Open Access
Add to BibSonomyAbstract
This paper deals with massive number of optical code (OC) label generation and recognition for scalable optical packet switching (OPS) networks. In order to expand the system scalability of code label processing, we develop a record port count 128 x 128 optical encoder/decoder (E/D) and propose a novel three-dimensional (3-D) optical label combining code label with wavelength and polarization. In the experiment, we conduct a proof-of-concept demonstration of 4-code x 2-wavelength x 2-polarization and validate that the 3-D labeling scheme can consequently increase the available number of code label up to more than 1,000 labels. Real-time labeling performance using a field programmable gate array (FPGA)-based processor and crosstalk influence at an optical switch are also experimentally evaluated.
© 2015 Optical Society of America
1. Introduction
Network traffic is exponentially expanding by emerging applications such as cloud computing, videoconferencing, and video streaming. In particular, data centers are seriously facing the rapid increase of network traffic and requiring an alternative solution to current electrical packet switching networks [1]. One possible technology to overcome the bottleneck of electrical routers is an optical packet switching (OPS) [2, 3]. OPS has the potential to offer high-capacity and energy-efficient networks by operating three fundamental subsystems directly in optical domain: the label processer, optical switch, and buffer. Further important point of OPS is a high network utilization via the flow management [4]. In flow management networks, high throughput is guaranteed by appropriately discarding and encapsulating packets for network demands. For instance, a flow manager can carry about 1,000 flows x 100 Mbit/s in 100 Gbit/s data link with fine data granularity. Highly scalable data centers target flow control of 1,000 flows to ensure flexible packet transport [5], that is, an OPS node comprised of 1,000 ports is needed for efficient network resource utilization [6, 7]. However, major barrier to address this requirement is the scalability of label processing, optical switching, and buffering. Recently, the optical switching and buffering technologies for scalable OPS network are maturing [8, 9], while the system expansion of optical labeling is still a remaining issue.
Several types of optical label processing such as time division multiplexing (TDM) [10], wavelength division multiplexing (WDM) [11], subcarrier multiplexing (SCM) [12], phased shift keying (PSK) [13], and orthogonal modulating schemes [14, 15] have been proposed and demonstrated in the literatures. Among these schemes, the WDM label system has been an attractive technology that enabled the number of optical label potentially up to 256 labels [11]. Another approach is code labeling scheme based on the optical encoding and decoding [16, 17]. Extracting optical labels by optical correlation, the code label system can support asynchronous OPS networks on low-latency. In the code label processing, a key device is optical enoder/decoder (E/D) to generate and recognize codes. A multi-port E/D has the unique feature simultaneously processing multiple codes in a single device [18, 19], which can aggregate the code labeling system in an OPS node. It is also worth noting on this device that high frequency efficiency is realized by overlapping each code spectrum like orthogonal frequency division multiplexing (OFDM) signals. In the previous experiment, we proposed a two-dimensional (2-D) optical label system combining code label with wavelength, and demonstrated label processing of 250 labels (25-code x 10-wavelength) using a 50-port optical E/D [20]. However, new device development and labeling scheme are necessary to further enhance system expandability for scalable OPS networks. Real-time label processing also becomes a challenging issue to dynamically manage optical switches and schedulers.
In this paper, we develop a high-port count 128 x 128 optical E/D to accommodate the massive number of code label. Table 1 summarizes the device characteristics of multi-port E/Ds that have been ever developed for code label systems, where spectral efficiency is defined as the bandwidth per code. By the benefit of narrow channel spacing, the highest spectral efficiency of 5 GHz/code is realized in the developed 128-port device. For fully exploiting resource of the developed 128-port device, we propose and experimentally demonstrate a novel three-dimensional (3-D) labeling scheme with code, wavelength, and polarization. In the experiment, a successful operation (an error probability of less than 10−9) for 4-code, 2-wavelength, and 2-polarization is achieved as a proof-of-concept, and the feasibility of 3-D label system scaling up to 1,800 labels is attested. Figure 1 shows the maximum number of optical label as a function of the spectral range for each labeling scheme [19–21]. To the best of our knowledge, this study is the first demonstration for scalable labeling system envisaging more than 1,000 labels. To enable the practical realization of the labeling system in scalable OPS networks, we also implement the real-time 2-D label processing with FPGA platform. The real-time operation of label recognition in the FPGA is successfully verified through the experiment. Finally, we evaluate the influence of switching crosstalk between code labels in an optical switch. The experimental result shows that the crosstalk should be less than −11 dB to keep an error probability under 10−9.
