A method for time delay equalization of a large-capacity backbone optical communication link based on wavelength division multiplexing

Abstract

The application discloses a kind of based on large-capacity backbone optical communication link delay equalization method of wavelength division multiplexing, belong to optical communication technical field, comprising the following steps: S1, collection WDM backbone link on optical domain, electric domain, environment and nonlinear parameter, after pre-processing, upload to control center;S2, construct delay prediction model, utilize the data of collection and train model, the delay variation trend and delay difference of future each wavelength channel are predicted in real time, if the delay difference of prediction exceeds service threshold then trigger equalization decision;S3, with minimum delay difference, reduce energy consumption, satisfy service demand as goal, generate optical domain-electric domain-cross node collaborative equalization strategy;S4, according to equalization strategy, respectively carry out optical domain equalization, electric domain equalization and cross node collaborative scheduling;S5, carry out equalization effect feedback and adaptive optimization.The application has significantly improved equalization precision and reduced system energy consumption, and has strong dynamic adaptability.

Description

Technical Field

[0001] This invention relates to the field of optical communication technology, and in particular to a delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing. Background Technology

[0002] With the rapid popularization of services such as 5G enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (uRLLC), backbone optical communication networks face the dual challenges of "high-capacity transmission" and "low-latency jitter." Wavelength division multiplexing (WDM) technology, by transmitting multiple wavelength channels in a single optical fiber to multiply capacity, has become a core technology for backbone networks. However, existing WDM backbone links suffer from significant latency imbalances in actual operation, mainly due to the following factors: Wavelength-dependent differences: When different wavelength channels propagate in optical fibers, the time delay varies nonlinearly with wavelength due to the effects of group velocity dispersion (GVD) and polarization mode dispersion (PMD), and the time delay difference continues to accumulate as the transmission distance increases. The equalization methods are limited: existing technologies mostly rely on single optical domain compensation (such as fixed dispersion compensation modules) or electrical domain digital signal processing (DSP). Optical domain compensation lacks flexibility, while electrical domain processing faces bottlenecks of high energy consumption and high complexity, and the two lack coordinated scheduling. Dynamic response lag: Dynamic fluctuations in backbone network traffic (such as tidal effects) and changes in ambient temperature can cause real-time changes in latency characteristics. Traditional static balancing strategies cannot adapt quickly and are prone to excessive latency jitter.

[0003] Existing technologies, such as delay tuning techniques based on fiber Bragg gratings (FBGs), can only provide static compensation for specific wavelengths, making them difficult to adapt to dynamic multi-wavelength scenarios. While reinforcement learning-based spectrum allocation algorithms focus on delay optimization, they lack a multi-dimensional equalization mechanism encompassing optical, electrical, and cross-node aspects. MIMO-DSP equalization techniques, limited by computational complexity, exhibit significant energy consumption and delay trade-offs in high-capacity scenarios. Therefore, a high-capacity WDM backbone delay equalization method that balances accuracy, dynamism, and low energy consumption is urgently needed to overcome the bottlenecks of existing technologies. Summary of the Invention

[0004] The purpose of this invention is to provide a delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing (WDM). This method aims to solve the technical problems of low delay equalization accuracy, poor dynamic adaptability, high energy consumption, and difficulty in adapting to high-capacity transmission scenarios in existing WDM backbone optical communication links, and to achieve the unification of precise delay difference control, dynamic adaptive scheduling, and low-energy operation.

[0005] To achieve the above objectives, this invention provides a delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing, comprising the following steps: S1. Collect optical, electrical, environmental, and nonlinear parameters on the WDM backbone link, and upload them to the control center after preprocessing. S2. Construct a latency prediction model based on the CNN-LSTM hybrid deep learning architecture, train the model using the data collected in step S1, and predict the latency change trend and latency difference of each wavelength channel in real time. If the predicted latency difference exceeds the service threshold, trigger the balancing decision. S3. Construct a reinforcement learning agent based on DDPG to generate a balanced strategy for optical-electrical-cross-node collaboration with the goal of minimizing latency difference, reducing energy consumption, and meeting business needs. S4. According to the equalization strategy, optical domain equalization is achieved through tunable optical delay lines and FBG arrays, electrical domain equalization is achieved through DSP time-domain / frequency-domain filtering, and collaborative scheduling is achieved through cross-node resource allocation and segmented equalization. S5 monitors the latency difference and transmission performance indicators after equalization in real time, feeds them back to the control center for analysis and processing, dynamically updates the prediction model parameters and equalization strategy, and adjusts the latency difference threshold to adapt to changes in services.

