Satellite laser communication signal and noise joint suppression method and device

Abstract

The application relates to the technical field of noise suppression, and provides a satellite laser communication signal noise joint suppression method and device, the first amplitude spectrum graph is corrected by using an amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics; the first phase spectrum graph is corrected by using a phase spectrum morphology discrimination index and phase spectrum physical characteristics, various types of abnormalities in the amplitude and phase spectrum graphs can be accurately identified and repaired, the data after correction can strictly conform to physical realizability, and high-quality and reasonable-structure data bases for noise suppression are provided. The first amplitude spectrum graph after correction and the first phase spectrum graph after correction are input into two input ends of a denoising model, a second amplitude spectrum graph and a second phase spectrum graph are output from an output end of the denoising model, the denoising model is integrated with a phase perception Transformer module and a codec, noise can be effectively suppressed under extreme working conditions, and signal fidelity is improved.

Description

Technical Field

[0001] This application relates to the field of noise suppression technology, and in particular to a method and apparatus for joint noise suppression of satellite laser communication signals. Background Technology

[0002] Satellite laser communication is an advanced technology that uses laser beams as information carriers to achieve high-speed data transmission between satellites or between satellites and ground stations. Its core principle is to encode information into optical signals by jointly modulating the amplitude and phase of the laser, and then transmit it over long distances in free space. Compared with traditional microwave communication, satellite laser communication has significant advantages such as extremely large bandwidth, high transmission rate, strong resistance to electromagnetic interference, small equipment size and low power consumption. It has become a key supporting technology for building future integrated air-space-ground information networks and low-Earth orbit mega-constellations.

[0003] However, in actual operating environments, satellite laser communication signals are highly susceptible to combined interference from multiple noise sources. In satellite-to-ground links, atmospheric turbulence can cause light intensity flicker and time-frequency energy dissipation; sunlight and cosmic background radiation introduce broadband photon noise; platform micro-vibrations cause beam pointing jitter, disrupting inter-frame continuity; Doppler frequency shift induces phase non-stationarity, causing phase jumps and global monotonicity instability; in addition, the detector's own shot noise, thermal noise, and quantization errors further degrade the quality of the received signal. All these interferences together lead to low quality of the received signal.

[0004] To address these challenges, existing satellite laser communication signal denoising primarily relies on physical layer compensation and signal processing algorithms. For example, adaptive optics is used to suppress atmospheric turbulence, narrowband and spatial filtering is used to suppress background light, high-precision acquisition and tracking alignment systems are used to mitigate pointing errors, and phase-locked loops, carrier recovery algorithms, and forward error correction are deployed at the receiver to correct phase noise and improve fault tolerance. However, these methods are mostly based on linear models or fixed threshold criteria, making it difficult to effectively handle non-Gaussian, non-stationary, and strongly nonlinear interferences caused by various noise couplings.

[0005] With the development of artificial intelligence, deep learning technology has been introduced into the field of laser communication denoising. In existing intelligent denoising methods, the amplitude spectrum and phase spectrum of satellite laser communication signals are usually spliced ​​together and then directly sent to a denoising model based on neural networks for denoising. Summary of the Invention

[0006] This application provides a method and apparatus for joint noise suppression of satellite laser communication signals, which addresses the problem that existing methods for denoising satellite laser communication signals do not consider the unique physical constraints of laser communication signals, resulting in a lack of physical constraint guidance and interpretability in the denoising process, making it unable to cope with unknown interference conditions and failing to meet the requirements of signal fidelity and robustness.

[0007] To address the aforementioned technical problems, the first aspect of this application provides a method for joint noise suppression of satellite laser communication signals, comprising: A short-time Fourier transform is performed on the time-domain signal of satellite laser communication to obtain the first amplitude spectrum and the first phase spectrum; The amplitude spectrum is verified by using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain the amplitude quality judgment result of each time frame in the first amplitude spectrum; the first amplitude spectrum is corrected according to the amplitude quality judgment result of each time frame in the first amplitude spectrum. The phase spectrum morphology discrimination index and phase spectrum physical characteristics are used to verify each time frame of the first phase spectrum to obtain the phase quality judgment result of each time frame of the first phase spectrum; the first phase spectrum is corrected based on the phase quality judgment result of each time frame of the first phase spectrum. The corrected first amplitude spectrum and the corrected first phase spectrum are input to the two input terminals of the pre-established denoising model, and the second amplitude spectrum and the second phase spectrum are output from the output terminal of the denoising model. Perform inverse short-time Fourier transform on the second amplitude spectrum and the second phase spectrum to obtain the denoised satellite laser communication time-domain signal; The denoising model includes: a first encoder, a second encoder, a fusion module, a phase-aware Transformer module, a first decoder, and a second decoder. The input terminals of the first encoder and the second encoder constitute the input terminals of the denoising model, and are used to extract features from the corrected first amplitude spectrum and the corrected first phase spectrum, respectively, to obtain a first feature map and a second feature map. The fusion module is used to concatenate the first feature map and the second feature map to obtain a first fused feature tensor. The phase-aware Transformer module is used to obtain a second fused feature tensor based on the first fused feature tensor. The output terminals of the first decoder and the second decoder constitute the output terminals of the denoising model, and are used to generate the second amplitude spectrum and the second phase spectrum based on the second fused feature tensor, respectively.

[0008] A second aspect of this application provides a satellite laser communication signal noise joint suppression device, comprising: The first transformation unit is configured to perform a short-time Fourier transform on the time-domain signal of satellite laser communication to obtain a first amplitude spectrum and a first phase spectrum. The first correction unit is configured to verify each time frame of the first amplitude spectrum using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain amplitude quality judgment result of each time frame of the first amplitude spectrum; and to correct the first amplitude spectrum based on the amplitude quality judgment result of each time frame of the first amplitude spectrum. The second correction unit is configured to verify each time frame of the first phase spectrum using phase spectrum morphology discrimination index and phase spectrum physical characteristics to obtain the phase quality judgment result of each time frame of the first phase spectrum; and to correct the first phase spectrum based on the phase quality judgment result of each time frame of the first phase spectrum. The denoising unit is configured to input the corrected first amplitude spectrum and the corrected first phase spectrum to the two input terminals of a pre-established denoising model, and output the second amplitude spectrum and the second phase spectrum from the output terminal of the denoising model. The second transformation unit is configured to perform inverse short-time Fourier transform on the second amplitude spectrum and the second phase spectrum to obtain the denoised satellite laser communication time domain signal; The denoising model includes: a first encoder, a second encoder, a fusion module, a phase-aware Transformer module, a first decoder, and a second decoder. The input terminals of the first encoder and the second encoder constitute the input terminals of the denoising model, and are used to extract features from the corrected first amplitude spectrum and the corrected first phase spectrum, respectively, to obtain a first feature map and a second feature map. The fusion module is used to concatenate the first feature map and the second feature map to obtain a first fused feature tensor. The phase-aware Transformer module is used to obtain a second fused feature tensor based on the first fused feature tensor. The output terminals of the first decoder and the second decoder constitute the output terminals of the denoising model, and are used to generate the second amplitude spectrum and the second phase spectrum based on the second fused feature tensor, respectively.

[0009] A third aspect of this application provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement an embodiment of the aforementioned method.

[0010] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor of a computer device, implements an embodiment of the aforementioned method.

[0011] The fifth aspect of this application provides a computer program product comprising a computer program that, when executed by a processor of a computer device, implements an embodiment of the aforementioned method.

[0012] The satellite laser communication signal noise joint suppression method and apparatus provided in this application overcomes the shortcomings of existing satellite laser communication signal noise suppression methods in terms of insufficient modeling of the unique physical constraints of laser communication signals. This ensures that the corrected first amplitude spectrum and first phase spectrum strictly conform to the physical constraints of the laser communication signal, providing a high-quality and structurally sound data foundation for noise suppression in the denoising model. The method involves: performing a short-time Fourier transform on the time-domain signal of the satellite laser communication signal to obtain a first amplitude spectrum and a first phase spectrum; verifying each time frame in the first amplitude spectrum using amplitude spectrum morphology discrimination indexes and phase spectrum physical characteristics to obtain an amplitude quality judgment result for each time frame; and correcting the first amplitude spectrum based on the amplitude quality judgment result for each time frame.

[0013] By inputting the corrected first amplitude spectrum and the corrected first phase spectrum into the two input terminals of a pre-established denoising model, and outputting the second amplitude spectrum and the second phase spectrum from the output terminal of the denoising model, the denoising model is designed to use two encoders to extract features from the first amplitude spectrum and the second phase spectrum respectively, use a phase-aware Transformer module to process the fused feature tensor, and use two decoders to decode the fused feature tensor output by the phase-aware Transformer module into the denoised second amplitude spectrum and the second phase spectrum respectively. This allows the model to learn the most suitable feature representations for amplitude and phase respectively. At the same time, it can also explicitly model the global dependency of phase information on long-term sequences (such as monotonicity and continuity), thereby effectively suppressing noise, restoring the phase structure damaged by noise, and improving signal fidelity even under extreme conditions.

[0014] This application combines the correction of amplitude spectrum diagrams using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics, the correction of phase spectrum diagrams using phase spectrum morphology discrimination index and phase spectrum physical characteristics, and intelligent denoising. This approach can increase physical interpretability and structural guidance during the denoising process, thereby enabling satellite laser communication signals to achieve accurate denoising even when facing unknown interference or extreme conditions.

