Satellite-ground laser communication physical consistent multi-time domain prediction modulation coding control method

Abstract

This invention discloses a physically consistent multi-temporal predictive modulation and coding control method for space-to-ground laser communication, belonging to the fields of space laser communication, free-space optical communication link prediction, and adaptive modulation and coding control. The method acquires historical link state data, independent calibration data, modulation and coding switching delays, a prediction time domain set, and a threshold table to construct a multimodal time token sequence; trains a physically consistent multi-temporal prediction model and determines the effective prediction time domain based on the switching delay, obtaining predicted values ​​for the Strehl ratio, effective signal-to-noise ratio, and logarithmic bit error rate; forms a calibration sample set based on independent calibration data and generates an effective signal-to-noise ratio prediction interval, taking the lower bound of the interval as a conservative effective signal-to-noise ratio; and then combines the threshold table to determine the target modulation and coding scheme and output the control results. This invention can reduce the risk of modulation and coding mismatch caused by delay and turbulence variations, improving link availability.

Description

Technical Field

[0001] This invention relates to the fields of space laser communication, free space optical communication link prediction, and adaptive modulation and coding control technology, specifically to a physical consistency multi-temporal-domain predictive modulation and coding control method for satellite-to-ground laser communication. 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 constellations, inter-satellite data backhaul, and high-speed satellite-to-ground downlink transmission scenarios. Compared to radio frequency (RF) communication, satellite-to-ground laser communication offers higher spectral efficiency and stronger anti-interception capabilities; however, its link quality is highly sensitive to atmospheric turbulence, beam drift, scintillation, wavefront distortion, and intensity fluctuations within the receiving aperture. On millisecond timescales, atmospheric turbulence can cause a decrease in the Strehl ratio, rapid fluctuations in the effective signal-to-noise ratio, and an increase in the bit error rate, thus affecting the selection of modulation and coding schemes.

[0003] Existing adaptive modulation and coding techniques typically select the modulation and coding scheme based on the current signal-to-noise ratio (SNR), historical channel quality, and short-term predicted channel quality fed back from the receiver. When the link quality is good, the system selects a modulation and coding scheme with higher spectral efficiency; when the link quality is poor, the system selects a more robust modulation and coding scheme. This type of method has been applied in terrestrial wireless communications and some satellite radio frequency links.

[0004] However, in satellite-to-ground laser communication links, receiver observation, feedback transmission, transmitter calculation, modulation and coding configuration updates, and frame boundary waiting all introduce latency. If the transmitter selects modulation and coding based solely on the current observation and the feedback value from the previous moment, the selection result will lag behind the link changes caused by turbulence by the time it actually takes effect in the future, leading to misselection of higher-order modulation, frame loss, and a decrease in link availability.

[0005] While existing predictive adaptive modulation and coding methods can extrapolate future channel quality using historical signal-to-noise ratios (SNR), their prediction targets are typically limited to a single SNR scalar, failing to simultaneously constrain the optical communication physical relationships between Strehl ratio, Fried parameter, phase distortion, bit error rate, and SNR. Consequently, the prediction results may exhibit small statistical errors but inconsistent optical physical meanings, leading to a lack of reliable basis for subsequent control decisions.

[0006] Furthermore, existing methods often employ a fixed safety margin to adjust the prediction signal-to-noise ratio in order to reduce the risk of misselection of higher-order modulation. However, a fixed safety margin cannot adapt to the differences in prediction error distribution across different prediction time domains, turbulence intensities, and link conditions. If the safety margin is too small, the risk of link interruption is high; if the safety margin is too large, the system will consistently select lower-order modulation and coding schemes, leading to throughput loss.

[0007] Therefore, there is an urgent need for an adaptive modulation control method for satellite-to-ground laser communication links, which can utilize multi-temporal link performance prediction results that conform to the physical relationship of the optical link before the modulation and coding scheme takes effect in the future, and replace the fixed empirical margin with a calibrated lower bound of the prediction interval, thereby forming a controllable balance between link availability and transmission efficiency. Summary of the Invention

[0008] To address the aforementioned problems, this invention proposes a physically consistent multi-temporal-domain predictive modulation and coding control method for satellite-to-ground laser communication. The technical problem this invention aims to solve is: in satellite-to-ground laser communication links, under conditions of rapid changes in atmospheric turbulence, feedback delay, and modulation and coding switching delay, how to determine the modulation and coding scheme that can meet the link reliability requirements at the future effective time, based on physically consistent future link performance prediction results and a calibrated lower bound of the signal-to-noise ratio prediction interval? This invention provides a prediction interval-driven adaptive modulation and coding control method for satellite-to-ground laser communication. This invention is applied to satellite-to-ground laser communication links affected by rapid changes in atmospheric turbulence. By performing physically consistent predictions of future link performance, calibrating the prediction results within an interval, and determining the modulation and coding scheme based on the lower bound of the prediction interval, the modulation and coding scheme is matched to the link state at the actual future effective time.

[0009] To solve the above-mentioned technical problems, the present invention provides a physically consistent multi-temporal-domain predictive modulation and coding control method for satellite-to-ground laser communication, the method comprising the following steps:

[0010] Step 1: Obtain historical link status data, independent calibration data, modulation and coding switching delay, predicted time domain set, and modulation and coding threshold table of the satellite-to-ground laser communication link. Construct a historical time window link status sequence based on the historical link status data, and determine the future effective time of the modulation and coding mode based on the modulation and coding switching delay.

[0011] Step 2: Based on the historical time window link state sequence constructed in Step 1, multimodal grouping feature mapping is introduced to improve the problem that the single link scalar input is insufficient to express the turbulent state, forming a multimodal time token sequence.

[0012] Step 3: Based on the multimodal time token sequence formed in Step 2, construct a physically consistent multitemporal prediction model. The physically consistent multitemporal prediction model includes a multimodal grouping mapping layer, a position encoding layer, a temporal attention encoding layer, and a multi-task prediction head.