[0006] Preferably, the optical domain parameters collected in S1 are the optical domain transmission distance, link loss, PMD, PDL, center wavelength drift, and SNR; the electrical domain parameters collected are the electrical domain symbol rate, BER, time-domain pulse broadening, real-time flow rate, and flow type; the environmental parameters collected are the environmental temperature and vibration data; and the nonlinear parameters collected are the nonlinear SPM and XPM characteristic parameters. Preprocessing includes wavelet transform denoising and normalization of the acquired parameters.

[0007] Preferably, in S2, during the training of the delay prediction model, the Adam optimization algorithm is used to train the model with the goal of minimizing the mean square error between the predicted delay value and the measured delay value; an attention mechanism is introduced to strengthen the weights of the flow characteristics and nonlinear effect parameters, thereby improving the prediction accuracy. The formula for calculating the mean square error is: ; in, MSE Here, N is the mean squared error value, and N is the sample size. y i For the first i The measured time delay value of each sample. The model predicts the time delay value; The formula for calculating the weights of the attention mechanism is: ; in, For the first i Attention weights for each feature s i For the firsti The score value of each feature, K The total dimension of the input features. For feature vectors, V For the feature transformation weight matrix, b For bias vectors, w T This is the transpose of the weight vector; The formula for calculating the time delay difference between any two wavelength channels is: ; in, Let m be the prediction delay difference between the m-th and n-th wavelength channels. Let be the prediction delay for the m-th wavelength channel. Let be the prediction delay for the nth wavelength channel. like This triggers the balancing process. This is the time delay difference threshold.

[0008] Preferably, in S3, a multi-objective reward function calculation formula is designed with the goals of minimizing latency difference, reducing energy consumption, and meeting business requirements: ; ; ; ; in, R To enhance the total reward value of the learning agent, Optimize the reward weighting coefficient for latency difference. This is the energy consumption reward weighting coefficient. This is the weighting coefficient for business quality rewards. R d Optimize rewards for latency differences. R e As an energy consumption reward, R q As a reward for business quality, This represents the actual time delay difference after equalization. For the target delay difference, E a The actual energy consumption of the equilibrium process. E max For maximum allowable energy consumption, BER Bit error rate; uRLLC business weight =0.5, =0.2, =0.3; eMBB service weight =0.4, =0.3, =0.3; Based on the current link state and latency prediction results, the agent generates an initial equalization strategy through the Actor network; after evaluation by the Critic network, the final strategy is obtained through optimization. The Actor network output calculation formula is: ; in, a To balance the action vectors, s For state vectors, W a This is the weight matrix. b a denoted as the bias vector of the output layer of the Actor network, and tanh is the activation function.

[0009] Preferably, in S4, optical domain equalization is achieved through a tunable delay line and an FBG array as follows: Tunable optical delay line adjustment: Array-type tunable optical delay lines are deployed at each optical amplification site. Delay compensation is performed on wavelength channels with short delays based on the delay adjustment amount in the equalization strategy. The adjustment accuracy is 1 ps, and the adjustment range is 0-1000 ps. The adjustment amount is calculated using the following formula: ; in, This is the TODL adjustment amount. For target latency, This represents the current wavelength channel delay. FBG Array Auxiliary Compensation: A dispersion-compensated FBG array is deployed at the transmitting end for precise calibration of the fixed dispersion delay difference of a specific wavelength channel; according to the equalization strategy, the center wavelength of the FBG is adjusted through a temperature tuning module, and the tuning amount is calculated as follows: ; in, This is the tuning amount of the FBG center wavelength. D The fiber dispersion coefficient, L For transmission distance, The time delay difference is due to dispersion. c The speed of light; Optical amplification gain equalization: A gain flattening filter is introduced into the EDFA amplifier to dynamically adjust the amplification gain according to the loss characteristics of each wavelength channel.