[0015] To make the above and other objects, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of a satellite laser communication signal noise joint suppression method according to an embodiment of this application is shown; Figure 2 A structural diagram of the denoising model according to an embodiment of this application is shown; Figure 3 A flowchart of the first amplitude spectrum correction process according to an embodiment of this application is shown; Figure 4 A flowchart illustrating the amplitude anomaly frame correction process according to an embodiment of this application is shown; Figure 5 A flowchart of the first phase spectrum correction process according to an embodiment of this application is shown; Figure 6 A flowchart illustrating the phase aberration frame correction process according to an embodiment of this application is shown; Figure 7 A structural diagram of a satellite laser communication signal noise joint suppression device according to an embodiment of this application is shown; Figure 8 A structural diagram of a computer device according to an embodiment of this application is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0020] This specification provides the operational steps of the methods described in the embodiments or flowcharts, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or device products, the methods shown in the embodiments or drawings can be executed sequentially or in parallel.

[0021] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all information and data authorized by the user or fully authorized by all parties, and the acquisition, transmission, storage, use and processing of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0022] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0023] Current intelligent denoising methods typically concatenate the amplitude and phase spectra of a signal and directly feed them into a denoising model based on a neural network. However, these methods do not explicitly incorporate the specific prior physical constraints of laser communication signals, such as the nonnegativity, power limitations, and time-frequency energy concentration of the amplitude spectrum, as well as the continuity, monotonicity, and local stability of the phase spectrum. This makes it difficult to effectively address the multi-dimensional anomalies in laser communication signals. Consequently, existing denoising models lack physical interpretability and structural guidance when applied to satellite laser communication signal denoising. This leads to potential issues such as phase distortion, amplitude distortion, or accidental deletion of key time-frequency structures when facing unknown interference or extreme conditions, failing to meet the stringent requirements of high-reliability satellite laser communication systems for signal fidelity and robustness. In addition, existing denoising models use the concatenation of amplitude and phase spectra as input, which means they cannot learn the most suitable amplitude and phase feature representations separately, thus failing to meet the signal fidelity requirements of high-reliability satellite laser communication systems.

[0024] To address the aforementioned technical problems in the existing technology, some embodiments of this application, such as... Figure 1 As shown, a method for joint noise suppression of satellite laser communication signals is provided, comprising: Step 101: Perform a short-time Fourier transform on the time-domain signal of the satellite laser communication to obtain the first amplitude spectrum and the first phase spectrum.

[0025] It is worth noting that in this embodiment, firstly, the satellite laser communication time-domain signal is received. Specifically, the satellite laser communication time-domain signal from the free-space channel is captured by the photodetector of the satellite laser communication receiving terminal. The satellite laser communication time-domain signal is a one-dimensional continuous or discrete time series, characterizing the dynamic process of the received light intensity changing over time. Subsequently, the satellite laser communication time-domain signal is preprocessed, including sampling rate normalization, DC component removal, and necessary time alignment operations, to ensure the accuracy and consistency of subsequent spectrum analysis.

[0026] Next, the preprocessed satellite laser communication time-domain signal is decomposed into time-frequency components using the short-time Fourier transform method: the entire signal is divided into multiple overlapping time frames, a window function is applied to each frame to reduce boundary effects, and a Fourier transform is performed independently on each frame to obtain the complex spectral coefficients at the corresponding frequency points. For each frequency point in each time frame, the magnitude and argument are extracted from the corresponding complex spectral coefficients. The magnitude constitutes the amplitude response of the time frame at that frequency point, and the argument constitutes the phase response of the time frame at that frequency point. The amplitude responses of all time frames across the entire frequency band are arranged in chronological order to form a two-dimensional amplitude spectrum. Similarly, the phase responses of all time frames across the entire frequency band are arranged in chronological order to form a two-dimensional phase spectrum.

[0027] Ultimately, the obtained amplitude spectrum and phase spectrum are strictly aligned in the time and frequency dimensions, and are both represented in matrix form, with each row corresponding to a time frame and each column corresponding to a frequency point, thus completely preserving the time-frequency structure information of the original satellite laser communication time-domain signal.

[0028] Step 102: Verify each time frame of the first amplitude spectrum using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain amplitude quality judgment results for each time frame of the first amplitude spectrum; correct the first amplitude spectrum based on the amplitude quality judgment results for each time frame of the first amplitude spectrum.

[0029] It is worth noting that in this embodiment, the aforementioned dimensions are not derived through direct calculation. For example, the value of time-frequency energy concentration is not directly calculated. Instead, a corresponding quantifiable criterion is invoked or constructed, and the numerical result of this criterion is compared with a preset threshold to determine whether the dimension meets the quality requirements. In other words, each dimension corresponds to a specific, executable discrimination operation process, which outputs a logical judgment result—whether it meets or not—rather than outputting the dimension itself. In summary, all dimensions are discriminated by invoking underlying criterion calculations and logical comparison operations corresponding to their semantics. Essentially, quantifiable operations support dimension judgment, rather than directly generating the numerical value of the dimension itself.

[0030] The amplitude spectrum morphology discrimination indicators include the following dimensions: time-frequency energy concentration, inter-frame amplitude variation gradient, and frequency band edge attenuation smoothness. The amplitude spectrum physical characteristics include the following dimensions: amplitude non-negativity and transmit power dynamic range limitation.

[0031] In some implementations, for each time frame in the amplitude spectrum, based on a comprehensive consideration of signal processing logic hierarchy and anomaly detection efficiency, the following verification sequence is proposed: first, perform verification of the time-frequency energy concentration dimension; second, perform verification of the inter-frame amplitude change gradient dimension; third, perform verification of the frequency band edge attenuation smoothness dimension; then, perform verification of the self-amplitude non-negativity dimension; and finally, perform verification of the transmit power dynamic range limitation dimension.

[0032] The first step in verifying the time-frequency energy concentration dimension is because this dimension reflects the basic existence and structural rationality of the signal in the frequency domain. If the energy of a time frame signal is highly diffuse and the entropy value is too high, it is very likely to be an invalid frame dominated by noise or completely lacking effective communication content. In this case, subsequent more detailed analyses of inter-frame amplitude changes, frequency band edge characteristics, etc., will lose physical meaning. Therefore, placing it first can enable the rapid screening of obviously invalid frames and avoid unnecessary computational overhead.

[0033] The second step is to verify the gradient dimension of the amplitude change between frames. This is to further determine whether the dynamic continuity between the current frame and the historical frames is reasonable, on the premise that the current frame has a basic energy structure. If this step is skipped and the edge smoothness is checked first, it is easy to misjudge isolated abnormal frames such as sudden interference as edge distortion. In fact, the fundamental problem is that the inter-frame mutation is too large. Therefore, this step is used as a second layer of filtering to remove pseudo-valid frames that are not dynamically consistent but have concentrated energy.

[0034] The third step is to verify the smoothness of the frequency band edge attenuation. At this point, it is ensured that the signal is reasonable in terms of overall energy distribution and time continuity. Only then can we focus on the detailed features of the sensitive area of ​​the frequency band boundary. If this step is performed in advance, the edge criteria may fail due to strong noise frames that have not been filtered out in the previous step, resulting in a large number of misjudgments.

[0035] The verification of the fourth dimension, the non-negativity of the amplitude, is a hard physical constraint that is usually implicitly satisfied in hardware or front-end processing. However, to ensure the robustness of the algorithm, it still needs to be explicitly verified. It is placed in a later position because even if the amplitude is negative, it is necessary to first confirm whether the frame has analytical value in the first three dimensions. If the previous steps have already determined it to be invalid, there is no need to check this item again.

[0036] Finally, the verification of the transmit power dynamic range limit dimension is performed. Since it involves global power calibration, it usually depends on system configuration parameters and has a relatively mild impact on the validity of a single frame. For example, a slight overload or weak signal may still contain usable information. Placing it at the end can complete the final compliance check while retaining as many candidate frames as possible.

[0037] If the execution is performed in any other order, such as prioritizing amplitude nonnegativity or power range verification, a large number of invalid frames that could have been efficiently filtered out in the preceding steps will enter the subsequent complex criterion calculations, significantly increasing the computational burden. If edge smoothness or inter-frame gradient is moved forward to before energy concentration, it is easy to make incorrect local reasonable judgments on noisy frames without actual communication content, thereby misleading subsequent correction operations to repair invalid data and introducing false structures. Therefore, this sequential execution design takes into account discrimination accuracy, computational efficiency, and engineering robustness, and is a key guarantee for achieving high-reliability amplitude spectrum quality determination.

[0038] Step 103: Verify each time frame in the first phase spectrum using the phase spectrum morphology discrimination index and phase spectrum physical characteristics to obtain the phase quality judgment result of each time frame in the first phase spectrum; and correct the first phase spectrum based on the phase quality judgment result of each time frame in the first phase spectrum.

[0039] It is worth noting that in this embodiment, the aforementioned dimensions are not derived directly through calculation. For example, the phase continuity deviation is not calculated directly. Instead, a corresponding quantifiable criterion is invoked or constructed, and the numerical result of this criterion is compared with a preset threshold to determine whether the dimension meets the quality requirements. In other words, each dimension corresponds to a specific, executable discrimination operation process, which outputs a logical judgment result—whether it meets or not—rather than outputting the dimension itself. In summary, all dimensions are discriminated by invoking underlying criterion calculations and logical comparison operations corresponding to their semantics. Essentially, quantifiable operations support dimension judgment, rather than directly generating the numerical value of the dimension itself.

[0040] The phase spectrum morphology discrimination indicators include the following dimensions: phase continuity deviation, phase jump frequency, and phase time-frequency stability. The phase spectrum physical characteristics include the following dimensions: phase continuity constraint and carrier phase global monotonicity. In some implementations, for each time frame in the phase spectrum, based on a comprehensive consideration of anomaly detection efficiency, computational dependency, and error propagation control, the following verification sequence is proposed: First, phase continuity deviation is the most basic and fastest initial screening indicator. It is used to measure whether the phase change between adjacent frequency points within the current time frame is smooth and reflects whether the local frequency domain structure is reasonable. If severe discontinuity is found at this step, such as severe phase jitter caused by FFT window function misalignment or data truncation, it means that the underlying data of the frame is likely corrupted. There is no need to perform more complex analysis. Placing it first can remove obviously invalid frames as early as possible and save computing resources.