Table 1. Device characteristics for three different types of multi-port E/Ds.
The reminder of this paper is organized as follows. Section II describes the detailed parameters and device characterization of fabricated 128-port device. The 2-D and 3-D label system using the 128-port device are experimentally demonstrated in Section III. In Section IV, we discuss the crosstalk contribution for code labels, followed by our concluding remark in Section V.
2. Device characterization of test 128-port optical E/D
Figure 2 shows optical encoding and decoding operation by the multi-port E/D. At a multi-port encoder shown in Fig. 2(a), different codes are simultaneously generated at different output ports by injecting an optical pulse from an input port. A multi-port decoder processes the code label while suppressing the payload [see Fig. 2(b)]; a distinctive autocorrelation emerges at the only matched port, whereas unnoticeable cross-correlations are measured at the other ports. In this way, code labels are asynchronously recognized on-the-fly at each parallel port in the multi-port decoder. The code embedded in the multi-port E/D is time-spread N-chip PSK code with same intensity. When a Gaussian waveform with standard deviation is sent into input port i, each PSK code generated from output port k is written as,
where donates chip interval, and code phase is represented as,
Fig. 2 Schematic of code generation and recognition in a multi-port device. (a) optical code encoding, (b) optical code decoding
For increasing the number of available code using a single device, we have developed a 128-port optical E/D on the basis of theoretical model and design guideline presented in [20]. Figure 3 shows a photograph of 128-port optical E/D which is fabricated device based on arryed waveguide grating (AWG) configuration. The module size without optical fibers and connectors is 150 mm width x100 mm height x 10 mm depth. The design parameters are listed in Table 2.
Table 2. Design parameter of 128-port optical encoder/decoder.
The port count is higher 128 than the previously fabricated 50-port optical E/D [20]. The codes generated by the 128-port optical E/D have 640 Gchip/s and 128 different phase level. The device insertion loss and adjacent channel crosstalk are 10 dB and −10 dB, respectively. Figure 4 is the spectra of the 128-port device with 5.1 nm (640 GHz) free spectral range (FSR). The spectra are simulated under MATLAB environment and observed by a spectral analyzer at 3 pm resolution. In terms of the spectral efficiency, noteworthy device characteristic is that the channel spacing between adjacent channels is extremely close, that is 5 GHz channel spacing. Owing to the 5GHz channel spacing, the 128-port device can achieve the high spectral efficiency of 5GHz/code. The resulting spectral efficiency is better than the 10GHz/channel realized in the highest resolution AWG for dense WDM (DWDM) [22].
We measured the encoded and decoded waveforms to confirm the accuracy of 128-PSK codes. A 2.4 ps optical pulse from a mode locked laser diode (MLLD) was fed into an input port 1 of 128-port optical encoder to generate codes. At output ports, 128 different codes were obtained and captured by an optical oscilloscope (Keysight 86100C) with 80GHz bandwidth. Figure 5(a) shows the encoded waveforms at the selected ports as an example. All measured waveforms of code signal are spread over 200 ps at time domain, the chip duration of 1.5 ps and chip rate of 640 Gchip/s. The unwanted amplitude variation of each encoded waveform is caused by the slab diffraction effect in the AWG. The coding impairments can be improved by optimizing device design [18] or adopting waveform reshaping technique [23]. However, thisamplitude non-uniformity due to the slab diffraction only slightly affects the code detection in the present device with long code. In the code recognition, we launched an encoded waveform into an input port 1 of 128-port optical decoder. The decoded waveforms from corresponding output ports selected in Fig. 5(a) are presented in Fig. 5(b). A distinctive autocorrelation waveform is detected at the only matched port 1, and cross-correlation waveforms of the lower intensity are measured at the other ports. To elaborate the coding performance, we evaluated the power contrast ratio (PCR) defined as the ration between the average power of autocorrelation and cross-correlation waveforms. The PCR at the port distance j from the matched port 1 is expressed as,
where Pac and Pcc-j (j = 1,2,..., 127) are the average power of autocorrelation and cross-correlation signals at the output of the 128-port decoder, respectively. Figure 6 illustrates the calculated PCR by numerical simulation and measured PCR by an optical power meter (Hewlett-Packard HP815361A) with the minimum responsibility of −70 dBm. The experimentally measured PCRs at near ports from an autocorrelation port (input port 1) agree well with the simulation result. By contrast, the experimental result at distant ports is worse than the estimated value by numerical simulation. This is the reason why the amplified spontaneous emission (ASE) noise added by an optical amplifier to compensate device loss dominates over low power cross-correlations. However, the coding performance is mainly determined by the highest crosstalk, and the −8.5 dB PCR at adjacent ports is sufficient to distinguish incoming codes via electrical thresholding [17]. This result shows that fabricated 128-port optical E/D is feasible to simultaneously generate and recognize independent codes. In the following demonstrations, the adjacent two or four codes are chosen to assume the worst condition that exhibits highest crosstalk scenario.Fig. 5 Measured waveforms at output ports of 128-port optical encoder/decoder. (a) encoded waveforms, (b) decoded waveforms.
