1. Introduction

Current and future internet-based services and devices place escalating pressure on the bandwidth capacity of optical communication systems. This demand significantly increases the cost of optical networks since additional transmission lines or multiplexing are necessary to handle the increased data traffic. Wavelength division multiplexing (WDM) allows for simple separation of channels through use of a spectral filter [1]. In comparison, optical time division multiplexing (OTDM) places lower data rate channels into unique time slots of a higher rate stream. This allows multiple channels to be placed on a single frequency where each channel can later be extracted for detection using a high-speed switch [1, 2]. Optimized optical networks will use a combination of WDM and OTDM to minimize resource requirements. For future higher bandwidth optical networks, greater single frequency data rates such as those provided using OTDM become more resource effective than adding WDM channels at lower rates [3]. To this end, OTDM demonstrations have reached symbol rates up to 1.28 Tbaud and, correspondingly, data rates up to 10.2 Tb/s with 16 level quadrature amplitude modulation (16-QAM) [4, 5]. While higher single frequency rates are helpful in reducing the network cost, current electronics and photodetectors cannot resolve data at such high rates. In order to properly extract the high speed data, all-optical signal processing is required to demultiplex the temporally encoded channels and attention must be paid to minimizing the added network resources to implement this operation.

A commonly employed demultiplexing scheme for OTDM channel separation uses an all-optical switch, such as a four-wave mixing (FWM) device, to extract a single lower data rate channel, allowing detection with conventional receivers [6–13]. Critically, the resource requirements (e.g. power, size, cost) of this approach scale with the number of channels since each channel requires a dedicated switch. This scaling makes full demultiplexing of an ultrahigh-capacity signal extremely demanding. For example, to fully convert a 1.28-Tbaud OTDM signal to 10-Gbaud channels, 128 devices would generally be required. To alleviate these scaling issues, much research has focused on creating optical demultiplexers that can switch multiple channels simultaneously in a single device [14–27]. These approaches have included multiple modulators and optical paths [16], multicasting the signal source [17], multicasting the pump source or using multiple pump sources [18, 19], and chirping the pump pulses [20–23].

A few previous approaches have demonstrated full demultiplexing of the OTDM signal in a single device at data rates above 100 Gb/s. Specifically, full demultiplexing of a 320-Gb/s OTDM signal to 8x40-Gb/s WDM channels was demonstrated using the multicasting parametric synchronous sampling (MPASS) architecture which first multicasts the OTDM signal [17]. Secondly, full demultiplexing of a 160-Gb/s OTDM signal to 4x40-Gb/s channels was demonstrated by first multicasting the pump source [18]. While generally effective in reducing the number of devices required for demultiplexing, both of these methods necessitate an extremely large amount of bandwidth since each demultiplexed channel requires a replica of the OTDM signal or the pump laser. For high capacity systems such as a 128 channel 1.28-Tb/s OTDM signal, assuming a 50% fill factor, these demonstrated methods would require well over 300 THz of bandwidth for full synchronous demultiplexing.

Older works, by Morioka et al. [20] and later from the same group, Uchiyama et al. [21–23], demonstrated multiple channel demultiplexing in a single device without multicasting. Morioka experimented with chirping a pump pulse across multiple channels of an OTDM test source and generated WDM channels via FWM or cross phase modulation. This approach was able to convert multiple channels simultaneously but for only a few channels since each converted idler necessitated a significant amount of bandwidth to avoid crosstalk. Recently, a related method to Morioka et al, has been proposed and demonstrated for simultaneous demultiplexing without multicasting the signal or pump [24–27]. These works improve upon Morioka’s design by using the principles of the space-time duality to create the temporal analogue of a Fourier processor. To create this temporal Fourier processor, a pre-chirping stage is added, which reduces the bandwidth and crosstalk of the generated WDM signal. Breakthrough work by Mulvad, et al. (2011) demonstrated simultaneous demultiplexing of 43 out of 64 tributaries with 25-GHz spacing in a 640-Gb/s signal using a single device. This performance and other works by the same group demonstrate demultiplexing more than half of the channels at once [25–27] as well as operation on various signaling formats [27]. However, since only a subset of channels is extracted at least two devices are necessary for full demultiplexing.

