Frequency-Shift Monitoring of Optical Filter Based on Optical Labels over FTN-WDM Transmission Systems

Abstract

Optical network monitoring and soft failure identification such as optical filter shifting and filter tightening are increasingly significant for the complex and dynamic optical networks of the future. Center frequency shift of optical filtering devices in optical networks has a serious impact on the performance of multi-span transmission, especially in high spectrum efficiency faster-than-Nyquist (FTN) transmission systems with various optical switching and add/drop nodes. Existing monitoring schemes generally have the problems of high cost, high complexity, and inability to realize multi-channel online monitoring, which makes it difficult for them to be applied in a wavelength division multiplexing (WDM) system with numerous nodes. In this paper, a monitoring scheme of frequency shift of optical filtering devices based on optical label (OL) is proposed and demonstrated. The signal spectrum of each channel is intentionally divided into many sub-bands with corresponding optical labels loading. The characteristics of spectrum power changing caused by frequency shift can be reflected on labels power changing of each sub-band, which are used to monitor and estimate the value of frequency shift via DSP algorithm. Simulation results show that the monitoring errors of frequency shift can be kept reasonably below 0.5 GHz after 10-span WDM transmission in FTN polarization multiplexing m-ary quadrature amplitude modulation (PM-mQAM) systems. In addition, 250 km fiber transmission experiments are also carried out, and similar results are obtained, which further verify the feasibility of our proposed scheme. The characteristics of low cost, high reliability, and efficiency make it a better candidate for practical application in future FTN-WDM networks.

1. Introduction

With the large-scale commercialization of 5G technology, emerging businesses and industries such as the Internet of Things, smart manufacturing, big data, and cloud computing are developing at a rapid pace [1,2]. The number of users and the volume of transmitted information have increased significantly, which has resulted in the continued development of WDM optical networks towards high capacity, high spectrum efficiency, and long distances [3,4]. Accordingly, the WDM network topology becomes more complex and variable. The reconfigurable optical multiplexers (ROADMs) and/or wavelength selective switches (WSSs) are widely arranged and applied in optical network nodes, making the wavelength connection of the WDM system more dynamic and random [5]. By allocating transmission bandwidth more flexibly and arranging transmission paths dynamically, transmission efficiency can be greatly improved, and WDM optical networks can be driven toward intelligence and efficiency [6,7,8]. However, as a series of ROADMs-based optical filtering devices are applied in WDM networks, the monitoring of soft failure such as filtering shift and filter tightening of devices in optical networks becomes an issue that must not be overlooked. When the filter failure occurs, the service signal will produce a corresponding filtering penalty. The quality of signal drops significantly with the bit error rate (BER) rising, which greatly affects the normal operation of the optical network system [9,10,11]. Therefore, it is necessary to monitor these soft failures to maintain normal and efficient operation of the optical network.

The problems of soft failure location and identification in optical networks have been explored in recent years, and most of exiting schemes are based on advanced machine learning and training algorithms [12]. In [13], the optical spectrum analysis (OSA) is used to acquire the optical spectrum features of service signals. Combining the classifying and training algorithms, the situations of the two most common filter-related soft failures, filter shift and tight filtering, can be obtained. In [14], a generative adversarial network (GAN)-based soft failure detection and a few-shot soft failure identification algorithm is verified. The scheme uses a number of failure samples from a coherent receiver for training to achieve soft failure identification. Furthermore, in [15], a novel two-stage soft failure identification scheme based on a convolutional neural network (CNN) and receiver digital signal processing (DSP) is proposed and demonstrated. The features from the power spectrum density (PSD) of received signals are used to identify different soft failures such as filter tightening, SNR degradation caused by amplified spontaneous emission (ASE) noise, and filter shift. Above all, most of the existing frequency-shift monitoring schemes obtain the relevant information about the service signal through OSAs or coherent receivers, and they then combine them with machine learning algorithms for pre-learning and training to realize failure identification and monitoring. However, the problems of high cost, difficulty in miniaturizing, and integration still exist, and only end-to-end status monitoring can be completed. They cannot fulfil the requirement of low-cost, online, and simultaneous monitoring of multi-channel at the transmission node, which makes them difficult to deploy in future dynamic and complex optical networks.

