Satellite-ground laser communication adaptive optical mode gain scheduling method

Abstract

This invention discloses an adaptive optics modal gain scheduling method for space-to-ground laser communication, belonging to the fields of space-to-ground laser communication, free-space optical communication, adaptive optics control, and optical communication link state awareness. The method constructs a training state sample set based on offline profile-link state sample data and represents the state samples as profile-link state vectors. It generates offline modal optimal gain labels through a modal group residual phase variance model composed of fitting residuals, servo hysteresis residuals, and noise propagation residuals. It constructs and trains a physical supervision proxy model. At the current scheduling moment, it acquires the real-time profile-link state vector, forming a slow modal gain scheduling result containing low-order, mid-order, and high-order modal control gains. This invention can reduce modal mismatch caused by uniform scalar gain and improve the optical compensation quality of space-to-ground laser communication links.

Description

Technical Field

[0001] This invention relates to the fields of satellite-to-ground laser communication, free-space optical communication, adaptive optics control, and optical communication link status awareness, specifically to a satellite-to-ground laser communication adaptive optics mode gain scheduling method. Background Technology

[0002] Satellite-to-ground laser communication features narrow beamwidth, high directionality, and high transmission rate, making it suitable for low-Earth orbit satellite data downlink, constellation data backhaul, and high-speed space communication scenarios. Compared to radio frequency (RF) communication, satellite-to-ground laser communication offers higher spectral efficiency; however, the downlink laser link needs to traverse a time-varying atmospheric turbulence medium. Fluctuations in atmospheric refractive index can lead to wavefront distortion, focal spot energy diffusion, and degraded optical quality at the receiver, thus affecting the link's signal-to-noise ratio and communication stability.

[0003] Adaptive optics systems are an important technique for compensating for wavefront distortion caused by atmospheric turbulence. These systems typically acquire wavefront error information using wavefront sensors and perform real-time correction via deformable mirrors. During adaptive optics control, the control gain directly affects the system's response speed to turbulence changes and the degree of noise amplification. Higher gain allows the system to track turbulent disturbances evolving with wind speed more quickly, but wavefront sensor noise is more easily transmitted to the deformable mirror control input; lower gain reduces noise amplification but increases servo hysteresis error.

[0004] Existing adaptive optics gain control methods typically employ fixed gain, empirically tuned gain, and scalar gain scheduling based on a single atmospheric coherence time. Fixed gain methods struggle to adapt to variations in turbulence profiles, lateral wind speeds, elevation angles, and photon availability during satellite transit. While scalar gain scheduling methods can adjust the gain based on coherence time, they apply the same gain simultaneously to low-, mid-, and high-order modes, failing to differentiate the sensitivity differences in servo hysteresis and noise propagation among different mode groups.

[0005] In satellite-to-ground laser communication scenarios, near-surface turbulence, high-altitude winds, wavefront sensor signal-to-noise ratio, photon-confined detection, and satellite overpass elevation angle all collectively affect the adaptive optics control performance. Lower-order modes typically require stronger time tracking capabilities, while higher-order modes are more susceptible to wavefront sensor noise propagation. Using a uniform scalar gain makes it difficult to simultaneously meet the control requirements of different mode groups, leading to increased residual wavefront error and decreased optical quality at the receiver.

[0006] Furthermore, directly performing multimodal gain grid selection for each link state involves a large computational load, which is not conducive to real-time use during the operation of space-to-ground laser communication links. Therefore, there is an urgent need for an adaptive optics modal gain scheduling method that can utilize atmospheric profiles and link sensing states, and can quickly output low-order, mid-order, and high-order modal gains, to provide a state-aware, slow outer-loop control basis for adaptive optics compensation in space-to-ground laser communication. Summary of the Invention

[0007] The technical problem this invention aims to solve is: in satellite-to-ground laser communication links, how to determine the gains of low-order, mid-order, and high-order adaptive optics modes under varying atmospheric turbulence profiles, stratified wind speeds, wavefront sensor signal-to-noise ratios, photon availability, and overpass elevation angles, so that the adaptive optics control process can simultaneously consider servo hysteresis suppression and noise propagation suppression, while avoiding online state-by-state multidimensional gain grid selection. This invention is used in the adaptive optics compensation process of satellite-to-ground laser communication links. By utilizing atmospheric turbulence profiles, stratified wind speeds, wavefront sensor signal-to-noise ratios, photon availability, and overpass elevation angles, it generates gain scheduling results corresponding to low-order, mid-order, and high-order adaptive optics modes, enabling the adaptive optics system to simultaneously consider servo hysteresis suppression and wavefront sensor noise propagation suppression under different link states.