Here, we experimentally investigate full OTDM demultiplexing using a single temporal Fourier processor. We demonstrate error-free (bit error rate < 10⁻⁹) demultiplexing of a 160-Gb/s OTDM signal to 16x10-Gb/s WDM channels for the first time using this architecture. To achieve full demultiplexing, the FWM pump pulses are chirped sufficiently to cause temporal overlap with the adjacent pump pulses. This allows all of the OTDM channels to be extracted in a single FWM interaction. Extending the temporal aperture of the pump pulses reduces crosstalk from adjacent channels that impact channels on the edges of the generated WDM spectrum. However, this reduction in crosstalk comes at the expense of additional crosstalk on channels in the center of the WDM spectrum due to non-degenerate FWM processes. We find the tradeoff between these two sources of crosstalk is the primary limiting factor in system performance. By striking a balance between these effects through the choice of the pump bandwidth, we are able to simultaneously achieve a bit error rate (BER) of less than 10⁻⁹ for all 16 generated WDM channels.

2. Principle of operation

A temporal Fourier processor is implemented using the principles of the space-time duality of electromagnetic waves [28]. This duality indicates that analogies can be drawn between spatial propagation of a beam of light and temporal dispersive propagation of a light pulse. One example of this duality is the temporal Fourier processor. Specifically, a lens positioned a focal length away from an object will produce an image at the opposite focal plane that is the two-dimensional Fourier transform of the object. Similarly, using the principles of the space-time duality, if a temporal optical waveform is dispersed through a focal length of dispersion, imparted with a temporally quadratic phase shift (known as a time lens [24–37]), and subsequently dispersed with a second focal length of dispersive propagation, the one-dimensional Fourier transform of the input waveform will be generated at the output of the system. This process is referred to as a temporal Fourier processor (Fig. 1 ) [24–27,30–36].

 

Fig. 1 A temporal Fourier processor is created using a FWM time lens. The incoming signal waveform is chirped prior to FWM, combine with a pump pulse chirped by twice the amount of the signal, and the resulting converted waveform propagates through an equivalent but opposite dispersion length to the signal. The resulting temporal waveform is a scaled version of the incident spectrum and likewise the generated spectrum is a scaled version of the incident temporal waveform.

Download Full Size | PDF

Here we use FWM with a chirped optical pump pulse to impose a quadratic temporal phase shift (a time lens) on a given signal [37]. This has been shown previously to enable optical waveform magnification, compression, and Fourier transformation [32–36]. The temporal Fourier processor performs the Fourier transform of the optical waveform over the temporal aperture of the chirped pump pulse used for FWM. This process converts the temporal features of the input waveform to the spectrum of the output waveform and the spectral features of the input waveform to the temporal waveform of the output as depicted in Fig. 1. Experimental realizations of this functionality originally focused on enabling measurement of ultrahigh-speed optical waveforms by converting the task into a relatively simple spectral measurement at the output of the system [30,32]. Recent research has extended the applications of the process to the conversion of multiple channels of an OTDM signal to parallel wavelengths channels and thereby conversion of OTDM to a WDM signaling scheme [24–27]. Importantly, the temporal and spectral spans at the output of the system can be controlled through the choice of the time-lens focal length (i.e. the dispersive lengths) and for OTDM to WDM conversion allows the WDM signal to be generated with a spectrally efficient industry standard channel spacing.

Previous research into temporal Fourier processing [25–27, 32–36] has employed temporally isolated optical pump pulses to perform the time-lens operation. Temporal guard bands between the chirped pump pulses ensure that distortions arising from temporally adjacent pump pulses are negligible. However, this design restricts the temporal Fourier processor from operating without interruption on a continuous input signal and therefore, for example, requires at least two devices for full OTDM to WDM conversion. Here, we focus on continuous operation of a temporal Fourier processor and demonstrate full OTDM to WDM conversion in a single nonlinear device. We use the term continuous operation to indicate that the device is able to capture and demultiplex all bits from the incoming data stream. This necessitates chirping the pump pulses used to generate the time lens such that they temporally overlap as depicted in the spectrogram shown in Fig. 2(a) . There are a variety of challenges that accompany operation in this regime and impact achievable error-rate, number of channels, and spectral efficiency. These challenges are briefly summarized in the following section.

 

Fig. 2 (a) A spectrogram depiction of the time lens FWM process (not to scale). (b) An example of non-degenerate FWM adding crosstalk to a center channel. The red ellipses outlined in black indicate the non-degenerate FWM conversion of channel 1 onto channel 3.

Download Full Size | PDF

3. Principal challenges for continuous operation

Crosstalk from adjacent channels

Channels that overlap with the temporal edge of the pump pulse will experience lower conversion efficiency and reduced spectral compression since the pump pulse intensity declines in this region. This causes increased crosstalk amongst adjacent channels and therefore leads to an increased power penalty for channels at the edge of the generated WDM spectrum. Increasing the overlap of adjacent pump pulses reduces the impact of this impairment.

