1. Introduction

Optical performance monitoring (OPM) is an indispensable function for the efficient operation and management of reconfigurable optical networks [1, 2]. In particular, for the system diagnosis, adaptive channel equalization, and impairment-aware routing, it is crucial to monitor the optical signal-to-noise ratio (OSNR) and chromatic dispersion (CD) of the optical signal. Numerous types of OPM techniques have been proposed so far. Among them, the techniques based on the asynchronous delay-tap sampling (DTS) [3–7] and asynchronous amplitude histogram (AAH) [8–12] have attracted significant attention since they can monitor multiple impairments simultaneously without using the clock-extraction circuitry. However, the DTS-based monitoring technique requires precise adjustment of the delay value between taps in accordance with the symbol rate, which, in turn, hinders the monitoring of the optical signals operating at various data rates [4]. Since future optical networks should accommodate these mixed line rates as well as multiple modulation formats, it would be highly desirable to have the capability of monitoring such various types of signals [13]. In this regard, the AAH-based technique can be an alternative solution for the multi-impairment monitoring since it is transparent to the data rates and modulation formats. In fact, it has been already demonstrated that this technique can be used for the multi-impairment monitoring of the phase-shift keying (PSK) and quadrature amplitude modulation (QAM) signals [8–12]. However, this technique cannot extract the desired monitoring parameters accurately without the prior knowledge of the rising/falling characteristics of transmitters. This is because the AAH-based monitoring technique asynchronously samples the signal’s intensity. As a result, the AAH is bound to have some unwanted samples collected at the symbol’s transitions. Thus, especially when the higher-level modulation format is used, the measured histogram can lose its distinct shape profile for the specific modulation format and channel impairment. To evaluate this effect of the symbol’s transitions on the performance of the AAH-based monitoring technique, we construct AAHs of the 10-Gbaud quadrature phase-shift keying (QPSK) and 16QAM signals generated by using a dual-parallel Mach-Zehnder modulator (MZM), and estimate their OSNRs by analyzing the AAH statistics. For this evaluation, we utilize two transmitters having different bandwidths: Transmitter 1 is modulated by using electrical amplifiers equipped with low-pass filters (LPFs) having a 3-dB bandwidth of 7.46 GHz, whereas no LPF is utilized for Transmitter 2. Figure 1(a) shows the OSNR monitoring parameter, F_OSNR, which is defined as the ratio of the extracted mean to the standard deviation of the sampled 10-Gbaud QPSK signal, as a function of the actual OSNR. The obtained amplitude samples and their corresponding AAH (when the actual OSNR is 20 dB) are also shown on the right-hand side of Fig. 1(a). These figures show that F_OSNR is severely affected by the transmitter bandwidth (i.e., symbol’s transition time). For example, F_OSNR is estimated to be 13.8 dB for Transmitter 2, but it is reduced to be 11.7 dB for Transmitter 1. The similar dependency is observed for the 16QAM signal, as shown in Fig. 1(b). These results imply that, to monitor the signals originating from different transmitters in a reconfigurable optical network, it is necessary to calibrate the monitoring parameters by using the prior knowledge of the transient characteristics of each transmitter. Recently, it has been demonstrated by computer simulations that the algorithm based on the artificial neural network (ANN) can be used for the monitoring of multiple impairments [14]. However, this technique utilizes the whole bins of AAH as the ANN input neurons and needs to train the monitoring module by using every possible combination of impairments (including the transient characteristics of the transmitters). Since this training process should be performed for every transmitter in the network, it may be too complicated for the practical applications.

 

Fig. 1 Measured OSNR monitoring parameter, F_OSNR, versus OSNR with and without 7.64-GHz-bandwidth LPFs at the modulator drivers for 10-Gbaud (a) QPSK and (b) 16QAM signals. The right-hand side of the figures show the obtained amplitude samples and their corresponding AAH when the OSNR is 20 dB.