[0008] To address the aforementioned technical problems, this invention provides an adaptive optics mode gain scheduling method for satellite-to-ground laser communication, the method comprising the following steps:

[0009] Step 1: Obtain profile-link state sample data of the satellite-to-ground laser communication link for offline modeling and training, construct an offline training state sample set, and represent each profile-link state sample in the offline training state sample set as a profile-link state vector; the profile-link state sample data includes three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed, Fried parameter, atmospheric coherence time, wavefront sensor signal-to-noise ratio, satellite overpass elevation angle, and photon availability; at the same time, construct low-order, mid-order, and high-order adaptive optics mode groups and corresponding candidate gain sets, and form a mode gain vector composed of low-order candidate gains, mid-order candidate gains, and high-order candidate gains.

[0010] Step 2: Based on the three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed and Fried parameter in the offline training state sample set obtained in Step 1, the effective time frequency calculation relationship of the mode group is introduced to determine the effective time frequency corresponding to the low-order, medium-order and high-order mode groups respectively; and based on the Fried parameter and the satellite overpass elevation angle, the turbulence intensity factor and elevation angle factor are generated.

[0011] Step 3: Based on the effective time frequency, turbulence intensity factor and elevation angle factor obtained in Step 2, introduce the mode group residual phase variance model composed of fitting residual, servo hysteresis residual and wavefront sensor noise propagation residual, and calculate the total residual phase variance corresponding to the mode gain vector formed in Step 1.

[0012] Step 4: Based on the candidate gain set constructed in Step 1 and the total residual phase variance obtained in Step 3, select the mode gain vector that minimizes the total residual phase variance from the Cartesian product of the low-order, mid-order, and high-order candidate gain sets to obtain the offline modal optimal gain label.

[0013] Step 5: Based on the offline training state sample set, profile-link state vector, and offline modal optimal gain label obtained in Step 4, construct a physical supervised modal gain proxy model, and train the physical supervised modal gain proxy model so that the physical supervised modal gain proxy model learns the mapping relationship between the profile-link state vector and the low-order, mid-order, and high-order modal gain vectors.

[0014] Step 6: After completing the training of the physical supervision modal gain surrogate model in Step 5, obtain the satellite-to-ground laser communication link profile-link state data at the current scheduling time, and form the current profile-link state vector according to the state vector construction method in Step 1; use the physical supervision modal gain surrogate model trained in Step 5 to perform gain inference on the current profile-link state vector to obtain the low-order modal gain, mid-order modal gain and high-order modal gain corresponding to the current link state.

[0015] Step 7: Based on the low-order mode gain, mid-order mode gain and high-order mode gain obtained in Step 6, the three types of gains are respectively assigned to the corresponding adaptive optics mode groups to form a slow mode gain scheduling result independent of the high-speed wavefront sensor-deformable mirror closed loop, and the slow mode gain scheduling result is output.

[0016] Further, in step 1, for any profile-link state sample in the offline training state sample set, its profile-link state vector is represented as: in, This represents the state vector corresponding to a profile-link state sample in the offline training state sample set. This represents the fraction of equivalent atmospheric turbulence intensity across three layers. This indicates the lateral wind speed at three levels. Indicates the Fried parameter. Indicates atmospheric coherence time. Indicates the signal-to-noise ratio of the wavefront sensor. Indicates the satellite's elevation angle over the horizon. This indicates the availability of photons.

[0017] Further, in step 1, the modal gain vector is represented as: in, This represents the modal gain vector composed of low-order candidate gains, mid-order candidate gains, and high-order candidate gains. Indicates the low-order mode gain. Indicates the intermediate-order mode gain. This represents the higher-order modal gain. The candidate gain set is represented as: in, , and These represent the candidate gain sets corresponding to the low-order, mid-order, and high-order mode groups, respectively. Let represent the Cartesian product of the three candidate gain sets.

