Multi-parameter real-time estimation and data traceability optical fiber link monitoring device

Abstract

The invention belongs to the technical field of optical fiber communication, discloses a multi-parameter real-time estimation and data traceability optical fiber link monitoring device, and aims to solve the problem that real-time performance and data depth are difficult to consider in the prior art. The device comprises a receiving module, an analog-to-digital conversion module, a cache module, a data processing module, an alarm analysis module and a data backtracking module. The core of the system is that the receiving module converts a link optical signal into four paths of analog signals; the analog-to-digital conversion module converts an analog signal into a digital signal; the data processing module extracts multi-dimensional key parameters in real time based on a parallel processing architecture and a full-blind digital signal processing technology; the alarm analysis module carries out state judgment according to a preset threshold value and triggers multi-gear alarms; and after the alarm is triggered, the data backtracking module analyzes the original data in the cache to realize accurate fault positioning and reason analysis. The device has the advantages of plug and play and no need of prior information, and the operation and maintenance efficiency of the optical fiber network is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of optical fiber communication technology, and more specifically, relates to an optical fiber link monitoring device with multi-parameter real-time estimation and data traceability. Background Technology

[0002] Optical performance monitoring (OPM) is a core technology for ensuring the reliability of fiber optic communication networks and realizing intelligent operation and maintenance. Its development is crucial for addressing fault prevention and resource optimization in dynamic optical networks. Although existing research has explored various technical approaches, significant challenges remain in the synergistic improvement of real-time performance, data fidelity, and multi-parameter synchronous monitoring capabilities.

[0003] On the one hand, offline post-processing schemes based on advanced algorithms such as deep neural networks (DNNs) can achieve high parameter estimation accuracy under ideal conditions, but their huge computational load and complex model inference process lead to inherent real-time insufficiency, which cannot meet the needs of online monitoring and instant response, and are difficult to apply to dynamic network control scenarios that require microsecond-level response.

[0004] On the other hand, some real-time monitoring schemes using low-bandwidth coherent receivers, while providing immediate feedback, often suffer from insufficient data recording depth due to their limited sampling rate and processing capabilities. This prevents the complete capture and retention of high-fidelity raw waveforms or long-term historical data. Consequently, post-event root cause analysis of critical events such as burst errors in passive optical networks (PONs) and transient disturbances in data center interconnects (DCIs) becomes extremely difficult, as there is a lack of sufficiently detailed data for retrospective diagnostics.

[0005] Therefore, there is an urgent need for a solution that can simultaneously achieve high-speed real-time processing and high-fidelity data recording, not only providing low-latency parameter estimation for network control, but also preserving a complete, high-quality dataset for later in-depth analysis. Summary of the Invention

[0006] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a fiber optic link monitoring method with multi-parameter real-time estimation and data traceability, which aims to realize intelligent monitoring and management of fiber optic links and ensure the reliability and stability of high-speed data transmission.

[0007] To achieve the above objectives, the present invention provides a fiber optic link monitoring device for real-time estimation of multiple parameters and traceable data, comprising: The receiving module is used to convert the link optical signal into four analog signals; An analog-to-digital converter module is used to convert the four analog signals into four digital signals and store them in a buffer module; the four digital signals are XI, XQ, YI, and YQ, where XI is the real part of the X-polarization state, XQ is the imaginary part of the X-polarization state, YI is the real part of the Y-polarization state, and YQ is the imaginary part of the Y-polarization state. The data processing module is used to perform digital signal processing on the four digital signals and extract multi-dimensional key parameters; the key parameters include dispersion value, polarization state, frequency deviation, phase deviation, signal-to-noise ratio, and bit error rate; The alarm analysis module is used to judge multi-dimensional key parameters based on preset parameter thresholds to determine whether the current state triggers an alarm. The data backtracking module is used to retrieve cached data to determine the location and cause of the fault if an alarm is triggered.

[0008] Preferably, digital signal processing does not require training sequences and prior link information, and the processing procedure includes the following steps: 1. Timing Deviation Calibration: Send an identical PRBS analog signal to the four analog-to-digital converters (ADCs) in the analog-to-digital conversion module. Due to differences in hardware paths, the signal arrives at each ADC at different times and completes the conversion.

[0009] The equivalent impulse response of each channel is defined as follows, where the reference channel is channel XI. Channel XI: XQ Road: YI Road: YQ Road: .in , , This represents the timing deviation of paths XQ, YI, and YQ relative to path XI.

[0010] The digital sequences acquired by the four ADCs are as follows: , , , Using path XI as a reference, calculate the cross-correlation functions between the other three paths and path XI.