3. Experimental demonstration of multi-dimensional optical label
Multi-dimensional labeling scheme realizes the efficient expansion of system scalability in optical label processing [20]. By using the 128-port optical E/D for multi-dimensional labeling scheme, we can drastically increase the possible number of code label. Figure 7 depicts the concept of 2-D labeling scheme using wavelength and code. Since the multi-port E/D has a unique feature of spectral periodicity on spectrum domain, K-code can be allocated in M different wavelength. Hence, the number of optical label in the 2-D label system becomes K x M. In this study, we can eventually handle 2-D optical label of 894 labels (128-code x 7-wavelength) in the C-band range from 1530 nm to 1565 nm.
To further enlarge the number of optical label, we propose a novel 3-D labeling scheme using code, wavelength and polarization, as shown in Fig. 8. By introducing 2 polarization states into 2-D labels with K-code and M-wavelength, this system scales up the label number to K x M x 2 labels. If full resources of the 128-port device within the C-band are exploited in the 3-D label system, we can accommodate independent 1,788 labels (128-code x 7-wavelength x 2-polarization) that satisfy the desired target of 1,000 labels. Therefore, the proposed 3-D labeling scheme is attractive solution for achieving scalable labeling system toward future OPS networks. In this section, we make an experiment to reveal the effectiveness of the proposed 3-D labeling scheme. The experimental demonstration of real-time 2-D label processing is also carried out to justify the practical solution for scalable labeling systems.
3.1 Experimental demonstration of 3-D label system
An experiment of 3-D label processing with 4-code, 2-wavelength, and 2-polarization was demonstrated as a proof-of-concept. In this experiment, the labeling performance that omitted payload part was evaluated for the simplicity. We also investigated the labeling performance of 2-D label including 4-code and 2-wavelength for system comparison. Figure 9 shows the experimental setup of 2-D and 3-D label system placing 128-port optical E/Ds. A 2.4 ps optical pulse with the center wavelength of 1550 nm is driven from a MLLD at 9.95328 GHz repetition. By using a LiNbO3 intensity modulator (LN-IM), the optical pulse train is modulated by 155.52 Mbit/s pseudo random binary sequence (PRBS) signal with pattern length of 27-1. The modulated stream is converted into a supercontinuum (SC) signal by passing through a 2 km dispersion flat fiber (DFF) to ensure the two FSR. The generated SC signal is filtered at the desired bandwidth from 1554.9 nm to 1565.1 nm as shown in Fig. 9(a), and launched into an input port 1 of a 128-port optical encoder. Each two FSR with 640 GHz is assigned at 1557.4 nm (FSR1) and 1562.5 nm (FSR2), whose crosstalk is highest due to the adjacent wavelength allocations. The encoded signals generated from output port 30 (OC1), 31 (OC2), 32 (OC3), and 33 (OC4) are combined with the same power and no contention by using variable optical attenuators (VOAs) and optical delay lines. The coded signal in Fig. 9(b) is separated to each FSR by a 3-dB coupler and optical band pass filters (OBPFs), then combined after the VOAs and delay lines. After combining all signals, different eight 2-D label is aligned in time domain as shown in Fig. 9(c), which is routed by a switch. The switch directly flows the 2-D label to label recognition part when the 2-D labeling performance is evaluated. In the case of 3-D label system, the 2-D label is split into two branches by an optical coupler, and each arm rotates the two polarization states to be orthogonal (X-POL and Y-POL) using polarization controllers (PCs). By propagating the two polarized signals through a polarization beam splitter (PBS), the 3-D optical label with 16 different patterns is obtained at a transmitter, as it is shown in Fig. 9 (d).