[0018] Further, in step 2, the effective time frequency is expressed as: in, Represents low-order, mid-order, and high-order mode groups. Represents mode group The corresponding effective time frequency, Represents mode group The corresponding modal time scale coefficients, Indicates the equivalent atmospheric layer number. , Indicates the first The percentage of turbulence intensity in each equivalent atmospheric layer Indicates the first The lateral wind speed of an equivalent atmospheric layer. This represents the Fried parameter.

[0019] Further, in step 3, the mode group residual phase variance model is composed of the fitting residuals, servo hysteresis residuals, and wavefront sensor noise propagation residuals of each mode group; based on the mode group residual phase variance model, the total residual phase variance is expressed as: in, State vector and mode gain vector The corresponding total residual phase variance, Represents mode group The fitting residuals, Represents mode group Servo hysteresis residuals Represents mode group The wavefront sensor noise propagation residual, Represents mode group The corresponding gain.

[0020] Further, in step 3, the servo hysteresis residual is expressed as: The noise propagation residual of the wavefront sensor is expressed as: in, Represents mode group Servo lag weight, Represents mode group Noise propagation weight, Indicates the basic control bandwidth. This indicates the gain control bandwidth scale. This represents the turbulence intensity factor corresponding to the Fried parameter. The elevation factor represents the elevation angle corresponding to the satellite's overhead elevation angle. This represents the index indicating the influence of the wavefront sensor's signal-to-noise ratio on the noise propagation residual.

[0021] Further, in step 4, the offline modal optimal gain label is represented as: in, State vector The corresponding offline modal optimal gain label.

[0022] Further, in step 5, the physical supervision modal gain surrogate model takes the profile-link state vector as input and outputs a modal gain vector composed of low-order modal gains, mid-order modal gains, and high-order modal gains. The physical supervision modal gain surrogate model is expressed as follows: in, This represents a physical supervision modal gain surrogate model. Indicates model parameters, The modal gain vector predicted by the surrogate model is represented as: The training objective of the surrogate model is expressed as: in, This represents the training loss of the surrogate model. This represents the number of training state samples. Indicates the first One training state sample, Indicates the first The offline modal optimal gain label corresponding to each training state sample. This represents the squared Euclidean distance.

[0023] Furthermore, in step 7, the slow mode gain scheduling result is expressed as follows: in, This represents the slow mode gain scheduling result formed at the current scheduling moment. Indicates the current scheduling time. , and These represent the control gains allocated to the low-order mode group, mid-order mode group, and high-order mode group at the current scheduling moment, respectively.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. This invention unifies the three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed, Fried parameter, atmospheric coherence time, wavefront sensor signal-to-noise ratio, satellite overpass elevation angle, and photon availability into a profile-link state vector, avoiding reliance on a single turbulence intensity and a single coherence time for gain scheduling, and improving the ability of adaptive optics gain selection to characterize changes in the state of satellite-to-ground laser communication links.

[0026] 2. This invention outputs low-order, mid-order, and high-order adaptive optical mode gains respectively, enabling low-order modes to obtain stronger turbulence time tracking capabilities and enabling high-order modes to reduce the risk of wavefront sensor noise propagation. Compared with a unified scalar gain scheduling method, it is more in line with the control requirements of different mode groups.

[0027] 3. This invention generates offline modal optimal gain labels by using a modal group residual phase variance model composed of fitting residuals, servo hysteresis residuals, and wavefront sensor noise propagation residuals. This provides clear physical supervision for the surrogate model training process and reduces the problem of lack of optical physical constraints in pure data-driven gain prediction.

[0028] 4. This invention utilizes the constructed and trained physical supervised modal gain proxy model to transform offline multidimensional gain grid screening into online fast state-to-gain inference, avoiding repeated low-order, mid-order, and high-order candidate gain vector screening during the operation of the satellite-to-ground laser communication link, thereby improving online scheduling efficiency.

