2-micron intelligent ultrafast laser spectrum real-time light detection device and method

Abstract

The invention provides a 2-micron intelligent ultrafast laser spectrum real-time light detection device and method. The 2-micron intelligent ultrafast laser spectrum real-time light detection device comprises a signal source module, a relay amplification module, a detection analysis module and an intelligent algorithm regulation and control module. The signal source module outputs intermediate infrared laser pulses; the relay amplification module realizes signal gain and dispersion broadening; the genetic algorithm sub-module dynamically optimizes transmission parameters such as pumping power and optical fiber length; the convolutional neural network algorithm sub-module carries out real-time feature recognition on the time-frequency signal output by the photoelectric detector and maps a time-domain waveform. And the detection and analysis module is used for intelligently monitoring the time-frequency composite signal. The problem of large intermediate infrared laser long-distance transmission loss is solved through the genetic algorithm, real-time analysis of a soliton pulse fine structure is realized by replacing a traditional spectrometer with the convolutional neural network, and time-frequency synchronous measurement of ultrafast laser in the 2-micron waveband can be completed without a complex test instrument. The system has the characteristics of adaptive parameter adjustment, high measurement precision and strong system expansibility.

Description

Technical Field

[0001] This invention relates to the field of spectroscopic analysis instrument technology, specifically to a 2-micron intelligent ultrafast laser spectroscopy real-time optical detection device and method. Background Technology

[0002] The mid-infrared band (especially 1.9μm-2.1μm, i.e., around 2μm) corresponds to the characteristic vibrational energy levels of various molecules such as methane, carbon dioxide, and water molecules, and has irreplaceable application value in fields such as environmental monitoring, medical diagnosis, materials science, and national security. Among them, ultrafast laser pulses in the 2μm band (covering nanosecond to femtosecond levels) have become a core tool in cutting-edge scientific research and industrial detection due to their high temporal resolution and strong peak power, enabling precise detection of transient physicochemical processes of matter (such as the evolution of chemical reaction intermediates and molecular vibrational relaxation).

[0003] However, existing ultrafast mid-infrared spectroscopy measurement technology in the 2μm band still faces many bottlenecks: when mid-infrared lasers in this band are transmitted over long distances through optical fibers, significant losses occur due to material absorption and scattering, which traditional single-stage amplification cannot effectively compensate for; at the same time, static settings of parameters such as pump power and fiber length cannot dynamically adapt to signal changes, easily causing pulse distortion (such as waveform distortion and abnormal spectral broadening), which seriously affects measurement accuracy. In addition, the soliton characteristics of ultrafast laser pulses (such as Kelly sidebands, dissipative soliton spectra, and soliton molecular pairs) are key to reflecting their physical nature, but traditional spectrometers rely on optical components such as gratings and interferometers, which not only result in large system size and high cost, but also lag in the resolution speed of transient fine structures (mostly on the order of milliseconds or more), failing to meet the real-time monitoring requirements of femtosecond-level dynamic processes. In terms of time-frequency synchronization measurement, the simultaneous measurement of the time domain (pulse width, repetition rate) and frequency domain (spectral distribution) of 2μm ultrafast lasers requires the use of multiple devices such as oscilloscopes, spectrometers, and interferometers. This results in low system integration, complex operation, and synchronization errors between devices further reduce measurement accuracy. Furthermore, parameters such as pump power and fiber dispersion in existing systems require manual adjustment, relying excessively on operator experience and making rapid optimization difficult in complex environments (such as temperature fluctuations and signal attenuation), leading to poor system adaptability and insufficient stability.

[0004] To address the aforementioned challenges, the development of a spectrometer capable of achieving low-loss transmission of ultrafast mid-infrared laser in the 2μm band, real-time analysis of fine structures, and time-frequency synchronized intelligent measurement has become an essential requirement. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to provide a 2-micron intelligent ultrafast laser spectroscopy real-time optical detection device and method, which enables real-time mapping of time and frequency signals, providing an efficient, intelligent, and integrated solution for ultrafast mid-infrared spectroscopy measurement in the 2μm band.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A 2-micron intelligent ultrafast laser spectroscopy real-time optical detection device includes:

[0008] The signal source module is used to output mid-infrared laser pulse signals with a wavelength range of 1.9μm-2.1μm;

[0009] The repeater amplification module contains three amplification units. Each amplification unit is connected in sequence to a pump source, a pump combiner, a 3-5m thulium-holmium co-doped fiber, a 2-3km single-mode fiber, and a polarization-independent isolator to achieve signal gain compensation and dispersion broadening.

[0010] The detection and analysis module, including a photodetector and an oscilloscope, is used to complete photoelectric signal conversion and real-time monitoring;

[0011] The intelligent algorithm control module includes a genetic algorithm submodule and a convolutional neural network algorithm submodule: the genetic algorithm submodule is used to dynamically optimize the pump power, the length of the thulium-holmium co-doped fiber, and the dispersion parameters of the single-mode fiber to balance the gain and loss during signal transmission; the convolutional neural network algorithm submodule is used to perform feature recognition on the time-frequency signal output by the photodetector, map the spectral information into a time-domain waveform, and display it in real time via an oscilloscope.

[0012] Preferably, the pulse modes output by the signal source module cover Q-switched pulses, conventional soliton pulses, dissipative soliton pulses, broadened soliton pulses, and soliton molecule pair pulses, with pulse widths ranging from nanoseconds to femtoseconds and repetition rates of 1kHz to 10GHz.

[0013] Preferably, the wavelength of the pump source is 793nm, the operating wavelength range of the pump combiner and the polarization-independent isolator is adapted to 2μm, and the single-mode fiber is an optical fiber with dispersion broadening function to realize dispersion Fourier transform.

[0014] Preferably, the optimization objective of the genetic algorithm submodule is to determine the optimal length of the thulium-holmium co-doped fiber and the optimal length of the single-mode fiber through iterative calculation, so that the signal transmission loss is controlled within the range of ≤3dB / km, while avoiding pulse distortion.