In recent years, another optical network monitoring scheme based on optical labels has attracted wide attention [16,17]. A low-frequency digital signal carrying the transmitted service information (baud rate, modulation format, transmission path, add/drop nodes, etc.) is attached to the corresponding service signal as optical labels, loading using DSP units at the transmitter side. After optical IQ modulation, each label signal follows the corresponding service wavelength along the road, experiencing the same channel fading and impairment as the service signal. At any monitoring node, the label information on all wavelengths can be obtained simultaneously by using only one low-bandwidth photoelectric detector (PD) and low-rate analog-to-digital converters (ADC), and the performance parameters can be obtained with DSP technology. There have been several reports of optical labels being used to monitor optical power, OSNR, and other performance parameters [18,19]. Since it can obtain information from all wavelength channels at the same time, the operation status of a WDM network can be reliably monitored online, so that the control platform can better optimize the allocation and scheduling of wavelength resources in the entire optical network. Meanwhile, the label-based monitoring scheme has characteristics of high processing efficiency and low capacity cost compared with the traditional scheme [19]. Therefore, it is necessary to propose a frequency-shift monitoring algorithm based on optical labels to adapt to the future optical networks with complex, variable, and dynamic characteristics.

In this paper, a scheme for frequency-shift monitoring of optical filter devices based on multi-band optical labels is proposed and demonstrated. The DSP algorithm is used at the transmitter side to divide the service spectrum into many sub-bands after Fast-Fourier-transform (FFT) for each service signal. For a single service wavelength signal, the symmetric service spectrum of each sub-band is modulated using different optical labels with different frequencies. The frequency shift of filter devices could be obtained by calculating the power ratio of the negative and positive outermost sub-band spectrums during label detection and procession at a monitoring node. Meanwhile, compared with existing schemes based on OSAs or coherent receivers to obtain spectrum information for a single channel, the proposed scheme uses only a low-bandwidth PD and a low-rate ADC to simultaneously acquire the label information of all transmission channels in the fiber transmission link. With these optical labels, the monitoring of the frequency-shift status of all channels can be obtained at any transmission node, which ensures efficient online monitoring and maintains normal operation of the optical transmission system. To verify our proposed scheme, we have carried out a transmission simulation from 1 to 10 spans (100 km per span) of fiber transmission and experiments testing 250 km fiber transmission. Both results show that the error of frequency-shift estimation can be kept below 0.5 GHz. With this scheme, network failure identification and localization can be detected quickly for timely handling afterwards. The characteristics of low cost, high reliability, and high efficiency lend the scheme a wide range of application prospects for future optical networks.

2. Principle

The optical label is known as a top modulation or a small amount of amplitude modulation, also called low-frequency perturbation or pilot tones in some studies [20]. The schematic diagram of the overall principle proposed in this scheme is shown in Figure 1. The service signal baseband information is generated at the transmitter side and mapped in the digital domain according to the preset modulation format and symbol rate. In the optical label generator module, the label bit information is determined based on the related parameter information of the service signal. This information is mapped, and pilot tones (PTs) are modulated to generate optical labels. Then, the optical labels of different frequencies are top-modulated in the digital domain in order to load them on the different sub-bands of the service signal according to the preset modulation depths. The service digital signal carrying the multi-frequency optical labels is generated. After the analog-to-digital conversion, the optical modulator is driven to convert the digital signal into the optical signal of the corresponding channel wavelength, which completes the optical label loading of one channel. For other optical channels of different wavelengths, the same modulation process is performed. After multiplexing the optical signal into the optical fiber transmission, the optical label and the original signal will be “bonded” into the optical network for transmission. The monitoring of optical labels can be set at any ROADM node, optical power amplifier input and output side, and other important location in the optical network. When monitoring is required, the front end of the optical performance monitoring (OPM) module extracts 1% of the power component of the transmitted optical signal in the fiber using an optical splitter. With the help of direct detection PD and analog-to-digital converter, the optical signal completes the process of receiving and sampling discretization. Optical labels transmitted in the measured fiber of all channels are obtained simultaneously. Combined with the subsequent DSP technology, the monitoring of relevant performances such as optical power, Optical Signal-to-Noise Ratio (OSNR), filtering costs, etc. are completed, and the recovery of service-related identity information is achieved. At the same time, the above monitoring results are uploaded to the control plane to reasonably confirm the optical network operation status so that the control plane can make a decision to dynamically adjust the network status if necessary.

