Intersatellite laser communication and signal scheduling method based on adaptive polarization modulation

Abstract

The present invention provides an inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation, which relates to the field of laser communication, including: obtaining the polarization state information of the transmitting and receiving light beams to calculate the mismatch degree and generate compensation parameters; using a deep learning neural network to build a polarization state prediction model for pre-correction; using a polarization modulator for real-time modulation and establishing a link quality assessment model; using coherent detection-based adaptive optical carrier recovery technology to achieve carrier synchronization and channel compensation; real-time monitoring of the link status and updating of the training data set.

Description

Technical Field

[0001] The present invention relates to the field of laser communication technology, and in particular to an inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation. Background Art

[0002] Intersatellite laser communication is a key technology for realizing space information networks. It offers advantages such as high bandwidth, high security, and low power consumption. In intersatellite laser communication systems, polarization matching between the transmitter and receiver has a significant impact on communication quality. The space environment is complex and changeable. Satellite attitude changes and atmospheric disturbances can cause the polarization state of the beam to drift, affecting the stability and reliability of the communication link.

[0003] Currently, polarization state control in intersatellite laser communications primarily relies on fixed compensation or simple feedback adjustment. Traditional polarization state control methods typically rely on real-time measurement and feedback, making them ineffective in responding to rapidly changing space environments. Furthermore, existing polarization state compensation technologies lack the ability to predict future polarization state trends, making it difficult to achieve advance compensation and optimized adjustment.

[0004] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the Invention

[0005] The embodiments of the present invention provide an inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation, which can at least solve some of the problems existing in the prior art.

[0006] A first aspect of an embodiment of the present invention provides an inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation, comprising:

[0007] Obtaining polarization state information of a transmitted light beam of the first intersatellite laser communication terminal and polarization state information of a received light beam of the second intersatellite laser communication terminal, calculating a polarization state mismatch according to the polarization state information of the light beam and the polarization state information of the received light beam, and generating an initial polarization state compensation parameter based on the polarization state mismatch;

[0008] Constructing a polarization state prediction model based on a deep learning neural network, wherein the polarization state prediction model inputs include atmospheric disturbance data, satellite attitude data, and historical polarization state data, and outputs a predicted polarization state drift trend, and pre-correcting the polarization state compensation parameters according to the polarization state drift trend;

[0009] Using a polarization modulator to modulate the polarization state of the transmitted light beam in real time, adjusting the modulation parameters of the polarization modulator according to a pre-calibrated polarization state compensation parameter, establishing an inter-satellite laser communication link quality assessment model based on the modulation parameters of the polarization modulator, and calculating the link quality parameters;

[0010] Based on the link quality parameters, an adaptive optical carrier recovery technology based on coherent detection uses an optical frequency comb as the local oscillator light source, achieves carrier synchronization through a digital phase-locked loop, and uses an adaptive equalization algorithm to compensate for channel dispersion and nonlinear effects, and dynamically adjusts equalizer parameters based on channel state information;

[0011] The demodulated communication signal is transmitted through the intersatellite laser communication link, and the link state parameters are monitored in real time. The training data set of the deep learning neural network is updated according to the link state parameters to achieve online optimization of the polarization state prediction model.

[0012] In an optional embodiment,

[0013] Calculating a polarization state mismatch degree according to the polarization state information of the light beam and the polarization state information of the received light beam, and generating an initial polarization state compensation parameter based on the polarization state mismatch degree includes:

[0014] Obtaining polarization state information of the transmitted light beam and the received light beam, wherein the polarization state information includes the total light intensity, the difference between the horizontal and vertical polarization components, the difference between the positive and negative forty-five degree polarization components, and the difference between the right-handed and left-handed circular polarization components;

[0015] Constructing a fourth-order Miller matrix, the fourth-order Miller matrix consisting of a phase delay matrix, a rotation matrix, and a depolarization matrix, performing eigenvalue decomposition on the fourth-order Miller matrix to obtain contribution components of each matrix, and calculating a polarization state mismatch based on the contribution components, where the polarization state mismatch is determined by a ratio of a square root of each contribution component to a sum of the square roots of all contribution components;

[0016] Based on the polarization state mismatch, a compensation parameter optimization objective function is constructed, where the compensation parameters include a phase compensation amount and a rotation angle compensation amount. The gradient iteration method is used to solve the optimization objective function to obtain an initial polarization state compensation parameter. According to the initial polarization state compensation parameter, the polarization state of the emitted light beam is compensated and modulated by a polarization modulator.

[0017] In an optional embodiment,

[0018] A polarization state prediction model is constructed based on a deep learning neural network, wherein the polarization state prediction model input includes atmospheric disturbance data, satellite attitude data, and historical polarization state data, and output includes a predicted polarization state drift trend. Pre-correcting the polarization state compensation parameter according to the polarization state drift trend includes:

[0019] Collect atmospheric temperature distribution, atmospheric pressure distribution, humidity distribution and wind speed distribution to form atmospheric disturbance data, collect roll angle, pitch angle, yaw angle and angular velocity vector to form satellite attitude data, obtain Stokes parameter vectors at multiple sampling moments to form a historical polarization state sequence, and the sampling interval of the historical polarization state sequence is a preset time interval;

[0020] Inputting the atmospheric disturbance data, the satellite attitude data, the historical polarization state sequence, and the initial polarization state compensation parameter into a recurrent neural network with a hybrid attention mechanism, the recurrent neural network calculating a hidden layer state at a current moment based on a long short-term memory unit, generating attention weights for the atmospheric disturbance data and the satellite attitude data according to the hidden layer state, performing weighted fusion of the attention weights with the atmospheric disturbance data and the satellite attitude data to obtain a first eigenvector and a second eigenvector, and performing feature fusion of the first eigenvector and the second eigenvector with the hidden layer state to obtain feature fusion data;

[0021] The feature fusion data is input into a pre-trained neural network mapping function to predict the Stokes parameters after the next preset time interval, the Stokes parameters are subtracted from the Stokes parameters at the latest moment in the historical polarization state sequence to obtain a polarization state difference, the polarization state difference is divided by the preset time interval to obtain a polarization state drift velocity vector, a pre-compensation objective function including a second-order regularization term of a pre-compensation parameter is constructed according to the polarization state drift velocity vector, the gradient descent method is used in combination with a deep reinforcement learning framework to iteratively optimize the pre-compensation objective function to obtain pre-compensation parameters, and the pre-compensation parameters are used to pre-compensate the Stokes parameters at the latest moment in the historical polarization state sequence.

[0022] In an optional embodiment,

[0023] A pre-compensation objective function including a second-order regularization term of a pre-compensation parameter is constructed according to the polarization state drift velocity vector. The pre-compensation objective function is iteratively optimized using a gradient descent method in combination with a deep reinforcement learning framework to obtain pre-compensation parameters including:

[0024] Constructing a system state vector including the current polarization state, the polarization state drift velocity vector, the compensation error, and the compensation parameter, and constructing the rotation angle parameter adjustment amount and the phase parameter adjustment amount as an action vector;

[0025] Calculating the square term of the difference between the product of the compensation matrix and the predicted polarization state and the target polarization state based on the system state vector, and constructing an immediate reward function in combination with the square terms of the rotation angle parameter adjustment amount and the phase parameter adjustment amount, and accumulating the immediate rewards within the future predicted steps using a discount factor to obtain a pre-compensation objective function;

[0026] Based on the pre-compensation objective function, a deep neural network is constructed as a value network and a policy network, the value network inputs the system state vector and the action vector and outputs a value evaluation, the policy network inputs the system state vector and outputs an action strategy, and the state transition samples are stored in an experience replay buffer;

[0027] A temporal difference objective function is constructed based on the instant reward function and the value network, a policy gradient objective function is constructed based on the value network and the policy network, and an entropy regularization objective function is constructed by introducing policy distribution entropy;

[0028] The temporal difference objective function is optimized by the gradient descent method to update the network parameters of the value network, the weighted sum of the policy gradient objective function and the entropy regularization objective function is optimized to update the network parameters of the policy network, and the target network parameters are updated by a soft update method;

[0029] Gaussian noise is injected into the action strategy output by the strategy network to obtain an actual execution action, the variance of the Gaussian noise is calculated based on the compensation error, and the compensation parameter is iteratively optimized using the actual execution action as the update amount of the compensation parameter to obtain a pre-compensation parameter.

[0030] In an optional embodiment,

[0031] A polarization modulator is used to perform real-time polarization state modulation on the transmitted light beam, a modulation parameter of the polarization modulator is adjusted according to a pre-calibrated polarization state compensation parameter, and an intersatellite laser communication link quality evaluation model is established based on the modulation parameter of the polarization modulator. Calculating the link quality parameter includes:

[0032] Using a polarization modulator to modulate the polarization state of the transmitted light beam in real time, and adjusting the modulation parameters of the polarization modulator according to the pre-calibrated polarization state compensation parameters, wherein the modulation parameters include the polarization state rotation angle and the polarization state ellipticity;

[0033] Establishing an intersatellite laser communication link quality assessment model based on the modulation parameters of the polarization modulator, wherein the link quality assessment model includes an influence factor of atmospheric disturbance on the polarization state, a receiving end polarization state analysis factor, and a system polarization extinction ratio factor;

[0034] A link quality parameter is calculated according to the link quality evaluation model, and the polarization state compensation parameter is updated in real time according to the link quality parameter.

