Low-noise wide-dynamic-range adaptive light amplification optimization system
By using an adaptive optical amplification optimization system that combines machine learning and deep reinforcement learning to identify scene and signal type and dynamically adjust gain and compensation, the system solves the problems of insufficient dynamic range and noise in high-end optical amplification technology, achieving efficient signal capture, low-distortion transmission and cross-scene adaptation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-03-27
AI Technical Summary
Existing optical amplification technologies suffer from insufficient dynamic range, signal delay, and increased noise levels in high-end fields such as high-speed coherent optical communication, long-distance distributed optical sensing, and quantum optical signal processing. They cannot simultaneously achieve signal capture, recognition, and low distortion, and their spectral gain is uneven across different application scenarios, making them unable to adapt in real time.
A low-noise, wide dynamic range adaptive optical amplification optimization system is adopted, including a preprocessing module, a photoelectric detection module, a balance control module, a dual-mode amplification module, an expansion module, and a self-calibration module. Through machine learning and deep reinforcement learning algorithms, the system identifies the scene and signal type, dynamically adjusts the gain and compensation, and achieves cross-scene adaptation and noise suppression.
It achieves high signal capture efficiency, high recognition accuracy, and low distortion transmission in sudden wide-range signal scenarios, with noise figure maintained at ≤3.0dB, dynamic range extended to ≥35dB, response latency controlled at the millisecond level, adaptable to multiple scenarios without manual debugging, and ensuring stable system performance.
Smart Images
Figure CN121743684A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of optical amplification technology, specifically a low-noise, wide dynamic range adaptive optical amplification optimization system. Background Technology
[0002] Optical amplification technology is a crucial supporting technology for high-end fields such as high-speed coherent optical communication, long-distance distributed optical sensing, and quantum optical signal processing. Its performance directly determines signal transmission distance, detection accuracy, and processing efficiency, and it has become one of the key foundational technologies driving the upgrading of information communication and sensing technologies. However, the following technical challenges still exist in the application of optical amplification technology in these high-end scenarios: Existing systems require compression of gain bandwidth to reduce ASE noise, resulting in insufficient dynamic range; multi-stage amplification structure is used to widen the dynamic range, which introduces significant signal delay and cannot adapt to burst signals; simplifying the adjustment logic in pursuit of fast response leads to an increase in noise figure. These three factors create an essential performance contradiction, making it impossible to balance signal capture, recognition and low distortion in burst wide-range signal scenarios. The existing system is designed for a single scenario and cannot dynamically identify scenario type, signal characteristics and noise sources. When applied across scenarios, the fixed parameters lead to uneven spectral gain and failure of noise suppression, resulting in significant performance degradation. Furthermore, manual debugging is required, which cannot meet the application requirements of real-time adaptation across scenarios without manual debugging. Summary of the Invention
[0003] The purpose of this invention is to provide a low-noise, wide dynamic range adaptive optical amplification optimization system to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: a low-noise, wide dynamic range adaptive optical amplification optimization system, the system comprising: Preprocessing module: Acquires multi-dimensional features of the signal, identifies the application scenario, signal type and noise type, and filters out the noise; Photoelectric detection module: Switches detection modes according to the scene and signal type, and collects signal characteristics, mode parameters and noise data; Balance control module: Establish a multi-objective optimization model and a dynamic mapping relationship between input signals and optimal adjustment parameters, predict signal changes and dynamically allocate resource weights; Dual-mode amplification module: It adopts a dual-mode collaborative amplification structure, automatically switches modes according to the mode crosstalk coefficient to adapt to different scenarios, suppresses corresponding noise through matching algorithms, and collects noise data and mode gain data in real time; Expansion module: Sets up a continuously adjustable amplification structure including a pre-amplifier unit, a post-amplifier unit, and a mode gain equalization compensation unit, adapting to different types of signals and dynamically adjusting the gain and compensation; Self-calibration module: Constructs a digital twin, receives data from each module in real time, achieves cross-scenario adaptation through a transfer model, establishes a quality assessment system to determine system performance, and issues calibration instructions to each module when calibration conditions are triggered; Shaping module: Corrects waveform distortion during amplification and adds amplitude smoothing processing for sudden signal scenarios.
[0005] Preferably, the preprocessing module is as follows: It integrates a dual-dimensional scene recognition subunit, a pattern feature acquisition subunit, and a noise source classification and suppression unit; by acquiring the spectral features, temporal features, and pattern features of the input optical signal, it identifies the application scene and signal type. The noise source is classified and identified based on a machine learning algorithm, and a dedicated suppression mechanism is enabled. The machine learning algorithm specifically adopts a lightweight convolutional neural network + random forest fusion model. The input features are spectral distribution, temporal correlation, and mode power difference. The training iterations are 500 times, and the accuracy threshold is ≥95%. Simultaneously, the filtering bandwidth and mode adaptation parameters are dynamically adjusted according to the scene type, and the identified scene type, signal characteristics, noise source category and adapted filtering parameters are transmitted to the photoelectric detection module in real time.