[0010] Preferably, in S4, the electrical domain equalization achieved by DSP time-domain / frequency-domain filtering is specifically as follows: in the receiver DSP unit, according to the filtering coefficients in the equalization strategy, the time-domain equalization and frequency-domain equalization collaborative working mode is activated. Among them, the time-domain equalization eliminates inter-symbol interference through an FIR filter, and the output calculation formula is: ; In the formula, y ( n The output signal is the filtered signal. This refers to the number of taps in the FIR filter. w ( k ) represents the filter coefficients. x ( n- k () represents the input signal at time nk; Frequency domain equalization compensates for time delay differences at each frequency point in the frequency domain through FFT transformation. The formula for frequency domain response correction is: ; In the formula, X ( f () represents the frequency domain response of the input signal. H eq ( f This represents the balanced frequency domain response. Y ( f The frequency domain response after equalization is... h eq ( n This represents the equilibrium time-domain impulse response; Nonlinear effect compensation: The time delay disturbance caused by the nonlinear effects of SPM and XPM is compensated by executing a nonlinear equalization algorithm based on Volterra series through DSP.

[0011] Preferably, in S4, the collaborative scheduling is achieved through cross-node resource allocation and segmented balancing as follows: Global allocation of balanced resources: The control center dynamically allocates balancing tasks to nodes with lower loads based on the load status of the balancing devices on each node. Long-distance link segmented equalization: For backbone links with a transmission distance of more than 1000km, a segmented equalization strategy is adopted, which divides the link into several equalization segments. Each segment performs equalization operations independently, and the control center coordinates the equalization parameters of each segment. Fault redundancy equalization: When the equalization device of a certain node fails, the redundant node is automatically triggered to take over the equalization task.

[0012] Preferably, in S5, the time delay difference and transmission performance indicators after equalization are monitored in real time and fed back to the control center for analysis and processing, specifically as follows: The equalization delay difference and performance index data are fed back to the control center in real time and compared with the prediction results of step S2 and the equalization strategy objectives of step S3. Simultaneously, the total system energy consumption is calculated using the following formula: ; in, Ez To balance the total energy consumption of the devices, E TODL For the power consumption of tunable delay lines, E FBG For FBG array power consumption, E DSP For DSP processing power consumption, E node Energy consumption for cross-node scheduling; If the latency difference after equalization still exceeds the threshold, or the performance indicators fail to meet the standards, it is marked as an equalization failure scenario.

[0013] Preferably, in S5, the prediction model parameters and equilibrium strategy are dynamically updated, specifically as follows: Adaptive Model Parameter Update: The training dataset of the multi-field coupled time delay prediction model is periodically updated using feedback data, and the model parameters are retrained using the Adam optimization algorithm. The parameter update calculation formula is as follows: ; in, For the updated set of model parameters, The set of model parameters before the update. For learning rate, For first-order moment estimation, For second-order moment estimation, It is a tiny constant; Dynamic adjustment of equilibrium strategy: Based on feedback analysis results, adjust the reward function weights and action space range of the reinforcement learning agent; optimize the balanced resource allocation ratio of optical-electrical-cross-node.

[0014] Preferably, in S5, the latency difference threshold is dynamically adjusted based on real-time changes in the service type. The calculation formula for adjusting the latency difference threshold is: ; in, For the new latency difference threshold, The percentage of uRLLC business. This represents the proportion of eMBB business.

[0015] Therefore, the beneficial effects of the above-mentioned delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing in this invention are as follows: (1) This invention significantly improves the equalization accuracy and effectively reduces inter-symbol interference and bit error rate by using multi-dimensional perception and coupled modeling; (2) The present invention is based on the intelligent decision-making and real-time feedback mechanism of reinforcement learning, which can quickly adapt to dynamic scenarios such as traffic fluctuations, environmental changes, and link failures. It has strong dynamic adaptability and can meet the real-time balancing needs of large-capacity backbone networks. (3) This invention significantly reduces system energy consumption through optical-electrical-cross-node collaborative balancing and resource optimization allocation.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the steps of an embodiment of a delay equalization method for a large-capacity backbone optical communication link based on wavelength division multiplexing according to the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example 1: like Figure 1 As shown, this embodiment provides a delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing, including the following steps: S1. Collect optical, electrical, environmental, and nonlinear parameters on the WDM backbone link, and upload them to the control center after preprocessing.

[0021] Distributed optical sensing modules and electrical parameter acquisition units are deployed at the transmitting and receiving ends of the WDM backbone link and at optical amplification sites (EDFA / Raman amplifiers) every 10-20km. The sensing nodes and the core control center communicate through 5G bearer network or optical over-the-air (RoF) technology.