[0029] 5. In the test results corresponding to the embodiments of the present invention, the average residual phase variance of the physically supervised modal gain surrogate model is 1.891 rad², which is close to the 1.887 rad² corresponding to the offline modal optimal gain label, and better than the 2.096 rad² of fixed modal gain, the 2.121 rad² of coherent time scalar scheduling, and the 2.022 rad² of scalar optimal gain. In the three-layer phase screen overpass simulation, the 5th percentile Strehl of the physically supervised modal gain surrogate model is 0.714, which is higher than the 0.608 of fixed modal gain and the 0.477 of coherent time scalar scheduling, indicating that the present invention can improve the low-tail optical quality. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall architecture of the adaptive optics mode gain scheduling method for satellite-to-ground laser communication according to the present invention.

[0031] Figure 2 This is a schematic diagram of the offline modal optimal gain label generation and physically supervised modal gain surrogate model training of the present invention.

[0032] Figure 3 This is a schematic diagram of the online profile-link state reasoning and low-order, mid-order, and high-order mode gain allocation of the present invention.

[0033] Figure 4 This is a schematic diagram comparing the residual phase variance and Strehl performance of different gain scheduling methods in the embodiments of the present invention. Detailed Implementation

[0034] The present invention will now be described with reference to an embodiment. This embodiment is used to illustrate the technical solution of the present invention and is not intended to limit the scope of protection of the present invention.

[0035] This embodiment focuses on the downlink of satellite-to-ground laser communication between a low-Earth orbit satellite and a ground receiving station. The adaptive optics system includes a wavefront sensor, a deformable mirror, and a mode controller. The mode controller divides the correction modes into low-order mode groups, mid-order mode groups, and high-order mode groups, and performs slow gain scheduling outside the high-speed wavefront sensor-deformable mirror closed loop.

[0036] like Figure 1 As shown, the adaptive optics mode gain scheduling method for satellite-to-ground laser communication of the present invention includes an offline modeling and training stage and an online slow scheduling stage. This embodiment is executed according to steps 1 to 7.

[0037] Step 1: Obtain profile-link state sample data of the satellite-to-ground laser communication link for offline modeling and training, and construct an offline training state sample set. The profile-link state sample data includes three-layer equivalent atmospheric turbulence intensity fractions, three-layer lateral wind speeds, Fried parameters, atmospheric coherence time, wavefront sensor signal-to-noise ratio, satellite overpass elevation angle, and photon availability. Each profile-link state sample in the offline training state sample set is represented as a profile-link state vector.

[0038] In this step, the three equivalent atmospheric layers correspond to the near-surface layer, the free troposphere, and the upper atmosphere, respectively. For any profile-link state sample in the offline training state sample set, its profile-link state vector is represented as: in, This represents the fraction of equivalent atmospheric turbulence intensity across three layers. This indicates the lateral wind speed at three levels. Indicates the Fried parameter. Indicates atmospheric coherence time. Indicates the signal-to-noise ratio of the wavefront sensor. Indicates the satellite's elevation angle over the horizon. This indicates the availability of photons.

[0039] Simultaneously, low-order, mid-order, and high-order adaptive optics mode sets and corresponding candidate gain sets are constructed. In this embodiment, the mode sets include low-order, mid-order, and high-order mode sets, and the candidate gains are uniformly selected from 0.10 to 0.90, taking 17 values. The Cartesian product of the low-order, mid-order, and high-order candidate gain sets forms the candidate gain set. Any low-order candidate gain, mid-order candidate gain, and high-order candidate gain form a modal gain vector, and the candidate gain set is... The number of modal gain vectors is 4913.

[0040] Step 2: Using the three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed, and Fried parameter from the offline training state sample set, the effective time frequency calculation relationship of the mode group is introduced to determine the effective time frequency corresponding to the low-order, mid-order, and high-order mode groups respectively.

[0041] In this step, the effective time frequency of the mode group is calculated according to the following relationship: in, Represents mode group The modal time scale coefficients are as follows. In this embodiment, the modal time scale coefficients for the low-order, mid-order, and high-order mode groups are 1.25, 1.00, and 0.75, respectively.

[0042] Simultaneously, a turbulence intensity factor is generated based on the Fried parameter, and an elevation angle factor is generated based on the satellite's transit elevation angle. These turbulence intensity and elevation angle factors are used for servo hysteresis residual calculation in step 3, ensuring that the time tracking cost under low elevation angle and strong turbulence conditions is reflected in the residual phase variance.