[0035] In an optional embodiment,

[0036] Based on the link quality parameters, adaptive optical carrier recovery technology based on coherent detection uses an optical frequency comb as the local oscillator light source, achieves carrier synchronization through a digital phase-locked loop, and uses an adaptive equalization algorithm to compensate for channel dispersion and nonlinear effects. The equalizer parameters are dynamically adjusted according to the channel state information, including:

[0037] Acquiring link quality parameters, using an optical frequency comb as a local oscillator light source, generating frequency comb spectral lines based on a preset center frequency and comb tooth spacing, wherein the frequency comb spectral lines are determined by the amplitude, frequency, and initial phase of the frequency components, monitoring temperature drift and driving voltage fluctuation, calculating frequency drift based on the product of a temperature sensitivity coefficient and the temperature drift, and the product of a voltage sensitivity coefficient and the driving voltage fluctuation, and performing real-time compensation for the frequency of the optical frequency comb;

[0038] Performing coherent detection on a received optical signal and the optical frequency comb to obtain a beat signal, performing complex conjugate multiplication on the beat signal and a local reference signal to obtain a phase error, inputting the phase error into a digital phase-locked loop, filtering the phase error using a proportional-integral controller, and calculating a control voltage based on a weighted sum of a current phase error multiplied by a proportional coefficient and a cumulative sum of historical phase errors multiplied by an integral coefficient;

[0039] Calculating an output phase of a digitally controlled oscillator according to the control voltage, the output phase being obtained by summing the output phase at a previous moment, the product of the center frequency and the sampling period, and the product of the control voltage and the gain coefficient, and using the output phase as a phase parameter of the local reference signal to achieve carrier synchronization;

[0040] Establishing a channel transmission model, the channel transmission model including a dispersion compensation model and a nonlinear effect compensation model, wherein the dispersion compensation model calculates a dispersion compensation coefficient based on the wavelength of the optical signal, the transmission distance, and the speed of light, and the nonlinear effect compensation model calculates a nonlinear compensation coefficient based on the input signal, the nonlinear coefficient, the signal power, and the effective length, performing feature extraction and optimizing loss function design through deep learning technology, and combining an adaptive equalization algorithm to compensate for the channel dispersion and the nonlinear effect, and multiplying multiple delayed signals of the input signal by the equalizer coefficients and summing the results to obtain an output signal of the adaptive equalization algorithm;

[0041] An adaptive step factor is calculated according to the link quality parameter, and the equalizer coefficient is updated by taking the product of the adaptive step factor and the conjugate signal of the error signal and the input signal delay signal as the update amount. The size of the adaptive step factor is dynamically adjusted based on the link quality parameter to achieve adaptive optimization of the equalizer coefficient and output the demodulated communication signal.

[0042] In an optional embodiment,

[0043] The method uses deep learning technology to extract features and optimize loss function design, combines an adaptive equalization algorithm to compensate for the channel dispersion and the nonlinear effect, and multiplies multiple delayed signals of the input signal by the equalizer coefficients and sums them to obtain the output signal of the adaptive equalization algorithm, including:

[0044] Using a sliding time window of a preset length, the input signal is subjected to time domain feature extraction to obtain signal amplitude, signal phase change, and signal instantaneous power; performing a fast Fourier transform on the signal data within the sliding time window to obtain a frequency domain signal; extracting the spectrum amplitude, channel dispersion characteristics, and nonlinear effect characteristics of the frequency domain signal; and forming a feature vector from the signal amplitude, signal phase change, signal instantaneous power, spectrum amplitude, channel dispersion characteristics, and nonlinear effect characteristics;

[0045] Inputting the feature vector into a first neural network for feature dimensionality reduction to obtain a first feature, the first neural network including a weight matrix and a bias vector, inputting the first feature into a long short-term memory network to obtain a latent state feature, the long short-term memory network being used to model temporal features, and inputting the latent state feature into a fully connected neural network to obtain an equalizer coefficient;

[0046] Compensating for the channel dispersion and the nonlinear effect, multiplying the multiple delayed signals of the input signal by the equalizer coefficients and summing the results to obtain an output signal of the adaptive equalization algorithm, subtracting the output signal of the adaptive equalization algorithm from the desired output signal to obtain an error signal, and updating the equalizer coefficients according to the error signal, an adaptive step size factor, a conjugate signal of the delayed signal of the input signal, and a fusion factor;

[0047] Constructing a comprehensive loss function including a mean square error loss term, a bit error rate loss term, and a peak-to-average ratio loss term, and optimizing the network parameters of the first neural network, the network parameters of the long short-term memory network, and the network parameters of the fully connected neural network according to the comprehensive loss function using a gradient descent method;

[0048] A channel state index is calculated according to the signal-to-noise ratio, the error vector magnitude, and the bit error rate, the adaptive step factor is dynamically adjusted according to the channel state index and the error signal, and the fusion factor is updated according to the channel state index and the comprehensive loss function.

[0049] In an optional embodiment,

[0050] Transmitting the demodulated communication signal through the intersatellite laser communication link, monitoring the link state parameters in real time, and updating the training data set of the deep learning neural network according to the link state parameters to achieve online optimization of the polarization state prediction model include:

[0051] Transmitting the demodulated communication signal through an intersatellite laser communication link, and using a polarization state analyzer to monitor link state parameters of the intersatellite laser communication link in real time, wherein the link state parameters include signal-to-noise ratio, bit error rate, and polarization extinction ratio;

[0052] Combining the link state parameters and corresponding polarization state parameters into training sample pairs, and updating a training data set of a deep learning neural network based on the training sample pairs, wherein the deep learning neural network is used to establish a polarization state prediction model;

[0053] The deep learning neural network is trained using the training data set, the network learning rate is adjusted according to the signal-to-noise ratio in the link state parameter, the weight coefficient of the network optimization objective function is adjusted according to the bit error rate in the link state parameter, and the network structure parameters are dynamically adjusted according to the polarization extinction ratio in the link state parameter to achieve online optimization of the polarization state prediction model.

[0054] In the present invention, a method combining adaptive polarization modulation and deep learning prediction can effectively compensate for polarization state mismatch in intersatellite laser communication, significantly improve the stability and reliability of the communication link, and reduce the probability of communication interruption. An optical frequency comb is used as the local oscillator light source, combined with a digital phase-locked loop and an adaptive equalization algorithm to achieve high-precision carrier synchronization and channel compensation, effectively overcome the influence of atmospheric disturbances and nonlinear effects, and improve the anti-interference ability and signal quality of the communication system. By real-time monitoring of the link status and updating the training data of the deep learning model, dynamic optimization and adaptive adjustment of the polarization state prediction model are achieved, so that the system has strong environmental adaptability and self-learning ability, can continuously improve communication performance, and extend the effective communication time. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Schematic diagram of the flow of an inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation according to an embodiment of the present invention;

[0056] Figure 2 This is a compensation error convergence curve diagram of the inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation according to an embodiment of the present invention;

[0057] Figure 3 This is a comparison chart of the system convergence performance of the inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0059] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0060] Figure 1 FIG. 1 is a flow chart of an intersatellite laser communication and signal scheduling method based on adaptive polarization modulation according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0061] Obtaining polarization state information of a transmitted light beam of the first intersatellite laser communication terminal and polarization state information of a received light beam of the second intersatellite laser communication terminal, calculating a polarization state mismatch according to the polarization state information of the light beam and the polarization state information of the received light beam, and generating an initial polarization state compensation parameter based on the polarization state mismatch;

[0062] Constructing a polarization state prediction model based on a deep learning neural network, wherein the polarization state prediction model inputs include atmospheric disturbance data, satellite attitude data, and historical polarization state data, and outputs a predicted polarization state drift trend, and pre-correcting the polarization state compensation parameters according to the polarization state drift trend;

[0063] Using a polarization modulator to modulate the polarization state of the transmitted light beam in real time, adjusting the modulation parameters of the polarization modulator according to a pre-calibrated polarization state compensation parameter, establishing an inter-satellite laser communication link quality assessment model based on the modulation parameters of the polarization modulator, and calculating the link quality parameters;

[0064] Based on the link quality parameters, an adaptive optical carrier recovery technology based on coherent detection uses an optical frequency comb as the local oscillator light source, achieves carrier synchronization through a digital phase-locked loop, and uses an adaptive equalization algorithm to compensate for channel dispersion and nonlinear effects, and dynamically adjusts equalizer parameters based on channel state information;

[0065] The demodulated communication signal is transmitted through the intersatellite laser communication link, and the link state parameters are monitored in real time. The training data set of the deep learning neural network is updated according to the link state parameters to achieve online optimization of the polarization state prediction model.

[0066] In an optional embodiment,

[0067] Calculating a polarization state mismatch degree according to the polarization state information of the light beam and the polarization state information of the received light beam, and generating an initial polarization state compensation parameter based on the polarization state mismatch degree includes:

[0068] Obtaining polarization state information of the transmitted light beam and the received light beam, wherein the polarization state information includes the total light intensity, the difference between the horizontal and vertical polarization components, the difference between the positive and negative forty-five degree polarization components, and the difference between the right-handed and left-handed circular polarization components;

[0069] Constructing a fourth-order Miller matrix, the fourth-order Miller matrix consisting of a phase delay matrix, a rotation matrix, and a depolarization matrix, performing eigenvalue decomposition on the fourth-order Miller matrix to obtain contribution components of each matrix, and calculating a polarization state mismatch based on the contribution components, where the polarization state mismatch is determined by a ratio of a square root of each contribution component to a sum of the square roots of all contribution components;

[0070] Based on the polarization state mismatch, a compensation parameter optimization objective function is constructed, where the compensation parameters include a phase compensation amount and a rotation angle compensation amount. The gradient iteration method is used to solve the optimization objective function to obtain an initial polarization state compensation parameter. According to the initial polarization state compensation parameter, the polarization state of the emitted light beam is compensated and modulated by a polarization modulator.