[0006] Preferably, the photoelectric detection module is as follows: It receives the output signals and parameter instructions from the preprocessing module and integrates a three-modal detection unit, a mode parameter acquisition unit, a noise feature extraction subunit, and an analog-to-digital converter. The three-modal detection unit automatically switches its working mode according to the scene and signal type instructions identified by the preprocessing module: low-noise high-sensitivity mode is suitable for weak signal detection, high-linearity anti-saturation mode is suitable for strong signal detection, and high-speed burst capture mode is suitable for burst signal detection. The mode parameter acquisition unit detects the power difference and crosstalk coefficient of each mode in real time for the LP mode type identified by the front end; the noise feature extraction subunit combines the noise source classification results to directionally acquire the spectral distribution, temporal correlation and mode-related noise data of the corresponding noise; the analog-to-digital converter realizes the distortion-free acquisition of broadband, multi-mode signal details; The collected signal characteristics, mode parameters, and noise data are uploaded in real time to the balance control module and the self-calibration module.
[0007] Preferably, the balance control module is as follows: The system receives three-dimensional information from the photoelectric detection module and adopts a deep reinforcement learning architecture with a master agent and slave agents. The master agent aims to minimize noise figure, dynamic range adaptation, and response delay, while the slave agent aims to achieve optimal mode balance. Through training with single-mode and multi-mode signal samples, a dynamic mapping relationship between input signal features and optimal adjustment parameters is established. In the deep reinforcement learning architecture, the master agent adopts the DQN network with 3 hidden layers, each with 128, 64 and 32 neurons, and the activation function is ReLU; the slave agent adopts the PPO network with 2 hidden layers, 64 and 32 neurons, and the training sample size is ≥100,000 sets, covering single-mode / multi-mode and continuous / burst signal types. A signal prediction and pre-adjustment mechanism based on wavelet transform is introduced. By analyzing the time-domain trend of the signal, the changes in signal amplitude and mode characteristics are predicted, and the back-end amplification gain, pump power and mode equalization parameters are pre-adjusted. Resource weights are dynamically allocated according to the scene priority. The generated amplification gain command, pump power parameters, and mode equalization compensation value are sent to the dual-mode amplification module and the expansion module, respectively, and the optimized target threshold is synchronized to the self-calibration module.
[0008] Preferably, the dual-mode amplification module is as follows: It receives amplification gain commands from the balance control module and adopts a dual-mode cooperative amplification structure, including two working modes: low noise - single mode and wide dynamic range - multi-mode. When the mode crosstalk coefficient exceeds the preset threshold, it automatically triggers mode switching. The low-noise single-mode mode uses single-mode erbium-doped fiber and a tilted grating structure to achieve low-noise amplification through fiber core pumping, making it suitable for single-mode narrow-spectrum communication scenarios. The wide dynamic range multimode mode uses few-mode erbium-doped fiber and a multi-port ring cladding pumping structure to balance multimode gain through fiber core and cladding coordinated pumping, making it suitable for multimode wide-spectrum sensing scenarios. A dynamic matching algorithm for noise sources and pump parameters is designed. The noise dominance type is determined by the data collected by the noise feature extraction subunit. When ASE noise is dominant, the pump power density is reduced; when Rayleigh scattering is dominant, the fiber orientation is optimized and the pump power is increased; when mode crosstalk noise is dominant, pump light frequency modulation technology is enabled. The module integrates a distributed noise monitoring unit to collect noise data and mode gain data at each amplification stage in real time. The data is fed back to the balance control module, and the pre-amplified signal and monitoring data are synchronously transmitted to the expansion module.
[0009] ASE noise threshold: noise spectrum width ≥ 50 nm and no temporal correlation; Rayleigh scattering noise threshold: mode crosstalk coefficient < -30 dB and temporal correlation ≥ 0.8; mode crosstalk noise threshold: mode power difference ≥ 2 dB.
[0010] Preferably, the extended module is as follows: It receives the initial amplified signal and monitoring data from the dual-mode amplification module, and simultaneously receives the gain allocation command from the balance control module. The gain adjustment range of the pre-amplifier unit is 0-20dB, which is suitable for strong signals, burst signals and single-mode signals. The fast gain switching ensures the response speed. The gain adjustment range of the post-amplifier unit is 20-60dB, which is suitable for weak signals, wide-range signals and multi-mode signals. The low-noise amplification link ensures the signal purity.
[0011] The mode gain equalization compensation unit dynamically adjusts the gain compensation amount of each LP mode based on the mode parameter acquisition results of the photoelectric detection module and the mode gain data of the dual-mode amplification module to suppress differential mode gain; the gain weight of the pre-amplifier unit and the post-amplifier unit is allocated by the balance control module; the integrated amplitude prediction compensation unit predicts the amplitude peak and mode power fluctuation based on signal characteristics and adjusts the pre-amplifier gain threshold and mode compensation parameters in advance; the wide dynamic range amplified signal is transmitted to the scene-adaptive output shaping module, and the final signal amplitude and mode equalization data are fed back to the self-calibration module.
