1 x 4 all-optical packet switch at 160 Gb/s employing optical processing of scalable in-band address labels N. Calabretta, H.D. Jung, J. Herrera, E. Tangdiongga, A.M.J. Koonen and H J.S. Dorren COBRA Research Institute, Eindhoven University of Technology, PO. Box 512, 5600MB – Eindhoven, The Netherlands
[email protected]
Abstract: We demonstrate for the first time a 1x4 all-optical packet switch that that utilizes a highly scalable and asynchronous header processor. Error-free operation at 160 Gb/s while wavelength routing over 20 nm is demonstrated. 2008 Optical Society of America OCIS codes: (060.6719) Switching, packet; (200.4740) Optical processing; (060.1155) All-optical networks processing
1. Introduction All-optical packet switching based on all-optical signal processing has been envisioned as a technology to solve the mismatch between the fibre bandwidth and the router forwarding capacity, at bit-rates where electronics is too slow to directly process the data [1]. In an all-optical packet switch the packets are routed based on address information that is encoded by attached labels. In [2] a 1 x 2 all-optical packet switch is demonstrated, that employs a single inband optical label. The processed label is stored by an optical flip-flop and the packet is routed by a wavelength converter. A key-issue is the implementation of a highly scalable header processing technique. Here, we present a 1 x 4 optical packet switch that employs a header processor that is highly scalable. A key issue is that building up a large optical packet switched cross-connect out of 1 x 2 optical packet switches requires a large number of packet switches. For instance, in a WDM switching node that has 16 input and 16 output fibres that each carry 16 wavelength channels one needs 16 x 16 = 256 1 x 2 packet switches per input fibre. This brings the total number of 1 x 2 packet switches for the entire switching node to 4096. If the node however contains 32 input fibres and 32 output fibres carrying each 32 wavelength channels, the number of packet switches increases to 32768. If one succeeds however to realize a 1 x N optical packet switch, the number of switches required in a node can be drastically reduced. Suppose that a 1 x 16 optical packet switch exists, a node that consists out of 16 input and 16 output fibres that carry each 16 wavelengths can be built-up out of 256 packet switches. To realize a 1 x N optical packet switch, it is essential to process N labels in parallel. In this paper we present a 1 x 4 optical packet switch that employs encoded in-band labelling addresses and a novel label processor. 2. System Operation Figure 1 shows the packet switch configuration. Packet payload is generated by time-multiplexing a 40 Gb/s datastream consisting of 2540 bits of pre-defined return-to-zero bits at λp=1553.8 nm up to 160 Gb/s data-stream using a passive fibre-based pulse interleaver. Each bit has duration of 1.6 ps making the 20 dB bandwidth of the payload to be 5 nm. The resulting packet payload consists of a 254 ns data burst. The packet-to-packet guard time is 2 ns making the packet repetition rate 256 ns. With this we mean, that the labels have a wavelength that is within the 20 dB bandwidth of the packet payload. We encode addresses by combining using different wavelength labels. For instance, by using labels at two different wavelengths, we can encode 4 different addresses. Figure 1 also shows that each of the optical labels have the same duration as the packet payload. Using in-band labels has as an advantage that the labels can be extracted by using narrow-band optical band-pass filters at the expense of a very low penalty. Moreover, using labels that have the same duration as the packet-payload make the use of an optical flip-flop redundant. In the demonstration we used two labels that allow for encoding four addresses. The labels are at wavelengths λL1=1551.9 nm and λL2=1552.5 nm which are within the 20-dB bandwidth of the packet-payload.
Figure 1. Experimental set-up.