[0041] Secondly, after passing the initial smoothness check, the phase jump frequency is determined. This step counts the number of phase abrupt changes per unit time. However, it should be noted that the number of jumps is only meaningful when the overall phase is relatively continuous. If the previous step has determined that the phase is extremely disordered, the statistical results of the jump frequency will lose their reference value and may even be misjudged as low jump due to noise interference. Therefore, it is necessary to confirm the basic continuity first and then evaluate the jump characteristics.

[0042] The third step is phase time-frequency stability, which examines whether the phase change in the two-dimensional time-frequency plane conforms to the slow evolution law of the physical channel. If serious anomalies have been found in the first two steps, continuing to execute this step will not only waste computing power, but may also lead to inaccurate stability indicators due to input data distortion. Therefore, it is placed in the middle position to ensure input quality and avoid premature investment in high-cost calculations.

[0043] The fourth step, phase continuity constraint, is to verify the legality of the overall range of phase values. Such problems often only become apparent after the aforementioned dynamic characteristic verification. For example, a seemingly stable phase may actually exhibit a false trend due to un-unwinding and long-term accumulation. Therefore, it is necessary to check whether its numerical representation is standardized after the dynamic behavior has been initially confirmed to be normal.

[0044] Finally, the global monotonicity of the carrier phase focuses on whether the carrier phase exhibits a reasonable monotonic change throughout the entire observation period. This is the highest level of semantic verification, which depends on the normal performance of all the preceding dimensions. If the local continuity is poor, the jumps are frequent, or the time and frequency are unstable, the global monotonicity will inevitably fail. Placing it last ensures that the final confirmation of the high-level semantics is only made under the condition that the lower-level structure is reliable.

[0045] By designing the verification sequence described above, efficiency can be optimized and the reliability of the judgment logic can be guaranteed. In actual implementation, if the verification sequence is not followed, for example, if a global monotonicity judgment is performed first and then the local continuity is checked, two serious problems will occur: First, a large amount of computing resources will be wasted on obviously abnormal data; second, the high-level judgment may be contaminated by the low-level error, leading to a seemingly reasonable but actually completely wrong conclusion. For example, a noisy phase curve may accidentally show an approximately linear trend and be mistakenly judged as qualified.

[0046] Step 104: Input the corrected first amplitude spectrum and the corrected first phase spectrum to the two input terminals of the pre-established denoising model, and output the second amplitude spectrum and the second phase spectrum from the output terminal of the denoising model.

[0047] Step 105: Perform inverse short-time Fourier transform on the second amplitude spectrum and the second phase spectrum to obtain the denoised satellite laser communication time-domain signal.

[0048] The second amplitude spectrum and the second phase spectrum are reconstructed spectra relative to the original first amplitude spectrum and the first phase spectrum.

[0049] It is worth noting that in this embodiment, firstly, the reconstructed amplitude spectrum and the reconstructed phase spectrum are aligned one by one in terms of time frames and frequency points to ensure that they have exactly the same number of time frames and frequency points.

[0050] Next, for each time frame, perform the following operations: For each frequency point in the frame, the value corresponding to that point in the reconstructed amplitude spectrum is taken as the amplitude, and the value corresponding to that point in the reconstructed phase spectrum is taken as the phase angle. According to the trigonometric function relationship, the complex number representation corresponding to that frequency point is calculated, that is, the cosine of the amplitude multiplied by the phase angle is taken as the real part, and the sine of the amplitude multiplied by the phase angle is taken as the imaginary part, so as to obtain the complex value of that frequency point.

[0051] Next, the complex values ​​of all frequency points within the time frame are arranged in frequency order to form a complete complex spectrum sequence.

[0052] Then, an inverse short-time Fourier transform operation is performed on the complex spectrum sequence. Specifically, the complex spectrum sequence is input into a pre-defined inverse transform calculation unit. This unit reconstructs the time-domain waveform segment corresponding to the time frame by superimposing the sine and cosine components of each frequency component at the corresponding time position according to the inverse process of the short-time Fourier transform. The parameters such as the window function type, window length, and frame shift step size used in the superposition process are completely consistent with the short-time Fourier transform parameters used when the original spectrum was initially generated.

[0053] Subsequently, the complex number synthesis and inverse transform operations described above are repeated for all time frames to obtain a series of continuous time-domain waveform segments. Finally, these time-domain waveform segments are spliced ​​together according to their corresponding time order using an overlapping and adding method: that is, the overlapping area of ​​adjacent segments is determined according to the frame shift step size, and the values ​​of the corresponding time points are added within the overlapping area, while the non-overlapping parts are directly retained, thereby synthesizing a complete and continuous time-domain signal waveform. The synthesized time-domain signal is the denoised satellite laser communication time-domain signal. Since its amplitude spectrum and phase spectrum have been corrected and reconstructed by the neural network as described above, the abnormal interference in the original signal has been suppressed, while the time-frequency structure characteristics of the effective communication signal are preserved. Therefore, this time-domain signal has a higher signal-to-noise ratio and waveform fidelity.

[0054] This embodiment, through the implementation of steps 102 and 103, can overcome the shortcomings of existing satellite laser communication signal noise suppression methods in terms of insufficient modeling of the unique physical constraints of laser communication signals, ensuring that the corrected first amplitude spectrum and first phase spectrum strictly conform to the physical constraints of laser communication signals, and providing a high-quality and structurally reasonable data foundation for noise suppression of the denoising model.

[0055] like Figure 2 As shown, the denoising model includes: a first encoder 201, a second encoder 202, a fusion module 203, a phase-aware Transformer module 204, a first decoder 205, and a second decoder 206.

[0056] The input terminals of the first encoder 201 and the second encoder 202 constitute the input terminals of the denoising model, and are used to extract features from the corrected first amplitude spectrum and the corrected first phase spectrum to obtain the first feature map and the second feature map, respectively.

[0057] The fusion module 203 is used to concatenate the first feature map and the second feature map to obtain the first fused feature tensor.

[0058] The phase-aware Transformer module 204 is used to obtain the second fusion feature tensor based on the first fusion feature tensor.

[0059] The outputs of the first decoder 205 and the second decoder 206 constitute the outputs of the denoising model, and are used to generate the second amplitude spectrum and the second phase spectrum based on the second fused feature tensor, respectively.

[0060] based on Figure 2 The denoising model structure shown includes the following internal processing flow: (1) Input the corrected first amplitude spectrum and the corrected first phase spectrum into a pre-constructed two-branch U-shaped neural network. The two encoder paths of the pre-constructed two-branch U-shaped neural network output the first feature map and the second feature map respectively. Then, the first feature map and the second feature map are spliced ​​together to obtain the fused feature tensor.

[0061] It is worth noting that in this embodiment, firstly, a pre-constructed dual-branch U-shaped neural network is prepared. This dual-branch U-shaped neural network consists of two encoder paths with identical structures, used to process amplitude information and phase information respectively; the two encoder paths are independent of each other and do not share weights; each encoder path adopts the encoding part structure of a standard U-shaped network, that is, it sequentially contains several downsampling stages, each stage consisting of a convolutional layer, an activation function, and a pooling operation, used to extract multi-scale spatial features of the input data step by step.

[0062] Secondly, the corrected first amplitude spectrum obtained after the aforementioned correction steps is used as the first input and fed into the first encoder path of the dual-branch U-shaped neural network; simultaneously, the corrected first phase spectrum is used as the second input and fed into the second encoder path. Both inputs have identical dimensions in both the time and frequency dimensions, ensuring subsequent feature alignment.

[0063] Subsequently, the first encoder path performs stepwise feature extraction on the corrected first amplitude spectrum, ultimately outputting a high-dimensional array, denoted as the first feature map; the second encoder path performs the same stepwise feature extraction on the corrected first phase spectrum, ultimately outputting another high-dimensional array, denoted as the second feature map. The two feature maps are consistent in spatial dimension, i.e., time-frequency location, only differing in channel dimension due to network design.

[0064] Next, the first feature map and the second feature map are concatenated along the channel dimension. Specifically, without changing the time-frequency position, all channels of the first feature map are sequentially connected with all channels of the second feature map to form a new high-dimensional array.

[0065] Finally, the concatenated array is used as the final output to obtain the fused feature tensor. This tensor contains both the intensity distribution features from the corrected amplitude spectrum and the phase evolution features from the corrected phase spectrum, and the two are strictly aligned in time and frequency positions, which can be used for subsequent joint analysis or reconstruction tasks.

[0066] (2) Input the fused feature tensor into the phase-aware Transformer module embedded in the bottleneck layer of the pre-built dual-branch U-shaped neural network, and input the fused feature tensor processed by the phase-aware Transformer module into the two decoder paths of the pre-built dual-branch U-shaped neural network to output the reconstructed amplitude spectrum and the reconstructed phase spectrum respectively.

[0067] It is worth noting that in this embodiment, firstly, in the pre-built dual-branch U-shaped neural network, a bottleneck layer is set between its encoder path and decoder path; a phase-aware Transformer module is embedded inside the bottleneck layer. This module has been fixedly integrated into the bottleneck position during the neural network construction stage as an enhancement processing unit for fused features.

[0068] Secondly, the obtained fused feature tensor is directly input into the phase-aware Transformer module in the bottleneck layer. This module performs internal transformation processing on the fused feature tensor. Specifically, it divides the fused feature tensor into several feature sequence units according to a preset method, performs global interactive modeling on these units using pre-trained parameters within the module, and outputs a processed fused feature tensor with the same dimensions as the input. This process does not change the time-frequency spatial structure of the fused feature tensor, but only adjusts its channel content with context awareness.

