A wavelength stable output method of a femtosecond laser pumped optical parametric amplifier

Abstract

The application relates to the field of laser technology and discloses a wavelength stable output method of a femtosecond laser pumped optical parametric amplifier. The method realizes real-time acquisition of a full spectrum of pump laser by a spectrometer, calculates a spectral barycenter frequency of the pump laser, combines a signal light frequency fed back by a wavemeter, constructs a double-variable deviation input, generates a phase matching angle correction amount by using a pre-trained neural network model, drives a piezoelectric ceramic platform to dynamically adjust the angle of a nonlinear crystal, and maintains a phase matching condition of Delta k=0. The application synchronously monitors the full spectrum spectral barycenter of the femtosecond pump laser and the output wavelength of signal light, constructs a double-variable feedback mechanism for a wide spectrum pump source, and breaks through the limitation that traditional single-point frequency locking technology can only stabilize a single frequency.

Description

Technical Field

[0001] This invention belongs to the field of laser technology, specifically relating to a method for stabilizing the wavelength output of a femtosecond laser-pumped optical parametric amplifier. Background Technology

[0002] Femtosecond laser-pumped optical parametric amplifiers (OPAs), as key components in ultrafast nonlinear optical systems, are widely used in high-resolution spectroscopy, attosecond science, and the generation of multi-band coherent light sources. Their working principle relies on a three-wave mixing process between the pump light, signal light, and idler light in a nonlinear crystal, achieving broadband tunable output through energy and phase matching. However, this process is extremely sensitive to the frequency stability of the pump laser; even minute frequency fluctuations can directly cause wavelength shifts in the signal and idler lights, disrupting the repeatability and coherence of the output spectrum and severely impacting the accuracy and reliability of downstream applications.

[0003] Wavelength stabilization control has become a core technological direction for improving the performance of femtosecond OPA systems. Traditional solutions often employ frequency-locking techniques based on feedback loops, such as using Fabry-Perot cavities or saturable absorbers to narrowly lock the pump laser and then indirectly stabilize the parametric output. These methods are highly effective in continuous-wave or narrow-linewidth laser systems, but when faced with the inherent wide-spectral characteristics of femtosecond lasers, their locking bandwidth is limited, their response speed is insufficient, and they struggle to cover the frequency noise components across the entire gain spectrum, resulting in significant fluctuations in the wide-spectral output.

[0004] Existing technologies for addressing the wavelength stability problem of femtosecond OPAs generally suffer from drawbacks such as mismatch between the locking mechanism and the light source characteristics, weak anti-interference capabilities, and poor dynamic adaptability. Traditional frequency locking strategies treat frequency noise as an interference source to be suppressed, passively filtering it rather than actively utilizing it; they lack coordinated control over the internal dynamic processes of the nonlinear crystal, making it impossible to achieve global synchronous locking over a wide spectrum. Especially under complex environmental disturbances or long-term operating conditions, the system is susceptible to the coupled effects of multiple factors such as temperature drift, mechanical vibration, and pump power fluctuations, leading to continuous output wavelength drift or even loss of lock. Therefore, there is an urgent need for a novel wavelength stabilization output method that can deeply integrate nonlinear dynamic characteristics, actively manage noise resources, and possess adaptive adjustment capabilities to overcome the bottlenecks of existing technologies in terms of wide-spectrum locking accuracy and robustness. Summary of the Invention

[0005] This invention provides a method for stabilizing the wavelength output of a femtosecond laser-pumped optical parametric amplifier, aiming to solve the problem of wavelength drift in the idler and signal light outputs of an optical parametric amplifier under femtosecond laser pumping conditions due to fluctuations in the pump laser frequency. In existing technologies, traditional frequency-locking techniques are mainly designed for narrow-linewidth continuous or quasi-continuous lasers, and their feedback control mechanisms rely on the extraction of error signals at a single frequency point, making it difficult to effectively cope with the ultra-wide spectral bandwidth and instantaneous frequency jitter characteristics of femtosecond laser pulses.

[0006] When such frequency-locking systems are applied to femtosecond pump sources, the inability to globally sense and coordinately correct phase and frequency disturbances across the entire spectral range leads to a mismatch in the phase-matching conditions of the nonlinear crystal during optical parametric amplification as the centroid of the pump light spectrum shifts. This results in uncontrollable drift of the output wavelengths of the signal light and idler light, severely affecting wavelength repeatability and the reliability of experimental data.