Additionally, the performance of the temporal Fourier processor relies on achieving pure group-velocity dispersion in the dispersive paths. Third-order dispersion will cause aberrations in the time lens leading to reduced spectral compression and therefore increased crosstalk from adjacent channels at impacted regions of the WDM spectrum. To mitigate this impairment we use matched optical fibers with opposing third-order dispersion in the dispersive paths.

Crosstalk from non-degenerate FWM

Figure 2(b) illustrates the origin of crosstalk due to non-degenerate FWM. Since the pump pulses temporally overlap, the overlapping regions can act as unique non-degenerate pumps in the FWM stage. As depicted in Fig. 2(b), due to energy conservation, these two separate pumps mixing with the OTDM signal will reproduce channel 1 onto channel 3′s output frequency, for example. In general, the non-degenerate FWM processes cause crosstalk from channels at the edge of the WDM spectrum onto channels at the center of the WDM spectrum [24]. Interestingly, as shown in Fig. 2(b), this crosstalk appears between the bit slots of the impacted channels. Decreasing the overlap of adjacent pump pulses reduces the impact of this impairment.

Crosstalk balance

As described in the previous two subsections, increasing the pump overlap decreases crosstalk from adjacent channels while increasing crosstalk from non-degenerate FWM. For this reason, a balance between the two sources of crosstalk must be found through the choice of pump overlap, which is set by the dispersion and pump bandwidth. Figure 3 shows the results of a simulation based on the methods of Ref. 24 for a 160 Gb/s and 16 channel system to evaluate the effects pump bandwidth on crosstalk. The solid curves for adjacent channel crosstalk show the crosstalk power on channel 2 from channels 1 and 3 relative to the signal power of channel 2. The dotted curves for non-degenerate FWM crosstalk show the crosstalk power on channel 9 from channel 1 relative to the signal power of channel 9. As seen in Fig. 3, there exists an optimum pump bandwidth to minimize the overall impact of crosstalk on the system. Note that since non-degenerate FWM crosstalk appears between the bit slots, the system is more tolerant of this crosstalk than crosstalk from the adjacent channels. This indicates that the ideal operating condition is at slightly larger pump bandwidth than the intersection of the curves depending on the receiver bandwidth.

 

Fig. 3 Simulated adjacent channel crosstalk (solid) and non-degenerate FWM crosstalk (dotted) as a function of 3-dB pump bandwidth and pump fill factor for no timing jitter (red) and RMS timing jitter between the pump and OTDM signal of 2 ps (blue), 4 ps (green), 6 ps (black). We define pump fill factor as the dispersed 3-dB pump pulsewidth divided by the pump period.

Download Full Size | PDF

Ancillary nonlinear effects

Partially-degenerate FWM amongst the temporally overlapping regions of the pump pulse will convert pump energy to wavelengths immediately surrounding the pump spectrum. Additionally, self-phase modulation (SPM) and cross-phase modulation (XPM) will cause aberrations in the time-lens performance as well as spectral broadening of the pump source during the interaction. To mitigate these impairments, the pump power is kept sufficiently low and a spectral guard band between the pump laser and the generated WDM spectrum is utilized.

4. Experimental setup

The experimental setup to demonstrate full 160-Gb/s OTDM to 16x10-Gb/s WDM conversion is shown in Fig. 4 . An 80/20 coupler splits the output of a 10-GHz harmonically mode locked erbium fiber laser tuned to 1560 nm to generate both the OTDM data source and the FWM time-lens pump source. The OTDM signal and the FWM pump are generated at unique wavelengths through subsequent spectral broadening and filtering in each arm as follows.

 

Fig. 4 Experimental setup showing the generation of the OTDM test source as well as the temporal Fourier processor. BPF and TBPF represent optical bandpass filter and tunable optical bandpass filter, respectively. For BPFs, the optical wavelength range passed through the filter are indicated under the block as 1529nm→1538nm and 1554nm→1563nm. WDM stands for wavelength division multiplexer, APD is an avalanche photodiode, BERT is a bit error rate tester, EOM is an electrooptic modulator, MZI is a highly asymmetric Mach-Zehnder interferometer, and D3 and D38 represent optical fiber with dispersion parameters of 3 ps/nm-km (Vascade LS + ) and 38 ps/nm-km (Vascade S2000) respectively. The autocorrelation trace of the 160-Gb/s OTDM test source is shown as an inset above the OTDM test source stage.

