Open Access
Add to BibSonomyAbstract
A new class of digital signal processing (DSP)-based fiber-longitudinal optical power profile monitor has recently been proposed and demonstrated toward optical network tomography that captures the whole physical status of an optical network, including in-span and wavelength-specific power profiles over a multi-span transmission light path. In this invited paper, we review the monitor that disentangles signal waveforms received by a standard digital coherent receiver to a distance-wise power profile over a multi-span transmission link and discuss its implementation aspect, including the advantages and limitations of its cloud/edge implementation, the dependency of the number representation in its algorithm, and a feasibility study on field-programmable gate array implementation.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. INTRODUCTION
The community of fiber-optic communication has developed new technologies such as advanced modulation formats, forward error correction, and various digital equalizers to accommodate growing traffic demands. Although these newly developed technologies enable us to approach the theoretical limit of capacity in fiber-optic communications, they have also introduced many parameters to be optimized for maximizing system performance.
The optimization of multiple parameters is a crucial issue in terms of the operating expenditure for such sophisticated systems. The concept of an autonomous optical network has been proposed and investigated to solve this issue [1–5]. Figure 1 shows a schematic of the autonomous optical network. The idea behind this concept is to utilize a closed loop containing a perception for a physical state, a network controller, and a programmable network infrastructure.
The perception, also known as the optical monitoring and analysis part, is an important part that recognizes the current status of optics in networks. Although various monitoring techniques using specially designed optical components have been investigated, one of the remaining issues in the approach is how to deploy and manage the enormous number of points ubiquitously spread over nationwide optical networks in a cost-effective manner. Recently, another approach that uses already-deployed digital coherent receivers in networks has been proposed and investigated to solve this issue. One of the most straightforward implementations of the digital-coherent-based approach is to use intermediate digital signal processing (DSP) parameters of the main signal demodulation on a digital coherent receiver, e.g., tap coefficients of an adaptive equalizer (AEQ) [6]. Emerging machine learning and deep learning techniques have recently advanced this digital-coherent-based approach in terms of the availability of various, unstructured input and automatic learning without domain knowledge [7–13]. This digital-coherent-based monitoring can estimate various physical statuses in optical networks, such as the modulation formats of received signals, optical signal-to-noise ratio (OSNR), chromatic dispersion (CD), polarization mode dispersion (PMD), and fiber nonlinearity, without additional optical components.
Although digital-coherent-based monitoring is indispensable in constructing future autonomous optical networks, existing digital-coherent-based monitors placed at the link-end can only estimate cumulative quantities through the entire fiber transmission line, e.g., cumulative CD, cumulative PMD, and cumulative OSNR. Thus, it should be useful if a receiver could determine not only the cumulative amount of parameters but also the longitudinal profile of such parameters in a distance-resolved manner. Such functionality helps the network owner efficiently identify where physical layer anomalies are generated and what type of anomalies they are.
Combining multiple light paths and intelligent measurement procedures, and localizing anomalies in span-by-span resolution monitoring have been demonstrated [14,15]. However, even with multiple optical monitors and light paths, a much more refined localization through in-span resolution monitoring remains an issue.
Recently, we proposed and experimentally demonstrated a new class of DSP-based fiber-longitudinal monitor in multi-span fiber-optic transmission links for optical network tomography [16–18]. This monitor delivers distance-wise optical power throughout the entire multi-span link by utilizing distributed nonlinearity over a fiber transmission link. Unlike existing fiber-longitudinal monitors, the proposed monitor does not need any additional optical component and transmitter-side information. This monitor works with only receiver-side DSP and signal waveform obtained by a standard coherent receiver placed at the link end. More recently, a similar type was proposed to extend this idea to extract passband narrowing responses with finite impulse response filters [19].
In this work, in addition to the reviews on the principle and experimental validation of the fiber-longitudinal optical power profile monitoring for fiber-optic transmission lines demonstrated in our previous papers [16–18], this study aims to add the following new contributions.
First, we present a discussion on the comparison of advantages and disadvantages of where this fiber-longitudinal monitor should be deployed in the overall optical fiber network. In particular, we discuss whether the monitor should be deployed on an edge side close to the optical transceiver or on a cloud side close to the upper layers of the network.
Second, we present an initial investigation for a realistic implementation of the monitor. In particular, we show examples and discuss how a numerical accuracy-limited implementation of the monitor algorithm affects the monitoring results. In addition, we discuss and present an initial evaluation of the computation time and computational complexity of the field-programmable gate array (FPGA) implementation of the monitor method. This investigation is beneficial especially for an edge-side implementation of the fiber-longitudinal monitor, which can provide real-time anomaly surveillance over the entire transmission link without concerns about security protection and increasing supervisory channel capacity but suffers from the limitation of computational resources.
Finally, we expand a description of the concept of one promising application enabled by the fiber-longitudinal monitor, which was pioneered in a previous work [17]. The application provides network-level power profile monitoring, including wavelength dependence, which we call optical network tomography.
In Section 2, we review the concept of the DSP-based fiber-longitudinal monitor proposed in [16]. In Section 3, we review details of the DSP-based fiber-longitudinal monitor algorithm. In Section 4, we discuss two implementation options of the monitor, i.e., edge- and cloud-based implementations, and highlight their advantages and limitations. In Section 5, we describe the experimental setup. In Section 6, we show the results and accompanying discussion. Finally, after the application discussion, including optical network tomography, in Section 8, we conclude the paper with a brief summary.