At the receiver side, the 3-D label is split into two 2-D labels with 2 polarization states by the PCs. The 2-D labels indicated in Figs. 9(e) and 9(f) are selected at an input port of a 128-port decoder and identified after passing through the decoder and OBPF. The identified labels shown in Figs. 9(g) and 9(h) are detected by photodetectors (U2Photonics XPDV2020R) with 50 GHz bandwidth, and system performance is evaluated using bit error rate testers (BERTs). In this evaluation, we qualify the performance with respect to packet loss probability (PLP), i.e., packet error rate attributed to wrong label recognition [18]. For this reason, the measured BER from received labels is directly interpreted to PLP.
Figure 10 shows the measured PLP of 2-D and 3-D labels for each FSR. The received optical power is monitored at input of the PD, and the power penalty between 2-D and 3-D label arose from polarization interference at the label recognition. Although there is 1.6 dB power penalty for PLP = 10−9, the PLP of less than 10−9 has been achieved for all of the optical labels. All the experiments were examined on worst condition with the highest crosstalk codes and wavelengths. The label interference from the other codes and wavelengths of 3-D label is practically negligible, so that it is possible to smoothly extend the available number of 3-D label by utilizing the all labels of a 128-port device. The wavelength expandability is grounded on the experimental result that multi-port E/D enabled wavelength independent operation over the wide range of wavelength [24]. From the future point, the components to control polarization variation is a crucial task. For instance, the polarization maintenance with polarization-maintaining and absorption-reducing (PANDA) fiber and the polarization compensation with optical power monitoring will be possible solutions [25, 26].
3.2 Real-time 2-D label system with FPGA-based processor
The experimental demonstration of real-time label processing is critical for not only evaluating labeling technique but also establishing the solid platform of dynamic flow control in OPS networks. To validate practical operation of the 2-D label system, we demonstrated real-time label processing of 2-code x 2-wavelength. Figure 11 shows the experimental setup and measured waveforms. The experimental setup is almost same as Fig. 9 except for the label detection part by a FPGA-based processor. An optical pulse sequence is modulated by a LN-IM with decimated signal from a pulse pattern generator (PPG) and emitted from a SC light source contained of a 2 km DFF. In the PPG, a signal pattern of 11000000 is prepared to operate toggle flip flops (T-FFs) at the specific frequency of 77.76 MHz. Insets (a) and (b) of Fig. 11 show the measured optical pulses from the MLLD and DFF, respectively. A 128-port device encodes the SC signal at the output ports 30 (OC1) and 31 (OC2), that are combined together at an optical coupler. The coupled encoded signal is sent to wavelength regulation part, and fourdifferent 2-D labels are generated after OBPF filtering and wavelength combining. The combined 2-D label is shown in Fig. 11(c), and the resulting signal pattern is 11111111 with 77.76 MHz repetition.
At label recognition part, the 2-D label is split into two branches to distinguish label information: one is to decode OCs, and the other is to select FSRs. Insets (d) and (e) of Fig. 11 indicate extracted optical labels by a 128-port decoder and wavelength selective switch (WSS). The each signal is detected by a PD and T-FF as can be seen in Figs. 11(f) and 11(g), whose signal patterns become 10100000 and 10001000 at the repetition rate of 77.76 MHz, respectively. Finally, the electrical signals are received by a FPGA-based label detector, in which the received signals are analog-to-digital converted and processed by a FPGA that consists of the signal decision, AND operation, and label counting functions. The implemented FPGA-based label detector is shown in Fig. 12. It comprises of one FPGA (Xilinx Virtex-II, XC2V1000) and 16-bit analog-to-digital converter (ADC) module for counting the received number of 2-D label. If the sequences of 10100000 and 10001000 are departed from T-FFs for this label detector, the FPGA obtains the signal pattern of 10000000 after AND calculation and counts the existence of “1”. As the result, the label counts of 2 and 1 can be confirmed before and after AND operation, respectively. The FPGA-based label detector can also reduce the number of PD by separately recognizing code and wavelength.