[0011] XQ and XI are cross-correlated:

[0012] YI and XI are cross-correlated:

[0013] YQ and XI are cross-correlated:

[0014] In each cross-correlation function In the process, find the time delay that maximizes it. This allows us to obtain the timing deviation of each channel relative to channel XI. ,in This is the ADC sampling period.

[0015] Based on the calculated time delay difference The four digital signals are calibrated in the digital domain, which is equivalent to advancing the signal of the lagging channel (or delaying the leading channel).

[0016] 2. Dispersion Estimation and Compensation: Dispersion causes the light pulse to broaden in the time domain, thereby changing the peak power of the signal and affecting the peak-to-average power ratio (PAPR). This property is used to estimate the dispersion value through inversion.

[0017] Fiber dispersion can be mathematically represented as a frequency domain filter. For a cumulative dispersion value of... The link has the following transfer function: ,in The fiber dispersion coefficient, The length of the optical fiber. The wavelength of light At the speed of light, This is the baseband frequency. This indicates that dispersion introduces a phase distortion in the frequency domain that is proportional to the square of the frequency. .

[0018] Theoretically, an ideal time-domain pulse After transmission through optical fiber, its waveform will become ,Right now The inverse Fourier transform. Time-domain waveform. Pulse broadening and cumulative dispersion value Directly related. A complex digital signal can be viewed as a superposition of a series of pulses. Dispersion causes each pulse to broaden and overlap with adjacent pulses; for the signal as a whole, dispersion causes a decrease in peak power, but the average power remains unchanged.

[0019] PAPR is the ratio of peak power to average power, i.e.: .

[0020] When performing dispersion compensation on a signal in the digital domain, the transfer function of the compensation filter is: ,in This is the currently given compensation value. The compensated signal spectrum is: .

[0021] The scanning range for setting the dispersion compensation value is The scan step value is Within the scanning range, calculate the current step sequentially. The corresponding PAPR.

[0022] when At this time, the dispersion is not fully compensated, the signal is still in a broadened state, the peak power is low, and therefore the PAPR is low. hour, The compensation filter precisely cancels out the dispersion effect of the link. At this point, the signal recovers to its compact original waveform in the time domain, and the peak power reaches its maximum value, thus PAPR reaches its maximum value. Therefore, the extreme point of PAPR directly indicates the optimal dispersion compensation value. And corresponding signal dispersion compensation is performed.

[0023] 3. Clock Recovery: The local clock at the monitoring end deviates from the clock at the transmitting end in both frequency and phase. Directly using this deviated local clock for sampling results in an asynchronous sampling sequence, severely degrading signal quality. By detecting the clock error in the digital domain and feeding it back to the voltage-controlled oscillator (VCO) in the analog domain, the VCO directly adjusts the ADC's sampling clock, forming a closed-loop control.

[0024] The ADC uses the current local clock to sample the input analog signal to obtain a digital signal. .

[0025] By analyzing the timing error of digital signals This is used to estimate the deviation between the current sampling time and the ideal optimal sampling time. It is the decision point of the current symbol. and It is the boundary point of the current symbol.

[0026] When the sampling clock is completely correct, the decision point is... Located at the extreme point of the waveform, then the boundary point and When the values ​​are equal, at this time When the sampling clock is off-center and the decision point is not at the ideal optimal sampling time, the boundary point values ​​will be unequal. It will be a positive or negative value, the sign of which indicates the direction of the deviation, and the magnitude of which reflects the degree of deviation.

[0027] Due to the output error signal Typically containing high-frequency noise, a loop filter is used to smooth it out. The digital signal output from the loop filter is then converted into an analog voltage through a digital-to-analog converter. To adjust the output frequency and phase of the VCO, the output frequency of the VCO is: ,in It is the center frequency. This is the gain of the VCO. From this, a new sampling clock can be obtained.

[0028] The newly generated clock from the VCO is fed into the ADC as the sampling clock, forming a closed-loop negative feedback. When the sampling clock lags, timing errors occur. The VCO is driven to increase its frequency, thus advancing the clock; conversely, it is driven to decrease its frequency. Ultimately, this stabilizes the dynamic timing error to zero, at which point the local clock at the monitoring end is synchronized with the clock at the transmitting end.

[0029] 4. Channel Equalization: After transmission through optical fiber, the originally separate X-polarization and Y-polarization states at the transmitting end will mix together. Therefore, it is necessary to separate the aliased X and Y polarization states and compensate for the inter-symbol interference caused by this. Ideally, all symbol points of a signal in the complex plane are distributed on concentric circles, meaning its magnitude is constant. By adjusting the coefficients of an adaptive filter, making the magnitude of the equalizer output signal as close as possible to a constant, channel equalization can be achieved.