Download Full Size | PDF

The problem associated with the transmitter’s transient characteristics on the performance of the AAH-based monitoring technique can be alleviated by using the synchronous amplitude histogram (SAH). This histogram is constructed by using the samples captured only at the symbol’s centers. Thus, the histogram exhibits clear separation between amplitude levels and facilitates the accurate extraction of the monitoring parameters. However, it would not be desirable if a hardware-based clock extraction circuitry should be used to obtain SAH, since it could not only increase the cost of OPM module but also make it opaque to the data rate and modulation format. Recently, we have proposed an OSNR monitoring technique for the non-return-to-zero (NRZ)-QPSK and NRZ-16QAM signals based on the SAH implemented by using the software-based synchronization technique [15]. This technique enables us to obtain the SAH from the asynchronously sampled data without using the clock extraction circuitry. Thus, the proposed technique is transparent to the data rates and modulation formats. In this paper, we extend this technique for the simultaneous monitoring of OSNR and CD. We experimentally demonstrate the OSNR and CD monitoring of 10-Gbaud NRZ-QPSK and NRZ-16QAM signals. The OSNR and CD monitoring ranges of this technique are measured to be 20 dB and 338 ps/nm, respectively. We also show that the proposed technique is robust against the transient characteristics of the optical transmitters.

2. Operating principle

Figure 2 shows a schematic diagram of the proposed monitoring technique. Our objective is to obtain the SAH from the asynchronously sampled data by utilizing the software-based synchronization technique, and then extract the monitoring parameters. For this purpose, we first asynchronously sample the intensity of the received signal by using a free-running sample-and-hold amplifier (SHA) and an analog-to-digital converter (ADC). The sampling rate of the SHA f_S, is not correlated with the signal’s symbol rate B. When f_S is much lower than B, the clock component of the received signal sampled at f_S is aliased at the frequency of f_a = |(B/f_S)-round(B/f_S)| × f_S, where the ‘round’ function rounds the number to its nearest integer. To extract f_a, we transform the sampled data by using fast Fourier transformation and estimate the coarse clock frequency by identifying the peak frequency f_p in the range of 0 to f_S/2. However, the eye diagram obtained by using this f_p would suffer from a severe clock drift as large as |1/2B|. Thus, for the accurate estimation of f_a, we apply the phase-reference detection method [16]. We extract the aliased clock component by applying a band-pass filter centered at f_p, and then detect the phase of the aliased clock by comparing its phase with that of a sinusoidal wave with a frequency f_p. As a result, we can rearrange the sampled data to be accurately synchronized to the symbol frame and reconstruct the eye diagram.

 

Fig. 2 Schematic diagram of the proposed monitoring technique using software-based SAH analysis.

Download Full Size | PDF

Figures 3(a) and 3(b) show examples of the asynchronously sampled data and reconstructed eye diagram of the 20-Gb/s NRZ-QPSK signal, respectively. Since we utilize a dual-parallel MZM for the generation of the signal, intensity dips are observed between the symbol’s transitions in the obtained eye diagram. Thus, for the evaluation of the signal’s quality without being affected by these dips [i.e., leading and trailing edges in the eye diagram in Fig. 3(b)], we estimate the signal’s quality at the symbol’s center. For this purpose, we divide the samples into 20 sections in time. Since the signal amplitude has the largest value at the symbol’s center, we can easily identify it by comparing the mean amplitude of each section and obtain the SAH. For example, the dotted curve in Fig. 3(c) shows the obtained SAH of the NRZ-QPSK signal. The solid curve in this figure represents the Gaussian-fitted amplitude histogram. To estimate the OSNR from this amplitude histogram, we define the OSNR monitoring parameter for the NRZ-QPSK signal as F_OSNR = μ_fit/σ, where μ_fit is the mean value of the peak in the Gaussian-fitted amplitude histogram and σ is its standard deviation [11]. For comparison, the obtained AAH is also shown in Fig. 3(d). When we use AAH, a couple of small peaks are also observed in the measured amplitude histogram, which makes the extraction of the statistical average for a specific peak difficult and inaccurate.