It is worth noting that since a single fiber in a WDM system often transmits service information for multiple wavelength channels, different optical labels are loaded at different frequencies for each wavelength channel. Here, unlike with a regular label-loading scheme, we have chosen to load multiple labels on a single channel with multiple sub-bands. The attenuation of optical label power caused by the chromatic dispersion (CD) effect can be mitigated effectively with the narrower spectrum bandwidth of optical label loading. Meanwhile, to improve the transmission efficiency and the signal-to-noise ratio of the label signal, a PM-FTN service is used, and the same frequency that the optical label uses is loaded at the same position in each polarization state. Next, a channel is used as an example to illustrate the specific process of the muti-band label-loading DSP module.

2.1. Muti-Band Optical Label Loding

Figure 2 shows the specific DSP flow after the label as well as how the service signal is generated for a single channel. After mapping of the service signal and the label is completed, the mapped service signal is subjected to FFT processing. After that, the spectrum of the service signal is divided into multiple sub-bands. Meanwhile, the low-rate digital bit sequence (containing service signal identity information) label signals are multiplied with different pilot tone frequencies, respectively, after symbol mapping to obtain multiple optical labels. These optical labels are multiplied with the time domain parts corresponding to different sub-bands of the service signal after IFFT, respectively, to complete the top modulation, thus realizing the label loading of the service signal. After that, the service signal containing the optical labels will be sent to the digital-to-analog (DAC) to complete the digital-to-analog conversion, and the converted digital signal will be sent to the IQ modulator to complete the optical modulation. At this time, for any service signals, a single wavelength is loaded with N optical label information. The single wavelength service signal can be expressed as:

$$\begin{array}{l}{E}_{i\_x(y)}(t)=A_{i\_1,x(y)}\left(t\right)\exp \left\{j\left(2\pi f_{0,i}t+{\theta }_{0,x(y)}\left(t\right)\right)\right\}\left[1+m\cos \left(2\pi f_{1}t\right)D_i\left(t\right)\right]\\ A_{i\_2,x(y)}\left(t\right)\exp \left\{j\left(2\pi f_{0,i}t+{\theta }_{0,x(y)}\left(t\right)\right)\right\}\left[1+m\cos \left(2\pi f_{2}t\right)D_i\left(t\right)\right]+\cdots +\\ A_{i\_N,x(y)}\left(t\right)\exp \left\{j\left(2\pi f_{0,i}t+{\theta }_{0,x(y)}\left(t\right)\right)\right\}\left[1+m\cos \left(2\pi f_{N}t\right)D_i\left(t\right)\right].\end{array}$$

(1)

where $E_{i\_x(y)}(t)$ represents the X/Y polarization state of the service signal after label loading. $A_{i\_1,x(y)}\left(t\right)$...$A_{i\_N,x(y)}\left(t\right)$ represents the magnitude of the time domain corresponding to the $N_{th}$ sub-band in the frequency domain of the $i_{th}$ channel after IFFT. $f_{0,i}$ denotes the center wavelength of the $i_{th}$ channel. ${\theta }_{0,x(y)}\left(t\right)$ represents the X/Y phase size of the polarization state. m is the modulation depth of the label signal, whose value is defined as the ratio of the optical label signal magnitude to the pure service signal magnitude. In practice, it generally does not exceed 15%. $f_1$...$f_n$ denotes the frequency of the pilot tone multiplied by the 1st to $N_{th}$ optical label. $D_i\left(t\right)$ represents the digital bits sequence, which carries information about the service signal in each optical label.

