1. Introduction

With the scalability and power consumption issues faced by electrical routers, optical packet switching (OPS) has been raised as a promising solution for future communications networks [1,2]. OPS technologies have the potential to enable flexible networks with low latency, low power consumption, and high bandwidth utilization. Realizing OPS networks requires optical packet routers that perform forwarding functions of label processing, switching, and buffering for high-speed asynchronous variable-length optical packets and would thus enable flexible traffic engineering on a packet-by-packet basis with low latency and power consumption.

We have developed a hybrid optoelectronic router prototype that efficiently utilizes both optical and electrical technologies. The router is implemented by combining key optical/optoelectronic device/sub-system technologies and complementary metal–oxide–semiconductor (CMOS) electronics in a novel router architecture. The router performs the above mentioned functions with reduced power and latency [3,4]. As shown in Fig. 1, the packet data is organized into wavelength layers, separated and combined with arrayed waveguide gratings (AWGs) at the system inputs and outputs ports. The incoming packet passes through the label processor [5] that is followed by an optical switch. If there is no contention caused by other incoming packets, the packet is routed through the switch to the desired output port without buffering. If there is contention, the packet is forwarded to a shared buffer. The shared buffer consists of optoelectronic interface devices and an electronic (CMOS) memory at the core. The interface devices are all-optical serial-to-parallel converters (SPCs) [1,6] at the inputs, parallel-to-serial converters (PSCs) [5] at the outputs, and optical clock pulse-train generators (OCPTGs) [7] to drive these converters. In the prototype router for 10-Gbit/s optical packets, the SPC, PSC, and OCPTG transform 10-Gbit/s burst optical packet signals into successive 16-parallel electrical signals at 625 Mbit/s and vice versa. Lowering the bit rate of the high-speed optical packets enables buffering the packets with CMOS circuitry [8]. Beside performing buffering for contention resolution, the shared buffer can also perform various functions based on the CMOS capabilities, such as switching between wavelength planes, 3R regeneration with error correction [9], and other high-level processing for various services [multicast, Quality of Services (QoS), etc.]. The selective buffering using the shared buffer reduces the power and latency of the router.

 

Fig. 1 Hybrid optoelectronic router architecture. AWG: arrayed waveguide grating. SPC: serial-to-parallel converter. PSC: parallel-to-serial converter. OCPTG: optical clock pulse-train generator.

Download Full Size | PDF

The OCPTG consists of an optical clock pulse generator (OCG) and a pulse train generator (PTG). The OCG generates a single optical pulse synchronized to the incoming preamble-free, asynchronous optical packet [7]. The fiber-loop-based PTG then converts the single pulse to a pulse train having a repetition rate the same as the line rate divided by the number of SPC/PSC channels and a duration covering the whole packet length. In the prototype router, the PTG with a 6.4-ns repetition cycle consisted of a ~128-cm-long fiber loop that includes an erbium-doped fiber amplifier (EDFA) and a 2x2 lithium niobate (LN) switch [7]. The EDFA’s gain cannot be modulated fast because of its slow transient response. Therefore, a fast modulator or switch like the LN is needed to define the duration of the output pulse train by controlling the loop gain according to the packet length. Although the EDFA-based PTG provides low-noise characteristics, it requires further additional control circuits to be able to maintain stable operation, such as a loop gain stabilizer and an LN bias controller, resulting in unwanted increases in size and power consumption. It is also difficult to shorten the loop length to cope with optical packets at higher bit rates.

In this paper, we propose a novel PTG that utilizes a semiconductor optical amplifier (SOA) and a spin-polarized saturable absorber (SA) [6]. The SOA is directly driven by the packet envelope signal to set the length of the output pulse train, thus eliminating the need for the LN switch. By means of a negative feedback mechanism due to gain saturation in the SOA, the PTG is self-stabilizing without any other external control circuits. The SA and a backward optical pulse launched into the SOA effectively suppress the amplified spontaneous emission (ASE), and thus enable output pulses with reduced noise despite the large ASE of the SOA. Moreover, since the PTG uses an SOA instead of the long EDFA, it is possible to reduce the loop length for optical packets at higher bit rates.