[0007] To overcome the above-mentioned technical defects, this invention proposes a method for stabilizing the wavelength output of a femtosecond laser-pumped optical parametric amplifier, comprising: Polarization stabilization of femtosecond pump laser; The output beam of the femtosecond pump laser was sampled by a spectrometer to obtain its complete spectral intensity distribution data. Based on the spectral intensity distribution data, the spectral centroid frequency of the pump laser is calculated, and the spectral centroid frequency is defined as the spectral intensity weighted average frequency. The center wavelength of the signal light output by the optical parametric amplifier is monitored in real time using a wavelength meter and converted into the corresponding signal light frequency. The pump frequency deviation is obtained by comparing the spectral centroid frequency of the pump laser with the preset target pump frequency. The signal light frequency is compared with the preset target signal light frequency to obtain the signal light frequency deviation. The pump frequency deviation and the signal light frequency deviation are input together into a pre-trained phase matching compensation model, which outputs the phase matching angle correction amount required for the nonlinear crystal under the current deviation state. Based on the phase matching angle correction, a high-precision piezoelectric ceramic rotating platform mounted on the nonlinear crystal base is driven to finely adjust the incident angle of the nonlinear crystal in real time, so that the pump light, signal light and idler light satisfy the momentum conservation condition.

[0008] Preferably, polarization stabilization processing of the femtosecond pump laser includes: By combining a Faraday isolator with a quarter-wave plate, the interference of back-reflected light on the pump laser resonator is eliminated, ensuring that the polarization direction of the pump light maintains a fixed angle with the principal axis of the nonlinear crystal.

[0009] Preferably, calculating the spectral centroid frequency of the pump laser based on the spectral intensity distribution data includes: Background subtraction and nonlinear response correction are performed on the discrete wavelength channel intensity data output by the spectrometer; According to the formula Calculate the frequency of the spectral barycenter , For the first The center frequency of each channel For the corresponding strength value, This represents the total number of valid channels.

[0010] Preferably, the method involves real-time monitoring of the center wavelength of the signal light using a wavelength meter and converting it into the signal light frequency, including: A wavelength meter based on the Michelson or Fabry-Perot interferometry principle is used; The measured center wavelength of the signal light Through the speed of light constant Convert to frequency ; Preferably, the phase matching compensation model adopts a multi-layer feedforward neural network structure. The input layer contains two nodes corresponding to the pump frequency deviation and the signal light frequency deviation, respectively. The hidden layer contains three nodes. The activation function is the hyperbolic tangent function. The output layer contains the phase matching angle correction corresponding to the nodes.

[0011] Preferably, the phase-matching compensation model is trained using an offline calibration dataset during the system initialization phase. The calibration dataset generates a known frequency drift by actively perturbing the pump laser cavity length or temperature, and records the corresponding optimal phase-matching angle adjustment value of the nonlinear crystal to form input-output sample pairs.

[0012] Preferably, driving the piezoelectric ceramic rotating platform according to the phase matching angle correction includes: The nonlinear crystal is mounted on a 5D adjustment frame, and its rotational degree of freedom about the axis perpendicular to the incident surface is controlled by a piezoelectric ceramic actuator. The central controller generates a driving voltage through a digital-to-analog converter, which is then amplified by a high voltage and applied to the piezoelectric ceramic to achieve fine-tuning of the angle.

[0013] Preferably, it also includes an environmental parameter compensation step: Ambient temperature and air pressure are collected using temperature and humidity sensors and a barometer; Environmental parameters are input into the atmospheric dispersion modifier model to calculate the correction amount for the effective refractive index of the nonlinear crystal. Based on the aforementioned correction amount, an additional correction amount for the phase matching angle is derived and superimposed with the main correction amount to form the final drive command.

[0014] Preferably, the method further includes periodically performing online updates of the phase-matching compensation model: At preset time intervals, the nonlinear crystal is controlled to perform step angle scanning near the current angle; Record the angular position corresponding to the peak value of the signal light output intensity as the measured optimal angle; The measured optimal angle is compared with the model's predicted angle. If the deviation is greater than a preset threshold, the incremental learning algorithm is triggered to fine-tune the neural network weights.

