Femtosecond Pulse Single-Frame Measurement Method, System and Medium Based on Neural Network

Abstract

The present invention provides a femtosecond pulse single-frame measurement method, system and medium based on a neural network, relating to the field of ultrafast laser technology. The method includes: Step S1: The laser pulse output by the femtosecond pulse laser is first broadened through the first dispersion medium; Step S2: Record the pulse intensity data after the first broadening; Step S3: The broadened laser pulse is secondarily broadened through the second dispersion medium; Step S4: Record the pulse intensity data after the second broadening; Step S5: Input the pulse intensity data after the first broadening and the second broadening into the neural network for training to restore the pulse intensity information and phase information at the target point. The present invention can solve the problems of edge phase loss in the intensity transport equation (TIE) algorithm and extremely high requirements for the device sampling rate and bandwidth by increasing dispersion broadening and combining neural network recognition, and realize single-frame global measurement of femtosecond pulses.

Description

Technical Field

[0001] The present invention relates to the field of ultrafast laser technology, and in particular, to a femtosecond pulse single-frame measurement method, system and medium based on a neural network. Background Technique

[0002] Femtosecond laser pulses have important application prospects in fields such as fine processing and precise measurement. The accurate measurement of femtosecond laser pulses has become a very important research content. Researchers first thought of using femtosecond optical pulses themselves as measurement tools and developed femtosecond optical pulse measurement methods such as the autocorrelation method, the Frequency-Resolved Optical Gating (FROG) method, and the Spectral Phase Interferometry for Direct Electric-field Reconstruction (SPIDER) method based on the interference between two femtosecond optical pulses. Among them, the autocorrelation method cannot obtain the amplitude and phase of the pulse and can only roughly obtain the pulse width; both FROG and SPIDER can perform global measurement of femtosecond pulses, that is, they can measure the amplitude and phase of the pulse.

[0003] However, commercially available autocorrelators, FROG and SPIDER devices all adopt the method of multi-frame averaging and cannot meet the requirements of single-frame global measurement of femtosecond pulses. In addition, traditional autocorrelation methods, FROG and SPIDER all utilize nonlinear effects and cannot directly measure low-energy pulses. Low-energy pulses output by femtosecond seed sources often need to be measured after optical amplification, and the pulses have changed during the amplification process, resulting in inaccurate measurement results. In recent years, researchers have proposed using the Time Stretch-Dispersive Fourier Transform (TS-DFT) technology and the time lens technology, combined with the Gerchberg–Saxton iterative algorithm to perform global measurement on picosecond pulses. This method is similar to FROG and also has the disadvantages of requiring nonlinearity and slow phase recovery of the iterative algorithm. Therefore, the single-frame global measurement of femtosecond pulses without nonlinear effects and without iterative algorithms is crucial for accurately measuring low-energy femtosecond pulse sequences.

[0004] In the prior art, when recovering the femtosecond pulse phase based on the Transport-of-Intensity Equation (TIE), by solving the differential of the signal phase according to the change of the signal intensity with the transmission distance to obtain the signal phase, there is a problem of edge phase distortion. When the pulse is narrow, the recovery error will be relatively large because a narrower pulse corresponds to a wider spectrum and is more severely broadened under the same dispersion parameter. Therefore, there is more phase information in the edge part, and the missing edge phase has a greater impact. On the other hand, the femtosecond pulse phase recovery based on time-domain TIE has extremely high requirements for the sampling rate and bandwidth of the device (requiring the order of TSa / s and THz), and the current real-time oscilloscopes cannot meet the requirements, so single-frame femtosecond pulse measurement cannot be achieved.

[0005] The invention patent with the publication number CN111006777B discloses a method and device for measuring femtosecond pulses, including: generating high-order harmonics in two channels through the interaction of a pulsed laser driving field and a rare gas; interfering the high-order harmonics in the two channels; introducing the femtosecond pulse to be measured into the process of the interaction between the pulsed laser and the rare gas to change the accumulated phase of the high-order harmonics generated in the two channels, so that the high-order harmonics in each channel generate energy drift; determining the function of the energy drift of the high-order harmonics in each channel changing with the delay according to the energy drift information generated by the high-order harmonics in the two channels respectively, and determining the structural information of the first component of the femtosecond pulse to be measured according to the function; the first component is the component of the femtosecond pulse to be measured with the same polarization direction as the pulsed laser driving field. Summary of the Invention

[0006] Aiming at the defects in the prior art, the present invention provides a single-frame measurement method, system and medium for femtosecond pulses based on a neural network.