[0012] Preferably, the self-calibration module is as follows: Establish a digital twin with a four-dimensional mapping of scene, noise, mode, and parameters; receive scene parameters from the preprocessing module, feature data from the photoelectric detection module, and operating parameters from the dual-mode amplification module and the expansion module in real time; and store the optimal parameter combination under different application scenarios, different noise sources, and different mode parameters. The twin layer dynamically maps the coupling relationship between erbium ion doping distribution, pump light field and noise-mode based on the received multi-module data, and achieves cross-scenario adaptation through a transfer model based on residual network; the calibration layer establishes a multi-dimensional signal quality evaluation system, and determines whether the system performance meets the preset indicators by combining the optimization target threshold issued by the balance control module. The residual network consists of 8 convolutional layers and 3 residual blocks. The input dimension is 4-dimensional (scene + noise + pattern + parameters), and the output dimension is the optimal parameter combination vector. The transfer learning fine-tunes the learning rate by 0.001.
[0013] When the system detects a scene switch or when the gain or noise figure of the twin's predicted differential mode exceeds the preset range, it triggers a self-calibration process: it generates a standard reference light with LP01 mode power stability ≤ ±0.1dB through the mode converter, compares the twin simulation data with the actual detection data, and sends calibration instructions to the preprocessing module, photoelectric detection module, dual-mode amplification module and expansion module respectively.
[0014] Preferably, the shaping module receives the wide dynamic range signal output by the expansion module, and integrates a patterned programmable optical waveform correction unit, an adaptive spectral shaping filter, and a robust enhancement unit, based on the application scenario type identified by the preprocessing module; and corrects waveform distortion during the amplification process using digital signal processing algorithms. The robust enhancement unit incorporates amplitude smoothing for burst signal scenarios and modal dispersion compensation for long-distance multimode transmission scenarios. It compensates for modal dispersion loss during fiber transmission and amplification through digital signal processing algorithms and feeds back the output signal quality assessment data to the self-calibration module.
[0015] The beneficial effects of this invention are as follows: 1. The balance control module of this invention adopts a master-slave intelligent agent deep reinforcement learning architecture, with the core optimization goals of minimizing noise figure, dynamic range adaptation, and response latency. Combined with a wavelet transform signal prediction mechanism, it pre-adjusts the back-end amplification gain, pump power, and mode equalization parameters. The dual-mode amplification module can automatically switch between single-mode and multi-mode operation modes according to the mode crosstalk coefficient, and with a dynamic matching algorithm for noise sources and pump parameters, it accurately suppresses interference for different noise types. The expansion module adopts a front-end and back-end graded gain design, with the front-end adapting to strong and burst signals, and the back-end adapting to weak and wide-amplitude signals, achieving continuous adjustable gain. It can maintain the noise figure at a low level of ≤3.0dB, extend the dynamic range to ≥35dB, and control the response latency in milliseconds, successfully adapting to burst wide-amplitude signal scenarios, and solving the technical problem of simultaneously achieving signal capture efficiency, recognition accuracy, and low-distortion transmission.
[0016] 2. The preprocessing module of this invention collects the spectral, temporal, and mode characteristics of the input optical signal to accurately identify the application scenario, signal type, and noise source category; the self-calibration module constructs a digital twin with a four-dimensional mapping of scenario-noise-mode-parameter, storing the optimal parameter combination under different application scenarios, noise sources, and mode parameters, and achieving rapid cross-scenario adaptation with the help of a transfer model; the mode gain equalization compensation unit dynamically adjusts the gain compensation amount of each LP mode according to the mode parameters collected by the front end and the gain data of the amplification module, so that the differential mode gain is effectively suppressed; it can automatically adapt to various scenarios such as 5G base station backhaul, distributed sensing of oil and gas pipelines, and quantum optical signal transmission, and can automatically trigger the calibration process when switching scenarios without manual intervention, ensuring the stability of system performance throughout the process.
