1. Introduction

Optical performance monitoring (OPM) is essential for next generation reconfigurable optical communication systems. Typical roles for OPM include enabling both quality of service and fault management in highly transparent networks [1,2]. For example OPM is used to monitor transmitted signals and quantify their degradation. These monitoring features are then exploited for impairment mitigation, or to trigger alarms for traffic re-routing. Ultimately an OPM scheme that can provide high sensitivity to impairments and multi-impairment monitoring capability at all operating bitrates is highly desirable.

Current OPM methods based on conventional opto-electronic devices, such as coherent detection [3,4], asynchronous delay tap sampling [5], in-band tone monitoring [6], orthogonal delayed-homodyne [7] and single in-band subcarrier tone methods [8] are mature and relatively effective for the measurement of signal impairments. These techniques, however, rely on fast opto-electronic devices to perform optical-to-electrical conversion as part of the monitoring process. Therefore, when data rates exceed 40 Gb/s per channel these electronic OPM schemes become impractical due to the limitation imposed by the electronic bandwidth.

All-optical performance monitoring based on nonlinear optics, which does not require high-speed electronic devices, is an alternative and attractive solution for high bitrates of 40 Gb/s and beyond which provides ultrahigh resolution due to the femtosecond response time of Kerr nonlinear effect [9]. Various all-optical architectures have been demonstrated to measure impairments of a signal under test (SUT), namely radio-frequency (RF) spectrum monitoring based on cross-phase modulation (XPM) [10,11], cascaded four-wave mixing [12], semiconductor optical amplifier based optical regeneration [13], a nonlinear optical loop mirror [14] and data stream intensity autocorrelation [15]. However these techniques have limited dynamic range or do not perform simultaneous multi-impairment monitoring, which is critical in optical networks.

We recently demonstrated an OPM technique to simultaneously monitor group velocity dispersion (GVD) and in-band optical signal to noise ratio (OSNR) of an optical signal [16]. Our approach was based on measurement of the RF spectrum obtained using cross-phase modulation. We then retrieved the autocorrelation (AC) waveform of the SUT which yielded the individual impairments. Although impressive results were obtained, the low nonlinear response of silica highly nonlinear fiber (HNLF) meant its length had to be hundreds of meters to provide sufficient nonlinearity and this led to significant walk-off between the SUT and the cw probe [9] notably reducing the measurement bandwidth. Furthermore, a fiber approach is incompatible with integration and does not offer a low cost solution. Photonic integration, on the other hand, would offer a compact and low-cost approach and the potential to incorporate multiple functions on a monolithic platform.

In this paper we present a novel multiple impairment all-optical performance scheme based on a photonic-chip based RF spectrum analyzer (PC-RFSA) for ultrahigh bitrates. We employ a 7-cm highly nonlinear (γ ≈9900 /W/km), dispersion-engineered ChG waveguide chip, which offers dramatically increased bandwidth due to reduced walk-off [11]. We apply this scheme to monitor GVD, in-band OSNR and timing jitter of 640 Gb/s return-to-zero (RZ) signals with an OSNR measurement dynamic range of greater than 30 dB. Our approach allows flexible wavelength operation, high measurement sensitivity without complex interferometric alignment. It also offers the capability for multi-impairment monitoring of up to Tb/s optical signals from a single measurement.

2. Working principle and waveguide

There are two main components to this multi-impairment monitoring approach. The first is to capture the RF spectrum and autocorrelation (AC) trace of a signal under test (SUT) from a single measurement with the PC-RFSA. The second is the monitoring algorithm used to extract impairment signatures from the RF spectra/AC traces.

Figure 1(a) shows the conceptual diagram to obtain the RF spectrum and AC waveform of the SUT [10,11,16]. Firstly, the RF spectrum is captured by co-propagating the signal at frequency f_s with a weaker cw probe at frequency f_p through the nonlinear waveguide. The sidebands generated around the probe due to XPM [9] are proportional to the power spectrum of the signal intensity [10,11]. Thus the RF spectrum of a SUT can be fully captured on an optical spectrum analyzer (OSA). Based on the Weiner-Khintchine theorem, we utilize the complete RF spectrum to generate the AC waveform by a numerical inverse Fourier transform (IFT) operation [17,18].

