Method and equipment for improving signal-to-noise ratio of low earth orbit satellite signal

Abstract

The invention relates to the technical field of low earth orbit satellites, and discloses a method and equipment for improving the signal-to-noise ratio of a low earth orbit satellite signal, and the method comprises the steps: carrying out the multi-dimensional signal collection and processing of a received signal of a low earth orbit satellite downlink, and obtaining a signal tensor model; executing structured decomposition based on the signal tensor model to obtain a channel factor matrix group; performing multi-stage sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; inputting the low-orbit satellite channel state information into an interference identification module for feature extraction and interference type analysis to obtain an interference type and probability distribution thereof; performing multi-dimensional adaptive filtering on the received signal to obtain a target signal; and performing modulation coding and beam forming parameter joint optimization on the target signal and the low-earth-orbit satellite channel state information to obtain a low-earth-orbit satellite communication link, thereby solving the problem of large fluctuation of the signal-to-noise ratio in a satellite rapid moving scene, and realizing balance of power efficiency and anti-interference performance at the same time.

Description

Technical Field

[0001] The present invention relates to the field of low-orbit satellite technology, and in particular to a method and device for improving the signal-to-noise ratio of a low-orbit satellite signal. Background Art

[0002] Low-Earth Orbit (LEO) satellites typically operate at an altitude of 500-1200 km. Compared to geosynchronous satellites, they offer advantages such as low transmission latency and wide coverage. However, they also face technical challenges such as rapidly time-varying channels, significant Doppler effects, and complex multipath propagation. Existing channel estimation methods, most of which are derived from terrestrial or geosynchronous satellite communication systems, often assume relatively stable or slowly varying channels. When applied to LEO satellite scenarios, these methods struggle to accurately capture channel characteristics in highly dynamic environments, resulting in a significant degradation of the signal-to-noise ratio (SNR), limiting the capacity and reliability of communication systems.

[0003] Low-orbit satellite communication systems also face the impact of multiple complex interference sources, including ground station interference, co-frequency satellite interference, and atmospheric scintillation. These interferences exhibit highly non-stationary characteristics in a fast-moving satellite environment. Traditional interference suppression technologies rely heavily on hardware filtering and fixed algorithm recognition, lacking intelligent adaptive mechanisms. Recognition accuracy plummets under Doppler shift and low signal-to-noise ratio conditions, making it impossible to effectively distinguish valid signals from interference sources, further deteriorating signal quality. Furthermore, existing interference suppression methods are typically designed for a single type of interference, with limited adaptability to complex interference environments. Furthermore, the suppression process often introduces additional signal distortion. Summary of the Invention

[0004] The present invention provides a method and device for improving the signal-to-noise ratio of low-orbit satellite signals. The present invention solves the problem of large fluctuations in the signal-to-noise ratio in scenarios where satellites move rapidly, while achieving a balance between power efficiency and anti-interference performance.

[0005] In a first aspect, the present invention provides a method for improving the signal-to-noise ratio of a low-orbit satellite signal, the method comprising: Perform multi-dimensional signal acquisition and processing on the received signals of the low-orbit satellite downlink to obtain a signal tensor model; performing structured decomposition based on the signal tensor model to obtain a channel factor matrix group; Performing multi-level sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; Inputting the low-orbit satellite channel state information into an interference identification module for feature extraction and interference type analysis to obtain the interference type and its probability distribution; Performing multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain a target signal; Modulation coding and beamforming parameters are jointly optimized for the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link.

[0006] In a second aspect, the present invention provides a low-orbit satellite signal signal-to-noise ratio improvement device, the low-orbit satellite signal signal-to-noise ratio improvement device comprising: The signal acquisition module is used to perform multi-dimensional signal acquisition and processing on the received signals of the low-orbit satellite downlink to obtain a signal tensor model; a structured decomposition module, configured to perform structured decomposition based on the signal tensor model to obtain a channel factor matrix group; a processing module, configured to perform multi-stage sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; A type analysis module is used to input the low-orbit satellite channel state information into an interference identification module for feature extraction and interference type analysis to obtain the interference type and its probability distribution; An adaptive filtering module, configured to perform multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain a target signal; The adaptive filtering module is used to perform modulation coding and beamforming parameter adaptive filtering on the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link.

[0007] The technical solution provided by the present invention constructs a five-dimensional signal tensor model that can more comprehensively capture the channel characteristics of low-orbit satellites, compared to existing three-dimensional models. In particular, it provides a more accurate description of the Doppler effect and polarization characteristics, providing a more complete information foundation for subsequent signal processing. A fifth-order Vandermonde structured CP decomposition algorithm is employed, and an adaptive step-size adjustment mechanism effectively addresses the estimation bias problem of traditional algorithms in scenarios with rapidly moving satellites, improving the accuracy of channel parameter estimation. A sequential three-stage processing strategy is employed, significantly reducing the real-time processing burden while ensuring parameter estimation accuracy. An interference identification method based on supervised contrastive learning addresses the problem of drastic performance degradation in existing identification algorithms under low signal-to-noise ratio and Doppler shift environments, maintaining high identification accuracy even under extreme conditions. Multi-dimensional adaptive filtering technology dynamically adjusts the suppression strategy based on the interference type and characteristics identified in real time, offering greater adaptability than traditional fixed-mode filtering and effectively addressing complex and changing interference environments. The present invention's joint optimization framework for modulation coding and beamforming, by introducing a dynamic spectrum sensing mechanism and orbit prediction model, addresses the challenge of large signal-to-noise ratio fluctuations in scenarios with rapidly moving satellites, while achieving a balance between power efficiency and anti-interference performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 Schematic diagram of the steps of a method for improving the signal-to-noise ratio of a low-orbit satellite signal according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of a low-orbit satellite signal noise ratio improvement device in an embodiment of the present invention. DETAILED DESCRIPTION

[0010] An embodiment of the present invention provides a method and apparatus for improving the signal-to-noise ratio of a low-orbit satellite signal. The terms "first," "second," "third," "fourth," and so on (if any) in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0011] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 An embodiment of a method for improving the signal-to-noise ratio of a low-orbit satellite signal in an embodiment of the present invention includes: Step S1: performing multi-dimensional signal acquisition and processing on the received signal of the low-orbit satellite downlink to obtain a signal tensor model; It is understandable that the execution subject of the present invention may be a low-orbit satellite signal noise ratio improvement device, or a terminal or a server, and the specific implementation is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0012] Specifically, a multi-channel receiving array is used to parallelly acquire time-domain waveform data of downlink signals from low-orbit satellites. Multiple antenna elements within the array simultaneously collect raw signal data from their respective channels. The collected raw signals are converted to digital signals via a high-precision analog-to-digital conversion unit and then input into a preprocessing module. This preprocessing process includes DC component removal, bandpass filtering, and noise whitening. This process initially suppresses ambient noise and system spurious interference while preserving the core signal information, resulting in a preprocessed signal. The preprocessed signal is reconstructed along the time, frequency, spatial, Doppler, and polarization dimensions to construct an initial five-dimensional signal structure. A sliding window technique is used to segment the five-dimensional signal structure. By setting appropriate window lengths and window overlap ratios, the signal stream is continuously segmented into continuous signal segments. Furthermore, to account for Doppler variations and sudden changes in signal characteristics caused by the high-speed operation of low-orbit satellites, the sliding window sampling rate needs to be dynamically adjusted as the satellite's relative velocity changes. An adaptive sampling rate adjustment mechanism optimizes the sampling interval within each window, resulting in signal data with optimized sampling intervals. The signal data after sampling interval optimization is normalized. By standardizing parameters such as signal amplitude and power in each dimension, the data magnitude differences caused by different physical dimensions and channel conditions are eliminated, and a signal tensor model is obtained.

