Method for driving laser diode analog modulation waveform

Abstract

The application belongs to the technical field of laser driving, and specifically discloses a method for driving an analog modulation waveform of a laser diode, which comprises the following steps: first, generating an initial composite analog modulation signal containing multiple frequency components; then, adjusting the amplitudes and phases of the components by a pre-compensator according to pre-compensation parameters, and driving the laser diode; collecting the driving current and / or voltage as an electrical feedback signal, and calculating an intermodulation distortion performance index through spectrum analysis; constructing and updating a Gaussian process regression model with the pre-compensation parameters as the input and the index as the output, and generating new pre-compensation parameters and updating them in combination with Bayesian optimization; and repeating the iteration until the index converges or the number of iterations reaches a limit. The application can actively suppress intermodulation distortion, guarantee the signal bandwidth and amplitude, does not require an additional optical detector, and improves the fidelity, safety and adaptability of laser diode modulation.

Description

Technical Field

[0001] This invention relates to the field of laser driving technology, specifically to a method for driving a laser diode to simulate a modulation waveform. Background Technology

[0002] In fields such as optical communication, spectral imaging, and precision ranging that rely on analog modulation using laser diodes, when a laser diode is modulated by multiple frequency components, different frequencies interact in their nonlinear regions, causing intermodulation distortion (IMD). This results in stray frequency components in the output optical signal, compromising the fidelity of the modulated waveform and severely impacting system performance.

[0003] There are two main traditional methods for suppressing intermodulation distortion: one is to use a bandpass filter to eliminate spurious spectra, but this will lead to waveform distortion; the other is to reduce the modulation depth or limit the frequency range, but at the cost of sacrificing system bandwidth and amplitude performance.

[0004] Existing technologies cannot actively counteract the distortion caused by nonlinear response; relying on fixed parameters or manual correction makes it difficult to adapt to individual differences in lasers and the drift of nonlinear characteristics caused by temperature and current changes; some solutions require additional deployment of optical detectors, increasing hardware complexity and cost; the optimization process lacks safety constraints, and devices are easily damaged due to improper parameter adjustment, which seriously restricts the practicality and safety of laser diode analog modulation. Summary of the Invention

[0005] The purpose of this invention is to provide a method for driving a laser diode to simulate a modulation waveform, so as to solve the problems mentioned in the background art.

[0006] According to one aspect of the present invention, a method for driving a laser diode to simulate a modulation waveform is provided, comprising the following method steps: S1: Generate an initial composite analog modulated electrical signal, which is synthesized from multiple frequency components with specific initial amplitude and initial phase; S2: The initial modulation signal is input to the pre-compensator. The pre-compensator applies an adjustment to the amplitude and phase of each frequency component in the initial modulation signal according to a set of pre-compensation parameters, and generates and outputs a pre-compensated drive signal. The pre-compensation parameters include the amplitude adjustment and phase adjustment for each frequency component. S3: Drive the laser diode using the pre-compensated drive signal; S4: Acquire the waveform of the driving current and / or driving voltage flowing through the laser diode as an electrical feedback signal; S5: Perform spectral analysis on the electrical feedback signal and calculate an intermodulation distortion performance index based on the analysis results; wherein, the intermodulation distortion performance index characterizes the sum of spectral energy at one or more preset intermodulation frequencies; S6: Using the pre-compensation parameter vector as input and the intermodulation distortion performance index as output, construct and update the Gaussian process regression model; S7: Based on the updated Gaussian process regression model, a new set of pre-compensation parameters is generated using a Bayesian optimization algorithm to minimize the intermodulation distortion performance index; S8: Update the pre-compensation parameters of the pre-compensator to the new pre-compensation parameters; S9: Repeat steps S2 to S8 until the intermodulation distortion performance index meets the preset convergence condition or reaches the preset number of iterations.

[0007] Preferably, in step S2, the amplitude adjustment for each frequency component is limited to a maximum positive or negative adjustment percentage range set based on the safe driving range of the laser diode, and / or, the phase adjustment for each frequency component is limited to a maximum positive or negative angle range set based on the safe driving range of the laser diode.

[0008] Preferably, the spectral analysis of the electrical feedback signal in step S5 specifically includes: windowing and fast Fourier transforming the electrical feedback signal to obtain its frequency amplitude spectrum. The calculation of the intermodulation distortion performance index is specifically as follows: for each frequency in the predetermined set of intermodulation frequencies to be suppressed, the squared value of its amplitude at the corresponding position in the frequency amplitude spectrum is multiplied by an optional weighting coefficient and then summed.

[0009] Preferably, in step S6, the Gaussian process regression model uses a combination of radial basis functions and white noise kernels as its kernel function. The radial basis functions are used to model the smooth nonlinear relationship between the pre-compensation parameters and the intermodulation distortion performance index, and the white noise kernel is used to characterize the noise in the observation data.

[0010] Preferably, in step S7, the Bayesian optimization algorithm uses an expected boost function or an upper confidence bound function as its acquisition function.

[0011] Preferably, step S7 includes a safety constraint check when generating new pre-compensation parameters. Specifically, it determines whether the driving signal generated using the new parameters will cause at least one of the peak value of the driving current, the average power, or the rate of change of the current of the laser diode to exceed its safety threshold. If so, a penalty term is applied or the parameter is directly rejected to ensure that the optimization process is carried out within the safety boundary.

[0012] Preferably, the training data used in step S6 to construct the Gaussian process regression model is initially obtained by collecting and evaluating multiple sample points within the allowed value space of the pre-compensation parameter vector using Latin hypercube sampling or grid sampling.

[0013] Preferably, when the number of frequency components is large, resulting in a high dimension of the pre-compensation parameter vector, a sparse Gaussian process regression model is used to replace the standard Gaussian process regression model in order to reduce computational complexity.

[0014] Preferably, the Bayesian optimization process in step S7 supports the parallel evaluation of multiple candidate parameter combinations to accelerate the optimization process.