[0030] The mixed signal received by the monitoring terminal can be represented as: ,in , These are the X and Y polarization states of the transmitting end signal. , These are the X and Y polarization state signals from the monitoring end. It is a fiber optic channel impairment model.

[0031] The goal of equilibrium is to find a matrix , making Ideally, .

[0032] Cost function: This is used to measure the degree to which the magnitude of the output signal deviates from the ideal constant, where... Expressing expectations, It is a constant (e.g., for a QPSK signal). ).

[0033] First, the equalizer matrix can be initialized. It is a simple identity matrix, that is .

[0034] For each sampling point entering the equalizer Its output is:

[0035]

[0036] in, This represents the convolution operation.

[0037] According to the cost function, the errors of the X and Y polarization states are respectively:

[0038]

[0039] As can be seen from the formula, when When the error is zero, the ideal state is achieved.

[0040] According to error Update the equalizer coefficients by a preset step size. Control the convergence speed and stability.

[0041]

[0042]

[0043]

[0044]

[0045] Continuously update the equalizer coefficients until the error... If the coefficients gradually decrease and stabilize, it indicates that the equalizer coefficients have converged to their optimal values. Afterward, the coefficients are further fine-tuned to track the slow changes in the channel, ensuring dynamic stability of the error.

[0046] 5. Polarization state monitoring: The Jones vector is estimated by using the equalizer coefficients obtained in the channel equalization, and the polarization state (SOP) is quantized.

[0047] The equalizer approximates the inverse channel matrix by adaptively updating the tap coefficients, so the average value of the converged coefficients can represent the channel polarization characteristics at this time.

[0048] Total luminous intensity:

[0049] Horizontal / vertical linear polarization components:

[0050] 45° linear polarization component:

[0051] Circular polarization component:

[0052] Use equalizer coefficients and As an equivalent electric field component, after normalization to eliminate the influence of light intensity, focusing on the SOP itself, the Stokes coefficient is calculated:

[0053]

[0054]

[0055]

[0056] Stokes parameters It can be mapped to points on the corresponding surface of the Poincaré sphere, intuitively displaying the Standard Operating Procedure (SOP).

[0057] 6. Carrier recovery: First calculate the signal... The power is used to remove the phase deviation, where , Let be the modulation order of the signal. Then, the frequency deviation is obtained by calculating the phase rotation rate. Finally, the phase deviation is estimated by averaging multiple adjacent symbols.

[0058] Considering the effects of frequency offset and phase offset, the received first individual code element It can be represented as ,in The original signal phase, The frequency offset between the signal and the local oscillator laser. The symbol sampling interval, This represents the phase noise of the laser.

[0059] Since frequency offset causes a phase difference between adjacent samples, the frequency offset can be calculated simply by estimating the phase difference between consecutive samples. (The phase difference between two adjacent symbols...) Adjacent symbols can be approximated as... .

[0060] Removing the encoded information contained in the signal phase only requires... conduct To the power of 1. We get: .in, Therefore, there is .

[0061]

[0062] The average value is calculated by accumulating dozens of adjacent symbols to eliminate burst errors and further reduce the influence of residual modulation phase. We obtain: The phase bias estimate is Frequency offset estimation is ,in It is a symbol period.

[0063] Finally, the original signal is compensated based on the estimated deviation. .

[0064] 7. Signal-to-noise ratio monitoring: The signal-to-noise ratio (SNR) is inferred by comparing the difference between the received signal and the ideal signal.

[0065] Error vector magnitude (EVM) is the square root of the ratio of error vector power to reference signal power, used to measure the difference between the actual received signal point and the ideal constellation point.

[0066]

[0067] in, It is the first Error vector of symbols, It is the first The reference vector of each symbol, It is the average error power. It is the average reference signal power.

[0068] In an ideal AWGN channel with no other distortions, the error vector It's equivalent to noise. .

[0069]

[0070] in, It is noise power. It is the signal power.

[0071] SNR is defined as the ratio of signal power to noise power, i.e. We can conclude that: .

[0072] Convert SNR to decibels (dB): .

[0073] 8. Bit Error Rate Monitoring: The theoretical bit error rate (BER) under the AWGN channel is calculated by using the calculated SNR.

[0074]

[0075] in, It is the symbol signal-to-noise ratio. It's the baud rate. It is the equivalent noise bandwidth. Ideally, Therefore, it can be approximated as... .