[0015] Preferably, the convolutional neural network algorithm submodule is pre-trained in a time-frequency feature library containing Kelly sidebands, dissipative soliton spectra, and fine structure of soliton molecule pairs, with a real-time recognition bandwidth of 100MHz-10GHz and a recognition accuracy of ≥95%.

[0016] Preferably, the wavelength response range of the photodetector is 1.8μm-2.1μm, the sensitivity is 0.5-1A / W, and the oscilloscope integrates a convolutional neural network algorithm submodule to display a time-domain waveform that is identical to the incident signal spectrum.

[0017] The fabrication method of a 2-micron intelligent ultrafast laser spectroscopy real-time optical detection device includes the following steps:

[0018] S1. Based on the design parameters, use 3D modeling software to construct 3D models of the signal source module, relay amplification module, and detection and analysis module, and perform virtual assembly to check the compatibility of each component.

[0019] S2. Assemble the relay amplification module: According to the composition order of each amplification unit, connect and fix the pump source, pump combiner, thulium-holmium co-doped fiber, single-mode fiber and polarization-independent isolator in sequence to ensure optical alignment accuracy. After assembling the three amplification units, connect them in a cascade manner.

[0020] S3. Optically couple the output of the signal source module to the input of the pump combiner of the first group of relay amplifier modules to ensure efficient transmission of optical signals.

[0021] S4. Assemble the detection and analysis module: Connect the signal output terminal of the photodetector to the signal input terminal of the oscilloscope through a high-frequency coaxial cable to ensure the integrity of the electrical signal transmission;

[0022] S5. Deploy intelligent algorithm control module: Load genetic algorithm submodule and convolutional neural network algorithm submodule into industrial control computer, and establish communication connection with pump source of relay amplification module and oscilloscope of detection and analysis module through data interface to realize linkage between parameter control and signal analysis;

[0023] S6. Fix the hardware carriers of the signal source module, the assembled relay amplification module, the detection and analysis module and the intelligent algorithm control module on the integrated frame according to the positional relationship of the virtual assembly. Adjust the relative positions of each module through a precision calibration tool to ensure the coaxiality and stability of the optical path.

[0024] S7. Power on and debug the entire system: Start the signal source module to output mid-infrared laser pulses, optimize the pump power parameters through the genetic algorithm submodule of the intelligent algorithm control module, observe the time-domain waveform displayed on the oscilloscope of the detection and analysis module, verify the accuracy and real-time performance of the spectral analysis, and complete the production.

[0025] Preferably, the optimization process of the genetic algorithm submodule in step S2 is as follows: construct a mathematical model of transmission loss and pulse distortion, and use selection, crossover, and mutation operations to iteratively generate the optimal parameter combination so that the signal energy fluctuation after multi-level relay is controlled within ≤5%.

[0026] Preferably, the real-time recognition process of the convolutional neural network algorithm submodule in step S4 is as follows: convert the electrical signal into a time-domain sequence, extract spectral features through convolutional layers and pooling layers, and map them into a time-domain waveform with a spectral similarity of ≥98% to the incident signal.

[0027] Preferably, the method is adapted to the wavelength range of 800nm-2100nm. By adjusting the pump wavelength optimization parameters of the genetic algorithm submodule and the feature library of the convolutional neural network algorithm submodule, it is adapted to ultrafast laser sources in the near-infrared to mid-infrared band.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. High temporal resolution characteristics

[0030] The 2μm mid-infrared intelligent spectrometer involved in this patent possesses excellent high temporal resolution. Its detection and analysis module integrates a convolutional neural network algorithm submodule into the oscilloscope, enabling real-time monitoring of time-frequency composite signals. This submodule performs real-time feature recognition on the time-frequency signal output from the photodetector, with a real-time recognition bandwidth of 100MHz-10GHz. It can quickly map spectral information into a time-domain waveform and display it in real time on the oscilloscope. This real-time processing and display capability allows the spectrometer to rapidly respond to and capture the time-frequency changes of ultrafast lasers in the 2μm band, achieving a "fast measurement" effect and meeting the demand for rapid measurement of ultrafast laser pulses.

[0031] 2. High measurement accuracy characteristics

[0032] This spectrometer exhibits excellent measurement accuracy, achieving "precise measurement." The convolutional neural network algorithm submodule is pre-trained in a time-frequency feature library containing Kelley sidebands, dissipative soliton spectra, and fine structure data of soliton molecules, achieving a recognition accuracy of ≥95%. During use, when the convolutional neural network algorithm submodule performs time-frequency feature recognition on electrical signals, it can map time-domain waveforms with a spectral similarity of ≥98% to the incident signal. Simultaneously, the genetic algorithm submodule dynamically optimizes transmission parameters such as pump power and fiber length, balancing gain and loss to control signal transmission loss within ≤3dB / km, avoiding pulse distortion. Signal energy fluctuations after multi-stage relays are controlled within ≤5%, further ensuring measurement accuracy.

[0033] 3. Wide spectral measurement range characteristic

[0034] This spectrometer boasts a wide measurement range, meaning it offers broad measurement coverage. The signal source module can output mid-infrared laser pulse signals with wavelengths ranging from 1.9 μm to 2.1 μm. Its output pulse modes include Q-switched pulses, conventional soliton pulses, dissipative soliton pulses, broadened soliton pulses, and soliton-molecule pair pulses. Pulse widths range from nanoseconds to femtoseconds, and repetition rates are from 1 kHz to 10 GHz. Furthermore, its application can be extended to the 800 nm to 2100 nm wavelength range. By adjusting the pump wavelength optimization parameters of the genetic algorithm submodule and the feature library of the convolutional neural network algorithm submodule, it can be adapted to ultrafast laser sources in the near-infrared to mid-infrared bands, significantly broadening its spectral range of applications.