[0015] In another aspect, this application also provides a system for driving an analog modulation waveform of a laser diode, comprising: An initial modulation signal generation module is used to generate an initial composite analog modulation electrical signal, which is synthesized from multiple frequency components with specific initial amplitude and initial phase. The pre-compensation module is used to input the initial modulation signal to the pre-compensator. The pre-compensator applies an adjustment amount to the amplitude and phase of each frequency component in the initial modulation signal according to a set of pre-compensation parameters, and generates and outputs a pre-compensated drive signal. The pre-compensation parameters include the amplitude adjustment amount and phase adjustment amount for each frequency component. A driving module for driving a laser diode using the pre-compensated driving signal; The feedback signal acquisition module is used to acquire the waveform of the driving current and / or driving voltage flowing through the laser diode as an electrical feedback signal. The performance index setting module is used to perform spectrum analysis on the electrical feedback signal and calculate an intermodulation distortion performance index based on the analysis results; wherein, the intermodulation distortion performance index represents the sum of spectral energy at one or more preset intermodulation frequencies; The model building module is used to construct and update the Gaussian process regression model with the pre-compensation parameter vector as input and the intermodulation distortion performance index as output. An optimization module is used to generate a new set of pre-compensation parameters based on the updated Gaussian process regression model using a Bayesian optimization algorithm to minimize the intermodulation distortion performance index. The parameter update module is used to update the pre-compensation parameters of the pre-compensator to the new pre-compensation parameters; The parameter determination module is used to repeatedly execute steps S2 to S8 until the intermodulation distortion performance index meets the preset convergence condition or reaches the preset number of iterations.

[0016] This invention constructs a closed loop of "pre-compensation-drive-feedback-modeling-optimization," combining Gaussian process regression modeling and Bayesian optimization to actively suppress intermodulation distortion in multi-frequency modulation of laser diodes. This significantly reduces stray frequency components and greatly improves waveform fidelity without sacrificing signal bandwidth and amplitude. It utilizes only drive current / voltage electrical feedback, eliminating the need for additional optical detectors and reducing hardware costs. Safety constraints are incorporated into the optimization process to ensure device safety. Furthermore, it supports parallel parameter evaluation and high-dimensional adaptation, effectively enhancing the practicality, safety, and adaptability of analog modulation of laser diodes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for driving a laser diode to simulate a modulation waveform, provided as an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the spectrum analysis process provided in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the model training process provided in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram illustrating the implementation process of the sparse Gaussian process regression model provided in an embodiment of the present invention.

[0022] Figure 5 This is a schematic diagram illustrating the implementation process of parallel optimization provided in an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of a system for driving an analog modulation waveform of a laser diode, provided as an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] like Figure 1 As shown, this embodiment of the invention discloses a method 100 for driving a laser diode to simulate a modulation waveform, comprising the following method steps: S1: Generate an initial composite analog modulated electrical signal, which is synthesized from multiple frequency components with specific initial amplitude and initial phase; S2: The initial modulation signal is input to the pre-compensator. The pre-compensator applies an adjustment to the amplitude and phase of each frequency component in the initial modulation signal according to a set of pre-compensation parameters, and generates and outputs a pre-compensated drive signal. The pre-compensation parameters include the amplitude adjustment and phase adjustment for each frequency component. S3: Drive the laser diode using the pre-compensated drive signal; S4: Acquire the waveform of the driving current and / or driving voltage flowing through the laser diode as an electrical feedback signal; S5: Perform spectral analysis on the electrical feedback signal and calculate an intermodulation distortion performance index based on the analysis results; wherein, the intermodulation distortion performance index characterizes the sum of spectral energy at one or more preset intermodulation frequencies; S6: Using the pre-compensation parameter vector as input and the intermodulation distortion performance index as output, construct and update the Gaussian process regression model; S7: Based on the updated Gaussian process regression model, a new set of pre-compensation parameters is generated using a Bayesian optimization algorithm to minimize the intermodulation distortion performance index; S8: Update the pre-compensation parameters of the pre-compensator to the new pre-compensation parameters; S9: Repeat steps S2 to S8 until the intermodulation distortion performance index meets the preset convergence condition or reaches the preset number of iterations.

[0027] In some embodiments, for step S1, according to the present invention, the generation of the initial composite analog modulated electrical signal is mainly based on the application scenario requirements of the laser diode, synthesizing an electrical signal with a clear frequency component, initial amplitude and initial phase, so as to provide initial input for subsequent pre-compensation optimization and low distortion driving.

[0028] Specifically, the first step is to determine the frequency components of the signal. For example, for different application scenarios such as optical communication, spectral imaging, or precision ranging, multiple frequency components to be superimposed are determined based on the modulation target (e.g., signal transmission rate, imaging resolution, ranging accuracy), denoted as... ( The total number of frequency components can be set from 2 to dozens depending on actual needs. Each frequency component needs to be pre-assigned a unique identifier for subsequent amplitude and phase adjustments and parameter traceability.

[0029] Secondly, set the initial amplitude for each frequency component. ( ) and initial phase ( Initial amplitude The initial phase setting must strictly refer to the rated drive current range of the laser diode, combined with the gain characteristics of the drive circuit, to ensure that the drive current corresponding to the initial signal does not exceed the device's safe drive threshold, thus avoiding damage to the laser diode due to current overload in the initial state. Simultaneously, the initial amplitude must meet the margin requirements for subsequent pre-compensation adjustments; it should not be set too high or too low. Typically, the initial amplitude is taken as 50%-80% of the amplitude corresponding to the device's rated maximum drive current. The initial phase can be set according to the waveform fidelity requirements of the modulation scenario. For example, in scenarios without special phase requirements, the initial phase of all frequency components can be uniformly set to 0; in modulation scenarios requiring specific phase relationships (such as quadrature modulation), the initial phase can be allocated according to a preset phase difference (such as 90°, 180°). This ensures that the initial synthesized waveform conforms to the basic modulation logic.

[0030] Specifically, the initial composite analog modulated electrical signal is generated by a waveform synthesizer, which can be based on a DAC (digital-to-analog converter) or a high-speed waveform generator. Its sampling rate and bandwidth need to match the requirements of the highest modulation frequency component to ensure the accuracy of signal synthesis.

[0031] For example, the synthesis process follows the formula ,in This is the initial modulation signal. This is a time variable. When generating a signal, the waveform synthesizer accurately superimposes the frequency components to avoid extra spurious components in the initial signal due to defects in the synthesis algorithm, which would affect the subsequent intermodulation distortion suppression effect.

[0032] In some embodiments, for step S2, the pre-compensator first receives a set of pre-compensation parameter vectors, which are denoted as... ,in For the first The amplitude adjustment amount for each frequency component For the first Phase adjustment amount for each frequency component.