[0076] Nonlinear transformations are used to highlight the periodicity implicit in the symbol sequence, which manifests as discrete spectral lines in the frequency domain. The baud rate is estimated by detecting the intervals between these spectral lines.

[0077] Digital signals can be represented as: ,in It is the sequence of symbols sent, i.e., the signal. It is a shaping pulse. It is the symbol period (baud rate) ), where n is the sequence number and N is the total number of sequences.

[0078] The statistical expectation after the fourth power operation can be simplified to a convolution of a periodic sequence related to the symbol period T: ,here It is a constant. It is a A periodic signal.

[0079] For periodic signals Performing a Fourier transform yields the fourth-order spectrum: ,in, , It is the Dirac function. This indicates that the spectrum is... ( Discrete spectral lines will appear at (integer values), calculate the frequency intervals between these spectral lines. Its reciprocal is the baud rate. .

[0080] For M-QAM signals, the symbol error rate For M-PSK signals, the symbol misalignment rate .

[0081] Based on Gray code mapping, the bit error rate (BER) can be approximated as: ,in, The number of bits carried for each symbol.

[0082] Furthermore, the processing steps in the data backtracking module include: The function of the mother wavelet Through scaling factor Translation factor Scaling and translating the mother wavelet yields a family of wavelet basis functions: ,in, It is an energy normalization factor used to ensure that wavelets at different scales have the same energy.

[0083] Signal The continuous wavelet transform is the inner product of the signal and the family of wavelet basis functions: Among them, scaling factor It is inversely proportional to the frequency, that is The smaller the value, the higher the frequency of the corresponding analysis. Translation factor The corresponding position on the time axis is used to analyze the signal characteristics near a specific moment; here, the moment is taken within ±50 ms of the alarm time. The extracted time-frequency features are compared with a known fault mode database to ultimately determine the specific location and root cause of the fault.

[0084] In summary, compared with the prior art, the technical solutions conceived in this invention have the following beneficial effects: The core advantage of the fiber optic link monitoring device provided by this invention lies in its innovative construction of a collaborative architecture that integrates high-speed real-time processing, fully blind adaptive analysis, and high-fidelity data traceability capabilities, effectively solving the industry pain point that it is difficult to balance real-time performance and data depth in traditional solutions.

[0085] Firstly, in terms of real-time processing, a parallel architecture is adopted, dividing the continuous data stream into blocks, which are then processed simultaneously by multiple processing units. This ensures high throughput and low latency in data processing, enabling the real-time completion of a series of complex and computationally intensive digital signal processing steps, ultimately achieving synchronous parameter extraction and data caching.

[0086] Secondly, in terms of fully blind processing, digital signal processing requires no training sequences or prior link information, directly extracting key parameters such as dispersion, polarization state, and frequency / phase offset from the received signal itself in real time. This ensures plug-and-play adaptive monitoring, significantly improving deployment efficiency and expanding applicability, making it suitable for dynamic network environments with frequent changes in network topology and a lack of prior knowledge.

[0087] Finally, regarding data backtracking, a combination of high-bandwidth caching and intelligent triggering mechanisms is employed. High-bandwidth caching continuously preserves complete original waveform data, while data labeling and saving are dynamically controlled based on real-time analysis parameters. This ensures the complete capture of high-value data related to abnormal events, thus providing a solid data foundation for accurate fault location and cause analysis.

[0088] In summary, through the organic synergy of the three core capabilities mentioned above, the device has constructed a full-link monitoring system that integrates real-time perception, adaptive analysis, and intelligent diagnosis, achieving complete coverage from instantaneous anomaly detection to in-depth fault tracing. Attached Figure Description

[0089] Figure 1 This is a block diagram of the internal structure of a fiber optic link monitoring device for real-time multi-parameter estimation and data traceability proposed in this invention. Figure 2 This is a flowchart of a fiber optic link monitoring method with multi-parameter real-time estimation and data traceability proposed in this invention. Figure 3 This is a framework diagram of a fiber optic link monitoring system with multi-parameter real-time estimation and data traceability proposed in this invention. Detailed Implementation

[0090] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0091] This invention provides a fiber optic link monitoring device with real-time multi-parameter estimation and data traceability, such as... Figure 1 As shown, it includes: The receiving module is used to convert the link optical signal into four analog signals; An analog-to-digital converter module is used to convert the four analog signals into four digital signals and store them in a buffer module; the four digital signals are XI, XQ, YI, and YQ, where XI is the real part of the X-polarized state, XQ is the imaginary part of the X-polarized state, YI is the real part of the Y-polarized state, and YQ is the imaginary part of the Y-polarized state. The data processing module is used to perform digital signal processing on the four digital signals and extract multi-dimensional key parameters; the key parameters include dispersion value, polarization state, frequency deviation, phase deviation, signal-to-noise ratio, and bit error rate; The alarm analysis module is used to judge multi-dimensional key parameters based on preset parameter thresholds to determine whether the current state triggers an alarm. The data backtracking module is used to retrieve cached data to determine the location and cause of the fault if an alarm is triggered.

