Laser communication signal attenuation prediction and stabilization method and system based on multi-modal data fusion

Abstract

The application discloses a laser communication signal attenuation prediction and stable control method and system based on multi-modal data fusion, and relates to the technical field of laser communication, and comprises the following steps: acquiring multi-modal data of a laser communication link, taking a sampling point as a graph node, determining an edge weight according to environmental parameter similarity, and constructing graph structure data; inputting the graph structure data into a graph neural network to perform feature fusion and obtain a joint feature vector; inputting the joint feature vector into a clustering model containing an environmental variance compensation term to obtain an attenuation density distribution; and adjusting the defocusing amount of a transmitting end and the focal length of a receiving end according to the attenuation density distribution and a received power feedback closed loop, so that the received power is not lower than a preset threshold P_th and the cat-eye effect is suppressed. Through four key links of multi-source data synchronous acquisition, graph neural network (GNN) feature fusion, optimized K-means clustering and dynamic control of an optical tracking matrix, high-precision prediction and adaptive optimization of laser communication signal attenuation are realized, and the prediction precision can be improved.

Description

Technical Field

[0001] This invention relates to the field of laser communication technology, specifically to a method and system for predicting and stabilizing laser communication signal attenuation through multimodal data fusion. Background Technology

[0002] Laser communication technology, with its advantages of high bandwidth, strong anti-interference capability, and high security, has become an important development direction in the modern communication field, and is widely used in military communication, satellite communication, deep-sea exploration, and other fields. However, laser signals are affected by many factors during long-distance transmission. For example, atmospheric turbulence causes random fluctuations in refractive index, leading to laser wavefront distortion; ambient temperature and humidity affect changes in atmospheric refractive index. These uncertainties lead to a decrease in signal strength and a deterioration in communication quality, severely restricting the reliability and operating range of laser communication systems. Currently, methods for predicting laser communication signal attenuation are mainly divided into two categories: traditional methods based on physical models and data-driven machine learning methods. However, both have significant shortcomings, such as overly complex calculation methods or inability to adapt to real-time changing transmission conditions. Therefore, how to accurately predict the attenuation characteristics of laser signals and dynamically optimize communication system parameters urgently requires a prediction method that can integrate multi-source environmental data and signal characteristics. This has become a key research issue. To address this, we propose a multi-modal data fusion-based laser communication signal attenuation prediction and stabilization method and system. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for predicting and stabilizing laser communication signal attenuation through multimodal data fusion, which solves the problems mentioned in the background section.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a laser communication signal attenuation prediction and stabilization method based on multimodal data fusion, comprising the following steps: S1. Acquire multimodal data of the laser communication link. The multimodal data includes at least one or more of the following: signal attenuation value, communication distance, operating wavelength, temperature, visibility, and turbulence intensity. S2. Construct graph structure data by using sampling points as graph nodes and determining edge weights based on environmental parameter similarity. S3. Input the graph structure data into the graph neural network to perform feature fusion and obtain a joint feature vector; S4. Input the joint feature vector into a clustering model that includes an environmental variance compensation term to obtain the attenuation density distribution; S5. Adjust the defocus of the transmitter and the focal length of the receiver based on the attenuation density distribution and the feedback closed loop of the received power to ensure that the received power is not lower than the preset threshold P_th and to suppress the cat's eye effect.

[0005] Preferably, the edge weights are calculated using a Gaussian kernel function. , Where e i and ej σ is the environmental parameter vector for the corresponding node, and σ is the scale parameter adaptively determined based on the variance of the environmental parameters.

[0006] Preferably, the graph neural network includes at least two graph convolutional layers with hidden dimensions in the range of 32 to 256, uses ReLU activation, and is trained with a joint loss consisting of graph Laplacian regularization and mean squared error.

[0007] Preferably, the clustering model is a K-means model that incorporates an environmental parameter variance compensation term λ, and its optimization objective is: , in λ represents the variance measure of the intra-cluster environmental parameters, where λ is a non-negative real number.

[0008] Preferably, the multimodal data is synchronized and denoised via timestamp alignment and Kalman filtering before being input into the graph neural network, thereby reducing multi-sensor clock skew and measurement noise.