[0069] Subsequently, the fused feature tensor processed by the phase-aware Transformer module is simultaneously fed into the two decoder paths of the pre-constructed dual-branch U-shaped neural network. These two decoder paths correspond to the amplitude branch and the phase branch, respectively, and are structurally mirror images of each other. Both adopt the decoding part structure of the standard U-shaped network, which sequentially contains several upsampling stages. Each stage consists of upsampling operations, skip connections (i.e., intermediate features from the corresponding encoder path), convolutional layers, and activation functions, used to gradually restore the spatial resolution of the original spectrum.

[0070] Next, the first decoder path takes the processed fused feature tensor as input, combines it with the jump connection information provided by its corresponding amplitude encoder path, upsamples step by step and reconstructs a two-dimensional array with the same size as the original input, denoted as the reconstructed amplitude spectrum; the second decoder path also takes the processed fused feature tensor as input, combines it with the jump connection information provided by its corresponding phase encoder path, and reconstructs another two-dimensional array of the same size, denoted as the reconstructed phase spectrum.

[0071] Finally, the two reconstruction results are output: the reconstructed amplitude spectrum and the reconstructed phase spectrum. The two are strictly aligned in the time and frequency dimensions and retain the main structural characteristics of the corrected amplitude and phase, respectively. At the same time, the time-frequency consistency is enhanced by the global interaction of the bottleneck layer.

[0072] This embodiment fully considers the significant differences between amplitude and phase in terms of physical nature and statistical characteristics. By designing a first encoder and a second encoder in the denoising model, the denoising model can learn the optimal feature representations of the amplitude spectrum and the phase spectrum, respectively.

[0073] This embodiment concatenates the output feature maps of the two encoders along the channel dimension to form a preliminary fused feature tensor. This fused feature tensor is then fed into a specially designed phase-aware Transformer module. Leveraging the phase-aware Transformer module's phase-aware capabilities, its internal self-attention mechanism is specifically optimized to more effectively capture the global dependencies and monotonic evolution trends of phase information over long time sequences. The fused feature tensor, enhanced by the phase-aware Transformer model, is then simultaneously fed to two symmetrical decoders. During the upsampling and resolution recovery process, each decoder obtains multi-scale shallow detail information from its corresponding encoder branch through skip connections, thereby accurately reconstructing the final amplitude and phase spectra respectively.

[0074] This embodiment combines the use of amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to correct the amplitude spectrum, the use of phase spectrum morphology discrimination index and phase spectrum physical characteristics to correct the phase spectrum, and intelligent denoising. This approach can increase physical interpretability and structural guidance during the denoising process, thereby enabling satellite laser communication signals to achieve accurate denoising even when facing unknown interference or extreme conditions.

[0075] In some embodiments of this application, the process of establishing the denoising model includes: (1) Data preparation stage: Construct a high-quality, large-scale training dataset.

[0076] First, a large number of noise samples and the corresponding noise-free ground truth signals are collected or simulated under various typical satellite laser communication channel environments.

[0077] Each noise sample represents a time-domain signal from satellite laser communication. The satellite laser communication channel environment includes, for example, atmospheric turbulence of varying intensities, background sunlight noise of varying levels, and Doppler shifts caused by different velocities.

[0078] Secondly, a short-time Fourier transform is performed on the collected or simulated noise samples to obtain the first amplitude spectrum and the first phase spectrum of the noise samples. Each time frame of the first amplitude spectrum of the noise samples is verified using amplitude spectrum morphology discrimination indexes and amplitude spectrum physical characteristics to obtain the amplitude quality judgment result for each time frame of the first amplitude spectrum of the noise samples. Based on the amplitude quality judgment results for each time frame of the first amplitude spectrum of the noise samples, the first amplitude spectrum of the noise samples is corrected. Similarly, each time frame of the first phase spectrum of the noise samples is verified using phase spectrum morphology discrimination indexes and phase spectrum physical characteristics to obtain the phase quality judgment result for each time frame of the first phase spectrum of the noise samples. Based on the phase quality judgment results for each time frame of the first phase spectrum of the noise samples, the first phase spectrum of the noise samples is corrected. Finally, the corrected amplitude spectrum and the corrected phase spectrum are obtained.

[0079] Next, a short-time Fourier transform is performed on the true signal of the noise sample to obtain the amplitude spectrum and phase spectrum of the true signal; a training dataset is established from the corrected amplitude spectrum and phase spectrum of the noise sample, and the amplitude spectrum and phase spectrum of the true signal.

[0080] The above method of establishing the training dataset ensures that the goal of network learning is to map a high-quality input, but still with residual complex noise, to a perfect output, rather than forcibly learning from messy raw data, thus improving the accuracy of the denoising model.

[0081] (2) Training phase: The pre-built denoising model is trained using the training dataset. The structure of the denoising model is as follows: Figure 2 As shown. The first encoder, the second encoder, the first decoder, and the second decoder can be implemented using neural networks.

[0082] During model training, the corrected amplitude and phase spectra of the noise samples are used as input data, and the amplitude and phase spectra of the true signal are used as output data.

[0083] During the training and optimization phases of the denoising model, an end-to-end supervised learning approach is adopted. In implementation, multiple loss terms can be combined to construct the loss function, such as L1 / L2 reconstruction loss for the amplitude spectrum and cycle-consistent loss for the phase spectrum (to maintain phase continuity). The loss function and training dataset are used for iterative optimization, gradually adjusting the weight parameters in the denoising model until it can stably and efficiently map from the corrected spectrum to the ideal spectrum.

[0084] In some embodiments of this application, when constructing the loss function, soft constraint loss based on one or more dimensions of amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics, phase spectrum morphology discrimination index and phase spectrum physical characteristics may be introduced to further guide the denoising model to output denoised amplitude spectrum and phase spectrum that strictly conform to the physical constraint law.

[0085] In some embodiments of this application, the amplitude spectrum morphology discrimination indicators include: time-frequency energy concentration, inter-frame amplitude variation gradient, and band edge attenuation smoothness. The amplitude spectrum physical characteristics include: amplitude non-negativity and dynamic range limitation of transmit power.

[0086] Step 102 above verifies each time frame in the first amplitude spectrum using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain the amplitude quality judgment result for each time frame in the first amplitude spectrum, including: Iterate through each time frame in the first amplitude spectrum, and perform the following operation on the current time frame: (1) Calculate the energy distribution entropy value of the current time frame in the first amplitude spectrum across the entire frequency band. Compare the energy distribution entropy value with the preset concentration reference value. When the energy distribution entropy value is less than or equal to the preset concentration reference value, it is determined that the current time frame in the first amplitude spectrum meets the requirements in the time-frequency energy concentration dimension. When the energy distribution entropy value is greater than the preset concentration reference value, it is determined that the current time frame in the first amplitude spectrum does not meet the requirements in the time-frequency energy concentration dimension, and the amplitude quality judgment result of the current time frame in the first amplitude spectrum is abnormal.

[0087] It is worth noting that, in this embodiment, the acquisition of the energy distribution entropy value is as follows: First, the amplitude values ​​of all frequency points in the current time frame in the first amplitude spectrum are squared to obtain the energy value of each frequency point; then, the energy values ​​of all frequency points are summed to obtain the total energy of the frame; next, the energy value of each frequency point is divided by the total energy of the frame to obtain the normalized energy distribution ratio; subsequently, the natural logarithm of each non-zero energy distribution ratio is taken, multiplied by itself, and the product of all frequency points is summed and negative to obtain the energy distribution entropy value of the frame. The preset concentration reference value can be predetermined based on the modulation method and channel bandwidth used by the communication system. For example, in a system using orthogonal frequency division multiplexing modulation with a subcarrier spacing of 15kHz and an effective bandwidth of 10MHz, the theoretical minimum energy entropy value can be calculated using an ideal single-tone signal, while the maximum permissible entropy value is set to 80% of its theoretical upper limit based on the flat spectrum characteristics of the Nyquist band-limited signal.

[0088] (2) Calculate the mean of the absolute values ​​of the amplitude differences between the current time frame and the previous time frame at the same frequency point in the first amplitude spectrum. Compare the mean with the preset gradient upper limit. When the mean is less than or equal to the preset gradient upper limit, it is determined that the gradient dimension of the inter-frame amplitude change in the current time frame in the first amplitude spectrum meets the requirements. When the mean is greater than the preset gradient upper limit, it is determined that the gradient dimension of the inter-frame amplitude change in the current time frame in the first amplitude spectrum does not meet the requirements, and the amplitude quality judgment result of the current time frame in the first amplitude spectrum is abnormal.

[0089] It is worth noting that, in this embodiment, to obtain the average absolute value of the amplitude difference between the current time frame and its previous time frame at the same frequency point in the first amplitude spectrum, the amplitude values ​​of the current time frame and its immediately preceding time frame at the same frequency point should first be obtained; for each frequency point, the absolute value of the difference between the amplitude value of the current frame and the amplitude value of the previous frame is calculated; the absolute values ​​of all frequency points are added together and divided by the total number of frequency points to obtain the average absolute difference of the amplitude change between frames. The preset gradient upper limit is derived jointly based on the system sampling rate, the maximum Doppler spread, and the modulation symbol duration. For example, in a scenario where the moving speed does not exceed 120 km / h and the carrier frequency is 2.4 GHz, the maximum Doppler frequency shift is approximately 220 Hz, from which the maximum possible rate of change of the channel response between adjacent frames can be estimated; combined with the system frame interval (e.g., one millisecond), the theoretical maximum slope of the amplitude change can be calculated, and then multiplied by a safety factor of 1.5 as the preset gradient upper limit.