[0013] Step S2: performing structured decomposition based on the signal tensor model to obtain a channel factor matrix group; Specifically, the normalized five-dimensional signal tensor model is represented by a canonical aggregation. Using the CP decomposition method from tensor decomposition theory, the high-order tensor is reconstructed as a linear superposition of multiple rank-1 tensors. Within this decomposition framework, the signal tensor is approximated as the sum of R rank-1 tensors, each composed of the outer products of vectors in five different physical dimensions. These factor vectors correspond to key characteristics such as time, frequency, space, Doppler, and polarization. To better adapt to the physical propagation characteristics of low-orbit satellite signals, all factor vectors are constructed as complex exponential sequences with a Vandermonde structure. This structure conforms to the array manifold properties commonly used in array signal processing and facilitates the subsequent analytical extraction of parameters such as direction and delay. A high-order singular value decomposition (SSD) is performed on the factor vector model to obtain the principal subspace structure of the signal in the five physical dimensions. The optimal decomposition rank R is determined based on the energy distribution of the decomposed tensor. This process automatically selects the principal components of the signal in complex noise and multipath interference environments, thereby improving the convergence and resolution of subsequent processing. Based on the initial subspace structure and rank R, the algorithm incorporates an optimization strategy combining accelerated gradient descent and line search to efficiently numerically optimize the initial factor matrix. Through iterative computation, the algorithm continuously searches for the optimal direction within the global parameter space. At each step, the step size is adaptively adjusted using a line search method to improve decomposition speed and ensure global optimality. During the optimization process, an adaptive step-size adjustment mechanism is introduced for the factor matrix. This mechanism automatically adjusts the iteration step size based on the signal change rate and error gradient, increasing the step size to improve tracking capability when the signal changes dramatically and reducing it as the parameters converge to ensure the accuracy and stability of the final solution. The entire decomposition and optimization process is accompanied by a residual energy monitoring mechanism. The energy of the decomposition residual is calculated in real time during each iteration. Convergence is considered achieved when the rate of change of the residual drops below a set threshold after several consecutive iterations. This avoids unnecessary over-computation and improves efficiency in practical engineering operations. The resulting channel factor matrix contains parameters for time, frequency, space, Doppler, and polarization dimensions.

[0014] Step S3: performing multi-level sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; Specifically, the system deeply integrates the spatial and polarization-dimensional factor matrices within the channel factor matrix group. By constructing an augmented spatial-polarization covariance matrix, the statistical properties between different antenna elements and polarization components are integrated to obtain a joint spatial-polarization feature representation. This representation reflects the energy distribution and correlation of the LEO satellite channel across multiple physical attributes, including azimuth, elevation, and polarization. This joint spatial-polarization feature representation is then subjected to eigendecomposition and noise subspace projection. Using the noise subspace projection method, combined with high-resolution algorithms such as multiple signal classification, high-precision estimation of the signal arrival angle (azimuth and elevation) and polarization state parameters is achieved. This step effectively distinguishes multipath components from true signals, providing the system with spatial positioning and polarization awareness capabilities. Furthermore, a joint time-frequency analysis method is employed for the frequency and Doppler-dimensional factor matrices within the channel factor matrix group to simultaneously model the multipath structure and Doppler effect in the frequency domain. By inversely transforming the frequency-dimensional factor matrix, the impulse response characteristics of each multipath component in the time domain are obtained, and then the multipath delay parameters are accurately extracted using techniques such as peak detection. At the same time, the Doppler frequency shift is resolved using a spectral estimation algorithm for the Doppler-dimensional factor matrix, thereby capturing the frequency shift patterns of the signal in high-speed motion environments. Considering the time-varying characteristics of the channel, a recursive least squares filtering process is introduced for the time-dimensional factor matrix. By performing real-time recursive estimation of the channel response coefficients at each time sampling point, the amplitude and phase fluctuations of the channel over time are dynamically tracked, effectively improving the system's adaptability and robustness in rapidly changing channel environments. The signal arrival angle, polarization state parameters, multipath delay, Doppler frequency shift, and time-varying channel coefficients obtained from each of the above steps are fused and comprehensively organized to form low-orbit satellite channel state information containing subcarriers, time sampling points, and receiving antenna dimensions.

[0015] Step S4: Input the low-orbit satellite channel state information into the interference identification module for feature extraction and interference type analysis to obtain the interference type and its probability distribution; Specifically, the received signal and low-orbit satellite channel state information are subjected to time-frequency transformation. Using techniques such as the short-time Fourier transform (SFT), the original time-domain signal is converted into a two-dimensional time-spectrogram matrix, reflecting the distribution of signal energy in both time and frequency dimensions and revealing the unique manifestations of different interferences in the time and frequency domains. The time-spectrogram matrix is ​​input into the residual network structure of the interference identification module for feature extraction. Through the synergistic effect of multi-layer convolution and skip connections, the network effectively extracts the complex time-frequency patterns and local detail features contained in the signal, automatically filters out noise and irrelevant background, and outputs a primary signal feature map. The primary signal feature map is compressed using the global average pooling layer of the interference identification module, mapping the two-dimensional or multi-dimensional feature map into a one-dimensional feature vector. The one-dimensional feature vector is then input into the multi-head self-attention layer of the interference identification module for attention calculation. The multi-head self-attention layer reevaluates the dependencies and relative importance of each component within the feature vector from a global perspective. By simultaneously focusing on different attributes and subtle changes in the signal through multiple parallel attention heads, the model improves its ability to discern subtle differences in interference types in low signal-to-noise ratio and mixed interference environments, thereby generating an enhanced feature vector. The enhanced feature vector is then identified and mapped into a high-dimensional space by a fully connected classification layer. Each output node corresponds to a specific interference type, ultimately outputting a raw score for each type of interference. Softmax normalization is performed on the raw scores to construct a probability distribution for multiple typical interference types, including ground station interference, co-frequency satellite interference, atmospheric scintillation interference, multipath interference, and thermal noise. Each probability value reflects the likelihood of the corresponding type of interference occurring in the current received signal.