[0017] 3. The shaping module of this invention integrates a programmable optical waveform correction unit, an adaptive spectral shaping filter, and a robustness enhancement unit. It uses digital signal processing algorithms to correct waveform distortion generated during amplification. For communication scenarios, it focuses on correcting phase distortion and mode crosstalk of coherent optical signals, while for sensing scenarios, it focuses on compensating for edge attenuation of broadband multimode signals. The robustness enhancement unit adds amplitude smoothing processing for sudden signal scenarios to avoid damage to back-end equipment due to signal abrupt changes, and adds mode dispersion compensation function for long-distance multimode transmission scenarios to offset mode dispersion loss generated during fiber transmission and amplification. The self-calibration module establishes a closed-loop optimization mechanism based on feedback data from each module to continuously optimize signal quality. The waveform stability, spectral purity, and mode uniformity of the final output signal are significantly improved, making it fully adaptable to high-end fields such as high-speed coherent optical communication, long-distance distributed optical sensing, and quantum optical signal processing. Attached Figure Description
[0018] Figure 1 This is a flowchart of the low-noise wide dynamic range adaptive optical amplification optimization system of the present invention; Figure 2 This is a flowchart of the mode switching process for the dual-mode amplification module of the present invention; Figure 3 This is a flowchart of the closed-loop calibration process for the self-calibration module of the present invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] like Figures 1 to 3 As shown, this embodiment of the invention provides a low-noise, wide dynamic range adaptive optical amplification optimization system, which includes: The preprocessing module, serving as the core of the system's front-end perception, integrates a dual-dimensional scene recognition subunit, a pattern feature acquisition subunit, and a noise source classification and suppression unit. It identifies application scenarios and signal types by acquiring the spectral, temporal, and pattern characteristics of the input optical signal. For example, in single-mode communication scenarios, it can identify 5G base station backhaul links; in multi-mode sensing scenarios, it can identify distributed sensing in oil and gas pipelines; and in mode-division multiplexing transmission scenarios, it can identify high-density interconnection in data centers. Signal types include narrow-spectrum / wide-spectrum, continuous / burst, and single-mode / multi-mode. Spectral features include spectral width and center wavelength; temporal features include continuous / burst states and pulse width; and pattern characteristics include the number of modes and LP mode type. Noise sources are classified and identified based on machine learning algorithms, and dedicated suppression mechanisms are activated. For example, in long-distance pipeline sensing scenarios, Rayleigh scattering noise is reduced by polarization-mode diversity suppression technology to reduce scattering interference; in communication scenarios near power lines, interference noise is filtered out by adaptive notch filtering technology to remove power frequency interference; in weak signal reception scenarios, ASE noise is purified by narrowband frequency selective suppression technology; and in multimode fiber transmission scenarios, multimode crosstalk noise is reduced by mode isolation filtering technology to reduce intermode interference. The noise source classification includes ASE noise, Rayleigh scattering noise, mode crosstalk noise, and interference noise. The system dynamically adjusts the filtering bandwidth and mode adaptation parameters according to the scene type. For single-mode scenes, it adapts to 10nm-50nm narrowband filtering, and for multi-mode scenes, it adapts to 100nm-200nm broadband noise reduction + mode filtering. The identified scene type, signal characteristics, noise source category and adapted filtering parameters are transmitted to the photoelectric detection module in real time, providing a basis for decision-making for its detection mode switching and data acquisition accuracy optimization.
[0021] The photoelectric detection module receives the output signals and parameter instructions from the preprocessing module and integrates a three-modal detection unit, a mode parameter acquisition unit, a noise feature extraction subunit, and a high-speed analog-to-digital converter. The three-modal detection unit automatically switches its working mode according to the scene and signal type instructions identified by the preprocessing module: a low-noise, high-sensitivity mode is suitable for weak signal detection (such as microwatt-level signals from long-distance sensing feedback), a high-linearity, anti-saturation mode is suitable for strong signal detection (such as kilowatt-level optical signals in data centers), and a high-speed burst capture mode is suitable for burst signal detection (such as short-frame burst signals in satellite communication). The mode parameter acquisition unit detects the power difference and crosstalk coefficient of each mode in real time based on the LP mode type identified by the front end; the noise feature extraction subunit combines the noise source classification results to directionally acquire the spectral distribution, temporal correlation, and mode-related noise data of the corresponding noise; the high-speed analog-to-digital converter has a sampling rate of 2GSps, a resolution of 14 bits, an input bandwidth of ≥2GHz, and a quantization error of ≤±1LSB, realizing distortion-free acquisition of broadband and multi-mode signal details, and the overall module has a detection dynamic range of 70dB; broadband signals are defined as having a spectral span of ≥200nm, and multi-mode signals are defined as having ≤6 LP modes; The collected signal characteristics, mode parameters, and noise data are uploaded in real time to the balance control module and the self-calibration module.
[0022] The balance control module receives three-dimensional information from the photoelectric detection module and adopts a deep reinforcement learning architecture with a master agent and slave agents. The master agent aims to minimize noise figure, dynamic range adaptation, and response delay, while the slave agent aims to achieve optimal mode balance. Through training with single-mode and multi-mode signal samples covering a temperature range of -40℃ to 85℃, a dynamic mapping relationship between input signal features and optimal adjustment parameters is established. The signal features include amplitude, frequency, noise type, and mode parameters. A signal prediction and pre-adjustment mechanism based on wavelet transform is introduced. By analyzing the temporal trends of signals, such as the rise slope of burst signals to predict the power peak of satellite burst frames, the amplitude fluctuation of continuous signals to predict the signal stability of power grid communication, and the mode power change rate to predict the mode imbalance risk of multimode transmission, changes in signal amplitude and mode characteristics can be predicted, and back-end amplification gain, pump power, and mode equalization parameters can be pre-adjusted. Resource weights are dynamically allocated according to scenario priority. For example, in burst signal scenarios, response speed is prioritized, with a weight of 60%, noise suppression weight of 30%, and mode equalization weight of 10%, such as in satellite short frame data transmission. In multimode signal scenarios, mode equalization is prioritized, with a weight of 50%, dynamic range weight of 35%, and response speed weight of 15%, such as in multimode fiber distributed sensing. In low-noise requirement scenarios, ASE noise suppression is prioritized, with a weight of 65%, dynamic range weight of 25%, and response speed weight of 10%, such as in quantum optical signal transmission. The generated amplification gain command, pump power parameters, and mode equalization compensation value are sent to the dual-mode amplification module and the expansion module, respectively. At the same time, the real-time operating data (noise value, mode gain difference, signal amplitude) fed back by the two modules are received, and the control strategy is iteratively optimized. The optimization target threshold (such as the upper limit of differential mode gain and the upper limit of noise figure) is synchronized to the self-calibration module as the basis for its calibration trigger.