 

Fig. 3 (a) Asynchronously sampled data, (b) reconstructed eye diagram, (c) SAH, and (d) AAH of the 20-Gb/s NRZ-QPSK signal.

Download Full Size | PDF

We can also monitor the accumulated CD by using SAH. Figure 4(a) shows the obtained eye diagram when the NRZ-QPSK signal experiences 338 ps/nm of dispersion. As the signal experiences a larger amount of CD, the amplitude level at the symbol’s center increases while the leading and trailing edges split into multiple levels. During the symbol’s transitions, the blue- and red-shifted chirps are introduced depending on the phase difference between the adjacent symbols. As these frequency chirps interact with CD in the anomalous dispersion regime, the blue- and red-shifted components (at the symbol’s transitions) are advanced and retarded in time, respectively, and induce an increase of the amplitude at the symbol’s centers. The same amplitude increase at the symbol’s centers occurs in the normal dispersion regime although the red-shifted components travel faster than the blue-shifted ones. These amplitude variations result in the distribution changes of amplitude histogram, as shown in Fig. 4(b). The blue and red symbols show the SAHs obtained when the amounts of CD are 0 and 338 ps/nm, respectively. As shown in this figure, the amplitude level of the peak rises with the accumulated CD. Thus, it is possible to estimate the amount of accumulated CD by introducing a dispersion monitoring parameter, F_CD = μ_unfit/μ_avg, where μ_unfit is the mean amplitude of the samples obtained at the symbol’s center and μ_avg is the mean amplitude of the entire samples.

 

Fig. 4 (a) Eye diagram of the 20-Gb/s NRZ-QPSK signal with CD of 338 ps/nm. (b) SAH of the 20-Gb/s NRZ-QPSK signal with CD of 0 ps/nm (the blue squares) and 338 ps/nm (the red circles).

Download Full Size | PDF

Similarly, we can also monitor the OSNR and CD of the NRZ-16QAM signal by using SAH. Figure 5(a) shows the eye diagram of the 40-Gb/s NRZ-16QAM signal obtained from the software-based synchronization technique. The SAH obtained by using the center portion of the eye diagram is shown in Fig. 5(b). As expected, the amplitude histogram of the NRZ-16QAM signal has three peaks. We identify each peak and then calculate its mean and standard deviation after the Gaussian fitting. The detailed procedures are as follows. We first normalize the amplitude histogram by using the total number of samples used in this histogram. For equiprobable 16QAM signals, the probabilities of each intensity peak should be 0.25, 0.5, and 0.25 for the first, second, and third peaks, respectively. Thus, we can set the decision thresholds by integrating the histogram from left to right until the integrated value becomes 0.25 and 0.75. We then separate each peak one by one by using these decision thresholds and fit each peak with a Gaussian function. For the NRZ-16QAM signal, we define the OSNR monitoring parameter as F_OSNR = ([(|μ₁-μ₂|)/(σ₁ + σ₂)] + [(|μ₂-μ₃|)/(σ₂ + σ₃)])/2, where μ_k and σ_k (k = 1,2,3) are the means and standard deviations of the Gaussian-fitted first, second, and third peaks, respectively.

 

Fig. 5 (a) Eye diagram and (b) SAH of the 40-Gb/s NRZ-16QAM signal.

Download Full Size | PDF

For the CD monitoring parameter of the NRZ-16QAM signal, we first observe the effect of CD on SAH. The eye diagram and SAH of the NRZ-16QAM signal impaired by the accumulated CD are shown in Figs. 6(a) and 6(b), respectively. The red and blue symbols in Fig. 6(b) show the SAHs of the NRZ-16QAM signal measured with and without CD, respectively. As CD is accumulated, the amplitude level at the symbol’s center rises. Thus, we define the CD monitoring parameter for the NRZ-16QAM signal as F_CD = μ_unfit/μ_avg, which is identical with that of the NRZ-QPSK signal.

