Open Access
Add to BibSonomyAbstract
We propose an optical performance monitoring scheme that combines a tunable narrowband optical filter with asynchronous histograms to create density plots that represent probability maps of both optical power and frequency. Simulations are used to investigate the technique, and show that measurements at transmitters and filters can directly detect optical chirp. The effects of several optical impairments are also shown, comprising misaligned filters in the optical path, dispersion, nonlinear effects and optical signal to noise ratio. The density plots are affected differently by each of these impairments.
©2005 Optical Society of America
1. Introduction
Optical performance monitoring is becoming increasingly important for both current and future optical communication systems [1]. The main drivers for signal monitoring are to identify changes in transmitted signals, diagnose the cause and location of the underlying faults, optimize the performance of tunable elements, and estimate the bit-error rate (BER).
Histogram techniques are one type of performance monitor that has attracted significant attention in recent times [2]. Their general principle of operation is to tap off some WDM power, filter out a single channel, and repeatedly sample the instantaneous received optical power, which is then plotted as a probability distribution. Sampling at the bit rate gives a synchronous histogram, which represents the power distribution at a particular point within the bit period. Sampling at a different rate gives an asynchronous histogram, which shows the power distribution across the entire bit period.
The shapes of both synchronous and asynchronous histograms tend to change as the signal becomes degraded, and significant effort has been put into correlating these changes with various degradation mechanisms. Some of these phenomena have been shown to have clearly identifiable effects on histogram shape [3–6]. However differentiating between some others, such as RIN and ASE, is challenging because they cause similar changes in histogram shape [7]. To compound this, the amount of optical power available is small, and so receiver noise may mask subtle changes in the histogram shapes [8].
One limitation of histogram techniques is that they only extract the power from the optical waveform: information on frequency, phase and polarization is lost at the receiver. However, some causes of signal degradation do not affect the power distribution at the point of origin, but interact with other mechanisms along the fiber to cause severe changes at the receiver.
In this paper, we propose creating a broadband histogram monitor by combining a narrowband tunable optical filter with an asynchronous histogram monitor and filtering part of the signal sidebands. The aim of the work is to investigate the potential of histogram techniques by extracting more information about the optical signal, while acknowledging that cost and slower response times may reduce its practicality. We present preliminary results for various sources of chirp in transmission systems, which show that the monitor is capable of detecting phase effects such as chirp at their point of origin. We also show that a side benefit of the technique is the potential to measure the out-of-band optical signal to noise ratio.
2. Broadband monitor concept
The optical spectrum of a modulated signal contains many frequency components with different relative phases, which interfere at the receiver. Applying sideband filtering to the signal preferentially removes some of these frequency components, preventing some of this interference occurring and causing amplitude distortion.
The filter effect can also be interpreted as adding chirp to the optical waveform [9], increasing the optical power in the sidebands relative to the carrier frequency. The resulting chirp interacts with existing phase distortions in the optical signal, converting phase noise to amplitude distortion. Sideband filters have previously been used to monitor dispersion by measuring the relative clock phase-shift of the two sidebands [10], but this is the first application to histogram monitoring that we have seen.
2.1 Monitor Structure
Figure 1 shows the monitor structure, which is based on a standard histogram configuration with the WDM channel band-pass filter replaced by a tunable narrowband Fabry-Perot filter. The passband of the filter is the same as the data transmission rate, as we empirically found this to give the best feature extraction. We found that broader filter bandwidths tend to reduce the contrast of features of interest in the channel sidebands, while narrower bandwidths excessively reduce the sideband powers. In general, we expect the optimal filter bandwidth to be determined by the modulation format and bit rate. The filter finesse must be sufficiently large to give a free spectral range that exceeds the WDM bandwidth of the multi-channel comb. The receiver used in the monitor is a PIN diode followed by a transimpedance amplifier (TIA) matched to the simulated data bit rate of 10 Gbit/s [11].