[0009] Preferably, the defocusing amount Δ is a monotonic function of the attenuation density distribution intensity Δ= f ( p )calculate, f It is one of linear, piecewise linear, or lookup table mapping, and sets the rate of change and amplitude limit; the focal length is adjusted by a PID controller with a sampling period Ts, the controller has a limiting and anti-integral saturation strategy, and uses the received power as the feedback quantity to achieve steady-state control, so that the received power is kept not lower than the threshold Pth.

[0010] A laser communication signal attenuation prediction and stabilization system based on multimodal data fusion includes an environmental perception module, a signal acquisition module, a fusion processing module, a cluster analysis module, and a control execution module.

[0011] Preferably, it includes an environmental sensing module for collecting at least one of temperature, visibility, and turbulence intensity; The signal acquisition module is used to acquire signal attenuation value, communication distance and operating wavelength; The fusion processing module is used to construct graph-structured data and run a graph neural network to obtain joint feature vectors; The clustering analysis module is used to run a clustering model that includes an environmental variance compensation term to output the attenuation density distribution; The control execution module is used to adjust the defocusing amount of the transmitter and the focal length of the receiver based on the attenuation density distribution and the received power feedback so that the received power is not lower than the threshold Pth.

[0012] This invention provides a method and system for predicting and stabilizing laser communication signal attenuation based on multimodal data fusion, which has the following advantages: This multimodal data fusion-based laser communication signal attenuation prediction and stabilization method and system achieves high-precision prediction and adaptive optimization of laser communication signal attenuation through four key steps: synchronous acquisition of multi-source data, graph neural network (GNN) feature fusion, optimized K-means clustering, and dynamic control of the optical tracking matrix. This improves prediction accuracy, reduces errors in low visibility and atmospheric turbulence scenarios, and extends communication distance. While ensuring signal strength, it achieves the longest possible communication distance. Furthermore, through the dynamic control matrix, signal strength fluctuations are reduced by 60%. The system also has strong real-time performance and compatibility; the hardware only requires the addition of a low-cost sensor (such as a temperature and humidity module) to be compatible with existing inverse modulation systems. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall system structure and data flow (including environmental perception, signal acquisition, fusion processing, and control layer) of the present invention. Figure 2 This is a schematic diagram of the GNN feature fusion process of the present invention; Figure 3 This is a schematic diagram showing the comparison of the optimized K-means clustering effect of the present invention (showing the reduction in inter-class spacing when K=3); Figure 4 This is a bar chart comparing the prediction errors of the present invention in different scenarios; Figure 5 This is a flowchart of the method of the present invention. Detailed Implementation

[0014] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0015] Please see Figures 1 to 4 This invention provides a technical solution: a laser communication signal attenuation prediction and stabilization method based on multimodal data fusion, comprising the following steps: S1. Acquire multimodal data of the laser communication link. The multimodal data includes at least one or more of the following: signal attenuation value, communication distance, operating wavelength, temperature, visibility, and turbulence intensity. S2. Construct graph structure data by using sampling points as graph nodes and determining edge weights based on environmental parameter similarity. S3. Input the graph structure data into the graph neural network to perform feature fusion and obtain a joint feature vector; S4. Input the joint feature vector into a clustering model that includes an environmental variance compensation term to obtain the attenuation density distribution; S5. Adjust the defocus of the transmitter and the focal length of the receiver based on the attenuation density distribution and the feedback closed loop of the received power to ensure that the received power is not lower than the preset threshold P_th and to suppress the cat's eye effect.

[0016] Edge weights are calculated using the Gaussian kernel function. , Where e i and ej σ is the environmental parameter vector for the corresponding node, and σ is the scale parameter adaptively determined based on the variance of the environmental parameters.

[0017] Graph neural networks consist of at least two graph convolutional layers with hidden dimensions ranging from 32 to 256. They employ ReLU activation and are trained using a joint loss consisting of graph Laplacian regularization and mean squared error.

[0018] The clustering model is a K-means model that incorporates an environmental parameter variance compensation term λ, and its optimization objective is: , in λ represents the variance measure of the intra-cluster environmental parameters, where λ is a non-negative real number.

[0019] Multimodal data is synchronized and denoised via timestamp alignment and Kalman filtering before being input into the graph neural network, thereby reducing multi-sensor clock skew and measurement noise.