 

Fig. 6 (a) Eye diagram of the 40-Gb/s NRZ-16QAM signal with CD of 338 ps/nm. (b) SAH of the 40-Gb/s NRZ-16QAM signal with CD of 0 ps/nm (the blue squares) and 338 ps/nm (the red circles).

Download Full Size | PDF

It is interesting to note that the proposed monitoring technique can also be used for the polarization-multiplexed signals (including the dual-polarization (DP) QPSK and DP-16QAM signals) without modifying the OPM module. For example, we can detect the polarization-multiplexed signals by using a single photo-detector as in the proposed monitoring module without polarization demultiplexing. In the case of monitoring the DP-QPSK signal by using this technique, the obtained SAH should be similar to that of the single-polarization QPSK signal. Thus, it is possible to monitor the OSNR and CD of the DP-QPSK signal by using the same monitoring parameter used for the single-polarization QPSK signal. However, in the case of monitoring the DP-16QAM signal, the number of intensity levels is increased to 5 (instead of 3 for the single-polarization 16QAM signal as shown in Fig. 5). Thus, in this case, it would be necessary to define new monitoring parameters by using the method described in this paper.

3. Experiments and results

Figure 7 shows the experimental setup used to evaluate the performance of the proposed monitoring technique. The NRZ-QPSK signal was generated by applying two uncorrelated 10-Gbaud pseudo-random bit sequences (PRBSs) (length = 2¹⁵-1) to a dual-parallel MZM. In the case of the NRZ-16QAM signal, we generated an electrical 4-level pulse-amplitude modulation (4PAM) signal by adding two uncorrelated PRBSs (of which one PRBS is attenuated by 6 dB with respect to the other PRBS), and then by applying two uncorrelated 10-Gbaud 4PAM signals to the dual-parallel MZM. We adjusted the OSNRs of these signals by adding amplified spontaneous emission (ASE) noises from an erbium-doped fiber amplifier (EDFA). The accumulated CD was emulated by using a standard single-mode fiber. A small portion of the generated signal was tapped and sent to an optical spectrum analyzer (OSA) for the reference OSNR measurement with 0.1-nm resolution. The rest of the signal was directed to the proposed monitor, which is composed of an optical band-pass filter (bandwidth = 1 nm), a photo-detector, a sampling oscilloscope, and an off-line DSP part. The received optical power was kept to be −9 dBm throughout this measurement. In the monitor, the detected optical signal was sampled by using a sampling scope operating at 25 Msample/s. This scope was operated by the internal clock source, which was not correlated with the signal rate. For the practical implementation of the proposed monitor, the sampling oscilloscope could be replaced with an SHA and a low-speed ADC. One million sampled data were finally sent to a personal computer for the offline processing including the software-based synchronization, parameter extraction, and estimation of monitoring parameters.

 

Fig. 7 Experimental setup.