[0043] Step 3: Based on the effective time frequency, turbulence intensity factor and elevation angle factor obtained in Step 2, construct the mode group residual phase variance model, and calculate the total residual phase variance corresponding to the different mode gain vectors formed in Step 1.

[0044] In this step, the mode group residual phase variance model is composed of the fitting residuals, servo hysteresis residuals, and wavefront sensor noise propagation residuals of each mode group; for any profile-link state vector and any candidate mode gain vector Calculate the total residual phase variance: in, State vector and mode gain vector The corresponding total residual phase variance, Represents mode group The fitting residuals, Represents mode group Servo hysteresis residuals Represents mode group The wavefront sensor noise propagation residual, Represents mode group The corresponding gain.

[0045] The fitting residual is used to characterize the residual error caused by the finite space order correction; the servo hysteresis residual is used to characterize the error caused by insufficient tracking of turbulent time changes when the gain is insufficient; the wavefront sensor noise propagation residual is used to characterize the error caused by the amplification of sensor noise when the gain is too high.

[0046] The servo hysteresis residual is expressed as: The noise propagation residual of the wavefront sensor is expressed as: in, Represents mode group Servo lag weight, Represents mode group Noise propagation weight, Indicates the basic control bandwidth. This indicates the gain control bandwidth scale. This represents the turbulence intensity factor corresponding to the Fried parameter. The elevation factor represents the elevation angle corresponding to the satellite's overhead elevation angle. This represents the index indicating the influence of the wavefront sensor's signal-to-noise ratio on the noise propagation residual.

[0047] In this embodiment, the modal weights of the low-order, mid-order, and high-order mode groups are 0.45, 0.34, and 0.21, respectively; the noise scales of the low-order, mid-order, and high-order mode groups are 0.45, 1.60, and 5.00, respectively; and the basic control bandwidth... 5Hz, gain control bandwidth scale It is 75Hz.

[0048] From the above relationship, it can be seen that the gain As the hysteresis residual increases, the servo hysteresis residual decreases, while the wavefront sensor noise propagation residual increases. Therefore, this step provides the physical basis for the modal gain optimization in step 4.

[0049] Step 4: Based on the candidate gain set constructed in Step 1 and the total residual phase variance corresponding to each modal gain vector obtained in Step 3, generate offline modal optimal gain labels.

[0050] In this step, the candidate gain set is... Each mode gain vector in the model is calculated using the mode group residual phase variance model from step 3, and the mode gain vector with the smallest residual phase variance is selected. This yields the profile-link state vector. Corresponding offline modal optimal gain label .like Figure 2 As shown, in the offline stage, the present invention iterates through different modal gain vectors based on the offline training state sample set and the candidate gain set. The modal gain vector with the smallest total residual phase variance is selected by the modal group residual phase variance model in step 3. The profile-link state vector and the offline modal optimal gain label are used to form a training sample pair, which provides supervision labels for the training of the physical supervision modal gain proxy model in step 5.

[0051] Step 5: Based on the offline training state sample set constructed in Step 1, the profile-link state vector, and the offline modal optimal gain label obtained in Step 4, construct and train the physical supervised modal gain surrogate model.

[0052] In this step, a physically supervised modal gain surrogate model is constructed, taking the profile-link state vector as input and the modal gain vector composed of low-order, mid-order, and high-order modal gains as output. The profile-link state vector from the offline training state sample set is used as the model input, and the offline modal optimal gain label is used as the supervision label to train the physically supervised modal gain surrogate model. in, This represents a physical supervision modal gain surrogate model. Indicates model parameters, This represents the modal gain vector predicted by the surrogate model.

[0053] The training objective is: in, This represents the training loss of the surrogate model. This represents the number of training state samples. Indicates the first One training state sample, Indicates the first The offline modal optimal gain label corresponding to each training state sample. This represents the squared Euclidean distance.

[0054] In this embodiment, the number of training samples is 3600, the number of test samples is 1200, and the experiment is repeated on five random seeds. The physical supervised modal gain surrogate model adopts a three-layer multilayer perceptron with 128 hidden units per layer, and is trained using the GELU activation function and the AdamW optimizer.