The packet switch consists of an all-optical label extractor, an all-optical label processor and a wavelength converter. The packets that input the packet switch are firstly processed by the label extractor. The label extractor consists of two fiber (reflective) Bragg gratings (FBG) with a 3 dB bandwidth of 0.12 nm and 0.432 nm centred around λL1 and λL2 respectively. The 20 dB bandwidths of the FBGs were of 0.8 nm and the insertion losses were 0.7 dB. The filter profiles were shown in Figure 2a, and the filtered 160 Gb/s payload in figure 2b. The label extractor separates the labels from the payload. If a label at wavelength λL1 is present, an output signal appears that OC1; if a label at wavelength λL2 is present, an output signal appears that OC2. If both labels are present, output signals appear at both outputs. The packet payload passes through the label extractor and is fed into the wavelength converter. The extracted labels are fed into the label processor that consists out of two cascaded SOA-MZIs. SOA-MZI 1 is biased with four CW-signals. Two CW-signals enter via the first port and two CWsignals enter via the other port. The SOA-MZI acts as a very fast wavelength selective switch. If a label at wavelength λL1 is present at the output SOA-MZI1, the pairs of CW-signals output the fists SOA-MZI at reversed order. At the SOA-MZI1, an AWG and two 50:50 optical couplers are used to separate the pair of CW-signals and fed to the two distinct ports of the SOA-MZI2. The SOA-MZI2, based on the label value, will provide at label processor output a distinct wavelength. The selected wavelength is then fed into the 160 Gb/s wavelength converter and represents the packet’s routing wavelength. The experimental set-up of the 1x4 packet switching demonstration employing two bit labels is illustrated in fig.1. We processed four packets with two label bits with pattern ‘0 0’, ‘0 1, ‘1 0’, and ‘1 1’ to cover the all possible combinations. The label bits extracted by the FBGs are shown in fig. 2c-d. The two labels are fed in the optical switches with an optical power of 1.5 dBm for label 1 and 0.3 dBm for label 2. We employed as optical switches one monolithically integrated SOA-MZI (SOA-SOA1) and one hybrid integrated version (SOA-MZI2). The input optical power per channel in the SOA-MZI1 was -2.5 dBm. The bias current of SSOA-MZI1 was 204 mA and 216 mA for SOA1 and SOA2, respectively. The dynamic extinction ratio was higher than 16 dB, and the efficiency was around -2 dB. The measured OSNR at the SOA-MZI1 output was 37 dB. The insertion loss was around 6 dB and the isolation between channels higher than 25 dB. The SOA1 and SOA2 of the second MZI-SOA2 were driven by 259 mA and 282 mA of current, respectively. The dynamic extinction ratio was 13 dB. The measured OSNR at the SOA-MZI2 output was 32 dB. The output power per channel (measured by using the OSA) after SOA-MZI2 was 0.2 dBm, which results in an amplification of 5 dB and thus losses compensation of the periodic demultiplexer. The output of SOA-MZI2 is coupled with the 160 Gb/s payload in the wavelength converter based on ultra-fast chirp dynamics in a single SOA [3]. At the receiver side the converted packets are demultiplexed by using an EAM to create 5 ps switching window and then analysed. 3. Results and Discussions Figure 2c-d report the time-domain extracted label 1 and label2, respectively. The measured extinction ratio was above 15 dB. Figure Figure 2e-h show the output of label processor. Figure 2i-l show the switched packets at different wavelengths for different label bits combination (data were recorded tuning the filter of the wavelength conversion at each of the packet’s wavelength routing). Figure 3a-d report the eye diagrams in different point of the packet switch. Small degradation and broadening of the pulses is observed after the label extraction (fig. 3b) with respect to the 160 Gb/s input payload pulses (fig 3a). This is confirmed and quantified by the BER measurement reported in fig 3e. The label extractor produces a penalty of less than 0.5 dB. After the wavelength converter, error-
free operation was obtained with 5.5-7 dB of penalty compared to the input payload, and 1.5-3 dB of additional penalty if compared with the back-to-back 160 Gb/s wavelength converter payload. The extra penalty can be ascribed to the pulse broadening after the label extractor which affects the wavelength converter performance and than produces cross- talk between adjacent time channels. This is also visible by comparing the eye diagrams in fig 3c-d. We demonstrated error-free operation of 1x4 all-optical packet switch at 160 Gb/s data payload by all-optically processing two label bits in combination with wavelength conversion over 20 nm of band.
Figure 2. Traces of (a-b) spectra of the packet at the input and after the label extractor; (c-d) the recovered bit1 and bits labels; (e-h) Switched packets at different wavelengths for different label bits combination. The vertical scale are in Volts.
Figure 3. Eye diagram of (a) input data payload; (b) 160 Gb/s after the label extractor; (c) back-to-back converted 160 Gb/s payload; (d) Typical 160 Gb/s data payoad after the packet switch. (e) BER measurements for different point in the packet switch system.
4. References [1] D. J. Blumenthal et al., “All-optical label swapping networks and technologies”, JLT 18, 2058 (2000). [2] J. Herrera et al., “160 Gb/s all-optical packet switched network operation over 110km of field installed fiber”, Proc. OFC2007, PDP4(2007). [3] Y. Liu et al., “Error-free all-optical wavelength conversion at 160 Gb/s using a semiconductor optical amplifier and an optical band-pass filter”, JLT 24, 230 (2006).