[0071] To obtain the polarization state information of the transmitted and received beams, a polarization beam splitter is used to separate the incident light into horizontal and vertical polarization components, and a photodetector is used to measure the light intensity of the two components. At the same time, a wave plate and a polarization analyzer are used to obtain the polarization components at + / - 45 degrees, and a quarter-wave plate is used to obtain the right-hand and left-hand circular polarization components. Based on the measurement results, the total light intensity, the difference between the horizontal and vertical polarization components, the difference between the + / - 45 degree polarization components, and the difference between the right-hand and left-hand circular polarization components are obtained, forming the complete polarization state information.

[0072] After obtaining the polarization state information, a fourth-order Miller matrix is ​​constructed. This matrix is ​​obtained by multiplying three basic matrices: the phase delay matrix, the rotation matrix, and the depolarization matrix. The phase delay matrix describes the phase delay characteristics of the light wave during transmission, the rotation matrix characterizes the rotational change of the polarization state, and the depolarization matrix reflects the depolarization effect during channel transmission. Eigenvalue decomposition of the constructed fourth-order Miller matrix can be used to obtain the contribution components of each of the three basic matrices. By calculating the square root of each contribution component and then comparing it to the sum of the square roots of all contribution components, the accurate polarization state mismatch is obtained.

[0073] Based on the calculated polarization state mismatch, a compensation parameter optimization objective function is constructed. This function uses phase compensation and rotation angle compensation as optimization variables, with the goal of minimizing the polarization state mismatch. A gradient iteration method is used to solve this optimization objective function. The specific process is as follows: first, the gradient of the objective function with respect to the two compensation parameters is calculated. Then, the compensation parameters are iteratively updated according to the gradient direction until the algorithm converges to the optimal compensation parameter value. Finally, these optimized compensation parameters are input into the polarization modulator to perform real-time compensation modulation of the transmitted beam's polarization state.

[0074] For example, in a satellite laser communication experiment, the polarization state of the transmitted beam was first measured: the total intensity was 1.0 milliwatt, the difference between the horizontal and vertical polarization components was 0.3 milliwatts, the difference between the positive and negative 45-degree polarization components was -0.2 milliwatts, and the difference between the right-handed and left-handed circular polarization components was 0.1 milliwatts. Based on these measurement data, a fourth-order Miller matrix was constructed to describe the channel characteristics.

[0075] Performing eigenvalue decomposition on this fourth-order Miller matrix yielded four eigenvalues: 0.95, 0.85, 0.80, and 0.75. Taking the square root of each eigenvalue yielded the following contributions: 0.975, 0.922, 0.894, and 0.866, with a total sum of 3.657. Further calculation yielded normalized contributions of 0.267, 0.252, 0.244, and 0.237, respectively.

[0076] Based on this data, an optimization objective function was constructed, and the optimal compensation parameters were obtained through a gradient iteration method: the phase compensation was 0.35π, and the rotation angle compensation was 0.28π. ​​After applying these compensation parameters to the polarization modulator, the system's polarization state mismatch was reduced from 0.45 to 0.12.

[0077] In this embodiment, by introducing complete polarization state characterization parameters, the system's perception of channel polarization characteristics is improved. The Miller matrix decomposition method is used to achieve decoupling analysis of complex channel effects. The compensation parameters are solved by the optimization algorithm to ensure the optimality of the compensation effect. The systematic improvement significantly improves the accuracy of polarization state compensation, enhances the system's adaptability to channel disturbances, and effectively reduces the degradation of communication performance caused by polarization state mismatch.

[0078] In an optional embodiment,

[0079] A polarization state prediction model is constructed based on a deep learning neural network, wherein the polarization state prediction model input includes atmospheric disturbance data, satellite attitude data, and historical polarization state data, and output includes a predicted polarization state drift trend. Pre-correcting the polarization state compensation parameter according to the polarization state drift trend includes:

[0080] Collect atmospheric temperature distribution, atmospheric pressure distribution, humidity distribution and wind speed distribution to form atmospheric disturbance data, collect roll angle, pitch angle, yaw angle and angular velocity vector to form satellite attitude data, obtain Stokes parameter vectors at multiple sampling moments to form a historical polarization state sequence, and the sampling interval of the historical polarization state sequence is a preset time interval;

[0081] Inputting the atmospheric disturbance data, the satellite attitude data, the historical polarization state sequence, and the initial polarization state compensation parameter into a recurrent neural network with a hybrid attention mechanism, the recurrent neural network calculating a hidden layer state at a current moment based on a long short-term memory unit, generating attention weights for the atmospheric disturbance data and the satellite attitude data according to the hidden layer state, performing weighted fusion of the attention weights with the atmospheric disturbance data and the satellite attitude data to obtain a first eigenvector and a second eigenvector, and performing feature fusion of the first eigenvector and the second eigenvector with the hidden layer state to obtain feature fusion data;

[0082] The feature fusion data is input into a pre-trained neural network mapping function to predict the Stokes parameters after the next preset time interval, the Stokes parameters are subtracted from the Stokes parameters at the latest moment in the historical polarization state sequence to obtain a polarization state difference, the polarization state difference is divided by the preset time interval to obtain a polarization state drift velocity vector, a pre-compensation objective function including a second-order regularization term of a pre-compensation parameter is constructed according to the polarization state drift velocity vector, the gradient descent method is used in combination with a deep reinforcement learning framework to iteratively optimize the pre-compensation objective function to obtain pre-compensation parameters, and the pre-compensation parameters are used to pre-compensate the Stokes parameters at the latest moment in the historical polarization state sequence.

[0083] The system then performs data collection and preprocessing. Using a distributed atmospheric sensor network, the system collects comprehensive atmospheric disturbance data, including atmospheric temperature, pressure, humidity, and wind speed distributions. Simultaneously, it acquires real-time roll, pitch, and yaw angle data from the satellite attitude control system, along with the satellite's angular velocity vector, to form a satellite attitude dataset. Furthermore, the Stokes parameter vectors are continuously collected at pre-set intervals to form a historical polarization state sequence database.

[0084] The collected multi-source data is input into a recurrent neural network with a hybrid attention mechanism for processing. This network uses long short-term memory (LSTM) units as basic computational units, effectively capturing long-term dependencies in data sequences. It then calculates the current hidden state and then, based on this hidden state, calculates attention weights for the atmospheric disturbance data and satellite attitude data. This attention mechanism enables the network to adaptively focus on the importance of different data features. The system then performs a weighted fusion of the calculated attention weights with the atmospheric disturbance data and satellite attitude data, respectively, to produce two feature vectors. Finally, these feature vectors are subjected to deep feature fusion with the hidden state to generate feature-fused data containing multidimensional information.

[0085] The feature fusion data is input into a pre-trained neural network mapping function. This mapping function, by learning from historical data, can predict the Stokes parameters after the next time interval. The system calculates the difference between the predicted Stokes parameters and the most recently recorded Stokes parameters in the historical polarization state sequence and divides the difference by the preset time interval to obtain a drift velocity vector that reflects the trend of polarization state changes. Based on the drift velocity vector, an objective function is constructed that includes a second-order regularization term for the pre-compensation parameters. This objective function is optimized using a gradient descent method combined with a deep reinforcement learning framework to obtain the optimal pre-compensation parameters, which are then used to pre-compensate the current polarization state.

[0086] For example, assume the following data is collected during a satellite laser communication mission: atmospheric temperature fluctuates between -60°C and 30°C, pressure decreases from 1 atmosphere to 0.1 atmosphere, relative humidity varies between 0-80%, and wind speed reaches a maximum of 15 meters per second. Satellite attitude data shows that the roll angle fluctuates within ±2 degrees, the pitch angle is maintained at approximately 45 degrees, the yaw angle is controlled within ±1 degree, and the maximum component of the angular velocity vector does not exceed 0.1 degrees per second. The system continuously collects 1000 sets of Stokes parameter vectors at 1 millisecond intervals.

[0087] The parameter vector was input into a hybrid attention recurrent neural network consisting of 128 LSTM units, with an attention layer dimension of 64. The network assigned a weight of 0.6 to the atmospheric disturbance data and a weight of 0.4 to the attitude data. After feature fusion, the data dimension was compressed to 32. The pretrained neural network mapping function predicted the Stokes parameters at the next time point. The polarization state drift trend was obtained by subtracting the Stokes parameters from the current parameters and dividing by the 1 millisecond time interval. Finally, after 500 iterative optimizations, the optimal pre-compensation parameters were obtained, achieving pre-compensation of the polarization state.

[0088] In this embodiment, the system's perception of the communication environment is improved through the fusion analysis of multi-source data. The introduction of the hybrid attention mechanism enhances the system's ability to extract key features. Through the establishment of a prediction model, accurate prediction of the polarization state change trend is achieved.