[0013] Step 4 introduces a joint training objective consisting of link performance prediction loss, Maréchal coupling constraint, phase structure function constraint, and cross-temporal Kolmogorov evolution constraint to train the physically consistent multi-temporal prediction model constructed in Step 3.

[0014] Step 5: Based on the physically consistent multi-temporal prediction model trained in Step 4, inference is performed on the multimodal time token sequence formed in Step 2 to generate the Strehl ratio prediction value, effective signal-to-noise ratio prediction value, and logarithmic bit error rate prediction value corresponding to each prediction time domain in the prediction time domain set determined in Step 1; and based on the modulation and coding switching delay determined in Step 1, the effective prediction time domain corresponding to the future effective time is determined from the prediction time domain set, and the Strehl ratio prediction value, effective signal-to-noise ratio prediction value, and logarithmic bit error rate prediction value corresponding to the effective prediction time domain are obtained.

[0015] Step 6: Based on the independent calibration data obtained in Step 1, form a calibration sample set, and perform conformal prediction interval calibration on the effective signal-to-noise ratio prediction value corresponding to the effective prediction time domain obtained in Step 5 to obtain the effective signal-to-noise ratio prediction interval, and take the lower bound of the effective signal-to-noise ratio prediction interval as the conservative effective signal-to-noise ratio.

[0016] Step 7: Based on the conservative effective signal-to-noise ratio obtained in Step 6 and the modulation and coding threshold table obtained in Step 1, determine the candidate modulation and coding schemes that meet the signal-to-noise ratio threshold constraints, and determine the target modulation and coding scheme with the highest spectral efficiency from the candidate modulation and coding schemes.

[0017] Step 8: Based on the future effective time determined in Step 1, associate the target modulation and coding scheme determined in Step 7 with the corresponding frame number and timestamp, and output the adaptive modulation control result for the satellite-to-ground laser communication link.

[0018] Furthermore, step 1 specifically includes the following sub-steps:

[0019] Step 1.1: Obtain historical link status data of the satellite-to-ground laser communication link. The historical link status data includes low-order Zernike aberration coefficients, receiver aperture statistics, Fried parameters, Rytov variance, Gamma-Gamma distribution parameters, Strehl ratio, effective signal-to-noise ratio, logarithmic bit error rate, and meteorological covariates. The meteorological covariates include air temperature, air pressure, relative humidity, horizontal wind speed component, solar shortwave radiation, and boundary layer height. Based on the historical link status data, construct training samples for model parameter training in Step 4, and divide the training samples into a validation set.

[0020] Step 1.2: Obtain independent calibration data for the satellite-to-ground laser communication link. The independent calibration data includes the historical time window link state sequence that does not participate in the model parameter training and its corresponding effective signal-to-noise ratio true value in the predicted time domain, which is used to form a calibration sample set in step 6.

[0021] Step 1.3: Obtain the modulation and coding threshold table, which includes multiple candidate modulation and coding schemes, the signal-to-noise ratio (SNR) threshold corresponding to each candidate modulation and coding scheme, and the spectral efficiency corresponding to each candidate modulation and coding scheme. The SNR threshold is used in step 7 to compare with the conservative effective SNR obtained in step 6 to determine the candidate modulation and coding schemes that meet the SNR threshold constraints. The spectral efficiency is used to determine the target modulation and coding scheme from the candidate modulation and coding schemes that meet the SNR threshold constraints.

[0022] Step 1.4: Construct a historical time window link state sequence based on the historical link state data: the current decision moment is denoted as... The length of the historical time window is denoted as The sampling interval is denoted as Number of time steps within a historical time window Represented as: Historical time window link state sequence Represented as: in, This indicates the number of time steps within a historical time window. Indicates the length of the historical time window. Indicates the sampling interval. Indicates the current decision-making moment Available historical time window link state sequences, Indicates the first The link state vector corresponding to each time step.

[0023] Step 1.5: Obtain the modulation and coding switching delay of the satellite-to-ground laser communication link, and determine the future effective time of the modulation and coding scheme based on the modulation and coding switching delay. Delay observed at the receiving end Adaptive optics processing delay Modulation and coding decision delay Modulation coding configuration switching delay The sum is obtained as follows: Future effective date Represented as: in, This represents the total switching delay required from the current decision-making moment to the actual implementation of the modulation and coding scheme. This indicates the moment when the target modulation and coding scheme actually begins to take effect on the satellite-to-ground laser communication link.

[0024] Step 1.6: Determine the prediction time domain set, wherein the shortest prediction time domain in the prediction time domain set is equal to the sampling interval. Correspondingly, the longest prediction time domain is not less than the modulation and coding switching delay. This enables subsequent step 5 to select the time of future effectiveness from the predicted time domain set. The corresponding effective prediction time domain.

[0025] Furthermore, the specific method of step 2 is to use the historical time window link state sequence constructed in step 1. Based on their physical meaning, they are divided into low-order aberration feature groups. Link turbulence and performance scalar characteristics and meteorological covariate characteristic group ; for time step Low-order aberration features Link turbulence and performance scalar characteristics Characteristics of meteorological covariates Perform linear mappings separately: in, , and These represent the linear mapping matrices corresponding to the low-order aberration feature group, the link turbulence and performance scalar feature group, and the meteorological covariate feature group, respectively. , and These represent the corresponding biases. , and These represent the embedding vectors after mapping the low-order aberration feature set, the link turbulence and performance scalar feature set, and the meteorological covariate feature set, respectively. , and Combined with modal bias and temporal position encoding, the time step is obtained. Corresponding multimodal time token : in, , and These represent the modal biases corresponding to the low-order aberration feature group, the link turbulence and performance scalar feature group, and the meteorological covariate feature group, respectively. Indicates time step Corresponding time location code;

[0026] The multimodal time token sequence is composed of multimodal time tokens from all time steps. .

[0027] Further, in step 3, the physically consistent multi-temporal prediction model includes a multimodal packet mapping layer, a location encoding layer, a temporal attention encoding layer, and a multi-task prediction head; the multimodal packet mapping layer is used to receive the multimodal time token sequence formed in step 2, the location encoding layer is used to maintain the order information of different time steps within the historical time window, the temporal attention encoding layer is used to encode the link state evolution relationship within the historical time window, and the multi-task prediction head is set according to the prediction time domain set determined in step 1 to form the link performance prediction output channel corresponding to each prediction time domain.