[0020] The defocusing amount Δ is a monotonic function of the attenuation density distribution intensity Δ= f ( p )calculate, f It can be one of linear, piecewise linear, or lookup table mapping, and the rate of change and amplitude limit are set; the focal length is adjusted by a PID controller with a sampling period Ts. The controller has a limiting and anti-integral saturation strategy, and uses the received power as the feedback quantity to achieve steady-state control so that the received power is not lower than the threshold Pth.

[0021] A multimodal data fusion-based laser communication signal attenuation prediction and stabilization system includes an environmental sensing module, a signal acquisition module, a fusion processing module, a cluster analysis module, and a control execution module. The environmental sensing module is used to collect at least one of temperature, visibility and turbulence intensity; The signal acquisition module is used to acquire signal attenuation values, communication distance, and operating wavelength. The fusion processing module is used to construct graph-structured data and run graph neural networks to obtain joint feature vectors; The clustering analysis module is used to run a clustering model that includes an environmental variance compensation term to output the attenuation density distribution; The control execution module is used to adjust the defocusing amount of the transmitter and the focal length of the receiver based on the attenuation density distribution and the received power feedback so that the received power is not lower than the threshold Pth.

[0022] The specific implementation method is attached. Figure 1 As shown, the hardware system consists of a closed loop comprising an environmental perception layer, a signal acquisition layer, a fusion processing layer, and a control layer.

[0023] In the environmental perception layer, temperature and humidity sensors (SHT35) are installed on both sides of the communication equipment to measure air temperature and humidity in real time. There is also a laser visibility meter that illuminates a 100-meter path and judges visibility by the degree of light scattering, as well as turbulence radar to detect the intensity of atmospheric turbulence.

[0024] At the signal acquisition layer, a high-sensitivity light receiver captures the intensity of the laser signal tens of thousands of times per second, and further automatically calculates the signal attenuation value, which can be obtained according to the light transmission formula:

[0025] Accurate fusion of multi-source data and generation of attenuation maps are performed in the fusion processing layer. This layer receives heterogeneous data streams from the environmental perception layer (temperature, humidity, visibility, turbulence intensity) and the signal acquisition layer, and achieves multimodal fusion through three key technologies: Spatiotemporal alignment: Add a uniform timestamp to all data to eliminate clock skew between sensors.

[0026] Error correction: The Kalman filter algorithm dynamically compensates for spatial location differences.

[0027] Data dimensionality reduction: The signal sampling frequency and environmental data are packaged into a standardized data packet with the following format: timestamp, temperature, visibility, turbulence intensity, attenuation value, communication distance, and wavelength; the processing result is output to a graph neural network (GNN) to generate an attenuation density heatmap along the communication path, providing a decision-making basis for the control layer.

[0028] The optical parameters are dynamically optimized and stabilized in real time at the control layer. Based on the heatmap and real-time signal strength output from the fusion processing layer, this layer achieves millisecond-level response through a dual-module approach. Cat's Eye Suppression Module: The laser emitter is dynamically adjusted according to the formula (defocus Δ(mm) = 0.15 × current attenuation density) to reduce the risk of reflection exposure caused by atmospheric disturbance.

[0029] Focal length stabilization module: Employs a PID controller (proportional coefficient 0.8 / integral coefficient 0.2 / derivative coefficient 0.1) to adjust the receiving lens focal length every 20 milliseconds, aiming to stabilize the signal strength >15 dBm. The two modules work together to form a closed loop.

[0030] like Figure 2 The diagram illustrates the specific implementation of the GNN dynamic feature fusion algorithm. The core of feature fusion is to weave environmental data and signal data into a "relationship network" and then extract key information through an intelligent network. The detailed steps are as follows.

[0031] 1. Pack the data into nodes: Convert the 1,000 signal points collected every millisecond into graph nodes. Each node carries six parameters (temperature, visibility, turbulence intensity, signal attenuation, communication distance, and wavelength), which is equivalent to attaching a complete environmental label to each data point.

[0032] 2. Establishing Relationships: Calculating the similarity between nodes—if the environmental parameters (temperature / visibility / turbulence) of two nodes are similar, they are connected by a "thick line" (weight value close to 1); those with large differences are connected by a "thin line" (weight value close to 0). Furthermore, a Gaussian kernel function is used for automatic calculation; the more similar the environments, the stronger the connection.