At the monitoring node, 1% of the optical power is divided using a 1:99 optical coupler for the OPM module to monitor and process. The optical label information of all wavelength channels can be obtained by using low-bandwidth PD to complete the optoelectronic conversion simultaneously. Since the PD direct detection uses the square law detector, the received signals can be expressed as:

$$\begin{array}{l}I(t)={\displaystyle \sum _{i=1}^R\left({\left|E_{i\_x}(t)\right|}^2+{\left|E_{i\_y}(t)\right|}^2\right)}\\ ={\displaystyle \sum _{i=1}^R\left\{\begin{array}{l}{\left\{A_{i\_1,x}\left(t\right)\left[1+m\cos \left(2\pi f_{1}t\right)D_i\left(t\right)\right]+\cdots +A_{i\_N,x}\left(t\right)\left[1+m\cos \left(2\pi f_{1}t\right)D_i\left(t\right)\right]\right\}}^2+\\ {\left\{A_{i\_1,y}\left(t\right)\left[1+m\cos \left(2\pi f_{N}t\right)D_i\left(t\right)\right]+\cdots +A_{i\_N,y}\left(t\right)\left[1+m\cos \left(2\pi f_{N}t\right)D_i\left(t\right)\right]\right\}}^2\end{array}\right\}}\\ ={{\displaystyle \sum _{i=1}^R\left\{\begin{array}{l}{{\displaystyle \sum _{k=1}^N\left\{A_{i\_k,x}\left(t\right)\left[1+m\cos \left(2\pi f_{k}t\right)D_i\left(t\right)\right]\right\}}}^2+\\ 2{\displaystyle \sum _{k=1}^N{A}_{i\_k,x}\left(t\right)\left[1+m\cos \left(2\pi f_{k}t\right)D_i\left(t\right)\right]{\displaystyle \sum _{l>1,l<k}{A}_{i\_l,x}\left(t\right)}}\left[1+m\cos \left(2\pi f_{l}t\right)D_i\left(t\right)\right]+\\ {{\displaystyle \sum _{k=1}^N\left\{A_{i\_k,y}\left(t\right)\left[1+m\cos \left(2\pi f_{k}t\right)D_i\left(t\right)\right]\right\}}}^2+\\ 2{\displaystyle \sum _{k=1}^N{A}_{i\_k,y}\left(t\right)\left[1+m\cos \left(2\pi f_{k}t\right)D_i\left(t\right)\right]{\displaystyle \sum _{l>1,l<k}{A}_{i\_l,y}\left(t\right)}}\left[1+m\cos \left(2\pi f_{l}t\right)D_i\left(t\right)\right]\end{array}\right\}}}_{·}\end{array}$$

(2)

where I(t) represents the electrical signal after square-law detector, and R represents the number of received channels. $A_{i\_k,x}\left(t\right)$ denotes the time domain amplitude of the $k_{th}$ sub-band of the service signal in channel i after IFFT. At this point, the received signal spectrum can be analyzed and processed using the DSP algorithm. Since the labels loaded on each channel are not similar, the label information of a single channel can be filtered out using a digital band-pass filter. For the optical label with the frequency $f_k$ of the $i_{th}$ channel, after removing the direct current (DC), it can be expressed as:

$$\begin{array}{l}{I}_k(t)=2m{A}_{i\_k,x}{\left(t\right)}^2\cos \left(2\pi f_{k}t\right)D_i\left(t\right)+2m{A}_{i\_k,x}\left(t\right)\cos \left(2\pi f_{k}t\right)D_i\left(t\right){\displaystyle \sum _{l\ne k}^N{A}_{i\_l,x}\left(t\right)}\\ +2m{A}_{i\_k,y}{\left(t\right)}^2\cos \left(2\pi f_{k}t\right)D_i\left(t\right)+2m{A}_{i\_k,y}\left(t\right)\cos \left(2\pi f_{k}t\right)D_i\left(t\right){\displaystyle \sum _{l\ne k}^N{A}_{i\_l,y}\left(t\right)}\\ =2m{A}_{i\_k,x}\left(t\right){\displaystyle \sum _{l=1}^N{A}_{i\_l,x}\left(t\right)}\cos \left(2\pi f_{k}t\right)D_i\left(t\right)+2m{A}_{i\_k,y}\left(t\right){\displaystyle \sum _{l=1}^N{A}_{i\_l,y}\left(t\right)}\cos \left(2\pi f_{k}t\right)D_i\left(t\right)\\ =2m\left[A_{i\_k,x}\left(t\right)A_{i,x}\left(t\right)+A_{i\_k,y}\left(t\right)A_{i,y}\left(t\right)\right]\cos \left(2\pi f_{k}t\right)D_i{\left(t\right)}_{·}\end{array}$$

(3)

where $A_{i,x}\left(t\right)$, $A_{i,y}\left(t\right)$ denotes the magnitude of the time domain amplitude corresponding to the X and Y polarization states in the $i_{th}$ channel, respectively.

