Non-intrusive OSNR measurement of polarization-multiplexed signals with spectral shaping and subject to fiber non-linearity with minimum channel spacing of 37.5GHz

Daniel Gariépy; Steven Searcy; Gang He; Sorin Tibuleac

doi:10.1364/OE.24.020156

1. Introduction

Performance monitoring in optical networks is important for assessing signal quality and identifying sources of signal degradation. The contribution of amplified spontaneous emission (ASE) noise from optical amplifiers is a critical system impairment. Historically, optical signal-to-noise ratio (OSNR) has been a key metric to quantify the ASE noise level relative to the signal. In most legacy systems, OSNR was straightforward to measure since WDM signals were not very closely spaced, allowing ASE noise to be measured optically by interpolating the noise level from the spectral gaps between signals [1,2]. This interpolation method was rendered ineffective by the rise of closer channel spacing in dense wavelength division multiplexing (DWDM) systems and in-line filtering from add/drop filters or reconfigurable optical add/drop multiplexers (ROADMs). However, legacy direct-detection signals generally occupied a single polarization state, so useful OSNR measurement techniques were developed which rely on distinguishing the polarized signal from unpolarized in-band ASE noise [3–5]. However, the introduction of coherent detection has also enabled the increased use of polarization-multiplexed (PM) signals, which cannot be easily distinguished from unpolarized noise, rendering polarization-nulling techniques ineffective. As a result, current and future optical systems require new in-band OSNR measurement techniques which can easily and effectively be used on densely-spaced polarization-multiplexed signals.

The continued increase in adoption of coherent optical transceivers has eased the measurement of several optical performance metrics through the capabilities of the advanced digital signal processing (DSP) embedded in the coherent receiver. Coherent DSP can readily measure signal parameters such as chromatic dispersion (CD) and polarization mode dispersion (PMD). However, other impairments such as in-band crosstalk, passband narrowing due to optical filtering, and fiber nonlinear effects are more challenging to directly measure. Additionally, coherent receivers are generally incapable of distinguishing between different additive noise sources with approximately Gaussian statistical properties. A key example is long-haul transmission systems without in-line CD compensation, where the distortions generated by fiber nonlinearity can be modeled as additive Gaussian noise [6,7]. While this observation is useful for developing analytical models to predict nonlinear system performance, it also indicates the difficulty of distinguishing between ASE noise and nonlinear “noise” at the receiver as both share similar statistical properties. Since ASE and nonlinear noise can each degrade the received SNR in the same way, it is often useful to quantify these noise sources separately, in order to identify the relative impact of the different impairments. Thus the “legacy” OSNR which measures the ratio of signal to ASE noise is still a relevant and useful performance metric for coherent signals.

In recent years, many techniques have been proposed to measure in-band OSNR for polarization-multiplexed coherent signals. One simple method, commonly known as the “Signal On/Off” method, requires temporarily turning off the channel of interest and measuring the in-band ASE noise in the absence of the signal. This technique can be highly accurate as long as all the elements in the signal path remain relatively stable during the process of turning the signal off and back on. Optical amplifiers will generally maintain a relatively steady state as long as enough channels are present along with the test signal as it is switched on and off. Thus the Signal On/Off method is generally a good reference measurement as long as there are more than a few WDM channels. However, this method’s primary drawback is that it is service-affecting for the test channel and thus cannot be used on signals carrying live traffic. A number of other non-service-affecting techniques to measure OSNR have also been proposed [8]. These various methods are based on several different principles of measurement, which include interferometric devices [9], characterization of beat noise [10], and Stokes-space based polarization measurements (“polarization disc”) [11]. Each of these methods has some merits but also some drawbacks, including the need for new specialized hardware or lack of applicability under certain system conditions, such as chromatic dispersion or fiber nonlinearity. DSP-based algorithms can be integrated with the existing coherent receiver DSP, and recent efforts have yielded methods for OSNR estimation which are tolerant to nonlinearity [12,13]. However, such methods still have drawbacks, which can include the need for additional overhead and capability to measure OSNR only at the location of the coherent receiver.

We have proposed a non-intrusive in-band OSNR measurement technique based on spectral analysis which utilizes widely-deployed conventional optical spectrum analyzer (OSA) [14]. This reference-based method relies on a detailed spectral comparison of a reference (noise-free) signal spectrum, taken at the transmitter, with the signal spectrum acquired at the desired OSNR measurement location (which can be anywhere along the signal path, not just at the receiver). Since the method does not rely on absolute power measurements, the necessary OSA traces can be acquired non-intrusively via taps in the system. The method also takes into account the spectral deformation induced by fiber nonlinear effects [15,16], and thus can be used on signals after nonlinear fiber transmission, as typically found in deployed systems.