[0033] Specifically, the pre-compensator adjusts the amplitude and phase of each frequency component separately: 1. Amplitude adjustment: for the first... The initial amplitude of each frequency component Apply adjustment amount The adjusted amplitude was obtained. During the adjustment process, the pre-compensator achieves continuous amplitude adjustment through a high-precision attenuator or gain controller, ensuring precise application of the amplitude adjustment and avoiding deviation of the pre-compensation effect from expectations due to adjustment errors. 2. Phase adjustment: For the first... Initial phase of each frequency component Apply adjustment amount The adjusted phase is obtained Phase adjustment is achieved through a high-precision phase shifter, supporting continuous phase adjustment from 0° to 360°, ensuring that the phase adjustment amount can accurately offset the phase shift caused by the nonlinear response of the subsequent laser diode.

[0034] Preferably, to ensure that the adjustment process does not exceed the safe operating range of the laser diode and the driving circuit, this embodiment of the invention sets safety limits on the amplitude adjustment amount and the phase adjustment amount. Specifically, the amplitude adjustment amount is limited as follows: The maximum positive and negative adjustment percentage range is limited to a range set based on the safe driving range of the laser diode. For example, considering the rated power of the laser diode, the carrying capacity of the driving circuit, and the signal amplitude requirements of the actual application scenario, a common maximum positive and negative adjustment percentage range is set as follows: to The reason for this limitation is that excessive amplitude adjustment may cause the drive current to exceed the rated maximum current of the laser diode, leading to overheating and damage to the device, or causing the drive circuit to enter the nonlinear operating region, generating additional distortion components; while an adjustment range that is too narrow will not be able to adequately cancel intermodulation distortion. to The range is the optimal choice that balances pre-compensation effectiveness and safety.

[0035] Phase adjustment amount limit: Phase adjustment amount It is limited to a maximum positive and negative angle range based on the safe driving range of the laser diode. For example, considering that a phase change may cause instantaneous fluctuations in the drive current, thereby impacting the laser diode chip, a common maximum positive and negative angle range is set as follows: to This limitation effectively avoids sudden current changes caused by excessive phase adjustment, reducing the risk of device damage, while ensuring that the phase adjustment amount can cover the typical phase shift range caused by the nonlinear response of the laser diode, thus guaranteeing the effectiveness of pre-compensation.

[0036] In some embodiments, the safe limit range of the adjustment amount can be adaptively adjusted according to the characteristic parameters (such as rated current and maximum withstand phase fluctuation) of different types of laser diodes. The pre-compensator automatically matches the corresponding adjustment amount limit range by reading the device parameter configuration file of the laser diode.

[0037] The pre-compensator is according to the formula Generate pre-compensated drive signals The signal is then output to the laser diode driver circuit. During the output process, the pre-compensator must ensure the integrity of the signal and avoid distortion of the pre-compensated signal due to impedance mismatch or signal attenuation at the output interface. This is usually achieved by using an impedance matching circuit (such as a 50Ω impedance matching circuit) and a signal buffer amplifier to improve the driving capability and stability of the output signal, ensuring that the pre-compensated driving signal can be accurately transmitted to the driver circuit.

[0038] In some embodiments, for step S3, the pre-compensated drive signal is converted into a drive current that meets the operating requirements of the laser diode, while a built-in protection mechanism ensures that the laser diode operates within a safe range.

[0039] Specifically, the laser diode driving and protection circuit receives the pre-compensation driving signal output by the pre-compensator. Then, the voltage-to-current conversion module converts the voltage-type drive signal into a current-type drive signal. The conversion process must ensure linearity to avoid introducing additional distortion due to nonlinearity. The output current range of the drive circuit must match the rated drive current of the laser diode. For example, for a laser diode with a rated maximum current of 200mA, the output current range of the drive circuit can be set to 0-250mA, leaving a certain safety margin.

[0040] The driver circuit has a built-in current limiting module to limit the current within a safe range, preventing the laser diode from burning out due to overcurrent. It also has a built-in rate limiting module to control the rate of change of the drive current. For example, the rate of change of current is limited to less than 10mA / μs to avoid the instantaneous high voltage caused by the sudden change of current from impacting the PN junction of the laser diode, which could lead to reverse breakdown or shortened lifespan of the device.

[0041] In some embodiments, the driving circuit may also have a built-in temperature compensation module to monitor the operating temperature of the laser diode in real time (collected by a temperature sensor mounted on the surface of the device). When the temperature exceeds a preset threshold (e.g., 50°C), the amplitude of the driving current is automatically adjusted to reduce the power loss of the device and avoid the output characteristics of the laser diode from drifting due to overheating, which would affect the intermodulation distortion suppression effect.

[0042] Finally, the driving circuit inputs the processed stable current signal into the laser diode, causing the laser diode to output an optical signal according to the modulation law of the pre-compensated driving signal.

[0043] In some embodiments, for step S4, the intermodulation distortion information of the laser diode output optical signal can be indirectly obtained by simply collecting the electrical signal (driving current or voltage) in the driving circuit as a feedback proxy, which reduces the hardware complexity and cost of the system, while ensuring the real-time performance and reliability of the feedback signal.

[0044] Specifically, a small-value sampling resistor is connected in series in the laser diode driving circuit. The selection of this sampling resistor needs to meet the requirements of low power consumption, low heat generation, and high accuracy. For example, an alloy sampling resistor with a resistance of 0.05Ω and a power rating of 1W is selected. The small resistance design ensures that the power loss of the sampling resistor itself is minimal, avoiding affecting the current stability of the driving circuit; at the same time, the low temperature coefficient (e.g., ±50ppm / ℃) ensures that the sampling accuracy is not affected by changes in ambient temperature.

[0045] A differential amplifier and an ADC (analog-to-digital converter) are connected across the sampling resistor to form a complete sampling channel. Specifically, a differential amplifier is used to amplify the weak voltage signal across the sampling resistor (because the sampling resistor has a small resistance value, the corresponding voltage signal is usually in the millivolt range). The amplification factor can be adjusted according to actual needs (such as 100x or 200x), while suppressing common-mode interference (such as power frequency interference and power supply noise) and improving the signal-to-noise ratio. The bandwidth of the differential amplifier must match the bandwidth of the ADC to ensure that the high-frequency modulated signal can be fully amplified.

[0046] The ADC is responsible for converting the amplified analog voltage signal into a digital signal for subsequent spectrum analysis.

[0047] The signal acquisition type can be selected according to the system's requirements for feedback accuracy, specifically including: Drive current acquisition: The current signal is converted into a voltage signal through a sampling resistor, amplified by a differential amplifier, and then the drive current waveform data is acquired by an ADC. This method directly reflects the current change driving the laser diode and has the strongest correlation with the amplitude of the laser diode's output optical signal, making it the preferred acquisition type.