[0016] Step S5: performing multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain the target signal; Specifically, the interference type and its probability distribution output by the previous interference identification link are used as the decision basis. Combined with the system's built-in interference suppression decision rules, a weight matrix containing spatial suppression weights, frequency domain suppression weights, and time domain suppression weights is dynamically constructed. This weight matrix is ​​intelligently adjusted based on the actual impact of different interference types and their probability distribution in the current signal environment, ensuring that the main manifestation dimensions of each type of interference receive sufficient suppression weights. For example, when the probability of interference from ground stations is high, the spatial suppression weight is increased accordingly to highlight the spatial selectivity of beamforming; when multipath interference or narrowband interference dominates, the frequency domain suppression weight is significantly enhanced to highlight the role of notch or band-stop filtering. Based on the above weight matrix and the specific interference probability distribution, a comprehensive suppression weight vector is calculated to determine the relative proportion of subsequent multi-dimensional filtering results in the final target signal fusion, enabling flexible allocation of filtering strategies under different interference scenarios. Based on this, the minimum variance distortionless response beamforming algorithm processes the received signal, dynamically adjusting the beam direction and response weights to maximize the suppression of spurious signals from the interference source while maintaining distortion-free reception of the target satellite signal, achieving a filtered output in the spatial dimension. Simultaneously, adaptive notch filtering and band-stop filtering are performed on the received signal, based on the frequency domain suppression weights. For typical scenarios such as co-channel satellite interference, multipath reflections, and narrowband terrestrial interference, the interference frequency and bandwidth are automatically detected. By adjusting the notch center and bandwidth, harmful interference signals in the target frequency band are suppressed, effectively improving the frequency domain purity of the primary signal. In the time domain, adaptive Kalman filtering is employed, based on the time domain suppression weights, to track and estimate the statistical characteristics and noise state of the signal in real time, continuously optimizing the filter parameters. This allows for targeted dynamic suppression of random pulses, time-varying noise, or non-stationary interference in the signal, enhancing the time domain smoothness and stability of the primary signal. After obtaining the filtering results of the spatial dimension, frequency domain dimension and time domain dimension, weighted fusion is performed on these three filtering results based on the aforementioned comprehensive suppression weight vector. Through a linear or nonlinear combination mechanism, the filtering advantages of different dimensions are effectively integrated to obtain the target signal.

[0017] Step S6: Jointly optimize the modulation coding and beamforming parameters of the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link.

[0018] Specifically, a two-layer dynamic spectrum sensing process is performed based on the target signal. By performing wideband energy detection on the target signal across a wide bandwidth, a comprehensive assessment of the energy distribution and signal occupancy of different frequency resources is achieved. Simultaneously, high-resolution, fine-scale feature extraction is performed on the detected available frequency bands to accurately calibrate the availability of each subband and identify weak interference, residual signals, and sudden energy changes within the band, resulting in a spectrum sensing result that combines global coverage with local details. The spectrum sensing results, synchronously acquired low-orbit satellite channel state information, and the filtered target signal are input into a LSTM-CNN hybrid neural network model. Combining the temporal memory properties of LSTM with the local feature extraction capabilities of CNN, this model enables in-depth analysis of the dynamic changes in the communication link environment, automatically capturing key physical parameters such as time-varying channel gain fluctuations, multipath delay spread, Doppler shift variations, and residual interference, and outputting a highly integrated and richly structured channel state vector. After obtaining the channel state vector, the system constructs an intelligent modulation and coding selection strategy based on the principle of link adaptation. This strategy establishes a state-action mapping and comprehensively analyzes the current channel state, signal-to-noise ratio (SNR), interference environment, and target communication quality requirements. It dynamically selects the optimal solution from a selection of available modulation order and coding rate combinations, achieving a multi-objective balance between data rate, robustness, and energy consumption. For example, when channel quality is good and interference is low, the system prioritizes high-order modulation and high-code rate combinations to improve link throughput. In poor channel conditions or under strong interference, it adaptively switches to low-order modulation and low-code rate combinations to ensure communication reliability. To overcome the Doppler shift caused by the high-speed motion of low-orbit satellites, the relative motion parameters between the satellite and the ground station are calculated in real time based on a satellite orbit prediction model. This calculation is combined with the Doppler shift information extracted from the current channel state vector to generate a feedforward compensation value. This compensation mechanism effectively offsets the frequency drift caused by the Doppler effect, improving signal demodulation accuracy and synchronization performance, thereby ensuring stable link quality under high-order modulation modes. The channel state vector, optimal modulation order, and coding rate combination are then input into a multi-objective genetic algorithm optimization module. By simulating the crossover, mutation, and survival of the fittest mechanisms in biological evolution, the algorithm conducts a global search and multi-objective trade-off on beamforming parameters, balancing signal-to-noise ratio maximization with power efficiency improvement. During each generation of evolution, the algorithm continuously evaluates the comprehensive performance of candidate solutions in multiple dimensions, including anti-interference capability, signal coverage, and energy consumption, ultimately converging on a set of beam weights, azimuths, and beamwidth parameters that combine high signal-to-noise ratio with high energy efficiency, achieving optimal matching of physical layer resource allocation. The optimized beamforming parameters, Doppler shift feedforward compensation values, and adaptively selected modulation order and coding rate are applied to the actual link processing of the target signal, achieving synchronous parameter configuration of the satellite transmitter and receiver, and obtaining a low-orbit satellite communication link for improving the signal-to-noise ratio.

[0019] In embodiments of the present invention, a five-dimensional signal tensor model is constructed, which can more comprehensively capture the channel characteristics of low-orbit satellites than existing three-dimensional models. In particular, it provides a more accurate description of the Doppler effect and polarization characteristics, providing a more complete information foundation for subsequent signal processing. A fifth-order Vandermonde structured CP decomposition algorithm, with an adaptive step-size adjustment mechanism, effectively addresses the estimation bias problem of traditional algorithms in scenarios with rapidly moving satellites, improving the accuracy of channel parameter estimation. A sequential three-stage processing strategy significantly reduces the real-time processing burden while ensuring parameter estimation accuracy. An interference identification method based on supervised contrastive learning addresses the sharp performance degradation of existing identification algorithms in low signal-to-noise ratio and Doppler shift environments, maintaining high identification accuracy even under extreme conditions. Multi-dimensional adaptive filtering technology dynamically adjusts the suppression strategy based on the real-time identified interference type and characteristics, offering greater adaptability than traditional fixed-mode filtering and effectively addressing complex and changing interference environments. The present invention's joint optimization framework for modulation coding and beamforming, by introducing a dynamic spectrum sensing mechanism and orbit prediction model, addresses the significant signal-to-noise ratio fluctuations in scenarios with rapidly moving satellites, while achieving a balance between power efficiency and anti-interference performance.