[0023] Multi-objective optimization of reward function:
[0024] In the formula: This represents the instantaneous reward value of the multi-agent DRL. The larger the value, the better the current control strategy. The negative sign in the formula is used to transform the minimization objective (such as minimizing the noise figure or dynamic range deviation) into maximizing the reward. Indicates noise figure The weighting coefficient ranges from 0.3 to 0.5. The specific value is adjusted according to the needs of the scenario. For low-noise scenarios, the upper limit (0.4 to 0.5) is used to prioritize noise suppression; for sudden signal scenarios, the lower limit (0.3 to 0.4) is used to prioritize response speed. The actual noise index of the system is expressed in dB. The core target value is ≤3.0dB. It is optimized and output by a multi-agent deep reinforcement learning algorithm. It corresponds to the optimization target of "lowest noise index" in the balance control module and reflects the system's ability to suppress signal noise. The weighting coefficient representing the dynamic range deviation ranges from 0.4 to 0.6. For wide-range signal scenarios (such as multi-mode distributed sensing), the upper limit (0.5 to 0.6) is used to prioritize adapting to signal amplitude fluctuations; for weak signal scenarios (such as long-distance weak sensing feedback), the lower limit (0.4 to 0.5) is used to prioritize ensuring signal amplification gain. This represents the actual dynamic range of the system, in dB, with a target value of ≥35dB. It is output by the expansion module and corresponds to the core objective of dynamic range adaptation, reflecting the system's ability to adapt to signals of different amplitudes, from microwatt-level weak signals to hundreds of milliwatt-level strong signals. This indicates the dynamic range of scenario requirements, in dB. It is determined by the preprocessing module after identifying the application scenario. The requirement for communication scenarios (such as 5G base station backhaul) is 25~35dB, and the requirement for sensing scenarios (such as distributed monitoring of oil and gas pipelines) is 35~45dB. Indicates mode gain difference The weighting coefficient ranges from 0.2 to 0.4. For multi-mode scenarios (such as modulus-multiplexed data transmission), the upper limit (0.3 to 0.4) is used to prioritize gain balance between modes. For single-mode scenarios (such as single-mode coherent communication), the lower limit (0.2 to 0.3) is used because only the LPO1 mode exists, resulting in a smaller gain difference between modes. =0, the weight has little impact; This represents the maximum gain difference between multiple LP modes, expressed in dB. The target value is ≤0.3dB, and in typical scenarios, it ranges from 0.1 to 0.5dB. It usually refers to the gain difference between LPO1 and LP11 modes, for example, when the LPO1 mode gain is 25dB and the LP11 mode gain is 24.8dB. =0.28, which is detected in real time by the mode parameter acquisition unit and reflects the degree of gain balance between modes in multi-mode scenarios.
[0025] The dual-mode amplification module receives amplification gain commands from the balance control module and adopts a dual-mode collaborative amplification structure, including two working modes: low-noise single-mode and wide dynamic range multi-mode. When the mode crosstalk coefficient is greater than -20dB, mode switching is automatically triggered. For example, when the crosstalk coefficient between LP01 and LP11 modes in the multi-mode signal reaches -18dB, the mode switches from mode A to mode B. The greater than -20dB threshold is set based on the mode isolation requirements in the multi-mode scenario. The crosstalk coefficient threshold in the single-mode scenario is less than -30dB, and switching is only triggered when the crosstalk of the multi-mode signal exceeds the standard.
[0026] The low-noise single-mode mode uses single-mode erbium-doped fiber (erbium doping concentration 300ppm) and a tilted grating structure with a center wavelength deviation ≤0.5nm. Low-noise amplification is achieved through core pumping (pump power 50-100mW, matching the power parameters issued by DRL), making it suitable for single-mode narrow-spectrum communication scenarios (1530-1565nm, such as 100G coherent communication in backbone networks). The wide dynamic range multimode mode uses few-mode erbium-doped fiber (erbium doping concentration 500ppm, supporting LP01 / LP11 modes) and a multi-port ring cladding pump structure (pump light uniformity ≥90%). Multimode gain is balanced through core and cladding co-pumping (total power 100-200mW, dynamically adjusted according to DRL commands), making it suitable for multimode wide-spectrum sensing scenarios (1470-1670nm, such as cross-basin hydrological monitoring sensing). The design incorporates a dynamic matching algorithm between noise sources and pump parameters. When ASE noise dominates (based on front-end noise classification results, such as in weak signal amplification scenarios), the pump power density is reduced. When Rayleigh scattering dominates (such as in long-distance fiber optic transmission scenarios), the fiber optic routing is optimized and the pump power is increased. When mode crosstalk noise dominates (such as in multimode dense transmission scenarios), pump optical frequency modulation technology is enabled. When interference noise dominates (such as power frequency interference and external electromagnetic interference), adaptive notch filtering combined with pump power anti-interference modulation technology is enabled to suppress narrowband interference signals. The module integrates a distributed noise monitoring unit to collect noise data and mode gain data at each amplification stage in real time. The data is fed back to the balance control module for iterative parameter adjustment, and the initially amplified signal and monitoring data are synchronously transmitted to the expansion module to provide real-time reference for its gain equalization compensation.