Download Full Size | PDF

OTDM signal

The 20% output of the coupler seeds the OTDM signal pulses. In this path, the 1560-nm laser spectrum is broadened through SPM in 620-m of highly nonlinear fiber (HNLF) and is spectrally filtered to create the test signal at 1536 nm. The HNLF used in this experiment (OFS HNLF Zero-Slope) has a nonlinear coefficient of 11 W⁻¹km⁻¹, zero dispersion-slope, and zero-dispersion at 1550 nm. This signal is filtered to a bandwidth of 2.25 nm to create transform limited pulses of 2.1 ps and therefore a 33% fill factor when the signal is later multiplexed to a 160-Gb/s return to zero format. A lithium niobate electro-optic amplitude modulator encodes the test signal with a 2³¹-1 pseudorandom bit sequence (PRBS) generated by the bit error rate tester (BERT). Four stages of highly asymmetric Mach-Zehnder interferometers (MZI) with delays set to maintain a PRBS length of 2⁹-1 and with amplifiers before the MZI stages and after the second MZI stage, multiplex the PRBS test signal to 160 Gb/s. Prior to the dispersive components of the time lens, the signal is compressed in single mode fiber (SMF-28) to the transform limit. An autocorrelation trace of this test signal is shown as an inset of Fig. 4 above the FWM stage. Two stages of dispersion fiber (165-m of Corning Vascade LS + and 415-m of Vascade S2000) provide the proper dispersion for the optical Fourier transform (OFT). These fibers possess opposing third-order dispersion and the combination of lengths is chosen to eliminate third-order dispersion from this dispersive path while achieving the proper amount of group-velocity dispersion. The 160-Gb/s test signal is then amplified to have a power of about 175 mW in the FWM stage and combined with the control pump in a wavelength division multiplexer (WDM).

Time lens pump source

The 80% output of the 80/20 coupler is used to generate the pump source for the FWM time lens. The laser pulses are spectrally broadened through SPM in 190-m of the Vascade LS + to create a spectrally flat pump source. An approximately 10-nm spectral range centered at 1560 nm is filtered with a tunable bandpass filter to create pump pulses with sufficient bandwidth for the OFT. Two stages of the dispersive fiber (310-m of Corning Vascade LS + and 1086-m of Vascade S2000) provide proper pump dispersion to create the OFT while eliminating third-order dispersion from the path. A tunable delay is used to align the pump pulses with the OTDM data and an erbium doped fiber amplifier (EDFA) boosts the dispersed pump prior to combining with the test signal in a WDM which yields 800 mW of optical pump power in the FWM stage.

Time lens FWM stage and detection

The FWM stage of the OFT is performed using 30-m of HNLF which is placed after the wavelength division multiplexer. An L-band filter separates the converted WDM signal (spanning from 1577 nm to 1590 nm) from the test source and control pump pulse before the signal is amplified using an L-band EDFA. Finally, each individual WDM channel is isolated with a tunable bandpass filter and detected on an avalanche photodiode (APD) and BERT. The received power is measured immediately prior to the APD for BER characterization.

5. Experimental results

Figure 5(a) shows the full spectrum exiting the FWM stage of the system. The 160-Gb/s OTDM test signal is centered at 1536 nm and spans 2.25 nm at full width half maximum (FWHM). This test signal undergoes FWM with the pump, which is centered at 1560 nm, spans 4 nm at 3-dB bandwidth, and 8 nm at 10-dB bandwidth. After dispersion, the 3-dB bandwidth stretches to 60% of the pump period and the 10-dB bandwidth stretches to 120% of the pump period. The temporal overlap of the pump pulses causes partially-degenerate FWM amongst the pump pulses and generates the spectral features at 1550 nm and 1570 nm. After FWM with the pump, the OTDM signal is converted to a WDM signal as observed in the 16 idlers toward the long wavelength side of the spectrum. The 16 idlers have a spacing of 108 GHz, a span from 1577 nm to 1590 nm and a 7.5-dB variation in amplitude among the channels. The majority of this variation is attributed to the flatness of the pump pulse’s spectrum which can be improved through more sophisticated spectral shaping [25–27]. Figure 5(b) shows an expanded view of the generated WDM spectrum after passing through the L-band amplifier (red solid line). The purple dotted line shows the filtered spectrum when detecting channel 16. Similar bandwidth was used to detect each channel.