[0028] Furthermore, in step 4, the joint training objective is represented as: in, Indicates the joint training objective. The link performance prediction loss is represented by Strehl ratio, effective signal-to-noise ratio, and logarithm of bit error rate. This represents the cross-temporal Kolmogorov evolution constraint loss. This represents the phase structure function constraint loss. This represents the Maréchal coupling constraint loss. , and This represents the weight of the corresponding constraint loss.

[0029] Furthermore, in step 4, the Maréchal coupling constraint loss The formula used to constrain the coupling relationship between the effective signal-to-noise ratio prediction and the Strehl ratio prediction is as follows: in, This indicates the number of prediction time domains in the prediction time domain set. Represents a prediction time domain in the set of prediction time domains. Represents the prediction time domain The corresponding effective signal-to-noise ratio prediction value, Represents the prediction time domain The corresponding Strehl ratio predicted value, and This represents the link budget constant obtained based on the training samples.

[0030] Furthermore, in step 4, the phase structure function constraint loss The complete formula for constraining the phase structure function relationship between the short-time domain Strehl ratio prediction and the Fried parameter is as follows: in, Represents the shortest prediction time domain. This represents the Strehl ratio prediction value corresponding to the shortest prediction time domain. Indicates the diameter of the receiving aperture. Indicates the Fried parameter. and This represents the phase structure function constant obtained based on the training samples calibration.

[0031] Furthermore, in step 4, the cross-temporal Kolmogorov evolution constraint loss is used to constrain the smoothness of turbulent evolution across multiple prediction time domains. First, the following variables are constructed: and make and Satisfies a linear fit relationship: The cross-temporal Kolmogorov evolutionary constraint loss The complete formula is: in, Represents the prediction time domain The corresponding phase variance characterization quantity, This represents the Kolmogorov power-law scale after time-domain normalization for prediction. Represents the maximum prediction time domain. and This represents the linear parameters obtained by fitting the same training sample across multiple prediction time domains.

[0032] Furthermore, the specific method for step 5 is as follows:

[0033] Step 5.1, use the multimodal time token sequence obtained in Step 2. Input the physically consistent multi-temporal prediction model trained in step 4, and obtain the temporal representation of the historical time window through the temporal attention encoding layer. : in, This represents the temporal attention encoding operation. Represents the time sequence of a historical time window; [take / extract] The encoding vector corresponding to the last time step is used as the representation of the current link state. And by predicting the time domain Corresponding multi-task prediction head Generate multiple link performance predictions in the prediction time domain: in, Represents the prediction time domain set, Represents a prediction time domain in the set of prediction time domains. Represents the prediction time domain The corresponding Strehl ratio predicted value, Represents the prediction time domain The corresponding effective signal-to-noise ratio prediction value, Represents the prediction time domain The corresponding logarithmic prediction of the bit error rate, Represents the prediction time domain The corresponding prediction head.

[0034] Step 5.2, Effective prediction of the time domain Determine as follows: in, Represents the prediction time domain set, This represents the total switching delay required from the current decision-making moment to the actual implementation of the modulation and coding scheme. This represents the effective prediction time domain used for modulation and coding selection.

[0035] Step 5.3, based on the effective prediction time domain From the link performance prediction values ​​generated by the physically consistent multi-temporal prediction model trained in step 4 across multiple prediction temporal domains, determine the link performance prediction results corresponding to the effective prediction temporal domains: in, Indicates the effective prediction time domain The corresponding Strehl ratio predicted value, Indicates the effective prediction time domain The corresponding effective signal-to-noise ratio prediction value, Indicates the effective prediction time domain The corresponding logarithmic prediction of the bit error rate.

[0036] Further, in step 6, the independent calibration data obtained in step 1 is input into the physically consistent multi-temporal prediction model trained in step 4 to obtain the predicted effective signal-to-noise ratio (SNR) value in the prediction time domain corresponding to each calibration sample in the independent calibration data, and these predicted values ​​together with the corresponding true effective SNR values ​​in the independent calibration data form a calibration sample set; the absolute residual between the true effective SNR value and the predicted effective SNR value is calculated in the calibration sample set, and the result is determined according to a preset coverage rate. Determine the effective prediction time domain Corresponding empirical quantiles Then, based on the effective signal-to-noise ratio prediction values ​​obtained in step 5, an effective signal-to-noise ratio prediction interval is generated: in, This represents the effective signal-to-noise ratio prediction interval corresponding to the effective prediction time domain. This represents the lower bound of the effective signal-to-noise ratio prediction interval. This represents the upper bound of the effective signal-to-noise ratio prediction interval;

[0037] The lower bound of the effective signal-to-noise ratio prediction interval is taken as the conservative effective signal-to-noise ratio: in, This represents the conservative effective signal-to-noise ratio corresponding to the effective prediction time domain. This represents the effective signal-to-noise ratio prediction value in the effective prediction time domain obtained in step 5. This represents the empirical quantile of the prediction residual corresponding to the effective prediction time domain in the calibration sample set.

[0038] Furthermore, the candidate modulation and coding schemes that satisfy the signal-to-noise ratio threshold constraint mentioned in step 7... Determine as follows: in, This represents the candidate modulation and coding schemes that satisfy the signal-to-noise ratio threshold constraint. This represents the candidate modulation and coding schemes in the modulation and coding threshold table. Indicates candidate modulation and coding schemes The corresponding spectral efficiency, Indicates candidate modulation and coding schemes The corresponding signal-to-noise ratio threshold, This represents the conservative effective signal-to-noise ratio corresponding to the effective prediction time domain.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. This invention unifies the low-order Zernike aberration coefficients, link turbulence and performance scalars, and meteorological covariates into a multimodal time token sequence, avoiding extrapolation of future link quality based solely on a single historical signal-to-noise ratio. This enables the prediction model to simultaneously utilize the correlation information between wavefront distortion, atmospheric turbulence, and link performance, thereby improving the ability to characterize millisecond-level link fluctuations.