[0092] Figure 2 This is a flowchart of a fiber optic link monitoring method with multi-parameter real-time estimation and data traceability proposed in this invention.

[0093] Specifically, the digital signal processing does not require training sequences or prior link information, and the processing procedure includes the following steps: Send the same PRBS analog signal as the four channels to the analog-to-digital conversion module, acquire the four digital sequences, take one of them as the reference channel, calculate the cross-correlation function between the other three channels and the reference channel, find the time delay that makes it reach the maximum value in each cross-correlation function, thereby obtaining the timing deviation of each channel relative to the reference channel, and calibrate the four digital signals according to the timing deviation. Dispersion values ​​are estimated for the four calibrated digital signals. The scanning range of the dispersion compensation value is set. The peak-to-average power ratio corresponding to the current dispersion compensation value is calculated sequentially according to the preset step. The dispersion compensation value corresponding to the maximum peak-to-average power ratio within the scanning range is the link dispersion value. The dispersion compensation value is used to perform dispersion compensation on the four digital signals. The clock error detected in the digital domain is fed back to the voltage-controlled oscillator in the analog domain to form a closed-loop control, adjust the local clock to zero error, and then perform channel equalization. Quantize the polarization state based on the adaptive equalizer coefficients obtained from channel equalization; The frequency deviation is obtained by calculating the phase rotation rate, the phase deviation is estimated by averaging multiple adjacent symbols, and the digital signal after channel equalization is compensated based on the frequency deviation and the phase deviation. The signal-to-noise ratio (SNR) is calculated by comparing the difference between the received signal and the ideal signal, and the bit error rate (BER) is calculated using the SNR.

[0094] Specifically, the processing steps in the data backtracking module include: With wavelet function As a basis function, through scaling factor Translation factor Scaling and translating the mother wavelet yields a family of wavelet basis functions: ,in, It is the energy normalization factor; Cached signals The continuous wavelet transform is the inner product of the signal and the wavelet basis functions: The time-frequency features are extracted and compared with a known fault mode database to ultimately determine the specific location and cause of the fault.

[0095] Specifically, the data processing module includes multiple digital signal processors (DSPs), each of which independently processes the allocated data blocks in real time. When the Kth DSP is processing the current data block, the programmable logic device allocates the next data block to the (K+1)th DSP, where K is an integer. The multiple DSPs output the processing results in real time.

[0096] Specifically, the alarm analysis module also adjusts the thresholds of various key parameters used for status determination according to the fiber type of the target fiber link and the service requirements it undertakes; performs correlation analysis on multi-dimensional key parameters to locate link faults; and generates multiple alarm alerts of different levels based on the results of the correlation analysis.

[0097] Specifically, the association analysis is implemented through a trained multilayer perceptron neural network model, including the following steps: Employing the aforementioned multi-dimensional key parameters ,in The number of parameters is given, and historical data is used as the training sample set. Each sample is labeled with the corresponding known state and fault type. The training sample set is input into a multilayer perceptron neural network model, wherein the first layer perceptron neural network model... Layer The output of each neuron is represented as: ,in It's weight. It's a bias. It's an activation function, using a damage function. Measuring the probability distribution of predictions Distribution of real labels The differences between them, With the goal of minimization, the weights and biases are iteratively optimized until the model converges; The final output is transformed into a probability distribution vector. ,in This represents the total number of link failure types. This indicates that the current link state is at the [number]th [level]. The probability of a class; The multi-dimensional key parameter vectors acquired in real time are input into the model, and the output probability distribution vector is determined to be the current fault type.

[0098] Specifically, generating multiple alarm signals includes: setting early warning thresholds and critical alarm thresholds; generating a prompt-level alarm when the deviation of a key parameter is lower than the early warning threshold; generating a warning-level alarm when the deviation of a key parameter exceeds the early warning threshold but does not reach the critical alarm threshold; and generating an emergency-level alarm when the deviation of a key parameter exceeds the critical alarm threshold.

[0099] Specifically, the formula for calculating the synergistic deviation of the multidimensional key parameters is as follows: ,in, Assign weights to each parameter. For the fluctuation of each parameter within a specified time period, The threshold values ​​set for each parameter.