[0022] Optical domain parameter acquisition: Transmission distance and link loss data for each wavelength channel are acquired using an optical time domain reflectometer (OTDR); polarization mode dispersion (PMD) and polarization correlation loss (PDL) data are acquired using a polarization diversity receiver module; and center wavelength drift and signal-to-noise ratio (SNR) for each wavelength channel are monitored in real time using a spectral analyzer. The acquisition frequency is 100Hz.

[0023] Electrical Parameter Acquisition: At the opto-to-electric conversion unit (O / E), the symbol rate, bit error rate (BER), and time-domain pulse broadening data of each channel are acquired through a high-speed analog-to-digital converter (ADC); the real-time flow rate and flow type (such as eMBB / uRLLC) of each wavelength channel are acquired through the flow monitoring module to distinguish between burst flow and steady-state flow.

[0024] Environmental and nonlinear parameter acquisition: Environmental temperature and mechanical vibration data along the link are collected through temperature and vibration sensors; characteristic parameters of nonlinear effects such as self-phase modulation (SPM) and cross-phase modulation (XPM) are extracted based on digital signal processing (DSP).

[0025] The collected multi-dimensional data is denoised and normalized to remove outlier data points. The denoising process uses a wavelet transform algorithm, calculated as follows: In the formula, These are wavelet coefficients. x ( n (This refers to the original collected data.) For db4 wavelet basis functions, 2 j This is the scaling factor.

[0026] The normalized calculation formula is: In the formula, x norm For the normalized data, x The original data, x max and x min These are the maximum and minimum values ​​of the parameter, respectively.

[0027] The data is initially compressed by the edge computing nodes, and the key parameters (latency raw data, dispersion value, traffic characteristics, nonlinear parameters) are uploaded to the control center at a time interval of 1ms.

[0028] S2. Construct a latency prediction model based on the CNN-LSTM hybrid deep learning architecture. Train the model using the data collected in step S1. Predict the latency change trend and latency difference of each wavelength channel in real time. If the predicted latency difference exceeds the service threshold, trigger the balancing decision.

[0029] Based on the historical data collected in step S1 (including multi-dimensional parameters under different traffic scenarios, environmental conditions, and link states), a training dataset is constructed; the channel delay values ​​of each wavelength measured at the receiving end (calibrated by the clock synchronization module with an accuracy of ps) are used as labels to divide the dataset into a training set (70%), a validation set (20%), and a test set (10%).

[0030] A prediction model is constructed using a CNN-LSTM hybrid deep learning architecture, where: the CNN module (convolutional neural network) is used to extract the coupling relationship of spatial feature parameters such as wavelength dispersion and nonlinear effects; the LSTM module (long short-term memory network) is used to capture time series features such as dynamic changes in traffic and environmental drift; the model input is the multi-dimensional parameters preprocessed in step S1, and the output is the prediction delay value and delay difference threshold of each wavelength channel.

[0031] During the training of the latency prediction model, the goal is to minimize the mean square error between the predicted latency value and the measured latency value. The Adam optimization algorithm is used to train the model. An attention mechanism is introduced to strengthen the weights of traffic characteristics and nonlinear effect parameters to improve prediction accuracy. After training, the model prediction error is controlled within ±5ps.

[0032] The formula for calculating the mean square error is: in, MSE Here, N is the mean squared error value, and N is the sample size. y i For the first i The measured time delay value of each sample. The model predicts the time delay value; The formula for calculating the weights of the attention mechanism is: ; in, For the first i Attention weights for each feature s i For the first i The score value of each feature, K The total dimension of the input features. For feature vectors, V For the feature transformation weight matrix, b For bias vectors, w T This is the transpose of the weight vector.

[0033] The control center calls the trained prediction model, inputs the multi-dimensional parameters uploaded in real time in step S1, and predicts the latency change trend of each wavelength channel within the next 10-100ms. The latency difference between any two wavelength channels is calculated using the following formula: in, Let m be the prediction delay difference between the m-th and n-th wavelength channels. Let be the prediction delay for the m-th wavelength channel. Let be the prediction delay for the nth wavelength channel.