[0090] (3) Extract the amplitude attenuation slope of the current time frame in the transition frequency bands on both sides of the effective bandwidth of the channel in the first amplitude spectrum. Compare the absolute value of the amplitude attenuation slope with the preset smoothness lower limit. When the absolute value of the amplitude attenuation slope is greater than or equal to the preset smoothness lower limit, it is determined that the attenuation smoothness dimension of the current time frame in the first amplitude spectrum meets the requirements at the frequency band edge. When the absolute value of the amplitude attenuation slope is less than the preset smoothness lower limit, it is determined that the attenuation smoothness dimension of the current time frame in the first amplitude spectrum does not meet the requirements, and the amplitude quality judgment result of the current time frame in the first amplitude spectrum is abnormal.

[0091] It is worth noting that in this embodiment, firstly, the left and right boundary frequencies of the effective bandwidth of the channel are determined; a certain number of frequency points, such as ten points, are extended from the left boundary towards lower frequencies to form the left transition band; the same number of frequency points are extended from the right boundary towards higher frequencies to form the right transition band; the amplitude values ​​of each frequency point in the left transition band are sorted from low to high frequency, and the average value of the amplitude difference between adjacent points is calculated as the left attenuation slope; similarly, the attenuation slope of the right transition band is calculated, which is from high to low frequency; the smaller of the absolute values ​​of the left and right attenuation slopes is taken as the final amplitude attenuation slope. The preset smoothness lower limit is preset based on the roll-off characteristics of the transmit filter. For example, if a raised cosine filter with a roll-off factor of 0.22 is used, the amplitude in the transition band should attenuate from the approximate maximum value at the edge of the passband to the approximate zero value at the start of the stopband. Its theoretical attenuation rate can be analytically obtained by the Fourier transform of the filter impulse response; 70% of this theoretical attenuation rate is taken as the preset smoothness lower limit to tolerate quantization error and finite word length effect.

[0092] (4) Traverse the amplitude values ​​of all frequency points in the current time frame in the first amplitude spectrum. When the amplitude values ​​of all frequency points are greater than or equal to zero, it is determined that the current time frame in the first amplitude spectrum meets the requirement in the dimension of amplitude non-negativity. When there is a frequency point with an amplitude value less than zero, it is determined that the current time frame in the first amplitude spectrum does not meet the requirement in the dimension of amplitude non-negativity, and the amplitude quality judgment result of the current time frame in the first amplitude spectrum is abnormal.

[0093] It is worth noting that in this embodiment, zero is the lower limit determined by the mathematical definition of the amplitude spectrum. The amplitude spectrum is obtained by taking the square root of the sum of the squares of the real and imaginary parts of the complex spectrum. The result is always a value greater than or equal to zero in the real number domain. Therefore, any amplitude value less than zero violates the basic mathematical principle, indicating that an error has occurred in the data acquisition, transmission or parsing process, such as sign bit flipping, memory overflow or illegal type conversion. Using zero as the judgment boundary has a clear physical and mathematical basis and does not need to rely on empirical data.

[0094] (5) When the amplitude values ​​of all frequency points are greater than or equal to zero, and the amplitude values ​​of all frequency points are less than or equal to the preset upper limit of transmit power, the current time frame in the first amplitude spectrum is determined to meet the requirements in the dynamic range limitation dimension of transmit power. When there is a frequency point with an amplitude value less than zero, or a frequency point with an amplitude value greater than the preset upper limit of transmit power, the current time frame in the first amplitude spectrum is determined to not meet the requirements in the dynamic range limitation dimension of transmit power, and the amplitude quality judgment result of the current time frame in the first amplitude spectrum is abnormal.

[0095] It is worth noting that in this embodiment, the preset upper limit of the transmit power is a positive real number that is strictly greater than zero, which is determined by the maximum linear output capability of the radio frequency front end and the maximum allowable radiated power specified by the communication protocol.

[0096] (6) If the current time frame in the first amplitude spectrum satisfies all five dimensions mentioned above, then the amplitude quality judgment result of the current time frame in the first amplitude spectrum is determined to be qualified; otherwise, the amplitude quality judgment result of the current time frame in the first amplitude spectrum is determined to be abnormal.

[0097] It is worth noting that in this embodiment, the full satisfaction mechanism ensures that only frames that simultaneously meet the requirements of energy structure rationality, time continuity, frequency domain detail specifications, mathematical validity, and system power constraints are considered valid data that can be used for subsequent processing. This strict joint discrimination avoids situations where one dimension passes while other dimensions have serious defects, such as concentrated energy but excessive power, or non-negative but drastic inter-frame jumps. This fundamentally prevents abnormal data from contaminating subsequent correction operations and significantly improves the overall robustness and reliability of the system.

[0098] In some embodiments of this application, such as Figure 3 As shown, step 102 above corrects the first amplitude spectrum based on the amplitude quality determination result of each time frame in the first amplitude spectrum, including: Step 301: Based on the amplitude quality judgment result of each time frame in the first amplitude spectrum, divide the time frames in the first amplitude spectrum into a set of amplitude qualified frames and a set of amplitude abnormal frames.

[0099] In some implementation methods, this step includes the following process: Traverse the amplitude quality assessment results of each time frame in the first amplitude spectrum; add the time frames in the first amplitude spectrum with qualified amplitude quality assessment results to the set of qualified amplitude frames; add the time frames in the first amplitude spectrum with abnormal amplitude quality assessment results to the set of abnormal amplitude frames.

[0100] Step 302: Traverse each amplitude abnormal frame in the amplitude abnormal frame set, and perform the following operation on the current amplitude abnormal frame: use the verification results of the current amplitude abnormal frame in the amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics that do not meet the requirements of the dimension to perform correction processing on the current amplitude abnormal frame to obtain the corrected amplitude frame.

[0101] In practice, the amplitude correction operation is first generated based on the verification results of the current amplitude abnormal frame in the amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics that do not meet the requirements. The correction operation is then executed to correct the amplitude abnormal frame.

[0102] Step 303: Recombine the amplitude qualified frames and the corrected amplitude frames in the amplitude qualified frame set to obtain the corrected first amplitude spectrum.

[0103] In this step, the amplitude-qualified frames in the amplitude-qualified frame set are recombined with all the corrected amplitude frames according to the original time frame order to obtain the complete corrected first amplitude spectrum.

[0104] In some embodiments of this application, such as Figure 4 As shown, step 302 utilizes the verification results of the current amplitude anomaly frame in terms of the dimensions that do not meet the requirements in the amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to perform correction processing on the current amplitude anomaly frame, obtaining the corrected amplitude frame, including: Step 401: Based on the verification results of the current amplitude anomaly frame in amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics, determine the dimensions that do not meet the requirements. That is, identify the specific dimensions that do not meet the requirements in time-frequency energy concentration, inter-frame amplitude change gradient, frequency band edge attenuation smoothness, amplitude non-negativity, and transmit power dynamic range limitation.

[0105] Step 402: When the required dimensions are not met, including amplitude non-negativity, transmit power dynamic range limitation, inter-frame amplitude change gradient, and frequency band edge attenuation smoothness, the current amplitude abnormal frame is corrected according to the correction strategy corresponding to the unmet requirement dimension.

[0106] In some implementation methods, the specific implementation process of this step includes: (1) If the only dimension that does not meet the requirement is non-negativity of amplitude, and all other dimensions meet the requirements, then set all negative elements in the current amplitude abnormal frame to zero to complete the correction.

[0107] It is worth noting that in this embodiment, the amplitude value should not be negative in a physical sense. When only this condition is not met, but other dimensions, such as whether the energy is concentrated, whether the inter-frame changes are gradual, whether the frequency band edges are smooth, and whether the power limit is exceeded, are all normal, it indicates that the overall structure and numerical range of the frame are reasonable. The negative value is likely caused by a small error in the calculation process. In this case, setting the negative value to zero is a minimal change. It eliminates the non-definitive value and fully preserves the effective information of the frame in terms of time-frequency distribution, energy form, and dynamic range without introducing new distortion.

[0108] (2) If the only dimension that does not meet the requirements is the dynamic range limit of the transmit power, and all other dimensions meet the requirements, then all out-of-limit elements in the current amplitude abnormal frame are truncated to the preset transmit power upper limit to complete the correction.

[0109] It is worth noting that in this embodiment, the preset transmit power upper limit represents the maximum amplitude value that the signal is allowed to reach under normal working conditions. When only this item is not met, but other dimensions, such as energy concentration, smooth inter-frame changes, smooth edges, and no negative values ​​are normal, it indicates that the overall signal form of the frame is reliable, and only the amplitude of individual points is slightly higher than the reasonable upper limit. This over-limit is usually caused by instantaneous fluctuations and is not a structural error. Cutting the over-limit value to the upper limit can bring the amplitude value of the frame back to a reasonable range, while not changing its original time-frequency position, energy distribution and change trend, so as to make it conform to the physical constraints of amplitude value while maintaining data availability.

[0110] (3) If the requirement is not met in the dimension of inter-frame amplitude change gradient or frequency band edge attenuation smoothness, but the other dimensions meet the requirements, then the amplitude qualified frames adjacent to the current amplitude abnormal frame shall be used as references: If there are frames with acceptable amplitudes before and after the current frame with abnormal amplitude, then perform time-domain linear interpolation on the current frame with abnormal amplitude.

[0111] If there is a frame with acceptable amplitude on only one side, then perform unidirectional time-domain copying or frequency-domain edge smoothing filtering on the current frame with abnormal amplitude to obtain the corrected amplitude frame.

[0112] It is worth noting that in this embodiment, the inter-frame amplitude change gradient reflects the severity of amplitude changes between adjacent time frames, and the frequency band edge attenuation smoothness reflects whether the two ends of the spectrum attenuate naturally. When only one of these two items is abnormal, while the core attributes such as energy concentration, no negative values, and no exceeding limits are normal, it indicates that the main body of the signal is intact, and the problem only occurs in the details of time continuity or frequency domain boundary. By using the information of adjacent qualified frames for interpolation or duplication, a reasonable time transition can be restored. Smoothing the edges separately can repair local unnatural abrupt changes. These operations are all based on existing qualified data, do not generate information out of thin air, and do not destroy the main structure within the frame.