[0027] Noise source-pump power matching formula:
[0028] In the formula: This indicates the actual output pump power, including both core-pumped and core-cladding co-pumped forms, in mW. The specific value corresponds to the design of the dual-mode amplifier module: 50~100mW in low-noise single-mode scenarios (1530-1565nm band) and 100~200mW in wide dynamic range multi-mode scenarios (1470-1670nm band), which can be dynamically adjusted according to the type of noise source. The reference pump power for a given scenario is expressed in mW and is determined by both the scenario type and the signal power. The reference value for a single-mode communication scenario (such as 100G coherent communication in a backbone network) is 80mW, and the reference value for a multi-mode sensing scenario (such as cross-basin hydrological monitoring sensing) is 150mW. This represents the noise source correction factor, which is dynamically determined based on the type of noise source: when ASE noise (amplified spontaneous emission noise) is dominant, k is 0.7~0.9, which reduces noise excitation by lowering the pump power density; when Rayleigh scattering noise is dominant, k is 1.1~1.3, which enhances signal gain by increasing the pump power and offsets scattering loss; when mode crosstalk noise is dominant, k is 0.9~1.1, which suppresses inter-mode crosstalk and ensures signal purity by fine-tuning the power in conjunction with pump light frequency modulation technology.
[0029] The expansion module receives the initial amplified signal and monitoring data from the dual-mode amplification module, and simultaneously receives the gain allocation command from the balance control module. The gain adjustment range of the pre-amplifier unit is 0-20dB, adapting to strong signals, burst signals, and single-mode signals, such as the hundreds of milliwatt-level signals transmitted back from the base station. Combined with the signal amplitude data from the dual-mode module, the response speed is ensured through rapid gain switching. The gain adjustment range of the post-amplifier unit is 20-60dB, adapting to weak signals, wide-amplitude signals, and multi-mode signals, such as the microwatt-level signals from sensors in remote areas. The signal purity is ensured through a low-noise amplification link.
[0030] The mode gain equalization compensation unit dynamically adjusts the gain compensation amount of each LP mode based on the mode parameter acquisition results of the photoelectric detection module and the mode gain data of the dual-mode amplification module. The compensation range is -5dB to +5dB, suppressing differential mode gain. For example, if the gain of LP01 mode is 3dB higher than that of LP11 mode, an additional 3dB compensation is added to LP11 mode. The gain weight of the pre-amplifier unit and the post-amplifier unit is allocated by the balance control module to avoid the delay accumulation of traditional multi-stage amplification. The integrated amplitude prediction compensation unit predicts the peak amplitude and mode power fluctuation based on signal characteristics. For example, if the peak value of a burst signal is predicted to reach 100mW, the pre-amplifier gain is reduced in advance, and the pre-amplifier gain threshold and mode compensation parameters are adjusted in advance to avoid strong signal saturation distortion and multi-mode signal gain imbalance. The wide dynamic range amplified signal is transmitted to the shaping module, and the final signal amplitude and mode equalization data are fed back to the self-calibration module.
[0031] Mode gain equalization compensation formula:
[0032]
[0033] In the formula: This represents the final gain of the LP mode after compensation, in dB. It is calculated by adding the original gain of the LP mode to the gain compensation amount. The core goal is to make the final gain of all LP modes more consistent. For example, the final gain of LP01 and LP11 are close to 24.5dB, so as to avoid signal transmission distortion caused by gain imbalance between modes. This represents the original gain of the LP mode (gain before compensation), in dB. It is output by the dual-mode amplifier module and detected in real time by the mode parameter acquisition unit. For example, the original gain of LPO1 mode is 25dB and the original gain of LP11 mode is 22dB. This indicates the gain compensation amount for LP mode, in dB, with a value range of -5dB to +5dB, corresponding to the 0-5dB gain compensation design in the expansion module; positive compensation (+1dB to +5dB) is used to improve modes with low gain, such as LP11 with an original gain of 22dB, which is compensated by +2.5dB; negative compensation (-1dB to -5dB) is used to suppress modes with excessive gain, such as LP01 with an original gain of 27dB, which is compensated by -2.5dB, thus achieving mode gain balance; This represents the target equalization gain of the LP mode, in dB. It is calculated as the average of the original gains of all LP modes. For example, if the original gain of LPO1 is 25dB and the original gain of LP11 is 22dB, the target equalization gain is (25+22) / 2=23.5dB, which serves as a benchmark for determining whether compensation is needed and how much compensation is required.