[0034] Set latency difference threshold (Dynamically adjusted according to business needs; uRLLC service threshold ≤ 20ps, eMBB service threshold ≤ 50ps); if the predicted latency difference exceeds the threshold, i.e. This triggers the equilibrium decision-making process.

[0035] S3. Construct a reinforcement learning agent based on DDPG, aiming to minimize latency difference, reduce energy consumption, and meet business requirements, and generate an equilibrium strategy for optical-electrical-cross-node collaboration, specifically: The decision objective is set as "minimize the channel delay difference of each wavelength + the lowest equalization energy consumption + meet the service delay requirements"; the decision state space is defined as (current delay difference, dispersion value of each wavelength, traffic type, equalization device working status), and the action space is (optical domain delay adjustment amount, electrical domain DSP equalization parameters, cross-node scheduling instructions).

[0036] Construct a reinforcement learning agent based on Deep Deterministic Policy Gradient (DDPG), where: the Actor network is responsible for generating continuous and balanced action instructions (such as the delay adjustment amount of the tunable delay line and the filtering coefficient of the DSP); the Critic network is responsible for evaluating the value of the action (based on the optimization effect of time delay difference and energy consumption cost).

[0037] A multi-objective reward function calculation formula was designed with the goals of minimizing latency difference, reducing energy consumption, and meeting business requirements: in, R To enhance the total reward value of the learning agent, Optimize the reward weighting coefficient for latency difference. This is the energy consumption reward weighting coefficient. This is the weighting coefficient for business quality rewards. Rd Optimize rewards for latency differences. R e As an energy consumption reward, R q As a reward for business quality, This represents the actual time delay difference after equalization. For the target delay difference, E a The actual energy consumption of the equilibrium process. E max For maximum allowable energy consumption, BER This represents the bit error rate.

[0038] In general, uRLLC service weights are defined. =0.5, =0.2, =0.3; eMBB service weight =0.4, =0.3, =0.3.

[0039] Based on the current link state and latency prediction results, the agent generates an initial equalization strategy through the Actor network. The Actor network output formula is: in, a To balance the action vectors, s For state vectors, W a This is the weight matrix. b a denoted as the bias vector of the output layer of the Actor network, and tanh is the activation function.

[0040] After evaluation by the Critic network, the final strategy was optimized, including: optical domain equalization instructions (adjustment amount of tunable optical delay line, center wavelength tuning parameters of FBG), electrical domain equalization instructions (time domain / frequency domain filtering coefficients of DSP, symbol timing calibration parameters), and cross-node coordination instructions (equal resource allocation scheme for adjacent optical amplification sites / optical cross connectors (OXCs).

[0041] S4. According to the equalization strategy, optical domain equalization is achieved through tunable optical delay lines and FBG arrays, electrical domain equalization is achieved through DSP time-domain / frequency-domain filtering, and collaborative scheduling is achieved through cross-node resource allocation and segmented equalization. Specifically: Optical domain equalization: Tunable optical delay line adjustment: Arrayed tunable optical delay lines are deployed at each optical amplification site. Delay compensation is performed on wavelength channels with short delays based on the delay adjustment amount in the equalization strategy. The adjustment accuracy is 1 ps, and the adjustment range is 0-1000 ps. Fast response (response time ≤ 10 μs) is achieved through a voltage control module. The adjustment amount is calculated as follows: in, This is the TODL adjustment amount. For target latency, This represents the current wavelength channel delay.

[0042] FBG array-assisted compensation: A dispersion-compensated FBG array is deployed at the transmitting end for precise calibration of the fixed dispersion delay difference of a specific wavelength channel. Based on an equalization strategy, the center wavelength of the FBG is adjusted via a temperature tuning module to achieve dynamic compensation of the dispersion delay, with a compensation accuracy of ±2 ps. The tuning amount is calculated as follows: in, This is the tuning amount of the FBG center wavelength. D The fiber dispersion coefficient, L For transmission distance, The time delay difference is due to dispersion. c It is the speed of light.

[0043] Optical amplification gain equalization: A gain flattening filter is introduced into the EDFA amplifier to dynamically adjust the amplification gain according to the loss characteristics of each wavelength channel, thereby avoiding additional time delay differences caused by gain imbalance.

[0044] Electric domain equalization: In the receiver DSP unit, the time-domain equalization and frequency-domain equalization co-working mode is activated according to the filtering coefficients in the equalization strategy.