 

Fig. 1 (a) Schematic of simultaneous multi-impairment monitoring based on THz bandwidth PC-RFSA [11,16]. IFT relates the RF spectrum to the AC waveform. (b) (Upper) Optical micrograph of ChG planar waveguide cleaved facet and (lower) circuit schematic of the 7cm chip.

Download Full Size | PDF

We employ two techniques to enhance the AC waveform reconstruction. Firstly, we perform a measurement calibration to eliminate the background in the retrieved AC trace of the un-distorted SUT (with no GVD and high OSNR) based on the fact that the ratio between autocorrelation and cross-correlation (XC) peaks is approximately 2 in a true PRBS sequence. We then applied this for other AC waveform reconstructions, without the need of considering the autocorrelation and XC peaks ratio again. Secondly, the temporal resolution of the experimental AC waveform, which is determined by the sampling point time step [18], is improved by extending the bandwidth of the captured RF spectrum with a numerically generated noise floor.

The key device to our approach is a dispersion-shifted As₂S₃ planar waveguide which offers both high Kerr nonlinearity and low dispersion over the C- and L-bands. Chip fabrication involves depositing a 0.85 μm thick film of As₂S₃ on an oxidized silicon wafer. A 7 cm long straight rib (2 μm wide), whose optical micrograph of cleaved facet and circuit schematic are shown in Fig. 1(b), is formed by photolithography and dry-etching. The chip is protected by a polymer layer before depositing an anti-reflection coating on the cleaved facets to minimize Fabry-Perot reflections. The effective core area A_eff of the fundamental transverse magnetic (TM) mode of this waveguide is reduced to ~1 μm² to enhance its nonlinearity (γ ~9900 /W/km) and shift the zero dispersion wavelength closer to 1550 nm. This short and low dispersion waveguide reduces the walk-off [9], permitting high measurement bandwidth. This permits a broad bandwidth and results in low signal waveform distortion, thus high RF spectral measurement accuracy [18].

Our monitoring approach is based on the relationship between signal impairments and their RF spectra/AC traces, as shown in Fig. 2(a) –2(c). Firstly, GVD disperses the SUT, thus reducing the peak intensity, which leads to interference between adjacent pulses. This produces ripples on the RF spectrum and results in the peak intensity at zero delay dropping together with an increase of the dips at T/2 and 3T/2 delays on the AC waveform [15]. ASE noise, on the other hand, decreases XPM efficiency and increases the noise floor on the RF spectrum, thus creating a constant background intensity on the AC trace [15]. Timing jitter of a SUT produces phase noise [19] reducing the power of higher order tones on the RF spectrum. It also broadens the width of the cross-correlated pulses whilst not affecting the zero-delay autocorrelation waveform [15,20]. Note the simulation results are discussed assuming that the pulse shape is Gaussian.

 

Fig. 2 Numerical (upper) RF spectra and (lower) AC waveforms of undistorted (solid) and distorted (dotted) signal due to (a) dispersion, (b) ASE noise, and (c) timing jitter. T is the bit period of the simulated SUT.

Download Full Size | PDF

We propose an algorithm to determine the impairments in three key steps. In step one, the dispersion is obtained from the AC trace. Once GVD is determined, the OSNR is detected in step two by measuring the optical clock tone captured from the PC-RFSA to take advantage of its much higher dynamic range. In step three, the root-mean-square (rms) value of timing jitter is calculated from the reconstructed AC waveform as reported in [20,21]. Here we introduce a more robust timing jitter calculation (assuming a SUT has Gaussian shape) which is expressed in terms of full-width at 90% maximum T_90%-AC and T_90%-XC of the autocorrelation and cross-correlation peaks$J_{rms}=1/\left[4\sqrt{-\ln \left(0.9\right)}\right]\times \sqrt{T_{90\%-XC}^2-T_{90\%-AC}^2}$.

