This is a continuation from the previous tutorial -Photonic integrations
1.INTRODUCTION
Fiber-optic communication has seen tremendous growth over the last decade fueled mainly by the incessant and relentless demand for high capacity. This insatiable demand is spurred by the Internet traffic growth both in terms of number of users and the bandwidth consumed by each user. This trend of multifold increase in data traffic every year is expected to continue in the foreseeable future. In order to comply with the enormous bandwidth requirements posed by the growth in data traffic, fiber-optic communication networks have evolved drastically.
For example, dense wavelength-division multiplexing \(\text{(DWDM)}\) technology, which increases the number of carrier wavelengths and hence data rates, has been introduced.
Advanced optical modulation formats offering high spectral efficiencies have been successfully employed in conjunction with digital coherent receivers. The transmission distances between regenerative repeaters have been increased significantly with theintroduction of optical amplifiers, efficient dispersion compensation techniques, and forward-error-correction \(\text{(FEC)}\) coding.
Furthermore, complex network architectures utilizing reconfigurable optical add-drop multiplexers \(\text{(ROADMs)}\) have been incorporated in order to promote dynamicity, flexibility, and better utilization of available transmission capacity.
All these developments have paved the way for a multifold increase in data rates currently supported by the modern fiber-optic networks.
The performance of optical networks operating at ultra-high data rates and over long transmission distances strongly depends on the extent of impairments introduced into the signals by the optical fiber as well as other network elements. Conventionally, the deleterious channel effects are handled by either introducing some safety margins while designing optical networks or by making attempts to compensate some of these impairments manually.
Unfortunately, due to stochastic nature of some of these impairments, it is difficult if not impossible to compensate them entirely. The advent of \(\text{ROADMs}\) makes the situation overwhelmingly more complicated since it allows the signal path to change dynamically, thereby making the impairments in reconfigurable fiber-optic networks become path dependent, dynamic and hence, random in nature.
As the optical networks grow larger, faster and more dynamic in nature, the control and management of these networks is rapidly becoming an arduous task. Unlike wireless networks where all the necessary networking issues such as link setup, optimization, and testing are performed automatically, such tasks are currently handled manually in optical networks requiring substantial human resources and time.
This is due to the fact that the existing fiber-optic networks are not capable of acquiring real-time information about the physical state of the network and the health of the signals propagating through the network. The price paid for this lacking of vital information is that the network designers are forced to keep considerable safety margins, in order to provide reasonable level of reliability, resulting in wastage of precious network resources. The designers may also be compelled to use more aggressive component specifications.
In order to reduce the operating costs, ensure optimum resources utilization and guarantee adequate operation and management of dynamic optical networks, it is essential to have the capability of continuous monitoring of network performance parameters.
Optical Performance Monitoring and Their Roles in Optical Networks
The capacity of optical communication systems is increasing unremittingly and the architectures of optical networks are continuously becoming more complex, transparent, and dynamic in nature.
These high-capacity fiber-optic networks are vulnerable to several transmission impairments, which can alter over time due to dynamic nature of these networks. Since each fiber carries an enormous amount of data traffic, even a brief disruption of services may result in disastrous consequences.
Therefore, it is imperative to incorporate effective monitoring mechanisms across the whole fiber-optic network, which could provide precise and real-time information about the health of each individual \(\text{DWDM}\) channel. Optical performance monitoring \(\text{(OPM)}\) is an enabling technology and a potential mechanism for the control,management, and maintenance of existing and future high-speed reconfigurable optical networks.
Quintessentially, \(\text{OPMis}\) a set of measurements performed on an optical signal at the intermediate network nodes or inside the receiver itself so as to estimate the performance of a transmission network. Performance monitoring has been a part of fiber-optic transmission systems from the very beginning, for example, bit/block-error-rate \(\text{(BER)}\) monitoring and other quality-of-service (QoS) measurements.
However, such monitoring has fundamentally been conducted in the electronic domain, that is, after optical-to-electronic \(\text{(O/E)}\) conversion of the signal being monitored. For the efficient operation of dynamic \(\text{DWDM}\) networks, it is crucial to monitor the key performance parameters directly in the optical domain.
That is why, \(\text{OPM}\) is expected to assess the quality of a data channel by estimating its optical characteristics without having a priori knowledge of the transmitted sequence of bits as shown in Figure 1.
It is envisagedthat \(\text{OPM}\) will be indispensable for the efficient operation and management of existing as well as future complex and dynamic optical networks. The incorporation of \(\text{OPM}\) in fiber-optic networks can enhance the network reliability and minimize the network down time. \(\text{OPM}\) can also facilitate the efficient utilization of available networks resources through significant reduction in systems’ safety margins currently used to ensure error-free operation.
Other important benefits of \(\text{OPM}\) include \(\text{(i)}\) reduced network operation and maintenance costs, \(\text{(ii)}\) effective control of network elements, that is, fault detection, localization, and troubleshooting of failures, and \(\text{(iii}\)) link setup and optimization. \(\text{OPM}\) can also empower the carriers and network operators to certify service-level agreements \(\text{(SLA)}\) and to guarantee certain \(\text{QoS}\) provision to their clients.
Network Functionalities Enabled by OPM
The installation of monitoring mechanisms across the whole optical network, which could continuously monitor the health of network as well as the data signals, may
enable several advanced network functionalities. Some of the areas where \(\text{OPM}\) can play a significant role are discussed in this section.
Adaptive Impairments Compensation The transmission impairments in dynamic optical network are inherently time-variant in nature. In fixed point-to-point links, these changes occur due to certain physical effects while in dynamic networks, the impairments vary because the paths traversed by the optical signals change continuously due to network reconfiguration.
This implies that the compensation techniques used in dynamic optical networks must be adaptive in nature. The realization of adaptive compensation techniques requires precise information about the extent of impairments introduced by the link. \(\text{OPM}\) can effectively assess the degradations caused by an optical link. This information available through \(\text{OPM}\) can then be utilized to provide feedback signals for the adaptive compensation of these impairments as shown in Figure 2.
Reliable Network Operation \(\text{OPM}\) can provide continuous and real-time information about the physical condition of an optical network and is thus capable of identifying the faults’ locations as well as their causes. Furthermore, it can facilitate the acquisition of information about the extent of individual impairments contributed by the network as well as their distribution.
This allows the network providers to know when the data signals are beginning to deteriorate. Therefore, preventive measures can be taken at right time to fix the problem before it starts to cause serious degradation to system performance, thus enabling reliable network operation as shown in Figure 3.
Efficient Resources Allocation \(\text{OPM}\) can facilitate efficient utilization of available network resources. For example, if the link quality is too good in a dynamic optical network, this information can be conveyed to the transmitter, which may effectively reduce the transmitted signal power and thus decrease optical signal-to-noise
ratio \(\text{(OSNR)}\), while still meeting the desired \(\text{BER}\) requirements. The reduction in transmitted optical power may enhance the signal’s robustness against the undesirable nonlinear distortions. On the contrary, the transmitted power of each channel may be increased (and hence, increase \(\text{OSNR})\) accordingly, when the \(\text{BER}\) rises above the required specifications. Alternatively, the data rate can be increased by using higher modulation formats if the \(\text{OSNR}\) of the link is monitored to be good.
Impairment-Aware Routing The data-routing algorithms used in existing static optical networks either route traffic on shortest paths (i.e., fewest hops between source and destination) or on paths that satisfy certain minimum \(\text{QoS}\) constraints (e.g., delay, packet loss, and data rate).
