A 200g psm8 optical module mt-fa protection device and method

Abstract

The application relates to a 200G PSM8 optical module MT-FA protection device and method, which is suitable for the application scene of connecting MT-FA optical fiber short jumpers of high-speed optical modules with multi-channel parallel output; wherein the device comprises a modeling unit, a confirmation unit, a regulation and control unit, a determination unit and an execution unit; the modeling unit constructs a channel stability evaluation tensor which integrates temperature, current, voltage and optical power parameters; the confirmation unit extracts a channel response sensitivity difference based on a periodic perturbation, and judges a thermal instability approaching state; the regulation and control unit generates a predicted driving inhibition curve to segmentally back off a driving current accordingly; the determination unit generates a data packet containing a light-out instruction and fault reporting according to a thermal response delay and heat recovery condition; and the execution unit is responsible for the specific implementation of the protection instruction set; the application can realize real-time identification and dynamic protection of the thermal instability risk of a high-speed optical channel, and improves the operation safety and reliability of the optical module.

Description

Technical Field

[0001] This invention relates to the field of optical fiber communication technology, and in particular to a 200G PSM8 optical module MT-FA protection device and method. Background Technology

[0002] In existing technologies, 200G PSM8 optical modules, as high-density parallel optical communication devices, are widely used in data centers and high-speed interconnect scenarios. They achieve high-speed transmission of multi-channel optical signals through MT-FA fiber optic patch cords. These optical modules typically employ parallel vertical-emitting laser arrays (VCSELs) with multi-core fiber array output interfaces to achieve simultaneous transmission of up to eight channels. With the continuous increase in transmission rate and power density, optical modules face higher thermal loads and driving stresses during operation, requiring effective protection mechanisms to ensure stable operation.

[0003] However, most existing protection mechanisms rely on single-point temperature detection and static threshold judgment, which cannot accurately reflect the dynamic thermal evolution of multiple transmission channels under complex operating conditions, nor can they quantitatively identify early signs of thermal instability. In addition, although some solutions introduce power control, their adjustment strategies are based on fixed preset curves, which are difficult to adapt to real-time changing thermal response characteristics, easily leading to over-protection or protection delay, and reducing system performance and reliability.

[0004] Therefore, there is an urgent need to propose a protection device and method that can be oriented towards multi-channel state fusion analysis, has disturbance confirmation capability and adaptive control strategy, so as to meet the stability guarantee requirements of high-bandwidth optical modules in complex operating environments. Summary of the Invention

[0005] This application provides a 200G PSM8 optical module MT-FA protection device and method to improve the operational safety and reliability of the optical module.

[0006] This application provides a 200G PSM8 optical module MT-FA protection device, including:

[0007] The modeling unit is used to construct a channel stability evaluation tensor based on the real-time temperature parameters, driving current parameters, bias voltage parameters and output optical power parameters of multiple emitted laser channels connected to the MT-FA fiber jumper in the optical module. The channel stability evaluation tensor is processed by nonlinear normalization transformation to handle the dimensional differences between parameters and achieve feature fusion at the same scale.

[0008] The confirmation unit is used to apply periodic micro-amplitude current perturbation to the channel stability evaluation tensor, and extract the channel response sensitivity difference based on the changes in the channel stability evaluation tensor before and after the perturbation. The channel response sensitivity difference is calculated by high-order gradient matrix projection and is used to generate the channel thermal instability approaching state.

[0009] The control unit is used to generate a predictive drive suppression curve based on the channel response sensitivity difference and the channel power consumption change rate recorded in the channel stability evaluation tensor, combined with the load history curve within the optical module's operating cycle. The predictive drive suppression curve defines multiple drive current back-off stages and corresponding delay timings.

[0010] The determination unit is used to perform the drive current back-off operation defined by the predicted drive suppression curve, detect the response delay time and heat recovery amplitude of the corresponding channel, and compare the detection results with the preset convergence criteria. If the convergence criteria are not met, a protection instruction set is generated. The protection instruction set includes channel light-off control instructions and fault reporting data packets.

[0011] An execution unit is used to receive and execute the protection instruction set.

[0012] This application provides a method for protecting the MT-FA of a 200G PSM8 optical module, including:

[0013] Based on the real-time temperature parameters, driving current parameters, bias voltage parameters, and output optical power parameters of multiple emitted laser channels connected to the MT-FA fiber jumper in the optical module, a channel stability evaluation tensor is constructed. The channel stability evaluation tensor is processed by nonlinear normalization transformation to handle the dimensional differences between parameters and achieve feature fusion at the same scale.

[0014] A periodic micro-amplitude current perturbation is applied to the channel stability assessment tensor, and the channel response sensitivity difference is extracted based on the changes in the channel stability assessment tensor before and after the perturbation. The channel response sensitivity difference is calculated by projection of a high-order gradient matrix and is used to generate the channel thermal instability approaching state.

[0015] Based on the channel response sensitivity difference and the channel power consumption change rate recorded in the channel stability evaluation tensor, combined with the load history curve during the optical module's operating cycle, a predictive drive suppression curve is generated. The predictive drive suppression curve defines multiple drive current back-off stages and corresponding delay timings.

[0016] After performing the drive current back-off operation defined by the predicted drive suppression curve, the response delay time and heat recovery amplitude of the corresponding channel are detected, and the detection results are compared with the preset convergence criteria. If the convergence criteria are not met, a protection instruction set is generated, which includes channel light-off control instructions and fault reporting data packets.

[0017] Execute the protection instruction set.

[0018] The beneficial effects of this application mainly include: (1) By constructing a channel stability evaluation tensor and introducing a nonlinear normalization transformation, the expression scale of heterogeneous parameters such as temperature, current, voltage and optical power is effectively unified, improving the comprehensiveness and accuracy of state evaluation. (2) By introducing periodic micro-amplitude current disturbances and high-order gradient matrix projection, the sensitivity of the channel to disturbances is accurately quantified, thereby enabling early identification before thermal instability occurs and providing stronger early warning capabilities. (3) By combining the power consumption change rate and load historical data to generate a predictive drive suppression curve, a phased and time-sequential drive current back-off strategy is supported, improving the adaptability and refinement of regulation and avoiding over-protection or response lag. (4) A closed-loop protection mechanism is provided, which realizes the light-off control and fault reporting through the execution instruction set, enhancing the timeliness and controllability of fault handling and effectively ensuring the long-term stable operation of the optical module in a high-speed and high-load environment. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a 200G PSM8 optical module MT-FA protection device provided in the first embodiment of this application.

[0020] Figure 2 This is a flowchart of a 200G PSM8 optical module MT-FA protection method provided in the second embodiment of this application. Detailed Implementation

[0021] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0022] The first embodiment of this application provides a 200G PSM8 optical module MT-FA protection device. Please refer to... Figure 1 This figure is a schematic diagram of the first embodiment of this application. The following is in conjunction with... Figure 1 The first embodiment of this application provides a detailed description of a 200G PSM8 optical module MT-FA protection device.

[0023] The 200G PSM8 optical module MT-FA protection device includes a modeling unit 101, a confirmation unit 102, a control unit 103, a judgment unit 104, and an execution unit 105.

[0024] Modeling unit 101 is used to construct a channel stability evaluation tensor based on the real-time temperature parameters, driving current parameters, bias voltage parameters and output optical power parameters of multiple emitted laser channels connected to the MT-FA fiber jumper in the optical module. The channel stability evaluation tensor is processed by nonlinear normalization transformation to handle the dimensional differences between parameters and achieve feature fusion at the same scale.

[0025] Modeling unit 101 is used to construct a channel stability evaluation tensor that reflects the thermal stability state of multiple laser emission channels in the optical module. Its operation process includes multiple sub-processes such as data acquisition, parameter alignment, dimension mapping, fusion calculation and tensor construction, ensuring that the generated evaluation tensor can not only realistically reflect the state of multi-channel optical and electrical parameters, but also has mathematical completeness and usability for subsequent confirmation analysis and driving control.

[0026] Specifically, the modeling unit 101 first collects four types of parameters corresponding to each emitted laser channel in real time from the control interface, drive circuit, and power monitoring circuit of the optical module. These parameters are: channel temperature parameter, channel drive current parameter, channel bias voltage parameter, and channel output optical power parameter. The temperature parameter is collected by a temperature-sensitive element integrated near each laser and converted to a standard value using an analog-to-digital converter. The drive current and bias voltage parameters are collected by current and voltage sensors at the driver output, and synchronous sampling is used to ensure alignment with the temperature parameter. The output optical power parameter is measured in real time by a miniature monitoring photodetector located at the front end of the MT-FA fiber array interface and input to the modeling unit processing logic through multiple A / D channels.