Download Full Size | PDF

Figure 8(a) shows the measured F_OSNR of the 20-Gb/s NRZ-QPSK signal as a function of the actual OSNR. During this measurement, we also varied the accumulated CD from 0 to 338 ps/nm to evaluate the effects of CD on F_OSNR. The results show that F_OSNR increases monotonically with the actual OSNR in the presence of CD. The slopes of these monotonic curves indicate the OSNR monitoring sensitivity (i.e., dF_OSNR/dOSNR). We obtained the highest OSNR monitoring sensitivity of 0.6 when the OSNR was ~16 dB and CD was 0 ps/nm. Thus, the monitoring error would be minimized at these OSNR and CD values. However, a reasonable OSNR monitoring sensitivity of 0.2 could be obtained even when the OSNR was merely 5 dB. The results also show that the OSNR monitoring sensitivity decreases as the accumulated CD increases due to the dispersion-induced pulse distortion. As a result, when the accumulated CD exceeded ~400 ps/nm, the OSNR monitoring sensitivity was deteriorated to be <0.05 at the high OSNR region and limited the performance of the proposed monitoring technique. Nevertheless, the one-to-one relationship between the F_OSNR and actual OSNR is maintained in the OSNR range of 5 ~28 dB and the CD range of 0 ~338 ps/nm. Figure 8(b) shows the measured F_CD of the NRZ-QPSK signal as a function of the actual OSNR. The measured F_CD gradually increases with the OSNR since the average signal power is also increased with the OSNR when the total received optical power (i.e., signal power plus noise power) is maintained to be constant. In the reconfigurable optical network, the optical power incident on the OPM module may not be a constant since the signals can be originated from different transmitters or traversed through different routes. Thus, to obtain the similar relationship between F_CD and OSNR, it is necessary to perform the normalization with the average received power at the OPM module. Figure 8(b) also shows that F_CD increases with the accumulated CD since the signal amplitude at the symbol’s center rises due to the chirp-dispersion interaction (as described in Section 2). The CD monitoring sensitivity (i.e., dF_CD/dCD) was measured to be 3.1 × 10⁻⁴ (ps/nm)⁻¹ when the CD was 0 ps/nm and OSNR was 28 dB. This value was increased with the accumulated CD, as shown in Fig. 8(b). For example, the CD monitoring sensitivity was measured to be 8.5 × 10⁻⁴ (ps/nm)⁻¹ when the CD was 338 ps/nm. From these results, we confirm that F_CD can be used as a monitoring parameter of the accumulated CD. Also, as shown in Figs. 8(a) and 8(b), each monitoring parameter was dependent on both OSNR and CD. Thus, for the accurate monitoring of OSNR and CD, it is necessary to measure both F_OSNR and F_CD at the same time and perform calibrations according to the functional relationships shown in Fig. 9. In this figure, the solid and dashed lines represent the OSNRs and accumulated CDs measured as a function of F_OSNR and F_CD, respectively. We estimate the OSNR and CD from the measured monitoring parameters (i.e., F_OSNR and F_CD) using this figure. For example, the OSNR and accumulated CD are estimated to be 17 dB and 166 ps/nm when F_OSNR and F_CD are measured to be 13.8 dB and 1.21, respectively.

 

Fig. 8 Measured (a) F_OSNR and (b) F_CD as a function of OSNR with various amounts of CD for the 20-Gb/s NRZ-QPSK signal.

Download Full Size | PDF

 

Fig. 9 Relationships between the monitoring parameters, F_OSNR and F_CD, and the monitoring attributes, actual OSNR and accumulated CD for the 20-Gb/s NRZ-QPSK signal. Solid lines: OSNR. Dashed lines: CD.

Download Full Size | PDF

We also evaluate the performances of the proposed SAH-based monitoring technique for the 40-Gb/s NRZ-16QAM signal. Figures 10(a) and 10(b) show the measured F_OSNR and F_CD as a function of the actual OSNR, respectively. Similar to the results of the 20-Gb/s NRZ-QPSK signal shown in Fig. 8, F_OSNR and F_CD are dependent on both OSNR and CD. Since F_OSNR of the NRZ-16QAM signal is essentially an arithmetic mean of the Q-factors of each eye opening, it should be increased with the actual OSNR. The CD dependency of F_OSNR is attributed to the dispersion-induced pulse distortion. From Fig. 10(a), we confirm that the proposed technique can monitor the OSNR of the 40-Gb/s NRZ-16QAM signal as low as 8 dB. This figure also shows that, when the accumulated CD is 0 ps/nm, the OSNR monitoring sensitivities (dF_OSNR/dOSNR) are 0.1, 0.6, 0.5, and 0.1 at the OSNRs of 8, 14, 21, and 28 dB, respectively. The CD monitoring parameter F_CD is strongly dependent on the accumulated CD, as shown in Fig. 10(b), although its dependence on the OSNR is not that strong (as in the case of the NRZ-QPSK signal). The CD monitoring sensitivities (dF_CD/dCD) are measured to be 1.8 × 10⁻⁴, 6.2 × 10⁻⁴, and 6.7 × 10⁻⁴ (ps/nm)⁻¹ for accumulated CDs of 0, 166, and 338 ps/nm, respectively (@ OSNR = 18 dB). Figure 11 shows the one-to-one relationship between F_OSNR and F_CD. As in the previous case, we can estimate the actual OSNR and CD from this figure since only one set of F_OSNR and F_CD is mapped onto one set of the actual OSNR and CD (in the ranges of 8 to 28 dB of OSNR and 0 to 338 ps/nm of the accumulated CD). Thus, we can monitor the actual OSNR and CD of the NRZ-16QAM signal by measuring two monitoring parameters (i.e., F_OSNR and F_CD).