However, such routing algorithms will perform far from optimum in dynamic optical networks since they do not take the variable physical state of the network into account. Therefore, in order to have better routing capabilities, the routing tables must be updated by taking optical layer parameters (e.g., fiber length, signal distortion, amplifier noise and transients) into consideration as shown in Figure 4.
The valuable information available through \(\text{OPM}\) can be provided to the network controllers, which may assign different weights to all possible paths considering various parameters. The routing decisions can subsequently be made based on these calculations.
Network Parameters Requiring \(\text{OPM}\)
Despite being a terrific transmission medium, an optical fiber introduces several signal impairments, which may degrade the performance of an optical network. Apart from optical fiber, other network components may also contribute several degradations as shown in Figure 5. The effects of all these impairments in downgrading the quality of the transmission link are strongly dependent on the data rates of the signals being transmitted. Impairments in optical networks can generally be categorized as catastrophic and noncatastrophic in nature. Catastrophic impairments result
in loss of optical power and include fiber breaks, network components failures, inappropriately installed network equipment, and so on. Noncatastrophic impairments do not necessarily decrease the optical signal power but they may severely distort the signal. Such distortions may be linear or nonlinear in nature and must be minimized or properly compensated so as to guarantee the desired network performance.
Optical Power The most fundamental parameter to be monitored in an optical network is the optical power. The optical power may decrease substantially due to fiber attenuation (caused by absorption and scattering) and losses encountered at the fiber connectors, splices, and couplers. In wavelength-division multiplexed \(\text{(WDM)}\) systems, the information about each channel’s power is required so as to dynamically equalize the power in each of these channels through a feedback mechanism, thus ensuring a stable system performance.
This can be accomplished either by demultiplexing all optical channels and monitoring their powers using several photodetectors or alternatively in a cost-effective manner by making use of tunable optical filters in conjunction with a single photodetector.
\(\text{OSNR}\) Optical amplifiers such as erbium-doped fiber amplifiers \(\text{(EDFAs)}\) are normally employed in optical networks to compensate for the transmission losses over long distances. However, besides providing optical gain, \(\text{EDFAs}\) also add unwanted amplified spontaneous emission \(\text{(ASE)}\) noise into the optical signal. Furthermore, the cascading of \(\text{EDFAs}\) results in accumulation of \(\text{ASE}\) noise. \(\text{ASE}\) noise is typically quantified by \(\text{OSNR}\) and is one of the most important parameters to be monitored in optical networks since the \(\text{BER}\) is directly related to the signal \(\text{OSNR}\). Furthermore, it also plays a pivotal role in fault diagnosis and as a measure of general health of links in an optical network.
Chromatic Dispersion (CD) \(\text{CD}\) belongs to the category of dispersive impairments and is one of the most crucial parameters to be monitored in dynamic optical networks. It arises due to the frequency-dependent nature of refractive index in an optical fiber and is one of the major limitations in high-speed, long-distance fiber-optic transmission systems. In reconfigurable optical networks, changes in path lengths for a given channel (due to switching) may result in variable amounts of accumulated \(\text{CD}\).
Therefore, fixed \(\text{CD}\) compensation techniques are less effective in such scenarios. For robust high-speed systems, it is essential to compensate \(\text{CD}\) adaptively within tight tolerances, making \(\text{CD}\) monitoring imperative in such systems.
Polarization-Mode Dispersion (PMD) Polarization-mode dispersion \(\text{(PMD)}\) is another dispersive impairment that needs to be effectively monitored since it is a major limitation in fiber-optic networks operating at data rates in excess of 40 Gbps. \(\text{PMD}\) is caused by the ellipticity of fiber geometry.
Since the same spectral component of an optical signal splits into two orthogonal states-of-polarization \(\text{(SOP)}\) inside a fiber, these two spectral copies have slightly different propagation speeds due to fiber birefringence. Hence, they reach the receiver at slightly different times resulting in pulse broadening. \(\text{PMD}\) effects are stochastic, time-variant, and temperature and data rate dependent.
Fiber Nonlinearity Fiber nonlinearity is also a critical parameter to be monitored. The nonlinear effects arise in an optical fiber due to the power dependence of refractive index and result in interference and crosstalk between different \(\text{WDM}\) channels when the transmitted power surpasses a certain limit.
Fiber nonlinearities deteriorate the performance of high-speed \(\text{WDM}\) systems employing advanced modulation formats. Therefore, fiber nonlinearities need to be well managed and for that objective, monitoring is mandatory.
Quality-Factor (Q-Factor) \(Q\)-factor is an important parameter to assess the overall performance of an optical network. It is used to analyze the performance of transmission systems for which direct measurement of \(\text{BER}\) is impractical. \(Q\)-factor is an indicator of the quality of an optical signal due to strong correlation between \(Q\)-factor and \(\text{BER}\).
It can be used to evaluate the effects of channel degradations such as \(\text{ASE}\) noise, \(\text{CD}\), \(\text{PMD}\), fiber nonlinearities as well as the impairments introduced by the transmitter and receiver, thus allowing effective \(\text{BER}\) estimation. Factors including non-Gaussian nature of noise, crosstalk and signal distortions may culminate in an inaccurate \(\text{BER}\) estimation using \(Q\)-factor.
BER The \(\text{BER}\) is an ultimate measure of the quality of an optical link. It is statistically defined as the time-averaged fraction of erroneous bits contained in a given bit stream. \(\text{BER}\) monitoring has traditionally been used as a preferable tool to characterize the overall performance of a system.
However, \(\text{BER}\) is merely a number and does not provide any insight into the individual contributions of different impairments toward the degradation of system performance. In addition, \(\text{BER}\) monitoring requires expensive equipments (such as clock and data recovery systems).
Wavelength Shift The ever-increasing demand for larger link capacity has resulted in a significant reduction in channel spacing in \(\text{DWDM}\) systems. As the interchannel spacing decreases, the requirements for wavelength control of optical components become extremely stringent.
This demands the incorporation of sophisticated monitoring mechanisms in optical networks that can detect or compare the wavelengths because a relative offset between the centre frequencies of \(\text{DWDM}\) channels, and the optical filters may cause significant power loss and may also result in crosstalk between the channels.
Apart from the aforementioned parameters, other factors such as optical amplifier gain and distortions, crosstalk and interference effects, \(\text{SOP}\) and polarization-dependent effects, pulse shape and timing jitter are also useful to be monitored in optical networks.
Desirable Features of OPM Techniques
The features that a given \(\text{OPM}\) technique is expected to demonstrate are determined by several factors such as the nature of the optical network in which the \(\text{OPM}\) module is anticipated to be deployed, types and extents of impairments prevalent in the network, data rates, implementation cost, and the degree of intelligence sought to be incorporated with the inclusion of \(\text{OPM}\) modules. Some common features expected from the \(\text{OPM}\) techniques are discussed in the following sections.
Accuracy, Sensitivity and Dynamic Range The precise compensation of impairments in dynamic optical networks depends on the degree of accuracy of the monitoring technique being used. Therefore, the monitoring techniques are expected to meet the desired accuracy requirements. Apart from accuracy, the techniques are also anticipated to demonstrate good sensitivities in the whole monitoring range.
Inaddition, the techniques are expected to exhibit broad monitoring ranges in order to enable appropriate compensation of impairments occurring in wide dynamic ranges. The accuracy, sensitivity, and dynamic range of a monitoring technique may rely on a number of factors such as the methodology being employed in the \(\text{OPM}\) module and the amount of signal power tapped for monitoring purposes.