[0089] In an optional embodiment,

[0090] A pre-compensation objective function including a second-order regularization term of a pre-compensation parameter is constructed according to the polarization state drift velocity vector. The pre-compensation objective function is iteratively optimized using a gradient descent method in combination with a deep reinforcement learning framework to obtain pre-compensation parameters including:

[0091] Constructing a system state vector including the current polarization state, the polarization state drift velocity vector, the compensation error, and the compensation parameter, and constructing the rotation angle parameter adjustment amount and the phase parameter adjustment amount as an action vector;

[0092] Calculating the square term of the difference between the product of the compensation matrix and the predicted polarization state and the target polarization state based on the system state vector, and constructing an immediate reward function in combination with the square terms of the rotation angle parameter adjustment amount and the phase parameter adjustment amount, and accumulating the immediate rewards within the future predicted steps using a discount factor to obtain a pre-compensation objective function;

[0093] Based on the pre-compensation objective function, a deep neural network is constructed as a value network and a policy network, the value network inputs the system state vector and the action vector and outputs a value evaluation, the policy network inputs the system state vector and outputs an action strategy, and the state transition samples are stored in an experience replay buffer;

[0094] A temporal difference objective function is constructed based on the instant reward function and the value network, a policy gradient objective function is constructed based on the value network and the policy network, and an entropy regularization objective function is constructed by introducing policy distribution entropy;

[0095] The temporal difference objective function is optimized by the gradient descent method to update the network parameters of the value network, the weighted sum of the policy gradient objective function and the entropy regularization objective function is optimized to update the network parameters of the policy network, and the target network parameters are updated by a soft update method;

[0096] Gaussian noise is injected into the action strategy output by the strategy network to obtain an actual execution action, the variance of the Gaussian noise is calculated based on the compensation error, and the compensation parameter is iteratively optimized using the actual execution action as the update amount of the compensation parameter to obtain a pre-compensation parameter.

[0097] A complete system state vector is constructed, consisting of four key components: the currently measured polarization state, the calculated polarization state drift velocity vector, the real-time compensation error, and the currently used compensation parameters. The rotation angle and phase parameter adjustments to be optimized are combined into an action vector, which is then output as the control variable.

[0098] An immediate reward function is constructed. This function consists of two parts: the first is the squared difference between the product of the compensation matrix and the predicted polarization state and the target polarization state, which measures the compensation effect; the second is the squared terms of the rotation angle and phase parameter adjustments, which constrain the range of parameter changes. By introducing a discount factor, the immediate rewards within the future prediction steps are weighted and accumulated to form the pre-compensation objective function.

[0099] Within the deep reinforcement learning framework, a value network and a policy network are constructed separately. The value network receives the system state vector and action vector as input and outputs an assessment of the value of the current state-action combination. The policy network receives the system state vector as input and outputs the corresponding action policy. During training, state transition experience samples are stored in the experience replay buffer for subsequent offline learning.

[0100] Based on the constructed immediate reward function and value network, a temporal difference objective function is established to evaluate and optimize the predictive ability of the value network. Simultaneously, a policy gradient objective function is constructed based on the value network and policy network, and the entropy of the policy distribution is introduced to construct an entropy regularization objective function to improve exploration efficiency.

[0101] Gradient descent is used to optimize these objective functions separately: the temporal difference objective function is optimized to update the value network parameters; the weighted sum of the policy gradient objective function and the entropy regularization objective function is optimized to update the policy network parameters. Soft updates are also used to update the target network parameters to ensure the stability of the training process.

[0102] Gaussian noise is injected into the action policy output by the policy network to generate the actual execution action. The variance of the Gaussian noise is dynamically adjusted based on the compensation error. When the compensation error is large, the exploration intensity is increased, and vice versa. The actual execution action is used as the compensation parameter update, and the compensation parameter is iteratively optimized to obtain the pre-compensation parameter.

[0103] For example, assume that during a polarization state compensation process, the initial value of the state vector is: current polarization state [1.0, 0.5, -0.3, 0.2], polarization state drift velocity vector [0.01, -0.02, 0.015, -0.005], compensation error 0.1, and current compensation parameter [0.3, 0.4]. The initial value of the action vector is set to: rotation angle adjustment amount 0.05, phase parameter adjustment amount 0.03, and the constructed immediate reward function is: R=-||M(θ,φ)S_pred-S_target||²-α(θ²+φ²), where M is the compensation matrix, S_pred is the predicted polarization state, S_target is the target polarization state, and α is the weight factor. Set the discount factor γ=0.95 and the number of prediction steps to 10;

[0104] The value network uses a three-layer fully connected network structure, with the input layer containing nodes equal to the state vector dimension plus the action vector dimension, the hidden layer containing 64 nodes, and the output layer containing a single node. The policy network also uses a three-layer fully connected structure, with the input layer containing nodes equal to the state vector dimension, the hidden layer containing 64 nodes, and the output layer containing nodes equal to the action vector dimension.

[0105] The experience replay buffer capacity is set to 10,000, and the batch size of each sampling is 128. The temporal difference objective function is constructed using the TD(0) algorithm, the policy gradient objective function is estimated using the advantage function, and the entropy regularization coefficient is set to 0.01.

[0106] During training, the learning rates of the value network and the policy network were set to 0.001, and the soft update coefficient of the target network was 0.005. Gaussian noise with a mean of 0 was injected into the action policy output by the policy network, with an initial variance of 0.1, which decayed as the compensation error decreased.

[0107] After 1000 rounds of iterative optimization, the compensation error was reduced from an initial 0.1 to 0.02, resulting in the final pre-compensation parameters of 0.42 for rotation angle compensation and 0.35 for phase compensation. These parameters significantly reduced the mismatch between the actual and target polarization states, achieving precise pre-compensation.

[0108] This embodiment, by considering long-term benefits, breaks through the limitations of traditional methods that only focus on immediate effects and achieves better compensation effects. Leveraging the powerful expressive power of deep neural networks, it achieves accurate modeling of complex nonlinear mapping relationships. Through online learning of the policy network, the compensation strategy can be continuously optimized and adapt to environmental changes. By introducing an exploration mechanism, the algorithm's adaptability to unknown states is enhanced.

[0109] The polarization state compensation parameter optimization method in the existing technology mainly adopts a single gradient descent or heuristic algorithm for solution. The optimization process only considers the compensation error of the current state, lacks consideration of long-term benefits, and is prone to falling into local optimal solutions. The optimization strategy is fixed and lacks exploration capabilities, making it difficult to adapt to complex and changeable spatial channel environments, resulting in unstable compensation effects. In addition, traditional methods often regard the optimization of compensation parameters as an independent optimization problem, ignoring the continuity characteristics of state transfer during the compensation process, making it difficult for the optimization results to meet actual application requirements. This embodiment constructs a complete state vector containing the current polarization state, drift velocity vector, compensation error and compensation parameters, and uses the adjustment amount of the compensation parameters as the action vector to establish a mapping relationship from the state space to the action space. By designing an immediate reward function containing compensation effect terms and parameter constraint terms, and combining the discount factor to accumulate future benefits, the optimization of long-term compensation effects is achieved;

[0110] This embodiment adopts a dual-network architecture to separate value evaluation and strategy generation. The value network is responsible for evaluating the long-term value of the state-action combination, and the strategy network is responsible for generating the optimal compensation parameter adjustment strategy. Through the experience replay mechanism, the correlation between samples is broken, and the learning efficiency is improved. The strategy distribution entropy is introduced as a regularization term to maintain moderate exploration capabilities while ensuring strategy convergence. In summary, this embodiment significantly improves the overall performance of the satellite laser communication system, provides a new technical path for high-precision polarization state compensation, and is of great significance to improving the reliability of space laser communication.

[0111] In an optional embodiment,

[0112] A polarization modulator is used to perform real-time polarization state modulation on the transmitted light beam, a modulation parameter of the polarization modulator is adjusted according to a pre-calibrated polarization state compensation parameter, and an intersatellite laser communication link quality evaluation model is established based on the modulation parameter of the polarization modulator. Calculating the link quality parameter includes:

[0113] Using a polarization modulator to modulate the polarization state of the transmitted light beam in real time, and adjusting the modulation parameters of the polarization modulator according to the pre-calibrated polarization state compensation parameters, wherein the modulation parameters include the polarization state rotation angle and the polarization state ellipticity;

[0114] Establishing an intersatellite laser communication link quality assessment model based on the modulation parameters of the polarization modulator, wherein the link quality assessment model includes an influence factor of atmospheric disturbance on the polarization state, a receiving end polarization state analysis factor, and a system polarization extinction ratio factor;

[0115] A link quality parameter is calculated according to the link quality evaluation model, and the polarization state compensation parameter is updated in real time according to the link quality parameter.

[0116] The polarization modulation stage begins. Based on the polarization compensation parameters obtained through pre-calibration, the polarization modulator's modulation parameters are precisely set, including two key parameters: the polarization rotation angle and the polarization ellipticity. After receiving these parameters, the polarization modulator performs real-time polarization modulation on the transmitted beam, precisely controlling the polarization state of the transmitted beam by adjusting the relative intensity and phase difference of the beam in different polarization directions.

[0117] A quality assessment model for intersatellite laser communication links was established. This model incorporates three core evaluation factors: the first is the atmospheric perturbation factor on the polarization state, which accounts for the effects of atmospheric turbulence, scattering, and absorption on the polarization state; the second is the receiver polarization analysis factor, which describes the receiver optical system's ability to analyze the polarization state of the incident light; and the third is the system polarization extinction ratio factor, which reflects the isolation of the communication system between different polarization states. These three factors are combined through a mathematical model to form a complete link quality assessment model.

[0118] Calculate link quality parameters and update compensation parameters. Based on the established link quality assessment model, calculate the quality parameters of the current communication link. These parameters reflect the quality of the current polarization modulation. Based on the calculated link quality parameters, the polarization compensation parameters are updated in real time. When the link quality parameters fall below a preset threshold, the compensation parameter update process is triggered. New compensation parameter values ​​are calculated using an optimization algorithm and re-input into the polarization modulator, forming a closed-loop control loop.