2.2. Frequency-Shift Monitoring Scheme

It is clear that the relationship between the service signal optical power and the optical label power exists, and the power magnitude of each sub-band can then be reflected using the signal power through the label [21]. Thus, the label power information can be used to recover the general shape of the service signal. And the label power variation is used to reflect the service signal affected by the filtering damage. For the label power statistics, we can use the information from the received label spectrum from the PD detection and FFT transformation to perform an integral operation to obtain [19]. Since different frequency optical labels are loaded on different sub-bands of the service signal in the originating label loading, the power of these optical labels can reflect the status of the optical power at each sub-band spectrum at this time. When the service signal is sent through ROADMs, MUXs/DEMUXs, and other optical filter devices, and if the filter window center frequency shift occurs, the optical power at the high frequency of the service signal will be significantly reduced due to the asymmetric filtering effect, corresponding to a significant drop in label power. But regarding the other side of the label power, the impact is not significant. Therefore, a label power difference between negative and positive symmetric parts of the service spectrum occurred, and different frequency-shift states resulted in different power difference values. This feature can be used to complete the monitoring of the filter center frequency shift. As shown in Figure 3, the service signal has been modulated with multiple optical label signals with a frequency shift of filter at this time.

The power ratio of the negative and positive symmetric positions of the service spectrum can be defined as $\gamma$:

$$\gamma =\frac{P_{high\_label+}}{P_{high\_label-}}=10\lg \left(\frac{P_{high\_label+}}{P_{high\_label-}}\right)(dB).$$

(4)

where $P_{high\_label1-}$ and $P_{high\_label+}$ are the power of the corresponding labels loaded in the negative and positive outermost frequency parts of the service signal, that is, the power of the optical label $f_1$ and optical label $f_2$ in Figure 3 above. If there is no frequency shift in the center of the optical filters, there should be almost no difference in the optical power of the negative and positive high-frequency parts of the spectrum, and the power difference should be about 0 dB. However, if the center frequency of the optical filter is shifted relative to the service signal, then an asymmetric filtering effect will occur, which can be reflected in the change of the $\gamma$ value. By measuring the power difference brought about by the frequency shift of the optical filter relative to the center wavelength of the service in advance, the frequency-shift power difference reference curve is obtained, which can be used as a reference for monitoring the failure of filter devices at each monitoring node in the transmission link and network.

However, considering the uncertainty of the differential beat noise generated during PD detection, the received power of the optical labels in the actual multi-span transmission will fluctuate in a small range, resulting in uncertain fluctuations of the $\gamma$ value. To reduce the impact of the previous span transmission, the power difference value at this time can be subtracted from the power difference from the previous span transmission and then compared with the reference curve of this span to obtain more accurate monitoring results, as shown in Equation (5):

$${\gamma }_{test}={\gamma }_n-{\gamma }_{n-1}{}_{\cdot }$$

(5)

where ${\gamma }_{test}$ represents the label power difference used for monitoring after the nth span. ${\gamma }_n$ represents the label power difference after n-span transmissions. ${\gamma }_{n-1}$ represents the label power difference without frequency shift of optical filter devices after the first n − 1 span transmissions.