Figure 13 captures label counting result from the received signal that is filtered at the different FSR [e.g., Fig. 11(f)] and decoded by proper OC [e.g., Fig. 11(g)] during about 20 s. The optical received power was set to be −24.5 dBm which was the minimum sensitivity value of the T-FF. The measured time difference between each case is related to the timing variation in clock synchronization. In Fig. 13, total number of received codes and wavelengths before AND operation are described at the third column and row, respectively. On the other hand, each table segment reports the counted 2-D label after AND operation between the total code and wavelength. For example, in the case of OC1, FSR1 shown in Fig. 13(a), the total code and wavelength amount to 388,801,150 labels, and half number (194,400,575 labels) of the total count is confirmed at the only desired 2-D label. Referring the transmitted label count of 388,801,150 labels ( = 77.76Mbps / 4bit x 20.000005916s) and PLP = 10−10 at −25 dBm in Fig. 10, we can regard this performance as no label error. From similar inspection of Fig. 13, we conclude that the real-time operation of 2-D label recognition is successfully demonstrated. Moreover, the experimental condition of 77.76 MHz repetition is reasonable rate in case of considering 10 MHz packet repetition model reported in [19,27]. In the Section 3, we independently connected an 128-port decoder to recognize code and OBPFs to separate wavelength. These components can be aggregated to singe device by the integration through planner lightwave circuit (PLC) platform [28]. Another task is the reduction of PDs at the label receiver, which is solved by using the technique to serially detect the recognized signals in parallel line.
Fig. 13 Real-time label counting results. (a) OC1, FSR1, (b) OC2, FSR1, (c) OC1, FSR2, (d) OC2, FSR2
4. Crosstalk effect evaluation at an optical switch
In the previous analysis, transmission timing of code labels is completely adjusted so that label collision does not occur. In the actual OPS node, however, these time-serial code labels interfere with each other in an optical switch. Figure 14 explains the switching crosstalk generated from an optical switch in an actual node. The node architecture consists of a processor, an optical switch, OC convertors, and an optical buffer. Arrived packets at the node are split into twobranches and launched into a processor and optical switch. The optical switch routes the optical packet to the proper output port by using label information read by the processor. Although the code labels remain to reuse the label resources, the main purpose of the optical switch is the payload switching. The label switching or rewriting is performed by the OC convertors that replace switched labels to the desired labels. For instance, the flexible OC convertor has been recently proposed using highly nonlinear fiber (HNLF) [29]. Through these payload and label switching, this node appropriately forwards optical packets to next nodes. Moreover, the buffer has the role to avoid packet collisions, and the processor controls all components in the node. In this node, the switching crosstalk between code labels is generated when optical packets go through the optical switch. The switching crosstalk increase interference noise at the label recognition and severely restrict the labeling performance. Especially, the crosstalk interference in code labeling system is more critical than that in other labeling systems, because the code interference between narrow channel spacing enlarges the beat noise at the detection. In this section, we study the crosstalk tolerance of label system by elaborating the PLP measured from crosstalk-induced labels. This section also takes into account 1-D label processing for simplifying the discussion. In other words, this evaluation is performed by decoding the crosstalk-induced labels after an optical switch.