[0041] 2. This invention uses Maréchal coupling constraints, phase structure function constraints, and cross-temporal Kolmogorov evolution constraints to jointly train a physically consistent multi-temporal prediction model, ensuring that the predicted Strehl ratio, effective signal-to-noise ratio, and logarithmic bit error rate are consistent in the physical relationship of the optical link, thereby reducing the risk of physically inconsistent prediction results generated by a purely data-driven model.

[0042] 3. This invention forms a calibration sample set based on independent calibration data and generates an effective signal-to-noise ratio prediction interval. The lower bound of the effective signal-to-noise ratio prediction interval is taken as the conservative effective signal-to-noise ratio, so that the fixed empirical safety margin is replaced by the lower bound criterion jointly determined by the calibration samples, the effective prediction time domain, and the prediction residual distribution, thereby improving the adaptability of modulation and coding threshold decision under different turbulent conditions.

[0043] 4. Based on the modulation and coding switching delay, this invention determines the effective prediction time domain corresponding to the future effective time from multiple prediction time domains, so that the modulation and coding scheme is determined based on the link state at the actual effective time, rather than the current link state, thereby reducing the risk of modulation and coding mismatch caused by the switching delay of the receiver observation, feedback transmission, and transmitter calculation and configuration.

[0044] 5. This invention connects multi-time-domain link performance prediction, effective signal-to-noise ratio prediction interval calibration, and modulation coding threshold decision into a continuous control process. Without adding wavefront sensors, it is deployed as a software control layer in the satellite-to-ground laser communication terminal, improving link availability and providing an adjustable basis for the trade-off between transmission efficiency and link reliability. Attached Figure Description

[0045] Figure 1 The overall flowchart of the satellite-to-ground laser communication prediction interval driven adaptive modulation control method provided in the embodiments of the present invention is shown.

[0046] Figure 2 This is a schematic diagram illustrating the construction of a multimodal time token sequence provided in an embodiment of the present invention.

[0047] Figure 3 This is a schematic diagram of the structure of a physically consistent multi-temporal prediction model provided in an embodiment of the present invention.

[0048] Figure 4 This is a schematic diagram illustrating conformal prediction interval calibration and conservative effective signal-to-noise ratio generation provided in an embodiment of the present invention.

[0049] Figure 5 A bar chart comparing the availability of links under different control strategies provided in embodiments of the present invention. Detailed Implementation

[0050] 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.

[0051] This embodiment focuses on a laser downlink from a low-Earth orbit satellite to a ground optical station, with a working wavelength of 1550nm and a receiving aperture diameter of... The link sampling frequency is 1kHz, the historical time window length is 200ms, and the prediction time domain set is 0.4m. for The modulation and coding threshold table uses the signal-to-noise ratio thresholds and spectral efficiencies corresponding to 28 DVB-S2X modulation and coding schemes. Link state data is obtained from atmospheric phase screens, Hufnagel-Valley 5 / 7 turbulence profiles, von Karman phase screen evolution, and link budget calculations.

[0052] like Figure 1 As shown, this embodiment is executed according to steps 1 to 8.

[0053] Step 1: Obtain historical link status data, independent calibration data, modulation and coding switching delay, predicted time domain set, and modulation and coding threshold table of the satellite-to-ground laser communication link, and construct the historical time window link status sequence and future effective time.

[0054] In this step, the current decision moment is denoted as... The length of the historical time window is denoted as The sampling interval is denoted as Number of time steps within a historical time window Represented as: Historical time window link state sequence Represented as: in, This indicates the number of time steps within a historical time window. Indicates the length of the historical time window. Indicates the sampling interval. Indicates the current decision-making moment Available historical time window link state sequences, Indicates the first The link state vector corresponding to each time step.

[0055] In this embodiment, It includes 21 features, specifically 7 low-order Zernike aberration coefficients, 7 link turbulence and performance scalars, and 7 meteorological covariates. The 7 low-order Zernike aberration coefficients include correlation coefficients for tilt, defocus, astigmatism, and coma; the 7 link turbulence and performance scalars include Gamma-Gamma distribution parameters. , Fried parameters Rytov variance Strehl Effective signal-to-noise ratio And the logarithmic value of the bit error rate, wherein the receiver aperture statistic in this embodiment is determined by the Strehl ratio. The wavefront quality within the receiving aperture is characterized; the seven meteorological covariates include air temperature, air pressure, relative humidity, east-west wind speed, north-south wind speed, solar shortwave radiation, and boundary layer height.

[0056] Training samples for model parameter training in step 4 are constructed based on the historical link state data, and a validation set is partitioned from the training samples. The independent calibration data does not participate in the parameter training of the physically consistent multi-temporal prediction model. The independent calibration data includes multiple historical time window link state sequences and their corresponding effective signal-to-noise ratio (SNR) true values ​​in the prediction time domain, which are used in subsequent step 6 to generate effective SNR predicted values ​​on the trained physically consistent multi-temporal prediction model and form a calibration sample set for calculating the empirical quantile of the prediction residual.

[0057] The modulation and coding threshold table includes signal-to-noise ratio (SNR) thresholds and spectral efficiencies for 28 DVB-S2X modulation and coding schemes. Each candidate modulation and coding scheme corresponds to an SNR threshold and a spectral efficiency. The SNR threshold is used to determine whether the conservative effective SNR obtained in step 6 meets the link reliability requirements of the candidate modulation and coding scheme, and the spectral efficiency is used to select the target modulation and coding scheme with the highest transmission efficiency from among multiple candidate modulation and coding schemes that meet the SNR threshold constraints.

[0058] Modulation coding switching delay Represented as: in, This indicates the delay between the receiver's observation and feedback. This indicates the adaptive optics processing delay. This indicates the modulation and coding decision delay at the transmitting end. Indicates the delay in switching modulation and coding scheme configurations. Future effective date. Represented as: in, This indicates the moment when the determined target modulation and coding scheme actually begins to take effect on the satellite-to-ground laser communication link.