[0100] The system framework diagram is as follows Figure 3 Taking the fiber optic communication system shown as an example, a specific implementation is as follows: In the specific implementation of optical fiber communication systems, a method for real-time estimation of multiple parameters and traceable data in optical fiber link monitoring is realized through refined design for different application scenarios.

[0101] For the terrestrial backbone network, the monitoring device is deployed on a link using ITU-T G.652.D standard single-mode optical fiber. This fiber has a typical dispersion coefficient of 17 ps / (nm·km) and an attenuation coefficient of 0.2 dB / km in the 1550 nm window, providing a physical basis for parameter threshold setting. The device is connected to the main optical fiber link, and its input end receives the main link optical signal. The optical signal is first converted into four analog electrical signals by the receiving module, and then converted into four digital signals by a high-speed analog-to-digital converter module at a sampling rate of 64 GSa / s. The converted digital signals are simultaneously sent to two paths: one path enters a 16 GB high-bandwidth buffer module to continuously record the raw data with a storage depth of 50 ms, supporting high-fidelity backtracking; the other path is to a data processing module to extract multi-dimensional key parameters. The data processing module is composed of a field-programmable gate array (FPGA), which contains multiple digital signal processors that perform fully blind digital signal processing in parallel, extracting multi-dimensional parameters such as dispersion value, polarization state, frequency deviation, phase deviation, signal-to-noise ratio, and bit error rate in real time. In the dispersion value estimation part, the dispersion value scanning range is set to [-20000, 20000] ps / nm, with a step of 5 ps / nm.

[0102] Multidimensional key parameters (including dispersion value, polarization state, frequency deviation, phase deviation, signal-to-noise ratio, and bit error rate) are used synergistically through weighted correlation analysis. The synergistic weights are determined by the training data set. A dataset containing multidimensional key parameter vectors and corresponding fault type labels is used as the training samples. Fault types include fiber bending, joint degradation, and external interference. The neural network model has 6 nodes in the input layer, corresponding to 6 parameters, and 3 hidden layers (64 neurons per layer). When the real-time parameter vector is input into the trained model, it outputs a probability distribution. If the probability of a certain fault exceeds a threshold of 0.7, preliminary localization is triggered. For example, when the probability of "fiber bending" exceeds 0.7... At that time, the system flags potential bending faults.

[0103] The alarm analysis module implements three levels of alarms based on a dynamic threshold mechanism: 1. Alert-level alarm: When a single parameter is slightly abnormal, or the degree of coordination deviation And the neural network outputs the failure probability At this time, the system logs and marks trends, but does not trigger data backtracking. For example, if the dispersion value fluctuates briefly by about +20 ps / nm, but the polarization state is stable, the neural network outputs a "normal" probability of 0.9.

[0104] 2. Warning level alarm: When 2-4 parameters are abnormal, or the degree of coordination deviation And the neural network outputs the failure probability At that time, the original data 50 ms before and after the anomaly is retrieved, and the neural network initially locates the cause, providing guidance for data backtracking and initiating differential spectral analysis. For example, dispersion persists for 5 points exceeding ±50 ps / nm and the signal-to-noise ratio degrades by 3-5 dB.

[0105] 3. Emergency Alarm Level: When 5-6 parameters are severely abnormal, or the degree of coordination deviation is And the neural network outputs the failure probability In such cases, a highest-priority alarm is immediately sent to the network management system, triggering the data backtracking module to perform in-depth analysis. Combined with wavelet analysis, this directly guides fault location. Examples include: dispersion jumps exceeding ±100 ps / nm, signal-to-noise ratio degradation exceeding 7 dB, and polarization state change rate >10 krad / s.

[0106] The data backtracking module activates upon alarm triggering, using wavelet transform to accurately pinpoint the fault location and cause. This is achieved by analyzing the time corresponding to the extreme points of the wavelet coefficient amplitude. and scale Calculate the fault distance. If the extreme point appears... At +10 ms, the speed of light is 2 × 10 8 m / s, then the fault location is approximately 2 km from the monitoring point (accuracy ±2 m). Compare the time-frequency characteristics with the fault mode database: if the spectral energy is concentrated in the high frequency ( Small, short duration, matching the "fiber micro-bending" mode; if low-frequency oscillation ( The fault was characterized by large-scale, prolonged occurrence, matching the "loose joint" pattern. Combined with the preliminary neural network localization results (such as "joint deterioration"), the cause was further verified. For example, wavelet features showed low-frequency oscillations, with a 90% match to the joint failure patterns in the database, confirming joint deterioration as the cause of the fault.