[0015] Preferably, the central controller is implemented using a field-programmable gate array (FPGA) chip, which integrates a spectral centroid calculation unit, a deviation generation unit, a neural network inference engine, and a digital-to-analog conversion control unit. The neural network inference engine adopts a pipelined structure.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a bivariate feedback mechanism for broadband pump sources by synchronously monitoring the centroid of the full spectrum of the femtosecond pump laser and the output wavelength of the signal light, thus overcoming the limitation of traditional single-point frequency locking technology that can only stabilize a single frequency.

[0017] 2. The phase-matching compensation model based on neural networks can effectively learn and map the nonlinear relationship between pump spectrum perturbation and crystal angle adjustment, thus achieving dynamic and accurate maintenance of phase-matching conditions.

[0018] 3. By introducing environmental parameter compensation and online model update mechanisms, the robustness and adaptability of the system under long-term operation and environmental changes are further improved, providing reliable technical support for application scenarios with stringent requirements for wavelength stability, such as ultrafast spectroscopy, coherent anti-Stokes Raman scattering imaging, and quantum light sources. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the present invention, which combines full-spectrum pump light monitoring with nonlinear phase-matching dynamic compensation. Figure 3 This is a flowchart illustrating the logical process of calculating the centroid of the pump spectrum and collecting dual-variable acquisition and deviation generation of the signal light wavelength feedback in this invention. Figure 4 This is a schematic diagram of the neural network structure and input-output mapping relationship of the phase matching compensation model in this invention. Figure 5 This is a schematic diagram of the multi-level interaction and data flow between the nonlinear crystal angle adjustment actuator and the high-precision piezoelectric ceramic drive control in this invention; Figure 6 This is a logical flowchart of the collaborative compensation mechanism for environmental parameter monitoring, atmospheric dispersion correction, and online model updates in this invention. Detailed Implementation

[0020] refer to Figures 1 to 6 This invention provides a method for stabilizing the output wavelength of a femtosecond laser-pumped optical parametric amplifier, aiming to solve the problem of wavelength drift in the output of signal light and idler light caused by the ultra-wide spectral bandwidth and instantaneous frequency jitter characteristics of femtosecond laser as a pump source.

[0021] Traditional frequency locking technology is based on a single frequency point error signal extraction mechanism, which cannot globally perceive and coordinately correct phase and frequency disturbances across the entire spectrum of femtosecond lasers. This causes the phase matching condition of the nonlinear crystal to become mismatched due to the shift in the centroid of the pump light spectrum, leading to uncontrollable drift of the output wavelength.

[0022] This embodiment constructs a closed-loop feedback control system to simultaneously acquire the complete spectral information of the femtosecond pump laser and the center wavelength information of the signal light, and establishes a mapping relationship model between the centroid of the pump spectrum and the phase matching angle. It then drives the nonlinear crystal angle adjustment mechanism in real time to dynamically maintain the constant phase matching condition, thereby achieving highly stable output of the signal light and idler light wavelengths.

[0023] The method includes the following steps: S1, polarization stabilization processing is performed on the femtosecond pump laser; S2, use a spectrometer to split and sample the output beam of the femtosecond pump laser to obtain complete spectral intensity distribution data; S3, Based on the spectral intensity distribution data, calculate the spectral centroid frequency of the pump laser; S4 uses a wavelength meter to monitor the center wavelength of the signal light output by the optical parametric amplifier in real time and converts it into the corresponding signal light frequency; S5, compare the spectral centroid frequency of the pump laser with the preset target pump frequency to obtain the pump frequency deviation. S6, compare the signal light frequency with the preset target signal light frequency to obtain the signal light frequency deviation; S7, The pump frequency deviation and the signal light frequency deviation are input together into the pre-trained phase matching compensation model, which outputs the phase matching angle correction amount required to adjust the nonlinear crystal under the current deviation state. S8. Based on the phase matching angle correction, drive the high-precision piezoelectric ceramic rotating platform mounted on the nonlinear crystal base to finely adjust the incident angle of the nonlinear crystal in real time, so that the pump light, signal light and idler light satisfy the conditions of energy conservation and momentum conservation. S9 monitors ambient temperature and air pressure in real time and performs secondary compensation for the phase matching angle correction. S10, periodically perform online updates of the phase matching compensation model.

