All-optical clock recovery (CR), in general, will potentially be used when data rates of optical communication systems exceed electronic capabilities. One of the applications of all-optical clock recovery is OTDM demultiplexing, where the OTDM line rate is higher than the electronic bottleneck. Both optoelectronic phase-locked loops (PLL) [1, 2] and passively mode-locked semiconductor lasers MLL [3] have been demonstrated for this purpose. These approaches recover a subharmonic clock signal, and hence are not suitable for another application of all-optical clock recovery, namely, optical 3R (retiming, reshaping and reamplification) regeneration, which requires a recovered clock at the optical line rate.

High-speed optical CR at the optical line rate has been realized using actively mode locked fiber lasers [4, 5] and passively mode locked semiconductor lasers [6]. Fiber lasers have been shown to be able to operate above 100 Gb/s but suffer from high power requirement and are very complicated. Passively mode-locked semiconductor lasers have shown to operate up to 40 Gb/s. An alternative approach is to use self-pulsing gain-coupled DFB lasers [7]. The two-section gain-coupled DFB laser has two sections that share a common substrate. The contacts for the two sections are electrically isolated. Unlike index-coupled DFB lasers, the gain-coupled DFB grating is etched into the gain region [8]. Gain coupling leads to a large (> 40 dB) single-mode suppression ratio (SMSR) in single-section DFB lasers. For two-section gain-coupled DFB lasers, gain coupling results in relatively independent lasing in the two sections. As a result proper design and operation of the device lead to self pulsing: periodic intensity output with DC bias currents for both sections [9].

Figure 1. Schematic and mechanism of incoherent clock recovery.

Experiments have shown that self pulsing can be injection locked to optical clock signals embedded in data streams thus realizing the clock recovery function. There are three modes of operation for all-optical clock recovery using two-section gain-coupled DFB lasers [10]. The first mode is the so-called the incoherent clock recovery as shown in Figure 1. The CR circuitry is characterized by the following: 1) Injection data is carried on a wavelength that is at least 1 nm from that of the free running TS DFB, and 2) Injection power is >1 mW. The mechanism of CR is also shown in Figure 1. Injection data creates carrier modulation at the clock frequency in the front section of the two-section DFB laser. This carrier modulation creates sidebands from the mode in the front section. One of the resulting sidebands all-optically injection locks the mode in the back section. Beating of the two modes results in a recovered clock. The advantages of this mode of CR are 1) it is wavelength insensitive as the CR mechanism depends on intensity-induced carrier modulations, and 2) it can be potentially polarization insensitive if the quantum-well gain region is properly strained. The disadvantages of this mode of CR are 1) it requires relatively high injection power, i.e., the sensitivity of CR is low, and 2) jitter of the recovered clock is relatively high.

Figure 2. Schematic and mechanism of coherent clock recovery.

The second mode is the so-called coherent clock recovery as shown in Figure 2. The CR circuitry is characterized by the following: 1) Injection data carried on almost the same wavelength as that of the free-running TS DFB, and 2) Injection power is . The mechanism of CR is also shown in Figure 2. The injection data contains sidebands due to existence of the clock component. One of the sidebands is aligned with the front section and the optical carrier is aligned with the back section. Optical injection locking takes place in both sections. Beating of the two optically injection locked modes results in a recovered clock. The advantages of this mode of CR are 1) it requires extreme low injection power, i.e., the sensitivity of CR is high, independent of the bit rate and 2) the jitter of recovered clock is very small. The disadvantages of this mode of CR are 1) it is wavelength insensitive, and 2) it is polarization insensitive even if the quantum-well gain region is properly strained.

Figure 3. Schematic of wavelength- and polarization-insensitive coherent clock recovery.