The monitor scans the optical filter across the entire WDM spectrum in discrete frequency steps, generating an asynchronous histogram (probability distribution function or PDF) of the TIA voltage at each frequency, using a sampling window of 3.125 ps. The probability of each histogram bin is then multiplied by the bin TIA voltage. This step makes the histogram total proportional to the average TIA voltage at that frequency, which in turn is proportional to the received optical power, allowing simple out-of-band estimation of the optical signal to noise ratio (OSNR). It also de-emphasizes the low powers between channels, which would otherwise dominate the response. Finally, the histograms are plotted against optical frequency as a density plot, creating a three dimensional frequency, amplitude and probability map of relative optical power levels.
Monitor performance is investigated here using numerical simulations with commercial software [12], to scope out the potential of the technique. The simulations include phase and amplitude responses. Figure 2 shows a schematic of the system modeled, consisting of four data channels spaced at 100 GHz, each modulated with NRZ data at 10 Gbit/s. The channels are generated by Mach-Zehnder modulators with chirps corresponding to α values of -1, 0, 1 and 2. Initially, each channel has a modulated power of 0 dBm. The channels are multiplexed using an arrayed waveguide grating (AWG) with a 3dB channel bandwidth of 40 GHz, and amplitude and phase responses calculated following the method in [13]. An EDFA is used to overcome the AWG insertion loss. Points where the monitor is inserted are labeled (a)–(c).
To match this system, the Fabry-Perot filter in the monitor has a 3 dB bandwidth of 10 GHz and Finesse of 700, corresponding to a free spectral range of 7 THz. The filter model also includes both amplitude and phase responses. The filter is stepped across the WDM channels in 5 GHz steps. Finally, we note that filter stability effects were not included in the simulations, but will be an issue in a practical system, and will be included in later work.
2.2 Output Density Plots
Figure 3 shows the resulting broadband power density plot for the system shown in Fig. 2, measured at point (a). Figure 3 also includes a density scale, which is used for all the figures in this paper. As discussed above, the shading (z scale) corresponds to the TIA voltage multiplied by probability on a logarithmic scale, and so is proportional to the optical power in dB. Figure 3 includes labels for the chirp of each channel and the main features of interest. As the frequency is stepped across the WDM signal, the power in each channel causes a characteristic increase in TIA voltage (y scale). The voltage peaks corresponds to the channel centre frequencies, with upper and lower sidebands falling away on either side. The background between the channels is a combination of ASE noise and channel leakage from the filter response. Within each channel, the data 1’s (marks) correspond to the high densities (dark areas) at high voltage, while the data 0’s (spaces) correspond to the smaller density peaks at low voltage.
Fig. 3. Broadband density plot at Fig. 2, point (a), showing channel chirp and features of interest. Shading density corresponds to relative optical power in dB.
3. Monitor performance
To interpret the density plots, it is useful to recollect that the effect of optical chirp is to increase the optical power in the sidebands relative to the center frequency. Negative alpha values move lower frequencies the rising edge of pulses and higher frequencies to falling edge, while positive alphas have the reverse effect. These effects are sufficient to account for the features of interest in the plots.
3.1 Transmitter chirp
Figure 3 shows symmetric upper and lower sidebands for the channel with α=0, as the sideband powers are spread evenly through the pulses. Positive or negative α values cause an overshoot peak to appear systematically in the upper or lower sidebands respectively, due to the increased power in the sideband at the rising edge of the eye. As the amount of chirp increases, the voltage of the overshoot peak also increases, consistent with the increased relative optical power in the sidebands. The responses for α values of ±1 are reflections of each other, which is also expected from the chirp effects given above. Note that the voltage positions of the mark peaks in the sidebands are not affected by chirp, offering a potential method of frequency calibration.