This paper will describe the method with a detailed explanation of the algorithm, then present a series of experimental validation tests which confirm the method’s applicability and accuracy across a wide range of practical system conditions. Results are presented demonstrating the method’s accuracy for coherent signals with varying modulation formats, spectral shaping, channel spacing, and fiber transmission conditions, as well as statistical analysis of the OSNR measurement results.

2. Description of reference-based spectral OSNR-measurement method

The process begins with the acquisition of two spectra: P_Ref(λ) is the reference spectrum measured at a location where the ASE noise level is known (or negligible) so that a noise-free spectrum is obtained and $P_{M e a s} (λ) = P_{S i g} (λ) + N_{A S E} (λ)$ is the spectrum measured at the location where the OSNR measurement is to be performed. Upon propagation, the spectrum of the transmitted signal P_Ref(λ) is shaped by the link contributions such that received signal $P_{S i g} (λ) = κ (λ) \times P_{R e f} (λ)$ where κ(λ) is the link’s spectral transfer function comprising the linear part of net gain/loss profile within the signal’s spectral width κ_L(λ) and, in a NL transmission regime, κ_NL(λ) which represents the spectral deformation incurred by the signal. The spectral transfer function can be written as:

κ (λ) = κ_{L} (λ) \times κ_{N L} (λ) = κ_{0} \times [1 + Δ κ_{L} (λ)] \times [1 + Δ κ_{N L} (λ)]

where κ₀ is the constant transmission ratio between P_Sig(λ) and P_Ref(λ) at the signal’s peak wavelength λ_p, Δκ_L(λ) is the relative wavelength dependent linear transfer function resulting from the gain/loss profile of the fiber, amplifiers and other network elements while Δκ_NL(λ) is the relative wavelength dependent spectral deformation resulting from NL effects on the signal.

The method is based on calculating or estimating the optical spectrum difference (OSD) between P_Meas(λ) and P_Ref(λ) to obtain In-Band OSNR. OSD can be expressed as:

O S D (λ) = \frac{P_{M e a s} (λ) - κ_{0} P_{r e f} (λ)}{κ_{0} \times P_{R e f} (λ_{p})} = O S D_{S} (λ) + O S D_{A S E} (λ)

where

O S D_{S} (λ) = \frac{P_{S i g} (λ)}{P_{S i g} (λ_{p})} - \frac{P_{R e f} (λ)}{P_{R e f} (λ_{p})} = Δ κ \times \frac{P_{R e f} (λ)}{P_{R e f} (λ_{p})}

O S D_{A S E} (λ) = \frac{N_{A S E} (λ)}{P_{S i g} (λ_{p})} = \frac{1}{O S N R_{A S E} (λ)}

and

Δ κ = \frac{κ (λ)}{κ_{0}} - 1

is the relative “signal only” deformation factor. In Eq. (2), OSD_S is the optical spectrum difference resulting from deformations induced by linear transmission and NL effects. OSD_ASE is the optical spectrum difference induced by added ASE on the signal during transmission.

“ASE only” OSNR can be calculated as:

O S N R_{A S E} (λ) = \frac{1}{O S D_{A S E} (λ)} = \frac{1}{O S D (λ) - O S D_{S} (λ)}

In this paper, focusing the spectral analysis of the OSD in the central passband of optical filters where Δκ_L(λ) can be neglected ( $Δ κ_{L} (λ) \approx 0$ in the unfiltered transmission case), the transfer function of the signal is dominated by NL-induced deformation and becomes approximately $κ (λ) \approx κ_{0} [1 + Δ κ_{N L} (λ)]$ and $Δ κ \approx Δ κ_{N L} (λ)$ .