2. Self-stabilizing pulse-train generator

Figure 2 shows the proposed PTG. A single seed optical pulse generated by the OCG is coupled into the PTG main loop with a 3-dB coupler. The pulse keeps circulating in the forward direction and also generates a backward pulse at the beginning of each round trip by the same input coupler. The forward pulse is amplified by the SOA whereas the backward pulse is injected into the SOA from the backside through a circulator. The optical loop includes an isolator to prevent the reverse circulation of the backward pulse and a bandpass filter (BPF) to filter out the out-of-band ASE of the SOA. A 4:1 coupler is used to tap out a small part of the forward pulse as the output train pulse whereas the pulse remaining in the loop gets circularly polarized before arriving at the SA. The forward pulse reflected from the SA is returned to its original polarization state, passes through an attenuator (ATT) for the rough adjustment of the loop gain to unity, and then loops back to the 3-dB coupler to form the optical loop. As long as the SOA is activated by a packet envelope current signal, the circulation is sustained and an optical pulse is output periodically at a repetition rate determined by the loop length and with a duration corresponding to the packet length.

 

Fig. 2 Proposed SOA-SA-based pulse train generator. SOA: semiconductor optical amplifier. SA: saturable absorber. BPF: bandpass filter. PC: polarization controller. ATT: attenuator.

Download Full Size | PDF

The three conditions for the stable operation are as follows: 1) The loop gain against the circulating pulse is unity. 2) The loop gain against the ASE is lower than unity. 3) The slope of the loop gain with respect to input pulse energy is negative at the operation point where the loop gain is unity. In order to meet the first condition, the gain of the SOA against the circulating pulse must be larger than the gain that compensates for the loss of other components, such as the SA. The second condition is a crucial one to suppress the ASE noise. The third condition is essential for the self-stabilizing operation.

Figure 3(a) shows the SOA gain profile as a function of input pulse energy (1539.5-nm, 10-ps input pulse). The SOA used here has a small signal gain of over 20 dB at 250 mA and exhibits clear gain saturation with increasing pulse energy, i.e., a negative slope of the gain with respect to input pulse energy. The SA has an impedance-matched asymmetric Fabry-Perot (AFP) etalon structure [Fig. 3(b)], where 1-μm-thick 1% compressively strained InGaAs/InAlAs multiple quantum wells (MQWs) are sandwiched between a bottom Au mirror and an upper distributed Bragg reflector (DBR) with 14% reflectivity [1,6]. The compressive strain enhances absorption saturation because it lowers the density of states in the valence band of the MQWs; hence, absorption requires fewer carriers to reach saturation. In addition, the strain causes a split perfectly between the heavy-hole and light-hole excitons. This allows selective spin excitation of the heavy-hole excitons; i.e., only the spin-down (or spin-up) carriers are excited when circularly polarized optical pulses irradiate, enabling further enhancement of absorption saturation. As shown in the temporal gain profiles measured by a pump-probe technique [Fig. 4(b)], a large transmission peak is obtained with a circularly polarized pulse (solid line) in comparison with a linearly polarized pulse (dashed line). Furthermore, the impedance-matched AFP structure improves the extinction ratio because in the low-energy region, the light reflected from the DBR cancels out the light from the bottom mirror through destructive interference. As a result of these improvements, the SA provides a sufficiently high transmittance against the circulating pulse and an extremely high extinction ratio against the ASE [Fig. 3(c)].

 

Fig. 3 Gain as a function of input pulse energy for the (a) SOA, (c) SA, and (d) SOA followed by SA, respectively. (b) SA structure.

Download Full Size | PDF

 

Fig. 4 Temporal gain profiles measured by a pump-probe technique for the (a) SOA, (b) SA, and (c) SOA followed by SA. LG: loop gain.

Download Full Size | PDF

Combining the SOA and SA thus provides the desired gain profile as shown in Fig. 3(d). Here, the loop gain at an input pulse energy of 0.1 pJ is adjusted to unity by the ATT. In the low-energy region, the loop gain is below −10 dB due to the strong absorption by the SA; providing effective suppression for the ASE noise. With increasing the pulse energy, the loop gain increases due to absorption saturation in the SA, reaching a peak, followed by a negative slope due to the gain saturation of the SOA. Such a negative slope of the gain curve causes a negative feedback that creates a stable operation point where the loop gain is self-stabilized to unity. A circulating pulse, whose energy is higher/lower than what corresponds to the stable point, encounters a loop gain lower/higher than unity. Thus even if the input pulse energy is different from the value that corresponds to the stable operation point, the operation point converges to the stable point as the optical pulse circulates in the optical loop several times.