 

Fig. 10 Measured (a) F_OSNR and (b) F_CD as a function of OSNR with various amounts of CD for the 40-Gb/s 16QAM signal.

Download Full Size | PDF

 

Fig. 11 Relationships between the monitoring parameters, F_OSNR and F_CD, and the monitoring attributes, actual OSNR and accumulated CD for the 40-Gb/s NRZ-16QAM signal. Solid lines: OSNR. Dashed lines: CD.

Download Full Size | PDF

The proposed technique is expected to be quite robust against to the transmitter’s rising/falling characteristics since it is based on SAH achieved by using the software-based synchronization. To verify this, we measured F_OSNR as a function of the actual OSNR using two transmitters having different bandwidths, as shown in Fig. 12. Two transmitters were implemented by using a dual-parallel MZM. However, we inserted electrical LPFs (3-dB bandwidth: 7.46 GHz) at the modulator drivers of one transmitter, whereas no LPF was applied to the other transmitter. The rise/fall time (20% to 80%) of the bandwidth-limited transmitter was measured to be 33 ps, whereas it was measured to be 18 ps for the transmitter implemented without using the LPFs. The results show that, in contrast to the case of using the AAH-based technique shown in Fig. 1, the proposed SAH monitoring technique is insensitive to the transmitter’s bandwidth. For example, the difference in the values of F_OSNR for the 20-Gb/s NRZ-QPSK signal is measured to be less than 0.22 dB for these transmitters having different bandwidths. For the 20-Gb/s NRZ-QPSK signal, the OSNR monitoring error caused by this F_OSNR difference is less than 0.9 dB. The F_OSNR difference is also measured to be only 0.25 dB for the 40-Gb/s NRZ-16QAM signal in the OSNR range of 5 ~21 dB. The OSNR monitoring errors induced by this F_OSNR difference are estimated to be 0.4, 0.1, and 1.7 dB for the actual OSNRs of 8, 17, and 24 dB, respectively. Thus, the proposed technique can monitor the OSNR and accumulated CD accurately without using the transmitter-specific calibration process in the reconfigurable optical networks.

 

Fig. 12 The effect of transmitter bandwidth variations on the performance of the proposed monitoring technique for the (a) 20-Gb/s NRZ-QPSK and (b) 40-Gb/s NRZ-16QAM signals.

Download Full Size | PDF

4. Summary

We have proposed and demonstrated an optical performance monitoring technique based on the SAH obtained from the asynchronously sampled data of signal intensity. In this technique, we obtained the SAH by using the software-based synchronization technique instead of the expensive high-speed sampling devices and clock recovery circuitry. This technique cannot only monitor the OSNR and accumulated CD simultaneously, but also is transparent to the symbol rates and modulation formats. We evaluated the performances of the proposed technique experimentally by using 10-Gbaud NRZ-QPSK and NRZ-16QAM signals. The results show that we can monitor the OSNRs of the QPSK signal in the range of 5 ~28 dB and of the 16QAM signal in the range of 8 ~28 dB. This technique can also monitor the accumulated CD in the range of 0 ~338 ps/nm for the QPSK and 16QAM signals. Unlike the optical performance monitoring technique based on AAH, this technique is quite robust against the transient characteristics of the transmitters. Thus, we believe that the proposed technique is well suited for the use in the reconfigurable optical networks where numerous transmitters generate various types of signals modulated at different speeds and/or formats.