[0024] In step S1, the femtosecond pump laser undergoes polarization stabilization. This polarization stabilization is achieved through a combination of a Faraday isolator and a quarter-wave plate. The Faraday isolator is placed after the output of the femtosecond laser to block backlight reflected from subsequent optical elements back into the pump laser resonant cavity, preventing it from causing intracavity mode instability or frequency jumps.

[0025] A quarter-wave plate is positioned after the Faraday isolator, with its fast axis set at 45 degrees relative to the initial polarization direction of the pump laser. This ensures that after one round trip, the polarization state is rotated by 90 degrees, preventing it from returning through the Faraday isolator. This combined structure ensures that the pump light polarization direction remains constant before entering the nonlinear crystal and maintains a fixed angle with the principal axis of the nonlinear crystal.

[0026] The principal axis direction maintains a fixed angle, which is preset according to the type of nonlinear crystal and the target phase matching type. It is usually zero degrees or a specific angle to meet the collinear type I or type II phase matching conditions. The stability of the polarization state is directly related to the effective components of the nonlinear polarizability tensor. If drift occurs, it will introduce additional phase matching errors, resulting in a decrease in conversion efficiency or wavelength shift.

[0027] In step S2, the output beam of the femtosecond pump laser is sampled by a spectrometer. A high-damage-threshold fused silica beam splitter is used, with a splitting ratio set to 99%:1 to minimize energy loss in the main optical path while providing sufficient signal-to-noise ratio sampling light for the monitoring optical path. The spectrometer has a spectral resolution of 0.1 nm, covering a wavelength range of 750 nm to 850 nm, sufficient to resolve the complete spectral profile of a Ti:sapphire femtosecond laser with a center wavelength of 800 nm and a pulse width of 30 femtoseconds.

[0028] The spectrometer employs a two-dimensional area array detector, combined with a precision grating and slit system, to achieve a balance between high throughput and high resolution. The sampling frequency is greater than 1 kHz, ensuring the capture of spectral fluctuations in femtosecond pumped lasers within a single pulse cycle caused by gain medium thermal effects, intracavity dispersion fluctuations, or pump power fluctuations. The spectrometer outputs an intensity array containing 1024 wavelength channels per frame, with each channel corresponding to a discrete wavelength point and its corresponding intensity value.

[0029] In step S3, the spectral centroid frequency of the pump laser is calculated based on the spectral intensity distribution data. The spectral centroid frequency is defined as the intensity-weighted average frequency. Its mathematical expression is: ; For frequency The spectral intensity at point [x] is given. In practical calculations, the integral is discretized into a summation operation. Let the spectral intensity at point [x] be the spectral intensity output by the spectrometer. The center frequency corresponding to each channel is The measured relative intensity is The formula for calculating the spectral centroid frequency is: ; This represents the total number of effective spectral channels. The calculation is performed by the spectral centroid calculation unit inside the central controller. This unit employs a fixed-point arithmetic architecture, avoiding the latency and resource overhead introduced by floating-point operations, and supports processing 1,000 frames of spectral data per second. Before calculation, background subtraction and nonlinear response correction are performed on the raw intensity data. The correction parameters are obtained through factory calibration and stored in the controller's non-volatile memory. The calculated spectral centroid frequency is output in a 60-bit fixed-point format, preserving sufficient dynamic range and accuracy for subsequent deviation generation.

[0030] In step S4, the center wavelength of the signal light output from the optical parametric amplifier (OPA) is monitored in real time using a wavelength meter. The signal light is drawn from the OPA output port via another beam splitter and enters a separate wavelength meter channel. This signal light center wavelength meter employs the Michelson-Perot interferometry principle or the Fabry-Perot interferometry principle, achieving wavelength measurement accuracy better than the picometer level, with a repeatability standard deviation of less than 0.5 picometers. The response time is less than 1 millisecond, ensuring the ability to track rapid changes in the signal light wavelength.