For systems operating in the linear regime $Δ κ_{N L} (λ) = 0$ (leading to $O S D_{S} = 0$ and $κ (λ) = κ_{0}$ ) such that the transfer function becomes wavelength independent. This constant transmission ratio can be estimated as:

κ_{0 - E s t} = \frac{P_{M e a s} (λ_{1}) - P_{M e a s} (λ_{2})}{P_{R e f} (λ_{1}) - P_{R e f} (λ_{2})} = κ_{0}

with λ₁, λ₂ chosen such that

[N_{A S E} (λ_{1}) - N_{A S E} (λ_{2})] < < [P_{M e a s} (λ_{1}) - P_{M e a s} (λ_{2})]

. From Eqs. (2) and (4)-(5), the OSNR can thus be obtained from estimating OSD with

O S D_{S} = 0

, as:

O S N R_{E s t} (λ) = \frac{1}{O S D (λ)} = \frac{κ_{0 - E s t} \times P_{R e f} (λ_{p})}{P_{M e a s} (λ) - κ_{0 - E s t} \times P_{R e f} (λ)} = O S N R_{A S E} (λ)

When systems operate in the NL regime however, NL-induced spectral deformation of the signal generates a significant Δκ_NL(λ) contribution such that $κ_{0 - E s t} \neq κ_{0}$ and $O S D_{S} \neq 0$ . This becomes the dominant cause of spectral difference (OSD_S) when Δκ_L(λ) can be neglected, as is the case for the experimental results presented in this work where no significant in-line filter shaping is incurred on the signal spectrum. In these conditions, $κ (λ) = κ_{0} \times [1 + Δ κ_{N L} (λ)]$ and applying Eqs. (5) and (6) no longer yields OSNR_ASE(λ) but rather leads to an estimated level OSNR_Est(λ) which relates to OSNR_ASE(λ) as follows:

\frac{1}{O S N R_{E s t} (λ)} = \frac{1}{O S N R_{A S E} (λ)} + \frac{1}{Δ O S N R (λ)}

where ΔOSNR(λ) is determined by OSD_S(λ) and is proportional to Δκ_NL(λ) as shown in Eq. (3). For weak NL effects,

Δ κ_{N L} (λ) \approx 0

and consequently

O S D_{S} (λ) \approx 0

and

Δ O S N R (λ) \approx 0

, while for strong NL effects, a large Δκ_NL(λ) leads to a larger OSD_S(λ) and ΔOSNR(λ).

In order to determine ΔOSNR(λ), without a priori knowledge of the ASE level, Δκ_NL(λ) can be estimated indirectly from the ratio of the derivatives of P_Meas(λ) and P_Ref(λ):

\begin{matrix} R d e v (λ) = \frac{d P_{M e a s} (λ)}{d P_{R e f} (λ)} = κ (λ) + P_{R e f} (λ) \times \frac{d κ (λ)}{d P_{R e f} (λ)} + \frac{d N_{A S E} (λ)}{d P_{R e f} (λ)} \\ \approx κ (λ) + P_{R e f} (λ) \times \frac{d κ (λ)}{d P_{R e f} (λ)} \end{matrix}

where

d P_{M e a s} = \frac{d P_{M e a s} (λ)}{d λ}

,

d κ = \frac{d κ (λ)}{d λ}

and λ is chosen in the region from λ₁ to λ₂ such that

d N_{A S E} = \frac{d N_{A S E} (λ)}{d λ}

<<

d P_{R e f} = \frac{d P_{R e f} (λ)}{d λ}

and

\frac{d P_{R e f} (λ)}{d λ} \neq 0

. An empirical relation dependent on the signal properties (baud rate, spectral shaping) observed on dP_Ref, which applies to all the presented systems under test, can be determined to infer

Δ κ_{N L} (λ) = \frac{κ (λ)}{κ_{0}} - 1

from the measured

Δ R d e v (λ) = \frac{R d e v (λ)}{R d e v (λ_{1})} - 1

and thus ultimately ΔOSNR(λ). As will be shown in the experimental results section, the correspondence between Δκ_NL(λ) and ΔRdev(λ) holds well for moderate NL conditions but the approximation made in Eq. (8) leads to increasing deviations when larger NL-induced spectral deformations are present. The steps of the measurement process, as applied over a wide range of operating conditions presented in this paper, are summarized in Fig. 1.

Fig. 1 Simplified flowchart of algorithm.

Download Full Size | PDF

3. Experimental configuration

Several different sets of OSNR measurements were taken, in back-to-back configuration with added ASE and after transmission over a multi-span fiber link. The non-intrusive OSNR was measured via OSA traces taken at the beginning (Tx Reference trace) and end (Rx trace) of the link with a commercial EXFO OSA. A second OSA was also used to verify the measurement repeatability across multiple devices. To validate the non-intrusive OSNR measurement method, a reference OSNR measurement was also acquired using the signal On/Off method or by interpolating the ASE noise level from empty 50 GHz channel slots near the signals under test (within 250 GHz of the test channel).