[0035] 4. Intelligent adaptive control characteristics

[0036] This spectrometer possesses intelligent adaptive control capabilities. The genetic algorithm submodule within the intelligent algorithm control module dynamically optimizes pump power, thulium-holmium co-doped fiber length, and single-mode fiber dispersion parameters. Through iterative calculations, it determines the optimal parameter combination to balance gain and loss during signal transmission, achieving adaptive parameter adjustment. The convolutional neural network algorithm submodule performs real-time feature recognition on the time-frequency signal output from the photodetector, mapping spectral information into a time-domain waveform, thus replacing traditional spectrometers to achieve real-time analysis of the fine structure of soliton pulses. This intelligent control allows the system to perform time-frequency synchronous measurements of 2μm ultrafast lasers without complex testing instruments, and the system exhibits strong scalability, demonstrating significant intelligent characteristics. Attached Figure Description

[0037] Figure 1 This is a structural framework diagram of the present invention;

[0038] Figure 2 This is a schematic diagram of the relay amplification module device of the present invention;

[0039] Figure 3 This is a schematic diagram of the intelligent algorithm control module device of the present invention;

[0040] Figure 4 This is the spectral signal of the 1.5μm ultrafast soliton laser source in Example 1 of the present invention;

[0041] Figure 5 This is the oscilloscope signal of the 1.5μm band ultrafast soliton laser source in Embodiment 1 of the present invention;

[0042] Figure 6 This is a noise-like mode-locking method for the 2μm band ultrafast soliton laser source in Example 2 of the present invention, where (a) is a linear form and (b) is a logarithmic form;

[0043] Figure 7This refers to the soliton mode-locking of the 2μm band ultrafast soliton laser source in Example 2 of the present invention, where (a) is in linear form and (b) is in logarithmic form.

[0044] in:

[0045] 01. Signal Source Module; 02. Relay Amplification Module; 03. Detection and Analysis Module; 04. Intelligent Algorithm Control Module;

[0046] 021. Pump source; 022. Pump combiner; 023. 3-5m thulium-holmium co-doped fiber; 024. 2-3km single-mode fiber; 025. Polarization-independent isolator; 031. Photodetector; 032. Oscilloscope;

[0047] 041. Genetic Algorithm Submodule; 042. Convolutional Neural Network Algorithm Submodule. Detailed Implementation

[0048] The invention will now be further described with reference to the accompanying drawings.

[0049] like Figures 1 to 7 As shown, a 2-micron intelligent ultrafast laser spectral real-time optical detection device includes:

[0050] Signal source module 01: Used to output mid-infrared laser pulse signals with a wavelength range of 1.9μm-2.1μm, providing the raw detection signal for the entire spectral measurement system.

[0051] The repeater amplification module 02 comprises three amplification units, each connected sequentially to a pump source 021, a pump combiner 022, a 3-5m thulium-holmium co-doped fiber 023, a 2-3km single-mode fiber 024, and a polarization-independent isolator 025. The function of this module is to perform gain compensation on the laser pulses output from the signal source module to offset transmission losses, and to achieve dispersion broadening of the signal through the single-mode fiber, preparing for subsequent time-frequency analysis.

[0052] Detection and analysis module 03: Composed of photodetector 031 and oscilloscope 032. Photodetector 031 is responsible for converting the optical signal processed by the relay amplification module into an electrical signal; oscilloscope 032 is used to receive and display the electrical signal in real time, completing real-time monitoring of the signal.

[0053] The intelligent algorithm control module 04 integrates the genetic algorithm submodule 041 and the convolutional neural network algorithm submodule 042.

[0054] The genetic algorithm submodule 041 is used to dynamically optimize the pump power, the length of the thulium-holmium co-doped fiber 023, and the dispersion parameters of the single-mode fiber 024. By precisely adjusting these transmission parameters, the gain and loss of the signal during transmission are balanced.

[0055] The convolutional neural network algorithm submodule 042 is used to perform real-time feature recognition on the time-frequency signal output by the photodetector 031, extract key spectral feature information, and map the spectral information into a time-domain waveform to achieve rapid spectral analysis.

[0056] Furthermore, the pulse modes output by the signal source module 01 cover Q-switched pulses, traditional soliton pulses, dissipative soliton pulses, broadened soliton pulses, and soliton molecule pair pulses. The pulse width is in the nanosecond to femtosecond range, and the repetition rate is 1kHz-10GHz, which can meet the measurement needs of laser pulse characteristics in different scenarios.

[0057] Furthermore, in the repeater amplification module 02, the pump source 021 has a wavelength of 793nm, and the pump combiner 022 and the polarization-independent isolator 025 have a working wavelength range of 2μm to ensure matching with the signal light wavelength and improve the combining efficiency and isolation effect; the single-mode fiber 024 is an optical fiber with dispersion broadening function, which can realize dispersion Fourier transform and convert spectral information into a time-domain distinguishable signal.

[0058] Furthermore, in the intelligent algorithm control module 04, the optimization objective of the genetic algorithm submodule 041 is to determine the optimal length (3-5m) of the thulium-holmium co-doped fiber 023 and the optimal length (2-3km) of the single-mode fiber 024 through iterative calculation, so that the signal transmission loss is controlled within the range of ≤3dB / km, while avoiding pulse distortion and ensuring signal integrity.

[0059] The convolutional neural network algorithm submodule 042 is pre-trained in a time-frequency feature library containing Kelly sidebands, dissipative soliton spectra, and fine structure of soliton molecules. The trained model has a real-time recognition bandwidth of 100MHz-10GHz and a recognition accuracy of ≥95%, which can accurately capture subtle spectral features.

[0060] Furthermore, in the detection and analysis module 03, the photodetector 031 has a wavelength response range of 1.8μm-2.1μm and a sensitivity of 0.5-1A / W, which can efficiently receive optical signals in the mid-infrared band and convert them into electrical signals; the oscilloscope 032 integrates a convolutional neural network algorithm submodule 042, which can directly display a time-domain waveform that is the same shape as the incident signal spectrum, and realize the visualization of spectral information.