2. DSP-BASED FIBER-LONGITUDINAL MONITOR
This section discusses the concept of a DSP-based fiber-longitudinal monitor that captures the light-path-level physical characteristics, including the power profile over the light path with a certain wavelength slot.
In this section, we focus on a single light path of the optical network. An overview of the monitoring method is shown in Fig. 2. This method uses a standard digital coherent receiver to measure a received optical waveform after fiber transmission. The measured waveform contains the time-sampled and digitized amplitude and phase of the horizontal and vertical polarizations of incoming optical signals. The key feature of the monitor is to estimate an optical power profile over a multi-span fiber link using received information only. Note that the monitoring result is a power profile on the certain wavelength signal received by the digital coherent receiver with a wavelength resolution relevant to the channel grid, such as 12.5 or 50 GHz. A detailed algorithm to estimate a power profile is discussed in Section 3.
3. ALGORITHM OF THE DSP-BASED FIBER-LONGITUDINAL MONITOR
This section reviews and describes the DSP-based fiber-longitudinal monitor for optical network tomography based on our previous works [16,17]. The main idea behind the DSP-based fiber-longitudinal monitor is to leverage a fiber communication channel property: distributed nonlinearity due to the Kerr effect, in the digital domain. Summarily, the monitor aims to create a virtual transmission line model in the digital domain to mimic the real world’s transmission line. If we obtain a transmission line model as a copy of a real transmission line, then we can easily extract the fiber-longitudinal properties from the model.
In our early-stage study in [20], we used a digital backpropagation-based nonlinear compensator [21,22], which is also known as the split-step Fourier method, as the model in the digital domain. The nonlinear channel model in the digital domain is optimized to fit the real world’s actual fiber link and to minimize the received waveform error. The early-stage method’s issue is on the computational complexity for the optimization, which leads to an $N$-dimensional optimization problem with the link comprising ${N_{\rm section}}$. This problem is usually challenging to solve when ${N_{\rm section}}$ is large without prior information, such as well-chosen initial values. This issue is particularly critical for an edge-side implementation that has limited computational resources. Recently, an advanced monitoring method similar to [20], which uses a nonlinear transmission line model based on a backpropagation nonlinear compensator, has been proposed and demonstrated [23]. This method mitigates the optimization problem by the sophisticated optimization technique widely used for neural networks.
We recently solved this issue by avoiding the optimization process to extract a fiber-longitudinal power profile from received waveforms. The fiber-longitudinal monitor in [16] used a simplified backpropagation and a fixed formula to estimate the value of the optical power at a point corresponding to a certain distance. It is a deterministic approach without adaptive and iterative optimization. Unlike straightforward solutions for multidimensional optimization, the deterministic method can experience possible benefits, including less required computational complexity and/or easier parallel processing. Although the method in [16] does not divide the transmission line into many sections, such as in [20,23], and avoids the multidimensional optimization problem, this method still inherits the concept of using fiber nonlinearity in the digital domain.
Figure 3 shows details of the data processing part [16–18] introduced in Fig. 2. The processing has two tributaries: The first tributary of signal processing calculates the waveform replica of the transmitter side, which is the waveform obtained before propagating the transmission link, using only a received waveform. The second tributary provides a pointwise nonlinear equalized waveform calculated from the received one.
Fig. 3. Schematic of the DSP-based fiber-longitudinal power profile monitor. ADC, analog-to-digital converter; DSP, digital signal processing; CDC, chromatic dispersion compensator; AEQ, adaptive equalizer; FOC, frequency offset compensator; CPR, carrier phase recovery; FEC, forward error correction; Abs, absolute value of the complex number.
As a preprocessing step to obtain the transmitted waveform replica on the first tributary, the standard demodulation DSP, which includes a fixed CD compensation, AEQ, frequency offset compensation (FOC), and carrier phase recovery (CPR), and forward error correction (FEC) process the received waveform. In particular, DSP of fixed CD compensation, AEQ, FOC, and CPR equalize bulk CD through fiber propagation, time-variation waveform distortion (including polarization changing and PMD), frequency mismatching between the signal and local oscillator (LO) laser, and carrier phase difference between the signal and LO laser on the received waveform, respectively. After that, FEC corrects bit errors on the equalized waveform. Note that an error-free bit sequence can be obtained if the pre-FEC bit error ratio is lower than the FEC limit of the current FEC algorithm.
The demodulation DSP can be implemented in a blind manner. In fact, the demodulation DSP used in this paper was performed in a blind manner. Although some types of demodulation DSP implementation, such as AEQ, FOC, and CPR, require a proprietary training sequence that is inserted into the transmitted signal, the training sequence has already been shared between the transmitter and receiver when the communication connection is established.
Using the same DSP as the transmitter-side DSP (including mapping and Nyquist shaping) at the receiver side, we can calculate the transmitted waveform replica from the received error-free bit sequence. The resulting waveform from the first tributary is utilized as the reference waveform. Notably, there is no need to use any extra sequence for the monitoring.
On the second tributary, simplified backpropagation containing only one nonlinear phase shift is carried out to estimate point-wise optical power while avoiding the multidimensional optimization problem in [20,23]. First, fixed CD compensation and AEQ are also performed to align the received signals’ polarization state. This part’s equalized CD, which is a cumulative CD of ${c_{\rm{total}}}$ is digitally added again after the polarization demultiplexing. ${c_{\rm{total}}}$ is a summation of ${c_{\rm trans.}}$ and ${c_{\rm predist.}}$, corresponding to the entire transmission link and predistortion at the transmitter. After polarization demultiplexing, the process is independently performed for each horizontal and vertical polarization waveform. To estimate an optical power at a distance of $x = ({{c_{\rm{total}}} - {c_{\rm predist.}} - {c_{\rm{pre}}}})/D$ on the transmission link, the following procedure is performed. Note that ${c_{\rm{pre}}}$ is a parameter of the algorithm, and $D$ is the dispersion of the fiber per unit length.