Multichannel Operation Since fiber-optic networks encompass multiple data channels by incorporating \(\text{WDM}\), the \(\text{OPM}\) techniques used must be capable of monitoring several data channels. This can be achieved by either using a parallel bank of monitoring devices or a tunable optical filter to select a particular channel for sequential monitoring. The parallel operation requires more number of devices and thus involves higher hardware costs. On the contrary, the sequential operation may introduce measurement latency, especially in systems with large number of data channels.
Multi-Impairment Monitoring As discussed in this Section Network Parameters Requiring \(\text{OPM]\), several impairments may coexist in an optical network. If different techniques are employed for the monitoring of individual impairments then it will increase the monitoring costs immensely.
Therefore, the monitoring techniques must be capable of monitoring multiple network impairments simultaneously as well as independently.
Data Rate and Modulation Format Transparency The future optical networks are envisioned to contain data traffic with different modulation formats and data rates in individual channels.
Therefore, the developed \(\text{OPM}\) techniques must be transparent to data rates and modulation formats, thus avoiding the need for the modification of monitoring modules.
Cost-Effectiveness Since \(\text{OPM}\) may be needed at multiple locations in an optical network, a general requirement is that its cost must be relatively lower than that of conventional testing equipment.
The cost of \(\text{OPM}\) module may depend on the complexity of the technique being employed for monitoring. A reduction in cost can be achieved by using techniques capable of monitoring multiple network impairments independently for several data channels and for various data rates and modulation formats.
Operation at Low Input Power For monitoring purposes, part of the signal power needs to be tapped from the optical link. The \(\text{OPM}\) module must be capable of performing its operation by exploiting only a small fraction of the signal power while still meeting the accuracy and sensitivity requirements.
A general rule of thumb is that the power used for monitoring must not exceed a small percentage of the total signal power.
Fast Response Time In static optical networks, the response time of the \(\text{OPM}\) technique can be of the same order of magnitude as the network restoration time of 50 ms.
However, in case of dynamic optical networks, the response time must be much smaller than the network reconfiguration interval. A general rule of thumb is that the monitoring time must be in the range of a few milliseconds.
Passiveness The \(\text{OPM}\) technique must not have an adverse effect on the normal operation of an optical network. This requires the \(\text{OPM}\) technique to not modify the network components while it performs the monitoring task. Also, it should not insert additional monitoring signals into the network, which may interfere with the data signal resulting in the degradation of data signal quality.
2. OPM TECHNIQUES FOR DIRECT DETECTION SYSTEMS
The receivers in direct detection systems employ simple photodetectors to detect the intensity of the optical signal (e.g., in case of on-off-keying \(\text{(OOK)}\) modulation scheme) or use delay interferometers \(\text{(DI)}\) in conjunction with photodetectors to transform the differential phase information into amplitude information in the electrical domain (e.g., in case of differential binary/quaternary phase-shift keying \(\text{(DBPSK/DQPSK)}\) modulation schemes).
Due to square-law detection nature of the direct detection receivers, only limited information can be retrieved from the optical signals, whereas the receivers in coherent detection systems perform down-conversion by using a combination of local oscillator \(\text{(LO)}\) lasers, optical hybrids, and photodetectors in conjunction with digital signal processing \(\text{(DSP)}\) modules so as to retrieve the information contained in the amplitude, phase, frequency, and polarization of the carrier. In this case, it is possible to retrieve all the information from an optical signal in the electrical domain. Coherent detection techniques allow the use of advanced higher-order modulation formats such as \(m\)-\(\text{PSK}\) and \(m\)-\(\text{QAM}\) along with polarization-division multiplexing \(\text{(PDM)}\), thereby enabling higher spectral efficiencies.
Due to the absence of high-speed and economical sampling and \(\text{DSP}\) devices in the past, most of the currently deployed fiber-optic communication networks employ direct detection receivers. Irrespective of whether the optical networks are using direct detection or coherent detection techniques, \(\text{OPM}\) is equally important for the reliable and efficient operation of these networks. However, the requirements, nature of parameters requiring monitoring, and the scope and network functionalities enabled by \(\text{OPM}\) are generally different for these two types of networks.
OPM Requirements for Direct Detection Optical Networks
The crucial parameters to be monitored in networks employing direct detection include residual \(\text{CD}\), total power (i.e., signal plus noise power), \(\text{OSNR}\), \(\text{PMD}\), polarization-dependent loss \(\text{(PDL)}\), and fiber nonlinearities (i.e., self-phase modulation \(\text{(SPM)}\), cross-phase modulation \(\text{(XPM)}\), four-wave mixing \(\text{(FWM)}\), and scattering). The \(\text{OPM}\) techniques developed for existing direct detection systems are preferred to utilize simple direct detection.
This is due to the fact that the complexity and cost of a coherent receiver is relatively higher and thus it may not be ideal for use in monitoring units deployed at the intermediate network nodes where cost is a major limitation. However, the adverse effect of using direct detection in these monitoring devices is that only limited information about the optical signal can beretrieved in the electrical domain, thereby making the accurate estimation of various network parameters quite challenging.
Direct detection fiber-optic transmission systems are typically dispersion-compensated systems.
However, due to network reconfigurability enabled by \(\text{ROADMs}\) as well as due to temperature and other physical effects, the \(\text{CD}\) of the link varies dynamically. Therefore, despite the use of dispersion compensation techniques in these networks, there is always some residual \(\text{CD}\) present.
Furthermore, most of the currently deployed optical fibers have reasonably high \(\text{PMD}\) coefficient values, that is, of the order of \(0.5\) \(\text{ps}\)/\(\text{km}\)\(^{1/2}\). Hence, the optical signals are also affected by polarization-related distortions.
Finally, the optical signals in direct detection systems are also subjected to degradations caused by the \(\text{ASE}\) noise of \(\text{EDFAs}\) as well as fiber nonlinearities. Since all the above-mentioned deleterious channel effects may coexist, the \(\text{OPM}\) techniques developed for direct detection optical networks must be capable of monitoring these parameters independently. For example, the monitoring of \(\text{OSNR}\) should not be affected by the presence of \(\text{CD}\) and \(\text{PMD}\).
However, as mentioned earlier, the availability of limited information about the optical signal in the \(\text{OPM}\) devices (due to the use of simple direct detection) makes simultaneous and independent monitoring of multiple performance parameters overwhelmingly complicated in direct detection optical networks.
Overview of \(\text{OPM}\) Techniques for Existing Direct Detection Systems
Over the past few years, a plethora of techniques for monitoring optical signal quality parameters in direct detection fiber-optic communication networks have been proposed. The general classification of these techniques is depicted in Figure 6.
The existing monitoring techniques can be classified as either digital or analog in nature. Digital techniques exploit the digital information content of the signal waveform in the electrical domain. Digital \(\text{OPM}\) methods, for example, \(\text{BER}\) monitoring, provide information about the overall degradation of the system caused by the network impairments but are unable to isolate their individual contributions. Analog monitoring techniques make use of the specific characteristics of the analog signal waveform to extract information about the channel impairments.
These techniques can be further subdivided into time-domain, frequency-domain, and polarization-domain techniques depending upon whether the monitoring information is extracted from the signal waveform, signal spectrum or the signal polarization, respectively.