[0007] According to a single-frame measurement method, system and medium for femtosecond pulses based on a neural network provided by the present invention, the solution is as follows:

[0008] In the first aspect, a single-frame measurement method for femtosecond pulses based on a neural network is provided, and the method includes:

[0009] Step S1: The laser pulse output by the femtosecond pulse laser is first broadened through the first dispersion medium.

[0010] Step S2: Record the pulse intensity data after the first broadening.

[0011] Step S3: The broadened laser pulse is secondarily broadened through the second dispersion medium.

[0012] Step S4: Record the pulse intensity data after the second broadening.

[0013] Step S5: Input the pulse intensity data after the first broadening and the second broadening into a neural network for training to recover the pulse intensity information and phase information at the target point.

[0014] Preferably, the laser pulse output by the femtosecond pulse laser passes through two gratings in sequence to obtain dispersion stretching.

[0015] Preferably, under the real-time recording of a GHz instrument, the time-domain intensity data of the femtosecond pulse after dispersion stretching is obtained as the input data of the neural network.

[0016] Preferably, the training label of the neural network is the intensity and phase data of the femtosecond laser pulse measured by a FROG instrument.

[0017] In a second aspect, a femtosecond pulse single-frame measurement system based on a neural network is provided. The system includes:

[0018] Module M1: The laser pulse output by the femtosecond pulse laser is first broadened through a first-stage dispersion medium.

[0019] Module M2: Record the pulse intensity data after the first broadening.

[0020] Module M3: The broadened laser pulse is secondarily broadened through a second-stage dispersion medium.

[0021] Module M4: Record the pulse intensity data after the second broadening.

[0022] Module M5: Input the pulse intensity data after the first broadening and the second broadening into a neural network for training to recover the pulse intensity information and phase information at the target point.

[0023] Preferably, the laser pulse output by the femtosecond pulse laser passes through two gratings in sequence to obtain dispersion stretching.

[0024] Preferably, under the real-time recording of a GHz instrument, the time-domain intensity data of the femtosecond pulse after dispersion stretching is obtained as the input data of the neural network.

[0025] Preferably, the training label of the neural network is the intensity and phase data of the femtosecond laser pulse measured by a FROG instrument.

[0026] In a third aspect, a computer-readable storage medium storing a computer program is provided. When the computer program is executed by a processor, the steps in the method are implemented.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. The present invention changes the characteristics of traditional algorithms, such as slow phase recovery and missing edge phase in non-linear and iterative algorithms, by increasing dispersion and combining with neural networks, achieving fast, accurate, and intelligent single-frame global measurement of femtosecond pulses.

[0029] 2. After the femtosecond pulse undergoes significant dispersion broadening in the present invention, the pulse width is significantly increased. The intensity data of the time-domain waveform after two broadenings can be collected separately using an oscilloscope or an analog-to-digital converter, filling the defect that the TIE algorithm has extremely high requirements for the sampling rate and bandwidth of the device.

[0030] 3. The data is input into the neural network for training. The trained neural network can output the phase of the femtosecond pulse and the shape of the femtosecond pulse according to the input intensity signal, filling the defect of missing edge phase in the TIE algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Other features, objects, and advantages of the present invention will become more apparent by reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0032] Figure 1 It is a schematic diagram of a femtosecond pulse phase recovery scheme based on time-domain TIE;

[0033] Figure 2 It is a target chirp-free pulse;

[0034] Figure 3 It is the phase recovered using TIE after passing through L1;

[0035] Figure 4 It is the finally recovered pulse;

[0036] Figure 5 It is a schematic diagram of a femtosecond pulse single-frame measurement scheme based on neural network. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0037] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0038] The embodiment of the present invention provides a method for single-frame measurement of femtosecond pulses based on neural network, aiming to solve the problems of missing edge phase in the TIE algorithm and extremely high requirements for the sampling rate and bandwidth of the device by increasing dispersion broadening and combining with neural network recognition, realizing single-frame global measurement of femtosecond pulses.

[0039] For the femtosecond pulse phase recovery based on TIE, the schematic diagram is referred to Figure 1 as shown.

[0040] The principle of TIE is that when the signal undergoes the Fresnel Transform during transmission (the dispersion process is a kind of Fresnel Transform), there is a relationship shown in Equation (1) between the change of the signal intensity with the transmission distance and the phase differential:

[0041]

[0042] In Equation (1), I is the intensity, the phase to be solved, t is the time, z is the transmission distance, and β2 is the Group Velocity Dispersion (GVD) of the Fresnel transform medium. By measuring the intensity before and after transmission, the phase can be inversely calculated.