[0061] The fabrication method of a 2-micron intelligent ultrafast laser spectroscopy real-time optical detection device includes the following steps:

[0062] S1. Based on the design parameters, use 3D modeling software to construct 3D models of the signal source module 01, the relay amplification module 02 (pump source 021, pump combiner 022, thulium-holmium co-doped fiber 023, single-mode fiber 024, polarization-independent isolator 024) and the detection and analysis module 03 (photodetector 031, oscilloscope 032), and perform virtual assembly to check the compatibility of each component;

[0063] S2. Assemble the relay amplification module 02: According to the composition order of each amplification unit, connect and fix the pump source 021, pump combiner 022, 3-5m thulium-holmium co-doped fiber 023, 2-3km single-mode fiber 024 and polarization-independent isolator 025 in sequence to ensure optical alignment accuracy. After assembling the three amplification units, connect them in a cascade manner.

[0064] S3. Optically couple the output of the signal source module 01 to the input of the pump combiner 022 of the first relay amplifier module to ensure efficient transmission of optical signals.

[0065] S4. Assemble the detection and analysis module 03: Connect the signal output terminal of the photodetector 031 to the signal input terminal of the oscilloscope 032 through a high-frequency coaxial cable to ensure the integrity of the electrical signal transmission.

[0066] S5. Deploy intelligent algorithm control module 04: Load genetic algorithm submodule 041 and convolutional neural network algorithm submodule 042 into the industrial control computer, and establish communication connections with pump source 021 of relay amplification module 02 and oscilloscope 032 of detection and analysis module 03 through data interface to realize the linkage of parameter control and signal analysis.

[0067] S6. Fix the hardware carrier (industrial control computer) of the signal source module 01, the assembled relay amplification module 02, the detection and analysis module 03 and the intelligent algorithm control module 04 on the integrated frame according to the positional relationship of the virtual assembly. Adjust the relative position of each module through a precision calibration tool to ensure the coaxiality and stability of the optical path.

[0068] S7. Power on and debug the entire system: Start the signal source module to output mid-infrared laser pulses, optimize parameters such as pump power through the genetic algorithm submodule 041 of the intelligent algorithm control module, observe the time-domain waveform displayed on the oscilloscope of the detection and analysis module, verify the accuracy and real-time performance of spectral analysis, and complete the production.

[0069] Example 1

[0070] The existing 1.5μm real-time detection spectrometer is the SPL-COMETⅢ spectrometer. The COMETⅢ contains an embedded dual-channel oscilloscope 032 (bandwidth: 8GHz, sampling rate: 10GS / s) and a high-performance computer (i9 processor, 64GB RAM, 1TB NMVe hard drive).

[0071] The embodiment of this invention is a 1.5μm real-time detection spectrometer, a high-performance computer, pre-installed with VSCode software and Python 3.9 environment, which achieves real-time detection by sending signals to control the amplification ratio.

[0072] The signal source module 01 uses a self-developed 2μm mid-infrared intelligent fiber laser, which can output fiber laser signals with a wavelength range of 1.9μm-2.1μm, covering five modes: Q-switched pulse (pulse width 5ns), traditional soliton pulse (100fs), dissipative soliton pulse (500fs), broadened soliton pulse (2ps), and soliton molecule pair pulse (pulse interval 10ps). The mode can be quickly switched through the pulse mode switching knob (mechanical shutter structure).

[0073] The repetition rate is continuously adjustable in the range of 1kHz-10GHz via a built-in frequency regulator, with femtosecond-level pulse repetition rates ≥1GHz and nanosecond-level pulse repetition rates ≤100kHz, meeting the measurement requirements for pulse characteristics in different scenarios.

[0074] The intermediate amplification module contains three cascaded amplification units, each configured as follows:

[0075] Pump source 021: Uses a 793nm semiconductor laser (model DL-793-500), with continuously adjustable output power from 0-500mW, and wavelength stabilized by a temperature control module (accuracy ±0.1℃);

[0076] Pump combiner 022: It adopts a 2×1 type fiber optic combiner (operating wavelength 1.8μm-2.2μm), with a combining efficiency ≥90%. The input end is connected to the pump source 021 and the output fiber of the signal source module 01 respectively.

[0077] Thulium-holmium co-doped fiber: A 4m long double-clad thulium-holmium co-doped fiber (core diameter 10μm, numerical aperture 0.15) is selected. Signal gain is achieved through thulium ion energy level transitions, and the gain efficiency is improved by 45% compared with traditional erbium-doped fiber.

[0078] Single-mode fiber: 2.5km dispersion-shifted single-mode fiber (dispersion value -12ps / (nm·km)) is used, which has the function of dispersion Fourier transform to linearly map spectral information to the time domain;

[0079] Polarization-independent isolator 025: operating wavelength 1.9μm-2.1μm, isolation ≥35dB, preventing backlight from interfering with pump source 021 and pre-amplifier modules.

[0080] The three amplification units are cascaded via flanges, with a 10cm spacing between adjacent units to ensure the continuity of optical signal transmission.

[0081] The detection and analysis module 03 includes a photodetector 031 and an oscilloscope 032. The photodetector 031 is an indium gallium arsenide avalanche photodetector 031 (model APD-2100), with a response wavelength of 1.8μm-2.1μm, sensitivity of 0.8A / W, dark current ≤10nA, which can convert optical signals into 0-10V electrical signals and a conversion bandwidth ≥12GHz. The oscilloscope 032 is a 10GHz real-time oscilloscope 032 (model DSO-X-10104A), with a sampling rate of 50GS / s, equipped with a 10.1-inch touch screen, and a built-in convolutional neural network algorithm submodule 042 (integrated through firmware). It can directly display time-domain waveforms that are the same shape as the incident spectrum, with a waveform refresh rate ≥1000 frames / second, and supports USB3.0 and Ethernet data output. The photodetector 031 and the oscilloscope 032 are connected by an SMA high-frequency coaxial cable (impedance 50Ω, length 1m). The cable attenuation is ≤0.5dB / m to ensure distortion-free transmission of electrical signals.