[0059] In this embodiment, the prediction time domain set is determined according to the link sampling interval and modulation and coding switching delay, specifically taking values ​​of {1ms, 10ms, 50ms, 100ms}. Here, 1ms corresponds to the link sampling interval, 10ms is used to cover short-term feedback and configuration update delays, and 50ms and 100ms are used to cover the impact of atmospheric turbulence evolution under longer modulation and coding activation delays.

[0060] Step 2: Based on the historical time window link state sequence constructed in Step 1, a multimodal time token sequence is formed.

[0061] like Figure 2 As shown, this step will generate the historical time window link status sequence. Based on their physical meaning, they are divided into low-order aberration feature groups. Link turbulence and performance scalar characteristics and meteorological covariate characteristic group For any time step Perform linear mappings respectively: in, , , Representing time steps Low-order aberration features, link turbulence and performance scalar features, and meteorological covariate features. , , This represents the corresponding linear mapping matrix. , , This indicates the corresponding bias. , , This represents the corresponding embedding vector.

[0062] Subsequently, the three embedding vectors are combined with modal bias and temporal position encoding to obtain the time step. Corresponding multimodal time token : in, , , This represents the modal bias corresponding to the three feature groups. This represents the time position encoding. A multimodal time token sequence is formed by the multimodal time tokens of all time steps. .

[0063] Step 3: Construct a physically consistent multi-temporal prediction model based on the multimodal time token sequence formed in Step 2.

[0064] like Figure 3 As shown, the physically consistent multi-temporal prediction model in this embodiment includes a multimodal grouping mapping layer, a location encoding layer, a temporal attention encoding layer, and a multi-task prediction head. The temporal attention encoding layer employs a 4-layer Transformer encoder, with each layer including 4 attention heads, a hidden dimension of 128, and a feedforward network dimension of 256.

[0065] Among them, the multimodal group mapping layer is used to receive the multimodal time token sequence formed in step 2, the position encoding layer is used to maintain the order information of different time steps within the historical time window, the temporal attention encoding layer is used to encode the link state evolution relationship within the historical time window, and the multi-task prediction head is used to form the link performance prediction output channel corresponding to each prediction time domain in the prediction time domain set determined in step 1.

[0066] Step 4: Introduce a joint training objective to train the physically consistent multi-temporal prediction model constructed in Step 3.

[0067] This step incorporates both data prediction error and optical link physical consistency constraints into the model training. The joint training objective is: in, It consists of the mean square error of the Strehl ratio, the effective signal-to-noise ratio, and the logarithm of the bit error rate; Used to constrain the Strehl ratio prediction and the effective signal-to-noise ratio prediction to meet the calibrated link budget relationship; The Strehl ratio prediction used to constrain the shortest prediction time domain satisfies the phase structure function relationship with the Fried parameter. This is used to constrain the turbulence evolution across multiple prediction time domains to conform to the Kolmogorov power-law smoothing relationship.

[0068] Maréchal coupling constraints are constructed as follows: in, This indicates the number of prediction time domains in the prediction time domain set. Represents a prediction time domain in the set of prediction time domains. Represents the prediction time domain The corresponding effective signal-to-noise ratio prediction value, Represents the prediction time domain The corresponding Strehl ratio predicted value, and This represents the link budget constant obtained based on the training samples. In this embodiment, , .

[0069] The phase structure function constraints are constructed as follows: in, Represents the shortest prediction time domain. This represents the Strehl ratio prediction value corresponding to the shortest prediction time domain. Indicates the diameter of the receiving aperture. Indicates the Fried parameter. and This represents the phase structure function constant obtained based on the training samples calibration.

[0070] The cross-temporal Kolmogorov evolutionary constraints are constructed as follows: in, Represents the prediction time domain The corresponding phase variance characterization quantity, This represents the Kolmogorov power-law scale after time-domain normalization for prediction. Represents the maximum prediction time domain. and This represents the linear parameters obtained by fitting the same training sample across multiple prediction time domains.

[0071] In this embodiment, the weights of the three physical consistency constraints are all set to 0.1. The model is trained using the AdamW optimizer with a learning rate of... The batch size is 64, the maximum number of training rounds is 50, and an early stopping strategy is used to select the model parameters with the minimum loss on the validation set obtained by partitioning the training samples.

[0072] Step 5: Based on the physically consistent multi-temporal prediction model trained in Step 4, reason about the multimodal time token sequence formed in Step 2, generate the link performance prediction results corresponding to each prediction time domain in the prediction time domain set determined in Step 1, and determine the link performance prediction results corresponding to the effective prediction time domains.

[0073] In this step, the current historical time window link state sequence is transformed into a multimodal time token sequence in step 2. Input the physically consistent multi-temporal prediction model trained in step 4, and obtain the temporal representation of the historical time window through the temporal attention encoding layer. : in, This represents the temporal attention encoding operation. Represents the time sequence of a historical time window; [take / extract] The encoding vector corresponding to the last time step is used as the representation of the current link state. And through the multi-task prediction head corresponding to each prediction time domain in the prediction time domain set determined in step 1. The actual generated prediction time domain set Strehl ratio prediction, effective signal-to-noise ratio prediction, and logarithmic bit error rate prediction on milliseconds: in, Represents the prediction time domain The corresponding Strehl ratio predicted value, Represents the prediction time domain The corresponding effective signal-to-noise ratio prediction value, Represents the prediction time domain The corresponding logarithmic prediction of the bit error rate, Represents the prediction time domain The corresponding prediction head.

[0074] Subsequently, based on the modulation and coding switching delay obtained in step 1 Determine the effective prediction time domain: This yields the link performance prediction results used for modulation and coding control: in, This represents the Strehl ratio prediction value corresponding to the effective prediction time domain. This represents the effective signal-to-noise ratio (SNR) prediction value corresponding to the effective prediction time domain. This represents the logarithmic prediction value of the bit error rate corresponding to the effective prediction time domain.

[0075] Step 6: Based on independent calibration data, form a calibration sample set and generate the effective signal-to-noise ratio prediction range and conservative effective signal-to-noise ratio.