Time-domain monitoring techniques can be categorized into synchronous and asynchronous sampling-based techniques depending upon whether the sampling rate is synchronized with the symbol rate. Synchronous sampling techniques require clock recovery, which is a relatively complex operation especially in networks supporting multiple data rates.
Eye-diagram is a typical synchronous sampling-based technique, which qualitatively reflects the effects of all impairments on the signal quality. However, it is unable to quantify the effects of individual impairments. Similarly, \(Q\)-factor monitoring is another popular synchronous
sampling-based technique, which is commonly used in practice due to its strong correlation with \(\text{BER}\). Asynchronous sampling-based techniques, for example, asynchronous amplitude histograms \(\text{(AAHs)}\), asynchronous delay-tap plots \(\text{(ADTPs)}\), asynchronous two-tap plots \(\text{(ATTPs)}\), and asynchronous single-channel sampling \(\text{(ASCS)}\), are considered attractive due to the fact that they do not require clock information and they are also capable of monitoring multiple impairments simultaneously, thus being cost-effective.
Frequency-domain monitoring techniques can be subdivided into optical and radio frequency \(\text{(RF)}\) spectrum-based techniques. The optical spectrum analysis techniques can make use of an optical filter, which is tuned over the channel bandwidth and the optical power is recorded. The spectral resolution in this case is determined by the filter’s bandwidth.
Alternatively, optical spectrum can be analyzed by performing hom*odyne detection, where a tunable \(\text{LO}\) laser signal is mixed with the monitored signal and the interference signal is then analyzed for spectral analysis. The \(\text{LO}\) laser is swept across the channel bandwidth. The spectral resolution in this case is determined by the linewidth of the \(\text{LO}\) laser and is several orders of magnitude higher than that of tunable optical filter.
The optical spectrum-based techniques are capable of monitoring out-of-band \(\text{OSNR}\), total optical power, and wavelength drift but they cannot monitor \(\text{CD}\) and \(\text{PMD}\). These techniques can be used to monitor multiple WDM channels. Since the optical filter or the \(\text{LO}\) laser needs to be tuned for scanning the whole \(\text{WDM}\) spectrum (which may require some time), such techniques can introduce measurement latency.
Radio frequency spectrum-based monitoring techniques can provide better estimation of signal quality as compared with optical spectrum-based techniques because they analyze the spectrum of the signal that is encoded on the optical carrier. \(\text{RF}\) spectrum-based techniques can either make use of clock tones present inherently in the spectrum of various modulation formats or insert pilot tones of different frequencies in each channel at the transmitter. The clock tones-based monitoring techniques can measure \(\text{CD}\) and \(\text{PMD}\) and are data rate and modulation format dependent.
On the contrary, pilot tones-based schemes can measure various parameters such as wavelength, \(\text{OSNR}\), \(\text{CD}\), and \(\text{PMD}\) and are data rate as well as modulation format independent. However, the adverse effect of pilot tones-based techniques is that the tones interfere with the data signal resulting in the deterioration of \(\text{BER}\). Apart from monitoring the specific tones (i.e., clock and pilot tones) in the \(\text{RF}\) spectrum, changes in the spectral distribution of overall \(\text{RF}\) spectrum, due to various network impairments, may also be exploited for the monitoring of these impairments.
Polarization-domain monitoring techniques exploit the polarization properties of the optical signal. The alterations in the polarization characteristics due to various channel degradations can be utilized for the effective monitoring of these impairments.
These techniques can monitor signal and noise powers (and hence, \(\text{OSNR}\)), for example, through polarization nulling, as well as \(\text{PMD}\) of the fiber link, for example, by measuring the degree-of-polarization \(\text{(DOP)}\) of the received signal. These techniques have the advantage of being transparent to data rates and modulation formats. However, they cannot be applied to polarization-multiplexed signals, which severely limits their use in coherent transmission systems.
Electronic DSP-Based Multi-Impairment Monitoring Techniques for Direct Detection Systems
\(\text{OPM}\) techniques utilizing electronic \(\text{DSP}\) have gained substantial attention in recent years. \(\text{DSP}\)-based monitoring techniques exploit the statistical properties of the data signals after \(\text{O/E}\) conversion for the estimation of critical signal quality parameters.
The reason behind the popularity of \(\text{DSP}\)-based \(\text{OPM}\) techniques is that they can facilitate cost-effective monitoring of multiple signal quality parameters simultaneously for several data rates and modulation formats and without necessitating modifications of monitoring hardware.
Furthermore, the monitoring of a new parameter as well as monitoring for a different signal type can simply be enabled by downloading the relevant algorithm to the DSP-based monitor, thereby facilitating flexibility and cost-effectiveness. \(\text{DSP}\)-based monitoring techniques typically perform asynchronous sampling of electrical signal amplitude and then generate one-dimensional \(\text{(1D)}\) or two-dimensional \(\text{(2D)}\) histograms of the signal samples.
The statistical features of these histograms are then exploited using statistical signal processing, artificial intelligence, and digital image processing techniques for multi-impairment monitoring. Some prominent \(\text{DSP}\)-based
monitoring techniques for direct detection systems include \(\text{AAH}\), \(\text{ADTP}\), \(\text{ATTP}\), and \(\text{ASCS}\) techniques.
Electronic \(\text{DSP}\)-based techniques using \(\text{AAHs}\) are attractive due to their remarkable simplicity and flexibility. The concept of \(\text{AAH}\) is shown in Figure 7. To obtain an \(\text{AAH}\), the electrical signal amplitude is randomly sampled (without any clock information) at a rate much lower than the symbol rate.
It is important to note that the sampling period \(\text{T}_\text{sampling}\) has no relation with the symbol period \(\text{T}_\text{symbol}\). After the acquisition of numerous samples pi, a histogram is formed by first dividing the dynamic range of sample values into different uniformly spaced levels called histogram bins. Next, the samples are sorted out depending upon their values.
The values are then mapped onto the histogram bins and the number of samples falling into each of these bins is counted. Plotting the bin count against the bin value generates an \(\text{AAH}\). The number of samples used for the synthesis of \(\text{AAH}\) must be sufficient enough to obtain the complete statistics of the signal.
The shape of an \(\text{AAH}\) reflects the signal properties. Since the signal is distorted by several optical impairments, the statistical features of an AAH also vary accordingly. The variations in \(\text{AAH}\)’s statistical properties can thus be tracked to evaluate the levels of various impairments degrading the optical signal.
The impairments-sensitive features of \(\text{AAHs}\) have been successfully exploited using statistical signal processing and machine learning techniques for the monitoring of multiple signal quality parameters in fiber-optic networks.
The advantages of \(\text{AAH}\)-based monitoring techniques are that they are cost-effective, have less implementation complexity, and are modulation format independent. Since they do not require clock information, they are also transparent to data rates.
However, their drawback is that the effects of various impairments are often intermixed, thus prohibiting independent monitoring. The monitoring accuracy of \(\text{AAH}\)-based techniques depends on the number of samples acquired for monitoring purpose. Hence, there is a trade-off between accuracy and monitoring speed.
\(\text{ADTS}\)-based monitoring techniques are quite interesting since they offer the potential for data rate and modulation format-independent multi-impairment monitoring. Similar to \(\text{AAH}\), \(\text{ADTS}\)-based techniques also exploit the statistical properties of the asynchronously sampled signal. However, in contrast to \(\text{AAH}\),
which is a \(\text{1D}\) histogram of signal amplitudes, \(\text{ADTS}\) produces a \(\text{2D}\) histogram of closely located sample pairs. The concept of \(\text{ADTS}\) is illustrated in Figure 8. The electrical signal amplitude after direct detection is asynchronously sampled in pairs \((p_i, q_i)\) with a known constant time delay \(\Delta t\) between them called tap-delay. The sampling period Tsampling between the pairs \((p_i, q_i)\) is not related to the symbol period \(\text{T}_\text{symbol}\) and can be many orders of magnitude longer.