3.2 Filter misalignment chirp
Chirp can also be caused by misaligned filters in the transmission link. To investigate this effect, the previous simulation was rerun, with the AWG increasingly detuned relative to the channel frequencies in 5 GHz steps from 0 to +25 GHz. Figure 4 shows an animation of the changes to the density plots, which model either negative frequency laser drift or positive frequency filter drift. The strongest effect of the frequency drift is attenuation; however we assume here that the post-multiplexer amplifier has automatic gain control and so compensates for the additional loss, leaving only the optical field distortion.
Fig. 4. (658 KB) Animation of broadband density plots at Fig. 2, point (a) with AWG detuning from 0 to +25 GHz in 5 GHz steps, with compensation for additional loss
The density plot shows that the chirp from the filter misalignment has very similar effects to transmitter chirp, and indeed seems to have additive effects. For the channel with α= -1, the filter misalignment has partially cancelled the transmitter chirp, increasing the symmetry of the upper and lower sidebands. However both upper and lower sidebands now show small overshoot peaks. For the other channels, the overshoot of the upper sideband has been emphasized.
3.3 Dispersion
The effects of dispersion were investigated by recording density plots at 10km (170 ps/nm) steps along the single mode transmission fiber, corresponding to point (b) in Fig. 2. For ease of comparison, the fiber attenuation has been numerically compensated in order to maintain constant OSNR. Figure 5 shows an animation of the resulting changes to the density plots with increasing dispersion. The effects of dispersion on asynchronous histograms have been previously reported: reducing the intensity and broadening the range of the peaks corresponding to both the marks and spaces; and causing a subsidiary peak to appear above the spaces [3, 4]. We find that our technique appears to enhance these effects at the central channel frequency, and also produces a second subsidiary density peak around the marks voltage. The interaction between chirp and dispersion is clearly evident when comparing the four data channels: the peaks move inward at an increasing rate when α is zero or positive, and outward when α is negative. The overshoot peaks in the upper and lower sidebands also generally increase with dispersion. Again there is clear evidence of a strong chirp interaction, which appears complex and may be difficult to analyze quantitatively.
Fig. 5. (943 KB) Animation of broadband density plots at Fig. 2, point (b) for transmission distances from 0 to 100 km SMF in 10 km steps (dispersion from 0 to 1700 ps/nm in 170 ps/nm steps), with numerical compensation for attenuation.
3.4 Nonlinear Effects
To investigate the effects of fiber nonlinearity, the simulation was rerun with the monitor at point (c) in Fig. 2, and the modulated laser powers increased from 0 to 10.5 dB in 1.5 dB steps. In order to maintain constant OSNR and receiver noise for comparison purposes, the simulations only increased the optical power through the SMF and DisCo, but not the EDFAs or at the receiver. We expect SPM to the dominant nonlinear effect under the simulation conditions, causing additional chirp along the fiber. However, the simulation also includes cross-phase modulation and four wave mixing effects.
Fig. 6. (693 KB) Animation of broadband density plots at Fig. 2, point (c), with launch powers from 0 to 10.5 dBm/channel, in 1.5 dB steps and with constant OSNR and receiver noise.
An animation of the resulting density plots is shown in Fig. 6, and indicates that the main effect of SPM is to spread the density peaks over a broader voltage range and reduce the peak intensities, in agreement with [7]. Some asymmetry also appears in the sidebands, but these differences are subtle.
3.5 OSNR
Although not a chirp-related effect, it is useful to compare the effects of OSNR on the broadband density plots. The initial OSNR of 50 dB was varied by increasing the noise loading from the post-mux EDFA to give a range of 30 to 18 dB in 1.5 dB steps. Figure 7 shows an animation of the resulting density plots, recorded at point (a) in Fig. 2. At high OSNR, the main effect is broadening of the mark peak due to-signal spontaneous beat noise. As the OSNR drops, the ASE between the channels becomes directly measurable, giving two possible measurement methods. We also note that the changes in shape are noticeably different from those with nonlinear effects in Fig. 6.