1. Pre-CD Compensation: The CD of ${c_{\rm{pre}}}$ on the waveform, which is equivalent to the linear fiber backpropagation of a distance of ${c_{\rm{pre}}}/D$ from the receiver-end, is digitally compensated.
2. Nonlinear Probing: A point-wise nonlinear phase shift for the output signal of (1) is imposed. The point-wise nonlinear phase shift is DSP given as ${u_{\rm{out}}}(t) = {u_{\rm{in}}}(t)\exp(- jp|{u_{\rm{in}}}(t){|^2})$, where $p$ is a parameter for the nonlinear-phase probe, and ${u_{\rm{in}}}$ and ${u_{\rm{out}}}$ are the input and output complex-valued waveforms, respectively. The parameter $p$ that governs the point-wise self-phase modulation (SPM) mitigation is a small fixed constant, not a variable to be optimized.
3. Post-CD Compensation: The CD of ${c_{\rm{post}}} = {c_{\rm{total}}} - {c_{\rm{pre}}}$ on the signal after (2) is compensated. The resulting waveform is perfectly CD compensated and partially SPM equalized.
The results of the first and second tributaries were combined to calculate the correlation between their absolute values. We adopted the absolute value of the resulting complex-valued waveform to avoid the phase-slip issue in the CPR DSP algorithm. The resulting correlation at distance $x$ is
where ${{{u}}_{\rm{first}}}(t)$ and ${{{u}}_{\rm{second}}}(t)$ are the output waveforms from the first and second tributaries, respectively. The correlation ${{c}}(x)$ is an indicator of the point-wise nonlinear power, i.e., SPM, at distance $x$, which consists of an instant optical power and the nonlinear refractive index of the fiber at that point. Assuming that the nonlinear refractive index along the transmission line is constant, the correlation represents an instant optical power at distance $x$. To obtain an optical power profile over the entire link, parameter ${c_{\rm{pre}}}$ was swept with a step size of $\Delta c$ to cover the whole transmission link.One of the useful properties of this monitoring method is that the monitor does not require a dedicated extra monitor channel, such as a probe light channel for optical time-domain reflectometry (OTDR). The optical signal for data transmission, e.g., a 400 Gbit/s dual-polarization 16-quadrature amplitude modulation (DP-16QAM) optical signal, can also be used for monitoring the multi-span link through which this signal has passed.
The assumptions required for the power profile monitoring in this method are as follows: First, the monitor assumes that there is no chromatic dispersion compensation module over the transmission link. Thus, the monitor assumes that the given link is a dispersion uncompensated link. Although enhancing the method to include a dispersion compensated link is left for future work, this assumption is considered to be acceptable for most modern terrestrial fiber-optic systems. This is because a modern digital coherent transceiver simply does not require any dispersion compensation modules for its deployment.
Next, as discussed in detail in Ref. [17], the positional accuracy of the output of this monitor degrades as the signal baud rate decreases. This case may be attributed to the low baud rate signals resulting in relatively small waveform changes through the transmission with certain fiber lengths. Recently, the typical baud rate of optical signals for fiber-optic transmission has been increasing. Thus this issue should not become a major limitation for the use of the monitoring method in modern state-of-the-art and future transmission systems.
Finally, we assume that the chromatic dispersion per unit fiber length and the nonlinear refractive index of the optical fiber are known. The chromatic dispersion per unit fiber length is used to convert the cumulative chromatic dispersion into the transmission distance, and the nonlinear refractive index is used to convert the monitor output from the nonlinear equivalent to the optical power equivalent.
4. IMPLEMENTATION OPTIONS: EDGE VERSUS CLOUD
In this section, we discuss two implementation scenarios of the DSP-based fiber-longitudinal monitor and discuss their advantages and limitations.
Figure 4(a) shows a block diagram of the first implementation scenario: cloud-based implementation. In this scenario, the digital coherent receiver’s measured waveform data are immediately sent to the centralized data center located at a different place, e.g., a cloud data center, via a supervisory network. The supervisory network shown in Fig. 4 is a network used for monitoring and controlling remote optical transceivers and nodes, which is logically independent of the main optical channel, such as 100 or 400 Gbits/s. One advantage of the cloud-based implementation is that rich computational resources are available at the centralized data center. Compared to standard edge devices, well-equipped centralized cloud data centers can provide better computational efficiency. The other possible advantage of this scenario is the high reusability of data. Datasets collected at multiple digital coherent transceivers can be utilized for advanced multi-modal data analyses, such as network-wide fault isolation and prediction.
Conversely, the cloud-based implementation scenario may not be preferred in some use cases because of data security and latency in transmitting and/or storing waveform data between the data measurement location (digital coherent receiver) and cloud data center. Data sharing may pose a security concern on the operator. For instance, one can reproduce transmitted data itself from waveform data, i.e., a data payload can be eavesdropped on from the shared waveform data in principle. In addition, the implementation typically leads to a large latency due to the communication between the digital coherent receiver and cloud computing resource. The large latency may hinder the real-time processing and feedback using monitor results that may be desirable to enable some control use cases.