[0048] Drive voltage acquisition: The drive voltage waveform data across the laser diode is directly acquired via an ADC. This method can help monitor the operating status of the laser diode (such as whether it is reverse biased or whether there is a risk of voltage breakdown), but its correlation with optical signals is weaker than that with current signals. It is usually used as a supplementary acquisition type, combined with current signals, to improve the comprehensiveness of the feedback.

[0049] Synchronous acquisition: Simultaneously acquire drive current and drive voltage waveform data, and through correlation analysis between the two (such as calculating impedance changes), more accurately determine the nonlinear response state of the laser diode, further improving the accuracy of subsequent spectrum analysis.

[0050] During normal iterative optimization, the ADC continuously acquires electrical feedback signals to ensure that it can capture the dynamic changes of each frequency component and the generation process of intermodulation signals.

[0051] In some embodiments, to ensure the quality of the sampled data, in addition to the key parameters mentioned above, the number of FFT points and the sampling duration are also optimized. Specifically, the number of FFT points should be 4096 or 8192 to ensure that the spectral resolution meets the requirements for intermodulation frequency detection. A higher number of FFT points results in higher spectral resolution and makes it easier to distinguish adjacent fundamental frequency signals from intermodulation signal components, but it also increases computational complexity.

[0052] Sampling duration: The duration of each sample must cover at least one complete cycle of the lowest frequency component to ensure that the waveform characteristics of all frequency components can be fully captured. For example, when the lowest modulation frequency is 1MHz, the sampling duration can be set to 1μs or longer.

[0053] In some embodiments, for step S5, the intermodulation distortion information in the electrical feedback signal is transformed into a quantifiable and optimizable objective function, providing a clear optimization direction for subsequent Gaussian process regression modeling and Bayesian optimization.

[0054] Specifically, see Figure 2 The spectrum analysis process includes signal preprocessing and FFT transformation. Signal preprocessing includes step S201, DC removal, which uses a high-pass filter or directly subtracts the signal's mean to remove the DC offset component from the electrical feedback signal. The presence of a DC component causes DC peaks in the FFT-transformed spectrum, interfering with the detection of intermodulation frequency components; therefore, it needs to be removed. For example, this is achieved by calculating the average value of the sampled signal. ( (number of sampling points), each sampling point minus The signal after DC removal is obtained. .

[0055] S202, windowing processing, applies a window function to the DC-free signal to reduce spectral leakage. The choice of window function needs to balance the main lobe width and side lobe attenuation. The Hann window or Blackman window is preferred. Among them, the Hann window has a narrower main lobe width and a side lobe attenuation of about 31dB, which is suitable for scenarios with high requirements for spectral resolution. The Blackman window has higher sidelobe attenuation (approximately 57dB), but a slightly wider main lobe, making it suitable for scenarios with weak intermodulation signal components and where sidelobe interference needs to be suppressed.

[0056] The windowing process involves applying a signal With window function Multiplication is achieved, that is The length of the window function is related to the number of sampling points. Consistent.

[0057] S203, FFT transform, for the windowed signal Performing an NFFT-point FFT transform converts the time-domain signal into a frequency-domain signal, yielding the frequency amplitude spectrum. The FFT transformation process employs the Fast Fourier Transform algorithm to improve computational efficiency and ensure real-time output of the frequency spectrum results. The transformed frequency amplitude spectrum... In the diagram, the horizontal axis represents frequency. The vertical axis represents the amplitude value of the corresponding frequency. This amplitude value is directly proportional to the power of the signal and can directly reflect the energy of each frequency component.

[0058] In some embodiments, to further improve the stability of spectrum analysis, a multiple averaging FFT technique can be used. This involves performing FFT transformations on multiple consecutively acquired signals (e.g., 4 or 8 signals) and then averaging the resulting frequency amplitude spectra to obtain the final frequency amplitude spectrum. This technology can effectively suppress the influence of random noise and improve the signal-to-noise ratio of intermodulation frequency component detection.

[0059] Based on the frequency components of the initial composite modulation signal, a set of intermodulation frequencies to be suppressed is determined in advance. Intermodulation frequencies are combination frequencies generated by the interaction of different fundamental frequency components, and their generation follows a certain pattern. ( It is a positive integer. (where the intermodulation order is the order of intermodulation), of which the third order intermodulation ( ,like , ) and fifth-order intermodulation ( ,like , The impact on system performance is most significant, therefore the set These low-order intermodulation frequencies should be covered first.

[0060] For example, when the initial frequency component is and ( When ), the set of intermodulation frequencies to be suppressed .gather The specific elements need to be determined based on the application scenario's requirements for suppressing spurious signals. For scenarios with extremely high requirements for waveform fidelity (such as precision ranging), the frequency can be appropriately extended to the seventh-order intermodulation frequency; for general scenarios, coverage up to the fifth-order intermodulation frequency is sufficient to meet the requirements.

[0061] Intermodulation distortion performance indicators The calculation of the intermodulation distortion is the core of quantifying the degree of intermodulation distortion, which is defined as the set of intermodulation frequencies. The sum of the spectral energies of all frequency components in the formula is as follows: Execution: Among them, Intermodulation frequency set The first in One intermodulation frequency; In the frequency amplitude spectrum The amplitude square value at the corresponding position is proportional to the spectral power of the intermodulation frequency and can directly characterize the energy of the intermodulation frequency component. : Optional weighting coefficient, used to adjust the weight ratio of different intermodulation frequency components in the performance index.

[0062] When different intermodulation frequencies have varying impacts on system performance, a larger weighting factor (e.g., 1.2-1.5) should be assigned to intermodulation frequencies with a greater impact (such as those close to the fundamental frequency component or easily misinterpreted as useful signals), while a smaller weighting factor (e.g., 0.5-0.8) should be assigned to intermodulation frequencies with a smaller impact. If the impact of each intermodulation frequency on system performance is consistent, It can be uniformly set to 1 to simplify calculations.

[0063] During the calculation process, it is necessary to locate each intermodulation frequency. In frequency amplitude spectrum The corresponding position in the spectrum. Because the frequency resolution of the spectrum is... ( (where ADC sampling rate is used), when intermodulation frequency is used. When the frequency does not fall exactly at a frequency point in the spectrum, an interpolation algorithm (such as linear interpolation or sinc interpolation) is used to calculate the amplitude value corresponding to that frequency. This ensures the accuracy of the calculation results.