Binning the sample pairs \((p_i, q_i)\) into a \(\text{2D}\) histogram generates an \(\text{ADTP}\) or scatter plot as shown in Figure 8. An \(\text{ADTP}\) provides the information richness of an eye-diagram without requiring clock information for its generation. The shape and features of an \(\text{ADTP}\) depend on the modulation format, bit rate, tap-delay \(\Delta t\) as well as various signal distortions such as \(\text{ASE}\) noise, \(\text{CD}\), \(\text{PMD}\), and crosstalk. The unique signatures of various impairments reflected in \(\text{ADTPs}\) can be exploited for \(\text{OPM}\) purpose.
In addition, the \(\text{ADTPs}\) can even distinguish between the sign of accumulated link \(\text{CD}\) and Li et al. detailed how this property can be effectively used for signed CD monitoring. However, it is important to note that an \(\text{ADTP}\) is essentially a graphical representation allowing qualitative estimation of signal properties.
In order to extract the quantitative information from an \(\text{ADTP}\), a further analysis is required. Several approaches such as statistical signal processing, image processing, artificial intelligence, and machine learning have been proposed for the treatment of \(\text{ADTPs}\) for the purpose of multi-impairment monitoring in direct detection systems.
The advantage of \(\text{ADTS}\)-based \(\text{OPM}\) techniques is that they are capable of monitoring multiple impairments simultaneously and independently. Similar to \(\text{AAH}\)-based techniques, they are also transparent to data rates and modulation formats. However, the implementation complexity of \(\text{ADTS}\)-based techniques is higher than that of \(\text{AAH}\).
This is due to the fact that the tap-delay value is a function of symbol period and hence, it needs to be adjusted precisely depending upon the data rate of the signal being monitored. The monitoring accuracy of \(\text{ADTS}\)-based techniques depends on the number of sample pairs acquired for monitoring purpose as well as on the degree of correlation between the transponders used for the acquisition of calibration curves or for the training of artificial intelligence-based classifiers, and the transponders employed during the actual monitoring process.
In addition to \(\text{AAH}\)- and \(\text{ADTS}\)-based techniques, several other electronic DSP-based schemes have been proposed for \(\text{OPM}\) in direct detection optical networks. These include \(\text{ASCS}\), asynchronous two-tap sampling (ATTS), and empirical moments based monitoring techniques.
In \(\text{ASCS}\)-based technique, a \(\text{2D}\) scatter plot equivalent to \(\text{ADTP}\) is produced. However, in contrast to \(\text{ADTS}\) technique, which requires two-channel sampling for the acquisition of delay-tap sample pairs, this technique needs only single-channel sampling, whereby the original and shifted versions of the acquired samples are utilized as sample-pairs for the generation of \(\text{2D}\) scatter plot.
Consequently, the implementation complexity and cost of this technique is expected to be lower as compared with \(\text{ADTS}\)-based techniques. An \(\text{ATTS}\)-based technique is proposed for \(\text{CD}\) monitoring in direct detection systems. This technique estimates the \(\text{CD}\)-induced relative group delay between the two vestigial sideband \(\text{(VSB)}\) signals by computing the differences in the sampled amplitude levels of two \(\text{VSB}\) signals, which are sampled simultaneously but asynchronously.
The technique is shown to monitor \(\text{CD}\) for various modulation formats and data rates without requiring hardware modifications. Unlike \(\text{ADTS}\)-based techniques, no precise adjustment of tap-delay is needed in this case. The use of empirical moments of asynchronously sampled signal amplitudes in conjunction with artificial neural networks \(\text{(ANNs)}\) is proposed for multi-impairment (i.e., \(\text{OSNR}\), signed \(\text{CD}\), and \(\text{PMD})\) monitoring in direct detection systems.
This technique requires simple hardware for the acquisition of signal samples as compared with \(\text{ADTP}\)- and \(\text{ATTS}\)-based techniques. Furthermore, no hardware modifications are needed for the monitoring of several different data rates and modulation formats.
Bit Rate and Modulation Format Identification Techniques for Direct Detection Systems
The existing \(\text{OPM}\) techniques for direct detection fiber-optic networks assume either prior information about the signal’s bit rate and modulation format or the attainment of this knowledge from the upper layer protocols. However, practically it is not feasible to introduce additional cross-layer communication for \(\text{OPM}\) purposes at the intermediate network nodes because these nodes can only handle limited complexity.
Therefore, it is crucial to have the capability of joint bit rate and modulation format identification \(\text{(BR}\)-\(\text{MFI)}\) at the intermediate network nodes since the OPM techniques used at these nodes may be bit rate/modulation format dependent. The critical information about the signal’s bit rate and modulation format can enable the \(\text{OPM}\) devices deployed at the intermediate network nodes to select a monitoring technique most suitable for the identified signal type.
Recently, a few techniques for modulation format identification \(\text{(MFI)}\) as well as joint BR-MFI in direct detection receivers have been proposed. Khan et al. demonstrated a simple and cost-effective \(\text{MFI}\) technique utilizing an \(\text{ANN}\) trained with the features extracted from the \(\text{AAHs}\). This technique is shown to successfully identify six different widely used modulation formats in the presence of various channel impairments.
However, since an \(\text{AAH}\) lacks the timing/slope information necessary for distinguishing between different bit rates of the signals, this techniqueis limited to just MFI. A modification of this technique is presented in, whereby joint \(\text{BR}\)-\(\text{MFI}\) is demonstrated by using an \(\text{ANN}\) trained with the features extracted from \(\text{ADTPs}\).
Since \(\text{ADTPs}\) contain information about the slopes of the signal pulses (which in turn depend on the bit rate) as well as unique signatures of different modulation formats, this technique is capable of simultaneous identification of bit rates and modulation formats of the signals.
Tan et al. proposed a simple technique for joint \(\text{BR}\)-\(\text{MFI}\) as well as multi-impairment monitoring by using principal component analysis \(\text{(PCA)}\)-based pattern recognition on \(\text{ADTPs}\). This technique can successfully identify the bit rate and modulation format of the signal from a known set of bit rates and modulation formats.
In addition, it can enable simultaneous and independent monitoring of \(\text{OSNR}\), \(\text{CD}\), and \(\text{DGD}\) without necessitating information about the signal type during the online monitoring process.
Commercially Available \(\text{OPM}\) Devices for Direct Detection Systems
\(\text{OPM}\) is regarded as a key enabling technology for high-speed self-managed optical networks. However, currently, the commercially available \(\text{OPM}\) devices are nothing more than a simplified version of optical spectrum analyzers.
These devices are capable of monitoring a few parameters such as wavelengths of \(\text{WDM}\) channels, total optical power, and out-of-band \(\text{OSNR}\), and are normally referred to as optical channel monitors \(\text{(OCM)}\). These devices typically employ a tunable band-pass filter or a diffraction grating combined with a single detector to monitor the above-mentioned parameters.
The information about the power level of each \(\text{WDM}\) channel is typically used to balance the optical amplifier gain across the spectrum. Similarly, the wavelength measurement provides information about whether the optical signal is properly placed in its channel.
