DCF) and intrachannel dispersion compensation with a chirped fiber Bragg grating (FBG). Delay reconfiguration time is measured to be less than 300 ps.


Introduction
Optical packet switching holds the promise of highly efficient use of the available bandwidth in an optical network.As with electronic packet switching, a key enabling technology is the ability to controllably delay/buffer a data packet so as to synchronize packets and rapidly resolve output port contention.Desirable characteristics of this delay include large tuning range, continuous tunability, and rapid reconfiguration [1].Previously published results for optical delays used in packet switching have typically been chosen from a finite set of discrete optical path lengths.This methodology produces only a fixed set of delays, whether they are in fiber [2,3], waveguides [4], or free-space [5].However, a laudable goal for a truly flexible and efficient packet switch would be to generate any arbitrary delay value.Recently, tunable delays have been shown using wavelength conversion via self-phase modulation [7] or four wave mixing [8,9] coupled with dispersive elements.In this method, a data stream was converted to a slower-propagating wavelength, followed by conversion back to the original wavelength.Note that proper dispersion compensation was necessary.The result is that a data bit arrives at the output, delayed in time relative to a data bit that had always resided at the original, faster-propagating wavelength.Bit-level delays were shown [10], but little was reported relating to using these delays for time manipulation and synchronization of full data packets.In this paper, we experimentally demonstrate synchronizing and multiplexing of optical data packets using a 26-ns tunable optical delay based on wavelength conversion and inter-channel chromatic dispersion.We achieved tunability for 10-Gbit/s non-return-to-zero (NRZ) on-off keyed (OOK) data packets, multiplexed two packet streams, and measured a 10 -9 bit error rate (BER).The delay reconfiguration time from one delay value to another was measured to be <300 ps.

Concept of conversion/dispersion based delay module for packet synchronization
Shown in Fig. 1 is the packet-synchronizing scheme.Two packet streams on two different wavelengths are routed such that the packets requiring delay (packet-3 on λ 1 ) pass through the delay module.After a packet is delayed for synchronization, it is then multiplexed with the non-delayed stream (λ 2 ) using a 3-dB optical coupler.As shown in Fig. 2, the delay module uses wavelength-dependent chromatic dispersion generated by a dispersive element, such as dispersion compensating fiber (DCF).Using sum frequency generation (SFG) followed by difference frequency generation (DFG) in PPLN-1 [11], we convert λ 1 to the desired wavelength λ c .The optical delay equals the total wavelength shift (λ 1 -λ c ) multiplied by the total dispersion (ps/nm) of the DCF [9,10].We use a second PPLN to convert the delayed signal back to the original wavelength, thereby preserving the original wavelength at the output.Since the DCF also causes data-degrading intra-channel dispersion, we use a chirped fiber Bragg grating (FBG) centered at λ 1 and with the opposite dispersion of the DCF in order to perform output dispersion compensation [12].

Two stage wavelength conversion in PPLN
The wavelength conversion process is shown in Fig. 3. SFG in Fig. 3(a) is produced by using two laser pumps that are spectrally equidistant from the quasi-phase-matched (QPM) PPLN wavelength.In PPLN-1, the two pumps are the input signal at λ 1 and a local pump at λ pump-1 .These two pumps mix via the first χ 2 process of SFG to generate λ p /2 = {(λ 1 +λ pump-1 )/2}/2.This λ p /2 mixes with another input "dummy" wavelength λ dummy-1 to produce a converted output at λ c = 2λ p -λ dummy-1 via the second χ 2 process of DFG.By tuning λ dummy-1 , we can tune our converted signal λ c to almost any desired value within the PPLN bandwidth.At PPLN-2 in Fig. 3(b), λ c and λ pump-2 are both equidistant from the QPM of PPLN-2 and constitute the two pumps.By tuning λ dummy-2 , we can ensure that the output wavelength is equal to λ 1 again, thus preserving the original signal wavelength.Although PPLN devices can have a bandwidth of more than 70 nm [11], our bandwidth is limited to ~25 nm since we use erbium-doped fiber amplifiers (EDFAs) to amplify the signals.
We note that the "continuous" delay range has small "gaps", such that the converted wavelength cannot be spectrally located at the QPM or the local pump.However, since the QPM is temperature tunable [11], we can avoid this problem by slightly tuning the QPM and the local pump wavelengths such that the gaps can be removed.Furthermore, another inaccessible λ is the input signal λ 1 , and this corresponds to zero delay.