[0027] Because the aforementioned physical quantities have different units of dimension, resulting in differences in magnitude and inconsistent dynamic ranges, direct use would distort subsequent analysis models. Therefore, modeling unit 101 constructs a standardized model for each type of collected parameter, using the following nonlinear normalization transformation function for processing:

[0028] For the original physical parameters of any channel Construct the standard quantity after transformation The calculation formula is as follows:

[0029] ;

[0030] in, This is the historical average of this parameter for this channel. For historical standard deviation, This is the sensitivity coefficient for channel parameters, preset according to the parameter type, such as 0.2 for temperature, 0.5 for current, 1.0 for voltage, and 0.1 for optical power. This function can suppress the influence of extreme values, allowing all channels to be mapped to different physical parameter dimensions. Interval.

[0031] , and subscript This is used to represent a specific type of physical parameter item for a particular channel; it is a compressed representation of a combined index of channel number and parameter type. In constructing the channel stability evaluation tensor, multiple parameters from multiple emitted laser channels need to be collected, each... This represents one of the channel-parameter combinations. A more detailed explanation follows: Assume there are... Each laser emits a laser through a channel (e.g., 8 in the PSM8), and each channel acquires four types of physical parameters, including temperature, drive current, bias voltage, and output optical power.

[0032] So, the total number of processes needed is... One original physical parameter. Use a linear numbering. This represents each such channel-parameter pair. That is:

[0033] Indicates the first The original values ​​of each channel-parameter pair;

[0034] For example, if the first channel is channel 0, its temperature parameter can be expressed as: Its driving current is And so on; mapping rules can be defined: the first The first channel Each parameter corresponds to .

[0035] After completing the normalization process for all parameters, the modeling unit 101 uses the four normalized parameters of each channel as feature dimensions to form a model of the form:

[0036] ;

[0037] The channel stability evaluation tensor, in which This indicates the total number of laser emission channels in the current optical module (e.g., 8 for a PSM8 module). Each row corresponds to one channel, and each column corresponds to four types of characteristics: normalized temperature, drive current, bias voltage, and output optical power.

[0038] This tensor will serve as the input basis for subsequent disturbance confirmation, sensitivity analysis, and predictive control, and can support dynamic updates, incremental adjustments, and thermal stability trend monitoring in subsequent operations. To ensure data consistency, the modeling unit 101 automatically performs channel failure marking cleanup in each data update cycle. If any parameter of a channel fails to be acquired or exceeds the limit, that channel will be set to zero and marked as abnormal in the current cycle and will not participate in sensitivity analysis.

[0039] In summary, modeling unit 101 has constructed a complete multi-channel state modeling mechanism through high-frequency sampling, physical parameter fusion, nonlinear normalization and tensor mapping.

[0040] Furthermore, the modeling unit is specifically used for:

[0041] Based on the variation amplitude and sampling stability of the real-time temperature parameters, drive current parameters, bias voltage parameters and output optical power parameters during the historical operating cycle, a nonlinear normalization transformation method matching the dynamic characteristics of each parameter is selected. The nonlinear normalization transformation method includes a variety of preset function families, and the selected function is automatically switched according to the current rate of change during operation to improve the resolution of the normalization result in different parameter ranges.

[0042] Based on all parameter results after normalization, the joint change trend of multiple emission laser channels under multiple parameter dimensions is analyzed, and a channel state coordination graph structure is constructed. The channel state coordination graph structure is used to represent the degree of synchronization of multi-dimensional covariates between channels within a certain time window, and serves as a weighting factor to guide the fusion method of channel states in the subsequent tensor construction process.

[0043] The normalized multi-parameter feature vector of each emitted laser channel is mapped to an embedded space that has been structurally compressed and optimized. The construction direction of the embedded space is determined by the main direction of the edge weights in the inter-channel state collaboration graph structure, thereby enhancing its ability to express the association mode of other emitted laser channels while maintaining the integrity of the state information of each emitted laser channel.

[0044] The feature vectors of all emitted laser channels in the embedded space are combined into a channel stability evaluation tensor with a unified structure.

[0045] In the implementation of the MT-FA protection device for the 200G PSM8 optical module, the modeling unit's task is to collect physical parameters, including temperature, drive current, bias voltage, and output optical power, from multiple emitted laser channels in real time, and construct a multi-dimensional tensor structure that accurately reflects the stability of each channel's operating state based on these parameters. After parameter acquisition, the modeling process first normalizes various physical parameters to address the dimensional differences and scale inconsistencies between different physical quantities. Unlike traditional unified normalization function strategies, this device matches the most suitable nonlinear normalization transformation method for each type of parameter based on the dynamic change amplitude and sampling stability exhibited by each parameter over historical operating cycles.

[0046] This transformation method includes a pre-defined family of functions, such as logarithmic functions, exponential decay functions, hyperbolic tangent functions, or piecewise linear functions. The device dynamically adjusts the selected function based on the local rate of change and historical standard deviation of the currently collected data. For example, when the temperature changes steadily but the optical power fluctuates significantly, a more stable logarithmic normalization method is used for temperature, while a more responsive piecewise linear transformation is used for optical power. This ensures that the parameters maintain high resolution and strong discriminative power within their respective ranges throughout the normalization process. This dynamic function selection mechanism can significantly improve the robustness and accuracy of subsequent multidimensional modeling.

[0047] After normalizing all parameters, the system further analyzes the joint variation trends of multiple emitted laser channels across various parameter dimensions. To this end, the modeling unit constructs a channel-to-channel state coordination graph structure. Each node in this graph corresponds to an emitted laser channel, and each edge represents the degree of synchronization of the state covariates of two channels within a given time window, i.e., the similarity of their change directions and amplitudes across multiple parameter dimensions. The edge weights reflect the coupling strength between channels and can be calculated based on strategies such as dynamic time warping distance, cosine similarity, or local trend comparison. This graph structure not only describes the consistency of state evolution between channels but will also serve as a guiding factor in the subsequent tensor fusion stage, determining the weight allocation for information fusion between different channels.

[0048] Subsequently, the system maps the multidimensional normalized feature vector of each channel to an embedded space that has undergone structural compression optimization. This embedded space is not randomly generated but constructed based on the principal direction of the edge weights in the inter-channel state collaboration graph structure. By performing eigenvalue decomposition or principal component analysis on the edge weight matrix, the optimal dimension and direction of the embedded space can be obtained. This ensures that each channel not only retains its own key state information but also enhances its responsiveness to the changing patterns of other channels within this space. This mapping can minimize redundancy between channel states and highlight the channel state characteristics that significantly affect the overall stability of the system.

[0049] Finally, the system structurally combines the feature vectors of all channels in the embedding space to generate a channel stability evaluation tensor with a unified format. This tensor's dimensional design considers the number of channels, parameter types, and the time dimension, supporting subsequent analysis needs for system stability trends, local anomaly evolution, and cooperative response relationships. Since each element in the tensor has undergone nonlinear normalization, cooperative graph weighting, and embedding optimization during preprocessing, it can be directly used as the input basis for judging channel stability and triggering protection mechanisms, ensuring that the model's accuracy and real-time performance meet the stringent requirements of real-world deployment scenarios.

[0050] Taking a typical application scenario as an example, during long-term operation, the 3rd, 5th, and 7th emission laser channels of a 200G PSM8 optical module exhibited varying degrees of power drift and temperature increase trends. The modeling unit first collected real-time physical parameters of all eight emission laser channels at 100ms intervals. Each channel corresponds to a set of four-dimensional original parameter vectors, namely temperature, drive current, bias voltage, and output optical power. After accumulating sampling data over the past 10 minutes, the system obtained the historical mean and standard deviation of each parameter and dynamically assigned a nonlinear normalization method based on its changing trend. The temperature parameter of the 3rd channel changed slowly and the sampling was stable, so a logarithmic function normalization method was used. However, its output optical power changed drastically, so the system selected a piecewise linear function for transformation. Similarly, the drive current of the 5th channel fluctuated frequently, so a hyperbolic tangent function was used to compress its extreme value influence; for the 7th channel, due to the abrupt change in bias voltage at a specific time point, an exponential decay function was used for normalization.