[0082] The intelligent control module includes a genetic algorithm submodule 041 and a convolutional neural network algorithm submodule 042;

[0083] The genetic algorithm submodule 041 was written in Python 3.9, employing a roulette wheel selection operator, a single-point crossover operator (crossover probability 0.8), and a mutation operator (mutation probability 0.05), with 50 iterations and a population size of 40. It connects to the pump source 021 power controller via an RS485 interface. The actual working length of the thulium-holmium co-doped fiber is controlled by a fiber length adjustment motor (accuracy ±0.1m). The dynamic optimization objectives are: transmission loss ≤2.8dB / km, signal energy fluctuation ≤4%.

[0084] The convolutional neural network algorithm submodule 042 is trained based on the TensorFlow 2.10 framework. The input layer is a 128×128 pixel time-frequency signal, which includes 3 convolutional layers (3×3 convolutional kernels) and 2 fully connected layers. It is preloaded with a time-frequency feature library (containing 100,000 sets of Kelly sidebands, dissipative soliton spectra, and soliton molecule pair fine structure samples). The recognition bandwidth is 100MHz-10GHz, the recognition accuracy is 96.5%, the similarity with the original spectrum is ≥98.5%, and the recognition delay is ≤8μs.

[0085] The intelligent algorithm control module 04 communicates with the oscilloscope 032 via Ethernet (gigabit bandwidth) and is connected to the pump source 021 of the relay amplifier module 02 via RS485 bus to form a closed-loop control link.

[0086] In the initial state, the signal source module 01 outputs a 2μm Q-switched pulse (repetition rate 10kHz), which is coupled to the 793nm pump light by the pump combiner 022 of the first relay amplification unit and then injected into a 4m thulium-holmium co-doped fiber for gain compensation. The genetic algorithm submodule 041 starts and optimizes the pump power to 320mW through iterative calculation, locking the single-mode fiber dispersion parameter to -12ps / (nm·km), so that the signal transmission loss is stabilized at 2.5dB / km. After the signal is dispersion broadened by the 2.5km single-mode fiber, it is converted into an electrical signal by the photodetector 031. After being received by the oscilloscope 032, the built-in convolutional neural network algorithm submodule 042 identifies the signal characteristics in real time (identification time ≤7μs) and displays the time-domain waveform on the screen. If the pulse mode is switched (such as to soliton molecule pair pulse), the genetic algorithm submodule 041 re-optimizes the parameters within 50ms, and the convolutional neural network algorithm submodule 042 automatically calls the corresponding feature library to ensure that the resolution accuracy is always ≥95%.

[0087] The intermediate amplification module in this embodiment contains three identical cascaded amplification units. Only a single amplification unit needs to be designed to complete the relay amplification of the entire 2μm time-stretched ultrafast mid-infrared intelligent spectrometer. The design, fabrication, and installation steps of the signal source module 01 and the relay amplification in this embodiment are as follows:

[0088] (1) Modeling of each core component using SolidWorks software: The outer shell of signal source module 01 is designed as a cuboid structure (200mm×150mm×100mm), with a reserved laser mounting slot inside (50mm in diameter, 80mm in depth); pump source 021 adopts a cylindrical package (30mm in diameter, 100mm in length); pump combiner 022 is a cubic structure (40mm×40mm×40mm), with two fiber optic interfaces (3mm in diameter) reserved at the input end; thulium-holmium co-doped fiber is wound on a cylindrical bracket (100mm in diameter, 50mm in height), with three fiber fixing slots (2mm in width, 1mm in depth) evenly distributed on the side of the bracket; single-mode fiber is stored in a reel (200mm in diameter). The spool has a thickness of 50mm and a fiber optic guide hole (2mm in diameter) on its edge. The polarization-independent isolator 025 is cylindrical (20mm in diameter, 50mm in length) with FC / APC interfaces at both ends. The photodetector 031 has a cuboid housing (60mm×40mm×30mm), with a fiber optic interface at the input and an SMA interface at the output. The oscilloscope 032 is a standard rack-mount design (482mm×177mm×300mm), with a 10-inch display on the front panel and power and signal interfaces on the rear panel. The integrated frame of the support module is a cuboid (1000mm×600mm×80mm), with multiple M6 threaded holes (50mm×50mm spacing) machined on the surface for fixing the modules.

[0089] (2) Draw a three-dimensional model of the support module and connection structure: The base is made of aluminum alloy plate (1200mm×800mm×20mm) with a surface flatness of ≤0.05mm / m; the guide rail is a linear slide rail (length 800mm, width 30mm) with a slider load capacity ≥5kg; the fiber optic fixing seat is an L-shaped structure (50mm×50mm×10mm) with a V-shaped groove (angle 90°, depth 2mm) machined on the top; the cable trough is a U-shaped trough (width 40mm, depth 30mm) arranged along the edge of the base.

[0090] (3) Design of module connection structure: Signal source module 01 is connected to pump combiner 022 via FC / APC fiber optic patch cord, with a patch cord length of 1m and insertion loss ≤0.3dB; Pump source 021 is connected to pump combiner 022 via 793nm dedicated fiber optic cable with an outer diameter of 0.9mm; Thulium-holmium co-doped fiber and single-mode fiber are connected via fusion splice plate, with a fusion splice loss ≤0.5dB; Photodetector 031 is connected to oscilloscope 032 via SMA high-frequency coaxial cable with an impedance of 50Ω and a length of 2m; Industrial control computer is connected to pump source 021 via RS485 data line and to oscilloscope 032 via Ethernet cable.

[0091] (4) Overall assembly modeling: With the center of the base as the reference, the signal source module 01 is fixed on the left side (100mm from the edge), the repeater amplifier module 02 is arranged in sequence along the guide rail (150mm between adjacent units), the detection and analysis module 03 is located on the right side (100mm from the edge), and the industrial control computer is fixed at the rear end of the base (center position); the optical fiber path is laid out in a straight line, and the optical axis height of each module is consistent (150mm from the surface of the base); the cable tray is arranged along the edge of the module to avoid crossing the optical fiber path.