[0033] 3. Intelligent Feature Extraction: First, each node collects information from its neighbors (weighted by connection strength), and the resulting data is fused to generate 64 new feature values. These 64 features are then further refined, retaining the most critical information. Finally, the condensed features are concatenated with the original environmental parameters to output a "super feature vector" containing joint signal-environment information. The entire process can be completed with millisecond precision, processing hundreds of thousands of nodes per second. Particularly in practical deployments, when turbulence suddenly intensifies, this module can immediately identify associated nodes and output intensity fluctuation warning features, demonstrating high timeliness.

[0034] like Figure 3 The figure shows the multidimensional data clustering results after processing with the optimized K-means algorithm. The results are clearly presented in a feature space composed of visibility (X-axis), ambient temperature (Y-axis), and laser signal attenuation (color depth represents the Z-axis). Three significantly separated clusters can be observed in the figure: the blue clusters are concentrated in areas with visibility above 15km, suitable temperatures between 20-25°C, and low attenuation values ​​less than 0.3, representing stable transmission; the red clusters are significantly clustered in areas with visibility less than 5km, abnormal temperatures greater than 30°C or less than 10°C, and high attenuation values ​​greater than 0.8, corresponding to channel degradation; the green clusters are in a transitional region, with parameter characteristics between the two. The algorithm significantly reduces the dispersion within each cluster by introducing an environmental parameter variance compensation term (λ) (average reduction of 42% in intra-cluster variance) and improves the cluster center point positioning accuracy by 37%. This optimization enables the system to accurately identify high-risk grid areas (the concentrated area of ​​red clusters), providing a quantitative basis for the early compensation and adjustment of optical parameters.

[0035] like Figure 4As shown, the prediction performance and control effect are quantitatively evaluated and visualized. The image presents the error rates of signal strength fluctuations between the traditional method and the present invention under the same environmental conditions. Measured data shows that under low visibility conditions (visibility less than 3 km), the attenuation prediction error of the present invention is controlled within 8.2%, significantly lower than the 28.5% of the traditional method; under conditions of strong ocean turbulence, the prediction error decreases from 23.4% to 12.3%. Regarding communication distance, while ensuring a received signal strength greater than 15 dBm, the maximum effective communication distance is doubled compared to the traditional scheme. These data fully demonstrate that the present invention has better robustness and reliability compared to traditional methods under complex environmental conditions.

[0036] In summary, the laser communication signal attenuation prediction and stabilization method and system based on multimodal data fusion achieves high-precision prediction and adaptive optimization of laser communication signal attenuation through four key steps: synchronous acquisition of multi-source data, feature fusion of graph neural networks (GNN), optimized K-means clustering, and dynamic control of the optical tracking matrix.

[0037] The first step is to acquire multi-source data synchronously. First, determine the attenuation value (n), which can be measured using a high-sensitivity optical receiver and calculated based on the Beer-Lambert law.

[0038] Where z is the attenuation coefficient, m is the communication distance, v is the wavelength (e.g., 650 nm), and b is the visibility correction factor. Meanwhile, to ensure the capture of rapidly fluctuating signals (such as flicker caused by turbulence), the sampling frequency is set to 1 MHz.

[0039] Next, environmental parameters were collected: for temperature (T) and humidity, digital sensors (such as SHT35) were used with an accuracy of ±0.5℃ and a sampling rate of 10Hz, communicating with the main control unit via an I2C interface; for visibility (Xc), a laser visibility meter was used, and based on the forward scattering principle, the measurement distance was determined to be approximately 100m; for turbulence intensity (ξ), an atmospheric turbulence detection radar was used, outputting the structure constant. The sampling rate is set to 100 Hz.

[0040] Finally, Kalman filtering can be used to fuse multi-sensor data and eliminate position drift errors.

[0041] The second step is feature fusion in a graph neural network (GNN). First, a graph structure is constructed, taking each signal sampling point as a node, with an attribute vector as follows: , Next, based on the similarity of environmental parameters, a Gaussian kernel function is used:

[0042] Then, the GNN model is designed: Input layer: Node features v i + edge weight w ij Image convolutional layer: Layer 1: 64-dimensional hidden layer, ReLU activation function, aggregating neighborhood features:

[0043] Layer 2: 64-dimensional output layer, generating fused features.

[0044] Output layer: Concatenates environmental parameters to obtain the final feature vector.