[0051] After normalization, each channel corresponds to a four-dimensional feature vector. The system then compares the changing trends between channels over the most recent 30 time windows. Channels 3 and 5 show consistent changes in temperature and current dimensions, and the system assigns a high cosine similarity weight (e.g., 0.82) to the edge between them. Channels 3 and 7, however, only exhibit synchronous fluctuations in the optical power dimension, resulting in a lower edge weight (e.g., 0.35). Based on this level of coordination, the constructed inter-channel state coordination graph structure weight matrix is ​​fed into the embedding space construction process. The system uses principal component analysis to extract the first principal direction of the edge weight matrix as the principal axis of the embedding space. Within this space, the feature vectors of all channels are compressed into a unified six-dimensional embedding vector through linear projection, preserving the maximum amount of coordination information.

[0052] Finally, the system sequentially combines all embedded vectors according to channel number into a channel stability evaluation tensor of size [8,6], and assigns a tensor coupling factor to each row that matches the channel connection weights in the collaboration graph. This tensor is fed into the subsequent protection mechanism in real time as the basis for judgment, playing a crucial role in the subsequent triggering of channel confirmation and prediction of suppression curve calculation.

[0053] The confirmation unit 102 is used to apply periodic micro-amplitude current perturbation to the channel stability evaluation tensor, and extract the channel response sensitivity difference based on the changes in the channel stability evaluation tensor before and after the perturbation. The channel response sensitivity difference is calculated by projection of a high-order gradient matrix and is used to generate the channel thermal instability approaching state.

[0054] The function of the verification unit 102 is to extract the response sensitivity of each emitted laser channel to the disturbance by introducing a small disturbance and analyzing the trend of tensor changes, based on the channel stability evaluation tensor constructed by the modeling unit 101. Based on the result of this response sensitivity calculation, it determines whether the channel is in a critical approaching state of thermal instability. Its functionality involves specific operations such as disturbance signal injection, tensor change detection, difference calculation, high-order matrix projection transformation, and state assessment, and has a complete and repeatable logical flow.

[0055] In specific implementation, the verification unit 102 first uses the channel stability evaluation tensor from the modeling unit 101. This tensor contains four types of normalized characteristic values ​​for each channel in the current sampling period: temperature, drive current, bias voltage, and output optical power. The verification unit uses this tensor as the reference state before perturbation, and then injects a small current perturbation into each laser drive channel of the optical module. The perturbation amplitude is ±0.2 mA, the perturbation duration is set to 10 µs, and the perturbation period is controlled between 5 ms and 20 ms, which can be dynamically adjusted according to the actual thermal inertia of the module.

[0056] After the perturbation injection, the system immediately re-acquires the real-time physical parameters of the laser channel, updates the evaluation tensor, and obtains the perturbed state matrix. The verification unit takes the tensor difference before and after the perturbation as input, performs difference analysis on the four parameter vectors corresponding to each channel, and projects this difference vector onto a pre-trained high-order sensitivity direction matrix. This direction matrix is ​​the set of principal response directions in the four-dimensional feature space, denoted as […]. Extracted from historical stable operating data through principal component analysis (PCA) or singular value decomposition (SVD), it can map the dominant contribution relationship of each physical dimension to the system's instability trend.

[0057] For any channel The difference between the eigenvectors before and after the perturbation is The corresponding response sensitivity difference is calculated as follows:

[0058] ;

[0059] This is the feature vector before perturbation; this is the channel. At the moment before the disturbance is applied (i.e., the time point) The state feature vector is derived from the first state of the channel stability evaluation tensor constructed by the modeling unit. Line. This vector contains channels. The four normalized feature parameters are as follows:

[0060] Temperature normalized value (unit: dimensionless, range: 0~1);

[0061] Normalized value of drive current;

[0062] Normalized bias voltage value;

[0063] Normalized output optical power.

[0064] Sensitivity projection matrix transpose, It is The real matrix is ​​a predefined sensitivity direction transformation matrix. The transpose of the matrix. It is usually obtained by the following method:

[0065] For a large number of channel characteristic data samples collected during historical operation (with stable and unstable states labeled separately), principal component analysis (PCA) or singular value decomposition (SVD) is performed to extract the most discriminative direction vectors; alternatively, based on expert experience, response weights for each dimension (such as temperature change dominating thermal instability) can be set to construct an orthogonal direction matrix.

[0066] Transposed to form ,Will Mapping to the "sensitivity space" enables different types of physical responses to have a unified metric.

[0067] This is the perturbed feature vector, which represents the channel. A short time interval after the disturbance is applied After (e.g.) The state feature vector obtained by resampling (=10 μs) is composed in the same way as... The similarity stems from the synchronous measurement of the channel state after the disturbance in the confirmation unit. This ensures complete consistency with the vector before the disturbance in both dimension and standardization method.

[0068] in, That is, the channel The response sensitivity difference is a scalar that represents the overall thermo-optical response intensity of the channel to perturbations.

[0069] The confirmation unit is based on each channel's Values ​​and empirically set thermal instability sensitivity thresholds When comparing, When this occurs, the channel is determined to be approaching a thermal instability state. In actual deployment, the threshold... It can be set according to the channel type, driving characteristics and historical statistical features. For example, the edge channel with poor stability can be set to 0.18, while the center channel can be set to 0.22 to ensure that it is not misjudged due to background noise.

[0070] Through the entire process described above, the confirmation unit 102 not only completes the early identification of the critical state of thermal instability in each channel, but also confirms the critical state of thermal instability in each channel. The value and the result of the approach state determination are passed to the control unit 103 as a structured output to drive the generation of the prediction-driven suppression curve. This unit has the advantages of strong real-time performance, controllable perturbation amplitude, and standardized output form. It can operate synchronously in a multi-channel parallel structure without interfering with the normal transmission of optical signals.

[0071] Furthermore, the confirmation unit is specifically used for:

[0072] Based on the rate of change of temperature parameters, driving current parameters and output optical power parameters of multiple laser emission channels within a preset operating cycle, and combined with the thermal inertia characteristics of each laser emission channel, the injection frequency and disturbance amplitude of periodic micro-amplitude current disturbances are dynamically set.

[0073] Under the periodic micro-amplitude current disturbance, the normalized parameters of each emitted laser channel are collected for disturbance response, forming the channel stability evaluation tensor change value before and after the disturbance, and the channel disturbance response difference vector is constructed based on the change value.

[0074] The channel perturbation response difference vector is projected using a higher-order gradient matrix. This higher-order gradient matrix is ​​constructed from the historical response features of the channel stability evaluation tensor through principal component direction extraction and is used to generate the principal response projection value under the perturbation direction.

[0075] Based on the consistency index between the main response projection value and the disturbance direction, non-thermal disturbance components are eliminated to obtain the channel response sensitivity difference, which is used to generate the channel thermal instability approach state.

[0076] In this embodiment, in order to realize the function of the "confirmation unit" in the MT-FA protection device of the 200G PSM8 optical module, it is necessary to ensure that the unit can adaptively set appropriate disturbance strategies for different operating states of the emitted laser channels, accurately evaluate the thermal response differences caused by disturbances, and finally form a highly reliable channel response sensitivity difference as the basis for subsequent thermal instability judgment.

[0077] First, before implementing periodic micro-amplitude current perturbations, the verification unit needs to construct a basic parameter analysis model based on the historical operating data of each emitted laser channel within a preset operating cycle, focusing primarily on three physical dimensions: temperature parameters, driving current parameters, and output optical power parameters. Each parameter is a dynamic time-series signal, acquired through a sampling mechanism with sampling frequencies ranging from 1ms to 10ms, forming a multi-dimensional time-series dataset indexed by timestamps. To improve the targeting of subsequent perturbation strategies, this time-series data needs to calculate the rate of change of each channel in each parameter dimension, i.e., the differential or sliding window slope estimation results. Common implementation methods include Savitzky-Golay filtering fitting, three-point difference method, or local regression analysis. Based on this, the thermal inertia characteristics of each channel also need to be evaluated. Thermal inertia reflects the channel's response delay to temperature changes after current changes, and can be measured by analyzing the time lag of the temperature change inflection point after a current step event. Typical thermal inertia ranges from 30ms to 150ms.

[0078] Based on the above analysis, the verification unit will set the injection frequency and amplitude of periodic micro-current disturbances for different channels. The disturbance frequency needs to match the thermal inertia of the channel; it cannot be too high, preventing heat accumulation, nor too low, affecting real-time performance. For example, when the channel thermal inertia is 80ms, the disturbance frequency should be set to once every 200ms. The disturbance amplitude needs to be small enough to avoid affecting the service signal, but still trigger an observable thermal response. It is usually set between ±1% and ±3% of the drive current amplitude and can be further adaptively fine-tuned based on the current stable state of the channel. It is recommended to use an alternating positive and negative sawtooth disturbance sequence to improve the resolution of response differences while ensuring that the total disturbance energy does not accumulate.