Experimental setup and results
The experimental setup is shown in Fig. 4. The 196-bit, NRZ-OOK 10-Gbit/s data packets are 19.6-nslong, and the packet guard time of empty space is 8 bits (800 ps).Packets are generated electronically and drive Mach-Zehnder intensity modulators.We manually programmed the pulse pattern generator (PPG) in order to generate the stream of 196-bit data packets, and the BERs were measured by programming the PPG and error detector accordingly.The 19.74-km DCF has a total dispersion of -1742 ps/nm at 1550 nm, loss of 10.7 dB, and dispersion slope of -0.22 ps/nm 2 /km.The chirped FBG has a positive dispersion of +2020 ps/nm, a 0.456-nm bandwidth at 1546.4 nm, and peak reflectivity of 89%.PPLN-1 and PPLN-2 have QPM wavelengths of 1550.1 and 1554.7 nm, respectively, at 91.5 °C.Laser diode (LD) λ pump-1 and LD λ pump-2 are fixed at 1553.8 and 1562.8 nm, respectively.LD λ dummy-1 and LD λ dummy-2 are tuned according to the desired converted wavelength, λ c .Polarization controllers (PCs) are inserted in the input path of each PPLN since the converters are polarization dependent [13].The input powers are: (i) for PPLN-1, λ 1 and λ pump-1 are each 14 dBm and λ dummy-1 is 10 dBm, and (ii) for PPLN-2, λ pump-2 and λ dummy-2 are each 12 dBm and λ c is 5 dBm.The converted wavelength is -13 dBm for PPLN-1 and -19 dBm for PPLN-2.The filters at the output of PPLN-1 and PPLN-2 have 3-dB bandwidths of 0.8 and 1.2 nm, respectively.The receiver is a 10-Gbit/s p-i-n device.The packets on λ 1 (1546.4nm) are routed through the delay module.Figure 5 shows three example delay scenarios of 0, 10 and 26.4 ns, corresponding, respectively, to λ c at 1556.7, 1551.92 and 1542.5 nm.The inset eye diagram is for the packet stream that has been delayed by 26.4 ns.The two packet streams at λ 1 and λ 2 are initially offset by more than a packet length.By changing λ c from 1556.76 to 1542.5 nm, we introduce a delay of 26.4 ns that results in aligning the λ 1 packet into the vacant slot between packets 1 and 2 on λ 2 (1552 nm). Figure 7 shows the synchronized packets that are multiplexed together using a 3-dB optical coupler.The inset eye diagram of Fig. 7 shows that the delayed data packet is slightly more noisy than the non-delayed packet.We believe that this is due to the added ASE arising from the nonoptimally-filtered high-power EDFAs that were needed to overcome the PPLNs conversion efficiency of -25 dB.Note that any distortion can be minimized by employing PPLNs with higher conversion efficiency, low noise EDFAs and matched dispersion compensation.Note that the maximum residual dispersion between the DCF and FBG is around 310 ps/nm when the converted wavelength is 1542.5 nm.We measured the BER of a manually-programmed bit sequence for the back-to-back (nondelayed) λ 1 packet stream, delayed λ 1 packet stream, and multiplexed λ 1 +λ 2 packet stream.Figure 8 shows a ~2.5-dB power penalty at a BER=10 -9 for the multiplexed signal.This penalty is due to the non-idealities of the delay module.Since higher bit rates are much more sensitive to residual chromatic dispersion, finer control on dispersion compensation would be required.

Reconfiguration time of optical delay
The delay can be tuned by either changing the input wavelength or the converted wavelength.For our demonstration to dynamically reconfigure the optical delay, we keep λ c constant and change λ 1 between the two values of 1548.5 and 1551.76 nm.We use this approach for this section because we want to remove from our measurement the additional frequencydependent time-of-flight of different wavelengths inside the DCF.The modified approach and experimental setup is shown in Figs. 9 and 10.Two input lasers (1548.5 and 1551.76 nm) that are equidistant from the QPM of PPLN-1 are input to a multi-GHz 2x2 prism-based electro-optic switch.The output port-1 of the switch is connected to a 3-dB coupler.The continuous-wave signal on one output of the coupler is fed to PPLN-1 as λ pump-1 , while the other coupler output becomes λ dummy-2 of PPLN-2.Output from port-2 of the switch is routed through the modulator and input to PPLN-1 as λ 1 .We keep λ dummy-1 and λ pump-2 fixed at 1553.8 and 1562.8 nm, respectively.The final output from PPLN-2 is at two different wavelengths, 1560.2 and 1557.5 nm, depending on whether the case is the "switch-off" or "switch-on" position (see Fig. 9).We split the output of PPLN-2, put optical filters with center wavelengths located at the two possible final output values, and observed both of the outputs on a dual-channel optical oscilloscope.The path lengths for both of these outputs are kept equal.As shown in the inset of Fig. 10, the observed elapsed reconfiguration time is ~276 ps.We note that the reconfiguration time in our experiment includes the inherent delay of the switch.Furthermore, we chose the values of the switched lasers at 1548.5 nm and 1551.76 nm due to the particular QPM settings of the available PPLNs.

Fig. 2 .
Fig. 2. In the two scenarios, λ 1 is converted to different λc's, resulting in different group velocities due to inter-channel dispersion.The undesired intra-channel dispersion is compensated by a FBG.(PPLN : periodically poled lithium niobate waveguide)

Fig. 8 .
Fig. 8. Measured bit error rate (BER) for back-to-back, delayed single packet stream and multiplexed data stream.Power penalty of 2.5 dB is observed.

Fig. 9 .
Fig. 9. (a) Output spectra of the two PPLNs when the switch is in the OFF position.(b) Output spectra when the 2x2 switch is turned ON.

Fig. 10 .
Fig. 10.Experimental setup for measuring the reconfiguration time of the delay scheme.The inset shows that the reconfiguration time is 276 ps, which is the time when one delayed signal vanishes (blue-colored) and the new delayed signal (pink-colored) appears.Ovals are passive couplers.