 

Fig. 5 (a) The full experimental spectrum after the FWM stage. (b) The idler spectrum after L-band amplifier for all 16 channels (red solid line), with half of the channels blocked (black dashed line), and the filtered spectrum for channel 16 prior to detection (purple dotted line). The resolution bandwidth of (a) and (b) is 0.01 nm with (a) 0.2 nm and (b) 0.08 nm between data points.

Download Full Size | PDF

To illustrate the effect of non-degenerate FWM from the OTDM signal and the temporally overlapping pump regions we block half of the OTDM channels (channels 5 through 12) by disconnecting the delay line in the first MZI stage. The black dashed line in Fig. 5(b) shows the idler spectrum in this configuration. As is shown, four erroneous channels appear in the center of the generated WDM spectrum due to the non-degenerate FWM processes.

Figure 6 shows eye diagrams from selected channels taken on a digital sampling scope after detection. Channel 16 at a BER of 10⁻⁷ and channel 15 at a BER of 10⁻⁸ are shown in Fig. 6(a) and 6(c), respectively, as examples of properly performing channels. These channels are not affected by the non-degenerate FWM crosstalk. Note that channel 16 (Fig. 6(a)) exhibits typical BER performance (Fig. 7 ) and channel 15 (Fig. 6(c)) is one of the poorer performing channels. Channel 15 performs poorly due to crosstalk of adjacent channels. Figure 6(d) shows the eye diagram of channel 15 with the output filter opened wide enough to include greater crosstalk from channels 14 and 16 to illustrate its effect. As is shown, the energy from channels 14 and 16 appear within the center of the bit slot of channel 15. In comparison, Fig. 6(b) shows the eye-diagram for channel 9 which is a channel near the center of the WDM spectrum and therefore suffers from crosstalk due to the non-degenerate FWM process. In contrast to crosstalk from adjacent channels, the non-degenerate FWM crosstalk on channel 9 (Fig. 6(b)) appears directly between bit slots. This is in agreement with our previous numerical simulations of this system [24]. Non-degenerate FWM leads to poor performing channels in the center of the idler spectrum and is found to be equally limiting as the adjacent channel crosstalk and lower conversion efficiency that afflicts the channels at the edge of the spectrum.

 

Fig. 6 Eye diagrams for selected WDM channels. (a) channel 16, (b) channel 9, (c) channel 15, and (d) channel 15 with the filter bandwidth prior to detection widened to observe the effects of adjacent channel crosstalk.

Download Full Size | PDF

 

Fig. 7 (a) Bit error rate (BER) versus received optical power for all 16 channels and the back to back (B2B) measurement. (b) Spectrum of the generated WDM channels and their channel number. The resolution bandwidth of (b) is 0.01 nm with 0.08 nm between data points. (c) The received power necessary to achieve a BER of 10⁻⁶ and power penalties of each of the channels. (d) The received power necessary to achieve a BER of 10⁻⁹ and power penalties of each of the channels. In (c) and (d) the dashed line indicates the B2B power necessary to achieve the respective BER.

Download Full Size | PDF

As shown in Fig. 7, we are able to achieve error-free performance (BER < 10⁻⁹) for all of the generated WDM channels. Full demultiplexing of all 16 channels is acquired with power penalties between 1.5 dB and 12 dB at a BER of 10⁻⁹ (Fig. 7). The BER curves that correspond to the eye diagrams in Fig. 6 as well as the back to back (B2B) curve are indicated in the legend. The B2B curve is taken by bypassing the MZI stages and feeding the 10-Gb/s data signal directly to the detector. The rest of the WDM channels are shown as red solid lines. An expanded view of the idler spectrum with the corresponding channel numbers is shown in Fig. 7(b). The received power necessary to achieve a BER of 10⁻⁹ and the resulting power penalty relative to a back-to-back measurement for each channel are shown in Fig. 7(d). Note that the channels with higher power penalties appear both towards the edge and near the center of the WDM spectrum and are impacted by adjacent channel crosstalk (edges) and crosstalk due to non-degenerate FWM (center). The poorest performing channels are the center channels such as 9 and 7 and less focused edge channels such as 15 and 14. While there is a large spread in power penalties for BER < 10⁻⁹ performance, the BER curves do not significantly diverge until BER < 10⁻⁶. Therefore, for systems with a more relaxed bit error rate requirement, such as those with error-correction, all of the power penalties are within 3 dB of the B2B at a BER of 10⁻⁶, as shown in Fig. 7(c).