[0079] Once the current disturbance takes effect, the verification unit will collect parameter responses for each emitted laser channel. The data collected here are already normalized parameters processed by the modeling unit; that is, the temperature parameters, drive current parameters, bias voltage parameters, and output optical power parameters have undergone nonlinear normalization transformation and been mapped to a uniform scale. A set of time points is sampled before and after the disturbance (e.g., 5 data points are collected before and after the disturbance), and the change value of the channel stability evaluation tensor is calculated, which is called the tensor variation value before and after the disturbance. This variation value can be regarded as a difference vector in the multidimensional normalized parameter space, representing the direction and intensity of the overall parameter response of the current channel under the disturbance.

[0080] To further improve the discrimination accuracy of channel responses, the verification unit performs a high-order gradient matrix projection operation on the disturbance response difference vector of each channel. This high-order gradient matrix is ​​not manually set but automatically constructed from historical data. Its generation method is as follows: First, for all normalized parameters recorded during the historical operation cycle of a certain channel, multiple channel stability evaluation tensor snapshots are generated using a sliding window approach. Then, all tensor snapshots are uniformly stretched into a set of column vectors, forming a multi-dimensional feature space. Next, principal component analysis (PCA) is used to extract the main change directions in this feature space, constructing a transformation basis matrix with the eigenvectors as columns, i.e., the high-order gradient matrix.

[0081] After obtaining the high-order gradient matrix, the aforementioned disturbance response difference vector is projected onto the space spanned by this matrix to obtain the main response projection value. This main response projection value can be understood as the degree of response of the channel in the historically highly sensitive direction. Since different fault types exhibit different disturbance patterns in the parameter space, especially thermal disturbances and load disturbances which have different directions, the verification unit needs to further determine whether the response is dominated by thermal response. Therefore, the system will perform consistency analysis based on the angle or cosine similarity between the main response projection value and the original disturbance direction vector. Generally, if the angle is small (e.g., less than 30 degrees), it indicates that the current response highly matches the disturbance injection direction and belongs to the thermal response category; otherwise, it is considered that the response is greatly affected by other noise or load disturbances, has no discriminative significance, and will be rejected.

[0082] After the above processing, the retained disturbance response is defined as the channel response sensitivity difference, which is a scalar. The larger the value, the more susceptible the channel is to thermal instability excitation. This value will be directly used as the basis for determining the channel's thermal instability approach state and can also be used as the input for generating the predictive drive suppression curve in the subsequent control unit.

[0083] To ensure the operability of the above description, a specific embodiment is provided below.

[0084] Assume a 200G PSM8 optical module has a laser emission channel numbered as channel 3, with a current operating current of 60mA. The thermal inertia of this channel's temperature change during historical operation is approximately 90ms. The normalized parameters are: temperature 0.68, drive current 0.61, bias voltage 0.55, and output optical power 0.72. The system is configured to perform a micro-perturbation of this channel every 250ms with an amplitude of ±1.8%, i.e., an injection current change of ±1.08mA. Five sets of data are sampled before the perturbation begins and five sets of data are sampled after the perturbation, calculating the difference in the 10-dimensional channel stability evaluation tensor. This difference vector is then fed into a higher-order gradient matrix, which was previously constructed using PCA with 2000 historical tensor snapshots, where the cumulative contribution rate of the first three dimensions in the principal component directions reaches 91.3%.

[0085] The difference vector, after projection, yields a principal response value of approximately 0.23. This value has a cosine similarity of 0.96 with the perturbation direction vector, indicating a high degree of consistency between the response direction and the perturbation direction. Based on this, the system determines the response to be thermally induced and records the corresponding channel response sensitivity difference s3 as 0.23. If s3 continues to rise above 0.32 in subsequent cycles, the system marks it as a channel thermal instability approaching state and notifies the control unit to generate the corresponding predictive drive suppression curve.

[0086] In summary, the confirmation unit described in this embodiment constructs a complete process of disturbance injection, parameter acquisition, response extraction, projection discrimination, and sensitivity extraction.

[0087] The control unit 103 is used to generate a predictive drive suppression curve based on the channel response sensitivity difference and the channel power consumption change rate recorded in the channel stability evaluation tensor, combined with the load history curve during the optical module's operating cycle. The predictive drive suppression curve defines multiple drive current back-off stages and corresponding delay timings.

[0088] The function of the control unit 103 is to comprehensively judge the dynamic power consumption change characteristics of each channel based on the channel response sensitivity difference generated by the confirmation unit 102 and the channel stability evaluation tensor constructed by the modeling unit 101, and combine it with the load history curve of the optical module during the actual operation cycle to construct a predictive drive suppression curve for active drive current control. This suppression curve is not a simple threshold judgment or fixed strategy, but a time-continuous segmented control strategy generated for each channel approaching thermal instability. It has predictive, adaptive, and phased adjustment capabilities for the current operating condition, and can gradually adjust the drive behavior before thermal risks occur, thereby avoiding sudden light outages or ineffective protection.

[0089] In the specific implementation process, the control unit first obtains the channel response sensitivity difference value output by the confirmation unit. This value reflects the first The sensitivity of each channel to perturbations is assessed. Simultaneously, the control unit also utilizes the historical power consumption change rate of that channel, recorded in the tensor of the modeling unit. (Power consumption change rate) The calculation method is the rate of change of the product of the channel drive current and the output optical power within each cycle, that is:

[0090] ;

[0091] in, For the first Each channel in time The driving current, For the first Each channel in time 'output optical power' For the first Each channel in time 'output optical power' This represents the time interval between adjacent sampling periods. If this rate of change shows an accelerating upward trend in recent periods, it indicates that the heat dissipation of this channel is growing unstably, and adjustments should be made as soon as possible.

[0092] The control unit then calls up the historical load curves from the operating cycle, i.e., each channel in the first... The average load data over a period of time is defined as the cumulative output optical power per unit time. This historical data is stored in a time-series cache, allowing for moving average and trend fitting operations. Fitting methods typically employ exponential smoothing or multinomial regression to obtain the load trend within the prediction time window. Based on this, the control unit can predict the thermal risk for a period of time in the future.

[0093] Based on the above three inputs, the control unit generates the predicted drive suppression curve for this channel according to the following rules. First, the target drive current reduction range is set to be [percentage missing] of the original drive current. to And it is set to 3 to 5 discrete stages. For example, if the initial drive current is 8 mA, the control unit will generate, for example... Set the target value accordingly, and assign each a separate startup delay timing sequence, such as... These values ​​are based on and The joint scoring function is dynamically determined. This scoring function can be written as:

[0094] ;

[0095] in, This is a weighting factor that reflects the priority of response to sensitivity, power consumption slope, and load trends. For example, it can be set to... And the value can be adjusted based on testing.

[0096] Rating results These are mapped to a pre-defined current backoff strategy table, with each scoring interval corresponding to a backoff level and a start delay. For example:

[0097] No adjustment;

[0098] Slight rollback (Phase 1, delay 2 ms);

[0099] Moderate regression (3 stages, decreasing step by step);

[0100] Severe rollback (5 stages, 1 ms interval between each stage).

[0101] Ultimately, the predictive drive suppression curve output by the control unit is a set of time-driven control points, each point specifying a target drive current value and its execution time, for example:

[0102] ;

[0103] The curve is written to the driving buffer, which is then handed over to the decision unit for further execution and effect determination.

[0104] In summary, during actual implementation, the control unit 103 combines response sensitivity, power consumption change rate and historical load trend to generate a drive control curve with forward-looking and phased adjustment capabilities, and completes the control plan output through a clear data structure.

[0105] Furthermore, the control unit is specifically used for:

[0106] Based on the channel response sensitivity difference and the power consumption change rate recorded in each channel in the channel stability evaluation tensor, the channel stability evaluation tensor sequence of multiple emitted laser channels within a set sliding time window is extracted. The fluctuation trend function of each channel is calculated using a nonlinear time-weighted strategy. The fluctuation trend function is used to describe the joint evolution of different dimensional parameters in the local period.

[0107] The matching degree is calculated based on the fluctuation trend function and the load history curve of the corresponding channel, and a multi-channel driven risk aggregation graph structure is constructed. Each node in the graph structure represents the laser emission channel, each side represents the risk coupling relationship between channels, and the edge weight is assigned based on the risk synergy factor calculated by the channel power consumption change rate and sensitivity difference.

[0108] The drive risk aggregation graph structure is subjected to spectral clustering operation to extract the main direction projection features of drive risk, and the projection features are mapped to the suppression parameter domain to obtain the channel-level risk response index. The risk response index is used to define the hierarchical strategy of drive current back-off.