3. Simulation and Discussion

With the help of VPI TransmissionMaker 9.0 created by VPIphotonics (Berlin, Germany) and MATLAB R2020b created by MathWorks (Natick, MA, USA), we have constructed a simulation platform of a WDM transmission system with an OPM module based on DPSK-modulated digital optical labels to investigate the performance of the proposed scheme for WSS frequency-shift monitoring. Here we build an 8-channel 10-span WDM transmission system. Meanwhile, for convenience, we choose a wavelength channel with a center frequency of 193.1 THz to study the frequency-shift monitoring effect of the service signal after passing through different filter devices’ frequency-shift states. The service signal here is set as a 40 GBaud PM-FTN-QPSK signal. And considering their prevalence in practical applications, the roll-off factor of filter shaping is set to 0.1, and the FTN acceleration factor is set to 0.9. For label loading, we choose to load six labels with different frequencies on the QPSK signal using the 40 GBaud symbol rate, and each label symbol rate is set to 200 KBaud. The frequencies of the pilot tones for label modulation are set to 40 MHz/40.5 MHz, 45 MHz/45.5 MHz, and 50 MHz/50.5 MHz. The six labels are arranged from high- to low-frequency parts of the service spectrum and are symmetrically distributed with a central frequency of 193.1 THz. Meanwhile, the filter window shape of WSS uses a 4.5 order super Gaussian filter with a −3 dB bandwidth of 37.5 GHz. A PD with a −3 dB bandwidth of 200 MHz and an ADC of 400 MSa/s are used for label signal acquisition and processing at receiver or monitoring nodes.

Table 1. Schematic of key simulation parameters.

Service Signal Format	PM-QPSK	Modulation Depth	5%/10%/15%
Service Baud Rate	40 GBaud	WSS −3 dB Bandwidth	37.5 GHz
Roll-down factor	0.1	EDFA Noise Figure	5 dB
FTN Acceleration factor	0.9	Span Length	100 km
Launch Power	0 dBm	PMD Coefficient	0.1 ps/(km1/2)
Digital Label Format	DPSK	PD Bandwidth	200 MHz
Digital Label Rate	200 KBaud	ADC Sample Rate	400 MSa/s
Carrier Wavelength	193.1 THz	Fiber Attenuation	0.2 dB/km

lists the important simulation parameters.

Performance of Frequency-Shift Monitoring

Considering the use of WSSs in actual commercial use, a filtering window with a −3 dB bandwidth of 37.5 GHz is used in the simulation, and the frequency response is represented as shown in Figure 4a. However, due to the cascading effect of multiple WSSs in the long-distance transmission of the backbone network, the filtering window becomes narrower as shown in Figure 4b, which results in a serious filter-tightening effect. Therefore, the powser difference reference curves under different filtering window shapes should also be reset according to different filtering windows.

The simulation continuously changes the WSS center frequency under back-to-back transmission to produce a relative frequency shift. The reference value of the power difference (dB) can be obtained based on Equation (4). And to remove the small deviation in the label power calculation, the scheme is normalized to the outermost label power difference of the spectrum when there is no frequency shift in the WSS. That is, the power difference is 0 dB when there is no relative frequency shift. Similarly, because of the narrow band-filtering effect generated by multiple WSS cascades, the power difference reference curve will also change. By changing the number of passing WSSs, the reference curve after passing different numbers of WSSs can be obtained as shown in Figure 5.

From the results of the power difference seen in Figure 5, when the center frequency of the WSS is shifted, the power difference of the labels loaded to the negative and positive service spectrum changes as a consequence. As the frequency shift becomes larger and larger, the power difference changes more rapidly. When the number of passing WSSs increases, the labels at the high frequency of the service suffer a larger portion of the attenuation, which leads to a flatter ratio (subtracted dB) reference curve.

The service rate used in the simulation is 40 GBaud. Since the FTN modulation is used, the roll-off factor of the forming filter and the FTN acceleration factor are set to 0.1 and 0.9, respectively, causing the service signal bandwidth to be close to 39.6 GHz, which means that the bandwidth distribution is ±19.8 GHz with the optical carrier 193.1 THz as the center. Therefore, three kinds of label-loading width ratios of sub-bands are selected in the simulation; from the negative to the positive of the service spectrum, they are 9.8/5/5/5/5/5/9.8, 11.8/4/4/4/4/4/11.8, and 13.8/3/3/3/3/13.8. Meanwhile, the label modulation depths are set to 5%, 10%, and 15%, respectively. At this time, the center frequency of the passed WSSs is set to 2 GHz frequency shift after transmitting 1, 2...10 spans, respectively, to simulate the sudden device frequency-shift situation. The sub-band width ratio in Figure 6a below is 11.8/4/4/4/4/4/11.8, and the fixed label modulation depth in Figure 6b is 10%.