To evaluate the crosstalk tolerance, we examined two different systems: the interfering cases of same codes (e.g., OC#A and OC#A in Fig. 14) and different codes (e.g., OC#A and OC#B in Fig. 14). The case of same code interference is the assumption that OC#A arrived at two different switching inputs in Fig. 14 are respectively forwarded to one output port and the optical buffer to prevent packet collision. Figure 15 presents the experimental setup that investigates the crosstalk contribution by changing the applied voltage of a LN switch (LN-SW). A MLLD at 1550 nm is driven at 9.95328 GHz and the pulse stream is modulated by 27-1 PRBS signal using a LN-IM. The label rate is set at 2.488455 Gbit/s corresponding to the autocorrelation width of 400 ps. The modulated signal is guided to an input port 1 of 128-port encoder for the label generation. By changing the connections, it is possible to select the transmission of same codes or different codes, to measure the crosstalk contribution of each interfering case. The VOAs and delay lines are located to control the optical power and signal timing, respectively. The 2 x 1 LN-SW takes the signals at the 0 dBm optical power with same timing and carries the incoming signal from the input port 2 to an output port. In this experiment, the crosstalk is given by the power ratio of leakage from input port 1 to input port 2, which is loaded by adjusting applied voltage of the LN-SW. The crosstalk-induced signal is decoded at a propoer port of 128-port decoder, whose optical power is fixed at −9 dBm. The decoded signal is converted into the electrical signal by a PD, and then its PLP is tested by means of a BERT.
Figure 16 show the measured PLP and waveforms versus the crosstalk level added by the LN-SW. The performances of both systems suffer from the beat noise effect caused by the induced crosstalk. It is mentioned that same code case deteriorates the 4-dB crosstalk tolerance than different code case because of the strongest beat noise generated between autocorrelation signals. From inspection of the measured waveforms, the larger temporal variation is also confirmed in the case of same code. As a consequence, we have observed that the crosstalk of −11 dB is acceptable to keep the PLP under 10−9. The crosstalk evaluation given here can be a stepping stone for future deployment of the code label system.
5. Conclusions
A novel 128-port optical E/D and multi-dimensional labeling scheme have been presented. In order to enhance the system scalability, we have proposed and demonstrated for the first time a 3-D label system. The experimental results have shown the PLP of less than 10−9 for all 16 labels (4-code x 2-wavelgnh x 2-polarization) and verified the feasibility of scalable label system. From a practical viewpoint, we have also implemented FPGA-based processor and realized real-time 2-D label processing. Finally, the crosstalk tolerance for code labels was revealed by evaluating the crosstalk influence in an optical switch. To promise the performance of PLP < 10−9, the crosstalk should be kept below −11 dB. These demonstration could serve as a milestone for the future OPS network with high-capacity and large-scalability.
For further discussion, we should consider code label assignment and timing control in a packet generator and optical switch. In code label assignment, we can flexibly select code labels by installing an optical switch with the multi-port device [30]. As demonstrated in the testbed, the timing control between payloads and labels is realized by sharing a crock source and inserting a delay line [19]. For timing control in an optical switch, a node uses time-gate signals detected by T-FFs and can forward optical packets without strict timing control [17]. In the demonstration, we have not also considered the effect of fiber transmission. As the discussion of transmission characteristic, the previous research demonstrated 100 km single mode fiber (SMF) transmission in a 1 Gbit/s optical code system without chromatic dispersion (CD) compensation [31]. Therefore, the proposed multi-dimensional label system could also cover 40 km standardized in dater centers [32].
Acknowledgments
The authors would like to thank S. Tsuda, H. Kawashiri, Y. Nomoto, and S. Hara of NTT electronics for the device fabrication. We also thank S. Shimizu, Y. Tomiyama, and H.Sumimoto of NICT for their continuous supports throughout this work and M. Hayashi for him help with this experiment. This work has been supported by NICT R&D program, “Basic Technologies for High-Performance Opto-electronic Hybrid Packet Router” (2011~2016) and Japan Society for the Promotion of Science (JSPS).