Either two or three test channels were coherently detected at the end of the link for BER monitoring. These test channels were commercial coherent transceivers operating with either PM quadrature phase-shift keying (PM-QPSK) or 16-ary quadrature amplitude modulation (PM-16QAM), as described in more detail below. Additional neighboring channels for generating nonlinear crosstalk were either legacy intensity-modulated/direct-detection or coherent PM-QPSK signals. All neighbor channels were spaced on the standard 50 GHz grid and multiplexed using a 50 GHz arrayed waveguide grating (AWG). A general schematic of the experimental link with standard single-mode fiber (SSMF) and all-EDFA amplification is shown in Fig. 2. For longer test links, wavelength-selective switches (WSS) were used for periodic power equalization, but in all cases no in-line per-channel filtering was applied. Each of the different test link variants are described in detail below, and summarized in Table 1.

Fig. 2 Schematic of experimental fiber transmission link.

Download Full Size | PDF

Table 1. Summary of Various Experimental Test Conditions for OSNR Measurement

View Table

In one set of tests, three central test channels at 193.40-193.50 THz were 120 Gb/s PM-QPSK with no digital pulse shaping, i.e., non-return-to-zero (NRZ) PM-QPSK. Neighboring channels were either 120 Gb/s NRZ-PM-QPSK or 10 Gb/s NRZ On-Off Keying (OOK). The transmission link consisted of 16 SSMF spans, with WSS before the first and ninth spans for channel power equalization. Measurements were taken first with all-coherent 100G QPSK channels and no in-line CD compensation (CDC), then in an “upgrade” scenario with mixed 100G QPSK and 10G OOK channels and DCMs placed at in-line amplifiers to create a common legacy CD map. The 10G neighbors were either placed immediately surrounding the 100G test channels, or grouped separately with a 300 GHz spectral guard band (GB) between 100G and 10G channels, to represent two different channel plans used for such upgrade scenarios.

In another set of tests, two test channels at 193.25-193.30 THz were modulated with either PM-QPSK or PM-16QAM at ~34 Gbaud. The test channels had root-raised cosine (RRC) spectral shaping applied at the transmitter via digital-to-analog converter (DAC) driving the modulator, with varying roll-off factor (1.0, 0.3, and 0.1) covering a full range of practical shaping for deployed signals. Neighboring channels were unshaped 120 Gb/s PM-QPSK, as in the previous test. The link with 8 SSMF spans had no in-line CDC and no WSS for equalization.

Another set of tests used three 120 Gb/s PM-QPSK channels with RRC shaping (0.3 roll-off factor) spaced at 37.5 GHz. The central test channel at 193.30 THz was used for OSNR measurement to assess performance in a flexible-grid scenario with tighter channel spacing. The three 37.5-GHz-spaced test channels were passively coupled and co-propagated along with unshaped 120 Gb/s PM-QPSK neighbor channels on the standard 50 GHz grid. This set of tests was performed over 8 SSMF spans with no in-line CDC and no WSS for equalization.

In each set of tests, the following general procedure was followed. First, back-to-back measurements were taken to validate the OSNR measurement method with only added ASE noise (no fiber transmission). Next, OSNR measurements were taken after transmission through the fiber link, while varying the launch power into each span. By operating around the optimum launch power where minimum BER is achieved, the test channel OSNR can be measured in multiple nonlinear transmission conditions: weakly nonlinear/nearly-linear regime (down to 4 dB below optimum launch power), moderately nonlinear regime (at optimum launch power), and strongly nonlinear regime (up to 3 dB above optimum launch power). In a subset of representative cases extra ASE noise was added before the receiver, to vary the OSNR level for the same nonlinear transmission condition.

4. Experimental results and discussion

In one series of tests, ten 100G NRZ-PM-QPSK channels were characterized in the back-to-back condition with ASE loading to cover OSNR levels ranging between 10 and 30dB. Throughout this paper, OSNR_NI refers to the non-intrusive OSNR calculation described in section 2 and all OSNR measurements are computed in a 0.1nm reference bandwidth.

Figure 3(a) shows the combined results for all the unshaped QPSK channels in the back-to-back condition. Five series of measurements with the upper left inset and another with the bottom right inset channel plans were done on 10 channels with 2 OSAs for a total of over 800 measurement points in all covering a range from 10 to 30dB OSNR. The overall average deviation is −0.11dB and only 11 out of these 800 measurements points exceed ± 0.5dB (dotted lines) from the target OSNR_On-Off value (dashed line), all of which occur at OSNR levels above 26dB, with a maximum measured deviation below 0.8dB. Larger deviations are expected for higher OSNR levels since the information used to perform the analysis is inversely proportional to OSNR. For example, measuring 20dB OSNR involves discriminating 1% of the measured power and this percentage is reduced to 0.1% for a 30dB OSNR. Any technique attempting to discriminate ASE power from signal power has a fundamental uncertainty that will increase with the OSNR level, as the ASE power decreases. It is also worth noting here that a detailed comparison performed on a subset of these results for the two OSAs showed very good agreement with an average difference of −0.02dB and standard deviation of 0.03dB for 120 back-to-back measurements made simultaneously on both OSAs.