[0031] The raw data output by the wavelength meter is the center wavelength value. The unit is nanometers. The central controller receives this center wavelength value. Then, it is immediately converted to the wavelength in air using a pre-stored vacuum-to-air refractive index conversion table, and then the speed of light constant is used. Converted to the corresponding signal light frequency The frequency values ​​are also stored in a high-precision fixed-point number format for comparison with the target frequency.

[0032] In step S5, the spectral centroid frequency of the pump laser is... With the preset target pump frequency By comparison, the pump frequency deviation is obtained. Target pump frequency This corresponds to the ideal pump wavelength set during system design, such as the frequency corresponding to 800 nanometers. The target pump frequency is set by the user or loaded from a configuration file during system initialization and stored in the controller's read-only memory. Deviation calculation uses double-precision subtraction, retaining the sign bit to distinguish between positive and negative drift directions.

[0033] In step S6, the frequency of the signal light is... With the preset target signal light frequency By comparison, the frequency deviation of the signal light is obtained. Target signal light frequency This corresponds to the user's desired output signal wavelength, such as 1100 nanometers. The target signal frequency can be dynamically set by the host computer via the communication interface or stored locally. Two deviation values... and The input vector that constitutes the phase matching compensation model.

[0034] In step S7, the pump frequency deviation and the signal light frequency deviation are jointly input into a pre-trained phase-matching compensation model. This phase-matching compensation model is constructed using a multi-layer feedforward neural network structure. The input layer contains two nodes, which respectively receive… and The hidden layers consist of three layers with 32, 16, and 8 nodes respectively. The activation function for each hidden layer is the hyperbolic tangent function. , The input is a variable whose output range is -1 to +1, exhibiting good gradient characteristics and nonlinear expressive power. The output layer contains one node, outputting the phase matching angle correction. The unit is radians. The model is trained using an offline calibration dataset during the system initialization phase.

[0035] The calibration process is as follows: The cavity length of the femtosecond laser is actively adjusted using piezoelectric ceramics or a temperature control module to generate a frequency drift of known amplitude, with a step size of 10 GHz, covering a range of ±500 GHz. At each drift point, the nonlinear crystal angle is manually fine-tuned until the signal light output intensity reaches its maximum value; the angle position at this point is recorded as the optimal phase-matching angle. The pump frequency deviation, the signal light frequency deviation (which is zero or a known value at this point), and the optimal angle offset are used to form an input-output sample pair. This process is repeated at least 500 times to form a training set. The backpropagation algorithm and mean square error loss function are used to optimize the network weights until the validation set error is less than a preset threshold. The weight matrix and bias vector after training are stored in the read-only memory of the central controller.

[0036] In step S8, the phase matching angle correction amount is determined. This drives a high-precision piezoelectric ceramic rotating platform mounted on a nonlinear crystal base. The nonlinear crystal is... The barium borate crystal is pre-cut at an angle between 42 and 45 degrees, based on the target signal light wavelength range. The crystal is mounted on a 5D adjustment frame, and its rotational degree of freedom about the axis perpendicular to the incident plane is controlled by a piezoelectric ceramic actuator. This rotational degree of freedom, controlled by the piezoelectric ceramic actuator, has an angular displacement resolution better than 0.1 microradians and a stroke range of ±5 milliradians, sufficient to cover the range of phase matching angle variations caused by pump frequency drift.

[0037] The digital-to-analog converter control unit inside the central controller receives Then, based on the pre-stored piezoelectric ceramic voltage-displacement calibration curve, the corresponding driving voltage is calculated. The conversion resolution is 16 bits, corresponding to 65,536 levels of voltage output, ensuring precise angle control. The drive voltage is amplified by a high-voltage amplifier and applied to the piezoelectric ceramic, causing it to produce precise angular displacement. After the displacement is completed, the system waits for a preset stabilization time (usually 200 microseconds) to eliminate the effect of piezoelectric hysteresis before entering the next control cycle.

[0038] In step S9, ambient temperature and air pressure are monitored in real time, and secondary compensation is performed on the phase matching angle correction. The high-precision temperature and humidity sensor and barometer are installed inside the optical parametric amplifier housing, close to the nonlinear crystal. The temperature measurement accuracy is 0.01 degrees Celsius, and the air pressure measurement accuracy is 0.1 Pascals. The collected environmental parameters... and The data is input into the atmospheric dispersion correction sub-model. This sub-model calculates the refractive index of air under the current environment based on the Edlén or Ciddor formulas. . Represents the wavelength of light.