[0109] Based on the risk response index and its time change rate, the corresponding level of parameterized interpolation template library is called, and a channel-customized predictive drive suppression curve is constructed through a bidirectional variable slope spline function. The predictive drive suppression curve contains multiple drive current back-off stages and their corresponding delay sequences, which are used to apply progressive current regulation intervention to the emitted laser channel before the thermal instability approach state is fully triggered.

[0110] In the implementation of the MT-FA protection device for the 200G PSM8 optical module, the control unit is mainly used to perform predictive drive control based on multi-parameter time-series analysis results before the thermal instability risk fully emerges, thereby achieving active protection of the emitted laser channel. In this embodiment, the functional flow of the control unit closely depends on the channel response sensitivity difference and the power consumption change rate in the channel stability assessment tensor. Furthermore, by combining the channel load history curve and local trend analysis mechanism, a customized predictive drive suppression curve is generated for each channel.

[0111] First, based on the channel response sensitivity difference output by the aforementioned confirmation unit and the power consumption change rate in the channel stability evaluation tensor formed by the modeling unit, the control unit extracts the stability evaluation tensor sequence of the target emitted laser channel within a set sliding time window. This time window can be dynamically configured according to the heating cycle and sampling interval of the optical module, with a common configuration being 8 to 16 frames of tensor data continuously acquired within 1 second.

[0112] After extracting the tensor sequence within the time window, the control unit constructs a corresponding fluctuation trend function for each type of physical parameter (e.g., temperature, drive current, bias voltage, output optical power) for each channel. This function is constructed using a non-linear temporal weighting strategy. Specifically, the control unit assigns different weighting coefficients to the tensor instances of each frame within the time window. The weight distribution exhibits a time-decay characteristic, with frames closer to the current time having higher weights and frames further away having lower weights. For example, in a five-frame sequence, the weights might be set to 1.0, 0.8, 0.6, 0.4, and 0.2. This mechanism allows for a more sensitive capture of dynamic trends within the current time period while avoiding susceptibility to historical outliers.

[0113] Taking temperature as an example, assuming that the normalized temperature values ​​of a certain channel in the most recent five frames of the tensor are 0.52, 0.54, 0.58, 0.63, and 0.68 respectively, with the corresponding weights set as above, the weighted average trend value will be higher than the mean of 0.59, indicating that the channel temperature is in a stable upward trend. Simultaneously, the control unit further analyzes the rate of change of the temperature value between frames to determine whether it exhibits an accelerating upward trend. If so, the system will label this parameter dimension as having a "Level 3 Risk Upward Trend". Similar operations are also applied to power consumption parameters, optical power parameters, etc., ensuring that the local evolution of all parameters is accurately characterized. This trend label and trend strength will serve as input variables for subsequent risk modeling.

[0114] Once the fluctuation trend functions across multiple parameter dimensions are constructed, the control unit matches these functions with historical load curves. The historical load curves are constructed by recording the drive current and output optical power curves of each emitted laser channel over a relatively long period (e.g., 10 seconds), representing the typical thermal behavior pattern of that channel. The control unit calculates the matching degree between the current trend function and several feature templates in the historical curves to determine whether the channel is within a typical overheating risk evolution path.

[0115] The matching degree calculation result, together with the response sensitivity difference and power consumption change rate, is input into a multi-channel driven risk aggregation graph structure. Each node in this graph structure represents an emitted laser channel, and each edge represents the risk synergy relationship between two channels in the current cycle. Specifically, the edge weight is determined by the risk synergy factor calculated by jointly normalizing the response sensitivity difference and power consumption change rate of the two channels. The higher the synergy factor, the stronger the mutual influence between the two channels in the thermal instability evolution. For example, if two channels exhibit synchronous heating and synchronous power consumption surges in the last three cycles, their edge weights will be significantly increased.

[0116] After the risk aggregation graph structure is constructed, the control unit uses a spectral clustering algorithm to deconstruct the graph structure, extracting the principal direction feature vectors to form risk principal components. By projecting these principal components onto a predefined suppression parameter domain, the system can derive the risk response index for each emitted laser channel. This index reflects the thermal risk level of the channel after being subjected to multi-parameter perturbations in the current cycle; a higher value indicates that it is closer to the edge of thermal instability. To ensure system response sensitivity, this index also needs to be combined with its rate of change over time to determine whether the risk trend is rapidly approaching, slowly increasing, or gradually mitigating.

[0117] Based on the risk response index and its rate of change, the control unit calls the corresponding level of parameters into the interpolation template library. This template library pre-sets multiple drive back-off curves at different levels. The curve construction methods include, but are not limited to, bidirectional variable-slope spline interpolation functions, smooth transition functions based on Bézier curves, or empirical back-off curves constructed based on historical data. Each template contains multiple current back-off stages and their corresponding delay sequences, used to control the current adjustment amplitude and speed.

[0118] Finally, the system constructs a predictive drive suppression curve specific to this channel based on the selected template. This curve has the following characteristics: First, the current back-off is not a one-time large drop, but a gradual decrease in multiple stages, with clear time delay control between each stage to avoid affecting signal quality due to excessive intervention; Second, the back-off rate is proportional to the response exponent, and channels with higher exponents have steeper slopes in their suppression curves; Third, if the trend eases, the system can dynamically terminate the back-off process or switch to a maintenance-level suppression strategy to reduce unnecessary current reduction.

[0119] For example, a certain channel currently has a sensitivity difference of 0.88, a positive power consumption change rate greater than the threshold, and a trend function indicating a continuous and rapid increase in temperature and optical power, which highly matches the historical load curve. This also drives the channel to form a high-weighted edge with three other channels in the risk graph. In spectral clustering analysis, this channel is classified into a high-risk cluster, with a final risk response index of 0.93 and an increasing change rate. Based on this, the system selects the "red alert" template and performs a three-stage rollback, reducing the current by 5%, 8%, and 10% at 0.2 seconds, 0.6 seconds, and 1.0 seconds respectively, and inserting a small maintenance increase at 1.5 seconds to prevent over-rollback.

[0120] Through the complete process described above, the control unit realizes a predictive dynamic drive control mechanism that is not based on traditional static threshold control.

[0121] The determination unit 104 is used to perform the drive current back-off operation defined by the predicted drive suppression curve, detect the response delay time and heat recovery amplitude of the corresponding channel, and compare the detection results with the preset convergence criteria. If the convergence criteria are not met, a protection instruction set is generated. The protection instruction set includes a channel light-off control instruction and a fault reporting data packet.

[0122] The decision-making unit 104 plays a crucial role in the protection triggering and decision-making process within the entire protection device. Its core function is to execute the predicted drive suppression curve generated by the control unit 103, then perform thermal state convergence judgment based on the response behavior of each channel during the current back-off process, and determine whether to enter the protection state accordingly. The operation of this unit involves multiple steps, including drive control signal issuance, multi-physical quantity response sampling, thermal dynamic feature extraction, criterion judgment, and control command generation, forming a closed-loop protection decision-making process.

[0123] After the control unit 103 outputs the predicted drive suppression curve, the decision unit 104 first parses it into a set of drive current control commands, each command containing a target current value and an expected execution time. The decision unit sequentially sends these commands to the channel drive controller, causing each emitted laser channel to sequentially reduce its drive current according to a specified timing sequence. The current reduction behavior is typically handled by the DAC control module at the hardware level, with a specific accuracy controlled within ±0.05 mA. After each current backoff operation, the decision unit immediately initiates a fixed time window (e.g., 3 ms) for response observation.

[0124] Within this window, the decision unit reads two key response parameters of the channel in real time from the modeling unit or state buffer: one is the temperature change trend before and after current back-off, used to determine the effect of heat accumulation mitigation; the other is the recovery trend after power change before and after back-off, used to evaluate the interference effect of drive downsizing on output stability.

[0125] Specifically, the temperature response is measured in the form of a delay time, which is the time elapsed from the start of the current decrease to the appearance of the inflection point of the channel temperature curve. This delay time is denoted as the response delay time. The inflection point of the temperature curve is defined as "the time point at which the first derivative of the temperature sequence first changes from non-negative to negative and is greater than a specified threshold"; the temperature sampling period is set to 1ms; the start time of the decline is defined as the moment when the drive current control value first declines.

[0126] Power change is the rate of decrease in heat power per unit time. The calculation shows that the thermal power is approximately represented by the product of the drive current and the bias voltage.

[0127] The decision unit then observes each channel's... and Compare with the system's preset convergence criteria. Convergence criteria are empirically set threshold conditions, such as:

[0128] like and It is believed that the channel has generated effective heat release and the system is stabilizing;

[0129] like or If the thermal accumulation in the channel is not effectively suppressed, it may be in a stage of thermal instability evolution.