[0020] In a specific embodiment, the process of executing step S1 may specifically include the following steps: The time domain waveform data of the received signal of the low-orbit satellite downlink is collected through a multi-channel receiving array to obtain the original signal data; Perform analog-to-digital conversion and preprocessing on the original signal data to obtain a preprocessed signal, and reconstruct the preprocessed signal according to the time dimension, frequency dimension, space dimension, Doppler dimension and polarization dimension to obtain an initial five-dimensional signal structure; The sliding window technique is used to segment the data based on the initial five-dimensional signal structure to obtain continuous signal segments. The sampling rate of the continuous signal segments is adaptively adjusted according to the relative speed of the satellite to obtain signal data with optimized sampling intervals. The signal data after sampling interval optimization is normalized to obtain a signal tensor model.

[0021] Specifically, a multi-channel receiving array system is constructed, consisting of multiple spatially distributed antenna elements and a high-speed parallel sampling front-end receiver. When the satellite downlink signal reaches the ground terminal, it is independently collected by each array antenna, generating parallel multi-channel time-domain raw waveform data. The signal collected by each channel contains physical information such as minute phase delays, multipath superposition, and polarization state changes caused by the high-speed motion of low-orbit satellites. As the satellite moves at high speed in orbit, the signal sequences received by each antenna element in the array are synchronously fed into the analog-to-digital conversion unit. Using a high-resolution, high-sampling-rate ADC, the continuously changing analog signal is converted into a discrete digital signal. The raw digital signal data is then input into a preprocessing module, which includes DC component removal, bandpass filtering, anti-aliasing, gain equalization, and preliminary noise whitening. By removing the DC component, signal distortion caused by equipment baseline drift during subsequent analysis is avoided; bandpass filtering ensures that only useful signal components within the target frequency band are retained, while effectively suppressing the system's out-of-band noise and non-target interference; anti-aliasing processing and gain equalization ensure the consistency of all signal channels in amplitude and spectrum, eliminating the impact introduced by channel imbalance for subsequent joint modeling; the noise whitening algorithm uses statistical characteristics to break up highly correlated background noise, making it appear as approximately Gaussian white noise, improving the robustness of the system's subsequent adaptive decomposition and parameter estimation algorithms in low signal-to-noise ratio environments, and enhancing the separability of signals and noise. The preprocessed signal stream is imported into the structure reconstruction unit, and based on the preset five-dimensional signal model framework, the data is classified and stacked according to the five physical dimensions of time, frequency, space, Doppler, and polarization. The time dimension captures temporal variations in the signal caused by satellite orbits, Earth's rotation, and other factors. The frequency dimension reflects the signal's energy distribution and bandwidth utilization within each subcarrier or frequency band. The spatial dimension carries observation data from each element of the antenna array, reflecting the array's gain and spatial filtering characteristics. The Doppler dimension characterizes frequency shifts caused by the high-speed relative motion between the satellite and the ground, a phenomenon particularly pronounced in low-orbit satellite communications. The polarization dimension encompasses signal behavior under different polarization modes (such as horizontal, vertical, or circular polarization). Through reconstruction and encoding of these five dimensions, all raw signal data is effectively organized into an initial five-dimensional signal structure. A sliding window technique is used to segment the data based on this initial five-dimensional signal structure. By setting the window length and window overlap ratio for the signal data, a number of consecutive signal segments are sequentially separated from the signal sequence. The real-time relative velocity of the satellite in orbit can cause drastic changes in signal characteristics. An adaptive sampling rate adjustment mechanism is employed to flexibly adjust the data sampling interval within each sliding window based on the real-time velocity between the ground and satellite and the dynamic Doppler estimation results. When the satellite speed is high, the signal changes intensify and the sampling interval is shortened accordingly to capture sufficient details; when the satellite speed is low and the changes tend to be stable, the sampling interval is appropriately lengthened to reduce redundant sampling and system burden.After completing adaptive sampling optimization, normalization is performed on all signal data after sampling interval optimization, including amplitude normalization, power normalization, and high-order statistics normalization. Amplitude normalization ensures that the maximum signal values ​​of all channels are within the same range, preventing extreme amplitude values ​​of a single channel from dominating the overall processing. Power normalization adjusts the energy of a single frame or window based on the average power level of all channels and all windows to ensure uniform energy distribution across all dimensions. High-order statistics normalization uniformly adjusts signal parameters such as mean, variance, kurtosis, and skewness to account for complex non-Gaussian effects such as environmental noise and device drift in practice. This results in a signal tensor model.

[0022] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Perform canonical aggregation representation on the signal tensor model to obtain the sum form of R rank-1 tensors, and establish a factor vector model with Vandermonde structure based on the sum form of rank-1 tensors; Perform high-order singular value decomposition on the factor vector model to obtain the initial subspace structure and the rank R value of the decomposition tensor; Based on the initial subspace structure and rank R value, the accelerated gradient descent and line search algorithms are used for optimization to obtain the initial factor matrix; Adaptively adjust the step size of the initial factor matrix to obtain the optimized factor matrix under the rapid satellite movement environment; Residual energy monitoring is performed on the optimized factor matrix to obtain a channel factor matrix group including time dimension parameters, frequency dimension parameters, space dimension parameters, Doppler dimension parameters and polarization dimension parameters.

[0023] Specifically, a canonical aggregation process is performed on the five-dimensional signal tensor model. A highly complex five-dimensional signal tensor is reformulated as the sum of several groups of fundamental low-rank structures. Each fundamental structure is understood as a typical behavioral pattern in the five physical dimensions of time, frequency, space, Doppler, and polarization. This decomposes the signal tensor into a set of factor vectors with clear physical meaning. To accommodate the high-speed dynamic environment of low-orbit satellites and the propagation patterns of spatial array signals, each factor vector adopts a Vandermonde structure. That is, the factor vectors in different dimensions exhibit a recursive relationship or exponential growth structure, which is highly consistent with the characteristics of actual satellite signals in terms of array, frequency drift, and Doppler spread. High-order singular value decomposition is then performed on this factor vector model. This high-order singular value decomposition automatically identifies the main directions and main features of the signal in each physical dimension and effectively distinguishes useful signal components from background noise. This process also automatically estimates the most suitable decomposition rank, that is, the number of fundamental structural units required to fully describe the original signal tensor. After extracting the initial principal subspace and determining its rank, the system numerically optimizes all factor matrices based on this initial structure using accelerated gradient descent and line search algorithms. Accelerated gradient descent enables faster parameter convergence and effectively avoids the local minima often encountered by conventional optimization algorithms. Line search automatically selects the optimal step size for each parameter update, ensuring that each step maximizes decomposition accuracy. Due to the significant dynamics of signals in low-orbit satellite communications, the system dynamically adjusts the parameter update step size during the optimization iteration process. The adaptive step size adjustment strategy flexibly increases or decreases the parameter change amplitude at each step based on the current signal change rate and optimization convergence trend. For example, when satellite motion increases and signal parameters fluctuate rapidly, the system automatically increases the step size to improve tracking capability. However, when the signal stabilizes or approaches the optimal decomposition state, the step size is reduced to ensure convergence accuracy of the final solution. The entire decomposition and optimization process is accompanied by a residual energy monitoring mechanism. After each parameter iteration, the system calculates the fitting error between the current decomposition model and the original signal and continuously monitors the trend of the residual energy. When the residual energy changes very little over multiple consecutive iterations, it indicates that the decomposition has reached convergence conditions. At this time, the operation is automatically terminated early to avoid invalid repeated calculations, and a channel factor matrix group containing time dimension parameters, frequency dimension parameters, space dimension parameters, Doppler dimension parameters, and polarization dimension parameters is obtained.