[0039] Since the pump light, signal light, and idler light propagate in air, their effective wave vectors are affected by the air refractive index, thus altering the phase-matching condition. The sub-model calculates the equivalent correction for the effective refractive index of the nonlinear crystal. Based on the differential form of the Sellmeier equation, the additional correction for the phase matching angle is derived. .

[0040] This additional correction amount is related to the main correction amount. Superimposed to form the final driving command This is used to drive the piezoelectric ceramic platform. This compensation mechanism effectively eliminates the impact of laboratory environmental fluctuations on optical path difference, improving long-term stability.

[0041] In step S10, the phase-matching compensation model is periodically updated online. Every preset time interval (e.g., every 2 hours of operation), the system pauses external tasks and starts the wavelength scanning program. The program controls the piezoelectric ceramic driver to make the nonlinear crystal perform a step-by-step angle scan around the current angle, with a step size of 0.5 microradians and a total scan range of ±20 microradians.

[0042] At each step point, the signal light output intensity is recorded. After the scan is completed, the angular position corresponding to the intensity peak is determined by parabolic fitting. The measured optimal angle is compared with the model predicted angle. Compare them. Calculate the deviation based on the initial angle position. .like If the value exceeds a preset threshold (e.g., 5 microradians), a model parameter fine-tuning algorithm is triggered. The algorithm employs an incremental learning strategy, updating the network weights with small steps using only the new samples obtained in the current scan, with a learning rate set to 0.001 to avoid catastrophic forgetting. The updated model takes effect immediately and is used in subsequent control cycles. This mechanism can adapt to the slow drift of phase-matching characteristics in nonlinear crystals caused by thermal lensing effects, optical damage, or material aging, maintaining system performance.

[0043] The synchronization of the entire feedback loop is guaranteed by the master clock source. The trigger signals of the spectrometer and the wavelength meter are synchronized by the same master clock source, whose jitter is less than 10 picoseconds and whose frequency stability is better than 10. -9 The central controller is implemented using a field-programmable gate array (FPGA) chip, which integrates a spectral centroid calculation unit, a deviation generation unit, a neural network inference engine, and a digital-to-analog conversion control unit. The neural network inference engine employs a pipelined architecture, storing weights in block random access memory (BRAM), and the computation unit multiplexes a multiplier-accumulator, resulting in a single inference latency of less than 200 microseconds. Combined with the response time of spectral acquisition and the actuator, the entire closed-loop bandwidth is greater than 500 Hz, sufficient to suppress the low-frequency drift and mid-frequency jitter common in femtosecond lasers.

[0044] Using the above method, the long-term drift of the signal light wavelength is controlled within 50 picometers, and the short-term jitter standard deviation is less than 5 picometers. This performance is significantly superior to existing technologies that employ passive temperature control or open-loop angle setting, providing reliable technical support for applications with stringent wavelength stability requirements, such as ultrafast spectroscopy, coherent anti-Stokes Raman scattering imaging, and quantum light sources.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for wavelength-stable output of a femtosecond laser-pumped optical parametric amplifier, characterized by, include: Polarization stabilization of femtosecond pump laser; The output beam of the femtosecond pump laser was sampled by a spectrometer to obtain its complete spectral intensity distribution data. Based on the spectral intensity distribution data, the spectral centroid frequency of the pump laser is calculated, and the spectral centroid frequency is defined as the spectral intensity weighted average frequency. The center wavelength of the signal light output by the optical parametric amplifier is monitored in real time using a wavelength meter and converted into the corresponding signal light frequency. The pump frequency deviation is obtained by comparing the spectral centroid frequency of the pump laser with the preset target pump frequency. The signal light frequency is compared with the preset target signal light frequency to obtain the signal light frequency deviation. The pump frequency deviation and the signal light frequency deviation are input together into a pre-trained phase matching compensation model, which outputs the phase matching angle correction amount required for the nonlinear crystal under the current deviation state. Based on the phase matching angle correction, a high-precision piezoelectric ceramic rotating platform mounted on the nonlinear crystal base is driven to finely adjust the incident angle of the nonlinear crystal in real time, so that the pump light, signal light and idler light satisfy the momentum conservation condition.