[0107] This implementation case was validated in a 100 Gbps QPSK transmission experiment. The modular structure reduced deployment time to 15 minutes and lowered the bit error rate by two orders of magnitude. Through parameter coordination and neural networks, the false alarm rate for early warning levels was reduced by 40%; wavelet transform improved positioning accuracy to ±2 m, and the root cause identification accuracy reached 95%. In the implementation of the submarine communication system, the monitoring device adopted a distributed architecture, deploying multiple monitoring nodes along an ultra-long span of G.654.E large effective area optical fiber. Each node was spaced 100 km apart, forming a monitoring network covering the entire link. Each device synchronized key parameters to the central processing unit every 10 seconds via a communication protocol. The central unit built a comprehensive parameter database for collaborative analysis.

[0108] The device receives C+L band optical signals at its input, which are converted into four analog electrical signals (XI, XQ, YI, YQ) by the receiving module. These are then converted to digital signals via a high-speed analog-to-digital converter (ADC) at a sampling rate of 64 GSa / s, ensuring signal integrity. The converted digital signals are simultaneously fed into two paths: one path enters a 16 GB high-bandwidth buffer module to continuously record the original data at a depth of 50 ms, supporting high-fidelity backtracking; the other path is to a data processing module that extracts multi-dimensional key parameters. This data processing module is composed of a field-programmable gate array (FPGA), containing multiple digital signal processors that perform fully blind digital signal processing in parallel. The converted digital signals are then fed into the FPGA, which is equipped with 16 GB of high-bandwidth memory to buffer the original data for high-fidelity backtracking. The FPGA executes a fully blind digital signal processing algorithm to extract multi-dimensional parameters in real time, including dispersion value, polarization state, frequency deviation, phase deviation, signal-to-noise ratio (SNR), and bit error rate (BER). In the dispersion value estimation part, the dispersion value scanning range is set to [-10000, 10000] ps / nm. A segmented strategy is adopted to reduce the scanning time. First, a coarse adjustment step of 100 ps / nm is used to quickly find the interval where the dispersion value is located, and then a fine step of 5 ps / nm is used to accurately search.

[0109] Upon alarm triggering, the data backtracking module is activated to synchronously retrieve cached data from multiple devices. Fault location is achieved through wavelet transform: the time-frequency feature matrices of multiple devices are extracted, and the propagation delay of the fault signal is calculated using a cross-correlation algorithm. The time difference between the extreme points of the wavelet coefficients of multiple devices is then used to determine the fault location. Calculate the fault distance. For example, if node A detects an extreme point. =+10ms, node =+12 ms, the speed of light is 2×10 8 If the speed is m / s, the fault location is approximately 2 km from node A and 2.4 km from node B. Triangulation improves the accuracy to ±1 m. The time-frequency characteristics of multiple devices are compared with the fault mode database: if multiple devices display high-frequency, short-duration continuous characteristics, a "anchoring failure" pattern is matched; if low-frequency oscillations are long-duration, a "joint degradation" pattern is matched. The cause is further verified by combining the neural network output. For example, if the wavelet features match the anchoring failure pattern with 95% accuracy, and the pressure sensor reading is abnormal, the fault is confirmed to be anchoring failure.

[0110] This implementation case was validated in a 500 km submarine optical cable system. Multi-device collaboration reduced fault location time from 5 minutes per device to 2 minutes. Through multi-data fusion, the false alarm rate for warning-level alarms was reduced by 50%; wavelet transform combined with triangulation improved accuracy to ±1 m, and the root cause identification accuracy reached 98%.

[0111] The technical features of the embodiments described above can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. It should be noted that the terms "in one embodiment," "for example," and "again" in this invention are intended to illustrate the invention and are not intended to limit the invention.

[0112] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A fiber optic link monitoring device with multi-parameter real-time estimation and data traceability, characterized in that, Includes the following steps: The receiving module is used to convert the link optical signal into four analog signals; An analog-to-digital converter module is used to convert the four analog signals into four digital signals and store them in a buffer module; the four digital signals are XI, XQ, YI, and YQ, where XI is the real part of the X-polarized state, XQ is the imaginary part of the X-polarized state, YI is the real part of the Y-polarized state, and YQ is the imaginary part of the Y-polarized state. The data processing module is used to perform digital signal processing on the four digital signals and extract multi-dimensional key parameters; the key parameters include dispersion value, polarization state, frequency deviation, phase deviation, signal-to-noise ratio, and bit error rate; The alarm analysis module is used to judge multi-dimensional key parameters based on preset parameter thresholds to determine whether the current state triggers an alarm. The data backtracking module is used to retrieve cached data to determine the location and cause of the fault if an alarm is triggered.