The reduction of the ASE noise before the optical pulse is confirmed from Fig. 3(d). Another concern is how the ASE is reduced after the optical pulse. This issue arises from a recovery time difference between the SOA and SA. Figure 4 shows temporal gain profiles measured by a pump-probe technique. The gain of the SOA is suppressed by an input pulse (pump pulse) and recovers gradually at a time constant of 300~400 ps [Fig. 4(a)]. On the other hand, the absorption of the SA is rapidly decreased by the input pulse and recovered according to two time constants: a spin relaxation time of ~80 ps and an inter-band carrier recombination time of 1~2 ns [solid line in Fig. 4(b)]. Although the spin relaxation contributes to fast recovery of the SA, the SA still recovers more slowly than the SOA because of the slow recombination time. As a result, as shown in Fig. 4(c), the total gain of the loop including the SOA and SA exhibits a second peak after the incident pulse. Before the pulse, the loop gain is well suppressed to <<1 due to absorption by the SA. However, after the pulse the second peak causes unwanted amplification of the ASE noise. To avoid this, we introduce an additional pulse injected into the SOA from the backside (blue pulse in Fig. 2) just after the forward circulating pulse (red pulse in Fig. 2). This suppresses the SOA gain again so that the loop gain after the pulse remains lower than unity [solid line in Fig. 5(a)]. As a result, all light except the forward circulating pulse has a loop gain lower than unity and rapidly attenuates as it circulates in the optical loop. All three conditions for the stable operation are thus satisfied by using the SA combined with the SOA with the backward pulse injection.

 

Fig. 5 (a) Temporal gain profile measured by a pump-probe technique for the SOA followed by SA. LG: loop gain. (b)-(e) Output pulse waveforms.

Download Full Size | PDF

3. Experimental results

Experiments for proof of the principle were performed with the setup shown in Fig. 2. The round-trip time of the optical loop was ~32 ns because we used fiber-pigtailed components. Reduction to 1~2 ns will be possible by integrating the components without using long fibers. The input optical pulse was a 1539.5-nm, 10-ps wide pulse from a gain-switched laser. The laser was driven by an electrical pulse fed from a pulse pattern generator (PPG) at a repetition cycle of ~3 μs. The MQWs in the SA had an exciton absorption peak at 1539 nm. Polarization of the circulating pulse was set to be linear (TE mode) at the SOA and adjusted to be circular at the SA for selective spin-polarization excitation. The backward pulse was launched into the SOA ~100 ps after the forward circulating pulse. A rectangular current signal was fed to the SOA as a packet envelope signal that determines the duration of the output pulse train.

Figures 5(b)-5(e) show the output pulse waveforms. The currents fed to the SOA corresponding to the ‘on’ and ‘off’ states of the packet envelope signal were 250 and 0 mA, respectively. The input pulse energy was 0.1 pJ, which corresponded to the stable operation point. Figures 5(b) and 5(c) are the waveforms obtained without a backward pulse after 25 and 50 circulations, respectively. The ASE is well suppressed before the pulse. However, the ASE amplification is observed after the pulse because of the faster recovery of the SOA, as expected from Fig. 4(c). Figures 5(d) and 5(e) were obtained with the backward pulse. The ASE noise after the pulse was successfully suppressed by the backward pulse injection that keeps the loop gain after the pulse at less than unity, as indicated by the measured temporal gain profile [Fig. 5(a)].

Figures 6(a)-6(c) show the waveforms of the obtained optical pulse trains when the input pulse energies were 0.032, 0.1, and 0.32 pJ, respectively. As shown in Fig. 6(b), a stable pulse train consisting of more than fifty pulses with each pulse having almost the same pulse energy was generated when the input pulse energy corresponded to the stable operation point. The output pulse energy was ~0.4 pJ/pulse. As shown in Figs. 6(a) and 6(c), even when the input pulse energy is different from what corresponds to the stable point, the output pulse energies become the same after the first several circulations because of the negative feedback. Thus, the self-stabilizing operation was achieved for a 10-dB dynamic range of input pulse energy. The results confirm that the PTG tolerates an unexpected change in the input pulse energy and in the loop gain during operation as well, with no additional control circuits.

 

Fig. 6 Waveforms of generated pulse train with input pulse energy of (a) 0.032, (b) 0.1, and (c) 0.32 pJ.