[0043] Therefore, for a femtosecond laser pulse, it passes through a pre - stretching optical fiber (marked as L1) and an optical fiber for performing the Fresnel transform (marked as L2) in sequence, and the intensity information is measured respectively, denoted as I1 and I2.

[0044] According to what is shown in Equation (2):

[0045]

[0046] The one - dimensional TIE can be approximately solved to obtain the phase after passing through the pre - stretching optical fiber. Given the dispersion parameters of the pre - stretching optical fiber, the target pulse can be inversely calculated.

[0047] Since the amplitudes on both sides of the pulse intensity are very low (close to 0), TIE cannot extract phase information from them, but phase information still exists in these weakly - intense areas. When using the traditional Fourier algorithm to solve the time - domain TIE, a normalized edge truncation threshold is often set to constrain the pulse edge (that is, the part below the edge truncation threshold is forced to become the edge truncation threshold). Therefore, when using TIE for phase recovery, there is a problem of edge phase distortion. When the pulse is narrower, the recovery error will be larger because a narrower pulse corresponds to a wider spectrum and is more severely broadened under the same dispersion parameters, so there is more phase information in the edge part and the influence of missing edge phase is greater.

[0048] On the other hand, the pulse phase recovery based on time - domain TIE has extremely high requirements for the device sampling rate and bandwidth (requiring the order of TSa / s and THz), and the current real - time oscilloscopes cannot meet the requirements and cannot achieve single - frame femtosecond pulse measurement. Theoretically, the pulse can be broadened more by increasing the dispersion of L1 to reduce the requirements for the device sampling rate and bandwidth. However, after the dispersion of L1 becomes larger, the TIE solution process is highly correlated with the edge truncation threshold and shows a pathological trend. The simulation diagram is referred to Figure 2 、 3, as shown in Figure 4, can well demonstrate this problem. As shown in Table 1 below, after increasing the dispersion and reducing the sampling rate to 20 GSa / s, even if the normalized edge truncation threshold only changes by one-thousandth, the error of pulse recovery is already very large.

[0049] Table 1.

[0050]

[0051] When using the traditional Fourier algorithm to solve the TIE for pulse recovery, reducing the sampling rate by increasing the dispersion will cause the recovery accuracy to be highly correlated with the edge truncation threshold.

[0052] A femtosecond pulse single-frame measurement method based on neural network proposed by the present invention has a process as Figure 5 shown, and the specific steps include:

[0053] Step S1: The laser pulse output by the femtosecond pulse laser passes through a section of dispersion medium (which can be a single-mode fiber or a grating). The purpose is to broaden the laser pulse to a width that can be measured in a single frame by a real-time oscilloscope.

[0054] Step S2: Record the pulse intensity data I1(t) after broadening.

[0055] Step S3: The broadened laser pulse passes through a second section of dispersion medium (which can be a single-mode fiber or a grating) for a second broadening. The purpose is to make the laser pulse signal undergo a Fresnel transform during transmission to obtain the relationship between the intensity change with the transmission distance and the phase differential.

[0056] Step S4: Record the pulse intensity data I2(t) after broadening.

[0057] Step S5: Input the pulse intensity data after two broadenings into the neural network for training, and the neural network restores the pulse intensity information and phase information

[0058] Among them, after the femtosecond pulse is broadened by the first dispersion, the pulse width is significantly increased. The purpose is to reduce the requirements for the sampling rate and bandwidth of the device, and to enable the acquisition of the intensity data of the time-domain waveform after two broadenings through a real-time oscilloscope or an analog-to-digital converter (ADC). Utilizing the non-linear fitting ability of the neural network, directly inputting the time-domain intensity data into the neural network to restore the pulse intensity information and phase information at the target point, without introducing an edge truncation threshold and combining with the Fourier algorithm for calculation, which may lead to a reduction in recovery accuracy. Therefore, this solution can fill the defect that the TIE algorithm has extremely high requirements for the sampling rate and bandwidth of the device.

[0059] The data is input into a neural network for training. The trained neural network can output the phase of the femtosecond pulse and the shape of the femtosecond pulse according to the input intensity signal. By utilizing the powerful feature extraction ability and non-linear fitting ability of the neural network, the defect of the lost edge phase in the TIE algorithm can be filled.

[0060] Step S3: The neural network outputs the phase information of the femtosecond laser pulse and the time-domain intensity information I(t), realizing single-frame measurement of the femtosecond pulse.

[0061] The femtosecond laser pulse output by the fiber laser is successively subjected to dispersion stretching through two gratings. Grating 1 has a larger dispersion and can broaden the femtosecond laser to a width that can be measured by a GHz instrument (such as a GHz-level oscilloscope or a data acquisition card or a spectrometer); Grating 2 has a smaller dispersion and realizes the Fresnel transform.