Figure 4(b) shows a block diagram of the second implementation scenario, i.e., edge-based implementation. Compared to the cloud-based implementation, the processing of the measured waveform is carried out on the edge-side device, e.g., digital coherent receivers inside or their embedded peripherals such as FPGA. The main advantage of edge-based processing is a reduced latency by cutting the communication between measurement and processing sites. With the possibly low latency, a prompt response according to the monitor results can be achieved. It is valuable when the transceiver or other network equipment needs real-time feedback. As regards the security issue, the edge-based implementation can simply minimize the risk and concern by processing data locally.
Furthermore, most edge-based implementations tend to suffer from a limitation of available computation resources at the edge, which is usually limited rather than the cloud. This scenario requires lightweight methods for the edge-based implementations and introduces additional discussion points related to the trade-off between performance and computational complexity reduction, such as bit accuracy/quantization or approximated functions in the algorithm. Some initial considerations on the computation complexity of the DSP-based fiber-longitudinal monitor will be provided in Section 6.
5. EXPERIMENTAL SETUP FOR PROOF OF CONCEPT
The experimental setup to evaluate the DSP-based fiber-longitudinal monitor in a dense wavelength-division multiplexing (DWDM) transmission link is shown in Fig. 5
Fig. 5. Experimental and simulation setup. DAC, digital-to-analog converter; InP IQM, indium phosphide in-phase and quadrature-phase modulator; LD, laser diode; EDFA, erbium-doped fiber amplifier; VOA, variable optical attenuator; ASE, amplified spontaneous emission source; OBPF, optical bandpass filter; ADC, analog-to-digital converter; FPGA, field-programmable gate array.
An optical transmitter was implemented using an external cavity laser (${\sim}{{25}}\;{\rm{kHz}}$ linewidth), a dual-polarization (DP) in-phase and quadrature (IQ) modulator, and a four-channel digital-to-analog converter (DAC). The external cavity laser was emitted on the channel at 193.3 THz. The DP-IQ modulator was driven by the four-channel DAC’s drive signals with a sampling rate of 92 Gsamples/s and 8-bit physical resolution. The transmitter generated Nyquist-filtered 63.25-Gbaud (GBd) 506-Gbit/s DP-16QAM optical signals. The resulting signal was digitally predistorted with a CD of 1500 ps/nm. Other 34-channel 128-Gbit/s 32-Gbaud Nyquist-filtered DP quadrature phase-shift keying (QPSK) signals were used as DWDM neighboring channels. All signals were spectrally multiplexed and launched into a 260-km-long straight fiber line consisting of five spans of standard single-mode fiber (SSMF), each having the length of either 40 or 60 km without any in-line optical dispersion compensation. The fiber launch power into each span was ${+}{{5}}$ and 0 dBm/channel for the measured and neighboring channels, respectively. The fiber-launched power used in the experiment was chosen as the first proof of concept of this method and not optimized to maximize the monitoring accuracy and transmission performance. As such, the fiber-launched power optimization, including the lower launched power, is left for future work.
After a five-span transmission, the channel under measurement was received by a coherent receiver. A local oscillator (${\sim}{{25}}\;{\rm{kHz}}$ linewidth) was superimposed with the signal in a polarization-diversity optical hybrid. The outputs of the hybrid were connected to four balanced photodetectors. The resulting signals were digitized using four analog-to-digital converters (ADCs) with a sample rate of 80 Gsamples/s. The digital samples were sent to a desktop computer equipped with a Xilinx Alveo U200 FPGA accelerator card and used for the analysis of the fiber-longitudinal monitoring. We confirmed the correct data reception by evaluating the bit error rate after the demodulation DSP (not shown in Fig. 5).
6. RESULTS AND DISCUSSION
A. Verification of the Monitor Concept
First, we evaluated the fiber-longitudinal monitor with a cloud-based implementation scenario, which includes a calculation with a double-precision (64-bit) floating-point number representation to obtain the following results. The results are a reproduction of the results in [17], but we present them here to show the base performance. Note that the FEC block in Fig. 3 was skipped in this step and the following experiments, allowing the residual bit error rate of approximately ${{1}}{{{0}}^{- 3}}$ to generate the reference signal on the first tributary. The parameter ${c_{\rm{pre}}}$ was swept with a resolution $\Delta c$ of 16.75 ps/nm (equivalent to 1-km-long SSMF). We used approximately 64.5M data samples for each line and averaged the monitor outputs by each processing data block: 2048 data size with 33,000 averaging. The data size to be processed was a length of digitized data corresponding to HI, HQ, VI, or VQ by the digital coherent receiver at a double sampling rate, i.e., 126.5 GSamples/s in this experiment, with resampling from 80 GSamples/s to 126.5 GSamples/s digitally. HI, HQ, VI, and VQ denote the in-phase (I) and quadrature (Q) component of each horizontal (H) or vertical (V) polarization of the complex optical field.
Figure 6 shows the monitored power profile with the reference power measured via OTDR at each span (black lines) in the experiment. (This result corresponds to those in our previous works, i.e. [16,17].) Three different additional optical power attenuators (1.8, 3.3, and 5.0 dB) were inserted at the 20-km point of the third span, i.e., 120 km from the transceiver. The optical power profile measured by the DSP-based fiber-longitudinal monitor successfully visualized the amplification by the erbium-doped fiber amplifier (EDFA) and the attenuation through span-by-span fiber propagation. Comparing the power profile in a case without an excess attenuator (solid line) to the others (dotted, dashed, and long-dashed lines), one can find the clear power dropping around the transmission distance of 120 km, which is the point corresponding to the inserted excess optical attenuator.