[0064] Intermodulation distortion performance indicators The physical meaning of is the total energy of all intermodulation frequency components that need to be suppressed in the system output signal under the current pre-compensation parameters. The smaller the value, the less intermodulation distortion and the higher the waveform fidelity; conversely, the larger the value, the lower the intermodulation distortion and the higher the waveform fidelity. The larger the value, the more severe the intermodulation distortion. Therefore, subsequent Gaussian process regression modeling and Bayesian optimization both aim to minimize... The core objective is to achieve this.

[0065] In some embodiments, for step S6, the input to the Gaussian process regression model is the pre-compensation parameter vector. The output is the corresponding intermodulation distortion performance index. The core task of the model is to construct a model from the input space. (The allowed value space of the pre-compensation parameters) to the output space Mapping relationship of (the range of values ​​for intermodulation distortion performance index) This mapping relationship can accurately reflect the influence of changes in pre-compensation parameters on intermodulation distortion performance indicators.

[0066] To improve the model's generalization ability, the input parameters need to be normalized before entering the model, mapping the value range of all pre-compensation parameters to the [0,1] interval. For example, for the magnitude adjustment... (The range of values ​​is) , The normalization formula is: For phase adjustment amount (The range of values ​​is) , The normalization formula is: Normalization can prevent model training from being biased towards parameters with larger value ranges due to differences in the units and value ranges of different parameters, thereby improving the training effect and prediction accuracy of the model.

[0067] The kernel function is the core of the Gaussian process regression model, and its selection directly determines the model's ability to model nonlinear relationships. According to an embodiment of the present invention, the kernel function adopts a combination of radial basis functions (RBF / Squared Exponential) and white noise kernels. This combination can simultaneously balance the ability to model smooth nonlinear relationships and the ability to suppress observation noise. The specific formula is: Wherein, the radial basis function part: This is used to model the smooth nonlinear relationship between pre-compensation parameters and intermodulation distortion performance indicators. Radial basis functions have infinite smoothness, which can accurately fit the continuity characteristics of the nonlinear response of lasers. Although the nonlinear response of lasers is complex, it has good continuity and smoothness, which is highly compatible with the modeling characteristics of radial basis functions. For the autocorrelation scaling matrix, different pre-compensation parameters are allowed (e.g. and , and Each pre-compensation parameter affects the output according to its own scale. This characteristic can adapt to the different degrees of influence of different pre-compensation parameters on intermodulation distortion. For example, the amplitude adjustment may have a more significant impact on intermodulation distortion than the phase adjustment, and the model can automatically adjust the corresponding length scale. Adapting to these differences improves modeling accuracy.

[0068] White noise kernel portion: This is used to characterize the observation noise during the sampling and spectral analysis of the electrical feedback signal. To observe the standard deviation of noise, For the Dirac function (when hour, ;otherwise The purpose of this section is to add a reasonable noise term to the model's prediction results, so as to avoid the model overfitting the noise in the training data and improve the model's generalization ability and robustness.

[0069] According to an embodiment of the present invention, see Figure 3 The model training process includes, in step S301, obtaining an initial training dataset through sampling in the early stages of model training. To ensure that the initial dataset can uniformly cover the allowable value space of the pre-compensation parameters, sample points are selected using Latin hypercube sampling or grid sampling methods. Latin hypercube sampling: Dividing the value range of each pre-compensation parameter into... There are equal probability intervals ( The initial sample size is typically 8-16. A sample point is randomly selected from each interval to ensure that the sample points for each parameter are evenly distributed within their value range, while the combination of sample points for different parameters exhibits good randomness. This method can cover the entire parameter space with a smaller sample size, improving the modeling accuracy of the initial model.

[0070] Grid sampling: The value range of each pre-compensation parameter is divided into grids with a fixed step size, and grid nodes are selected as sample points. The advantage of this method is that the sample points are distributed regularly, which facilitates subsequent analysis. However, the number of samples will increase exponentially with the increase of parameter dimension, so it is only suitable for scenarios with low parameter dimension (such as 2N≤6).

[0071] After selecting sample points, the pre-compensation parameters corresponding to each sample point are evaluated (i.e., steps S2-S5 are performed to calculate the corresponding intermodulation distortion performance index). ), forming the initial training dataset .

[0072] S302, after each iteration (i.e., looping once from steps S2 to S8), the new pre-compensation parameters generated by Bayesian optimization are... and their corresponding intermodulation distortion performance indicators Add to the training dataset to form a new training dataset. ( (This represents the number of iterations completed). Subsequently, the Gaussian process regression model is retrained using a new training dataset, and the model's parameters (such as the mean function and covariance function) are updated.

[0073] S303 updates the model's hyperparameters using the Marginal Likelihood Maximization (MLE) method during each model update. , , ):in, (Signal Standard Deviation): Taken from the current training dataset The estimated standard deviation reflects the overall fluctuation range of the intermodulation distortion performance index; (Length scale): Take 10%-50% of the corresponding pre-compensation parameter value range. The initial value can be set according to experience. Iterative optimization is then carried out through the MLE method to enable the model to better adapt to the correlation between parameters and output. (Standard deviation of observation noise): Initially set to 0.01-0.05 times. The initial value is then dynamically adjusted based on the noise level of the training data to ensure that the model can effectively suppress the influence of observation noise.

[0074] Preferably, when the number of frequency components Larger (e.g.) When this happens, the dimension of the pre-compensation parameter vector increases (to 2N), causing the computational complexity of the standard Gaussian process regression model to increase exponentially (to O(n^2). , The current number of training samples is insufficient to meet the requirements of real-time optimization. To address this issue, this invention replaces the standard Gaussian process regression model with a sparse Gaussian process regression model (such as sparse GP based on the FITC algorithm). The core idea is to reduce the computational complexity by selecting a small number of representative inducing points to replace all training sample points for modeling. ( The number of induced points, ).

[0075] See Figure 4 Specifically, the implementation process of the sparse Gaussian process regression model is as follows: S401, Induction point selection: selecting from the training dataset. The selection of induction points must follow the coverage principle to ensure that they uniformly cover the entire parameter space. For example, the K-means clustering algorithm can be used to cluster the training sample points, and the center of each cluster can be selected as the induction point. The number of induction points... Typically, the number of training samples is taken. 10%-20%.

[0076] S402 constructs the covariance matrix of a sparse Gaussian process based on the induced points, and estimates the hyperparameters of the model and the function values ​​at the induced points by maximizing the marginal likelihood function, thus completing the model training.