[0045] Among them, the time-domain equalization eliminates inter-symbol interference through an FIR filter, and the output calculation formula is: In the formula, y ( n The output signal is the filtered signal. This refers to the number of taps in the FIR filter. w ( k ) represents the filter coefficients. x ( n- k ) represents the input signal at time nk.

[0046] Frequency domain equalization compensates for time delay differences at each frequency point in the frequency domain through FFT transformation. The formula for frequency domain response correction is: In the formula, X ( f () represents the frequency domain response of the input signal. H eq ( f This represents the balanced frequency domain response. Y ( f The frequency domain response after equalization is... h eq ( n ) represents the equilibrium time-domain impulse response.

[0047] Based on a high-speed clock synchronization module, the received symbols of each wavelength channel are time-aligned with a calibration accuracy of ps; for burst traffic scenarios, a fast timing calibration algorithm is activated with a response time ≤1μs. By executing a nonlinear equalization algorithm based on Volterra series using DSP, delay disturbances caused by the nonlinear effects of SPM and XPM are compensated, further reducing delay jitter.

[0048] Cross-node collaborative scheduling: Global allocation of balanced resources: The control center dynamically allocates balancing tasks to nodes with lower loads based on the load status of the balancing devices at each node (such as the adjustment margin of the tunable delay line and the DSP computing power utilization rate), so as to avoid overloading of a single node.

[0049] Long-distance link segmented equalization: For backbone links with a transmission distance of more than 1000km, a segmented equalization strategy is adopted, dividing the link into several equalization segments (each segment is 200-300km). Each segment performs equalization operation independently, and the control center coordinates the equalization parameters of each segment to avoid inter-segment delay accumulation.

[0050] Fault redundancy equalization: When the equalization device of a certain node fails, a redundant node is automatically triggered to take over the equalization task, ensuring that the equalization process is not interrupted and guaranteeing the stability of the link transmission.

[0051] S5. Real-time monitoring of latency difference and transmission performance indicators after equalization, feeding back to the control center for analysis and processing, dynamically updating prediction model parameters and equalization strategies, and adjusting latency difference thresholds to adapt to service changes. Specifically: Deploy a delay monitoring module at the receiving end to measure the equalized delay value of each wavelength channel in real time using a high-precision time-to-digital converter (TDC) and calculate the delay difference; at the same time, monitor performance indicators such as bit error rate (BER) and signal-to-noise ratio (SNR) to determine whether the equalization effect meets the service requirements.

[0052] The equalization delay difference and performance index data are fed back to the control center in real time and compared with the prediction results of step S2 and the equalization strategy objectives of step S3. Simultaneously, the total system energy consumption is calculated using the following formula: in, E z To balance the total energy consumption of the devices, E TODL For the power consumption of tunable delay lines, E FBG For FBG array power consumption, E DSP For DSP processing power consumption, E node Energy consumption for cross-node scheduling.

[0053] If the latency difference after equalization still exceeds the threshold, or the performance indicators fail to meet the standards, it is marked as an equalization failure scenario.

[0054] The training dataset of the multi-field coupling delay prediction model is updated regularly using feedback data. The model parameters are retrained using the Adam optimization algorithm to improve the prediction accuracy of the model under different link states. For balanced failure scenarios, the sample weights are increased to enhance the model's adaptability to complex scenarios.

[0055] The parameter update calculation formula is: in, For the updated set of model parameters, The set of model parameters before the update. For learning rate, For first-order moment estimation, For second-order moment estimation, It is a tiny constant.

[0056] Based on the feedback analysis results, the reward function weights and action space range of the reinforcement learning agent are adjusted; the balanced resource allocation ratio between optical, electrical, and cross-node domains is optimized. For example, in burst traffic scenarios, the weight of electrical domain balancing is increased to improve response speed; in long-distance steady-state transmission scenarios, the weight of optical domain balancing is increased to reduce energy consumption.

[0057] The latency difference threshold is dynamically adjusted based on real-time changes in the business type. The calculation formula for adjusting the latency difference threshold is as follows: in, For the new latency difference threshold, The percentage of uRLLC business. This refers to the proportion of eMBB services. For example, when the proportion of uRLLC services in the link increases, the latency difference threshold is automatically reduced to ensure low-latency service requirements; when the proportion of eMBB services is dominant, the threshold can be appropriately increased to optimize system energy consumption.