Fig. 6. DSP-based fiber-longitudinal monitor outputs as a baseline in the “normal” (no excess loss, solid line) and “anomaly” cases with three different insertion losses (dotted, dashed, and long-dashed lines) with the measured relative loss by OTDR corresponding to the solid line. This figure is based on Ref. [17].
To discuss a possible application of this monitor, we investigated a spatial identification of the excess optical attenuator, i.e., an unexpected optical loss point on a transmission link. This means that the DSP-based fiber-longitudinal monitor can detect the anomaly and localize this point in terms of the distance from the transceiver.
Fig. 7. Experimental results of the anomaly location indicator (ALI) with two resolution parameters for varying ${c_{\rm{pre}}}$. This figure is based on Ref. [17].
To highlight the anomaly point, we plotted the anomaly location indicator (ALI) as a function of the equivalent transmission distance in Fig. 7 for the same data used in Fig. 6. This result is also based on our previous work in Ref. [17]. The ALI is defined by
The ALI provides useful information to localize a point at which an unexpected optical attenuation has occurred. To investigate the spatial resolution of the ALI, we plotted the ALI calculated with two difference resolution parameters $\Delta c$ (corresponding to 1 km and 5 m SSMF) for varying ${c_{\rm{pre}}}$ in Fig. 7. The results show that the ALI provides an on-demand spatial resolution depending on the resolution parameter $\Delta c$. The ALI calculated with the finer resolution parameter has the potential to estimate the anomaly location within an error of less than 1 km.
B. Impact of Calculation Accuracy in the Monitor
This section discusses the feasibility study of the DSP-based fiber-longitudinal monitor with limited calculation accuracy. In the edge-implementation scenario, available computation resources would be usually limited because of a limitation in the energy/power consumption on embedded edge devices, such as ASIC or FPGA. Here, we investigated a trade-off between the computational resolution and performance on our DSP-based fiber-longitudinal monitor, which uses a sophisticated waveform transition due to fiber nonlinearity to estimate the power profile over the transmission link.
Figures 8(a) and 8(b) show the output results of the DSP-based fiber-longitudinal monitor with a double-precision floating-point format (64-bit) and single-precision floating-point format (32-bit) in its calculation.
Fig. 8. DSP-based fiber-longitudinal monitor outputs calculated by the single- and double-precision floating-point number representations: (a) without output averaging and (b) with 3000-times averaging.
As shown in Fig. 8(a), the DSP-based fiber-longitudinal monitor showed reasonable performance on the double-precision case and successfully extracted power profile information from the received waveform measured by the experiment. For plotting Fig. 8(a), we used approximately 6.1M data samples for each HI, HQ, VI, and VQ line and processed all the data at once. Although a detailed analysis to identify the required computation precision for this algorithm is left for future work, the monitor output was noisy when the computation precision was limited to single precision as shown in Fig. 8(a). The result indicates that we should consider the computation precision to design and implement the algorithm.
Next, we further investigated the design space in the fiber-longitudinal monitor. To eliminate the noise from insufficient calculation accuracy, as shown in Fig. 8(a), we simply used the averaging outputs of the monitor to mitigate noise on the outputs. For a fair comparison, we divided the total measured waveform data into small portions: the total 6.1M sample point data was divided into 3000 small data portions consisting of 2048 samples for each HI, HQ, VI, and VQ.
Figure 8(b) shows the fiber-longitudinal monitor results with averaging. The measured results in the single-precision case agreed well with those in the double-precision case. Note that the averaging process may increase the processing latency. Thus, we should consider the monitor’s design space to obtain the desired result even with the limited capability of the computation devices.
C. Considerations on the Computation Time of the Monitor
This section discusses the computation time of the DSP-based fiber-longitudinal monitor for implementation in actual optical networks. The computational complexity ($C$) of the DSP-based fiber-longitudinal monitor is proportional to the number of representative points, i.e., locations to be estimated with the optical power along the distance ($L/R + {{1}}$) and the averaging number $N$ that is determined from the required SNR of the optical signal:
where $L$ is the link distance and $R$ is the resolution along the distance ($R = \Delta c/D$).We benchmarked the computing time for the proposed optical network tomography programmed in C${++}$ on a PC server with the Intel Xeon Gold 5215 CPU operating at a clock frequency of 2.5 GHz where only one of the cores was used.
The power profile estimation for a 260-km link with five spans as it is described in Fig. 6, as an example, took 8.5 min for 352 representative points with an averaging number of $N = {{3000}}$.
Based on Eq. (3), to cope with a longer link distance, higher distance resolution, and higher noise tolerance at a smaller turnaround time, it is crucial to accelerate the computation speed at the edge devices. As a method for speeding up the computation, we considered hardware acceleration to a part of computation of the DSP-based fiber-longitudinal monitor.