2. The fiber optic link monitoring device as described in claim 1, characterized in that, The digital signal processing does not require training sequences or prior link information, and the processing procedure includes the following steps: Send the same PRBS analog signal as the four channels to the analog-to-digital conversion module, acquire the four digital sequences, take one of them as the reference channel, calculate the cross-correlation function between the other three channels and the reference channel, find the time delay that makes it reach the maximum value in each cross-correlation function, thereby obtaining the timing deviation of each channel relative to the reference channel, and calibrate the four digital signals according to the timing deviation. Dispersion values ​​are estimated for the four calibrated digital signals. The scanning range of the dispersion compensation value is set. The peak-to-average power ratio corresponding to the current dispersion compensation value is calculated sequentially according to the preset step. The dispersion compensation value corresponding to the maximum peak-to-average power ratio within the scanning range is the link dispersion value. The dispersion compensation value is used to perform dispersion compensation on the four digital signals. The clock error detected in the digital domain is fed back to the voltage-controlled oscillator in the analog domain to form a closed-loop control, adjust the local clock to zero error, and then perform channel equalization. Quantize the polarization state based on the adaptive equalizer coefficients obtained from channel equalization; The frequency deviation is obtained by calculating the phase rotation rate, the phase deviation is estimated by averaging multiple adjacent symbols, and the digital signal after channel equalization is compensated based on the frequency deviation and the phase deviation. The signal-to-noise ratio (SNR) is calculated by comparing the difference between the received signal and the ideal signal, and the bit error rate (BER) is calculated using the SNR.

3. The fiber optic link monitoring device as described in claim 1, characterized in that, The processing steps in the data backtracking module include: With wavelet function As a basis function, through scaling factor Translation factor Scaling and translating the mother wavelet yields a family of wavelet basis functions: ,in, It is the energy normalization factor; Cached signals The continuous wavelet transform is the inner product of the signal and the wavelet basis functions: The time-frequency features are extracted and compared with a known fault mode database to ultimately determine the specific location and cause of the fault.

4. The fiber optic link monitoring device as described in claim 1, characterized in that, The data processing module includes multiple digital signal processors (DSPs), each of which independently processes its assigned data block in real time. When the Kth DSP is processing the current data block, the programmable logic device allocates the next data block to the (K+1)th DSP, where K is an integer. The multiple DSPs output the processing results in real time.

5. The fiber optic link monitoring device as described in claim 1, characterized in that, The alarm analysis module also adjusts the thresholds of various key parameters used for status determination based on the fiber type of the target fiber link and the service requirements it undertakes; performs correlation analysis on multi-dimensional key parameters to locate link faults; and generates multiple alarm alerts of different levels based on the results of the correlation analysis.

6. The fiber optic link monitoring device as described in claim 5, characterized in that, The association analysis is implemented using a trained multilayer perceptron neural network model, and includes the following steps: Employing the aforementioned multi-dimensional key parameters ,in The number of parameters is given, and historical data is used as the training sample set. Each sample is labeled with the corresponding known state and fault type. The training sample set is input into a multilayer perceptron neural network model, wherein the first layer perceptron neural network model... Layer The output of each neuron is represented as: ,in It's weight. It's a bias. It's an activation function, using a damage function. Measuring the probability distribution of predictions Distribution of real labels The differences between them, With the goal of minimization, the weights and biases are iteratively optimized until the model converges; The final output is transformed into a probability distribution vector. ,in This represents the total number of link failure types. This indicates that the current link state is at the [number]th [level]. The probability of a class; The multi-dimensional key parameter vectors acquired in real time are input into the model, and the output probability distribution vector is determined to be the current fault type.

7. The fiber optic link monitoring device as described in claim 5, characterized in that, The generation of multiple alarm signals includes: setting early warning thresholds and critical alarm thresholds; generating a prompt-level alarm when the deviation of a key parameter is lower than the early warning threshold; generating a warning-level alarm when the deviation of a key parameter exceeds the early warning threshold but does not reach the critical alarm threshold; and generating an emergency-level alarm when the deviation of a key parameter exceeds the critical alarm threshold.

8. The fiber optic link monitoring method and apparatus as described in claim 7, characterized in that, The formula for calculating the co-discrepancy of multidimensional key parameters is as follows: ,in, Assign weights to each parameter. For the fluctuation of each parameter within a specified time period, The threshold values ​​set for each parameter.

Images