[0024] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Performing augmented spatial-polarization covariance analysis based on the spatial dimension factor matrix and the polarization dimension factor matrix in the channel factor matrix group to obtain a spatial-polarization joint feature representation; Perform feature decomposition and noise subspace projection on the spatial polarization joint feature representation to obtain the signal arrival angle and polarization state parameters; Based on the frequency dimension factor matrix and the Doppler dimension factor matrix in the channel factor matrix group, a joint time-frequency analysis is performed to obtain multipath delay and Doppler frequency shift parameters; Performing recursive least square filtering on the time dimension factor matrix in the channel factor matrix group to obtain time-varying channel coefficients; The signal arrival angle, polarization state parameters, multipath delay, Doppler frequency shift parameters and time-varying channel coefficients are integrated to obtain the low-orbit satellite channel state information including subcarriers, time sampling points and receiving antenna dimensions.

[0025] Specifically, augmented spatial-polarization covariance analysis is performed based on the spatial and polarization dimension factor matrices in the channel factor matrix group to obtain a joint spatial-polarization signature. This signature reflects the signal energy distribution and phase relationship between different antenna elements and reveals the energy distribution characteristics of the signal under different polarization states. This joint analysis enables the comprehensive utilization of spatial array gain and polarization diversity. Furthermore, eigendecomposition techniques are used to process the joint spatial-polarization signature. Through eigendecomposition of the matrix, the signal and noise spatial components are effectively distinguished, and the main direction information that best represents the signal characteristics is obtained. Using the noise subspace projection method, the orthogonal relationship between candidate spatial-polarization steering vectors and the noise subspace is measured to accurately extract parameters such as the signal angle of arrival and polarization state. This method is suitable for distinguishing the main signal from interference and spurious signals in multipath environments, and can improve the resolution and accuracy of parameter estimation. Extracting the signal angle of arrival helps identify the geometric relationship between the target satellite and the ground station, while obtaining the polarization state provides a direct physical basis for signal demodulation and polarization decorrelation processing. To recover the multipath delay and Doppler shift characteristics of the channel, the frequency- and Doppler-dimensional factor matrices are jointly analyzed. Due to the high-speed relative motion between the satellite and the ground terminal and the complex signal propagation path, the signal is affected by multipath and Doppler effects. Joint time-frequency analysis utilizes the energy distribution in the frequency-dimensional factor matrix to map the signal to the time domain, identifying the primary energy concentration interval for each path. The dynamic changes in the Doppler-dimensional factor matrix capture and separate the frequency shift characteristics of each path. Combining this information, the signal delay characteristics under different paths and the corresponding Doppler shift parameters for each path are determined. Recursive least-squares filtering is performed on the time-dimensional factor matrix in the channel factor matrix group. This adaptive filtering algorithm dynamically tracks and estimates the time-varying channel coefficients at each time sampling point. The introduction of recursive least-squares filtering eliminates the need for channel state estimation by relying solely on data at a single moment in time. Instead, it integrates historical data with current observations, improving sensitivity and adaptability to various dynamic scenarios, such as sudden channel changes and slow drift. The signal arrival angle, polarization state parameters, multipath delay, Doppler frequency shift parameters and time-varying channel coefficients are integrated to obtain the low-orbit satellite channel state information including subcarriers, time sampling points and receiving antenna dimensions.

[0026] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Perform time-frequency transformation on the received signal and the low-orbit satellite channel state information to obtain a time-frequency spectrum matrix; The time-frequency spectrum matrix is ​​input into the residual network structure of the interference recognition module for feature extraction to obtain the primary signal feature map; The primary signal feature map is compressed through the global average pooling layer of the interference recognition module to obtain a one-dimensional feature vector; The one-dimensional feature vector is input into the multi-head self-attention layer of the interference recognition module to perform attention calculation and obtain the enhanced feature vector; The enhanced feature vector is processed through the fully connected classification layer of the interference identification module to obtain the raw score of the interference type; The original scores are subjected to Softmax normalization to obtain interference types and their probability distributions, including ground station interference, co-frequency satellite interference, atmospheric scintillation interference, multipath interference, and thermal noise.

[0027] Specifically, the satellite received signal and real-time channel state information are jointly input into the time-frequency transformation processing module. Using time-frequency analysis methods such as short-time Fourier transform and continuous wavelet transform, the original time-domain signal is windowed and mapped into the frequency domain, resulting in a time-spectrogram matrix that reflects the dual time and frequency variations of signal energy. This time-spectrogram matrix is ​​then input into the interference identification module's residual network architecture for feature extraction. Coreized by multi-layer convolutional units and feature skip connections, the residual network effectively addresses the dual challenges of weakened signal features and complex interference patterns in low signal-to-noise ratio (SNR) environments by simultaneously extracting local features and global structure at varying network depths. The convolutional layers automatically capture key signal microstructures such as frequency variation patterns, periodicity mutations, bandwidth expansions, and transient energy bursts. The residual connections enable cross-layer perception and smooth gradient propagation, maintaining information integrity and robustness during deep network training. After multiple convolutional and residual blocks, the network outputs a primary signal feature map. This map condenses all high-order structural information of the input time-spectrogram in two dimensions, effectively separating and representing the characteristics of different interference types. The primary signal feature map is input to the global average pooling layer. The original two-dimensional or multi-channel feature map is compressed into a one-dimensional feature vector by taking the mean of all spatial dimensions. This one-dimensional feature vector is then fed into a multi-head self-attention mechanism for feature enhancement. The self-attention mechanism automatically focuses on the most critical local patterns for interference discrimination by calculating the dependencies and importance between positions within the feature vector. This multi-head structure allows the network to model multiple different sets of attention points in parallel within the same feature space, simultaneously focusing on sudden frequency peaks and specific energy distributions while also capturing slowly changing background trends or fine-grained modulation details. The enhanced feature vector is then fed into a fully connected classification layer. The fully connected layer maps the enhanced high-dimensional features to output nodes corresponding to the number of interference categories, assigning a raw score to each category. Each node corresponds to a specific interference type, such as ground station interference, co-frequency satellite interference, atmospheric scintillation interference, multipath interference, and thermal noise. At the classification output stage, the raw scores obtained from the fully connected layer are subjected to softmax normalization. Softmax is a normalization function that converts any real value into a standard probability distribution, such that the sum of all class probabilities is 1. Through normalization, the raw scores are mapped into a practical probability distribution. The probability value of each interference type directly reflects its relative likelihood of occurring in the current signal sample. This results in structured data that includes the probability distributions of multiple interference types, including ground station interference, co-frequency satellite interference, atmospheric scintillation interference, multipath interference, and thermal noise.