[0113] Step 403: When the dimension that does not meet the requirements is time-frequency energy concentration or there are two or more dimensions that do not meet the requirements, the current amplitude abnormal frame is corrected according to the amplitude qualified frame that is closest to the current amplitude abnormal frame in time.

[0114] In some implementation methods, the specific implementation process of this step includes: (1) If the set of amplitude qualified frames is not empty, select the amplitude qualified frame that is closest in time to the current amplitude abnormal frame from the set of amplitude qualified frames as the amplitude template, and replace the current amplitude abnormal frame with the amplitude template.

[0115] (2) If the set of frames with acceptable amplitude is empty, then retain the original data of the current frame with abnormal amplitude.

[0116] It is worth noting that in this embodiment, the time-frequency energy concentration measures whether the signal energy is concentrated in the expected time-frequency region. If this condition is not met, it indicates that the energy distribution of the frame is abnormally diffuse or misaligned, and the overall reliability is extremely low. Similarly, if two or more dimensions are abnormal at the same time, such as negative values, exceeding limits, and energy dispersion, it indicates that the frame has multiple inconsistencies and it is difficult to restore its rationality through local adjustments. In this case, if a qualified frame is available, the most recent qualified frame is used as a substitute, which can maintain the temporal continuity and morphological consistency of the data to the greatest extent. If there is no qualified frame, the original data is retained to avoid unreliable guesswork corrections in the absence of reliable references.

[0117] This embodiment implements a layered correction strategy based on the specific combination of abnormal dimensions (i.e., dimensions that do not meet requirements) of frames with abnormal amplitude. This includes targeted lightweight repair for single-dimensional anomalies and initiating a difference or template replacement mechanism based on adjacent qualified frames for multi-dimensional coupling anomalies or cases involving deterioration of time-frequency energy concentration. This can significantly improve the data quality input to the denoising model, enabling the subsequent denoising model composed of encoder-Transformer bottleneck layer-decoder to be jointly optimized in a high-fidelity, low-interference feature space. This avoids signal distortion or accidental deletion of key features caused by ignoring physical laws in traditional end-to-end denoising models.

[0118] In some embodiments of this application, the phase spectrum morphology discrimination indicators include: phase continuity deviation, phase jump frequency, and phase time-frequency stability. The phase spectrum physical characteristics include: phase continuity constraint and carrier phase global monotonicity.

[0119] Based on this, in some embodiments of this application, step 103 above verifies each time frame in the first phase spectrum using phase spectrum morphology discrimination index and phase spectrum physical characteristics to obtain the phase quality judgment result of each time frame in the first phase spectrum, including: Iterate through each time frame in the first phase spectrum, and perform the following operation on the current time frame: (1) Calculate the average absolute value of the phase difference between adjacent time sampling points in the current time frame in the first phase spectrum, and compare the average value with the preset continuity deviation upper limit. When the average value is less than or equal to the preset continuity deviation upper limit, it is determined that the current time frame in the first phase spectrum meets the requirements in the phase continuity deviation dimension. When the average value is greater than the preset continuity deviation upper limit, it is determined that the current time frame in the first phase spectrum meets the requirements in the phase continuity deviation dimension, and the phase quality judgment result of the current time frame in the first phase spectrum is determined to be abnormal.

[0120] It is worth noting that in this embodiment, all phase sampling points arranged in chronological order in the current time frame are obtained; starting from the second sampling point, the phase difference between it and the previous sampling point is calculated sequentially; the absolute value of each phase difference is taken; after adding all the absolute values, the average value of the absolute value of the phase difference between adjacent time sampling points is obtained by dividing by the total number of phase difference values, i.e., the total number of sampling points minus one.

[0121] The preset continuity deviation upper limit is determined by the system sampling interval and the maximum Doppler frequency shift. For example, in a scenario where the carrier frequency is 2.4 GHz and the moving speed does not exceed 60 km / h, the maximum Doppler frequency shift is about 110 Hz. If the intra-frame time sampling interval is 0.5 milliseconds, the maximum phase shift caused by channel changes between adjacent sampling points is about 0.34 radians. Multiplying this by a safety factor of 1.8, the preset continuity deviation upper limit is 0.61 radians.

[0122] (2) Calculate the proportion of sampling points in the first phase spectrum where the absolute value of the phase difference exceeds the preset transition threshold to the total number of sampling points. Compare this proportion with the preset upper limit of the transition frequency. If the proportion is less than or equal to the preset upper limit of the transition frequency, the current time frame in the first phase spectrum is determined to meet the requirements in the phase transition frequency dimension. If the proportion is greater than the preset upper limit of the transition frequency, the current time frame in the first phase spectrum is determined to not meet the requirements in the phase transition frequency dimension, and the phase quality judgment result of the current time frame in the first phase spectrum is determined to be abnormal.

[0123] It is worth noting that in this embodiment, the absolute value of the phase difference between all adjacent time sampling points calculated in the previous step (1) is used; the number of differences greater than the preset jump threshold is counted; and the number is divided by the total number of phase differences to obtain the proportion of jump sampling points.

[0124] The preset phase jump threshold is set to 3.0 radians. This value originates from the boundary characteristics of the phase principal value interval from negative π to positive π. When the true phase crosses positive or negative π, the undecoupled phase value will jump from positive π to negative π, or vice versa, resulting in a jump of approximately 6.28 radians. However, considering quantization errors and noise effects, the actual observed jump is usually slightly less than 6.28 radians. Therefore, setting the jump threshold to 3.0 radians can effectively distinguish between true phase jumps and normal slow changes, while avoiding misjudging small fluctuations as jumps. At the same time, the preset upper limit of the jump frequency is set to 5%. This value is determined based on the physical causes of phase jumps in typical communication channels, such as sudden interference and symbol synchronization loss. Under normal working conditions, phase jumps should be extremely rare. If the jump ratio exceeds 5%, it indicates that the data is very likely to be severely interfered with or have parsing errors.

[0125] (3) Calculate the standard deviation of the phase value within the time-frequency local window of the current time frame in the first phase spectrum, and compare the standard deviation with the preset stability upper limit. When the standard deviation is less than or equal to the preset stability upper limit, it is determined that the current time frame in the first phase spectrum meets the requirements in the phase-time-frequency stability dimension. When the standard deviation is greater than the preset stability upper limit, it is determined that the current time frame in the first phase spectrum does not meet the requirements in the phase-time-frequency stability dimension, and the phase quality judgment result of the current time frame in the first phase spectrum is abnormal.

[0126] It is worth noting that in this embodiment, on the time-frequency plane corresponding to the current time frame, a local window is selected with the current time as the center, containing 5 time sampling points and 10 frequency points before and after it, for a total of 110 phase values; the arithmetic mean of all phase values ​​in the window is calculated; then the square of the difference between each phase value and the average value is calculated; all square values ​​are added together and divided by the total number of phase values, and then the square root of the result is taken to obtain the standard deviation of the phase values.

[0127] The preset stability upper limit is set to 0.45 radians. This value is derived from the theoretical variance of phase noise under ideal static channel conditions and is corrected by the quantization noise limit of the analog-to-digital converter. Under ideal conditions of no interference and no frequency offset, the phase standard deviation is usually less than 0.3 radians. The upper limit of 0.45 radians can tolerate slight thermal noise and quantization error, while excluding obviously unstable frames.

[0128] (4) Check whether there is a case in the first phase spectrum where the absolute value of the phase difference between any adjacent time sampling points in the current time frame is greater than the preset continuity hard limit. If not, it is determined that the current time frame in the first phase spectrum meets the requirements in the phase continuity constraint dimension. If it does, it is determined that the current time frame in the first phase spectrum does not meet the requirements in the phase continuity constraint dimension, and the phase quality judgment result of the current time frame in the first phase spectrum is abnormal.

[0129] It is worth noting that in this embodiment, the absolute values ​​of the phase differences between all adjacent time sampling points calculated in step (1) above are used again, and it is checked whether any difference is greater than the preset continuity hard limit.

[0130] The preset continuity hard limit is set to 4.5 radians, which is significantly higher than the average deviation upper limit in step (1) above. It is used to capture extreme anomalies, such as data packet misalignment and phase change caused by memory out-of-bounds reading. In normal communication, even if there is an unwound transition, its single transition is close to 6.28 radians. However, such transitions should be identified by step (2) above. The 4.5 radian hard limit is intended to intercept non-physical, ultra-large transitions caused by system failures, such as phase values ​​being incorrectly assigned random numbers. This limit is determined by injecting fault tests.

[0131] (5) Extract the phase sequence of the carrier frequency in the current time frame of the first phase spectrum and fit a trend line of phase sequence change over time. If the derivative sign of the trend line remains consistent throughout the entire time interval, it is determined that the current time frame in the first phase spectrum meets the requirements in the global monotonicity dimension of the carrier phase. If the derivative sign of the trend line is inconsistent throughout the entire time interval, it is determined that the current time frame in the first phase spectrum does not meet the requirements in the global monotonicity dimension of the carrier phase, and the phase quality judgment result of the current time frame in the first phase spectrum is abnormal.