Fig. 3 a) Back-to-back OSNR_NI results for 10 unshaped QPSK channels for 0.1nm OSNR levels of 10 to 30dB. b) Same channels after transmission while varying launch power and thus non-linear conditions.

Download Full Size | PDF

With the back-to-back results confirming the method’s performance in the linear regime, the same sets of 10 channels were measured after transmission over 8 SSMF spans with no in-line CD compensation and no WSS for equalization. In Fig. 3(b), the measurement deviation (OSNR_Ni - OSNR_On-Off) is displayed for varying launch powers covering a wide range of non-linear conditions. The optimum per-channel launch power (0 dBm), corresponding to OSNR levels between 23 and 24dB depending on the configuration, was determined by measuring the BER on the central test channels for each configuration and the results are shown for varying powers relative to optimal launch conditions. All the data points are presented (green X’s) and since there were at least 20 data points for each OSA for a minimum of 40 data points per launch condition, statistical results for all channels for the average and ± 2σ upper and lower values are also shown. In the moderate non-linear regime up to the optimum launch power, only 2 deviation results exceed ± 0.5dB. At higher launch powers with stronger non-linearity, the deviations increase but typically remain < ± 1dB with 4 deviation results exceeding ± 1dB (out of one hundred data points), obtained at a per-channel launch power 3dB above the optimum launch power. It is important to highlight that this corresponds to a OSNR level >27dB where 0.2% ASE to signal power ratio must be discriminated accurately, and strong NL effects impact the correspondence between Δκ_NL(λ) and ΔRdev(λ). Both the low ASE power and strong NL effects contribute to the larger deviations observed.

The same processing was also applied to different channel plans and dispersion conditions. In the same format as Fig. 3(b) previously, Fig. 4 presents the deviation results at varied per-channel launch powers, normalized to the optimum launch power, after propagation through 16 spans of SSMF for three system configurations. The respective spectral plans are shown in inset. At the optimum launch power (0 dBm for all 100G and −1 dBm for mixed 100G and 10G), OSNR levels ranged from 18.5 to 20.5dB depending on the configuration. The filled color circles represent the average results (three 100G PM-QPSK test channels measured with two OSAs), while the colored X’s show all 6 deviation results for each configuration’s launch power. The first series (green) presents the deviation results for the all 100G channel plan with no in-line CDC. These are consistent with the results from Fig. 3(b) that were obtained on a shorter link for a similar configuration. The second series (red) presents the deviation results for a mixed 10G-100G channel plan without guard band and with CDC. This condition presents the highest deviations with a worst case slightly above 1dB at the highest launch power. The third series (blue) shows that adding a guardband between the 10G and 100G channels for a mixed chanel plan with dispersion compensation resulted in smaller measurement deviations similar to the all 100G chanel plan which did not have dispersion compensation. It is worth noting that the behavior of the spectral deformation was different for the mixed signals channel plans compared to the all 100G but the technique worked well with this range of different spectral deformations.

Fig. 4 OSNR_NI deviation results for 3 channels for different system configurations with and without CDC. The test channels are indicated by matching color arrows in the inset spectra.

Download Full Size | PDF

Spectrally shaped channels were then analyzed to evaluate the measurement technique’s performance. The same processing, which adaptively selects the working spectral region for performing the spectral comparison and determines a corresponding non-linear correction factor, was applied to two 200G PM-16QAM test channels with RRC shaping at 1, 0.3 and 0.1 roll-off conditions. The tests were also done with these same test channels configured for 100G PM-QPSK transmission with RRC shaping at 1 and 0.1 roll-off conditions. Figure 5(a) presents the back-to-back results for the two test channels with different shaping conditions for a range of OSNR levels of 10 to 20dB for 100G PM-QPSK and 20 to 30dB for 200G PM-16QAM.

Fig. 5 a) Back-to back OSNR_NI results for 5 shaping conditions of two tests channels for 0.1nm OSNR levels ranging from 10 to 30dB. b) Same channels after 8 spans of SSMF for varied launch power.