[0077] S403, in the prediction stage, utilizes the function values ​​and covariance information of the induction points to quickly calculate the predicted mean and variance of the new input parameters without relying on all training sample points, significantly improving the prediction speed.

[0078] Therefore, by applying the sparse Gaussian process regression model, this invention can effectively reduce the computational complexity in high-dimensional parameter scenarios while ensuring minimal loss of modeling accuracy, ensuring that the system can respond in real time and complete optimization iterations, and improving the practicality and adaptability of the method.

[0079] In some embodiments, for step S7, the core optimization objective of this step is to minimize the intermodulation distortion performance index. That is, to find the optimal pre-compensation parameter vector. ,in The allowable value space for the pre-compensation parameters is determined by the adjustment limit range in step S2. This optimization objective directly corresponds to the actual application requirements, namely, to maximize the suppression of intermodulation distortion components in the laser diode output signal and improve waveform fidelity by optimizing the pre-compensation parameters.

[0080] In the optimization process, in addition to the core objective, the system's safety constraints (such as peak drive current, average power, and rate of change of current) must also be considered to ensure that the optimization process is carried out within the safety boundaries and to avoid damage to the laser diode or drive circuit due to the pursuit of optimal intermodulation distortion suppression. Therefore, the actual optimization objective is a constrained minimization problem, i.e. ,in This is a set of safety constraints.

[0081] The acquisition function is the core of the Bayesian optimization algorithm, and its role is to predict the mean based on the prediction results (predicted mean) of the Gaussian process regression model. With prediction variance The method evaluates the "value" of each potential pre-compensation parameter and selects the parameter with the highest value as the next candidate parameter. This invention preferably uses the Expected Improvement Function (EI) or the Upper Confidence Bound Function (UCB) as the acquisition function; both functions have their advantages and can be selected according to the actual scenario.

[0082] Specifically, the Expected Improvement Function (EI) is based on the core idea of ​​calculating new parameters. Intermodulation distortion performance index Better than the current best performance index The expected value, that is It exhibits stronger robustness in small sample scenarios, requiring no additional parameter adjustments and automatically balancing "exploration" and "exploitation." When the model's prediction variance... When the variance is large (the parameter range has not been fully explored), the EI function tends to select that parameter for exploration; when the model prediction variance is large... When the range is small (the parameter range has been fully explored), the EI function tends to select the predicted mean. It utilizes relatively small parameters. Suitable for scenarios with few initial samples and a high degree of uncertainty in the parameter space, it is the preferred acquisition function of this invention.

[0083] The upper confidence bound function (UCB) combines the predicted mean with the predicted variance, and explores the coefficients... Adjust the weights for exploration and utilization, using the following formula: Since the optimization objective is to minimize In practical applications, the following methods are typically used. (Or take a negative value); It is flexible in adjustment and can be adjusted by changing the exploration coefficient. Adapt to different optimization stages. The larger the value, the higher the weight for exploring unknown areas; The smaller the value, the higher the weight of the region with known optimal parameters. This is suitable for scenarios where there are specific requirements for the exploration intensity of the optimization process, such as in the later stages of optimization. Reduce (e.g.) Focusing on utilizing known optimal parameter regions; in the early stages of optimization, one can... Increase (e.g.) This strengthens the exploration of unknown areas. Typically... The value range is 2-3, which can achieve a good balance between exploration and utilization in most scenarios.

[0084] The calculation process of the acquisition function is as follows: First, each potential pre-compensation parameter is predicted based on the Gaussian process regression model. mean With variance Then, the selected acquisition function formula is substituted to calculate the acquisition function value; finally, numerical optimization algorithms (such as L-BFGS, particle swarm optimization) are used in the parameter space. Find the parameter that maximizes the value of the acquisition function. , as the next candidate pre-compensation parameter.

[0085] To ensure that the optimization process does not exceed the system's safe operating range, new pre-compensation parameters are generated. Afterwards, a rigorous safety constraint check is required, the core of which is verification. Corresponding pre-compensation drive signal Whether the preset safety constraints are met, specifically including: Drive current peak constraint: ,in The peak value of the drive current corresponding to the new parameters. This refers to the laser diode's rated maximum current (e.g., 200mA, refer to the device datasheet for details). The peak drive current exceeds... This can cause the laser diode chip to overheat and the PN junction to burn out. Therefore, this constraint is a hard constraint and must be strictly met.

[0086] Average power / thermal limit constraint: ,in This is the long-term average value of the drive current. This is the forward voltage drop of the laser diode. This represents the maximum permissible power dissipation of the laser diode. This constraint is used to prevent thermal damage to the laser diode due to excessively high average power over a long period, ensuring that the device operates within its thermally stable range.

[0087] Current change rate constraint: ,in The maximum rate of change of the driving current. This is the maximum allowable rate of current change (e.g., 10 mA / μs) for the drive circuit and the laser diode. This constraint is used to prevent instantaneous high voltage surges caused by sudden current changes from impacting the laser diode, thus reducing the risk of reverse breakdown of the device.

[0088] The implementation process of safety constraint checking is as follows: Simulation calculation: based on new pre-compensation parameters The corresponding drive current waveform is calculated through circuit simulation model. And extract key indicators such as peak current, average current, and maximum current change rate.

[0089] Constraint verification: The extracted key indicators are compared with the preset safety thresholds to determine whether all constraints are met.

[0090] Exception handling: If If all constraints are met, the parameter is deemed safe and can be used for subsequent parameter updates and iterations; if... If any constraint condition is not met, the following processing method shall be adopted: Apply a penalty term: Set the objective function after penalty to... ,in For large constants (such as) ), Penalty value for exceeding the limit (when exceeding the limit) Take the maximum value, such as ; Not exceeding the time limit The penalized objective function is fed back to the acquisition function to regenerate candidate parameters; Direct refusal: If Serious violation of safety constraints (such as peak current exceeding) If the threshold is 120%, then the parameter is directly rejected without imposing a penalty term, saving computation time and allowing for the regeneration of candidate parameters.

[0091] By checking safety constraints, this invention can ensure that the optimization process is always carried out within the safety boundary, avoiding system damage caused by parameter optimization and improving the reliability and safety of the method.

[0092] To further improve optimization efficiency and shorten iteration convergence time, this invention supports parallel evaluation of the Bayesian optimization process, that is, simultaneously generating and evaluating multiple candidate pre-compensation parameter combinations. This feature is particularly suitable for scenarios with high parameter dimensionality or large sample budgets, and can significantly reduce the overall optimization time.