[0076] like Figure 4 As shown, this step first inputs the independent calibration data obtained in step 1 into the physically consistent multi-time-domain prediction model trained in step 4 to obtain the effective signal-to-noise ratio (SNR) prediction value of each calibration sample in the corresponding prediction time domain, and forms a calibration sample set together with the actual effective SNR values ​​in the independent calibration data. Then, the effective SNR prediction residual is calculated for each prediction time domain on the calibration sample set: in, Indicates calibration sample In the prediction time domain Effective signal-to-noise ratio prediction residuals Indicates calibration sample In the prediction time domain The effective signal-to-noise ratio is the true value. Indicates calibration sample In the prediction time domain The effective signal-to-noise ratio prediction value.

[0077] Based on the preset coverage rate Calculate empirical quantiles: in, Represents the prediction time domain Empirical quantiles of predicted residuals This represents the empirical quantile operation. Indicates the number of calibration samples. This indicates the preset coverage loss rate. In this embodiment, .

[0078] Effective signal-to-noise ratio prediction range Represented as: in, This represents the effective signal-to-noise ratio prediction interval corresponding to the effective prediction time domain, with the lower bound being... The upper bound of the interval is .

[0079] The conservative effective signal-to-noise ratio is expressed as: in, This represents the conservative effective signal-to-noise ratio corresponding to the effective prediction time domain.

[0080] In this embodiment, the number of calibration samples is 288, and the coverage loss rate is 0.05. For the prediction time domains of 1ms, 10ms, 50ms, and 100ms, the signal-to-noise ratio residual quantiles are 1.735dB, 4.151dB, 7.134dB, and 6.249dB, respectively.

[0081] Step 7: Determine the target modulation and coding scheme based on the conservative effective signal-to-noise ratio obtained in Step 6 and the modulation and coding threshold table obtained in Step 1.

[0082] This step will conservatively maintain the effective signal-to-noise ratio. The modulation and coding threshold table is compared with that of other modulation and coding schemes. The threshold table includes multiple candidate modulation and coding schemes, each corresponding to a signal-to-noise ratio threshold and a spectral efficiency. The target modulation and coding scheme is determined as follows: in, Indicates the target modulation and coding scheme. Indicates candidate modulation and coding schemes Spectral efficiency, Indicates candidate modulation and coding schemes The signal-to-noise ratio threshold.

[0083] When the conservative effective signal-to-noise ratio is lower than the lowest threshold in the modulation and coding threshold table, the lowest-order modulation and coding scheme is determined as the target modulation and coding scheme to keep the output result complete.

[0084] Step 8: Output the adaptive modulation control result based on the future effective time determined in Step 1.

[0085] In this step, the target modulation and coding scheme determined in step 7 is applied. Effective date in the future Corresponding frame number and current decision time Correlation, forming an adaptive modulation control result: in, This represents the adaptive modulation control result output at the current decision moment. Indicates the current decision-making moment. Indicates the actual effective time of the target modulation and coding scheme. Indicates the target modulation and coding scheme. This represents the conservative effective signal-to-noise ratio used to determine the target modulation and coding scheme.

[0086] like Figure 5 As shown, in the closed-loop adaptive modulation control test of this embodiment, link availability is used as the evaluation index to compare the reactive strategy, the point prediction strategy, the method of this embodiment, and the ideal upper bound strategy. The reactive strategy selects the modulation and coding scheme based solely on the current or historical signal-to-noise ratio (SNR); the point prediction strategy selects the SNR based on the point prediction value of the effective SNR output by the physically consistent multi-temporal prediction model; the method of this embodiment selects the SNR based on the conservative effective SNR obtained in step 6; and the ideal upper bound strategy represents the achievable comparison result when the actual effective SNR at the future effective time is known.

[0087] from Figure 5It can be seen that when the modulation and coding scheme switching delay is 10ms, the link availability of the reactive strategy is 65.3%, the link availability of the point prediction strategy is 70.1%, and the link availability of the method in this embodiment is 94.1%, which is close to the ideal upper bound strategy of 95.5%. When the modulation and coding scheme switching delay is 100ms, the link availability of the reactive strategy is 62.8%, the link availability of the point prediction strategy is 73.6%, and the link availability of the method in this embodiment is 95.8%, which is close to the ideal upper bound strategy of 97.2%.

[0088] The above results indicate that, under conditions of modulation and coding scheme switching delay and rapid changes in atmospheric turbulence, relying solely on historical signal-to-noise ratio (SNR) or predicted effective SNR values ​​for threshold determination can still easily lead to a mismatch between the modulation and coding scheme and the actual link state at the future effective time. This embodiment uses the lower bound of the effective SNR prediction interval after conformal calibration for threshold determination, ensuring that the target modulation and coding scheme is determined according to a more conservative approach that corresponds to the link quality at the future effective time. This reduces the probability that higher-order modulation and coding schemes will not meet the SNR threshold at the future effective time, thereby improving the availability of the space-to-ground laser communication link.