[0119] For example, assume that during an intersatellite laser communication, the initial polarization state compensation parameters are set to a rotation angle of π / 4 and an ellipticity of 0.5. The polarization modulator modulates the transmitted beam based on these parameters to achieve a specific polarization state distribution.

[0120] In the link quality assessment model, the atmospheric disturbance impact factor is expressed as exp(-σ²h / h0), where σ² represents the atmospheric turbulence intensity, h represents the transmission distance, and h0 represents the equivalent atmospheric altitude. The receiving-end polarization analysis factor is represented by the Miller matrix, which incorporates the polarization response characteristics of the receiving optical system. The system polarization extinction ratio factor is defined as the logarithm of the power ratio between orthogonal polarization states.

[0121] Real-time monitoring revealed that the link quality parameter gradually decreased from an initial value of 0.9 to 0.7, falling below the preset threshold of 0.8. This triggered the compensation parameter update mechanism. After algorithmic optimization, the rotation angle was adjusted to π / 3 and the ellipticity to 0.6. These updated compensation parameters were input into the polarization modulator, which readjusted the polarization state of the transmitted beam, restoring the link quality parameters to the ideal level.

[0122] This closed-loop control of real-time modulation and evaluation ensures the stable operation of the intersatellite laser communication link. The introduction of the evaluation model provides a reliable basis for updating the compensation parameters, improving the adaptability and reliability of the communication system.

[0123] In this embodiment, the polarization state quality of the transmitted light beam is improved through precise polarization state modulation, the accurate evaluation of the link quality is achieved through a multi-factor evaluation model, and the continuous optimization capability of the system is guaranteed through a real-time parameter update mechanism. In summary, this embodiment not only improves the accuracy of polarization state modulation, but also enhances the adaptive capability of the system, effectively solves various problems existing in traditional methods, provides a new technical path for high-quality intersatellite laser communication, and is of great significance to improving the overall performance of space communication systems.

[0124] In an optional embodiment,

[0125] Based on the link quality parameters, adaptive optical carrier recovery technology based on coherent detection uses an optical frequency comb as the local oscillator light source, achieves carrier synchronization through a digital phase-locked loop, and uses an adaptive equalization algorithm to compensate for channel dispersion and nonlinear effects. The equalizer parameters are dynamically adjusted according to the channel state information, including:

[0126] Acquiring link quality parameters, using an optical frequency comb as a local oscillator light source, generating frequency comb spectral lines based on a preset center frequency and comb tooth spacing, wherein the frequency comb spectral lines are determined by the amplitude, frequency, and initial phase of the frequency components, monitoring temperature drift and driving voltage fluctuation, calculating frequency drift based on the product of a temperature sensitivity coefficient and the temperature drift, and the product of a voltage sensitivity coefficient and the driving voltage fluctuation, and performing real-time compensation for the frequency of the optical frequency comb;

[0127] Performing coherent detection on a received optical signal and the optical frequency comb to obtain a beat signal, performing complex conjugate multiplication on the beat signal and a local reference signal to obtain a phase error, inputting the phase error into a digital phase-locked loop, filtering the phase error using a proportional-integral controller, and calculating a control voltage based on a weighted sum of a current phase error multiplied by a proportional coefficient and a cumulative sum of historical phase errors multiplied by an integral coefficient;

[0128] Calculating an output phase of a digitally controlled oscillator according to the control voltage, the output phase being obtained by summing the output phase at a previous moment, the product of the center frequency and the sampling period, and the product of the control voltage and the gain coefficient, and using the output phase as a phase parameter of the local reference signal to achieve carrier synchronization;

[0129] Establishing a channel transmission model, the channel transmission model including a dispersion compensation model and a nonlinear effect compensation model, wherein the dispersion compensation model calculates a dispersion compensation coefficient based on the wavelength of the optical signal, the transmission distance, and the speed of light, and the nonlinear effect compensation model calculates a nonlinear compensation coefficient based on the input signal, the nonlinear coefficient, the signal power, and the effective length, performing feature extraction and optimizing loss function design through deep learning technology, and combining an adaptive equalization algorithm to compensate for the channel dispersion and the nonlinear effect, and multiplying multiple delayed signals of the input signal by the equalizer coefficients and summing the results to obtain an output signal of the adaptive equalization algorithm;

[0130] An adaptive step factor is calculated according to the link quality parameter, and the equalizer coefficient is updated by taking the product of the adaptive step factor and the conjugate signal of the error signal and the input signal delay signal as the update amount. The size of the adaptive step factor is dynamically adjusted based on the link quality parameter to achieve adaptive optimization of the equalizer coefficient and output the demodulated communication signal.

[0131] The initial operating state of the optical frequency comb is set according to the preset center frequency and comb tooth spacing parameters. The amplitude, frequency, and initial phase of each frequency component need to be precisely controlled. A temperature sensor continuously monitors changes in the working environment temperature and records the temperature drift. At the same time, a high-precision voltage detection circuit is used to collect the fluctuation value of the driving voltage in real time. The temperature drift is multiplied by the temperature sensitivity coefficient to obtain the frequency offset caused by temperature, and the driving voltage fluctuation is multiplied by the voltage sensitivity coefficient to obtain the frequency offset caused by voltage. The two are added to obtain the total frequency drift. Based on the calculated frequency drift, the driving parameters of the optical frequency comb are adjusted to achieve real-time frequency compensation and ensure that the output characteristics of the frequency comb remain stable.

[0132] The received optical signal is coherently detected with the compensated optical frequency comb output, generating a beat frequency signal at the photodetector output. This beat frequency signal is complex-conjugate multiplied by a digitally generated local reference signal to extract phase error information. The phase error signal is input into a digital phase-locked loop (PLL) and processed by a proportional-integral controller (PIC). The controller utilizes a two-path parallel structure: the proportional path multiplies the current phase error by a proportional coefficient to obtain the instantaneous control variable; the integral path accumulates historical phase errors and multiplies them by an integral coefficient to obtain the cumulative control variable. The two control variables are summed to obtain the final control voltage.

[0133] After receiving the control voltage input, the digitally controlled oscillator first reads the output phase value at the previous moment. The product of the center frequency and the sampling period is used as the reference phase increment. The product of the control voltage and the oscillator gain coefficient is then used as the correction phase. These three values ​​are added together to obtain the current output phase. This output phase is used to generate a new local reference signal, completing a phase update process. Carrier synchronization is achieved through continuous phase tracking.

[0134] When establishing a channel transmission model, the first step is to construct a dispersion compensation model. This model takes into account the operating wavelength, actual transmission distance, and speed of light in a vacuum as inputs to the optical signal and calculates the dispersion compensation coefficient. A nonlinear effect compensation model is also constructed, which comprehensively considers the input signal characteristics, the fiber nonlinear coefficient, the signal transmission power, and the effective fiber length to calculate the nonlinear compensation coefficient. A multi-layer convolutional neural network is used to extract signal features, a loss function based on mean square error is designed, and a channel response prediction model is established in conjunction with a recursive neural network. The extracted features are then input into an adaptive equalization algorithm, which performs a multi-stage delay expansion on the input signal. These features are then multiplied by the corresponding equalizer coefficients and summed to produce the equalized output signal.

[0135] Based on real-time link quality parameters, an adaptive step factor is dynamically calculated. This step factor is multiplied by the error signal and the conjugate of the input signal's delayed component to determine the updated equalizer coefficients. When link quality parameters are low, the step factor is increased to accelerate convergence; when link quality parameters are high, the step factor is decreased to improve stability. This adaptive update mechanism optimizes the equalizer coefficients, ultimately outputting a stable demodulated communication signal.

[0136] For example, assume the center frequency of the optical frequency comb is set to 193.1 THz, with a comb tooth spacing of 25 GHz. A temperature drift of 0.1°C (a temperature sensitivity coefficient of -1.5 GHz / °C) is detected, while a drive voltage fluctuation of 0.01 V (a voltage sensitivity coefficient of 2 GHz / V) is detected. The calculated total frequency drift is -0.13 GHz, and the frequency comb drive parameters are adjusted accordingly.

[0137] In coherent detection, the beat signal frequency is 2 GHz and the amplitude is 0.5 V. The digital phase-locked loop's proportional coefficient is set to 0.2, and the integral coefficient is set to 0.05. When a phase error of 0.1 rad is detected, the calculated control voltage is 0.025 V.

[0138] The gain coefficient of the numerically controlled oscillator is set to 1 MHz / V, and the sampling period is 0.1 ns. When the output phase at the previous moment was 1.5 rad, the output phase at the current moment is calculated to be 1.5126 rad.

[0139] In the channel model, the optical signal wavelength is 1550 nm, the transmission distance is 100 km, and the calculated dispersion compensation coefficient is 17 ps / nm / km. The nonlinear coefficient is set to 2.6 W^(-1)km^(-1) and the signal power is 1 mW. The nonlinear compensation coefficient is calculated.

[0140] The equalization algorithm uses an 11-tap structure, and the initial equalizer coefficients are [0.1, 0.2, 0.3, 0.4, 0.5, 1, 0.5, 0.4, 0.3, 0.2, 0.1]. When the link quality parameter is 0.9, the step size factor is set to 0.001. When the quality parameter drops to 0.7, the step size factor is increased to 0.005 to accelerate the convergence of the equalizer coefficients.