The limited information provided by the current \(\text{OPM}\) devices can only facilitate the detection of sudden faults and thus enable system alarms and error warnings for lost or out-of-specification optical channels. These devices are far from achieving the real objectives of \(\text{OPM}\), that is, to enable error root cause analysis and to provide early failure identification so that the operators can initiate fast error cancellation.
Therefore, to fulfill the ultimate potential of \(\text{OPM}\), the commercial \(\text{OPM}\) devices need to undergo significant improvement in order to be able to constantly monitor the signal dynamics, observe system functionality, detect performance changes, and provide feedback information to the network elements for the optimization of operational performance of the optical networks.
Applications of OPM in Deployed Fiber-Optic Networks
Recently, there have been a few attempts to realize the OPM-enabled functionalities in practical fiber-optic networks. Morgan et al. reported the in-service measurement of \(\text{OSNR}\), \(\text{CD}\), and \(\text{PMD}\) on a live 140 km, 10 Gbps WDM optical link between Bromsgrove and Shrewsbury in Western England without interrupting the actual data traffic. An \(\text{ADTS}\)-based technique is used for the simultaneous monitoring of above-mentioned parameters in a field trial. The real-time information providedby the \(\text{OPM}\) module is then used to effectively diagnose the underlying system issues.
For example, it is revealed that the root cause of the higher than expected pre-\(\text{FEC}\) \(\text{BER}\) and reduced system margin of the monitored link is the high amount of residual \(\text{CD}\) resulting from the use of an incorrect dispersion compensation module. Thus, an in-service diagnosis of underlying network problems, as demonstrated in this work, can facilitate the early identification of root causes and permit the network operators to resolve the issues.
3. OPM FOR COHERENT DETECTION SYSTEMS
The explosive growth in data traffic and demand for higher bandwidth has led to the evolution of employing advanced optical modulation formats with coherent receiver and DSP in fiber-optic communication systems. Since it is possible to retrieve all the information from an optical signal in the electrical domain, coherent detection techniques allow the use of advanced higher-order modulation formats such asm-PSK and \(m\)-\(\text{QAM}\) along with \(\text{PDM}\), thereby enabling higher spectral efficiencies. On the contrary, since higher-order modulation signal with high baud rate are more sensitive to the channel impairments and noise, the link margin may be reduced.
The requirements and parameters to be monitored for coherent systems are quite different from noncoherent systems. With the help of \(\text{DSP}\), linear channel impairments such as \(\text{CD}\) and \(\text{PMD}\) can be fully compensated. Meanwhile, it also enables a promising and comprehensive built-in \(\text{OPM}\) at the receiver for free.
Assume that the optical channel is linear with the channel transfer matrix \(\text{H(f)}\), the equalization filter \(W(f)\) obtained through the zero-forcing \(\text{(ZF)}\) solution or minimum mean square estimation \(\text{(MMSE)}\) solution is the inverse impulse response of the channel that can be expressed as
\[\tag{1}w(f)=H^{-1}(f)=D^{-1}(f)\prod^1_{i=N,-1}U^{-1}_i(f)E^{-1}_i\]
where \(H(f)\) is composed of the transfer function \(D(f)\) and concatenated elements \(E_i\) and \(U_i(f)\) accounting for \(\text{PDL}\) and higher-order \(\text{PMD}\).
After some algebraic manipulations, the following equations can be obtained:
\[\tag{2}\text{arg}(\hat{H^{-1}_{CD}})=\text{arg}(\sqrt{det(W(f))})=-f^2\varphi\]
\[\tag{3}\qquad\qquad\quad\hat{H}_\text{PDL}(f)=\left|\sqrt{\text{det}(W(f))}\right|=|H_\text{AF}(f)|^{-1}\prod^N_{i=1}(k_i)^{-1/2}\]
\[\tag{4}\qquad\qquad\qquad W_{UE}(f)=\frac{W(f)}{\sqrt{\text{det(W(f))}}}=\prod\left(\begin{array}&u^*_i-v_i\\v_i^*\quad u_i\end{array}\right)\left(\begin{array}&k_i^{1/2}\;0\\0\;k_i^{-1/2}\end{array}\right)\]
which are responsible for residual \(\text{CD}\), \(\text{PDL}\), and \(\text{PMD}\). \(\text{W_{UE}(f)\) is the normalized \(W(f)\) by the square root of its determinant. \(u_i\) and \(v_i\) form the \(\text{PMD}\) matrices and \(k_i\) is the attenuation factor accounting for \(\text{PDL}\).
Since those linear impairments can be estimated and compensated, they no longer limit system performance. On the contrary, as \(\text{ASE}\) noise cannot be compensated, the system performance is largely determined by the \(\text{OSNR}\) of received signals. Accurate, reliable and low-cost in-band OSNR monitoring is still highly desired for coherent systems.
Although \(\text{OSNR}\) monitoring is not as easy as reading off filter taps, \(\text{ASE}\)-noise-induced distortions can be separated from all the other linear transmission impairments in a digital coherent receiver and reliable \(\text{OSNR}\) can still be estimated with further processing of the received signals. There are two types of approaches to \(\text{OSNR}\) monitoring: non-data-aided estimation and data-aided estimation that is based on the property of training sequences \(\text{(TS)}\). Data-aided estimation gives high estimation accuracy and faster estimation speed with the trade-off of reduced bandwidth efficiency compared with non-data-aided estimation.
Non-Data-Aided OSNR Monitoring for Digital Coherent Receivers
One simple approach for non-data-aided \(\text{OSNR}\) monitoring utilizes the statistical moments of the equalized signal, the signal used to estimate the \(\text{OSNR}\) is taken from just after the adaptive equalization, before the carrier phase recovery stage (as shown in Figure 9(a)). The adaptive equalization can either utilize “blind” non-data-aided channel acquisition by gradient algorithms such as constant-modulus algorithm \(\text{(CMA)}\) or data-aided channel estimation based on a periodically transmitted training sequence.
After adaptive equalization, the linear distortions such as \(\text{CD}\) and \(\text{PMD}\) ideally can be fully compensated, and thus the variations of the equalized signal envelope are mainly caused by the \(\text{ASE}\) noise. The envelope of the output signal from the adaptive filter (as shown in Figure 9(b))
\[\tag{5}\text y_n\approx\sqrt Ca_n^{ej\theta_n}+\sqrt N\text w_n\]
where \(a_n\) is the \(m\)-\(\text{PSK}\) or \(m\)-\(\text{QAM}\) symbol amplitude, \(C\) is the signal-power scale factor, \(N\) is the noise power scale factor, \(w_n\) is the \(\text{ASE}\) noise, \(\theta_n\) is the phase noise stemming from phase fluctuations of a transmitter laser and a local oscillator, and \(n\) is the number of samples.