[0130] For channels that fail to meet the convergence criteria, the decision unit will immediately generate a protection instruction set. This instruction set consists of two parts: first, a channel shutdown control instruction, which notifies the execution unit to immediately shut down the laser of that channel to prevent further heating; second, a fault reporting data packet, which encapsulates information such as the channel number, decision time, drive current sequence, response delay time, and heat recovery magnitude in a structured form, and sends it to the host computer system or system management module for diagnosis and recording.

[0131] To ensure real-time performance, the decision unit employs interrupt-driven execution logic and is synchronized with the system clock, ensuring that the response analysis delay for each channel does not exceed 1 ms. All convergence decisions and instruction generation are completed in on-chip logic units, avoiding any impact on protection effectiveness due to communication or buffer delays.

[0132] In summary, the judgment unit 104 performs closed-loop verification of the driving current regulation effect through real-time response analysis, thermal dynamic characteristic evaluation and convergence logic comparison, and thereby realizes timely generation and accurate issuance of protection commands.

[0133] Furthermore, the determination unit is specifically used for:

[0134] After performing the drive current back-off operation defined by the predicted drive suppression curve, the temperature parameters, drive current parameters and output optical power parameters recorded in the channel stability evaluation tensor of each emitted laser channel are continuously collected. The multidimensional response trajectory within the corresponding delay time sequence is extracted. Based on the set delay time of the drive current back-off stage, a thermal response sequence containing the parameter change trends before and after intervention is constructed for each emitted laser channel.

[0135] Based on the transient coupling change between the driving current and the output optical power in the thermal response sequence, the thermal power recovery amplitude of each emitted laser channel within the delay time range is calculated. At the same time, combined with the response delay time of the channel temperature parameter, a set of physical response indicators for convergence judgment is generated. This set of physical response indicators is used as input to construct the physical convergence vector required for subsequent comparison and judgment.

[0136] The physical convergence vector is weighted, and the weighting is based on the sensitivity level of each channel in the channel response sensitivity difference. The power consumption change rate in the channel stability evaluation tensor is introduced as a dynamic adjustment factor to form a comprehensive state score for convergence judgment.

[0137] The comprehensive state score is compared with a preset convergence criterion, which is constructed based on the parameter response change rate and the heat power recovery ratio. A segmented mapping method is used to output a convergence state label. If the convergence state label indicates that the current state does not meet the convergence criterion, the judgment unit generates a protection instruction set. The protection instruction set includes a channel light-out control instruction and a fault reporting data packet, and feeds back the current physical convergence vector to the control unit to update the next cycle back-off strategy in the prediction drive suppression curve.

[0138] During the implementation of the MT-FA protection device for the 200G PSM8 optical module, the role of the judgment unit is to monitor and provide dynamic feedback on the effect of the predicted drive suppression curve generated by the aforementioned control unit in real time. Its core objective is to evaluate whether each emitted laser channel exhibits a state change trend that conforms to the thermal stability evolution law after the execution of the drive current back-off strategy, thereby determining whether further protection operations need to be initiated.

[0139] Specifically, after the control unit generates a customized predictive drive suppression curve for each channel based on the channel response sensitivity difference and the power consumption change rate recorded in the channel stability evaluation tensor, the decision unit continuously tracks the actual execution effect of this drive current back-off curve. Since the drive current back-off strategy may be gradual, involving adjustments at multiple stages and current levels, and each stage has a preset response delay time window, the decision unit first needs to collect the time series of temperature parameters, drive current parameters, and output optical power parameters for each emitted laser channel within the delay time range corresponding to each back-off stage, and construct a multi-dimensional thermal response sequence. This sequence not only reflects the parameter change trends before and after the current back-off stage but also preserves the correlation and evolution rhythm between parameters, serving as the basic data source for subsequent analysis and judgment.

[0140] For example, in a typical drive current back-off phase, the current of a certain emitted laser channel gradually decreases from 110 mA to 90 mA, with a set response delay time of 2 seconds. During this time, the decision unit collects temperature change curves and output optical power change curves from 0.5 seconds before the current decrease to 2.5 seconds after the decrease, thus forming a thermal response sequence covering the critical change window. This sequence not only records the direct impact of current changes but also indirectly reflects the channel's dynamic recovery capability under the thermal control strategy.

[0141] Next, the decision unit analyzes the coupling degree between the drive current and the output optical power based on the thermal response sequence. The "thermal power recovery amplitude" refers to the magnitude of whether the optical power value shows a significant rebound or tends to stabilize after the current recedes within a certain delay time. This parameter can be calculated by comparing the mean change and peak recovery rate of the power fluctuation curves before and after the disturbance, and has strong intuitive physical meaning. Simultaneously, the response delay time of the temperature parameter is also recorded, i.e., the length of time required for the temperature to show a downward trend after the current recedes. This time period can be identified by the transition point from positive to negative using the first derivative, or by statistically capturing the inflection point of the temperature curve, thus obtaining an indicator representing the channel's thermal inertia and heat dissipation efficiency.

[0142] These physical response metrics, including heat power recovery magnitude, temperature response delay time, and optical power stability recovery rate, constitute a set of physical response metrics. This set will be used to construct a physical convergence vector in subsequent steps. To enhance the ability of this convergence vector to characterize the overall system state, the decision unit assigns weights to each emitted laser channel based on its level in the aforementioned channel response sensitivity difference. That is, channels with higher sensitivity differences have a higher weight in terms of their impact on stability and contribute more to the convergence decision result. Furthermore, to enhance real-time response capabilities, the decision unit also introduces the power consumption change rate in the channel stability evaluation tensor as a dynamic adjustment factor. This factor is used as a coefficient to correct the weights, adapting to the dynamic changes in the load of different channels during actual operation.

[0143] After completing the weighted processing described above, the system obtains a comprehensive state score. This score reflects whether the current channel, after the intervention of the driving current, is evolving towards the ideal thermal response trend of "thermal power decrease—optical power recovery—temperature stabilization." The judgment unit then compares this score with a preset convergence criterion. The convergence criterion here is a thermal behavior evaluation model constructed based on engineering experience and large-sample statistics, and typically includes, but is not limited to, multiple conditions such as: temperature decrease rate not lower than a certain threshold, optical power recovery exceeding the minimum power recovery line, and thermal power retreat ratio maintained within a certain range. These conditions are combined into a total score through a weighted function. This score interval is segmented and mapped to multiple convergence state labels, such as: mild convergence, moderate fluctuation, and severe instability.

[0144] If the comparison results show that the current comprehensive status score cannot meet the convergence criterion, i.e., the current behavior of the channel does not have thermal stability characteristics, the determination unit will immediately generate a protection instruction set. This instruction set contains at least two components: first, a channel shutdown control instruction, which directly shuts down the currently emitting laser channel to stop the heat source and prevent further deterioration; second, a fault reporting data packet, which includes key fields such as the fault channel number, current status label, thermal response trajectory sampling data summary, the version number of the predictive drive suppression curve used, and the fault occurrence timestamp, so that the upper-level system can promptly locate the root cause of the problem and take system-level redundancy switching measures.

[0145] It is worth noting that, to achieve a closed-loop response for thermal stability control, the decision unit, while generating the protection instruction set, also feeds back the currently constructed physical convergence vector to the control unit for updating the predictive drive suppression curve for subsequent cycles. This feedback mechanism has extremely high engineering value, as it can continuously correct the trend prediction error of the control unit for the current state through feedback, realizing a three-in-one adaptive control path of "determination-correction-control". For example, if insufficient heat power recovery and a continuous rise in temperature are found after a current backoff, the system can infer that the previous current drop magnitude or response speed setting was inappropriate, and adopt a more aggressive backoff strategy in the next cycle, while pre-setting additional power consumption compensation templates or dynamic skipping mechanisms.

[0146] In summary, the implementation scheme of the judgment unit not only realizes the thermal response judgment of the driving current backoff effect, but also forms a complete and rigorous judgment system by constructing a set of physical response indicators, comparing the comprehensive state score with the convergence criterion, and has the ability to drive a prediction strategy with feedback, demonstrating a high degree of autonomous adjustment and safety control capabilities.

[0147] Here is a specific example:

[0148] Assuming a 200G PSM8 optical module is operating under high load, and the third emission laser channel (CH3) consistently shows an excessively high channel response sensitivity difference, the system, based on the predicted drive suppression curve generated by the aforementioned control unit, has implemented a gradual back-off strategy for the drive current of channel CH3. Specifically, this curve is set within a 5-second window, gradually reducing the drive current of CH3 from the initial 110 mA to 95 mA in three stages: Stage 1: 110 mA → 105 mA; Stage 2: 105 mA → 100 mA; Stage 3: 100 mA → 95 mA, with a delay of 1.5 seconds for each stage.