References and links
1. C. Kachris and I. Tomkos, “A Survey on Optical Interconnects for Data Centers,” IEEE Comm. Surv. Tutor. 14(4), 1021–1036 (2012). [CrossRef]
2. H. J. S. Dorren, M. T. Hill, Y. Liu, N. Calabretta, A. Srivatsa, F. M. Huijskens, H. de Waardt, and G. D. Khoe, “Optical packet switching and buffering by using all-optical signal processing methods,” J. Lightwave Technol. 21(1), 2–12 (2003). [CrossRef]
3. S. J. B. Yoo, “Optical packet and burst switching technologies for the future photonic internet,” J. Lightwave Technol. 24(12), 4468–4492 (2006). [CrossRef]
4. L. G. Roberts, “A radical new router,” IEEE Spectrum Mag. 46(7), 34–39 (2009). [CrossRef]
5. H. H. Bazzaz, M. Tewari, G. Wang, G. Porter, T. S. E. Ng, D. G. Andersen, M. Kaminsky, M. A. Kozuch, and A. Vahdat, “Switching the optical divide: fundamental challenges for hybrid electrical/optical datacenter networks,” in Proc. ACM SIGCOMM 2011 Conf., pp. 30–37 (2011). [CrossRef]
6. S. Di Lucente, N. Calabretta, J. A. C. Resing, and H. J. S. Dorren, “Scaling low-latency optical packet switches to a thousand ports,” J. Opt. Commun. Netw. 4(9), A17–A28 (2012). [CrossRef]
7. N. Farrington, G. Porter, S. Radhakrishnan, H. H. Bazzaz, V. Subramanya, Y. Fainman, G. Papen, and A. Vahdat, “Helios: a hybrid electrical/optical switch architecture for modular data centers,” in Proc. ACM SIGCOMM 2010 Conf., pp. 339–350, (2010). [CrossRef]
8. S. Cheung, T. Su, K. Okamoto, and S. J. B. Yoo, “Ultra-compact silicon photonic 512 x 512 25GHz arrayed waveguide grating router,” IEEE J. Sel. Top. Quantum Electron. 20(4), 8202207 (2014).
9. K. Nozaki, A. Shinya, S. Matsuo, Y. Suzaki, T. Segawa, T. Sato, Y. Kawaguchi, R. Takahashi, and M. Notomi, “Ultralow-power all optical RAM based on nanocavities,” Nat. Photonics 6(4), 248–252 (2012). [CrossRef]
10. X. Yi, W. Shieh, and A. V. Tran, “Cascadability study for optical label swapping using synchronous phase modulation,” IEEE Phton. Technol. Lett. 18(13), 1445–1447 (2006). [CrossRef]
11. A. K. Medhin, M. Galili, and L. K. Oxenløwe, “Scalable in-band optical notch-filter labeling for ultrahigh bit rate optical packet switching,” J. Lightwave Technol. 24(15), 4269–4276 (2014).
12. S. J. B. Yoo, H. J. Lee, Z. Pan, J. Cao, Y. Zhang, K. Okamoto, and S. Kamei, “Rapidly switching all-optical packet routing system with optical-label swapping incorporating tunable wavelength conversion and a uniform-loss cyclic frequency AWGR,” IEEE Phton. Technol. Lett. 14(8), 1211–1213 (2002). [CrossRef]
13. G. Yazawa, S. Shimizu, and H. Uenohara, “Optical serial-to-parallel conversion technique with phase-shifted preamble for optical label switching systems,” IEEE J. Quantum Electron. 47(9), 1222–1229 (2011). [CrossRef]
14. N. Chi, J. Zhang, P. V. Holm-Nielsen, L. Xu, I. T. Monroy, C. Peucheret, K. Yvind, L. J. Christiansen, and P. Jeppesen, “Experimental demonstration of cascaded transmission and all-optical label swapping of orthogonal IM/FSK labelled signal,” Electron. Lett. 39(8), 676–678 (2003). [CrossRef]
15. C. W. Chow and H. K. Tsang, “Orthogonal label switching using polarization-shift-keying payload and amplitude-shift-keying label,” IEEE Photon. Technol. Lett. 17(11), 2475–2477 (2005). [CrossRef]
16. K. Kitayama and M. Murata, “Versatile optical code-based MPLS for circuit, burst, and packet switchings,” J. Lightwave Technol. 21(11), 2753–2764 (2003). [CrossRef]
17. N. Wada, H. Furukawa, and T. Miyazaki, “Prototype 160-Gbit/s/port optical packet switch based on optical code label processing and related technologies,” IEEE J. Sel. Top. Quantum Electron. 13(5), 1551–1559 (2007). [CrossRef]
18. G. Cincotti, N. Wada, and K. Kitayama, “Characterization of a full encoder/decoder in the AWG configuration for code-based photonic routers-part I: modeling and design,” J. Lightwave Technol. 24(1), 103–112 (2006). [CrossRef]
19. N. Wada, G. Cincotti, S. Yoshima, N. Kataoka, and K. Kitayama, “Characterization of a full encoder/decoder in the AWG Configuration for code-based photonic routers-part II: experiments and applications,” J. Lightwave Technol. 24(1), 113–121 (2006). [CrossRef]
20. N. Kataoka, G. Cincotti, N. Wada, and K. Kitayama, “Demonstration of optical label processing with 250 address-entry for scalable optical packet switched network,” in Proceedings of the Optoelectronics and Comumunication Conference (OECC 2011), paper 8A2–5, (2011).