[0028] In a specific embodiment, the process of executing step S5 may specifically include the following steps: A weight matrix including spatial suppression weights, frequency domain suppression weights and time domain suppression weights is constructed according to the interference type and its probability distribution; Calculate the comprehensive suppression weight vector based on the weight matrix and the probability distribution of interference types; Based on the spatial suppression weights and the low-orbit satellite channel state information, the minimum variance distortion-free response beamforming algorithm is performed on the received signal to obtain the spatial dimension filtering result; Based on the frequency domain suppression weight, the received signal is subjected to adaptive notch filtering and band-stop filtering to obtain the frequency domain dimension filtering result, and based on the time domain suppression weight, the received signal is subjected to adaptive Kalman filtering to obtain the time domain dimension filtering result; According to the comprehensive suppression weight vector, the spatial dimension filtering results, the frequency domain dimension filtering results and the time domain dimension filtering results are weightedly fused to obtain the target signal.

[0029] Specifically, a weight matrix consisting of spatial, frequency, and time domain suppression weights is constructed based on the interference type and its probability distribution. Each interference type has different impact weights on the signal in the spatial, frequency, and time domains. For example, strong interference from ground stations is primarily in a specific spatial direction, narrowband interference is prominent in the frequency domain, and random noise or multipath aliasing is most significant in the time domain. Therefore, based on the interference classification results, historical experience, and simulation data, a basic suppression weight value is assigned to each interference type, forming a structured representation in the weight matrix. For example, each row of the spatial weight matrix corresponds to an interference type, and the columns represent the contribution to spatial suppression. The frequency and time domain weight matrices respectively represent the suppression capabilities in their respective dimensions. A comprehensive suppression weight vector is calculated based on the weight matrix and the probability distribution of the interference type. The interference probability vector is multiplied by the weight matrix row by row to form a set of comprehensive suppression weight vectors that reflect the actual suppression requirements of the current signal environment. Using the spatial suppression weights and real-time spatial parameters derived from the channel state information, a minimum variance distortionless response beamforming algorithm is performed on the received signal. By analyzing the spatial manifold of satellite signals and the directional distribution of interference sources, the system dynamically adjusts beamforming parameters, aligning the mainlobe with the arrival direction of the target satellite signal while setting deep nulls in the interference direction to achieve spatial suppression of interference. During the beamforming process, the algorithm considers the spatial distribution of the signal and interference and adaptively adjusts spatial weights to maximize signal gain and minimize interference energy. Furthermore, based on frequency-domain suppression weights, adaptive notch filtering and band-stop filtering are used to process received signals for interference types with distinct frequency-domain characteristics, such as co-channel satellite interference and narrowband interference from ground stations. Notch filters automatically identify interference frequencies and achieve frequency-domain suppression, while band-stop filters attenuate energy in target frequency bands based on the detected interference bandwidth. Based on the interference probability and weight, the system dynamically adjusts the filter's center frequency, bandwidth, and attenuation depth, achieving adaptive control of frequency-domain filtering strength. In the time domain, an adaptive Kalman filter is used to process the received signal based on the time-domain suppression weights. The Kalman filter is a recursive optimal filtering method based on state estimation theory that adjusts filter parameters in real time according to signal statistics and noise models. To address issues such as multipath aliasing, burst pulses, and time-varying random interference, the Kalman filter can dynamically adjust the weights of the prediction and correction processes to accurately track and suppress non-stationary interference. During the time-domain filtering process, the Kalman gain and covariance parameters are adjusted in real time with environmental dynamics and interference probability, ensuring that the system can effectively isolate and weaken abnormal interference components in the time domain without losing the main signal information. After completing three independent filtering paths in the spatial, frequency, and time domains, the three filtering results are weighted and fused based on the aforementioned comprehensive suppression weight vector. The contribution of each filter output to the construction of the final target signal is determined by its comprehensive suppression weight.When spatial filtering plays a higher role, it indicates that interference threats from spatial directions dominate, and the system relies more on spatial beamforming results. When frequency or time domain weighting predominates, the filtering results from those dimensions are given greater weight in target signal fusion. The fusion process uses linear weighting or higher-order adaptive fusion mechanisms to complement the three paths, further suppressing residual interference components and maximizing the energy of the main signal. The target signal is then obtained.

[0030] In a specific embodiment, the process of executing step S6 may specifically include the following steps: Perform dual-layer dynamic spectrum sensing processing based on the target signal to obtain spectrum sensing results that include broadband energy distribution and frequency band characteristics; The spectrum sensing results, low-orbit satellite channel state information, and target signal are input into the LSTM-CNN hybrid network for channel state analysis, obtaining a channel state vector containing channel gain, multipath delay spread, Doppler frequency shift, and interference residual. Establishing a modulation and coding selection strategy based on the channel state vector, the modulation and coding selection strategy includes an optimal modulation order and coding rate combination; The Doppler frequency shift parameter in the channel state vector is calculated based on the satellite orbit prediction model to obtain the relative motion parameter and the Doppler frequency shift feedforward compensation value; Based on the channel state vector, optimal modulation order and coding rate combination, a multi-objective genetic algorithm is used to simultaneously optimize the signal-to-noise ratio and power efficiency to obtain the optimized beamforming parameters. The target signal is processed based on the optimized beamforming parameters, Doppler frequency shift feedforward compensation value, and the optimal modulation order and coding rate combination to obtain a low-orbit satellite communication link for improving the signal-to-noise ratio.