Figure 6 reflects the fluctuation of the monitoring error under the condition of a certain modulation bandwidth ratio. From the frequency-shift monitoring results, the average monitoring errors at 5%, 10%, and 15% modulation depths are 0.187 GHz, 0.12 GHz, and 0.099 GHz, respectively, with variances of 0.0172, 0.00448, and 0.00615, respectively, and the maximum monitoring error does not exceed 0.5 GHz. It is obvious that when the modulation depth reaches 10%, the monitoring error and fluctuation are in a smaller state. At this time, the label signal-to-noise ratio is already in a good state, so the modulation depth increasing from 10% to 15% does not have a significant monitoring performance improvement, and the maximum monitoring error does not exceed 0.35 GHz in these conditions. Figure 6b shows the monitoring effect under different label-loading sub-bandwidths; to some extent, this is also an impact of the label power. The average values of monitoring errors are 0.208 GHz, 0.12 GHz, and 0.099 GHz, and the maximum monitoring error does not exceed 0.6 GHz for label loading bandwidth ratios of 9.8/5/5/5/5/5/9.8, 11.8/4/4/4/4/4/11.8, and 13.8/3/3/3/3/13.8, respectively. When the power of the label used for frequency-shift monitoring increases, the monitoring accuracy at this time will also improve. However, when the signal-to-noise ratio of the label is increased to a better level, the improvement for the monitoring performance also becomes minimal. By controlling the modulation bandwidth and modulation depth, the monitoring error can be kept below 0.5 GHz or even lower. To verify the frequency-shift monitoring range of this scheme, the simulation is performed at 10% modulation depth. The frequency-shift cases of 1 GHz, 2 GHz, 3 GHz, 5 GHz, 8 GHz, and 10 GHz are set to occur after 1, 2...10 spans of transmission, respectively. The obtained results are shown in Figure 7.

The results depicted in in Figure 7 show that the maximum error generated by the monitoring of this scheme is not more than 0.5 GHz when the maximum frequency shift of the service signal 10-span transmission is 10 GHz and the average monitoring error is not more than 0.2 GHz. When the frequency shift of the filter pieces is more serious, there may be a relatively large monitoring error. This is due to the impact of asymmetric filtering caused by frequency shift. It produces great attenuation for the label power used for monitoring, which makes the signal-to-noise ratio of label reception decrease, and the beat frequency noise generated by PD direct detection creates greater interference for monitoring.

4. Experiment and Discussion

To verify the monitoring effect of this scheme in actual transmission, we built a single-channel 250 km transmission system. Its main scheme is shown in Figure 8. We set the service signal to 30 GBaud PM-FTN-QPSK with a roll-off factor of 0.1 and an FTN acceleration factor of 0.9. The number of service sub-bands is set to 4 in the experiment, and they are distributed in the negative and positive symmetric positions of the service spectrum. And the frequency of loaded optical labels are 40/44 MHz and 60/64 MHz, respectively. The modulation depths of the optical labels are both 15%. The ratio of the optical label modulation width is similar to that in the simulation as 11.8/8/8/11.8. The number of label information data bits is 8 bit, with an internal bit sequence [1 1 1 0 1 1 1 0 1]. And the digital information rate of the label here is set to 2 MBit/s, which means that the transmission time of a bit is 0.5 μs. All 4 labels are loaded onto the corresponding sub-bands of the spectrum via offline DSP processing at the transmitter side. After IQ modulation and standard single-mode fiber transmission, the service signal is compensated using an amplifier and passed through the filter window. A 4.5-order super-Gaussian filter with a −3 dB bandwidth of 25 GHz is used in the experiment and is shown in Figure 9. Here, an attenuator is used to simulate the situation of the spectrum splitting of 1:99. Then, the signal required for processing and monitoring is obtained through 200 MHz bandwidth PD after direct detection. This analog signal is sampled and quantized using a 500 MSa/s oscilloscope. It is worth noting that we choose a sampling time of 9 μs, which is more than two cycles of digital label sequence, to ensure that a complete digital label signal is obtained after sampling.