[0055] In the ablation tests of this embodiment, the physically supervised modal gain surrogate model under the complete profile-link state input outperforms the fixed modal gain, coherent time scalar scheduling, and scalar optimal gain methods. The ablation results show that the three-layer equivalent atmospheric turbulence intensity fraction, wavefront sensor signal-to-noise ratio, and photon availability play a crucial role in modal gain scheduling.

[0056] Step 6: After completing the training of the physical supervision mode gain surrogate model, obtain the star-ground laser communication link profile - link status data at the current scheduling time, and use the physical supervision mode gain surrogate model trained in Step 5 to obtain the current mode gain.

[0057] In this step, the three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed, Fried parameters, atmospheric coherence time, wavefront sensor signal-to-noise ratio, satellite overpass elevation angle, and photon availability obtained at the current scheduling time are used to form the current profile-link state vector according to the state vector construction method in step 1. And input it into the agent model trained in step 5: in, , and These represent the low-order, mid-order, and high-order mode gains corresponding to the current link state, respectively. For example... Figure 3 As shown, in the online phase, after the current profile-link state data is formed into a current profile-link state vector according to the state vector construction method in step 1, the physical-supervised modal gain surrogate model trained in step 5 infers the low-order, mid-order, and high-order current modal gains online and assigns them to the corresponding modal groups. Further tests show that the physical-supervised modal gain surrogate model is close to the offline modal optimal gain in terms of state-level residuals. The median residual difference between the surrogate model and the offline modal optimal result is 0.0007 rad², and the 95th percentile difference is 0.023 rad².

[0058] Step 7: Based on the low-order mode gain, mid-order mode gain and high-order mode gain obtained in Step 6, output the slow mode gain scheduling result.

[0059] In this step, low-order modal gains are assigned to the low-order mode group, mid-order modal gains are assigned to the mid-order mode group, and high-order modal gains are assigned to the high-order mode group, resulting in a slow modal gain scheduling outcome. in, This represents the slow-mode gain scheduling result formed at the current scheduling moment. The scheduling result contains... , and As the control gain of the corresponding mode group, it acts outside the high-speed wavefront sensor-deformed mirror closed loop to adjust the control gain of different mode groups without changing the basic structure of wavefront measurement and deformed mirror drive in the high-speed closed loop.

[0060] In the equivalent turbulence intensity variation test, the physically supervised modal gain surrogate model maintained a small residual difference relative to the offline modal optimal gain label, while the coherent time scalar scheduling and scalar optimal gain still showed a significant residual difference. This result indicates that the advantage of this embodiment is not a random phenomenon under a specific turbulence intensity, but rather that it can maintain the modal scheduling advantage under different turbulence intensities.

[0061] like Figure 4As shown, this embodiment of the invention compares the physical supervision modal gain proxy model used in this invention with the fixed modal gain method, the coherent time scalar scheduling method, the scalar optimal gain method, and the offline modal optimal gain label. Specifically, the fixed modal gain method represents a baseline method that does not adjust the modal gain according to changes in link state; the coherent time scalar scheduling method represents a scheduling method that generates a uniform scalar gain based solely on atmospheric coherence time; the scalar optimal gain method represents a comparative method that optimizes within a single scalar gain range; and the offline modal optimal gain label represents a reference upper limit obtained by offline traversal of candidate modal gain vectors and selecting the result with the minimum residual phase variance.

[0062] Figure 4 (a) is used to compare the average residual phase variance corresponding to different gain scheduling methods. The average residual phase variance is used to characterize the magnitude of the residual wavefront error after adaptive optics compensation. The lower the value, the smaller the residual after compensation. Figure 4 (b) is used to compare the low-tail Strehl performance of different gain scheduling methods under non-stationary three-layer phase screen overpass simulation. The low-tail Strehl performance is used to characterize the optical compensation quality under adverse conditions such as low elevation angle, strong turbulence and photon confinement during satellite overpass. The higher the height of the bar, the better the link optical quality under risk-sensitive conditions.