[0062] Under the real-time recording of the GHz instrument, the time-domain intensity data I1(t) and I2(t) of the femtosecond pulse after dispersion stretching are obtained as the input data of the neural network, and the training label is the intensity and phase data of the femtosecond laser pulse measured by the FROG instrument. The output result of the neural network is the predicted intensity and phase signals of the femtosecond laser pulse.

[0063] Next, the present invention will be described in more detail.

[0064] The implementation principle of the present invention: First, the laser pulse output by the femtosecond pulse laser passes through a dispersion medium (which can be a single-mode fiber or a grating), and the laser pulse is broadened to a width that can be measured in a single frame by a real-time oscilloscope. The intensity data of the broadened pulse is recorded by an oscilloscope or an ADC. The broadened laser pulse passes through a second dispersion medium (which can be a single-mode fiber or a grating) for a second broadening, so that the laser pulse signal undergoes a Fresnel transform during transmission, obtaining the relationship between the intensity change with the transmission distance and the phase differential. The intensity data of the pulse after the second broadening is recorded by an oscilloscope or an ADC. And the intensity data of the pulse after the two broadenings are input into the neural network for training, and the neural network restores the pulse intensity information and phase information at the target point.

[0065] A femtosecond pulse single-frame measurement method, system and medium based on a neural network provided by an embodiment of the present invention solve the problems of edge phase loss in the TIE algorithm and extremely high requirements for the sampling rate and bandwidth of equipment by increasing dispersion broadening and combining neural network recognition, and realize single-frame global measurement of femtosecond pulses. The laser pulses output by the femtosecond pulse laser are broadened twice through two sections of dispersion media, and the intensity data of the time-domain waveforms after the two broadenings are respectively collected by an oscilloscope or an ADC and input into the neural network for phase recognition, so that the phase and shape of the femtosecond pulses can be restored, providing a new single-frame global (i.e., including amplitude and phase) measurement scheme for femtosecond pulses.

[0066] Those skilled in the art know that in addition to implementing the system and its various devices, modules, and units provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to enable the system and its various devices, modules, and units provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc. to achieve the same functions. Therefore, the system and its various devices, modules, and units provided by the present invention can be regarded as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structures within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as both software modules for implementing the method and the structures within the hardware component.

[0067] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A femtosecond pulse single-frame measurement method based on a neural network, characterized in that Including: Step S1: The laser pulses output by the femtosecond pulsed laser are first broadened through the first dispersion medium. Step S2: Record the pulse intensity data after the first broadening. Step S3: The broadened laser pulses are secondarily broadened through the second dispersion medium. Step S4: Record the pulse intensity data after the second broadening. Step S5: Input the pulse intensity data after the first broadening and the second broadening into a neural network for training to restore the pulse intensity information and phase information at the target point. The laser pulses output by the femtosecond pulsed laser are dispersion-stretched successively through two gratings. Among them, grating 1 has a larger dispersion, broadening the femtosecond laser to a width that can be measured by a GHz instrument; grating 2 has a smaller dispersion to achieve a Fresnel transform. Through real-time recording by the GHz instrument, the time-domain intensity data I1(t) and I2(t) of the dispersion-stretched femtosecond pulses are obtained as the input data of the neural network. The training labels of the neural network are the intensity and phase data of the femtosecond laser pulses measured by the FROG instrument, and the output result of the neural network is the predicted intensity and phase signals of the femtosecond laser pulses.

2. A femtosecond pulse single-frame measurement system based on a neural network, characterized in that, Including: Module M1: The laser pulses output by the femtosecond pulsed laser are first broadened through the first dispersion medium. Module M2: Record the pulse intensity data after the first broadening. Module M3: The broadened laser pulses are secondarily broadened through the second dispersion medium. Module M4: Record the pulse intensity data after the second broadening. Module M5: Input the pulse intensity data after the first broadening and the second broadening into a neural network for training to restore the pulse intensity information and phase information at the target point. The laser pulses output by the femtosecond pulsed laser are dispersion-stretched successively through two gratings. Among them, grating 1 has a larger dispersion, broadening the femtosecond laser to a width that can be measured by a GHz instrument; grating 2 has a smaller dispersion to achieve a Fresnel transform. Through real-time recording by the GHz instrument, the time-domain intensity data I1(t) and I2(t) of the dispersion-stretched femtosecond pulses are obtained as the input data of the neural network. The training labels of the neural network are the intensity and phase data of the femtosecond laser pulses measured by the FROG instrument, and the output result of the neural network is the predicted intensity and phase signals of the femtosecond laser pulses.

3. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in claim 1.

Images