In our experiments, we conducted the optical label reception tests when the WSS had a 10 GHz relative shift after 80 km, 160 km, and 250 km fiber transmissions, respectively. The results depicted in Figure 10 show that even though the center frequency of the WSS filter window is shifted by 10 GHz, it only affects the signal on one side of the service spectrum. There is no significant effect on the demodulation of labels on the opposite side of the frequency shift. Even a large frequency shift occurs after 250 km of transmission as shown in Figure 10b, and the label information can still be demodulated correctly. The label sequence [1 1 1 0 1 1 1 0 1] can be easily identified and acquired, which is good news for WDM network system monitoring.

The results of the frequency-shift reference curve and monitoring in the experiments are shown in Figure 11. The reference curve here is slightly different from the simulation, mainly because the filtering conditions are stricter than in the simulation. When the frequency offset exceeds 9 GHz, most of the outside label used for monitoring is filtered out, so the reference curve becomes increasingly smooth. From the results after 80 km to 250 km transmission, the filtering shift generated by the WSS filters can be effectively monitored, and the maximum monitoring error of the experiment is not more than 0.5 GHz. It is worth noting that, due to the limitations of the experimental conditions, we reduce the service rate to 30 GBaud and increase the rate of digital labels to 2 Mbit/s compared with the simulations, which implies an increase of the label spectrum bandwidth. Under the premise of ensuring nearly similar incoming fiber power, the power spectral density of the service signals increases compared to 40 GBaud, which exacerbates the beat noise between the services and between the services and labels at the time of PD detection, resulting in a great decrease in the signal-to-noise ratio of the labels. This also makes the measurement of the label power more affected by noise when a higher frequency shift occurs, which leads to larger monitoring errors in the experiments, but overall, it can still be guaranteed to be within 0.5 GHz after a 250 km transmission. Furthermore, the monitoring error in the experiment fluctuates within a small range, which is mainly caused by other various types of damage such as chromatic dispersion, polarization dependent impairments, and nonlinear effects in the fiber transmission.

Meanwhile, compared with the previous scheme mentioned in [13,14,15,16], expensive devices such as OSAs or coherent receivers are no longer needed, and pretraining processing in various scenarios is not required, so the economic cost and complexity has been greatly reduced. It also ensures that the frequency-shift scheme we proposed is 100% identified with a no more than 0.5 GHz monitoring error. Not only can the monitoring of the frequency-shift failure caused by optical filters or other filtering devices be ensured, but the monitoring accuracy and reliability effectively can also be improved. And the capability of the multi-channel online monitoring can be also achieved.

5. Conclusions

In this paper, a highly efficient and reliable monitoring scheme for the center frequency shift of filter devices inside optical networks is proposed and demonstrated. In this scheme, the service spectrum is divided into different sub-bands, making full use of the characteristics of the filter frequency shift that has a serious impact on the power attenuation of the high-frequency service band. By calculating the power difference between the labels loaded on the positive and negative outermost bands of the service spectrum, the filter frequency-shift condition can be effectively monitored. Moreover, the reference curve of frequency shift is reasonably optimized according to the number of different filters in the transmission, which results in more accurate monitoring results and less error. The simulation results show that the monitoring error proposed in this scheme can be kept below 0.5 GHz if the filter center frequency shift occurs in the span after 10-span WDM transmission. In addition, the experimental results of fiber transmission from 80 km to 250 km also show that the monitoring error of frequency shifts can be kept below 0.5 GHz, and the bit information contained in the optical label can be correctly demodulated even when a 10 GHz relative frequency shift is generated. The scheme only uses low-bandwidth PD and low-rate ADC combined with a DSP algorithm to complete optical label detection and processing, leading to low cost, high efficiency, and reliable monitoring. These devices and algorithms can be easily integrated into monitoring modules to be placed at each switching and transmission node, as well as amplifier inputs and outputs, to complete the monitoring of the entire transmission link, which makes it more practical for future optical network implementations and applications. For the higher-order modulation of the service, the corresponding monitoring error may also increase due to interference caused by greater beat noise when the PD receives label signals during the transmission channel. How to mitigate the effect of beat noise generated to decrease the monitoring error in higher-order modulation format systems is also one of the directions of work we are interested in the future.