In a practical system, the second- and fourth-order moments from a received data block of \(L\) symbols can be calculated as
\[\tag{6}\mu_2\approx\frac{1}{L}\sum^{L-1}_{n=0}|\text y_n|^2\]
\[\tag{7}\mu_4\approx\frac{1}{L}\sum^{L-1}_{n=0}|\text y_n|^4\]
respectively. And the carrier-to-noise ratio \(\text{(CNR)}\) for \(\text{QPSK}\) is expressed as
\[\tag{8}\text{CNR}_\text{QPSK}=\frac{\sqrt{2\mu^2_2-\mu_4}}{\mu_2-\sqrt{2\mu^2_2-\mu_4}}\]
and \(\text{CNR}\) for \(\text{QPSK}\) is expressed as
\[\tag{9}\text{CNR}_{16-\text{QAM}}=\frac{\sqrt{2\mu^2_2-\mu_4}}{\mu_2\sqrt{0.68}-\sqrt{2\mu^2_2-\mu_4}}\]
When the launched power is so low that fiber nonlinear effects can be neglected, the \(\text{OSNR}\) value in \(dB\) can be estimated from the \(\text{CNR}\) value as
\[\tag{10}\text{OSNR}_{dB}=101\text{og}_{10}\text{CNR}+101\text{og}_{10}\left(\frac{R_s}{B_r}\right)\]
where \(R_s\) is the symbol rate and \(\text{R}_s∕\text{B}_r\) is a scaling factor adjusting the measured noise bandwidth to the reference bandwidth \(B_r\). The bandwidth \(B_r\) is usually set to 12.5 GHz, which is equivalent to the 0.1-nm \(\text{OSA}\) resolution bandwidth. As shown in Equations 8 and 9, measuring second- and fourth-order moments does not include any effect of the phase noise and thus the proposed scheme operates phase insensitively.
Another approach is to utilize the error vector magnitude \(\text{(EVM)}\) of fully equalized signals as an \(\text{OSNR}\) estimator. In this approach, the signal is taken after decoding and symbol decision stage where both the linear impairments such as \(\text{CD}\) and \(\text{PMD}\) and carrier-phase are recovered. The distributions of fully equalized \(\text{QPSK}\) signal are shown in Figure 10.
In this case, the \(kth\) received symbol in one particular polarization can be represented as
\[\tag{11}r_k=s_k+n_k\]
where \(s_k\) is the transmitted \(\text{QPSK}\) symbol and \(n_k\) models the collective \(\text{ASE}\) noise generated by inline optical amplifiers, which is a band-limited complex circularly
symmetric zero-mean Gaussian random process with covariance matrix \(\sigma^2I\). The \(\text{OSNR}\) can be estimated through an EVM-based approach
\[\tag{12}\text{OSNR}_\text{Estimated}=\frac{P_\text{in}}{P_\text{ASE}}=\frac{E(|\hat{s_k}|^2)}{E(|n_k|^2)}\]
where \(P_\text{in}\) is the signal power, \(P_\text{ASE}\) accounts for the \(\text{ASE}\) noise power, and \(\hat{S}_k\) is the symbol after decoding and symbol decision stage and \(E(\cdot)\) denotes expectation. However, since the signals used for estimation are taken after carrier-phase recovery, the accuracy of the estimation may be affected by the frequency offset and phase noise.
Non-data-aided \(\text{OSNR}\) estimation is not modulation format independent and suffers from a relatively longer acquisition time, preventing the receiver from fast switching protection.
Data-Aided (Pilot Symbols Based) OSNR Monitoring for Digital Coherent Receivers
Data-aided \(\text{OSNR}\) estimation has the advantage of being independent of modulation format as it only utilizes the properties of the equalized \(\text{TSs}\), and the modulation format can be altered arbitrarily in between the fixed training patterns. Furthermore, channel equalization based on the periodically transmitted \(\text{TSs}\) allows instantaneous filter acquisition and thus enables faster \(\text{OPM}\).
\(\text{OSNR}\) can then be estimated through the \(\text{SNR}\) from the training sequences. Figure 11 shows a constellation plot of the equalized training sequences for \(\text{QPSK}\) system before and after channel filtering using Golay sequences with \(\text{OSNR}=22\text{dB}\).
In case of no channel filtering, small variations in the equalized training sequences are due to misalignment between the overlap-cut equalizer and the training blocks, while it can be seen that, with estimation filtering, the noise information is reserved and can be used for \(\text{SNR}\) estimation.
After equalization of the training sequences using the estimated channel information with filtering, the added noise can then be subtracted from the signal and the \(\text{SNR}\) is measured as
\[\tag{13}\text{SNR}=\sum^N_{k=1}\frac{\text s[k]^2}{\text w[k]^2}\]
where \(\text{s[k]}\) is the original sequence and \(\text{w[k]}\) is the added noise sample. Finally, the \(\text{OSNR}\) is estimated by first measuring the electrical \(\text{RF}\) noise SNRRF at a reference \(\text{OSNR}\) point. This is done by measuring the reference point in a back-to-back configuration and without any added \(\text{ASE}\) and then the \(\text{OSNR}\) is calculated as
\[\tag{14}\text{OSNR}_\text{dB}=101\text{og}_{10}\left(\frac{B_\text{ref}}{R_s}\right)-101\text{og}_{10}(\text{SNR}^{-1}-\text{SNR}^{-1}_{RF})\]
where \(\text{B}_\text{ref}\) and \(\text{R}_{s}\) are the reference bandwidth and baud rate, respectively, and \(\text{SNR}_\text{RF}\) is the measured system \(\text{SNR}\) without any added \(\text{ASE}\) noise.
OPM at the Intermediate Network Nodes Using Low-Cost Structures
As discussed in the previous section, coherent receiver with \(\text{DSP}\) enables a comprehensive built-in \(\text{OPM}\) at the receiver end for free. However, \(\text{OPM}\) devices are supposed to be deployed ubiquitously across the network including intermediate nodes and it is simply too costly and impractical to use full digital coherent receivers with symbol-rate bandwidth for that purpose. A low-cost monitoring solution utilizing reduced-complexity and low speed hardware for distributed monitoring of optical network is in demand.
An in-band \(\text{OSNR}\) estimation technique with data-aided \(\text{DSP}\) utilizing low-bandwidth coherent receivers and low sampling rate is proposed and demonstrated. It is known that Golay sequences are a pair of complimentary sequences \(\text{S}_1[k]\) and \(\text{S}_2[k]\) that satisfy the power spectrum property:
\[\tag{15}G[k]=|S_1[k]|^2+|S_2[k]|^2=L\]
where \(S_1[k]\) and \(S_2[k]\) are the discrete Fourier transform of the original \(S_1\) and \(S_2\) sequence, respectively, and L is a constant related to the length of each sequence. Neglecting the effect of linear impairments, considering Gaussian distributed noise with zero mean and after some mathematical manipulation and simplification, it is shown that the variance of Golay power spectrum of the received \(\text{TS}_s\) \(G[k]\) is proportional to the expected value of noise power spectral density and thus is related to the system \(\text{SNR}\) as shown in Figure 12.
Since the proposed technique utilizes spectral property in frequency domain, \(\text{SNR}\) can be estimated by using a low sampling speed and low-bandwidth receiver. Experimental verification was demonstrated to monitoring the \(\text{OSNR}\) of 10 Gbaud \(\text{PDM}\)-\(\text{QPSK}\) and \(\text{PDM}\)-\(\text{16QAM}\) signals utilizing a low-bandwidth receiver with
800-MHz filter working at 2.5 GHz sampling rate. A wide \(\text{OSNR}\) monitoring range with accuracy within 1 \(\text{dB}\) for up to 1000-km transmission is achieved.
Another three cost-effective \(\text{OPM}\) techniques that can be deployed at network intermediate nodes for coherent systems are proposed. An \(\text{ANN}\) is used in combination with \(\text{RF}\) spectrum of a directly detected \(\text{PDM}\)-return-to-zero \(\text{(RZ)}\)-\(\text{QPSK}\) signal for \(\text{OSNR}\) monitoring in the presence of large \(\text{CD}\).