[0034] The self-calibration module establishes a digital twin with a four-dimensional mapping of scene, noise, mode, and parameters. It receives scene parameters (transmission distance, temperature) from the preprocessing module, characteristic data (signal spectral width, mode crosstalk) from the photoelectric detection module, and operating parameters (pump power, gain) from the dual-mode amplification module and the expansion module in real time. It synchronizes the actual system parameters every 10ms, and the synchronized data is used for real-time updates and performance evaluation of the digital twin. It stores the optimal parameter combinations (filter parameters, detection mode, amplification gain, pump power) under different application scenarios (single-mode / multi-mode, communication / sensing), different noise sources (ASE / Rayleigh scattering / mode crosstalk), and different mode parameters (number of modes, LP type). The digital twin dynamically maps the coupling relationship between erbium ion doping distribution, pump light field, and noise-mode based on received multi-module data. It achieves cross-scenario adaptation through a transfer model based on residual networks, such as switching from a single-mode metropolitan area communication scenario to a multi-mode seabed sensing scenario. The calibration layer establishes a multi-dimensional signal quality evaluation system, including signal-to-noise ratio, distortion, spectral equalization, and mode equalization. Combined with the optimization target threshold issued by the balance control module, it determines whether the system performance meets the preset indicators. When the system detects a scene switch or when the gain or noise figure of the twin's predicted differential mode exceeds the preset range, a self-calibration process is triggered: a standard reference light with LP01 mode power stability ≤ ±0.1dB is generated through the mode converter, and the twin simulation data is compared with the actual detection data. For example, when the simulation signal-to-noise ratio is 25dB and the actual detection signal-to-noise ratio is 23dB, the amplification gain is finely adjusted, and calibration commands are sent to the preprocessing module to adjust the filter bandwidth, to the photoelectric detection module to optimize the detection mode, to the dual-mode amplification module to correct the pump power, and to the expansion module to adjust the gain compensation. After calibration, the digital twin collects the calibrated data of each module in real time, such as the filter bandwidth of the preprocessing module and the pump power of the dual-mode amplification module, and compares the deviation between the actual data and the optimal parameter combination: if the deviation is ≤1%, the calibration is deemed to be compliant; if the deviation is >1%, a second calibration is triggered until the deviation meets the requirements, thus completing the logical closed loop of the calibration process; the deviation refers to the relative deviation between the actual operating parameters (such as pump power, filter bandwidth, and gain value) after calibration and the optimal parameter combination stored in the digital twin. The calibration is deemed compliant when the deviation of all key parameters is ≤1%.
[0035] The shaping module receives the wide dynamic range signal output by the expansion module and integrates a patterned programmable optical waveform correction unit, an adaptive spectral shaping filter, and a robustness enhancement unit, based on the application scenario type identified by the preprocessing module. It corrects waveform distortion during amplification using digital signal processing algorithms. In communication scenarios (single-mode / multi-mode), the focus is on correcting phase distortion and inter-mode crosstalk of coherent optical signals, such as correcting constellation map shifts caused by phase noise in 100G coherent communication. In sensing scenarios, the focus is on compensating for edge attenuation and mode gain differences in broadband multi-mode signals, such as compensating for attenuation of the 1620nm band edge signal in oil and gas pipeline sensing, and correcting spectral broadening using an adaptive spectral shaping filter.
[0036] The robustness enhancement unit incorporates amplitude smoothing for sudden signal scenarios (such as sudden satellite communication, combined with front-end scene recognition results) to prevent abrupt changes in output signal from damaging back-end equipment. For long-distance multimode transmission scenarios (such as multimode fiber transmission across mountainous areas), it adds mode dispersion compensation, using digital signal processing algorithms to compensate for mode dispersion loss during fiber transmission and amplification. For example, after 10km of multimode fiber transmission, it compensates for the time delay difference between LP01 and LP11 modes. Output signal quality assessment data (such as waveform distortion and spectral equalization) is fed back to the self-calibration module to assist it in optimizing subsequent calibration strategies, ultimately ensuring waveform stability, spectral purity, mode equalization, and transmission reliability of output signals in different scenarios and modes.
[0037] 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.
[0038] 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 low-noise, wide dynamic range adaptive optical amplification optimization system, characterized in that: The system includes: Preprocessing module: Acquires multi-dimensional features of the signal, identifies the application scenario, signal type and noise type, and filters out the noise; Photoelectric detection module: Switches detection modes according to the scene and signal type, and collects signal characteristics, mode parameters and noise data; Balance control module: Establish a multi-objective optimization model and a dynamic mapping relationship between input signals and optimal adjustment parameters, predict signal changes and dynamically allocate resource weights; Dual-mode amplification module: It adopts a dual-mode collaborative amplification structure, automatically switches modes according to the mode crosstalk coefficient to adapt to different scenarios, suppresses corresponding noise through matching algorithms, and collects noise data and mode gain data in real time; Expansion module: Sets up a continuously adjustable amplification structure including a pre-amplifier unit, a post-amplifier unit, and a mode gain equalization compensation unit, adapting to different types of signals and dynamically adjusting the gain and compensation; Self-calibration module: Constructs a digital twin, receives data from each module in real time, achieves cross-scenario adaptation through a transfer model, establishes a quality assessment system to determine system performance, and issues calibration instructions to each module when calibration conditions are triggered; Shaping module: Corrects waveform distortion during amplification and adds amplitude smoothing processing for sudden signal scenarios.
2. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 1, characterized in that: The preprocessing module identifies the application scenario and signal type by collecting the spectral characteristics, temporal characteristics, and mode characteristics of the input optical signal; The noise sources are classified and suppressed based on machine learning algorithms. At the same time, the filtering bandwidth and mode adaptation parameters are dynamically adjusted according to the scene type. The identified scene type, signal characteristics, noise source category and adapted filtering parameters are transmitted to the photoelectric detection module.
3. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 2, characterized in that: The photoelectric detection module automatically switches its working mode according to the output signal and parameter instructions of the preprocessing module, including low noise high sensitivity mode, high linearity anti-saturation mode, and high speed burst capture mode. Real-time detection of power differences and crosstalk coefficients between modes; combined with noise source classification results, collection of spectral distribution, temporal correlation and mode-related noise data of the corresponding noise. The collected signal characteristics, mode parameters, and three-dimensional noise data are uploaded to the balance control module and the self-calibration module.
4. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 3, characterized in that: The balance control module receives three-dimensional information from the photoelectric detection module and adopts a deep reinforcement learning architecture with a master agent and slave agents. The master agent aims to minimize noise figure, adapt dynamic range, and minimize response delay, while the slave agent aims to optimize mode balance. Through training with single-mode and multi-mode signal samples, a dynamic mapping relationship between input signal features and optimal adjustment parameters is established. A signal prediction and pre-adjustment mechanism based on wavelet transform is introduced to predict changes in signal amplitude and mode characteristics, pre-adjust back-end amplification gain, pump power and mode equalization parameters, and dynamically allocate resource weights according to scenario priority. The generated amplification gain command, pump power parameters, and mode equalization compensation value are sent to the dual-mode amplification module and the expansion module, respectively, and the optimized target threshold is synchronized to the self-calibration module.
5. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 4, characterized in that: The dual-mode amplification module receives amplification gain commands from the balance control module and adopts a dual-mode cooperative amplification structure, including two working modes: low noise - single-mode and wide dynamic range - multi-mode. When the mode crosstalk coefficient exceeds the preset threshold, the mode switching is automatically triggered. The module employs a dynamic matching algorithm between noise sources and pump parameters to suppress corresponding noise. It integrates a distributed noise monitoring unit to collect noise data and mode gain data at each amplification stage in real time, feeds the data back to the balance control module, and synchronously transmits the pre-amplified signal and monitoring data to the expansion module.
6. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 5, characterized in that: The expansion module receives the initial amplified signal and monitoring data from the dual-mode amplification module, and simultaneously receives the gain allocation command from the balance control module. The gain adjustment range of the pre-amplifier unit is 0-20dB, which is suitable for strong signals, burst signals and single-mode signals; the gain adjustment range of the post-amplifier unit is 20-60dB, which is suitable for weak signals, wide-amplitude signals and multi-mode signals. The mode gain equalization compensation unit dynamically adjusts the gain compensation amount of each LP mode based on the mode parameter acquisition results of the photoelectric detection module and the mode gain data of the dual-mode amplification module to suppress differential mode gain; the integrated amplitude prediction compensation unit predicts the amplitude peak and mode power fluctuation based on signal characteristics and adjusts the pre-stage gain threshold and mode compensation parameters in advance.
7. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 6, characterized in that: The self-calibration module establishes a digital twin with a four-dimensional mapping of scene, noise, mode, and parameters, receives data from each module in real time, and stores the optimal parameter combination under different application scenarios, different noise sources, and different mode parameters. By implementing cross-scenario adaptation based on the transfer model, a multi-dimensional signal quality evaluation system is established. Combined with the optimization target threshold issued by the balance control module, it is determined whether the system performance meets the preset indicators. When the system detects a scene switch or when the gain or noise figure of the twin prediction differential mode exceeds the preset range, it triggers a self-calibration process, compares the twin simulation data with the actual detection data, and sends calibration instructions to the corresponding modules respectively.
8. The low-noise wide dynamic range adaptive optical amplification optimization system according to claim 7, characterized in that: The shaping module receives the wide dynamic range signal output by the expansion module, and integrates a programmable optical waveform correction unit, an adaptive spectral shaping filter, and a robust enhancement unit, based on the application scenario type identified by the preprocessing module; it corrects waveform distortion during the amplification process using digital signal processing algorithms. The robust enhancement unit incorporates amplitude smoothing for burst signal scenarios and modal dispersion compensation for long-distance multimode transmission scenarios. It compensates for modal dispersion loss during fiber transmission and amplification through digital signal processing algorithms and feeds back the output signal quality assessment data to the self-calibration module.