[0149] After the backoff current is applied in each stage, the decision unit needs to collect three parameters of CH3 within the corresponding delay time: temperature parameter, drive current parameter, and output optical power parameter. Taking stage 2 as an example, after the current decreases from 105 mA to 100 mA, the system collects the following data within the following 1.5 seconds:

[0150] Temperature change: slowly decreasing from 65.2°C to 64.7°C;

[0151] Optical power change: dropped from -1.5 dBm to -1.7 dBm and then rebounded to -1.3 dBm;

[0152] Drive current data: The actual value remains stable at 100 mA.

[0153] The system combines these three sets of time-series data to form the thermal response sequence for this stage, and calculates the thermal power recovery amplitude according to the process. In this embodiment, the determination unit recognizes that the output optical power rapidly recovers to a level higher than before the pullback (from -1.5 dBm to -1.3 dBm) within 1 second after the drive current decreases, indicating that the channel has a strong thermal power recovery capability.

[0154] Meanwhile, the system further analyzed the derivative change of the temperature curve and found that it reached an inflection point approximately 0.8 seconds after the rollback (i.e., the rate of temperature change changed from positive to negative). This value was recorded as the temperature response delay time. Based on this, the judgment unit extracted several physical indicators, including "heat power recovery amplitude = +0.2 dBm", "temperature response delay time = 0.8 s", and "fluctuation amplitude of optical power change after rollback = ±0.4 dBm", to construct a set of physical response indicators for this stage.

[0155] Then, the system labels CH3 as "highly sensitive" in the "channel response sensitivity difference" and assigns it a high weighting factor (e.g., 0.85). It also incorporates the power consumption change rate recorded in the "channel stability assessment tensor" (e.g., 4.2 mW / s) as a weighting adjustment coefficient to weight each parameter in the aforementioned physical index set, ultimately forming a comprehensive state score. The score is assumed to be 0.68 (out of 1.0, with higher scores indicating greater stability).

[0156] Next, the judgment unit compares the overall state score with the preset convergence criteria. These criteria may include conditions such as an overall score greater than 0.75, a temperature response delay less than 1 second, a heat power recovery exceeding +0.3 dBm, and optical power stability greater than ±0.2 dBm. In this example, the overall score is 0.68, the heat power recovery is +0.2 dBm, and the optical power fluctuates significantly; therefore, the current state clearly does not meet the convergence criteria.

[0157] Therefore, the determination unit marks the current CH3 state as "thermal behavior not converged," thereby triggering the generation of a protection instruction set. This instruction set includes the following:

[0158] Channel shutdown control command: Immediately shut down the laser emitting device of CH3 to prevent further overheating and device failure;

[0159] Fault reporting data packet: Package and send a data packet containing the following fields:

[0160] Fault channel number: CH3;

[0161] Current rating: 0.68;

[0162] Convergence criterion: Not converged;

[0163] Optical power trend data (curve sampling within 1.5 seconds);

[0164] Temperature variation data (including derivative analysis results);

[0165] Drive current back-off record;

[0166] The current version of the prediction-driven inhibition curve (e.g., V3.2);

[0167] The timestamp of the fault occurrence (e.g., 2025-06-02 14:23:17), etc.

[0168] Furthermore, after generating the protection instruction set, the system feeds back the currently constructed physical convergence vector (i.e., the aforementioned weighted state scoring factor) to the control unit. Based on this, the control unit updates the predictive drive suppression curve for the next cycle, optimizing it by, for example, increasing the initial backoff amplitude, switching the current stage earlier, or extending the buffer delay time, to avoid convergence failure again.

[0169] The execution unit 105 is used to receive and execute the protection instruction set.

[0170] The execution unit 105 is responsible for the final response and actual action in the entire 200G PSM8 optical module MT-FA protection device. Its main task is to receive the protection command set generated by the decision unit 104 and implement the control content in the command set at the hardware level, including extinguishing the laser emitted by the corresponding channel and sending structured fault information to the upper layer of the system. In order to ensure the closed-loop execution of the protection strategy and the operational safety of the optical module, the execution unit must have stable signal response capability, accurate channel identification mechanism, programmable control path and low-latency communication link, and its design and implementation must ensure that it can maintain the accuracy and independence of execution even when multiple channels trigger protection at the same time.

[0171] In practice, the execution unit maintains a high-speed data link connection with the decision unit through a communication interface, typically using a serial inter-chip bus such as SPI or I²C, or exchanging data within an integrated chip via shared registers and an on-chip bus. When the decision unit identifies a channel that does not meet the convergence criteria and generates a protection instruction set, this instruction set is written to a designated communication register area in the form of a data packet. The data packet includes the channel number, control type (e.g., light-off), control parameters (e.g., delay, hold time), and status code. The execution unit receives this instruction packet via polling or interrupt and immediately initiates the channel identification and execution mapping process.

[0172] The channel identification mechanism relies on a channel numbering system consistent with the modeling unit and drive control unit. For example, for the PSM8 module, the laser emission channels are numbered from 0 to 7, corresponding to eight independent current-driven output channels. The execution unit activates the dimming control path of a channel using the channel number specified in the instruction set. Electrically, dimming is achieved by setting the drive current value to 0 mA or less than the minimum threshold current required for the laser to maintain emission, for example, setting it to below 0.3 mA. In the driver chip implementation, this process is typically achieved by setting the corresponding DAC channel to zero or by turning off the drive switch via a control register.

[0173] After control execution is completed, the execution unit reports the fault status. The fault reporting data packet is initially constructed by the judgment unit, and the execution unit is responsible for adding a timestamp to it and sending it to the host computer system or the module's main control logic through the communication channel. The data packet is usually structured and includes fields such as protection trigger time, light-off channel, execution result flag (success / failure), system power supply voltage, and temperature snapshot. If sent via I²C or SMBus, the data packet can occupy multiple register address fields and is read by the external system using an auto-incrementing address method; if uploaded via SPI, it is encapsulated using a frame data protocol and includes a frame header, frame trailer, and checksum.

[0174] To ensure execution stability, the execution unit incorporates a state cache and timeout protection mechanism. If action confirmation is not completed within a specified period (e.g., 2 ms) after receiving an instruction, the system will resend the instruction or issue a local alarm signal. Furthermore, to avoid multi-channel instruction conflicts, the execution unit sorts multiple instruction packets arriving simultaneously by time and employs a priority queue strategy to execute them in the order of their trigger times.

[0175] In summary, the execution unit 105 not only strictly executes the protection instructions issued by the judgment unit, but also undertakes the task of system cascade action feedback. Its implementation fully considers the accuracy of channel control, the timeliness of action response, and the completeness of fault information.

[0176] In the above embodiments, a 200G PSM8 optical module MT-FA protection device is provided. Correspondingly, this application also provides a 200G PSM8 optical module MT-FA protection method. Please refer to... Figure 2 This is a flowchart illustrating an embodiment of a protection method for a 200GPSM8 optical module MT-FA according to this application. Since this embodiment, namely the second embodiment, is basically similar to the first embodiment, it is described simply; relevant details can be found in the description of the first embodiment. The method embodiments described below are merely illustrative.

[0177] The second embodiment of this application provides a method for protecting the MT-FA of a 200G PSM8 optical module, including:

[0178] Step S201: Based on the real-time temperature parameters, driving current parameters, bias voltage parameters and output optical power parameters of multiple emitted laser channels connected to the MT-FA fiber jumper in the optical module, a channel stability evaluation tensor is constructed. The channel stability evaluation tensor is processed by nonlinear normalization transformation to handle the dimensional differences between parameters and achieve feature fusion at the same scale.

[0179] Step S202: Apply periodic micro-amplitude current perturbation to the channel stability evaluation tensor, and extract the channel response sensitivity difference based on the changes in the channel stability evaluation tensor before and after the perturbation. The channel response sensitivity difference is calculated by high-order gradient matrix projection and is used to generate the channel thermal instability approaching state.

[0180] Step S203: Based on the channel response sensitivity difference and the channel power consumption change rate recorded in the channel stability evaluation tensor, combined with the load history curve within the optical module's operating cycle, a predictive drive suppression curve is generated. The predictive drive suppression curve defines multiple drive current back-off stages and corresponding delay timings.