Claims

1. A physical consistent multi-time domain prediction modulation coding control method for satellite-ground laser communication, characterized in that, The method includes the following steps: Step 1: Obtain historical link status data, independent calibration data, modulation and coding switching delay, predicted time domain set and modulation and coding threshold table of the satellite-to-ground laser communication link, and construct a historical time window link status sequence based on the historical link status data, and determine the future effective time of the modulation and coding mode based on the modulation and coding switching delay; Step 2: Based on the historical time window link state sequence constructed in Step 1, multimodal grouping feature mapping is introduced to improve the problem that the single link scalar input is insufficient to express the turbulent state, forming a multimodal time token sequence; Step 3: Based on the multimodal time token sequence formed in Step 2, construct a physically consistent multitemporal prediction model. The physically consistent multitemporal prediction model includes a multimodal grouping mapping layer, a position encoding layer, a temporal attention encoding layer, and a multi-task prediction head. Step 4: Introduce a joint training objective consisting of link performance prediction loss, Maréchal coupling constraint, phase structure function constraint, and cross-temporal Kolmogorov evolution constraint to train the physically consistent multi-temporal prediction model constructed in Step 3. Step 5: Based on the physically consistent multi-temporal prediction model trained in Step 4, inference is performed on the multimodal time token sequence formed in Step 2 to generate the Strehl ratio prediction value, effective signal-to-noise ratio prediction value, and logarithmic bit error rate prediction value corresponding to each prediction time domain in the prediction time domain set determined in Step 1; and based on the modulation and coding switching delay determined in Step 1, the effective prediction time domain corresponding to the future effective time is determined from the prediction time domain set, and the Strehl ratio prediction value, effective signal-to-noise ratio prediction value, and logarithmic bit error rate prediction value corresponding to the effective prediction time domain are obtained; Step 6: Based on the independent calibration data obtained in Step 1, form a calibration sample set, and perform conformal prediction interval calibration on the effective signal-to-noise ratio prediction value corresponding to the effective prediction time domain obtained in Step 5 to obtain the effective signal-to-noise ratio prediction interval, and take the lower bound of the effective signal-to-noise ratio prediction interval as the conservative effective signal-to-noise ratio. Step 7: Based on the conservative effective signal-to-noise ratio obtained in Step 6 and the modulation and coding threshold table obtained in Step 1, determine the candidate modulation and coding schemes that meet the signal-to-noise ratio threshold constraints, and determine the target modulation and coding scheme with the highest spectral efficiency from the candidate modulation and coding schemes. Step 8: Based on the future effective time determined in Step 1, associate the target modulation and coding scheme determined in Step 7 with the corresponding frame number and timestamp, and output the adaptive modulation control result for the satellite-to-ground laser communication link.

2. The method according to claim 1, wherein, Step 1 specifically includes the following sub-steps: Step 1.1: Obtain historical link status data of the satellite-to-ground laser communication link. The historical link status data includes low-order Zernike aberration coefficients, receiver aperture statistics, Fried parameters, Rytov variance, Gamma-Gamma distribution parameters, Strehl ratio, effective signal-to-noise ratio, logarithmic bit error rate, and meteorological covariates. The meteorological covariates include air temperature, air pressure, relative humidity, horizontal wind speed component, solar shortwave radiation, and boundary layer height. Based on the historical link status data, construct training samples for model parameter training in Step 4, and divide the training samples into a validation set. Step 1.2: Obtain independent calibration data for the satellite-to-ground laser communication link. The independent calibration data includes the historical time window link state sequence that does not participate in the model parameter training and its corresponding effective signal-to-noise ratio true value in the predicted time domain, which is used to form a calibration sample set in step 6. Step 1.3: Obtain the modulation and coding threshold table, which includes multiple candidate modulation and coding schemes, the signal-to-noise ratio (SNR) threshold for each candidate SNR, and the spectral efficiency for each candidate SNR. The SNR threshold is used in Step 7 to compare with the conservative effective SNR obtained in Step 6 to determine the candidate SNR that meets the SNR threshold constraint. The spectral efficiency is used to determine the target SNR from the candidate SNR that meets the SNR threshold constraint. Step 1.4: Construct a historical time window link state sequence based on the historical link state data: the current decision moment is denoted as... The length of the historical time window is denoted as The sampling interval is denoted as Number of time steps within a historical time window Represented as: Historical time window link state sequence Represented as: in, This indicates the number of time steps within a historical time window. Indicates the length of the historical time window. Indicates the sampling interval. Indicates the current decision-making moment Available historical time window link state sequences, Indicates the first The link state vector corresponding to each time step; Step 1.5: Obtain the modulation and coding switching delay of the satellite-to-ground laser communication link, and determine the future effective time of the modulation and coding scheme based on the modulation and coding switching delay. Delay observed at the receiving end Adaptive optics processing delay Modulation and coding decision delay Modulation coding configuration switching delay The sum is obtained as follows: Future effective time is represented as: in, This represents the total switching delay required from the current decision-making moment to the actual implementation of the modulation and coding scheme. This indicates the moment when the target modulation and coding scheme actually begins to take effect on the satellite-to-ground laser communication link; Step 1.6: Determine the prediction time domain set, wherein the shortest prediction time domain in the prediction time domain set is equal to the sampling interval. Correspondingly, the longest prediction time domain is not less than the modulation and coding switching delay. This enables subsequent step 5 to select the time of future effectiveness from the predicted time domain set. The corresponding effective prediction time domain.

3. The method for physical consistency multi-temporal predictive modulation and coding control in satellite-to-ground laser communication according to claim 2, characterized in that, The specific method of step 2 is to use the historical time window link state sequence constructed in step 1. Based on their physical meaning, they are divided into low-order aberration feature groups. Link turbulence and performance scalar characteristics and meteorological covariate characteristic group ; for time step Low-order aberration features Link turbulence and performance scalar characteristics Characteristics of meteorological covariates Perform linear mappings separately: in, , and These represent the linear mapping matrices corresponding to the low-order aberration feature group, the link turbulence and performance scalar feature group, and the meteorological covariate feature group, respectively. , and These represent the corresponding biases. , and These represent the embedding vectors after mapping the low-order aberration feature group, the link turbulence and performance scalar feature group, and the meteorological covariate feature group, respectively. Will , and Combined with modal bias and temporal position encoding, the time step is obtained. Corresponding multimodal time token : in, , and These represent the modal biases corresponding to the low-order aberration feature group, the link turbulence and performance scalar feature group, and the meteorological covariate feature group, respectively. Indicates time step Corresponding time location code; The sequence of multi-modal time tokens is composed of multi-modal time tokens of all time steps .

4. The method according to claim 1 or 2 or 3, characterized in that, In step 3, the physically consistent multi-temporal prediction model includes a multimodal packet mapping layer, a location encoding layer, a temporal attention encoding layer, and a multi-task prediction head. The multimodal packet mapping layer is used to receive the multimodal time token sequence formed in step 2. The location encoding layer is used to maintain the order information of different time steps within the historical time window. The temporal attention encoding layer is used to encode the link state evolution relationship within the historical time window. The multi-task prediction head is set according to the prediction time domain set determined in step 1 and is used to form the link performance prediction output channel corresponding to each prediction time domain.