6. Discussion

The spectral width of the pump laser is critical to achieving error-free performance for all 16 channels. The degree of temporal overlap of the pump pulses after chirping is adjusted in the experiment through a tunable bandpass filter that controls the spectral bandwidth of the pump laser. When the pump is filtered to a narrow bandwidth to reduce the non-degenerate FWM crosstalk, the channels at the edge of the idler spectrum exhibit a poor BER performance due to adjacent channel crosstalk and diminished signal strength. In comparison, when a broader pump bandwidth is used the performance of the edge channels improves but the performance of the channels near the center of the WDM spectrum suffers from greater crosstalk due to non-degenerate FWM. For the greatest overall performance, a balance must be achieved between these two effects. This balance manifests itself in the “W”-shaped power penalty versus channel number curves in Fig. 7(c) and 7(d). In comparison previous work by Mulvad et al. [25], which convert only a subset of the channels, observe a “U”-shaped curve.

Thermal drift causes a slow wavelength shift in the idlers and was observed during operation. This was manually corrected using the tunable delay on the pump arm and will require a feedback circuit in a deployed system. Rapid fluctuations such as timing jitter due to clock recovery will lead to frequency jitter in the generated WDM channels and thereby increases the impact of crosstalk as depicted in Fig. 3. For minimal impact on the performance the RMS timing jitter between the pump laser and the OTDM signal should be maintained below approximately 2 ps.

The pump spectral shape for the simulations shown in Fig. 3 is 16th-order super-Gaussian. When the pump flatness is reduced and the spectral cutoff is more gradual (e.g. as seen in this experiment), our simulations indicate that the adjacent channel crosstalk is reduced while the FWM crosstalk is increased. This causes the optimal 3-dB pump pulse bandwidth to decrease and thereby decreases the signal power in the edge channels. As a result, the signal-to-noise ratio of the edge channels suffers yielding poorer BER performance. For this reason, we expect a flatter and sharper pump pulse shape, which can be achieved with more sophisticated pulse shaping techniques [25], to improve the power penalty difference among the channels in future systems.

The spectral components found at 1550 nm and 1570 nm from partially-degenerate FWM of temporally overlapping pump pulses necessitate the use of a spectral guard band between the pump laser and the generated WDM spectrum. This increases the bandwidth usage of the device relative to a system that switches only a subset of the OTDM channels and therefore does not have temporal overlap of the pump pulses. Despite this increase in bandwidth, the system presented here is still more efficient in its use of bandwidth (~6.9 THz) than other methods based on multicasting which we estimate to require over 15 THz for a 160-Gb/s OTDM to 16x10-Gb/s WDM system based on our OTDM fill factor of 33%. The spectral features found at 1530 nm and 1543 nm arise from SPM and cross-phase modulation on the signal pulses in the FWM stage. Since they are well separated from the pump and generated WDM idlers, no increase in bandwidth usage is necessary to accommodate these features.

Although limited to 10 Gb/s here due to the available BERT, full demultiplexing can be carried out in a similar manner with 40-Gb/s WDM channels. In our simulations, we find little change in the crosstalk characteristics when the entire OTDM stream is demultiplexed to 4x40 Gb/s with a 400-GHz WDM spacing. However, if the channel spacing relative to the channel bandwidth is modified (e.g. 40-Gb/s WDM channels with 100-GHz spacing) we find an increase in the adjacent channel crosstalk leading to reduced overall system performance.

7. Conclusion

We experimentally demonstrate full simultaneous error-free (BER < 10⁻⁹) demultiplexing of all channels of a 160-Gb/s OTDM data stream to 16x10-Gb/s WDM channels for the first time using a single OFT device. Unlike other approaches to full OTDM demultiplexing with a single device, the WDM channels are generated with a spectrally efficient 108-GHz spacing in the telecommunications L-band. The primary impairments to system performance are found to be adjacent channel crosstalk and crosstalk from non-degenerate FWM with the temporally overlapping regions of the pump pulses. These impairments primarily impact the edge channels and center channels, respectively. These two sources of crosstalk must be balanced though choice of the pump spectral width to maximize the overall performance of the device, which leads to error-free operation of all channels with a “W”-shaped power penalty versus channel number. This demonstration shows that OTDM signals can be demultiplexed in both a resource efficient (e.g. power, size, cost) and spectrally efficient manner and facilitates the incorporation of OTDM into future resource efficient ultrahigh-bandwidth optical communications architectures. Furthermore, this approach is compatible with any FWM device and we anticipate further advances in resource efficiency can be made through the use of power efficient and compact integrated nonlinear elements [9–13,25].