[0092] (5) Solid parts are made using a laser cutter and a 3D printer: Metal parts (base, guide rail, fiber optic mounting base) are made of 6061 aluminum alloy and processed by laser cutting, with surface anodizing treatment; Plastic parts (fiber optic bracket, spool, cable tray) are made of polytetrafluoroethylene material and printed by a Bambu X1 3D printer with the following printing parameters: layer height 0.1mm, infill density 50%, linear support structure, and printing accuracy ±0.1mm; All interface parts (fiber optic connector, cable plug) are standard commercial parts.

[0093] (6) Mechanical assembly process: Place the base horizontally on the optical platform and level it with a level (error ≤ 0.02 mm / m); fix the guide rail to the base surface with hexagonal screws (M6×20 mm) and pre-adjust the slider position to the middle scale; fix the signal source module 01, repeater amplification module 02, and detection and analysis module 03 on the slider in sequence, and adjust the height of each module with an optical axis calibrator to ensure that the optical axis coaxiality is ≤ 0.1 mm; wind the thulium-holmium co-doped fiber on the bracket, and connect the pump combiner 022 and the single-mode fiber at both ends respectively, and control the winding tension at 5-10 N; lead the single-mode fiber out from the spool and connect it to the polarization-independent isolator 025 through the guide hole to ensure that the fiber is not bent (curvature radius ≥ 30 mm).

[0094] (7) Hardware installation and connection: Fix the pump source 021 next to the pump combiner 022 and connect the power cord (12VDC); connect the output of the photodetector 031 to the input of the oscilloscope 032 through the SMA cable and tighten to a torque of 5N·m; connect the industrial control computer to the control interface of the pump source 021 through the RS485 data cable and connect it to the LAN port of the oscilloscope 032 through the Ethernet cable; check that all fiber optic interfaces are clean and wipe them with anhydrous ethanol if necessary; before powering on, check that the power supply voltages of each module are matched (signal source 220VAC, pump source 021 12VDC, detector 5VDC), and connect the main power supply after confirming that there are no errors to complete the assembly.

[0095] Example 2

[0096] This embodiment uses a YOKOGAWA AQ6377 spectrometer, covering the MWIR region, to accurately detect five wavelengths from 1900 to 5500 nm.

[0097] The steps for integrating and using the signal source module 01 and the relay amplifier based on Embodiment 1 are as follows:

[0098] (1) Modeling of each core component using SolidWorks software: The outer shell of signal source module 01 is designed as a cuboid structure (200mm×150mm×100mm), with a reserved laser mounting slot inside (50mm in diameter, 80mm in depth); pump source 021 adopts a cylindrical package (30mm in diameter, 100mm in length); pump combiner 022 is a cubic structure (40mm×40mm×40mm), with two fiber optic interfaces (3mm in diameter) reserved at the input end; thulium-holmium co-doped fiber is wound on a cylindrical bracket (100mm in diameter, 50mm in height), with three fiber fixing slots (2mm in width, 1mm in depth) evenly distributed on the side of the bracket; single-mode fiber is stored in a reel (200mm in diameter). The spool has a thickness of 50mm and a fiber optic guide hole (2mm in diameter) on its edge. The polarization-independent isolator 025 is cylindrical (20mm in diameter, 50mm in length) with FC / APC interfaces at both ends. The photodetector 031 has a cuboid housing (60mm×40mm×30mm), with a fiber optic interface at the input and an SMA interface at the output. The oscilloscope 032 is a standard rack-mount design (482mm×177mm×300mm), with a 10-inch display on the front panel and power and signal interfaces on the rear panel. The integrated frame of the support module is a cuboid (1000mm×600mm×80mm), with multiple M6 threaded holes (50mm×50mm spacing) machined on the surface for fixing the modules.

[0099] (2) Draw a three-dimensional model of the support module and connection structure: The base is made of aluminum alloy plate (1200mm×800mm×20mm) with a surface flatness of ≤0.05mm / m; the guide rail is a linear slide rail (length 800mm, width 30mm) with a slider load capacity ≥5kg; the fiber optic fixing seat is an L-shaped structure (50mm×50mm×10mm) with a V-shaped groove (angle 90°, depth 2mm) machined on the top; the cable trough is a U-shaped trough (width 40mm, depth 30mm) arranged along the edge of the base.

[0100] (3) Design of module connection structure: Signal source module 01 is connected to pump combiner 022 via FC / APC fiber optic patch cord, with a patch cord length of 1m and insertion loss ≤0.3dB; Pump source 021 is connected to pump combiner 022 via 793nm dedicated fiber optic cable with an outer diameter of 0.9mm; Thulium-holmium co-doped fiber and single-mode fiber are connected via fusion splice plate, with a fusion splice loss ≤0.5dB; Photodetector 031 is connected to oscilloscope 032 via SMA high-frequency coaxial cable with an impedance of 50Ω and a length of 2m; Industrial control computer is connected to pump source 021 via RS485 data line and to oscilloscope 032 via Ethernet cable.

[0101] (4) Overall assembly modeling: With the center of the base as the reference, the signal source module 01 is fixed on the left side (100mm from the edge), the repeater amplifier module 02 is arranged in sequence along the guide rail (150mm between adjacent units), the detection and analysis module 03 is located on the right side (100mm from the edge), and the industrial control computer is fixed at the rear end of the base (center position); the optical fiber path is laid out in a straight line, and the optical axis height of each module is consistent (150mm from the surface of the base); the cable tray is arranged along the edge of the module to avoid crossing the optical fiber path.

[0102] (5) Solid parts are made using a laser cutter and a 3D printer: Metal parts (base, guide rail, fiber optic mounting base) are made of 6061 aluminum alloy and processed by laser cutting, with surface anodizing treatment; Plastic parts (fiber optic bracket, spool, cable tray) are made of polytetrafluoroethylene material and printed by a Bambu X1 3D printer with the following printing parameters: layer height 0.1mm, infill density 50%, linear support structure, and printing accuracy ±0.1mm; All interface parts (fiber optic connector, cable plug) are standard commercial parts.