[0132] It is worth noting that in this embodiment, firstly, the phase values ​​of all time sampling points corresponding to the carrier frequency are extracted from the phase spectrum of the current time frame to form a one-dimensional phase sequence; then, the sequence is linearly fitted to obtain an optimal fitting line; the slope of the line is calculated; it is determined whether the slope is greater than zero or less than zero; if the slope is greater than zero, it is checked whether the phase difference between any two adjacent points in the original phase sequence is always greater than or equal to -0.2 radians; if the slope is less than zero, it is checked whether the phase difference between any two adjacent points is always less than or equal to +0.2 radians; if the above conditions are met, it is considered that the phase sequence as a whole shows a consistent increasing or decreasing trend; it is worth further noting that when there is a fixed frequency offset, the carrier phase should change approximately linearly with time, showing a monotonically increasing or decreasing trend, allowing small reverse fluctuations, such as ±0.2 radians, to accommodate measurement noise, but the overall trend must be consistent. If the slope is zero, that is, there is no frequency offset, the phase should be basically constant, and the adjacent difference should be within two radians of positive and negative zero, which is also considered monotonic.

[0133] (6) If the current time frame in the first phase spectrum meets the requirements in all five dimensions, the phase quality judgment result of the current time frame in the first phase spectrum is determined to be qualified; otherwise, the phase quality judgment result of the current time frame in the first phase spectrum is determined to be abnormal.

[0134] It is worth noting that in this embodiment, the five verifications construct a complete phase reliability evaluation system from five aspects: local smoothness, jump probability, spatiotemporal stability, extreme anomaly tolerance, and global semantic consistency. Failure in any dimension means that the phase information is unreliable and may mislead subsequent frequency offset estimation, channel equalization, or demodulation decisions. Therefore, adopting a full satisfaction mechanism is a necessary measure to ensure the robustness of the system.

[0135] This embodiment can perform refined quality assessment on each time frame in the phase spectrum.

[0136] In some embodiments of this application, such as Figure 5 As shown, step 103 above corrects the first phase spectrum based on the phase quality determination result of each time frame in the first phase spectrum, including: Step 501: Based on the phase quality determination result of each time frame in the first phase spectrum, divide the time frames in the first phase spectrum into a set of phase qualified frames and a set of phase abnormal frames.

[0137] In some implementation methods, this step includes the following process: Traverse the phase quality determination results of each time frame in the first phase spectrum; classify the time frames in the first phase spectrum with qualified phase quality determination results into the qualified phase frame set; classify the time frames in the first phase spectrum with abnormal phase quality determination results into the abnormal phase frame set.

[0138] Step 502: Traverse each phase anomalous frame in the set of phase anomalous frames, and perform the following operation on the current phase anomalous frame: use the verification results of the current phase anomalous frame in the phase spectrum morphology discrimination index and phase spectrum physical characteristics that do not meet the requirements of the dimension to perform correction processing on the phase anomalous frame, and obtain the corrected phase frame.

[0139] In practice, the phase correction operation is first generated based on the verification results of the current phase abnormal frame in terms of the phase spectrum morphology discrimination index and the phase spectrum physical characteristics that do not meet the requirements. The phase abnormal frame is then corrected by executing the correction operation.

[0140] Step 503: Recombine the phase-qualified frames and the corrected phase frames in the phase-qualified frame set to obtain the corrected first phase spectrum.

[0141] During this step, all phase-qualified frames in the phase-qualified frame set and the corrected phase frames are recombined according to the original time frame order to obtain the complete corrected first phase spectrum.

[0142] In some embodiments of this application, such as Figure 6 As shown, step 502 above utilizes the verification results of the current phase anomaly frame failing to meet the required dimensions in the phase spectrum morphology discrimination index and phase spectrum physical characteristics to perform correction processing on the phase anomaly frame, obtaining a corrected phase frame, including: Step 601: Based on the verification results of the current phase-abnormal frame in the phase spectrum morphology discrimination index and phase spectrum physical characteristics, determine the dimensions that do not meet the requirements. That is, identify the specific dimensions that do not meet the requirements in time-frequency energy concentration, inter-frame amplitude change gradient, frequency band edge attenuation smoothness, amplitude non-negativity, and transmit power dynamic range limitation.

[0143] Step 602: When the unmet requirement dimension includes one of the following dimensions: phase jump frequency, phase continuity deviation, phase continuity constraint, and global monotonicity of carrier phase, the current phase abnormal frame is corrected according to the correction strategy corresponding to the unmet requirement dimension.

[0144] In some implementation methods, the specific implementation process of this step includes: (1) If the only dimension that does not meet the requirements is the phase jump frequency, and all other dimensions meet the requirements, then perform local phase unwrapping and three-sampling median filtering on the phase jump points in the current phase abnormal frame to restore local continuity and complete the correction.

[0145] It is worth noting that in this embodiment, the phase jump frequency reflects the number of discontinuous phase jumps occurring per unit time. When only this aspect is abnormal, while other dimensions such as small overall continuity deviation, no local hard-limit jumps, stable time and frequency, and global monotonicity are normal, it indicates that the main phase structure is intact. The jump is likely an apparent change caused by incorrect unwrapping, rather than a true signal distortion. In this case, only unwrapping the jump point locally and correcting it with median filtering can eliminate false jumps without interfering with other stable phase values. This is a precise and low-disturbance repair method.

[0146] (2) If the only dimension that does not meet the requirements is the phase continuity deviation, and all other dimensions meet the requirements, then perform time-domain sliding window averaging filtering or low-pass phase smoothing filtering on the current phase abnormal frame to complete the correction.

[0147] It is worth noting that in this embodiment, the phase continuity deviation measures the average severity of phase changes between adjacent sampling points. When only this item slightly exceeds the limit, and there are no serious jumps, no extreme hard limit abrupt changes, and the characteristics such as time-frequency stability and global monotonicity are maintained well, it indicates that the overall phase trend is correct and there is only slight high-frequency noise or quantization jitter. At this time, by using moving average or low-pass smoothing, local fluctuations can be effectively suppressed, the average deviation can be reduced, and the long-term trend and monotonicity of phase changes can be preserved, avoiding information loss due to excessive smoothing.

[0148] (3) If the only dimension that does not meet the requirements is the phase continuity constraint, and all other dimensions meet the requirements, then perform global phase unwrapping on the current phase abnormal frame and force the phase difference between adjacent sampling points to be less than or equal to the preset continuity hard limit to complete the correction.

[0149] It is worth noting that in this embodiment, the phase continuity constraint is used to intercept extreme abnormal single jumps. When only this condition is not met, while other dimensions, including average deviation, jump frequency, stability, and monotonicity, are normal, it indicates that the problem is likely caused by a false large jump due to unwrapping, or by instantaneous interference at individual sampling points. The true continuous trajectory of the phase can be recovered by global unwrapping. Furthermore, by forcing the adjacent difference not to exceed the hard limit, non-physical abrupt changes can be completely eliminated. This operation ensures that local changes meet the basic continuity requirements while preserving the overall phase evolution law.

[0150] (4) If the only dimension that does not meet the requirements is the global monotonicity of the carrier phase, and all other dimensions meet the requirements, then extract the carrier phase trajectories of the two nearest qualified phase frames before and after the current phase abnormal frame, fit the extracted carrier phase trajectory to obtain the monotonic phase trend line at the current moment, and then shift and align the phase of all frequency points of the current phase abnormal frame along the monotonic phase trend line to complete the correction.

[0151] It is worth noting that in this embodiment, the global monotonicity of the carrier phase requires the carrier phase to show a consistent increasing or decreasing trend over time. This is a typical characteristic when there is a fixed frequency offset. When only this condition is not met, but the local continuity, jumps, stability, etc. are normal, it indicates that the phase data itself is of good quality, but the overall trend may deviate due to the initial phase offset or short-term disturbance. By reconstructing the trend line using qualified frames before and after and rigidly shifting the phase of the entire frame, the correct global monotonicity can be restored without changing the internal relative relationship, and no new discontinuities or distortions will be introduced.

[0152] Step 603: When the dimension that does not meet the requirements is phase time-frequency stability or there are two or more dimensions that do not meet the requirements, the current phase abnormal frame is corrected according to the phase qualified frame that is closest to the current phase abnormal frame in time.

[0153] In some implementations, this step includes the following process: (1) If there are phase-qualified frames before and after the current phase-abnormal frame, then the current phase-abnormal frame is subjected to temporal phase interpolation based on the two phase-qualified frames that are closest in time to each other.

[0154] (2) If there is a phase-qualified frame on only one side, then copy the nearest phase-qualified frame on that side as the replacement for the current phase-abnormal frame.

[0155] (3) If the set of qualified phase frames is empty, the original data of the current abnormal phase frame is retained.

[0156] It is worth noting that in this embodiment, phase time-frequency stability reflects the degree of phase fluctuation within the local time-frequency window. If this condition is not met, it indicates that there are uncontrollable random disturbances in both the time and frequency dimensions of the phase, resulting in low overall reliability. Similarly, if two or more anomalies occur simultaneously, such as high jumps, large deviations, and non-monotonicity, it indicates that the frame has multiple inconsistencies and it is difficult to restore its physical rationality through local correction. In this case, if a qualified frame is available, bilateral interpolation can be used to reconstruct a reasonable phase transition; unilateral copying provides a suboptimal but stable alternative; if there is no qualified frame, the original data is retained to avoid unreliable guesswork corrections in the absence of reliable references.

[0157] This embodiment implements a hierarchical correction strategy based on the specific combination of abnormal dimensions (i.e., dimensions that do not meet requirements) of the phase-abnormal frames. This includes targeted lightweight repair for single-dimensional anomalies and the initiation of a difference or template replacement mechanism based on adjacent qualified frames for multi-dimensional coupling anomalies or cases involving phase time-frequency stability degradation. This can significantly improve the data quality input to the denoising model, enabling the subsequent denoising model composed of encoder-Transformer bottleneck layer-decoder to be jointly optimized in a high-fidelity, low-interference feature space. This avoids signal distortion or accidental deletion of key features caused by ignoring physical laws in traditional end-to-end denoising models.