The third mode is the so-called wavelength- and polarization-insensitive coherent clock recovery as shown in Figure 3. The injection data is carried on arbitrary wavelength. Probe wavelength is nominally the same as free-running TS DFB Laser. The semiconductor optical amplifier (SOA) is both wavelength- and polarization-insensitive. The injection power into the SOA is < 100 mW. The mechanism for this wavelength- and polarization-insensitive coherent clock recovery scheme is straightforward. Clock component carried on the arbitrary wavelength of the injection data is converted to probe wavelength, which is the same as the free-running DFB laser. The wavelength conversion is insensitive to both wavelength and polarization. After the bandpass filter, CR operates in the coherent mode, identical to the coherent mode as shown in Figure 2.

The wavelength and polarization-insensitive coherent clock recovery is the preferred mode of operation since it combines the advantages of the previous two modes. It is both wavelength and polarization insensitive and it offers low jitter. Compared to the incoherent clock recovery scheme, the only additional component is the probe laser. This wavelength- and polarization-insensitive coherent clock recovery circuitry has demonstrated the following characteristics simultaneously:

• Recovered clock signal should have low jitter (<1 ps at 40 Gb/s and < 250 fs at 100 Gb/s)
• high sensitivity and large dynamic range (-10 dBm and 7 dB, respectively)
• high-speed operation above electronic speed limit (at least 100 Gb/s)
• wavelength insensitive
• polarization insensitive
• a fast lock up time for reduced latency (<1ns)
• a long hold time to eliminate pattern-dependent effects (5 ns)
• simple, compact, low-power consumption and reliable

Figure 4. Eye diagram of the input data signal (a) and waveform of the recovered clock (b) at 96 Gb/s.

Figure 4 contains time-domain measurements of the input data and the recovered clock at 96 Gb/s made possible by the recent availability of the Terascope by Agilent. The eye diagram of the input data is shown in Fig. 4(a). Some non-uniformity can be observed in the 8 optical time-division multiplexed (OTDM) channels at 12 Gb/s, each of which is a pseudo-random bit stream (PRBS) with a pattern length of . The recovered clock at 96 GHz is shown in Fig. 4 (b). The clock recovery circuitry successfully removed the pattern effect in the OTDM channels. The RMS timing jitters of the recovered clock was 499 fs compared with the 436 fs RMS jitter of the original OTDM signal. The jitter of the original OTDM signal is mainly due to OTDM multiplexing. Assuming that the jitter of the original OTDM signal and the intrinsic jitter induced by the clock recovery circuitry (wavelength converter and the TS-DFB laser) are the only jitter sources of the recovered clock and they are statistically independent, the intrinsic jitter of the clock recovery circuitry is fs, which is very close to the instrument timing jitter (200fs) of the Terascope.

This work has been supported by the National Science Foundation.

1. D. T. K. Tong, et. al., IEEE Photon. Technol. Lett., vol. 12, pp. 1064-1066, Aug. 2000.

2. O. Kamatani, et. al., IEEE Photon. Technol. Lett., vol. 8, pp. 1094 -1096, Aug. 1996.

3. I. Ogura, et. al., ECOC’97 Technical Dig., vol. 2, pp. 77-80.

4. S. Bigo, et. al., IEEE J. Select. Topics Quantum Electron., vol. 3, pp. 1208-1223, Oct. 1997.

5. H. Yokoyama, et. al., Proc. CLEO/pacific Rim 2001, vol. 2, pp. 498-499.

6. X. Wang, et. al., J. Opt. Soc. Am. B, vol.16, pp. 2030-2039, Nov. 1999.

7. W. Mao, et. al., Electron. Lett., vol. 37, pp. 1302 –1303, Oct. 2001.

8. B. Sartorius, et. al., IEEE J. Select. Topics Quantum Electron., vol. 1, pp.535-538, June

1995.

9. C. Bornholdt, et. al., OFC’2002 Technical Dig., Paper TuN6, pp. 87-89.
10. W. Mao, et. al., to appear in IEEE J. Lightwave Technol., Sept. 2002.