[0093] Specifically, see Figure 5 The parallel optimization implementation process is as follows: S501, Parallel Candidate Parameter Generation: Parallel acquisition functions (such as parallelized EI or q-EI) are used in the parameter space. Simultaneously generated indivual( (This represents the number of parallel evaluations, typically 2-4, determined based on the number of hardware channels and computing resources.) Candidate pre-compensation parameters During the generation process, it is necessary to ensure that there are certain differences between the candidate parameters to avoid them being concentrated in the same parameter region, which would affect the parallel optimization effect.

[0094] S502, Parallel Evaluation: Utilizing multi-channel drive circuits and multiple sampling modules, simultaneously... The candidate parameters are applied to the pre-compensator to generate... Road pre-compensation drive signal, respectively drive Using three identical laser diodes (or different drive cycles of the same laser diode), electrical feedback signals are simultaneously acquired through corresponding sampling channels, and the intermodulation distortion performance index corresponding to each candidate parameter is calculated. .

[0095] S503, Training Set Update: [The text appears to be incomplete and contains several typos. A more accurate translation would Each candidate parameter and its corresponding performance metric is added to the training dataset, and the Gaussian process regression model is retrained to enter the next iteration.

[0096] The core advantage of parallel optimization lies in its ability to acquire information from multiple sample points simultaneously in a single iteration, significantly improving the speed of training sample accumulation and shortening model convergence time. For example, when the number of parallel evaluations... In this way, the optimization iteration time can be shortened to about 1 / 4 of the original time, which greatly improves the practicality and engineering feasibility of the method.

[0097] In some embodiments, for step S8, before updating the parameters, the new pre-compensation parameters generated by Bayesian optimization need to be processed. Perform final verification to ensure that the parameter meets the following conditions: Security verification: Confirm again The safety constraint check has been passed, and the corresponding peak drive current, average power, current change rate and other indicators are all within the safe range; Validity verification: Confirmed Corresponding intermodulation distortion performance indicators Compared to the current parameters Some reduction (i.e.) This ensures that parameter updates lead to improved intermodulation distortion suppression. If... If the parameter is invalid, the process should be backtracked to step S7 to regenerate the candidate parameter. like After all verifications are passed, the pre-compensator will store the current pre-compensation parameters internally. Updated to Parameter updates are implemented via a high-speed bus (such as SPI or I2C) to ensure the speed of the update process (update time ≤ 1ms) and avoid instantaneous distortion of the drive signal due to parameter update delays; After the parameters are updated, the system automatically backs up the current pre-compensation parameters. and their corresponding intermodulation distortion performance indicators Information such as update time is uploaded to the database to form a parameter update log. This log can be used for subsequent optimization process tracking, fault diagnosis, and model improvement. For example, when an anomaly occurs during the optimization process, the log can be used to trace back to the historical optimal parameters and restore the normal operation of the system.

[0098] In some embodiments, to improve the reliability of parameter updates, a dual-buffering mechanism is adopted: the pre-compensator has two built-in parameter buffers (a main buffer and a backup buffer). The main buffer stores the parameters currently in use, and the backup buffer stores the newly generated parameters. When updating parameters, first... The data is written to the backup buffer, and then a hardware signal triggers a switch between the primary and backup buffers to ensure that the drive signals are uninterrupted and distortion-free during the parameter update process. After the switch is complete, the parameters in the primary buffer are verified, and the entire update process is completed once the parameters are confirmed to be written correctly.

[0099] In some embodiments, for step S9, to avoid the iteration process from going on indefinitely, a clear convergence criterion needs to be preset. The iteration can stop when any criterion is met. Specifically, this may include... Performance metrics convergence: Intermodulation distortion performance metrics during 3-5 consecutive iterations. Change ≤ preset threshold .in The value needs to be determined based on the waveform fidelity requirements of the application scenario. For example, Can be set to the current optimal The value is 5%. The physical meaning of this convergence condition is that when the improvement in intermodulation distortion in multiple consecutive iterations is very small, it indicates that the parameters are close to the optimal value, and continuing the iteration will not bring significant performance improvement. At this time, stopping the iteration can save computational resources and time.

[0100] Specific judgment formula: ,in This represents the current iteration number. To determine the number of consecutive rounds (3-5 rounds).

[0101] Iteration limit: The number of iterations reaches the preset maximum number of iterations. (Typically set to 30-80 iterations). This convergence condition is to prevent iterations from failing to converge due to complex parameter space or model prediction bias, ensuring that the system can output optimization results within a finite time.

[0102] Optionally, when the parameter dimension is low (e.g., 2N≤10), the optimization process is easier to converge, and the maximum number of iterations can be set to 30-40 times; when the parameter dimension is high (e.g., 2N>10), the optimization process requires more sample points to explore the parameter space, and the maximum number of iterations can be set to 60-80 times.

[0103] In some embodiments, a third convergence condition may also be set: model uncertainty convergence: when the Gaussian process regression model achieves optimal parameters... Prediction variance ( This is a preset variance threshold, such as 0.01 times. When the initial value is reached, it indicates that the model's prediction of the optimal parameters is sufficiently reliable, and further iteration is not worthwhile, so iteration can be stopped. This convergence condition can further improve the reliability of the optimization results, but it will increase the computational complexity to some extent. It can be selected whether to enable it according to actual needs.

[0104] Starting from step S2, repeat steps S2 to S8. In each iteration, the system generates better pre-compensation parameters based on the optimization results of the previous round, using a Gaussian process regression model and a Bayesian optimization algorithm, thereby gradually reducing the intermodulation distortion performance index.

[0105] The system stops iterating when any convergence condition is met, and outputs the final optimal pre-compensation parameters. and its corresponding pre-compensation drive signal This driving signal can minimize intermodulation distortion and maximize waveform fidelity of the optical signal output by the laser diode.