[0063] Depend on Figure 4 It is evident that the fixed-mode-gain method, due to its failure to utilize current profile-link state information, struggles to adapt to changes in turbulence profiles, lateral wind speeds, elevation angles, and photon availability during satellite transit. Consequently, it exhibits high residual phase variance and limited low-tail Strehl performance. While the coherent-time scalar scheduling method utilizes atmospheric coherence time to adjust the gain, it applies a uniform scalar gain to low-, mid-, and high-order mode groups, failing to reflect the varying sensitivities of different mode groups to servo hysteresis and wavefront sensor noise propagation. Therefore, it remains inadequate in residual control and low-tail optical quality preservation. Although the scalar-optimal gain method optimizes within a single scalar gain range, it still cannot determine control gains separately for different mode groups, thus making it difficult to simultaneously address the time tracking requirements of low-order modes and the noise suppression requirements of high-order modes.

[0064] Compared to the methods described above, the physical-supervised modal gain proxy model employed in this invention can output low-order, mid-order, and high-order modal gains based on the current profile-link state vector, enabling different modal groups to obtain gain configurations that match their control characteristics. Its performance trend closely approximates the offline modal optimal gain label and outperforms the fixed modal gain method, the coherent time scalar scheduling method, and the scalar optimal gain method in both average residual phase variance and low-tail Strehl performance dimensions. This demonstrates that this embodiment can maintain near-offline modal optimal scheduling performance while avoiding online traversal search of the candidate gain set, and improve the optical compensation quality in risk-sensitive areas during satellite-to-ground laser communication transit.

Claims

1. A method for adaptive optics mode gain scheduling in satellite-to-ground laser communication, characterized in that, The method includes the following steps: Step 1: Obtain profile-link state sample data of the satellite-to-ground laser communication link for offline modeling and training, construct an offline training state sample set, and represent each profile-link state sample in the offline training state sample set as a profile-link state vector; the profile-link state sample data includes three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed, Fried parameter, atmospheric coherence time, wavefront sensor signal-to-noise ratio, satellite overpass elevation angle, and photon availability; at the same time, construct low-order, mid-order, and high-order adaptive optics mode groups and corresponding candidate gain sets, and form a mode gain vector composed of low-order candidate gains, mid-order candidate gains, and high-order candidate gains; Step 2: Based on the three-layer equivalent atmospheric turbulence intensity fraction, three-layer lateral wind speed and Fried parameter in the offline training state sample set obtained in Step 1, the effective time frequency calculation relationship of the mode group is introduced to determine the effective time frequency corresponding to the low-order, medium-order and high-order mode groups respectively; and based on the Fried parameter and the satellite overpass elevation angle, the turbulence intensity factor and elevation angle factor are generated. Step 3: Based on the effective time frequency, turbulence intensity factor and elevation angle factor obtained in Step 2, introduce the mode group residual phase variance model composed of fitting residual, servo hysteresis residual and wavefront sensor noise propagation residual, and calculate the total residual phase variance corresponding to the mode gain vector formed in Step 1. Step 4: Based on the candidate gain set constructed in Step 1 and the total residual phase variance obtained in Step 3, select the mode gain vector that minimizes the total residual phase variance from the Cartesian product of the low-order, mid-order and high-order candidate gain sets to obtain the offline modal optimal gain label. Step 5: Based on the offline training state sample set, profile-link state vector, and offline modal optimal gain label obtained in Step 4, construct a physical supervised modal gain proxy model, and train the physical supervised modal gain proxy model so that the physical supervised modal gain proxy model learns the mapping relationship between the profile-link state vector and the low-order, mid-order, and high-order modal gain vectors. Step 6: After completing the training of the physical supervision modal gain proxy model in Step 5, obtain the satellite-to-ground laser communication link profile-link state data at the current scheduling time, and form the current profile-link state vector according to the state vector construction method in Step 1; use the physical supervision modal gain proxy model trained in Step 5 to perform gain inference on the current profile-link state vector to obtain the low-order modal gain, mid-order modal gain and high-order modal gain corresponding to the current link state; Step 7: Based on the low-order mode gain, mid-order mode gain and high-order mode gain obtained in Step 6, the three types of gains are respectively assigned to the corresponding adaptive optics mode groups to form a slow mode gain scheduling result independent of the high-speed wavefront sensor-deformable mirror closed loop, and the slow mode gain scheduling result is output.