By analyzing the contributions of each functional block in the DSP-based fiber-longitudinal monitor computation in the above case, the functional blocks for “Partial CDC of ${{{c}}_{\rm{pre}}}$,” “Non-linear probe,” and “Residual CDC of ${c_{\rm{post}}}$” were found to consume a major portion of the CPU time, i.e., 19.9%, 21.1%, and 36.5%, respectively (summarized in Table 1). Because these functional blocks are heavily replying on the fast Fourier transform (FFT), we chose to implement these operation blocks in FPGA. The register transfer level (RTL) was automatically generated from the C${++}$ code used in the benchmark described above through a high-level synthesis using the development tool Vivado HLS. The RTL was implemented in a Xilinx Alveo U200 FPGA accelerator card that is connected to the CPU environment for the benchmark described in the above by PCI Express 3.0. Then, the operation blocks that were not subject to hardware acceleration remained to be calculated in the CPU. Because of the acceleration, the operation time was reduced from 8.5 min in the operation only using the CPU down to 2.7 min, where the FPGA and CPU consumed 0.8 and 1.9 min, respectively.
Table 1. Portion of CPU Time of Functional Blocks
Because the proposed algorithm allows optical power estimation at one target point in a link to be independent of the estimation for the other target points on the same link, parallel processing can be easily adopted for further significant acceleration.
7. OPTICAL NETWORK TOMOGRAPHY: EXAMPLE APPLICATION ENABLED BY THE FIBER-LONGITUDINAL MONITOR
This section discusses an application enabled by the fiber-longitudinal monitor to capture all the network-level physical characteristics of optical networks. The idea behind the application is simple: overlaying multiple measured results of the power profile over the routing light paths and wavelength to construct network-level, in-span distance-wise, wavelength-dependent digital twins of optical networks. This application is referred to as optical network tomography by the fiber-longitudinal monitor in the following. Optical network tomography becomes a promising tool for handling and efficiently managing physical networks.
Figure 9 shows a block diagram to compute optical network tomography. First, multiple monitoring results by the DSP-based fiber-longitudinal monitors are overlaid on a single optical route, but the different wavelength channels are spectrally multiplexed to obtain a power profile spectrum over this optical route. Second, as shown in Fig. 9, multiple measured power profile spectra regarding with different optical routes are combined into one entity as a digital twin that represents the physical aspect of the optical network.
Considering the real application of optical network tomography, it is attractive to extract the power profile spectrum by the DSP-based fiber-longitudinal monitors. Note that the DSP-based fiber-longitudinal monitor should work independently from amplification methods, such as EDFA and Raman amplification. Thus, the power profile spectrum in optical network tomography can be worked into cutting-edge optical communication systems that utilize multiple bands [24,25] (e.g., the C-, L-, and S-bands) for transmission assisted with distributed Raman amplification (DRA). The use of DRA, which provides broadband amplification and improved SNR, is a promising way to extend the link capacity by extending utilized optical bands [26]. However, the DRA-assisted system introduces design and management difficulties on the DWDM signal power profile because of their wavelength dependency. Optical network tomography, which enables wavelength-specific and in-span distance-wise power profile monitoring, can mitigate the difficulty by providing the DRA gain profile at the actual signal wavelength. Although OTDR, one of the most common techniques to investigate fiber power profiles, can be considered in predicting the DRA performance in off-line measurements, it is not a perfect choice for the in situ monitoring of DRA characteristics. This is because the OTDR does not provide the gain profile at actual signal wavelengths.
The application of the DSP-based fiber longitudinal monitor for DRA-assisted multi-band transmission has been proposed and discussed in Ref. [17]. Following the discussion in Ref. [17], an impressive experimental demonstration to show the power profile spectrum for the DRA system was presented in Ref. [27] by using an alternative monitoring method to estimate the fiber-longitudinal power profile [23]. Although the authors of [27] used the other monitoring method based on multi-segment digital backpropagation for constructing the power profile spectrum, they cited the fiber-longitudinal monitor method reviewed in this paper and mentioned that “the in-situ power profile estimator that enables monitoring of signal power profile has been proposed and can also be used to calculate Raman-amplified power evolution” to imply that the monitor method reviewed in this paper can be used to estimate the power profile spectrum for DRA-assisted multi-band systems.
8. SUMMARY
In this study, we presented data-processing-based, distance-wise optical power monitoring toward optical network tomography that provides a digital twin of the entire optical network in terms of its physical aspect. We investigated its implementation scenario in edge devices. The monitoring method estimates the fiber-longitudinal power profile over multi-span optical transmission links using a receiver-end waveform observed by a single coherent receiver only. We believe that this technology is a valuable part for designing future autonomous optical networks.
Funding
National Institute of Information and Communications Technology.
Acknowledgment
This work was partly supported by the National Institute of Information and Communications Technology (NICT), Japan.
REFERENCES
1. S. Yan, A. Aguado, Y. Ou, R. Wang, R. Nejabati, and D. Simeonidou, “Multi-layer network analytics with SDN-based monitoring framework,” J. Opt. Commun. Netw.9, A271–A279 (2017). [CrossRef]
2. F. Meng, Y. Ou, S. Yan, K. Sideris, M. D. G. Pascual, R. Nejabati, and D. Simeonidou, “Field trial of a novel SDN enabled network restoration utilizing in-depth optical performance monitoring assisted network re-planning,” in Optical Fiber Communication Conference (OFC) (2017), paper Th1J.8.
3. S. Yan, F. N. Khan, A. Mavromatis, D. Gkounis, Q. Fan, F. Ntavou, K. Nikolovgenis, F. Meng, E. H. Salas, C. Guo, C. Lu, A. P. T. Lau, R. Nejabati, and D. Simeonidou, “Field trial of machine-learning-assisted and SDN-based optical network planning with network-scale monitoring database,” in 43rd European Conference on Optical Communication (ECOC) (2017).