[0106] Figure 6 A system 600 for driving an analog modulation waveform of a laser diode is shown. Embodiments of this system are... Figure 1 Corresponding to the illustrated method embodiments, the specific methods include: The initial modulation signal generation module 601 is used to generate an initial composite analog modulation electrical signal, which is synthesized from multiple frequency components with specific initial amplitude and initial phase. The pre-compensation module 602 is used to input the initial modulation signal to the pre-compensator. The pre-compensator applies an adjustment amount to the amplitude and phase of each frequency component in the initial modulation signal according to a set of pre-compensation parameters, and generates and outputs a pre-compensated drive signal. The pre-compensation parameters include the amplitude adjustment amount and phase adjustment amount of each frequency component. The driving module 603 is used to drive the laser diode using the pre-compensated driving signal; The feedback signal acquisition module 604 is used to acquire the waveform of the driving current and / or driving voltage flowing through the laser diode as an electrical feedback signal; The performance index setting module 605 is used to perform spectrum analysis on the electrical feedback signal and calculate an intermodulation distortion performance index based on the analysis results; wherein, the intermodulation distortion performance index represents the sum of spectral energy at one or more preset intermodulation frequencies; The model building module 606 is used to build and update the Gaussian process regression model with the pre-compensation parameter vector as input and the intermodulation distortion performance index as output. The optimization module 607 is used to generate a new set of pre-compensation parameters based on the updated Gaussian process regression model using a Bayesian optimization algorithm to minimize the intermodulation distortion performance index. The parameter update module 608 is used to update the pre-compensation parameters of the pre-compensator to the new pre-compensation parameters. The parameter determination module 609 is used to repeatedly execute steps S2 to S8 until the intermodulation distortion performance index meets the preset convergence condition or reaches the preset number of iterations.

[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for driving a laser diode to simulate a modulation waveform, characterized in that, Includes the following steps: S1: Generate an initial composite analog modulated electrical signal, which is synthesized from multiple frequency components with specific initial amplitude and initial phase; S2: The initial composite analog modulated electrical signal is input to the pre-compensator. The pre-compensator applies an adjustment amount to the amplitude and phase of each frequency component in the initial composite analog modulated electrical signal according to a set of pre-compensation parameters, and generates and outputs a pre-compensated drive signal; wherein, the pre-compensation parameters include the amplitude adjustment amount and phase adjustment amount of each frequency component. S3: Drive the laser diode using the pre-compensated drive signal; S4: Acquire the waveform of the driving current and / or driving voltage flowing through the laser diode as an electrical feedback signal; S5: Perform spectral analysis on the electrical feedback signal and calculate an intermodulation distortion performance index based on the analysis results; wherein, the intermodulation distortion performance index characterizes the sum of spectral energy at one or more preset intermodulation frequencies; S6: Using the pre-compensation parameter vector as input and the intermodulation distortion performance index as output, construct and update the Gaussian process regression model; S7: Based on the updated Gaussian process regression model, a new set of pre-compensation parameters is generated using a Bayesian optimization algorithm to minimize the intermodulation distortion performance index; S8: Update the pre-compensation parameters of the pre-compensator to the new pre-compensation parameters; S9: Repeat steps S2 to S8 until the intermodulation distortion performance index meets the preset convergence condition or reaches the preset number of iterations.

2. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, In step S2, the amplitude adjustment for each frequency component is limited to a maximum positive or negative adjustment percentage range set based on the safe driving range of the laser diode, and / or the phase adjustment for each frequency component is limited to a maximum positive or negative angle range set based on the safe driving range of the laser diode.

3. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, Step S5 involves performing spectral analysis on the electrical feedback signal, specifically including: windowing and fast Fourier transforming the electrical feedback signal to obtain its frequency amplitude spectrum. The calculation of the intermodulation distortion performance index is specifically as follows: for each frequency in the predetermined set of intermodulation frequencies to be suppressed, the squared value of its amplitude at the corresponding position in the frequency amplitude spectrum is multiplied by an optional weighting coefficient and then summed.

4. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, In step S6, the Gaussian process regression model uses a combination of radial basis functions and white noise kernels as its kernel function. The radial basis functions are used to model the smooth nonlinear relationship between the pre-compensation parameters and the intermodulation distortion performance index, and the white noise kernel is used to characterize the noise in the observed data.

5. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, In step S7, the Bayesian optimization algorithm uses the expected boost function or the upper confidence bound function as its acquisition function.

6. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, In step S7, when generating new pre-compensation parameters, a safety constraint check is also included. Specifically, it is determined whether the driving signal generated using the new pre-compensation parameters will cause at least one of the peak value of the driving current, average power, or rate of change of the laser diode to exceed its safety threshold. If so, a penalty term is applied or the parameter is directly rejected to ensure that the optimization process is carried out within the safety boundary.

7. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, The training data used in step S6 to construct the Gaussian process regression model is initially obtained by collecting and evaluating multiple sample points within the allowed value space of the pre-compensated parameter vector using Latin hypercube sampling or grid sampling.

8. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, When the number of frequency components is large, resulting in a high dimension of the pre-compensation parameter vector, a sparse Gaussian process regression model is used to replace the standard Gaussian process regression model to reduce computational complexity.

9. The method for driving a laser diode to simulate a modulation waveform according to claim 1, characterized in that, The Bayesian optimization process in step S7 supports the parallel evaluation of multiple candidate parameter combinations to accelerate the optimization process.

10. A system for driving an analog modulation waveform of a laser diode, characterized in that, include: The initial composite analog modulated electrical signal generation module is used to generate an initial composite analog modulated electrical signal, which is synthesized from multiple frequency components with specific initial amplitude and initial phase. The pre-compensation module is used to input the initial composite analog modulated electrical signal to the pre-compensator. The pre-compensator applies an adjustment amount to the amplitude and phase of each frequency component in the initial composite analog modulated electrical signal according to a set of pre-compensation parameters, and generates and outputs a pre-compensated drive signal. The pre-compensation parameters include the amplitude adjustment amount and phase adjustment amount for each frequency component. A driving module is used to drive a laser diode using the pre-compensated driving signal; The feedback signal acquisition module is used to acquire the waveform of the driving current and / or driving voltage flowing through the laser diode as an electrical feedback signal. The performance index setting module is used to perform spectrum analysis on the electrical feedback signal and calculate an intermodulation distortion performance index based on the analysis results; wherein, the intermodulation distortion performance index represents the sum of spectral energy at one or more preset intermodulation frequencies; The model building module is used to construct and update the Gaussian process regression model with the pre-compensation parameter vector as input and the intermodulation distortion performance index as output. An optimization module is used to generate a new set of pre-compensation parameters based on the updated Gaussian process regression model using a Bayesian optimization algorithm to minimize the intermodulation distortion performance index. The parameter update module is used to update the pre-compensation parameters of the pre-compensator to the new pre-compensation parameters; The parameter determination module is used to repeatedly execute steps S2 to S8 until the intermodulation distortion performance index meets the preset convergence condition or reaches the preset number of iterations.

Images