[0045] Finally, the loss function is used for training and optimization: Mean Squared Error (MSE) + Graph Laplacian Regularization (to prevent overfitting)

[0046] Of the dozens of datasets collected simulating scenarios such as low visibility and atmospheric turbulence in urban areas, 70% will be used for training and 30% for testing.

[0047] The third step is to optimize K-means clustering.

[0048] Compared to the traditional K-means minimization of intra-class distance:

[0049] The optimized version introduces an environmental compensation item:

[0050] The fourth step is dynamic control of the optical tracking matrix.

[0051] Input: Optimized attenuation density Output: Defocus amount Δ: based on the cat's eye effect suppression requirements.

[0052] Focal length P: Dynamically adjusted via a PID controller to maintain a target signal strength >15 dBm.

[0053] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A laser communication signal attenuation prediction and stabilization method based on multimodal data fusion, characterized in that: Includes the following steps: S1. Acquire multimodal data of the laser communication link. The multimodal data includes at least one or more of the following: signal attenuation value, communication distance, operating wavelength, temperature, visibility, and turbulence intensity. S2. Construct graph structure data by using sampling points as graph nodes and determining edge weights based on environmental parameter similarity. S3. Input the graph structure data into the graph neural network to perform feature fusion and obtain a joint feature vector; S4. Input the joint feature vector into a clustering model that includes an environmental variance compensation term to obtain the attenuation density distribution; S5. Adjust the defocus of the transmitter and the focal length of the receiver based on the attenuation density distribution and the feedback closed loop of the received power to ensure that the received power is not lower than the preset threshold P_th and to suppress the cat's eye effect.

2. The laser communication signal attenuation prediction and stabilization method based on multimodal data fusion according to claim 1, characterized in that: The edge weights are calculated using a Gaussian kernel function. , Where e i and ej σ is the environmental parameter vector for the corresponding node, and σ is the scale parameter adaptively determined based on the variance of the environmental parameters.

3. The laser communication signal attenuation prediction and stabilization method based on multimodal data fusion according to claim 1, characterized in that: The graph neural network includes at least two graph convolutional layers with hidden dimensions in the range of 32 to 256. It uses ReLU activation and is trained with a joint loss consisting of graph Laplacian regularization and mean squared error.

4. The laser communication signal attenuation prediction and stabilization method based on multimodal data fusion according to claim 1, characterized in that: The clustering model is a K-means model that incorporates an environmental parameter variance compensation term λ, and its optimization objective is: ， in λ represents the variance measure of the intra-cluster environmental parameters, where λ is a non-negative real number.

5. The laser communication signal attenuation prediction and stabilization method based on multimodal data fusion according to claim 1, characterized in that: The multimodal data is synchronized and denoised via timestamp alignment and Kalman filtering before being input into the graph neural network, thereby reducing multi-sensor clock skew and measurement noise.

6. The laser communication signal attenuation prediction and stabilization method based on multimodal data fusion according to claim 1, characterized in that: The defocusing amount Δ is a monotonic function of the attenuation density distribution intensity Δ= f ( ρ )calculate, f It is one of linear, piecewise linear, or lookup table mapping, and sets the rate of change and amplitude limit; the focal length is adjusted by a PID controller with a sampling period Ts, the controller has a limiting and anti-integral saturation strategy, and uses the received power as the feedback quantity to achieve steady-state control, so that the received power is kept not lower than the threshold Pth.

7. A laser communication signal attenuation prediction and stabilization system for implementing the multimodal data fusion laser communication signal attenuation prediction and stabilization method according to any one of claims 1 to 6, characterized in that: It includes an environmental perception module, a signal acquisition module, a fusion processing module, a cluster analysis module, and a control execution module.

8. The laser communication signal attenuation prediction and stabilization system based on multimodal data fusion according to claim 7, characterized in that: The environmental sensing module is used to collect at least one of temperature, visibility, and turbulence intensity; The signal acquisition module is used to acquire signal attenuation value, communication distance and operating wavelength; The fusion processing module is used to construct graph-structured data and run a graph neural network to obtain joint feature vectors; The clustering analysis module is used to run a clustering model that includes an environmental variance compensation term to output the attenuation density distribution; The control execution module is used to adjust the defocusing amount of the transmitter and the focal length of the receiver based on the attenuation density distribution and the received power feedback, so that the received power is not lower than the threshold Pth.

Images