[0058] Therefore, the present invention adopts the above-mentioned delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing to achieve synchronous perception of wavelength, dispersion, nonlinear effects and traffic characteristics. It breaks through the limitations of traditional single parameter detection, can accurately predict delay differences under different traffic scenarios, avoid passive compensation, can dynamically allocate optical and electrical domain equalization resources, achieve cross-node delay collaborative compensation, and at the same time reduce system energy consumption.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing, characterized in that, Includes the following steps: S1. Collect optical, electrical, environmental, and nonlinear parameters on the WDM backbone link, and upload them to the control center after preprocessing. S2. Construct a latency prediction model. Use the data collected in step S1 to train the model and predict the latency change trend and latency difference of each wavelength channel in real time. If the predicted latency difference exceeds the service threshold, trigger the equalization decision. S3. Construct a reinforcement learning agent based on DDPG to generate a balanced strategy for optical-electrical-cross-node collaboration with the goal of minimizing latency difference, reducing energy consumption, and meeting business needs. S4. According to the equalization strategy, optical domain equalization is achieved through tunable optical delay lines and FBG arrays, electrical domain equalization is achieved through DSP time-domain / frequency-domain filtering, and collaborative scheduling is achieved through cross-node resource allocation and segmented equalization. S5 monitors the latency difference and transmission performance indicators after equalization in real time, feeds them back to the control center for analysis and processing, dynamically updates the prediction model parameters and equalization strategy, and adjusts the latency difference threshold to adapt to changes in services.

2. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: The optical domain parameters collected in S1 are the transmission distance, link loss, PMD, PDL, center wavelength drift, and SNR; the electrical domain parameters are the symbol rate, BER, time-domain pulse broadening, real-time flow rate, and flow type; the environmental parameters are the ambient temperature and vibration data; and the nonlinear parameters are the nonlinear SPM and XPM characteristic parameters. Preprocessing includes wavelet transform denoising and normalization of the acquired parameters.

3. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S2, during the training of the latency prediction model, the goal is to minimize the mean square error between the predicted latency value and the measured latency value. The Adam optimization algorithm is used to train the model. An attention mechanism is introduced to strengthen the weights of traffic characteristics and nonlinear effect parameters, thereby improving the prediction accuracy. The formula for calculating the mean square error is: ； in, MSE Here, N is the mean squared error value, and N is the sample size. y i For the first i The measured time delay value of each sample. The model predicts the time delay value; The formula for calculating the weights of the attention mechanism is: ； in, For the first i Attention weights for each feature s i For the first i The score value of each feature, K The total dimension of the input features. For feature vectors, V For the feature transformation weight matrix, b For bias vectors, w T This is the transpose of the weight vector; The formula for calculating the time delay difference between any two wavelength channels is: ； in, Let m be the prediction delay difference between the m-th and n-th wavelength channels. Let be the prediction delay for the m-th wavelength channel. Let n be the prediction delay for the nth wavelength channel. like This triggers the balancing process. This is the time delay difference threshold.

4. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 3, characterized in that: In S3, a multi-objective reward function calculation formula is designed with the goals of minimizing latency difference, reducing energy consumption, and meeting business requirements: ； ； ； ； in, R To enhance the total reward value of the learning agent, Optimize the reward weighting coefficient for latency difference. This is the energy consumption reward weighting coefficient. This is the weighting coefficient for business quality rewards. R d Optimize rewards for latency differences. R e As an energy consumption reward, R q As a reward for business quality, This represents the actual time delay difference after equalization. For the target delay difference, E a The actual energy consumption of the equilibrium process. E max For maximum allowable energy consumption, BER Bit error rate; uRLLC business weight =0.5, =0.2, =0.3; eMBB business weight =0.4, =0.3, =0.3; Based on the current link state and latency prediction results, the agent generates an initial equalization strategy through the Actor network; after evaluation by the Critic network, the final strategy is obtained through optimization. The Actor network output calculation formula is: ； in, a To balance the action vectors, s For state vectors, W a This is the weight matrix. b a denoted as the bias vector of the output layer of the Actor network, and tanh is the activation function.

5. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S4, optical domain equalization is achieved through tunable delay lines and FBG arrays as follows: Tunable optical delay line adjustment: Array-type tunable optical delay lines are deployed at each optical amplification site. Delay compensation is performed on wavelength channels with short delays based on the delay adjustment amount in the equalization strategy. The adjustment accuracy is 1 ps, and the adjustment range is 0-1000 ps. The adjustment amount is calculated using the following formula: ； in, This is the TODL adjustment amount. For target latency, This represents the current wavelength channel delay. FBG Array Auxiliary Compensation: A dispersion-compensated FBG array is deployed at the transmitting end for precise calibration of the fixed dispersion delay difference of a specific wavelength channel; according to the equalization strategy, the center wavelength of the FBG is adjusted through a temperature tuning module, and the tuning amount is calculated as follows: ； in, This is the tuning amount of the FBG center wavelength. D The fiber dispersion coefficient, L For transmission distance, The time delay difference is due to dispersion. c The speed of light; Optical amplification gain equalization: A gain flattening filter is introduced into the EDFA amplifier to dynamically adjust the amplification gain according to the loss characteristics of each wavelength channel.

6. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S4, the electrical domain equalization is achieved through DSP time-domain / frequency-domain filtering as follows: In the DSP unit at the receiving end, the time-domain equalization and frequency-domain equalization collaborative working mode is started according to the filtering coefficients in the equalization strategy. Among them, the time-domain equalization eliminates inter-symbol interference through an FIR filter, and the output calculation formula is: ； In the formula, y ( n The output signal is the filtered signal. This refers to the number of taps in the FIR filter. w ( k ) represents the filter coefficients. x ( nk () represents the input signal at time nk; Frequency domain equalization compensates for time delay differences at each frequency point in the frequency domain through FFT transformation. The formula for frequency domain response correction is: ； In the formula, X ( f () represents the frequency domain response of the input signal. H eq ( f This represents the balanced frequency domain response. Y ( f The frequency domain response after equalization is... h eq ( n This represents the equilibrium time-domain impulse response; Nonlinear effect compensation: The time delay disturbance caused by the nonlinear effects of SPM and XPM is compensated by executing a nonlinear equalization algorithm based on Volterra series through DSP.

7. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S4, collaborative scheduling is achieved through cross-node resource allocation and segmented load balancing, specifically as follows: Global allocation of balanced resources: The control center dynamically allocates balancing tasks to nodes with lower loads based on the load status of the balancing devices on each node. Long-distance link segmented equalization: For backbone links with a transmission distance of more than 1000km, a segmented equalization strategy is adopted, which divides the link into several equalization segments. Each segment performs equalization operations independently, and the control center coordinates the equalization parameters of each segment. Fault redundancy equalization: When the equalization device of a certain node fails, the redundant node is automatically triggered to take over the equalization task.

8. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S5, the time delay difference and transmission performance indicators after equalization are monitored in real time and fed back to the control center for analysis and processing. Specifically: The equalization delay difference and performance index data are fed back to the control center in real time and compared with the prediction results of step S2 and the equalization strategy objectives of step S3. Simultaneously, the total system energy consumption is calculated using the following formula: ； in, E z To balance the total energy consumption of the devices, E TODL For the power consumption of tunable delay lines, E FBG For FBG array power consumption, E DSP For DSP processing power consumption, E node Energy consumption for cross-node scheduling; If the latency difference after equalization still exceeds the threshold, or the performance indicators fail to meet the standards, it is marked as an equalization failure scenario.

9. The delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S5, the prediction model parameters and equilibrium strategy are dynamically updated as follows: Adaptive Model Parameter Update: The training dataset of the multi-field coupled time delay prediction model is periodically updated using feedback data, and the model parameters are retrained using the Adam optimization algorithm. The parameter update calculation formula is as follows: ； in, For the updated set of model parameters, The set of model parameters before the update. For learning rate, For first-order moment estimation, For second-order moment estimation, It is a tiny constant; Dynamic adjustment of equilibrium strategy: Based on feedback analysis results, adjust the reward function weights and action space range of the reinforcement learning agent; optimize the balanced resource allocation ratio of optical-electrical-cross-node.

10. A delay equalization method for high-capacity backbone optical communication links based on wavelength division multiplexing according to claim 1, characterized in that: In S5, the latency difference threshold is dynamically adjusted based on real-time changes in the service type. The calculation formula for adjusting the latency difference threshold is as follows: ； in, For the new latency difference threshold, The percentage of uRLLC business. This represents the proportion of eMBB business.

Images