Download Full Size | PDF

We evaluated the pulse width, timing jitter, and optical spectrum of the generated pulse train. The current fed to the SOA corresponding to the ‘on’ state of the packet envelope signal was 180 mA, which was set as small as possible under the condition that a stable operation point could be created. With this SOA current, more than 10-dB tolerance in the input pulse energy was also achieved. Lowering the SOA current contributes to reduction of the power consumption and the ASE noise as well. Figures 7(a)-7(d) show oscilloscope traces of the input pulse and the 1st, 25th, and 50th output pulses taken by a 20-GHz sampling oscilloscope which is triggered by a pattern synchronization signal fed from the PPG at ~3-μs repetition cycle. These traces confirm that the multiple circulations did not add significant degradation of the pulse waveform. From the oscilloscope measurements, the timing jitter was evaluated for the individual output pulses as shown in Fig. 7(e). The figure showing the RMS timing jitter as a function of the number of circulations confirms that the jitter does not increase for the output pulse even after fifty circulations, maintaining an initial value of ~1.6 ps. Most of this value is attributed to the measurement system operating with the triggering signal at the fairly long repetition cycle of ~3μs as mentioned above. Figure 7(f) shows the autocorrelation traces for the input pulse (dashed line) and the output pulse train (solid line). Figures 7(g) and 7(h) show the optical spectra of the input pulse and the output pulse train, respectively. Figures 7(g) and 7(h) indicate that the optical spectrum is red-shifted after multiple circulations, which is likely to occur due to self-phase modulation (SPM) in the SOA. We used an 8-nm-bandwidth BPF in the optical loop so that all the circulating pulses that span the broad wavelength spectrum as shown in Fig. 7(h) are not filtered out during circulation. As a result, output pulses had almost the same pulse width as the input pulse, exhibiting only a slightly broaden shape [Fig. 7(f)]. In Fig. 7(e), the FWHM pulse width for the corresponding output pulse extracted from the generated pulse train is also plotted as a function of the number of circulations (assuming a Gaussian waveform). It is confirmed that the pulse width was maintained within the range of 10.0 to 11.5 ps. A stable optical pulse train consisting of more than fifty pulses with low jitter and maintained pulse width is thus generated from a single seed optical pulse by the PTG.

 

Fig. 7 Oscilloscope traces taken for the (a) input pulse and (b), (c), and (d) 1st, 25th, and 50th output pulses. (e) RMS timing jitter and FWHM pulse width as a function of number of circulations. (f) Autocorrelation traces for the input pulse and output pulses. Optical spectrum of the (g) input pulse and (h) output pulse train.

Download Full Size | PDF

We note that the pulse width might have been influenced by the long fibers or the unoptimized BPF used for the current setup. A ~6.4-m long fiber-loop (total transmission length of ~320 m for 50 circulations) could vary the pulse widths of the output pulses because of a chromatic dispersion in the fiber. The non-flat-top BPF could also have affected the pulse width. Integration of the components without using long fibers will not only increase the pulse repetition rate suitable for higher-bit-rate packets but also improve the PTG performance. The integration will reduce optical loss of the loop, which enables both optical pulses with higher energy fed to the SA and reduced SOA current. These will further reduce the ASE noise. The integration will also prevent pulse width variation via dispersion in the long fiber.

4. Conclusion

A novel optical clock pulse-train generator for processing preamble-free, asynchronous optical packets is proposed and demonstrated using an optical loop consisting of an SOA and an SA. The loop gain is self-stabilized to unity because of the negative feedback due to gain saturation in the SOA. The SA, consisting of an asymmetric Fabry-Perot structure with compressively strained spin-polarized MQWs, absorbs the ASE of the SOA with a high extinction ratio while providing a sufficiently high transmittance against the circulating optical pulse because of absorption saturation. The SA and a backward optical pulse injected into the SOA effectively suppress the ASE in the optical loop, resulting in output pulses with reduced noise. An optical pulse-train consisting of more than 50 pulses with low jitter and maintained pulse width is successfully generated from a 10-ps-wide seed optical pulse, with 10-dB tolerance in the seed pulse intensity. The SOA-SA-based PTG does not need additional control circuits for stable operation. In addition, it does not need long-fiber-based components like an EDFA, which enables reduction of the loop length for optical packets at higher bit rates. The PTG is thus promising as a low-power, compact clock generator for processing high-speed, preamble-free asynchronous variable-length optical packets.