[0103] (6) Mechanical assembly process: Place the base horizontally on the optical platform and level it with a level (error ≤ 0.02 mm / m); fix the guide rail to the base surface with hexagonal screws (M6×20 mm) and pre-adjust the slider position to the middle scale; fix the signal source module 01, repeater amplification module 02, and detection and analysis module 03 on the slider in sequence, and adjust the height of each module with an optical axis calibrator to ensure that the optical axis coaxiality is ≤ 0.1 mm; wind the thulium-holmium co-doped fiber on the bracket, and connect the pump combiner 022 and the single-mode fiber at both ends respectively, and control the winding tension at 5-10 N; lead the single-mode fiber out from the spool and connect it to the polarization-independent isolator 025 through the guide hole to ensure that the fiber is not bent (curvature radius ≥ 30 mm).

[0104] (7) Hardware installation and connection: Fix the pump source 021 next to the pump combiner 022 and connect the power cord (12VDC); connect the output of the photodetector 031 to the input of the oscilloscope 032 through the SMA cable and tighten to a torque of 5N·m; connect the industrial control computer to the control interface of the pump source 021 through the RS485 data cable and connect it to the LAN port of the oscilloscope 032 through the Ethernet cable; check that all fiber optic interfaces are clean and wipe them with anhydrous ethanol if necessary; before powering on, check that the power supply voltages of each module are matched (signal source 220VAC, pump source 021 12VDC, detector 5VDC), and connect the main power supply after confirming that there are no errors to complete the assembly.

[0105] (8) Each module is fixed to the optical platform with M6 screws, and the optical path is calibrated with a laser collimator (accuracy ±1μm) to ensure the signal source output.

[0106] (9) The coaxiality deviation of each unit center of the relay amplifier module 02 and the input terminal of the photodetector 031 is ≤5μm. During power-on debugging, the link integrity is first tested with a low-power (50mW) signal, and then the power is gradually increased to the working value (300-400mW) to verify the real-time performance and accuracy of spectral analysis.

[0107] The 2-micron intelligent ultrafast laser spectroscopy real-time optical detection device of this application has the following advantages:

[0108] 1. Intelligent adaptive parameter control: balancing gain and loss

[0109] Compared to traditional mid-infrared spectrometers that rely on manually preset fixed parameters and struggle to dynamically adapt to signal transmission conditions, this application utilizes a genetic algorithm submodule to achieve real-time dynamic optimization of pump power, thulium-holmium co-doped fiber length, and single-mode fiber dispersion parameters. The genetic algorithm generates optimal parameter combinations through iterative calculations, controlling the transmission loss of mid-infrared laser in the 2μm band to ≤3dB / km while avoiding pulse distortion. This resolves the contradiction in traditional systems where excessive gain leads to abnormal pulse broadening, while insufficient gain causes signal attenuation. Furthermore, the cascaded design of multi-stage repeater amplification modules, combined with genetic algorithm optimization, can control signal energy fluctuations to ≤5%, significantly improving signal stability during long-distance transmission.

[0110] 2. Ultrafast spectral analysis and real-time monitoring: Replacing complex traditional instruments

[0111] Compared to traditional spectrometers that rely on precision optical components such as gratings and interferometers for spectral splitting, and whose analysis of the fine structure of soliton pulses (such as Kelly septa and dissipative soliton spectra) requires offline processing and is time-consuming, this invention employs a convolutional neural network (CNN) algorithm submodule to achieve real-time feature recognition of time-frequency signals. The CNN, through a pre-trained time-frequency feature library (covering various soliton pulse structures), can directly map the electrical signal output by the photodetector into a time-domain waveform. The analysis process does not require complex optical splitting components, and the recognition bandwidth reaches 100MHz-10GHz, with an accuracy ≥95% and a similarity to the original spectrum ≥98%. Simultaneously, the oscilloscope integrating the CNN can display the waveform in real time, achieving a rapid "end-to-end" response from signal detection to spectral analysis, making it particularly suitable for dynamic monitoring scenarios involving ultrafast laser pulses.

[0112] 3. High-efficiency mid-infrared band signal transmission and amplification: overcoming the bottleneck of signal loss.

[0113] To address the issues of high loss and rapid signal attenuation caused by material absorption and dispersion during the transmission of traditional mid-infrared lasers in optical fibers, this invention designs three cascaded repeater amplification units. Each unit utilizes a 793nm pump source combined with 3-5m thulium-holmium co-doped fiber to achieve efficient gain compensation, and is further enhanced by 2-3km dispersion-broadening single-mode fiber to achieve time stretching. Through the energy level matching characteristics of the thulium-holmium co-doped fiber, the signal gain efficiency is improved by more than 40% compared to traditional erbium-doped fiber. The dispersion Fourier transform function of the single-mode fiber linearly maps the spectral information to the time domain, avoiding the signal distortion problem in traditional dispersion compensation schemes. This ensures that the 2μm band laser maintains spectral structural integrity after long-distance transmission, laying the foundation for subsequent high-precision analysis.

[0114] 4. Wideband compatibility and system scalability: Adaptable to multiple application scenarios.

[0115] Compared to traditional mid-infrared spectrometers, which are mostly designed for fixed wavelengths and require large-scale replacement of optical components for cross-band applications, this invention, through modular design and algorithm adaptability optimization, can be extended to a wavelength range of 800nm-2100nm. By adjusting the pump wavelength parameter library of the genetic algorithm submodule (such as adapting pump source power curves for different wavelengths) and the feature library of the convolutional neural network algorithm submodule (supplementing spectral features from near-infrared to mid-infrared), it can be adapted to various ultrafast laser sources from near-infrared to mid-infrared without replacing the core hardware. This scalability allows the system to be flexibly applied to multiple fields such as environmental monitoring (e.g., gas molecule absorption spectroscopy) and biomedicine (e.g., infrared feature analysis of biological tissues), reducing equipment costs for cross-scenario applications.