[0141] In this embodiment, the frequency stability of the optical frequency comb is significantly improved through a frequency compensation mechanism coupled with multiple factors. The application of a digital phase-locked loop achieves faster and more accurate carrier synchronization. The accuracy of channel compensation is improved through a deep learning-enhanced channel model. The adaptive equalization algorithm enhances the system's adaptability to channel changes.

[0142] Existing intersatellite laser communication systems typically use single temperature control or simple feedback regulation to maintain frequency stability, lacking comprehensive consideration of multi-source interference. Conventional methods often separate dispersion and nonlinear effects for channel compensation, employing fixed-parameter equalization algorithms that struggle to adapt to complex and changing spatial channel environments. Traditional carrier synchronization technology primarily relies on hardware phase-locked loops, which suffer from slow response and poor adaptability.

[0143] This embodiment decouples and analyzes the effects of temperature drift and driving voltage fluctuations, establishes an accurate frequency drift compensation model through their respective sensitivity coefficients, and achieves high-precision and stable control of the optical frequency comb. It realizes the transition from single compensation to multi-dimensional coordination, the improvement from fixed parameters to adaptive optimization, and the leap from independent processing to system integration. It not only improves the stability of the communication system, but also enhances the environmental adaptability of the system, effectively solves various problems existing in traditional methods, achieves an overall improvement in the performance of the communication system, and provides an innovative solution for the development of space laser communication systems.

[0144] In an optional embodiment,

[0145] The method uses deep learning technology to extract features and optimize loss function design, combines an adaptive equalization algorithm to compensate for the channel dispersion and the nonlinear effect, and multiplies multiple delayed signals of the input signal by the equalizer coefficients and sums them to obtain the output signal of the adaptive equalization algorithm, including:

[0146] Using a sliding time window of a preset length, the input signal is subjected to time domain feature extraction to obtain signal amplitude, signal phase change, and signal instantaneous power; performing a fast Fourier transform on the signal data within the sliding time window to obtain a frequency domain signal; extracting the spectrum amplitude, channel dispersion characteristics, and nonlinear effect characteristics of the frequency domain signal; and forming a feature vector from the signal amplitude, signal phase change, signal instantaneous power, spectrum amplitude, channel dispersion characteristics, and nonlinear effect characteristics;

[0147] Inputting the feature vector into a first neural network for feature dimensionality reduction to obtain a first feature, the first neural network including a weight matrix and a bias vector, inputting the first feature into a long short-term memory network to obtain a latent state feature, the long short-term memory network being used to model temporal features, and inputting the latent state feature into a fully connected neural network to obtain an equalizer coefficient;

[0148] Compensating for the channel dispersion and the nonlinear effect, multiplying the multiple delayed signals of the input signal by the equalizer coefficients and summing the results to obtain an output signal of the adaptive equalization algorithm, subtracting the output signal of the adaptive equalization algorithm from the desired output signal to obtain an error signal, and updating the equalizer coefficients according to the error signal, an adaptive step size factor, a conjugate signal of the delayed signal of the input signal, and a fusion factor;

[0149] Constructing a comprehensive loss function including a mean square error loss term, a bit error rate loss term, and a peak-to-average ratio loss term, and optimizing the network parameters of the first neural network, the network parameters of the long short-term memory network, and the network parameters of the fully connected neural network according to the comprehensive loss function using a gradient descent method;

[0150] A channel state index is calculated according to the signal-to-noise ratio, the error vector magnitude, and the bit error rate, the adaptive step factor is dynamically adjusted according to the channel state index and the error signal, and the fusion factor is updated according to the channel state index and the comprehensive loss function.

[0151] Perform signal feature extraction. Set a sliding time window of a preset length and perform time-domain analysis on the input signal to extract three basic features: signal amplitude by calculating the signal envelope; signal phase change by the phase difference between adjacent sampling points; and instantaneous power by squared signal amplitude. Perform a fast Fourier transform on the signal data within the sliding time window to extract three features in the frequency domain: spectral amplitude by calculating the frequency domain signal amplitude; channel dispersion characteristics by multiplying the dispersion coefficient and signal bandwidth; and nonlinear effect characteristics by multiplying the nonlinear coefficient and signal power. These six features are sequentially combined into a feature vector.

[0152] The feature vector is input into the first neural network for dimensionality reduction. This network consists of a multi-layer structure, each containing a weight matrix and a bias vector. A nonlinear activation function transforms the features to produce the reduced first feature. The first feature is then chronologically input into a long short-term memory network (LSTM). This network consists of three control units: an input gate, a forget gate, and an output gate. This gating mechanism models the temporal information and outputs latent state features. The latent state features are processed by a fully connected neural network, which maps the features to the same dimension as the number of equalizer taps, resulting in the equalizer coefficients.

[0153] The resulting equalizer coefficients are used to compensate the signal. The input signal is passed through multiple delay units to obtain a delayed signal sequence. Each delayed signal is multiplied by the corresponding equalizer coefficient, and all products are summed to obtain the output signal of the equalization algorithm. The output signal is subtracted from the predetermined desired output signal to obtain an error signal. Based on the error signal, the current adaptive step size factor, the conjugate signal of the input signal's delayed sequence, and the fusion factor, the update amount for the equalizer coefficients is calculated and updated.

[0154] A comprehensive loss function is constructed to optimize the neural network. The loss function consists of three terms: the mean squared error between the output signal and the expected signal, the system bit error rate, and the peak-to-average ratio of the output signal. Gradient descent is used to calculate the gradient of the loss function with respect to each network parameter, updating the weight matrix and bias vector of the first neural network, the gating parameters of the long short-term memory network, and the connection weights of the fully connected neural network.

[0155] Channel state indicators are calculated in real time. The signal-to-noise ratio is calculated from the signal-to-noise power ratio, the error vector magnitude is calculated from the deviation between the actual and ideal constellation points, and the bit error rate is calculated from the ratio of the number of bit errors after demodulation to the total number of bits. These three indicators are weighted to obtain the channel state indicator. The adaptive step size factor is dynamically adjusted based on the changing trend of the channel state indicator and the size of the error signal. Simultaneously, the fusion factor used to update the equalizer coefficients is updated based on the channel state indicator and the value of the comprehensive loss function.

[0156] For example, a sliding time window length of 64 sampling points is set to extract features from the input 16QAM modulated signal. In the time domain, the signal amplitude ranges from [0, 1], the phase variation range is from [-π, π], and the instantaneous power range is from [0, 1]. A 2048-point FFT is performed to obtain the frequency domain signal. The spectral amplitude, dispersion characteristics, and nonlinear features are extracted to form a feature vector with a dimension of 6.

[0157] The first neural network uses a three-layer structure, with 32 and 16 neurons in the hidden layer, respectively. The ReLU activation function reduces the 6-dimensional feature vector to 8-dimensional features. The long short-term memory network uses 16 storage units to establish temporal dependencies. The fully connected network maps the 16-dimensional features to an 11-dimensional output, corresponding to an 11-tap equalizer.

[0158] The initial coefficients of the equalizer are set to [0.1, 0.2, 0.3, 0.4, 0.5, 1, 0.5, 0.4, 0.3, 0.2, 0.1]. The input signal passes through 10 delay units, resulting in 11 delayed signals. The initial adaptive step size factor is set to 0.001, and the fusion factor is set to 0.5.

[0159] The weights of the three loss terms in the comprehensive loss function are 0.5, 0.3, and 0.2, respectively. The Adam optimizer is used for parameter updates, with a learning rate of 0.001. When the channel state index is less than 0.8, the step size factor is increased; when the loss function value is greater than the threshold, the fusion factor is reduced to achieve adaptive adjustment.

[0160] In this embodiment, the ability to characterize signal characteristics is improved through the fusion extraction of time-frequency domain features, the ability to compensate for nonlinear distortion is enhanced through the processing of deep neural networks, the comprehensive improvement of equalization performance is achieved through multi-objective optimization design, and the algorithm's adaptability to channel changes is improved through parameter adaptive adjustment.

[0161] Adaptive equalization algorithms in the existing technology often only focus on the characteristics of a single domain, either time domain characteristics or frequency domain characteristics, and are difficult to fully reflect signal characteristics. Conventional equalization algorithms adopt a fixed update strategy and lack the ability to adapt to channel conditions. Traditional methods are too single in the design of optimization objectives, usually only considering the mean square error, and cannot meet the multi-dimensional optimization requirements in complex communication environments. This embodiment integrates time domain characteristics and frequency domain characteristics, and realizes dynamic extraction of time domain characteristics such as signal amplitude, phase change, and instantaneous power through a sliding time window. At the same time, it uses fast Fourier transform to obtain spectrum amplitude, channel dispersion characteristics and nonlinear effect characteristics, and constructs a complete feature vector. In summary, this embodiment not only improves the equalization accuracy, but also enhances the robustness of the algorithm, effectively solves various problems existing in traditional methods, achieves an overall improvement in the performance of the equalization algorithm, and provides an innovative solution for performance optimization of communication systems.