[0181] Step S204: After performing the drive current back-off operation defined by the predicted drive suppression curve, the response delay time and heat recovery amplitude of the corresponding channel are detected, and the detection results are compared with the preset convergence criteria. If the convergence criteria are not met, a protection instruction set is generated. The protection instruction set includes channel light-off control instructions and fault reporting data packets.

[0182] Step S205: Execute the protection instruction set.

[0183] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A 200G PSM8 optical module MT-FA protection device, characterized in that, The method comprises the following steps: A modeling unit is configured to construct a channel stability evaluation tensor based on real-time temperature parameters, driving current parameters, bias voltage parameters and output optical power parameters of a plurality of emission laser channels connected to MT-FA optical fiber short jumpers in an optical module, and the channel stability evaluation tensor is processed by a nonlinear normalization transformation to handle dimensional differences between the parameters and realize the same scale feature fusion; A confirmation unit is configured to apply a periodic micro-amplitude current disturbance to the channel stability evaluation tensor, and extract a channel response sensitivity difference based on the variation of the channel stability evaluation tensor before and after the disturbance, wherein the channel response sensitivity difference is calculated by a high-order gradient matrix projection method to generate a channel thermal instability approaching state; A regulation unit is configured to generate a predicted driving suppression curve based on the channel response sensitivity difference and a channel power consumption change rate recorded in the channel stability evaluation tensor, in combination with a load history curve within an optical module operation period, wherein the predicted driving suppression curve defines a plurality of driving current rollback stages and corresponding delay timing; A determination unit is configured to detect a response delay time and a heat recovery amplitude of a corresponding channel after performing a driving current rollback operation defined by the predicted driving suppression curve, and compare the detection results with a preset convergence criterion, wherein if the convergence criterion is not met, a protection instruction set is generated, and the protection instruction set comprises a channel light-out control instruction and a fault reporting data packet; An execution unit is configured to receive and execute the protection instruction set.

2. The 200G PSM8 optical module MT-FA protection device according to claim 1, wherein, The modeling unit is specifically configured to: select a nonlinear normalization transformation mode matched with the dynamic characteristics of each parameter according to the variation amplitude and sampling stability of the real-time temperature parameters, driving current parameters, bias voltage parameters and output optical power parameters within a historical operation period, wherein the nonlinear normalization transformation mode comprises a plurality of preset function families, and the selected function is automatically switched according to the current variation rate during operation to improve the resolution capability of the normalization result in different parameter intervals; analyze the joint variation trend of a plurality of emission laser channels in a plurality of parameter dimensions based on all parameter results after normalization processing, and construct a channel state coordination graph structure, wherein the channel state coordination graph structure is used to represent the multi-dimensional covariant synchronization degree between channels within a certain time window, and is used as a weight factor to guide the fusion mode of each channel state in the subsequent tensor construction process; map the multi-parameter feature vector of each normalized emission laser channel to a structure-compressed embedding space, wherein the construction direction of the embedding space is determined by the main direction of the edge weight in the channel state coordination graph structure, so as to enhance the expression ability of each emission laser channel state information to other emission laser channel correlation patterns while maintaining the integrity of the state information of each emission laser channel; combine the feature vectors of all emission laser channels in the embedding space into a unified structure of the channel stability evaluation tensor.

3. The 200G PSM8 optical module MT-FA protection device according to claim 1, wherein, The confirmation unit is specifically configured to: Based on the change rates of the temperature parameters, the driving current parameters and the output optical power parameters of the multiple emission laser channels within a preset operation period, and in combination with the thermal inertia characteristics of each emission laser channel, the injection frequency and the perturbation amplitude of the periodic micro-amplitude current perturbation are dynamically set; Under the action of the periodic micro-amplitude current perturbation, the normalized parameters of each emission laser channel are collected for perturbation response, a channel stability evaluation tensor variation value before and after the perturbation is formed, and a channel perturbation response difference vector is constructed based on the variation value; The channel perturbation response difference vector is projected on a high-order gradient matrix, the high-order gradient matrix is constructed by extracting the principal component direction from the historical response characteristics of the channel stability evaluation tensor, and is used to generate a main response projection value in the perturbation direction; According to the consistency index of the main response projection value and the perturbation direction, the non-thermal type perturbation component is removed, and a channel response sensitivity difference is obtained, which is used to generate a channel thermal instability approaching state.

4. The 200G PSM8 optical module MT-FA protection device according to claim 1, wherein, The regulation unit is specifically used for: According to the channel response sensitivity difference and the power consumption change rate recorded in the channel stability evaluation tensor, a channel stability evaluation tensor sequence of the multiple emission laser channels within a set sliding time window is extracted, a fluctuation trend function of each channel is calculated by using a nonlinear time sequence weighting strategy, and the fluctuation trend function is used to describe the joint evolution form of different dimensional parameters within a local period; Based on the fluctuation trend function and the matching degree calculation of the corresponding channel load history curve, a multi-channel driving risk aggregation graph structure is constructed, each node in the graph structure represents an emission laser channel, each edge represents a risk mutual coupling relationship between channels, and the edge weight is valued based on a risk synergy factor calculated by combining the channel power consumption change rate and the sensitivity difference; Spectral graph clustering is performed on the driving risk aggregation graph structure, a driving risk main direction projection feature is extracted, and the projection feature is mapped to a suppression parameter domain to obtain a channel-level risk response index, which is used to define a hierarchical strategy of driving current rollback; According to the risk response index and its time change rate, a parameterized interpolation template library corresponding to the level is called, a prediction driving suppression curve customized for the channel is constructed by using a bidirectional variable slope spline function, the prediction driving suppression curve includes multiple driving current rollback stages and corresponding delay time sequences, and is used to apply gradual current regulation intervention to the emission laser channel before the thermal instability approaching state is completely triggered.

5. The 200G PSM8 optical module MT-FA protection device according to claim 1, wherein, The determination unit is specifically used for: After performing the driving current rollback operation defined by the prediction driving suppression curve, the temperature parameters, the driving current parameters and the output optical power parameters recorded in the channel stability evaluation tensor of each emission laser channel are continuously collected, a multi-dimensional response trajectory within the corresponding delay time sequence is extracted, a thermal response sequence including the parameter change trend before and after the intervention is constructed for each emission laser channel according to the set delay time of the driving current rollback stage; According to the transient coupling change of the driving current and the output optical power in the thermal response sequence, the thermal power recovery amplitude of each emission laser channel in the delay time range is calculated, and the physical response index set for convergence judgment is generated by combining the response delay time of the channel temperature parameter, which is used as input to construct the physical convergence vector required for subsequent comparison and judgment; The physical convergence vector is subjected to weighting processing, the weighting processing allocates weights according to the sensitivity level of each channel in the channel response sensitivity difference, and introduces the power consumption change rate in the channel stability evaluation tensor as a dynamic adjustment factor, and then forms a comprehensive state score for convergence judgment; The comprehensive state score is compared with the preset convergence criterion, the convergence criterion is constructed according to the parameter response change rate and the thermal power recovery ratio, and the convergence state label is output in a segmented mapping manner, if the convergence state label represents that the current state does not meet the convergence criterion, the protection instruction set is generated by the judgment unit, the protection instruction set includes channel light-out control instruction and fault reporting data packet, and the current physical convergence vector is fed back to the regulation and control unit for updating the next period rollback strategy in the predicted driving inhibition curve. 6.A 200G PSM8 optical module MT-FA protection method, characterized in that, Comprise: Based on the real-time temperature parameters, driving current parameters, bias voltage parameters and output optical power parameters of a plurality of emission laser channels connected to the MT-FA optical fiber short jumper in the optical module, a channel stability evaluation tensor is constructed, the channel stability evaluation tensor is processed by nonlinear normalization transformation to handle the dimensional difference between parameters and realize the same scale feature fusion; Periodic micro-current disturbance is applied to the channel stability evaluation tensor, and based on the change of the channel stability evaluation tensor before and after the disturbance, the channel response sensitivity difference is extracted, the channel response sensitivity difference is calculated by high-order gradient matrix projection method, and is used to generate the channel thermal instability approaching state; Based on the channel response sensitivity difference and the channel power consumption change rate recorded in the channel stability evaluation tensor, and combined with the load history curve in the optical module operation period, a predicted driving inhibition curve is generated, the predicted driving inhibition curve defines a plurality of driving current rollback stages and corresponding delay time sequences; After performing the driving current rollback operation defined by the predicted driving inhibition curve, the response delay time and the heat recovery amplitude of the corresponding channel are detected, and the detection results are compared with the preset convergence criterion, if the convergence criterion is not met, a protection instruction set is generated, the protection instruction set includes channel light-out control instruction and fault reporting data packet; The protection instruction set is executed.

Images