3. Experiment

Our experimental setup is shown in Fig. 3(a) . A 40 GHz train containing ~550 fs duration pulses (λ_s = 1543 nm) was generated from a mode-locked fiber laser (MLFL) followed by nonlinear pulse compression. This was encoded with data using a Mach-Zehnder (MZ) modulator driven electrically by a 40 Gb/s pseudo-random bit sequence (PRBS) of 2³¹ - 1 pattern length. Optical time-division multiplexing (MUX) in a four-stage fiber delay line interferometer of 2⁷ - 1 bit delay produced a 640 Gb/s signal (~35% duty cycle). The eye diagram and the optical spectrum of this 640 Gb/s signal is shown in Fig. 3(i) and Fig. 3(b), respectively. Impairments were simulated by adding GVD from a Finisar WaveShaper [22] and ASE noise from an erbium doped fiber amplifier (EDFA). The SUT and polarized ASE noise (after the in-line polarizer) were filtered with 8 nm bandwidth filters before co-propagating with a cw probe (that was also filtered by a 0.7 nm filter). The distorted 640 Gb/s optical signal (P_ave = 70 mW) and a cw probe (P_ave = 30 mW, λ_p = 1580 nm) were launched into a 7-cm-long dispersion-engineered As₂S₃ planar waveguide (γ ≈9900 /W/km, dispersion D ≈28 ps/nm/km at 1550 nm). Note the polarization controllers (PCs) help to couple to the fundamental TM mode of the waveguide (insertion loss ≈14.2 dB) via a pair of lensed fibers with 2.5 µm spot diameter. An example of the captured RF spectrum and the numerically reconstructed AC waveform of a 640 Gb/s optical signal is shown in Fig. 3(c), 3(d), respectively.

 

Fig. 3 (a) Experimental setup for simultaneous multi-impairment monitoring; (i) an eye diagram of a 640 Gb/s optical time division multiplexing source signal. (b) The optical spectrum, (c) the captured RF spectrum via PC-RFSA and (d) corresponding reconstructed AC waveform of a 640 Gb/s optical signal.

Download Full Size | PDF

Figure 4 experimentally demonstrates the effect of dispersion and ASE noise impairments on the AC waveforms. Here we show contours plots of the normalized AC traces reconstructed from the captured RF spectra for various GVD and OSNR values. We define I_a, I_b and I_c as AC power at zero, 0.5T and 1.5T delays respectively, where T is the bit period of a 640 Gb/s signal. In the case of increasing GVD from 0 to 0.16 ps/nm (with OSNR ~40 dB), the AC peak intensity (I_a - I_b) reduces while the bit asymmetry (I_b – I_c) increases as shown in Fig. 4(a). These trends reverse as the GVD is increased further from 0.16 to 0.21 ps/nm due to the reconstruction of periodic signals that can be explained by the temporal Talbot effect [23]. In contrast to GVD, the AC peak intensity increases while bit asymmetry remains fairly constant when the OSNR is reduced from 40 to 0 dB (with GVD ~0 ps/nm) [Fig. 4(b)]. Similar to Fig. 4(b), 4(c) depicts AC traces with varying OSNR, but with GVD ~0.08 ps/nm. Here we can see different features compared with the case of zero GVD. To summarize, different forms and amounts of impairments reflect differently on the corresponding AC waveforms, and this is exploited for multi-impairment OPM.

 

Fig. 4 (a) Experimental results showing the top view of the retrieval AC waveforms of 640 Gb/s optical signals while varying (a) GVD (OSNR ~40 dB), (b) OSNR (GVD ~0 ps/nm) and (c) OSNR (GVD = 0.08 ps/nm).

Download Full Size | PDF

We apply this monitoring scheme to simultaneously monitor impairments on a 640 Gb/s signal caused by GVD, ASE noise and timing jitter. Firstly, GVD is distinguished based on the specific features of the AC traces described above. In Fig. 5(a) , we map the AC peak intensity (I_a – I_b) against the AC bit asymmetry (I_b – I_c). Each point on this map uniquely defines both OSNR and GVD values; thus we can determine both GVD and OSNR values from this graph. However, the dynamic range of the OSNR measurement obtained from the AC traces is small (< 3 dB [15]) compared to > 30 dB output range measured from the fundamental 640 GHz clock tone of the RF spectrum, as shown in Fig. 5(b). Note that GVD = 0 ps/nm in the OSNR measurements shown in Fig. 5(b). Therefore, once GVD has been determined in step one; we enhance the OSNR monitoring performance by measuring the optical tone power (plotted in Fig. 5(c) as a function of OSNR). We also note that all GVD and OSNR used in these measurements are true values as they were calibrated and measured directly from the waveshaper and OSA respectively.