[0162] In an optional embodiment,

[0163] Transmitting the demodulated communication signal through the intersatellite laser communication link, monitoring the link state parameters in real time, and updating the training data set of the deep learning neural network according to the link state parameters to achieve online optimization of the polarization state prediction model include:

[0164] Transmitting the demodulated communication signal through an intersatellite laser communication link, and using a polarization state analyzer to monitor link state parameters of the intersatellite laser communication link in real time, wherein the link state parameters include signal-to-noise ratio, bit error rate, and polarization extinction ratio;

[0165] Combining the link state parameters and corresponding polarization state parameters into training sample pairs, and updating a training data set of a deep learning neural network based on the training sample pairs, wherein the deep learning neural network is used to establish a polarization state prediction model;

[0166] The deep learning neural network is trained using the training data set, the network learning rate is adjusted according to the signal-to-noise ratio in the link state parameter, the weight coefficient of the network optimization objective function is adjusted according to the bit error rate in the link state parameter, and the network structure parameters are dynamically adjusted according to the polarization extinction ratio in the link state parameter to achieve online optimization of the polarization state prediction model.

[0167] The demodulated communication signal is transmitted via an intersatellite laser communication link. During transmission, a polarization state analyzer continuously collects link status parameters. The signal-to-noise ratio (SNR) is calculated by calculating the ratio of received signal power to noise power. The bit error rate (BER) is calculated by comparing the number of bit errors to the total number of bits. The polarization extinction ratio (PER) is calculated by measuring the power ratio between orthogonal polarization states. These three parameters collectively reflect the transmission quality of the link.

[0168] At each sampling moment, the current polarization state parameters, including characteristic parameters such as polarization ellipticity and azimuth, are recorded. These polarization state parameters are paired with the link state parameters measured at the corresponding moment to form training sample pairs. Each training sample pair contains input features (link state parameters) and output labels (polarization state parameters). Newly collected training sample pairs are continuously added to the training dataset to achieve dynamic updating of the dataset.

[0169] A deep learning neural network is constructed using a multi-layered architecture consisting of an input layer, multiple hidden layers, and an output layer. The input layer receives link state parameters, which are transformed nonlinearly in the hidden layers to generate predicted polarization state parameters in the output layer. Each hidden layer contains multiple neurons, connected by a weight matrix and equipped with an activation function to implement nonlinear mapping.

[0170] Adjust the network learning rate based on changes in the signal-to-noise ratio: When the signal-to-noise ratio is high, indicating good channel conditions, a larger learning rate can be used to accelerate convergence; when the signal-to-noise ratio decreases, reduce the learning rate to improve training stability. Secondly, adjust the weight coefficients of each term in the optimization objective function based on changes in the bit error rate: When the bit error rate increases, increase the weight of the corresponding loss term to strengthen learning of samples with large errors. Finally, dynamically adjust the network structure parameters based on changes in the polarization extinction ratio: When the polarization extinction ratio decreases, appropriately increase the number of network layers or neurons to improve the network's expressive power.

[0171] For example, assume that during an intersatellite laser communication, a demodulated 16QAM modulated signal is transmitted over a link. The polarization state analyzer samples the signal every millisecond, measuring link state parameters such as the signal-to-noise ratio, bit error rate, and polarization extinction ratio. The polarization state parameters at that moment, including ellipticity and azimuth, are also recorded.

[0172] The initial configuration of the deep neural network is a four-layer structure: an input layer with three nodes (corresponding to the three link state parameters), two hidden layers with 64 and 32 nodes respectively, and an output layer with two nodes (corresponding to the polarization state parameters). The initial learning rate is set to 0.001, and the optimization objective function includes a mean square error term and a regularization term, with an initial weight ratio of 1:0.1.

[0173] When the signal-to-noise ratio dropped from 20dB to 15dB, the learning rate was adjusted to 0.0005. When the bit error rate increased from 1e-6 to 1e-4, the error term weight was adjusted to 2:0.1. When the polarization extinction ratio dropped from 30dB to 20dB, 16 neurons were added to the second hidden layer. After each adjustment, training was repeated until the network performance met the requirements.

[0174] The Adam optimizer is used for parameter updates, with batch training performed on 128 samples at a time. Model performance is evaluated through cross-validation, and a training round is completed when the prediction error on the validation set is less than a threshold. Continuous online optimization ensures that the model can adapt to dynamic changes in link status.

[0175] In this embodiment, the response speed of the prediction model to link changes is improved through real-time monitoring of link status and dynamic updating of data sets. The environmental adaptability of the model is enhanced through adaptive adjustment of multi-dimensional parameters. The continuous improvement of prediction performance is achieved through the online optimization mechanism. The transition from static model to dynamic model, from single parameter to multi-dimensional parameter, and from offline training to online optimization are realized. This not only improves the prediction accuracy, but also enhances the adaptability of the algorithm, effectively solving various problems existing in traditional methods.

[0176] Figure 2FIG. 1 is a compensation error convergence curve diagram of the inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation according to an embodiment of the present invention, as shown in FIG. Figure 2 As shown in the figure, the proposed technical solution (circle mark) shows the fastest convergence speed in the first 100 iterations, with the error dropping rapidly from 0.1 to 0.035. The adaptive filtering algorithm (square mark) converges slowly, requiring 300 iterations to reach an error level of 0.045. The traditional LMS algorithm (triangle mark) performs in the middle, reaching an error level of 0.04 after 250 iterations. Figure 2 The results demonstrate the dual advantages of this technical solution in terms of convergence speed and final accuracy, with the final error stabilized at 0.02.

[0177] Figure 3 This is a comparison chart of the system convergence performance of the inter-satellite laser communication and signal scheduling method based on adaptive polarization modulation according to an embodiment of the present invention, showing the performance comparison of this technical solution with the MMSE algorithm and the CMA algorithm in the equalizer coefficient convergence process. Figure 3 As shown in the figure, this technical solution demonstrates faster convergence and lower steady-state error, demonstrating the advantages of deep learning-based feature extraction methods in adaptive equalization. Although the MMSE and CMA algorithms can also achieve system convergence, their convergence speed is slower and the final steady-state error is larger, indicating that the performance of traditional algorithms is limited in complex channel environments.

[0178] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. Intersatellite laser communication and signal scheduling method based on adaptive polarization modulation, characterized in that: include: Obtaining polarization state information of a transmitted light beam of the first intersatellite laser communication terminal and polarization state information of a received light beam of the second intersatellite laser communication terminal, calculating a polarization state mismatch according to the polarization state information of the light beam and the polarization state information of the received light beam, and generating an initial polarization state compensation parameter based on the polarization state mismatch; Constructing a polarization state prediction model based on a deep learning neural network, wherein the polarization state prediction model inputs include atmospheric disturbance data, satellite attitude data, and historical polarization state data, and outputs a predicted polarization state drift trend, and pre-correcting the polarization state compensation parameters according to the polarization state drift trend; Using a polarization modulator to modulate the polarization state of the transmitted light beam in real time, adjusting the modulation parameters of the polarization modulator according to a pre-calibrated polarization state compensation parameter, establishing an inter-satellite laser communication link quality assessment model based on the modulation parameters of the polarization modulator, and calculating the link quality parameters; Based on the link quality parameters, an adaptive optical carrier recovery technology based on coherent detection uses an optical frequency comb as the local oscillator light source, achieves carrier synchronization through a digital phase-locked loop, and uses an adaptive equalization algorithm to compensate for channel dispersion and nonlinear effects, and dynamically adjusts equalizer parameters based on channel state information; The demodulated communication signal is transmitted through the intersatellite laser communication link, and the link state parameters are monitored in real time. The training data set of the deep learning neural network is updated according to the link state parameters to achieve online optimization of the polarization state prediction model.

2. The method according to claim 1, characterized in that Calculating a polarization state mismatch degree according to the polarization state information of the light beam and the polarization state information of the received light beam, and generating an initial polarization state compensation parameter based on the polarization state mismatch degree includes: Obtaining polarization state information of the transmitted light beam and the received light beam, wherein the polarization state information includes the total light intensity, the difference between the horizontal and vertical polarization components, the difference between the positive and negative forty-five degree polarization components, and the difference between the right-handed and left-handed circular polarization components; Constructing a fourth-order Miller matrix, the fourth-order Miller matrix consisting of a phase delay matrix, a rotation matrix, and a depolarization matrix, performing eigenvalue decomposition on the fourth-order Miller matrix to obtain contribution components of each matrix, and calculating a polarization state mismatch based on the contribution components, where the polarization state mismatch is determined by a ratio of a square root of each contribution component to a sum of the square roots of all contribution components; Based on the polarization state mismatch, a compensation parameter optimization objective function is constructed, where the compensation parameters include a phase compensation amount and a rotation angle compensation amount. The gradient iteration method is used to solve the optimization objective function to obtain an initial polarization state compensation parameter. According to the initial polarization state compensation parameter, the polarization state of the emitted light beam is compensated and modulated by a polarization modulator.

3. The method according to claim 1, characterized in that A polarization state prediction model is constructed based on a deep learning neural network, wherein the polarization state prediction model input includes atmospheric disturbance data, satellite attitude data, and historical polarization state data, and output includes a predicted polarization state drift trend. Pre-correcting the polarization state compensation parameter according to the polarization state drift trend includes: Collect atmospheric temperature distribution, atmospheric pressure distribution, humidity distribution and wind speed distribution to form atmospheric disturbance data, collect roll angle, pitch angle, yaw angle and angular velocity vector to form satellite attitude data, obtain Stokes parameter vectors at multiple sampling moments to form a historical polarization state sequence, and the sampling interval of the historical polarization state sequence is a preset time interval; Inputting the atmospheric disturbance data, the satellite attitude data, the historical polarization state sequence, and the initial polarization state compensation parameter into a recurrent neural network with a hybrid attention mechanism, the recurrent neural network calculating a hidden layer state at a current moment based on a long short-term memory unit, generating attention weights for the atmospheric disturbance data and the satellite attitude data according to the hidden layer state, performing weighted fusion of the attention weights with the atmospheric disturbance data and the satellite attitude data to obtain a first eigenvector and a second eigenvector, and performing feature fusion of the first eigenvector and the second eigenvector with the hidden layer state to obtain feature fusion data; The feature fusion data is input into a pre-trained neural network mapping function to predict the Stokes parameters after the next preset time interval, the Stokes parameters are subtracted from the Stokes parameters at the latest moment in the historical polarization state sequence to obtain a polarization state difference, the polarization state difference is divided by the preset time interval to obtain a polarization state drift velocity vector, a pre-compensation objective function including a second-order regularization term of a pre-compensation parameter is constructed according to the polarization state drift velocity vector, the gradient descent method is used in combination with a deep reinforcement learning framework to iteratively optimize the pre-compensation objective function to obtain pre-compensation parameters, and the pre-compensation parameters are used to pre-compensate the Stokes parameters at the latest moment in the historical polarization state sequence.