Fig. 7. (920 KB) Animation of broadband density plots at Fig. 2, point (a), with variable ASE loading to give OSNR of 50 dB, then from 30 to 18 dB in 1.5 dB steps.
Since the histogram data is presented in terms of the optical power probability, the OSNR can also be estimated from the histogram totals across the frequency range. However, the resolution of this third method is limited by both the response of the Fabry-Perot filter and receiver noise. For the 10 GHz bandwidth used here, the rejection at 50 GHz is 20 dB, giving a maximum detectable OSNR of 17 dB. This is unlikely to be adequate for normal telecommunication applications. However, this method can detect higher OSNRs by using narrower filters, leaving the optimum filter width as an issue still to be explored.
3.6 Discussion
Finally, we note that the density plots presented above appear to show different changes under the effects of various optical impairments. Taken alone, the ability to monitor chirp at a transmitter may be useful. However, the true test of the technique is blind identification and separability of impairments, which will be the subject of future work.
4. Conclusions
We have presented a novel technique for monitoring the performance of optical networks. We have shown that the use of a narrowband tunable optical filter with asynchronous histogram techniques produces density plots with clearly visible difference under the effects of transmitter and filter chirp, dispersion, SPM and OSNR.
Acknowledgments
National ICT Australia is funded by the Australian Government’s Department of Communications, Information Technology, and the Arts and the Australian Research Council through Backing Australia’s Ability and the ICT Research Centre of Excellence programs.
References and links
1 . C.P. Larsen and P.O. Andersson , “ Signal Quality Monitoring in Optical Networks ,” SPIE Optical Networks Magazine , Oct 2000, 17 ( 2000 ).
2 . J.D. Downie and D.J. Tebben , “ Performance monitoring of optical networks with synchronous and asynchronous sampling ,” OFC’01, paper WDD50 ( 2001 ).
3 . N. Hanik , et al ., “ Application of amplitude histograms to monitor performance of optical channels ,” Electron. Lett. 35 , 403 – 404 ( 1999 ). [CrossRef]
4 . I. Shake and H. Takara , “ Chromatic dispersion dependence of asynchronous amplitude histogram evaluation of NRZ signal ,” J. Lightwave Technol. 21 , 2154 – 2161 ( 2003 ). [CrossRef]
5 . C.M. Weinert , et al ., “ Histogram method for identification and evaluation of crosstalk ,” Electron. Lett. 36 , 558 – 559 ( 2000 ). [CrossRef]
6 . K. Hinton , “ Histogram analysis for optical monitoring ,” COIN/ACOFT’03 ( 2003 ).
7 . K. Clarke , et al ., “ Considerations for the histogram approach to optical performance monitoring ,” OECC/IOOC’01 ( 2001 ).
8 . W. Yang , “ Sensitivity issues of optical performance monitoring ,” IEEE Photonics Technol. Lett. 14 , 107 – 109 ( 2002 ). [CrossRef]
9 . D.F. Hewitt , “ Design and performance of optical vestigial sideband (VSB) filters for 40 Gbit/s modulated systems ,” COIN/ACOFT’03 ( 2003 ).
10 . Q. Yu , et al ., “ Chromatic dispersion monitoring technique using sideband optical filtering and clock phase-shift detection ,” J. Lightwave Technol. 20 , 2267 – 2271 ( 2002 ). [CrossRef]
11 . OMMIC, “ CGY2112UH/C1 TIA ”, http://www.ommic.com/data/circuits/pdf/fl_cgy2112uhc1.pdf .
12 . VPIsystems, “ VPItransmissionMaker TM ”, http://www.vpisystems.com .
13 . M. Smit and C.v. Dam , “ PHASAR-based WDM-devices: principles, design and applications ,” J. Sel. Top. Quantum Electron. 2 , 236 – 250 ( 1996 ). [CrossRef]