4. K. Christodoulopoulos, P. Kokkinos, A. Di Giglio, A. Pagano, N. Argyris, C. Spatharakis, S. Dris, H. Avramopoulos, J. C. Antona, C. Delezoide, P. Jennevé, J. Pesic, Y. Pointurier, N. Sambo, F. Cugini, P. Castoldi, G. Bernini, G. Carrozzo, and E. Varvarigos, “ORCHESTRA—Optical performance monitoring enabling flexible networking,” in 17th International Conference on Transparent Optical Networks (ICTON) (2015).
5. S. Oda, M. Miyabe, S. Yoshida, T. Katagiri, Y. Aoki, J. C. Rasmussen, M. Birk, and K. Tse, “A learning living network for open ROADM networks,” in 42nd European Conference on Optical Communication (ECOC) (2016).
6. F. N. Hauske, M. Kuschnerov, B. Spinnler, and B. Lankl, “Optical performance monitoring in digital coherent receivers,” J. Lightwave Technol.27, 3623–3631 (2009). [CrossRef]
7. A. S. Kashi, Q. Zhuge, J. C. Cartledge, S. A. Etemad, A. Borowiec, D. W. Charlton, C. Laperle, and M. O’Sullivan, “Nonlinear signal-to- noise ratio estimation in coherent optical fiber transmission systems using artificial neural networks,” J. Lightwave Technol.36, 5424–5431 (2018). [CrossRef]
8. D. Zibar, M. Piels, R. Jones, and C. G. Schaeffer, “Machine learning techniques in optical communication,” J. Lightwave Technol.34, 1442–1452 (2016). [CrossRef]
9. T. Tanimura, T. Hoshida, J. C. Rasmussen, M. Suzuki, and H. Morikawa, “OSNR monitoring by deep neural networks trained with asynchronously sampled data,” in Proceedings of OECC/PS (2016), paper TuB3-5.
10. T. Tanimura, T. Hoshida, T. Kato, S. Watanabe, and H. Morikawa, “Data-analytics-based optical performance monitoring technique for optical transport networks (invited),” in Optical Fiber Communication Conference (OFC) (2018), paper Tu3E.3.
11. T. Tanimura, T. Hoshida, T. Kato, S. Watanabe, and H. Morikawa, “Convolutional neural network-based optical performance monitoring for optical transport networks,” J. Opt. Commun. Netw.11, A52–A59 (2019). [CrossRef]
12. F. N. Khan, K. Zhong, X. Zhou, W. H. Al-Arashi, C. Yu, C. Lu, and A. P. T. Lau, “Joint OSNR monitoring and modulation format identification in digital coherent receivers using deep neural networks,” Opt. Express25, 17767–17776 (2017). [CrossRef]
13. D. Wang, M. Zhang, Z. Li, J. Li, M. Fu, Y. Cui, and X. Chen, “Modulation format recognition and OSNR estimation using CNN-based deep learning,” IEEE Photon. Technol. Lett.29, 1667–1670 (2017). [CrossRef]
14. A. P. Vela, B. Shariati, M. Ruiz, F. Cugini, A. Castro, H. Lu, R. Proietti, J. Comellas, P. Castoldi, S. J. B. Yoo, and L. Velasco, “Soft failure localization during commissioning testing and lightpath operation,” J. Opt. Commun. Netw.10, A27–A36 (2018). [CrossRef]
15. C. Mas, I. Tomkos, and O. K. Tonguz, “Failure location algorithm for transparent optical networks,” IEEE J. Sel. Areas Commun.23, 1508–1519 (2005). [CrossRef]
16. T. Tanimura, K. Tajima, S. Yoshida, S. Oda, and T. Hoshida, “Experimental demonstration of a coherent receiver that visualizes longitudinal signal power profile over multiple spans out of its incoming signal,” in European Conference on Optical Communication (ECOC) (2019), paper PD.3.4.
17. T. Tanimura, S. Yoshida, K. Tajima, S. Oda, and T. Hoshida, “Fiber-longitudinal anomaly position identification over multi-span transmission link out of receiver-end signals,” J. Lightwave Technol.38, 2726–2733 (2020). [CrossRef]
18. T. Tanimura, S. Yoshida, S. Oda, K. Tajima, and T. Hoshida, “Advanced data-analytics-based fiber-longitudinal monitoring for optical transport networks (invited),” in European Conference on Optical Communication (ECOC) (2020).
19. T. Sasai, M. Nakamura, E. Yamazaki, S. Yamamoto, H. Nishizawa, and Y. Kisaka, “Digital backpropagation for optical path monitoring: loss profile and passband narrowing estimation,” in European Conference on Optical Communication (ECOC) (2020), paper Tu2D-1.
20. T. Tanimura, T. Hoshida, T. Tanaka, L. Li, S. Oda, H. Nakashima, Z. Tao, and J. C. Rasmussen, “Semi-blind nonlinear equalization in coherent multi-span transmission system with inhomogeneous span parameters,” in Optical Fiber Communication Conference (OFC) (2010), paper OMR6.
21. E. Ip, “Nonlinear compensation using backpropagation for polarization-multiplexed transmission,” J. Lightwave Technol.28, 939–951 (2010). [CrossRef]
22. T. Tanimura, T. Hoshida, S. Oda, T. Tanaka, C. Ohshima, Z. Tao, and J. C. Rasmussen, “Systematic analysis on multi-segment dual-polarisation nonlinear compensation in 112 Gb/s DP-QPSK coherent receiver,” in European Conference on Optical Communication (ECOC) (2009), paper 9.4.5.