Claims

1. A 2-micron intelligent ultrafast laser spectral real-time optical detection device, characterized in that, include: The signal source module (01) is used to output mid-infrared laser pulse signals with a wavelength range of 1.9μm-2.1μm; The relay amplification module (02) includes three amplification units. Each amplification unit is connected in sequence to a pump source (021), a pump combiner (022), a 3-5m thulium-holmium co-doped fiber (023), a 2-3km single-mode fiber (024), and a polarization-independent isolator (025) to achieve signal gain compensation and dispersion broadening. The detection and analysis module (03) includes a photodetector (031) and an oscilloscope (032) for performing photoelectric signal conversion and real-time monitoring. The intelligent algorithm control module (04) includes a genetic algorithm submodule (041) and a convolutional neural network algorithm submodule (042): the genetic algorithm submodule (041) is used to dynamically optimize the pump power, the length of the thulium-holmium co-doped fiber (023) and the dispersion parameters of the single-mode fiber (024) to balance the gain and loss during signal transmission; the convolutional neural network algorithm submodule (042) is used to perform feature recognition on the time-frequency signal output by the photodetector (031), map the spectral information into a time-domain waveform, and display it in real time through an oscilloscope (032).

2. The 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 1, characterized in that, The pulse modes output by the signal source module (01) include Q-switched pulses, conventional soliton pulses, dissipative soliton pulses, broadened soliton pulses, and soliton pair pulses. The pulse width is in the range of nanoseconds to femtoseconds, and the repetition rate is 1kHz-10GHz.

3. The 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 1, characterized in that, The pump source (021) has a wavelength of 793nm, the pump combiner (022) and the polarization-independent isolator (025) have a working wavelength range of 2μm, and the single-mode fiber (024) is an optical fiber with dispersion broadening function, used to realize dispersion Fourier transform.

4. The 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 1, characterized in that, The optimization objective of the genetic algorithm submodule (041) is to determine the optimal length of the thulium-holmium co-doped fiber (023) and the optimal length of the single-mode fiber (024) through iterative calculation, so that the signal transmission loss is controlled within the range of ≤3dB / km, while avoiding pulse distortion.

5. The 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 1, characterized in that, The convolutional neural network algorithm submodule (042) is pre-trained in a time-frequency feature library containing Kelly sidebands, dissipative soliton spectra, and fine structure of soliton molecule pairs. The real-time recognition bandwidth is 100MHz-10GHz, and the recognition accuracy reaches ≥95%.

6. The 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 1, characterized in that, The wavelength response range of the photodetector (031) is 1.8μm-2.1μm, and the sensitivity is 0.5-1A / W. The oscilloscope (032) integrates a convolutional neural network algorithm submodule (042) to display a time-domain waveform that is identical to the incident signal spectrum.

7. The method for fabricating the 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in any one of claims 1 to 6, characterized in that, Includes the following steps: S1. Based on the design parameters, use 3D modeling software to construct 3D models of each component of the signal source module (01), relay amplification module (02), and detection and analysis module (03), and perform virtual assembly to check the compatibility of each component. S2. Assemble the relay amplification module (02): According to the composition order of each amplification unit, connect and fix the pump source (021), pump combiner (022), thulium-holmium co-doped fiber (023), single-mode fiber (024) and polarization-independent isolator (025) in sequence to ensure optical alignment accuracy. After assembling the three amplification units, connect them in a cascade manner. S3. Optically couple the output of the signal source module (01) to the input of the pump combiner (022) of the first relay amplifier module to ensure efficient transmission of optical signals. S4. Assemble the detection and analysis module (03): Connect the signal output terminal of the photodetector (031) to the signal input terminal of the oscilloscope (032) through a high-frequency coaxial cable to ensure the integrity of the electrical signal transmission; S5. Deploy the intelligent algorithm control module (04): Load the genetic algorithm submodule (041) and the convolutional neural network algorithm submodule (042) into the industrial control computer, and establish communication connections with the pump source (021) of the relay amplification module (02) and the oscilloscope (032) of the detection and analysis module (03) through the data interface to realize the linkage of parameter control and signal analysis. S6. Fix the hardware carriers of the signal source module (01), the assembled relay amplification module (02), the detection and analysis module (03) and the intelligent algorithm control module (04) on the integrated frame according to the positional relationship of the virtual assembly. Adjust the relative positions of each module through a precision calibration tool to ensure the coaxiality and stability of the optical path. S7. Power on and debug the entire system: Start the signal source module to output mid-infrared laser pulses, optimize the pump power parameters through the genetic algorithm submodule (041) of the intelligent algorithm control module (04), observe the time-domain waveform displayed on the oscilloscope of the detection and analysis module, verify the accuracy and real-time performance of the spectral analysis, and complete the production.

8. The method for fabricating the 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 7, characterized in that, The optimization process of the genetic algorithm submodule (041) in step S2 is as follows: construct a mathematical model of transmission loss and pulse distortion, and use selection, crossover and mutation operations to iteratively generate the optimal parameter combination so that the signal energy fluctuation after multi-level relay is controlled within ≤5%.

9. The method for fabricating the 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 7, characterized in that, The real-time recognition process of the convolutional neural network algorithm submodule (042) in step S4 is as follows: the electrical signal is converted into a time-domain sequence, and spectral features are extracted through convolutional layers and pooling layers, and mapped to a time-domain waveform with a spectral similarity of ≥98% with the incident signal.

10. The method for fabricating the 2-micron intelligent ultrafast laser spectral real-time optical detection device as described in claim 7, characterized in that, The method is adapted to the wavelength range of 800nm-2100nm. By adjusting the pump wavelength optimization parameters of the genetic algorithm submodule (041) and the feature library of the convolutional neural network algorithm submodule (042), it is adapted to ultrafast laser sources in the near-infrared to mid-infrared bands.

Images