Download Full Size | PDF

The deviations are typically within ± 0.5dB of the target OSNR value with 13 of the 120 data points exceeding ± 0.5dB, all of which occur at OSNR levels >25dB. A maximum deviation corresponding to a 1.24dB OSNR underestimation was observed for a 28dB target and a maximum overestimation of >3dB was obtained for a 29.6dB OSNR. It should be highlighted here that this overestimation was present only with one of the two OSAs and provides valuable information on key performances parameters of an OSA (e.g. power linearity at scale changes) that could limit the measurement capabilities for field deployment of this technique and that require further investigation.

Figure 5(b) shows the deviation results after transmission for the same channels while varying per-channel launch power, as before, relative to optimum. At optimum launch power of 0 dBm, the OSNR levels were close to 24dB for all cases. For moderate non-linear conditions observed up to the optimum launch power, the deviations are typically within ± 0.5dB with a worst case deviation of −0.9dB occurring at the optimum launch power. At higher launch powers, the deviations remain typically within ± 1dB except for the highest launch power case, at 3dB above optimum, where the worst deviation results in an almost 2dB underestimation of OSNR for the 100G QPSK channels and 1.5dB underestimation for the 16QAM channels. It is again worth noting however that this condition also corresponds to a OSNR of 27dB for which larger deviations are also observe for the back-to-back test case of Fig. 5(a). Overall the results are similar to those of the unshaped QPSK case presented earlier, showing a trend to underestimate the OSNR as non-linearity increases, but the measured deviations are slightly higher than for the unshaped QPSK case. The choice of a simple adaptive zone selection, and corresponding correction factor, based on dP_Ref appears slightly limiting in light of the range of shaping conditions and signal widths tested. More complex multifactored adaptative processes could be investigated to further improve the performance.

As a logical next step, the performance of the method was evaluated in a configuration with tighter channel spacing, where shaping is most likely to be used. For this test, three RRC shaped 100G PM-QPSK channels with 0.3 roll-off were set on a 37.5GHz grid and their OSNR was measured back-to-back and after 8 spans of SSMF fiber. The processing used in this case was very similar but it was necessary to develop additional steps for detecting the reduced channel spacing and, more importantly, for detecting the presence of non-linear spectral deformation to allow for a stronger correction of that deformation when the channel spacing is reduced (e.g. the same level of ΔRdev(λ) yields a larger ΔOSNR(λ)). The processing was applied to the central signal of the group which had two 37.5GHz spaced neighbor channels.

In a similar format, Fig. 6 presents the measurement results for the back-to-back and through fiber configurations. For the back-to-back case, Fig. 6(a) shows the results and the measured deviations typically remain within ± 0.5dB of the target value up to 25dB OSNR except for one measurement point with one OSA showing a deviation of 1.2dB at 25dB OSNR. At higher OSNR levels, larger deviations are observed with a maximum deviation of nearly 3dB again on the same OSA at the highest OSNR (almost 30dB). This again highlights a dependence of the measurement technique on key performance parameters of the OSA. Overall, the results for the tighter spacing also show larger deviations than those that were observed for the 50GHz spaced signals with similar shaping. This is expected since tighter channel spacing further limits the useable working range where the conditions of applicability for Eq. (5) are met and the adaptive processing limitations mentioned earlier are reached at lower OSNR levels.

Fig. 6 a) Back-to back OSNR_NI results 100G PM-QPSK RRC shaped 37.5GHz spaced channel for 0.1nm OSNR levels from 13 to 30dB. b) Same channel after transmission while varying launch power and after adding ASE.

Download Full Size | PDF

Figure 6(b) shows the results for the non-linear regime where the stronger applied correction leads to an overestimation of OSNR of up to almost 1dB for launch powers below optimum (corresponding to −2 dBm per channel power and a target OSNR just below 24dB). For higher launch powers, measurement deviations leading to underestimation of OSNR up to 2dB at the highest launch were measured. This behavior is similar to that observed with the RRC shaped QPSK signals presented earlier. These deviations are however happening at ~1dB lower target OSNR in the tightly spaced case. The stronger correction applied provides usable results with deviations within ± 1dB up to 2dB above optimum launch power but the fact that overcorrection is observed for lower launch powers suggests that applying the same constant correction factor on the entire range of non-linear cases present may no longer be adequate for the tighter spacing condition. A more sophisticated correction factor that increases when non-linearity increases could be more appropriate but this requires further investigation.