This is motivated by the fact that new coherent transmissions links will not have inline dispersion compensation and hence OSNR monitoring in the presence of a wide range of unknown \(\text{CD}\) become a new and unprecedented challenge in \(\text{OPM}\) research. The input to the \(\text{ANN}\) is part of the \(\text{RF}\) spectrum power which can be measured in practice by simple power meters and only direct detection is used as shown in Figure 13. Therefore, this proposed technique is low-cost and may serve as a key step toward practical realization of ubiquitous \(\text{OSNR}\) monitors for coherent links.
A delay-line interferometer \(\text{(DLI)}\)-based \(\text{OSNR}\) monitor is proposed. Since the signal is coherent and experiences constructive and destructive interference in the \(\text{DLI}\) whereas in-band noise is noncoherent and insensitive to constructive and destructive interference, the power splitting ratio of signal and noise between the constructive and the destructive ports of the \(\text{DLI}\) are different.
Therefore, the \(\text{OSNR}\) can be estimated from the power ratios of the two ports. The authors propose an \(\text{OSNR}\) monitoring approach based on uncorrelated signal generated by optical bandpass filtering and balanced subtraction after photodetection. The power ratio of the uncorrelated signal and original signal can be a measure of \(\text{OSNR}\). The method is experimentally verified in 100-\(\text{Gb}\)/\(\text{s}\) \(\text{PDM}\)-\(\text{QPSK}\) and 50-\(\text{Gb}\)/\(\text{s}\) \(\text{QPSK}\) systems in the presence of \(\text{CD}\) and \(\text{PMD}\) effects.
OSNR Monitoring in the Presence of Fiber Nonlinearity
Most of the currently deployed long-haul optical communication systems operate in the weakly nonlinear regime, which is a trade-off between mitigating the effect
of \(\text{ASE}\) noise and fiber nonlinearities. Nonlinear distortions are typically treated as noise and are indistinguishable from amplifier noise (as shown in Figure 14) by the standard \(\text{DSP}\) platform since fiber nonlinearity compensation algorithms such as digital backpropagation is too complex to be realized at present. Therefore, those above-mentioned \(\text{OSNR}\) estimation techniques based on the \(\text{SNR}\) of equalized signal or \(\text{TSs}\) will considerably underestimate the \(\text{OSNR}\) for long-haul transmission systems.
The fiber nonlinearity-induced amplitude noise correlation among neighboring symbols is characterized as a quantitative measure of nonlinear distortions which is shown to depend only on signal-launched power but not \(\text{OSNR}\) and hence fiber nonlinear distortions can be isolated from \(\text{ASE}\) noise. In this case, nonlinearity-insensitive \(\text{OSNR}\) monitoring is achieved by incorporating/calibrating such amplitude noise correlations into an \(\text{EVM}\)-based \(\text{OSNR}\) estimator.
Since nonlinear distortions can be modeled as complex circularly symmetric additive Gaussian noise with zero-mean for long-haul coherent transmission links without in-line dispersion compensation. Equation 11. can be rewritten as
\[\tag{16}r_k=s_k+n'_k=s_k+n_k+v_k\]
where \(n^′_k=n_k+v_k\) consists of \(\text{ASE}\) noise \(n_k\) and nonlinearity-induced distortions \(v_k\).
With the \(\text{EVM}\) methodology, \(v_k\) become addition distortions and thus if we naively use the \(\text{EVM}\) method by simply measuring the “size” of the “clouds” in the received signal distributions, the \(\text{OSNR}\) estimates in Equation 12. can be rewritten as
which can significantly underestimate the true \(\text{OSNR}\).
The interaction of fiber nonlinearity, \(\text{CD}\) and \(\text{ASE}\) noise will produce distortions such as \(\text{IFWM}\) that are shown to be correlated across neighboring symbols even after appropriate linear impairment compensation. Denoting \(\Delta k\) as the amplitude noise of the\(\Delta k\)received symbol, let the autocorrelation function \(\text{(ACF)}\) of amplitude noise across neighboring symbols be
\[\tag{18}R_\Delta(m)=E[\Delta_k\Delta_{k+m}]\]
It is shown in that the amplitude noise is correlated across neighboring symbols and \(|R_\Delta(1)|\) can be used by multiplied by a calibration factor\(\xi\) as a measure/estimate of the amount of nonlinear distortions \(\text{P}_{NL}\) in the received signal \(r_k\). The calibration factor \(\xi\) only depends on the transmission distance \(L\). Incorporate the term \(|R_\Delta(1)|\times\xi\) in the \(\text{OSNR}\) estimator in Equation 17 and thus a nonlinearity-insensitive \(\text{OSNR}\) estimation can be obtained by
\[\tag{19}\text{OSNR}_\text{Estimated}\frac{E(|\hat{s}_k|^2)}{E(|n'_k|^2)-|R_{\Delta}(1)|\times\xi}\]
Typical \(\text{OSNR}\) monitoring results before and after calibration are shown in Figure 15. When \(|R_{\Delta}(1)|\times\xi\) is not incorporated, the \(\text{OSNR}\) is significantly
FIGURE 15. Monitoring \(\text{OSNR}\) versus reference \(\text{OSNR}\) experimentally obtained from a 112 Gb/s \(\text{PDM}\)-\(\text{QPSK}\) system for various signal launched powers and \(\text{OSNR}\) values after 800 km transmission.
underestimated as the nonlinear distortions are treated as \(\text{ASE}\) noise in the OSNR estimates and the monitoring error generally increases with input power due to enhanced nonlinearity effects. With the calibration based on \(|R_{\Delta}(1)|\times\xi\), the \(\text{OSNR}\) monitoring error is largely reduced.
4.INTEGRATING OPM FUNCTIONALITIES IN NETWORKING
One of the major application and motivation for \(\text{OPM}\) is to obtain real-time detailed conditions of the network to realize impairment-aware routing and improve overall network efficiency. In this regard, it is vital that information from OPM devices be integrated in network management.
To this end, Lai et al.demonstrated the use of \(\text{OSNR}\) monitoring for efficient network routing and enabling packet protection for critical data flows. In their work, \(\text{OSNR}\) information obtained thorough a \(\text{DLI}\)-based OPM module is used to send feedback signals to the higher layers for effective packet rerouting and protection.
A low \(\text{OSNR}\) value detected by the \(\text{OPM}\) device indicates degraded signal quality and vice versa. Depending upon the measured \(\text{OSNR}\) and the packet-encoded priority, the optical packets are either discarded and rerouted on the alternate path, or forwarded to the final destination port.
This approach helps to reduce the penalties incurred due to re-transmission of critical, high-priority data packets. The \(\text{DLI}\) approach is also extended to \(\text{WDM}\) system setups with different signal modulation formats.
5.CONCLUSIONS AND OUTLOOK
\(\text{OPM}\) continues to be an integral part of optical network operation and imperative for their evolution toward higher speed and improved reliability. As we move towarddigital coherent transmissions and beyond, more tools are at our disposal for \(\text{OPM}\) and their underlying principles inherently merge with other well-researched disciplines such as channel estimation in traditional copper-wire/wireless communications.
In addition, the emerging dominance of data centers/cloud computing driven traffic has fueled the need for \(\text{OPM}\) to not only manage network faults, but also provide information on real-time network conditions for implement impairment-aware routing and software-defined networking \(\text{(SDN)}\).
\(\text{OPM}\) and related optical network functionalities are expected to play an increasing role in shaping next generation optical networks.