2. The adaptive optics mode gain scheduling method for satellite-to-ground laser communication according to claim 1, characterized in that, In step 1, for any profile-link state sample in the offline training state sample set, its profile-link state vector is represented as: in, This represents the state vector corresponding to a profile-link state sample in the offline training state sample set. This represents the fraction of equivalent atmospheric turbulence intensity across three layers. This indicates the lateral wind speed at three levels. Indicates the Fried parameter. Indicates atmospheric coherence time. Indicates the signal-to-noise ratio of the wavefront sensor. Indicates the satellite's elevation angle over the horizon. Indicates photon availability; In step 1, the modal gain vector is represented as: in, This represents the modal gain vector composed of low-order candidate gains, mid-order candidate gains, and high-order candidate gains. Indicates the low-order mode gain. Indicates the intermediate-order mode gain. Indicates higher-order mode gain; The candidate gain set is represented as follows: in, , and These represent the candidate gain sets corresponding to the low-order, mid-order, and high-order mode groups, respectively. This represents the Cartesian product of the three candidate gain sets.

3. A method for adaptive optics mode gain scheduling in satellite-to-ground laser communication according to claim 1 or 2, characterized in that, In step 2, the effective time frequency is expressed as: in, Represents low-order, mid-order, and high-order mode groups. Represents mode group The corresponding effective time frequency, Represents mode group The corresponding modal time scale coefficients, Indicates the equivalent atmospheric layer number. , Indicates the first The percentage of turbulence intensity in each equivalent atmospheric layer Indicates the first The lateral wind speed of an equivalent atmospheric layer. This represents the Fried parameter.

4. The adaptive optics mode gain scheduling method for satellite-to-ground laser communication according to claim 3, characterized in that, In step 3, the mode group residual phase variance model is composed of the fitting residuals, servo hysteresis residuals, and wavefront sensor noise propagation residuals of each mode group; based on the mode group residual phase variance model, the total residual phase variance is expressed as: in, Represents the state vector and mode gain vector The corresponding total residual phase variance, Represents mode group The fitting residuals, Represents mode group Servo hysteresis residuals Represents mode group The wavefront sensor noise propagation residual, Represents mode group The corresponding gain.

5. The adaptive optics mode gain scheduling method for satellite-to-ground laser communication according to claim 4, characterized in that, The servo hysteresis residual mentioned in step 3 is expressed as: The noise propagation residual of the wavefront sensor is expressed as: in, Represents mode group Servo lag weight, Represents mode group Noise propagation weight, Indicates the basic control bandwidth. This indicates the gain control bandwidth scale. This represents the turbulence intensity factor corresponding to the Fried parameter. The elevation factor represents the elevation angle corresponding to the satellite's overhead elevation angle. This represents the index indicating the influence of the wavefront sensor's signal-to-noise ratio on the noise propagation residual.

6. The adaptive optics mode gain scheduling method for satellite-to-ground laser communication according to claim 5, characterized in that, In step 4, the offline modal optimal gain label is represented as: in, Represents the state vector The corresponding offline modal optimal gain label.

7. The adaptive optics mode gain scheduling method for satellite-to-ground laser communication according to claim 6, characterized in that, In step 5, the physical supervision modal gain surrogate model takes the profile-link state vector as input and outputs a modal gain vector composed of low-order modal gains, mid-order modal gains, and high-order modal gains. The physical supervision modal gain surrogate model is expressed as follows: in, This represents a physical supervision modal gain surrogate model. Indicates model parameters, This represents the modal gain vector predicted by the surrogate model. The training objective of the surrogate model is represented as: in, This represents the training loss of the surrogate model. This represents the number of training state samples. Indicates the first One training state sample, Indicates the first The offline modal optimal gain label corresponding to each training state sample. This represents the squared Euclidean distance.

8. A method for adaptive optics mode gain scheduling in satellite-to-ground laser communication according to claim 1 or 2, characterized in that, In step 7, the slow mode gain scheduling result is expressed as follows: in, This represents the slow mode gain scheduling result formed at the current scheduling moment. Indicates the current scheduling time. , and These represent the control gains allocated to the low-order mode group, mid-order mode group, and high-order mode group at the current scheduling moment, respectively.

Images