23. T. Sasai, M. Nakamura, S. Okamoto, F. Hamaoka, S. Yamamoto, E. Yamazaki, A. M. H. Nishizawa, and Y. Kisaka, “Simultaneous detection of anomaly points and fiber types in multi-span transmission links only by receiver-side digital signal processing,” in Optical Fiber Communication Conference (OFC) (2020), paper Th1F.1.
24. A. Napoli, N. Costa, J. K. Fischer, J. Pedro, S. Abrate, N. Calabretta, W. Forysiak, E. Pincemin, J. P. F. Gimenez, C. Matrakidis, G. Roelkens, and V. Curri, “Towards multiband optical systems,” in Advanced Photonics (2018), paper NeTu3E.1.
25. T. Kato, “Real-time transmission of 240 × 200-Gb/s signal in S+C+L triple-band WDM without S- or L-band transceivers,” in 45th European Conference on Optical Communication (ECOC) (2019).
26. R. J. Essiambre, G. Kramer, P. J. Winzer, G. J. Foschini, and B. Goebel, “Capacity limits of optical fiber networks,” J. Lightwave Technol.28, 662–701 (2010). [CrossRef]
27. T. Sasai, M. Nakamura, T. Kobayashi, H. Kawakami, E. Yamazaki, and Y. Kisaka, “Revealing Raman-amplified power profile and Raman gain spectra with digital backpropagation,” in Optical Fiber Communication Conference (OFC) (2021), paper M3I.5.
Takahito Tanimura received his B.S. and M.S. degrees in physics from the Tokyo Institute of Technology (Tokyo Tech), Tokyo, Japan, in 2004 and 2006, respectively, and his Ph.D. degree in electrical engineering and information systems from the University of Tokyo, Tokyo, Japan, in 2018.
Since 2020, he has been with Hitachi Ltd., Tokyo, Japan, where he has been engaged in the research and development of machine learning technologies with high-speed communication systems. From 2019 to 2020, he was with Fujitsu Limited. From 2006 to 2019, he was with Fujitsu Laboratories Ltd., Kawasaki, Japan, where he was engaged in the research and development of intelligent digital coherent optical communication systems. From 2011 to 2012, he was with the Fraunhofer Institute for Telecommunications, Heinrich Hertz Institute, Berlin, Germany. His research interests include digital signal processing and machine learning for large-scale nonlinear systems.
Dr. Tanimura is a senior member of the Institute of Electronics, Information and Communication Engineers (IEICE) and a member of the Physical Society of Japan (JPS). He has served as an Editorial Committee Member of the IEICE Transactions on Communications and as a Technical Program Committee Member of the Optical Fiber Communication Conference. He has received the IEICE CS Excellent Paper Award for Tutorial Paper, Japan Patent Office Commissioner Award of the National Commendation for Invention, and Award for Science and Technology (Development Category) of the Commendation for Science and Technology by the Ministry of Education, Culture, Sports, Science and Technology (MEXT).
Takeshi Hoshida (S’97–M’98–SM’00) received his B.E., M.E., and Ph.D. degrees in electronic engineering from the University of Tokyo, Tokyo, Japan, in 1993, 1995, and 1998, respectively.
Since he joined Fujitsu Laboratories Ltd., Kawasaki, Japan, in 1998, he has been engaged in the research and development of dense wavelength-division multiplexing optical transmission systems. From 2000 to 2002, he was with Fujitsu Network Communications, Inc., Richardson, Texas. Since 2007, he has been with Fujitsu Limited, Kawasaki, Japan.
Dr. Hoshida is a senior member of the Institute of Electronics, Information and Communication Engineers (IEICE) and a member of the Japan Society of Applied Physics (JSAP).
Setuo Yoshida received his B.S. and M.S. degrees from Hokkaido University, Sapporo, Japan, in 1993 and 1995, respectively.
In 2000, he joined Fujitsu Laboratories Ltd., Kawasaki, Japan. Since 2005, he has been with Fujitsu Limited, Kawasaki, Japan, where he has been engaged in the research and development of the controlling technology of optical devices.
He is a member of the Institute of Electronics, Information and Communication Engineers of Japan.
Shouichiro Oda (S’02–M’06) received his B.E., M.E., and Ph.D. degrees in electronic, information systems, and energy engineering from Osaka University, Suita, Japan, in 2002, 2003, and 2006, respectively.
From 2006 to 2007, he was a postdoctoral researcher with the Chalmers University of Technology, Gothenburg, Sweden, where his research interests included fiber-optic parametric amplifiers and their applications. In 2007, he joined Fujitsu Laboratories, Ltd., Kawasaki, Japan, where he was involved in the research and development of long-haul wavelength-division-multiplexing transmission systems. Since 2019, he has been with Fujitsu Limited, Kawasaki, Japan, where he is involved in the research and development of intelligent optical networking.
He is a member of the Institute of Electronics, Information and Communication Engineers.
Kazuyuki Tajima received B.E. and M.E. degrees in electronic engineering from Utsunomiya University, Utsunomiya, Japan, in 1987 and 1989, respectively.
Starting in 1989, he was with Fujitsu Laboratories Ltd., Kawasaki, Japan, where he was engaged in the research and development of optical communication systems. In 2019, he joined Fujitsu Limited, Kawasaki, where he has been engaged in the development of intelligent optical communication systems.
He is a member of the Institute of Electronics, Information and Communication Engineers of Japan.