 

Fig. 5 (a) Step 1 – GVD measurement (GVD = 0 (diamond), 0.08 (circle) and 0.16 (triangle) ps/nm). The arrows show directions of increasing OSNR values. (b) OSNR monitoring using PC-RFSA and data stream AC [15] methods, highlighting very large dynamic range of PC-RFSA scheme. The lines are to guide the eye. (c) Step 2 – OSNR monitoring via optical tone power once GVD is determined. (d) Step 3 – timing jitter measurements once GVD and OSNR are obtained from step 1 and 2.

Download Full Size | PDF

Finally, timing jitter results are plotted in Fig. 5(d), showing a low timing jitter of ~100 fs (in the case of no other impairments) for our 640 Gb/s optical signal. Furthermore, the impact of the GVD and OSNR degradation on the quality of the SUT is also demonstrated. Figure 5(d) clearly shows that with increasing GVD and ASE noise, timing jitter increases as we expect. With the appearance of other impairments such as GVD and ASE noise, the true timing jitter J_true is obtained from both J_measure and J_distort as$J_{true}=\sqrt{J_{measure}^2-J_{distort}^2}$ [24], where J_measure is calculated from the experimental AC waveform and J_distort refers to a numerically calculated timing jitter of a simulated zero-jitter-source with the same GVD and OSNR as measured previously. For comparison, Fig. 5(d) also depicts timing jitter measurements using an optical sampling oscilloscope (OSO), showing a good agreement between two methods.

4. Discussion

There are significant performance advantages of this PC-RFSA based multi-impairment monitoring scheme. Firstly, analyzing the AC waveform obtained from a broadband RF spectrum measurement offers simultaneous multi-impairment measurement without ambiguities and a large OSNR measurement dynamic range of > 30 dB. Conventional AC measurements are limited by the dynamic range of the oscilloscope trace and are less robust due to complex interferometric alignment. Secondly, the PC-RFSA based on the short, highly nonlinear and dispersion-shifted planar waveguide offers broad bandwdith (> 2 THz) due to very low walk-off between the SUT and the cw probe leading to high measurement accuracy and good temporal resolution of tens of femtoseconds [18]. This allows the RF spectra/AC waveforms of 640 Gb/s signals to be captured accurately. Thirdly, the timing jitter results obtained using the PC-RFSA method represent the true timing jitter in the system, while the jitter measured using an optical sampling oscilloscope (OSO) is affected by the influence of amplitude fluctuations [25]. The limited temporal resolution of the OSO (> 600 fs) not only restricts the operating bitrates but also produces jitter measurement errors for data rates of ≥ 640 Gb/s.

We note that the GVD monitoring window for a 640 Gb/s RZ signal is currently limited to about ± 0.16 ps/nm because of the reconstruction of periodic signals. However, this monitoring window can be expanded by further analyzing the RF spectra/AC waveforms. The current monitoring speed is ultimately determined by the sweeping time of the OSA. Finally, the required input power can be significantly reduced by decreasing insertion loss and enhancing the nonlinear coefficient of the ChG planar waveguide. Placing the monitor after an in-line EDFA within the network will further reduce the power requirement.

5. Conclusion

We have demonstrated simultaneous multi-impairment monitoring of 640 Gb/s optical signals based on PC-RFSA. The novel feature of this approach is the combination of the captured RF spectra and AC traces, which provides excellent output dynamic range and resolves the measurement ambiguities arising from simply measuring the optical tone power from the RF spectrum. By exploiting the high nonlinearity and low dispersion of a ChG photonic chip, large measurement bandwidth, high accuracy/low signal distortion and operating wavelength flexibility have been achieved. These features allow the capture of RF spectra/AC waveforms of broadband optical signals within the whole C- and L-band correctly. Finally, we have demonstrated the effectiveness of this scheme as a useful diagnostic tool for multi-impairment monitoring of up to terabit data rates.