2. The method of claim 1, wherein the femtosecond laser pumped optical parametric amplifier is a wavelength stabilized output method, characterized in that, Polarization stabilization of femtosecond pump lasers includes: By combining a Faraday isolator with a quarter-wave plate, the interference of back-reflected light on the pump laser resonator is eliminated, ensuring that the polarization direction of the pump light maintains a fixed angle with the principal axis of the nonlinear crystal.

3. The method of claim 2, wherein the femtosecond laser pumped optical parametric amplifier is a wavelength stabilized output method, characterized in that, The spectral centroid frequency of the pump laser is calculated based on the spectral intensity distribution data, including: Background subtraction and nonlinear response correction are performed on the discrete wavelength channel intensity data output by the spectrometer; According to the formula The spectral centroid frequency is calculated , The center frequency of the first channel, The corresponding intensity value, The total number of effective channels.

4. The method of claim 3, wherein the femtosecond laser pumped optical parametric amplifier is a wavelength stabilized output method, characterized in that, The signal light center wavelength is monitored in real time using a wavelength meter and converted into the signal light frequency, including: A wavelength meter based on the Michelson or Fabry-Perot interferometry principle is used; The measured center wavelength of the signal light Through the speed of light constant Convert to frequency .

5. The wavelength-stabilized output method of the femtosecond laser-pumped optical parametric amplifier according to claim 4, characterized in that, The phase matching compensation model adopts a multi-layer feedforward neural network structure. The input layer contains two nodes corresponding to the pump frequency deviation and the signal light frequency deviation, respectively. The hidden layer contains three nodes. The activation function is the hyperbolic tangent function. The output layer contains the phase matching angle correction corresponding to the nodes.

6. The wavelength-stabilized output method of the femtosecond laser-pumped optical parametric amplifier according to claim 5, characterized in that, The phase-matching compensation model is trained using an offline calibration dataset during the system initialization phase. The calibration dataset generates a known frequency drift by actively perturbing the pump laser cavity length or temperature, and records the corresponding optimal phase-matching angle adjustment value of the nonlinear crystal to form input-output sample pairs.

7. The wavelength-stabilized output method of the femtosecond laser-pumped optical parametric amplifier according to claim 6, characterized in that, The piezoelectric ceramic rotating platform is driven according to the phase matching angle correction, including: The nonlinear crystal is mounted on a 5D adjustment frame, and its rotational degree of freedom about the axis perpendicular to the incident surface is controlled by a piezoelectric ceramic actuator. The central controller generates a driving voltage through a digital-to-analog converter, which is then amplified by a high voltage and applied to the piezoelectric ceramic to achieve fine-tuning of the angle.

8. The wavelength-stabilized output method of the femtosecond laser-pumped optical parametric amplifier according to claim 7, characterized in that, It also includes environmental parameter compensation steps: Ambient temperature and air pressure are collected using temperature and humidity sensors and a barometer; Environmental parameters are input into the atmospheric dispersion modifier model to calculate the correction amount for the effective refractive index of the nonlinear crystal. Based on the aforementioned correction amount, an additional correction amount for the phase matching angle is derived and superimposed with the main correction amount to form the final drive command.

9. The wavelength-stabilized output method of the femtosecond laser-pumped optical parametric amplifier according to claim 8, characterized in that, The method also includes periodically performing online updates to the phase-matching compensation model: At preset time intervals, the nonlinear crystal is controlled to perform step angle scanning near the current angle; Record the angular position corresponding to the peak value of the signal light output intensity as the measured optimal angle; The measured optimal angle is compared with the model's predicted angle. If the deviation is greater than a preset threshold, the incremental learning algorithm is triggered to fine-tune the neural network weights.

10. The wavelength-stabilized output method of the femtosecond laser-pumped optical parametric amplifier according to claim 9, characterized in that, The central controller is implemented using a field-programmable gate array (FPGA) chip, which integrates a spectral centroid calculation unit, a deviation generation unit, a neural network inference engine, and a digital-to-analog conversion control unit. The neural network inference engine adopts a pipelined structure.

Images