[0031] Specifically, a two-layer dynamic spectrum sensing process is performed based on the target signal, generating spectrum sensing results that include broadband energy distribution and frequency band characteristics. The first layer of spectrum sensing scans the entire operating bandwidth covered by the target signal, monitoring the energy distribution, noise floor, and abnormal energy concentration points in each frequency band in real time. This creates a broadband energy distribution view that reflects the overall strength and bandwidth utilization of the current signal, and reveals external strong interference, frequency holes, and changes in link availability in the system. The second layer of spectrum sensing performs refined segmented analysis on this broadband energy distribution map. Within the detected usable frequency band or suspected interference band, the frequency resolution is further improved, and features are extracted for signal energy, modulation characteristics, in-band noise, and weak interference within each small frequency band. The spectrum sensing results are synchronously input into an LSTM-CNN hybrid network structure along with channel state information acquired in real time from the low-orbit satellite link and the latest target signal data. The LSTM (Long Short-Term Memory) network is capable of modeling time series features, capturing the historical dependencies, sudden changes, and gradual trends of satellite channel evolution over time. The CNN (Convolutional Neural Network) focuses on extracting local patterns and spatial distribution patterns in the signal spectrum and spatiotemporal structure. By combining the LSTM unit's modeling of channel temporal dependencies with the CNN unit's in-depth extraction of the spectrum and spatial structure, the system comprehensively analyzes key channel physical parameters such as gain fluctuation, multipath delay spread, Doppler shift, and residual interference in the dynamically changing low-orbit satellite environment. It outputs a multidimensional channel state vector (CSV), reflecting the current channel reliability, interference resistance, and the system's remaining communication capability. Using the CSV as input, the system develops an adaptive modulation and coding selection strategy based on current communication requirements and the actual physical environment. Based on parameters such as signal-to-noise ratio, multipath delay, frequency shift, and residual interference in the CSV, it intelligently analyzes the most appropriate modulation and channel coding scheme for the current link. In environments with high signal-to-noise ratios and low interference, the system automatically recommends high-order modulation and high-code rate combinations to achieve high-throughput communication. In scenarios with degraded channels or increased multipath and interference, it adaptively selects low-order modulation and strong error correction coding to ensure link reliability and stability. This modulation and coding selection strategy not only optimizes the current state but also, through rolling prediction and dynamic updates, automatically adjusts to the evolving satellite orbit, channel, and interference conditions, ensuring the communication link maintains its optimal configuration at all stages of operation. Furthermore, because Doppler shift in low-orbit satellite systems is a critical physical phenomenon affecting signal synchronization and modulation and demodulation performance, the system uses a satellite orbit prediction model to physically model and predict Doppler shift parameters in the channel state vector. By inputting the current satellite orbit parameters, ground station location, and time, combined with the observed frequency shift values, the system calculates the relative motion parameters between the satellite and the ground, predicts the Doppler shift trend for future time periods, and generates feedforward compensation values ​​accordingly.This compensation value is directly applied to frequency calibration at both the transmitter and receiver ends, preemptively offsetting frequency offsets caused by the satellite's high-speed motion. This ensures that the signal remains well synchronized upon reaching the receiver, significantly improving link stability and modulation and demodulation efficiency. Based on the channel state vector, the optimal modulation order, and coding rate combination, a multi-objective genetic algorithm simultaneously optimizes both signal-to-noise ratio (SNR) and power efficiency, comprehensively searching for and optimizing the link's beamforming parameters. With the dual goals of maximizing SNR and improving power efficiency, the multi-objective genetic algorithm dynamically explores the optimal solution for parameters such as beam direction, weight, and beamwidth through an evolutionary mechanism involving population initialization, fitness evaluation, crossover mutation, and survival of the fittest. During this evolutionary process, the system evaluates in real time the impact of different solutions on link performance, energy consumption, and interference mitigation capabilities, ensuring that the output beamforming parameters maintain primary signal strength while minimizing interference and system energy consumption, achieving the optimal balance between high-quality communication and resource conservation. Based on the optimized beamforming parameters, Doppler shift feedforward compensation, and the optimal modulation order and coding rate combination, the target signal is processed in an integrated manner. This process includes beam steering and frequency pre-compensation on the transmitting side, demodulation and error correction parameter configuration on the receiving side, and end-to-end link resource allocation. All processes are dynamically adaptive, rapidly adjusting to real-time link changes and satellite motion, ensuring continuous improvement in the link's signal-to-noise ratio, reduction in communication bit error rates, and overall energy consumption control. Ultimately, a low-orbit satellite communication link with an improved signal-to-noise ratio is achieved.

[0032] The above describes the method for improving the signal-to-noise ratio of a low-orbit satellite signal in an embodiment of the present invention. The following describes the device for improving the signal-to-noise ratio of a low-orbit satellite signal in an embodiment of the present invention. Figure 2 In one embodiment of the present invention, a device for improving the signal-to-noise ratio of a low-orbit satellite signal includes: The signal acquisition module is used to perform multi-dimensional signal acquisition and processing on the received signals of the low-orbit satellite downlink to obtain a signal tensor model; A structured decomposition module, configured to perform structured decomposition based on a signal tensor model to obtain a channel factor matrix group; A processing module, configured to perform multi-level sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; The type analysis module is used to input the low-orbit satellite channel state information into the interference identification module for feature extraction and interference type analysis to obtain the interference type and its probability distribution; Adaptive filtering module, used to perform multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain the target signal; The adaptive filtering module is used to perform modulation coding and beamforming parameter adaptive filtering on the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link.

[0033] Through the collaborative efforts of these components, a five-dimensional signal tensor model is constructed that more comprehensively captures the channel characteristics of low-orbit satellites than existing three-dimensional models. In particular, it provides a more accurate description of the Doppler effect and polarization characteristics, providing a more complete information foundation for subsequent signal processing. A fifth-order Vandermonde structured CP decomposition algorithm, with an adaptive step-size adjustment mechanism, effectively addresses the estimation bias of traditional algorithms in scenarios with rapidly moving satellites, improving the accuracy of channel parameter estimation. A sequential three-stage processing strategy significantly reduces the real-time processing burden while ensuring parameter estimation accuracy. An interference identification method based on supervised contrastive learning addresses the sharp performance degradation of existing identification algorithms in low signal-to-noise ratio and Doppler shift environments, maintaining high identification accuracy even under extreme conditions. Multi-dimensional adaptive filtering technology dynamically adjusts the suppression strategy based on the real-time identified interference type and characteristics, offering greater adaptability than traditional fixed-mode filtering and effectively addressing complex and changing interference environments. The modulation coding and beamforming joint optimization framework of the present invention solves the problem of large signal-to-noise ratio fluctuations in scenarios with rapid satellite movement by introducing a dynamic spectrum sensing mechanism and an orbit prediction model, while achieving a balance between power efficiency and anti-interference performance.

[0034] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0035] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0036] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for improving the signal-to-noise ratio of a low-orbit satellite signal, characterized in that: include: Perform multi-dimensional signal acquisition and processing on the received signals of the low-orbit satellite downlink to obtain a signal tensor model; performing structured decomposition based on the signal tensor model to obtain a channel factor matrix group; Performing multi-level sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; Inputting the low-orbit satellite channel state information into an interference identification module for feature extraction and interference type analysis to obtain the interference type and its probability distribution; Performing multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain a target signal; Modulation coding and beamforming parameters are jointly optimized for the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link.