21. K. Morita, G. Weerasekara, T. Kodama, S. Shimizu, N. Wada, and K. Kitayama, “Massive generation/recognition of 2-D label using optical code and wavelength,” in Proceedings of the Photonics and Switching Conference (PS 2013) (2013), paper MT2–4.
22. F. M. Soares, N. K. Fontaine, R. P. Scott, J. H. Baek, X. Zhou, T. Su, S. Cheung, Y. Wang, C. Junesand, S. Lourdudoss, K. Y. Liou, R. A. Hamm, W. Wang, B. Patel, L. A. Gruezke, W. T. Tsang, J. P. Heritage, and S. J. B. Yoo, “Monolithic InP 100-Channel × 10-GHz device for optical arbitrary waveform generation,” IEEE Phton. J. 3(6), 975–985 (2011).
23. S. Shimizu, G. Cincotti, and N. Wada, “Demonstration and performance investigation of all-optical OFDM systems based on arrayed waveguide gratings,” Opt. Express 20(26), B525–B534 (2012). [CrossRef] [PubMed]
24. N. Kataoka, N. Wada, G. Cincotti, and K. Kitayama, “2.56 Tbps (40-Gbps x 8-wavelength x 4-OC x 2-POL) asynchronous WDM-OCDMA-PON using a multi-port encoder/decoder,” in Proc. European Conf. on Optical Commun. (ECOC 2011), paper PDP Th.13.B.6, (2011). [CrossRef]
25. J. Noda, K. Okamoto, and Y. Sasaki, “Polarization-maintaining fibers and their applications,” J. Lightwave Technol. 4(8), 1071–1089 (1986). [CrossRef]
26. R. Khosravani, A. Havstad, Y. W. Song, P. Ebrahimi, and A. E. Willner, “Polarization-mode dispersion compensation in WDM systems,” IEEE Photonics Technol. Lett. 13(12), 1370–1372 (2001). [CrossRef]
27. J. Luo, S. Di.Lucente, J. Ramirez, H. J. Dorren, and N. Calabretta, “Low latency and large port count optical packet switch with highly distributed control,” in Proceedings of the Optical Fiber Comumunications Conference and National Fiber Optics Engineering Conference (OFC/NFOEC 2012), paper OW3J.2, (2012). [CrossRef]
28. S. Kamei, A. Kaneko, M. Ishii, T. Shibata, Y. Inoue, and Y. Hibino, “Crosstalk reduction in arrayed-waveguide grating multiplexer/demultiplexeru using cascaded connection,” J. Lightwave Technol. 23(5), 1929–1938 (2005). [CrossRef]
29. T. Kodama, N. Wada, G. Cincotti, and K. Kitayama, “Tunable optical code converter using XPM and linear-slope pulse streams generated by FBGs,” Opt. Lett. 39(2), 355–358 (2014). [CrossRef] [PubMed]
30. T. Kodama, N. Nakagawa, N. Kataoka, N. Wada, G. Cincotti, X. Wang, and K. Kitayama, “Secure 2.5 Gbit/s, 16-ary OCDM block-ciphering with XOR using a single multi-port En/Decoder,” J. Lightwave Technol. 28(1), 181–187 (2010). [CrossRef]
31. Y. Tanaka, S. Yoshima, N. Kataoka, J. Nakagawa, N. Wada, and K. Kitayama, “100-km uplink of 10G- and 1G-ONU co-existing TDM-OCDMA-PON system using dual-rate burst-mode receiver,” in Proceedings of the Optical Fiber Comumunications Conference and National Fiber Optics Engineering Conference (OFC/NFOEC 2011), paper OThT5, (2011). [CrossRef]
32. IEEEP8023ba, “40Gb/s and 100 Gb/s ethernet task force,” http://www.ieee802.org/3/ba/.