The tight channel spacing case was tested in an 8 spans SSMF configuration with an optimum launch power achieved at an OSNR level of 24dB. Since this is relatively high for a system operating with only QPSK channels, ASE was added to set the target OSNR level at ~17dB which is a more common QPSK link condition. The measurement deviations for this test case are also presented at Fig. 6(b) and they remain typically within ± 0.5dB with only the highest launch power condition exceeding ± 0.5dB (at 0.9dB). These results illustrate the specific contribution of NL-induced spectral deformations affecting the correspondence between Δκ_NL(λ) and ΔRdev(λ) for a lower OSNR level where a similar 2% ASE to signal power ratio would have been easily discriminated in the linear regime. They also show that, despite the limitations to the useable spectral region and stronger non-linear spectral deformation observed with tighter spacings, the non-intrusive OSNR measurement technique presented can provide very good OSNR measurements when the OSNR levels are below 20dB.

5. Conclusion

A non intrusive OSNR measurement technique applicable to densely spaced spectrally shaped polarization multiplexed signals operating in the linear and non-linear regime of fiber transmission was described along with its underlying algorithmic processes. The key equations, validity conditions and algorithmic steps required to perform the detailed spectral comparison with a “noise-free” signal spectrum and to determine the ASE contribution (thus OSNR) were presented. Most importantly the algorithm yields accurate ASE-induced OSNR even when moderate non-linear induced spectral deformations also contribute to the measured spectral changes. This method can be used for in-service troubleshooting or long-term periodic maintenance, and does not require extra equipment beyond existing OSAs already widely used for OSNR measurement and monitoring.

The measurement technique’s performance was demonstrated over a wide range of operating conditions. The results for unshaped 100G PM-QPSK signals were typically within ± 0.5dB from the target for a wide range of OSNR levels in the back-to-back condition and in the moderate non-linear regime conditions, up to the optimum launch power. These results were confirmed in different system configurations including the presence of 10G NRZ neighbors and in-line dispersion compensation. The technique was also demonstrated on RRC shaped signals with similar deviations below ± 0.5dB in the back-to-back and moderate non-linear operating condtions. For shaped signals, larger deviations were observed at higher OSNR levels and for higher launch powers (i.e., stronger non-linearity), as well as a higher sensitivity to the OSA’s measurement performance. The latter provides useful information for specifying and optimizing OSA parameters for operational field use. Finally, the technique was demonstrated with 37.5GHz spaced RRC shaped signals representative of flexgrid configurations. In this configuration, the back-to-back condition yielded similar results to the previous 50GHz spaced RRC shaped condition but with a larger dependency on OSA intrinsic properties. When operating in the non-linear regime at lower OSNR levels around 17dB, the technique performed well with deviations similar to those observed with the previous signal configurations tested and remained within ± 0.5dB at launch powers up to 2dB above the optimum launch condition. However, in the moderate non-linear regime up to optimum launch power, larger deviations up to ± 1dB were measured at higher OSNR levels and deviations in excess of 2dB were obtained at the highest launch power condition where non-linear effects were larger. Further efforts will be directed at improving the performance of the technique for the tighter spacing shaped signals and to improve the OSA characteristics to reduce the OSA dependence of the measurements observed in some of the tests. Further investigations will also be undertaken to evaluate if knowledge of the ROADMs transfer function can be used and working region selection adapted in order to extend the applicability of the technique to cases where in-line filtering is present.

References and links

1. Optical monitoring for dense wavelength division multiplexing systems: ITU-T Recommendation G.697 (02/12).

2. Optical signal-to-noise ratio measurement for dense wavelength-division multiplexed systems: IEC 61280–2-9 (Edition 2.0 2009–02).

3. M. Rasztovist-Wiech, M. Danner, and W. R. Leeb, “Optical signal-to-noise ratio measurement in WDM networks using polarization extinction,”in ECOC 1998 (1998), pp. 549–550.

4. J. H. Lee and Y. C. Chung, “Improved OSNR monitoring technique based on polarization-nulling method,” Electron. Lett. 37(15), 792–793 (2001). [CrossRef]

5. D. Gariépy, G. He, Y. Breton, B. Dery, and G. W. Schinn, “Novel OSA-based method for in-band OSNR measurement,” in OFC (2010), paper JThA15.