4. The method according to claim 3, characterized in that A pre-compensation objective function including a second-order regularization term of a pre-compensation parameter is constructed according to the polarization state drift velocity vector. The pre-compensation objective function is iteratively optimized using a gradient descent method in combination with a deep reinforcement learning framework to obtain pre-compensation parameters including: Constructing a system state vector including the current polarization state, the polarization state drift velocity vector, the compensation error, and the compensation parameter, and constructing the rotation angle parameter adjustment amount and the phase parameter adjustment amount as an action vector; Calculating the square term of the difference between the product of the compensation matrix and the predicted polarization state and the target polarization state based on the system state vector, and constructing an immediate reward function in combination with the square terms of the rotation angle parameter adjustment amount and the phase parameter adjustment amount, and accumulating the immediate rewards within the future predicted steps using a discount factor to obtain a pre-compensation objective function; Based on the pre-compensation objective function, a deep neural network is constructed as a value network and a policy network, the value network inputs the system state vector and the action vector and outputs a value evaluation, the policy network inputs the system state vector and outputs an action strategy, and the state transition samples are stored in an experience replay buffer; A temporal difference objective function is constructed based on the instant reward function and the value network, a policy gradient objective function is constructed based on the value network and the policy network, and an entropy regularization objective function is constructed by introducing policy distribution entropy; The temporal difference objective function is optimized by the gradient descent method to update the network parameters of the value network, the weighted sum of the policy gradient objective function and the entropy regularization objective function is optimized to update the network parameters of the policy network, and the target network parameters are updated by a soft update method; Gaussian noise is injected into the action strategy output by the strategy network to obtain an actual execution action, the variance of the Gaussian noise is calculated based on the compensation error, and the compensation parameter is iteratively optimized using the actual execution action as the update amount of the compensation parameter to obtain a pre-compensation parameter.

5. The method according to claim 1, wherein A polarization modulator is used to perform real-time polarization state modulation on the transmitted light beam, a modulation parameter of the polarization modulator is adjusted according to a pre-calibrated polarization state compensation parameter, and an intersatellite laser communication link quality evaluation model is established based on the modulation parameter of the polarization modulator. Calculating the link quality parameter includes: Using a polarization modulator to modulate the polarization state of the transmitted light beam in real time, and adjusting the modulation parameters of the polarization modulator according to the pre-calibrated polarization state compensation parameters, wherein the modulation parameters include the polarization state rotation angle and the polarization state ellipticity; Establishing an intersatellite laser communication link quality assessment model based on the modulation parameters of the polarization modulator, wherein the link quality assessment model includes an influence factor of atmospheric disturbance on the polarization state, a receiving end polarization state analysis factor, and a system polarization extinction ratio factor; A link quality parameter is calculated according to the link quality evaluation model, and the polarization state compensation parameter is updated in real time according to the link quality parameter.

6. The method according to claim 1, characterized in that Based on the link quality parameters, adaptive optical carrier recovery technology based on coherent detection uses an optical frequency comb as the local oscillator light source, achieves carrier synchronization through a digital phase-locked loop, and uses an adaptive equalization algorithm to compensate for channel dispersion and nonlinear effects. The equalizer parameters are dynamically adjusted according to the channel state information, including: Acquiring link quality parameters, using an optical frequency comb as a local oscillator light source, generating frequency comb spectral lines based on a preset center frequency and comb tooth spacing, wherein the frequency comb spectral lines are determined by the amplitude, frequency, and initial phase of the frequency components, monitoring temperature drift and driving voltage fluctuation, calculating frequency drift based on the product of a temperature sensitivity coefficient and the temperature drift, and the product of a voltage sensitivity coefficient and the driving voltage fluctuation, and performing real-time compensation for the frequency of the optical frequency comb; Performing coherent detection on a received optical signal and the optical frequency comb to obtain a beat signal, performing complex conjugate multiplication on the beat signal and a local reference signal to obtain a phase error, inputting the phase error into a digital phase-locked loop, filtering the phase error using a proportional-integral controller, and calculating a control voltage based on a weighted sum of a current phase error multiplied by a proportional coefficient and a cumulative sum of historical phase errors multiplied by an integral coefficient; Calculating an output phase of a digitally controlled oscillator according to the control voltage, the output phase being obtained by summing the output phase at a previous moment, the product of the center frequency and the sampling period, and the product of the control voltage and the gain coefficient, and using the output phase as a phase parameter of the local reference signal to achieve carrier synchronization; Establishing a channel transmission model, the channel transmission model including a dispersion compensation model and a nonlinear effect compensation model, wherein the dispersion compensation model calculates a dispersion compensation coefficient based on the wavelength of the optical signal, the transmission distance, and the speed of light, and the nonlinear effect compensation model calculates a nonlinear compensation coefficient based on the input signal, the nonlinear coefficient, the signal power, and the effective length, performing feature extraction and optimizing loss function design through deep learning technology, and combining an adaptive equalization algorithm to compensate for the channel dispersion and the nonlinear effect, and multiplying multiple delayed signals of the input signal by the equalizer coefficients and summing the results to obtain an output signal of the adaptive equalization algorithm; An adaptive step factor is calculated according to the link quality parameter, and the equalizer coefficient is updated by taking the product of the adaptive step factor and the conjugate signal of the error signal and the input signal delay signal as the update amount. The size of the adaptive step factor is dynamically adjusted based on the link quality parameter to achieve adaptive optimization of the equalizer coefficient and output the demodulated communication signal.

7. The method according to claim 6, characterized in that The method uses deep learning technology to extract features and optimize loss function design, combines an adaptive equalization algorithm to compensate for the channel dispersion and the nonlinear effect, and multiplies multiple delayed signals of the input signal by the equalizer coefficients and sums them to obtain the output signal of the adaptive equalization algorithm, including: Using a sliding time window of a preset length, the input signal is subjected to time domain feature extraction to obtain signal amplitude, signal phase change, and signal instantaneous power; performing a fast Fourier transform on the signal data within the sliding time window to obtain a frequency domain signal; extracting the spectrum amplitude, channel dispersion characteristics, and nonlinear effect characteristics of the frequency domain signal; and forming a feature vector from the signal amplitude, signal phase change, signal instantaneous power, spectrum amplitude, channel dispersion characteristics, and nonlinear effect characteristics; Inputting the feature vector into a first neural network for feature dimensionality reduction to obtain a first feature, the first neural network including a weight matrix and a bias vector, inputting the first feature into a long short-term memory network to obtain a latent state feature, the long short-term memory network being used to model temporal features, and inputting the latent state feature into a fully connected neural network to obtain an equalizer coefficient; Compensating for the channel dispersion and the nonlinear effect, multiplying the multiple delayed signals of the input signal by the equalizer coefficients and summing the results to obtain an output signal of the adaptive equalization algorithm, subtracting the output signal of the adaptive equalization algorithm from the desired output signal to obtain an error signal, and updating the equalizer coefficients according to the error signal, an adaptive step size factor, a conjugate signal of the delayed signal of the input signal, and a fusion factor; Constructing a comprehensive loss function including a mean square error loss term, a bit error rate loss term, and a peak-to-average ratio loss term, and optimizing the network parameters of the first neural network, the network parameters of the long short-term memory network, and the network parameters of the fully connected neural network according to the comprehensive loss function using a gradient descent method; A channel state index is calculated according to the signal-to-noise ratio, the error vector magnitude, and the bit error rate, the adaptive step factor is dynamically adjusted according to the channel state index and the error signal, and the fusion factor is updated according to the channel state index and the comprehensive loss function.

8. The method according to claim 1, characterized in that Transmitting the demodulated communication signal through the intersatellite laser communication link, monitoring the link state parameters in real time, and updating the training data set of the deep learning neural network according to the link state parameters to achieve online optimization of the polarization state prediction model include: Transmitting the demodulated communication signal through an intersatellite laser communication link, and using a polarization state analyzer to monitor link state parameters of the intersatellite laser communication link in real time, wherein the link state parameters include signal-to-noise ratio, bit error rate, and polarization extinction ratio; Combining the link state parameters and corresponding polarization state parameters into training sample pairs, and updating a training data set of a deep learning neural network based on the training sample pairs, wherein the deep learning neural network is used to establish a polarization state prediction model; The deep learning neural network is trained using the training data set, the network learning rate is adjusted according to the signal-to-noise ratio in the link state parameter, the weight coefficient of the network optimization objective function is adjusted according to the bit error rate in the link state parameter, and the network structure parameters are dynamically adjusted according to the polarization extinction ratio in the link state parameter to achieve online optimization of the polarization state prediction model.

Images