5. The method according to claim 1 or 2 or 3, characterized in that, In step 4, the joint training objective is expressed as: in, Indicates the joint training objective. The link performance prediction loss is represented by Strehl ratio, effective signal-to-noise ratio, and logarithm of bit error rate. This represents the cross-temporal Kolmogorov evolution constraint loss. This represents the phase structure function constraint loss. This represents the Maréchal coupling constraint loss. , and This represents the weight of the corresponding constraint loss.

6. The method for physical consistency multi-temporal predictive modulation and coding control in satellite-to-ground laser communication according to claim 5, characterized in that, In step 4, the Maréchal coupling constraint loss The formula used to constrain the coupling relationship between the effective signal-to-noise ratio prediction and the Strehl ratio prediction is as follows: in, This indicates the number of prediction time domains in the prediction time domain set. Represents a prediction time domain in the set of prediction time domains. Represents the prediction time domain The corresponding effective signal-to-noise ratio prediction value, Represents the prediction time domain The corresponding Strehl ratio predicted value, and This represents the link budget constant obtained based on the training samples. The phase structure function constraint loss For constraining the phase structure function relationship between the short-term Strehl ratio prediction value and the Fried parameter, the calculation formula is: wherein, denotes the shortest prediction time domain, denotes the Strehl ratio prediction value corresponding to the shortest prediction time domain, denotes the receiving aperture diameter, denotes the Fried parameter, and denotes the phase structure function constant calibrated based on the training samples; The cross-temporal Kolmogorov evolution constraint loss is used to constrain the smoothness of turbulent evolution across multiple prediction time domains. The following variables are first constructed: and also with satisfy a linear fit relationship: The cross-time-domain Kolmogorov evolution constraint loss The calculation formula is: in, Represents the prediction time domain The corresponding phase variance characterization quantity, This represents the Kolmogorov power-law scale after time-domain normalization for prediction. Represents the maximum prediction time domain. and This represents the linear parameters obtained by fitting the same training sample across multiple prediction time domains.

7. The method according to claim 5, wherein, The specific method for step 5 is as follows: Step 5.1, obtain the multi-modal time token sequence from step 2 Input the trained physically consistent multi-time domain prediction model of step 4, and obtain the time sequence representation of the historical time window from the time sequence attention encoding layer : in, This represents the temporal attention encoding operation. Represents the time sequence of a historical time window; [take / extract] The encoding vector corresponding to the last time step is used as the representation of the current link state. And by predicting the time domain Corresponding multi-task prediction head Generate multiple link performance predictions in the prediction time domain: in, Represents the prediction time domain set, Represents a prediction time domain in the set of prediction time domains. Represents the prediction time domain The corresponding Strehl ratio predicted value, Represents the prediction time domain The corresponding effective signal-to-noise ratio prediction value, Represents the prediction time domain The corresponding logarithmic prediction of the bit error rate, Represents the prediction time domain The corresponding prediction head; Step 5.2, valid prediction time domain Is determined as follows: wherein, denotes the total switching delay required for the modulation coding scheme to take effect from the current decision time instant, denotes the valid prediction time domain for modulation coding selection; Step 5.3, based on the effective prediction time domain From the link performance prediction values ​​generated by the physically consistent multi-temporal prediction model trained in step 4 across multiple prediction temporal domains, determine the link performance prediction results corresponding to the effective prediction temporal domains: wherein, effective prediction time domain corresponding Strehl ratio prediction value, effective prediction time domain corresponding effective signal-to-noise ratio prediction value, effective prediction time domain corresponding log error rate prediction value.

8. The method for physical consistency multi-temporal predictive modulation and coding control in satellite-to-ground laser communication according to claim 7, characterized in that, In step 6, the independent calibration data obtained in step 1 is input into the physically consistent multi-temporal prediction model trained in step 4 to obtain the predicted effective signal-to-noise ratio (SNR) value in the prediction time domain for each calibration sample in the independent calibration data. This predicted value is then combined with the corresponding true effective SNR value in the independent calibration data to form a calibration sample set. The absolute residual between the true effective SNR value and the predicted effective SNR value is calculated in the calibration sample set, and a preset coverage rate is applied. Determine the effective prediction time domain Corresponding empirical quantiles Then, based on the effective signal-to-noise ratio prediction values ​​obtained in step 5, an effective signal-to-noise ratio prediction interval is generated: in, This represents the effective signal-to-noise ratio prediction interval corresponding to the effective prediction time domain. This represents the lower bound of the effective signal-to-noise ratio prediction interval. This represents the upper bound of the effective signal-to-noise ratio prediction interval; The lower bound of the effective signal-to-noise ratio prediction interval is taken as the conservative effective signal-to-noise ratio: in, This represents the conservative effective signal-to-noise ratio corresponding to the effective prediction time domain. This represents the effective signal-to-noise ratio prediction value in the effective prediction time domain obtained in step 5. This represents the empirical quantile of the prediction residual corresponding to the effective prediction time domain in the calibration sample set; Based on the conservative effective signal-to-noise ratio obtained in step 6 and the modulation and coding threshold table obtained in step 1, candidate modulation and coding schemes that meet the signal-to-noise ratio threshold constraints are determined, and the target modulation and coding scheme with the highest spectral efficiency is determined from the candidate modulation and coding schemes.

9. The method for physical consistency multi-temporal predictive modulation and coding control in satellite-to-ground laser communication according to claim 1, characterized in that, Candidate modulation and coding schemes that satisfy the signal-to-noise ratio threshold constraint as described in step 7 Determine as follows: in, This represents the candidate modulation and coding schemes that satisfy the signal-to-noise ratio threshold constraint. This represents the candidate modulation and coding schemes in the modulation and coding threshold table. Indicates candidate modulation and coding schemes The corresponding spectral efficiency, Indicates candidate modulation and coding schemes The corresponding signal-to-noise ratio threshold, This represents the conservative effective signal-to-noise ratio corresponding to the effective prediction time domain.

Images