6. F. Vacondio, O. Rival, C. Simonneau, E. Grellier, A. Bononi, L. Lorcy, J.-C. Antona, and S. Bigo, “On nonlinear distortions of highly dispersive optical coherent systems,” Opt. Express 20(2), 1022–1032 (2012). [CrossRef] [PubMed]

7. A. Carena, V. Curri, G. Bosco, P. Poggiolini, and F. Forghieri, “Modeling of the Impact of Nonlinear Propagation Effects in Uncompensated Optical Coherent Transmission Links,” J. Lightwave Technol. 30(10), 1524–1539 (2012). [CrossRef]

8. Z. Dong, F. N. Khan, Q. Sui, K. Zhong, C. Lu, and A. P. T. Lau, “Optical performance monitoring: A review of current and future technologies,” J. Lightwave Technol. 34(2), 525–543 (2016). [CrossRef]

9. M. R. Chitgarha, S. Khaleghi, W. Daab, M. Ziyadi, A. Mohajerin-Ariaei, D. Rogawski, M. Tur, J. D. Touch, V. Vusirikala, W. Zhao, and A. E. Willner, “ Demonstration of WDM OSNR Performance Monitoring and Operating Guidelines for Pol-Muxed 200-Gbit/s 16-QAM and 100-Gbit/s QPSK Data Channels,” in OFC/NFOEC (2013), paper OTh3B.6.

10. Y. C. Chung, S. K. Shin, and K. J. Park, “A novel optical signal-to-noise ratio monitoring technique for WDM networks,” in OFC (2000), paper WK6.

11. T. Saida, I. Ogawa, T. Mizuno, K. Sano, H. Fukuyama, Y. Muramoto, Y. Hashizume, H. Nosaka, S. Yamamoto, and K. Murata, “In-band OSNR monitor with high-speed integrated Stokes polarimeter for polarization division multiplexed signal,” Opt. Express 20(26), B165–B170 (2012). [CrossRef] [PubMed]

12. L. Dou, Z. Tao, Y. Zhao, S. Oda, Y. Aoki, T. Hoshida, and J. C. Rasmussen, “Differential pilots aided in-band OSNR monitor with large nonlinear tolerance,” in Proc. OFC (2015), paper W4D.3. [CrossRef]

13. H. G. Choi, J. H. Chang, H. Kim, and Y. C. Chang, “Nonlinearity-tolerant OSNR estimation technique for coherent optical systems,” in Proc. OFC (2015), paper W4D.2. [CrossRef]

14. D. Gariépy, G. He, and G. W. Schinn, “Non-intrusive measurement of In-band OSNR of high bitrate polarization-multiplexed signals,” Opt. Fiber Technol. 17(5), 518–522 (2011). [CrossRef]

15. D. Gariépy, S. Searcy, G. He, and S. Tibuleac, “Non-intrusive In-band OSNR Measurement of Polarization Multiplexed Signals Operating in the Non-Linear Regime,” in Proc. OFC (2015), paper W4D.6. [CrossRef]

16. D. Gariépy, S. Searcy, G. He, and S. Tibuleac, “Demonstration of non-intrusive in-band OSNR measurement technique for PM-16QAM signals with spectral shaping and subject to fiber nonlinearity,” in OFC (2016), paper Tu3G.5.

Test channel(s)	Neighbor channels	Spacing	SSMF Fiber Link	Results
100G PM-QPSK (unshaped)	100G PM-QPSK (unshaped)	50 GHz	8 spans, no CDC	Figure 3
100G PM-QPSK (unshaped)	100G PM-QPSK (unshaped)	50 GHz	16 spans, no CDC	Figure 4
100G PM-QPSK (unshaped)	10G NRZ-OOK (no GB)	50 GHz	16 spans, w/CDC	Figure 4
100G PM-QPSK (unshaped)	10G NRZ-OOK (300GHz GB)	50 GHz	16 spans, w/CDC	Figure 4
100G PM-QPSK (r = 1.0,0.1)	100G PM-QPSK (unshaped)	50 GHz	8 spans, no CDC	Figure 5
200G PM-16QAM (r = 1,0.3,0.1)	100G PM-QPSK (unshaped)	50 GHz	8 spans, no CDC	Figure 5
100G PM-QPSK (r = 0.3)	100G PM-QPSK (r = 0.3)	37.5GHz	8 spans, no CDC	Figure 6

Non-intrusive OSNR measurement of polarization-multiplexed signals with spectral shaping and subject to fiber non-linearity with minimum channel spacing of 37.5GHz

Abstract

1. Introduction

2. Description of reference-based spectral OSNR-measurement method

3. Experimental configuration

4. Experimental results and discussion

5. Conclusion

References and links

Cited By

Figures (6)

Tables (1)

Equations (9)

Optics Express