2. The method for improving the signal-to-noise ratio of a low-orbit satellite signal according to claim 1, wherein: The multi-dimensional signal acquisition and processing of the received signal of the low-orbit satellite downlink to obtain the signal tensor model includes: The time domain waveform data of the received signal of the low-orbit satellite downlink is collected through a multi-channel receiving array to obtain the original signal data; Performing analog-to-digital conversion and preprocessing on the original signal data to obtain a preprocessed signal, and reconstructing the preprocessed signal according to the time dimension, frequency dimension, spatial dimension, Doppler dimension, and polarization dimension to obtain an initial five-dimensional signal structure; Segmenting data based on the initial five-dimensional signal structure using a sliding window technique to obtain continuous signal segments, and adaptively adjusting the sampling rate of the continuous signal segments according to the relative velocity of the satellite to obtain signal data with optimized sampling intervals; The signal data after the sampling interval optimization is normalized to obtain a signal tensor model.

3. The method for improving the signal-to-noise ratio of a low-orbit satellite signal according to claim 1, wherein: The performing structured decomposition based on the signal tensor model to obtain a channel factor matrix group includes: Performing a canonical aggregation representation on the signal tensor model to obtain a sum form of R rank-1 tensors, and establishing a factor vector model with a Vandermonde structure based on the sum form of the rank-1 tensors; Performing high-order singular value decomposition on the factor vector model to obtain an initial subspace structure and a rank R value of the decomposition tensor; Based on the initial subspace structure and the rank R value, an accelerated gradient descent and line search algorithm are used for optimization to obtain an initial factor matrix; Adaptively adjusting the step size of the initial factor matrix to obtain an optimized factor matrix under a fast-moving satellite environment; Residual energy monitoring is performed on the optimization factor matrix to obtain a channel factor matrix group including time dimension parameters, frequency dimension parameters, space dimension parameters, Doppler dimension parameters and polarization dimension parameters.

4. The method for improving the signal-to-noise ratio of a low-orbit satellite signal according to claim 1, wherein: The performing multi-stage sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information includes: Performing augmented spatial-polarization covariance analysis based on the spatial dimension factor matrix and the polarization dimension factor matrix in the channel factor matrix group to obtain a spatial-polarization joint feature representation; Performing feature decomposition and noise subspace projection on the spatial polarization joint feature representation to obtain signal arrival angle and polarization state parameters; Performing a joint time-frequency analysis based on the frequency dimension factor matrix and the Doppler dimension factor matrix in the channel factor matrix group to obtain multipath delay and Doppler frequency shift parameters; Performing recursive least squares filtering on the time dimension factor matrix in the channel factor matrix group to obtain time-varying channel coefficients; The signal arrival angle, the polarization state parameter, the multipath delay, the Doppler frequency shift parameter and the time-varying channel coefficient are integrated to obtain low-orbit satellite channel state information including subcarriers, time sampling points and receiving antenna dimensions.

5. The method for improving the signal-to-noise ratio of a low-orbit satellite signal according to claim 1, wherein: The inputting the low-orbit satellite channel state information into the interference identification module to perform feature extraction and interference type analysis to obtain the interference type and its probability distribution includes: Performing time-frequency conversion on the received signal and the low-orbit satellite channel state information to obtain a time-frequency spectrum matrix; Inputting the time-frequency spectrum matrix into the residual network structure of the interference identification module for feature extraction to obtain a primary signal feature map; Performing feature compression on the primary signal feature map through the global average pooling layer of the interference identification module to obtain a one-dimensional feature vector; Inputting the one-dimensional feature vector into the multi-head self-attention layer of the interference recognition module to perform attention calculation to obtain an enhanced feature vector; Processing the enhanced feature vector through the fully connected classification layer of the interference identification module to obtain a raw score of the interference type; Softmax normalization processing is performed on the original scores to obtain interference types including ground station interference, co-frequency satellite interference, atmospheric scintillation interference, multipath interference and thermal noise and their probability distribution.

6. The method for improving the signal-to-noise ratio of a low-orbit satellite signal according to claim 1, wherein: The performing multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain a target signal includes: Constructing a weight matrix including spatial suppression weights, frequency domain suppression weights, and time domain suppression weights according to the interference type and its probability distribution; Calculating a comprehensive suppression weight vector based on the weight matrix and the probability distribution of the interference type; Performing a minimum variance distortionless response beamforming algorithm on the received signal according to the spatial suppression weight and the low-orbit satellite channel state information to obtain a spatial dimension filtering result; Performing adaptive notch filtering and band-stop filtering on the received signal based on the frequency domain suppression weight to obtain a frequency domain dimension filtering result, and performing adaptive Kalman filtering on the received signal based on the time domain suppression weight to obtain a time domain dimension filtering result; According to the comprehensive suppression weight vector, the spatial dimension filtering result, the frequency domain dimension filtering result and the time domain dimension filtering result are weightedly fused to obtain a target signal.

7. The method for improving the signal-to-noise ratio of a low-orbit satellite signal according to claim 1, wherein: The performing joint optimization of modulation coding and beamforming parameters on the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link includes: Performing double-layer dynamic spectrum sensing processing based on the target signal to obtain a spectrum sensing result including broadband energy distribution and frequency band characteristics; Inputting the spectrum sensing result, the low-orbit satellite channel state information, and the target signal into an LSTM-CNN hybrid network for channel state analysis to obtain a channel state vector including channel gain, multipath delay spread, Doppler frequency shift, and interference residual; Establishing a modulation and coding selection strategy according to the channel state vector, wherein the modulation and coding selection strategy includes an optimal modulation order and coding rate combination; Calculating the Doppler frequency shift parameter in the channel state vector based on the satellite orbit prediction model to obtain a relative motion parameter and a Doppler frequency shift feedforward compensation value; According to the channel state vector, the optimal modulation order and coding rate combination, a multi-objective genetic algorithm is used to simultaneously optimize the signal-to-noise ratio and power efficiency to obtain optimized beamforming parameters; The target signal is processed based on the optimized beamforming parameters, the Doppler frequency shift feedforward compensation value, and the optimal modulation order and coding rate combination to obtain a low-orbit satellite communication link for improving the signal-to-noise ratio.

8. A low-orbit satellite signal signal-to-noise ratio improvement device, characterized in that: Used to perform the method for improving the signal-to-noise ratio of a low-orbit satellite signal according to any one of claims 1 to 7, the low-orbit satellite signal signal-to-noise ratio improving device comprising: The signal acquisition module is used to perform multi-dimensional signal acquisition and processing on the received signals of the low-orbit satellite downlink to obtain a signal tensor model; a structured decomposition module, configured to perform structured decomposition based on the signal tensor model to obtain a channel factor matrix group; a processing module, configured to perform multi-stage sequential processing according to the channel factor matrix group to obtain low-orbit satellite channel state information; A type analysis module is used to input the low-orbit satellite channel state information into an interference identification module for feature extraction and interference type analysis to obtain the interference type and its probability distribution; An adaptive filtering module, configured to perform multi-dimensional adaptive filtering on the received signal based on the interference type and its probability distribution to obtain a target signal; The joint optimization module is used to jointly optimize the modulation coding and beamforming parameters of the target signal and the low-orbit satellite channel state information to obtain a low-orbit satellite communication link.

Images