[0158] Based on the same inventive concept, this application also provides a satellite laser communication signal noise joint suppression device, as described in the following embodiments. Since the principle of the satellite laser communication signal noise joint suppression device in solving the problem is similar to that of the satellite laser communication signal noise joint suppression method, the implementation of the satellite laser communication signal noise joint suppression device can refer to the satellite laser communication signal noise joint suppression method, and repeated details will not be elaborated further.

[0159] Specifically, such as Figure 7 As shown, the satellite laser communication signal noise joint suppression device includes: The first transformation unit 701 is configured to perform a short-time Fourier transform on the time-domain signal of satellite laser communication to obtain a first amplitude spectrum and a first phase spectrum. The first correction unit 702 is configured to verify each time frame of the first amplitude spectrum using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain amplitude quality judgment result of each time frame of the first amplitude spectrum; and to correct the first amplitude spectrum based on the amplitude quality judgment result of each time frame of the first amplitude spectrum. The second correction unit 703 is configured to verify each time frame in the first phase spectrum using phase spectrum morphology discrimination index and phase spectrum physical characteristics to obtain the phase quality judgment result of each time frame in the first phase spectrum; and to correct the first phase spectrum based on the phase quality judgment result of each time frame in the first phase spectrum. The denoising unit 704 is configured to input the corrected first amplitude spectrum and the corrected first phase spectrum to the two input terminals of the pre-established denoising model, and output the second amplitude spectrum and the second phase spectrum from the output terminal of the denoising model. The second transformation unit 705 is configured to perform inverse short-time Fourier transform on the second amplitude spectrum and the second phase spectrum to obtain the denoised satellite laser communication time domain signal; The denoising model includes: a first encoder, a second encoder, a fusion module, a phase-aware Transformer module, a first decoder, and a second decoder.

[0160] The input terminals of the first encoder and the second encoder constitute the input terminals of the denoising model, used to extract features from the corrected first amplitude spectrum and the corrected first phase spectrum, respectively, to obtain a first feature map and a second feature map. The fusion module concatenates the first and second feature maps to obtain a first fused feature tensor. The phase-aware Transformer module obtains a second fused feature tensor based on the first fused feature tensor. The output terminals of the first decoder and the second decoder constitute the output terminals of the denoising model, used to generate a second amplitude spectrum and a second phase spectrum based on the second fused feature tensor, respectively.

[0161] This application introduces a spectral layering correction mechanism based on multi-dimensional physical criteria (amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics, phase spectrum morphology discrimination index and phase spectrum physical characteristics) at the deep learning front end. This effectively overcomes the shortcomings of existing methods in modeling the unique structure of laser communication signals. It can not only accurately identify and repair various anomalies in the amplitude and phase spectra, ensuring that the input data strictly conforms to physical realizability, but also provide a high-quality and structurally reasonable feature foundation for the joint optimization of the U-shaped network and phase-aware Transformer model of the subsequent denoising model. This ensures that it can maintain excellent noise suppression capability and signal fidelity under extreme conditions such as strong turbulence and low signal-to-noise ratio, significantly reduce the risk of bit errors, improve link stability and robustness, and fully meet the stringent requirements of future high-reliability satellite laser communication systems for signal integrity and anti-interference performance.

[0162] In some embodiments of this application, a computer device is also provided, such as... Figure 8 As shown, computer device 802 may include one or more processors 804, such as one or more central processing units (CPUs), each of which may implement one or more hardware threads. Computer device 802 may also include any memory 806 for storing information of any kind, such as code, settings, data, etc. Non-limitingly, for example, memory 806 may include any type of RAM, any type of ROM, flash memory, hard disk, optical disk, etc. More generally, any memory can use any technology to store information. Furthermore, any memory may provide volatile or non-volatile retention of information. Furthermore, any memory may represent a fixed or removable component of computer device 802. In one case, when processor 804 executes associated instructions stored in any memory or combination of memories, computer device 802 may perform any operation of the associated instructions. Computer device 802 also includes one or more drive mechanisms 808 for interacting with any memory, such as hard disk drive mechanisms, optical disk drive mechanisms, etc.

[0163] Computer device 802 may also include an input / output module 810 (I / O) for receiving various inputs (via input device 812) and providing various outputs (via output device 814). A specific output mechanism may include a presentation device 816 and an associated graphical user interface 818 (GUI). In other embodiments, the input / output module 810 (I / O), input device 812, and output device 814 may be omitted, and the device may function solely as a computer device within a network. Computer device 802 may also include one or more network interfaces 820 for exchanging data with other devices via one or more communication links 822. One or more communication buses 824 couple the components described above together.

[0164] Communication link 822 can be implemented in any way, such as via a local area network, a wide area network (e.g., the Internet), a point-to-point connection, or any combination thereof. Communication link 822 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc., governed by any protocol or combination of protocols.

[0165] In some embodiments of this application, a computer-readable storage medium, such as a non-transient computer-readable storage medium, is also provided, on which a computer program is stored, which, when executed by a processor, performs the methods described in any of the above embodiments.

[0166] In some embodiments of this application, a computer-readable instruction is also provided, wherein when a processor executes the instruction, the program therein causes the processor to perform the method described in any of the above embodiments.

[0167] In some embodiments of this application, a computer program product is also provided, including a computer program that, when run by a processor, executes the methods described in any of the above embodiments.

[0168] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0169] It should also be understood that, in the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this application generally indicates that the preceding and following related objects have an "or" relationship.

[0170] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0171] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0172] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, apparatuses, or units, or they may be electrical, mechanical, or other forms of connection.

[0173] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0174] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0175] 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 this application, in essence, or the part 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0176] This application uses specific embodiments to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for joint noise suppression of satellite laser communication signals, characterized in that, include: A short-time Fourier transform is performed on the time-domain signal of satellite laser communication to obtain the first amplitude spectrum and the first phase spectrum; The amplitude spectrum is verified by using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain the amplitude quality judgment result of each time frame in the first amplitude spectrum; the first amplitude spectrum is corrected based on the amplitude quality judgment result of each time frame in the first amplitude spectrum. The phase spectrum morphology discrimination index and phase spectrum physical characteristics are used to verify each time frame of the first phase spectrum to obtain the phase quality judgment result of each time frame of the first phase spectrum; the first phase spectrum is corrected based on the phase quality judgment result of each time frame of the first phase spectrum. The corrected first amplitude spectrum and the corrected first phase spectrum are input to the two input terminals of the pre-established denoising model, and the second amplitude spectrum and the second phase spectrum are output from the output terminal of the denoising model. Perform inverse short-time Fourier transform on the second amplitude spectrum and the second phase spectrum to obtain the denoised satellite laser communication time-domain signal; The denoising model includes: a first encoder, a second encoder, a fusion module, a phase-aware Transformer module, a first decoder, and a second decoder. The input terminals of the first encoder and the second encoder constitute the input terminals of the denoising model, and are used to extract features from the corrected first amplitude spectrum and the corrected first phase spectrum, respectively, to obtain a first feature map and a second feature map. The fusion module is used to concatenate the first feature map and the second feature map to obtain a first fused feature tensor. The phase-aware Transformer module is used to obtain a second fused feature tensor based on the first fused feature tensor. The output terminals of the first decoder and the second decoder constitute the output terminals of the denoising model, and are used to generate the second amplitude spectrum and the second phase spectrum based on the second fused feature tensor, respectively.

2. The method as described in claim 1, characterized in that, The amplitude spectrum morphology discrimination indicators include: time-frequency energy concentration, inter-frame amplitude variation gradient, and frequency band edge attenuation smoothness; The amplitude spectrum physical characteristics include amplitude nonnegativity and dynamic range limitation of transmit power.

3. The method as described in claim 2, characterized in that, The amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics are used to verify each time frame of the first amplitude spectrum, and the amplitude quality judgment result of each time frame of the first amplitude spectrum is obtained, including: Iterate through each time frame in the first amplitude spectrum, and perform the following operation on the current time frame: Calculate the energy distribution entropy value of the current time frame in the first amplitude spectrum across the entire frequency band, compare the energy distribution entropy value with a preset concentration reference value, and determine that the current time frame in the first amplitude spectrum meets the requirements in the time-frequency energy concentration dimension when the energy distribution entropy value is less than or equal to the preset concentration reference value. Calculate the mean of the absolute value of the amplitude difference between the current time frame and the previous time frame at the same frequency point in the first amplitude spectrum, compare the mean with the preset gradient upper limit, and when the mean is less than or equal to the preset gradient upper limit, determine that the gradient dimension of the inter-frame amplitude change in the current time frame in the first amplitude spectrum meets the requirements. Extract the amplitude attenuation slope of the current time frame in the transition frequency band on both sides of the effective bandwidth of the channel in the first amplitude spectrum, compare the absolute value of the amplitude attenuation slope with a preset smoothness lower limit, and determine that the attenuation smoothness dimension of the current time frame in the first amplitude spectrum meets the requirements when the absolute value of the amplitude attenuation slope is greater than or equal to the preset smoothness lower limit. Traverse the amplitude values ​​of all frequency points in the first amplitude spectrum for the current time frame. When the amplitude values ​​of all frequency points are greater than or equal to zero, determine that the current time frame in the first amplitude spectrum meets the requirement in the dimension of amplitude non-negativity. When the amplitude values ​​of all frequency points are greater than or equal to zero, and the amplitude values ​​of all frequency points are less than or equal to the preset upper limit of transmit power, it is determined that the current time frame in the first amplitude spectrum meets the requirements in the dimension of transmit power dynamic range limitation. If the current time frame in the first amplitude spectrum meets all the above dimensions, then the amplitude quality judgment result of the current time frame in the first amplitude spectrum is determined to be qualified; otherwise, the amplitude quality judgment result of the current time frame in the first amplitude spectrum is determined to be abnormal.

4. The method as described in claim 2, characterized in that, Based on the amplitude quality determination result of each time frame in the first amplitude spectrum, the first amplitude spectrum is corrected, including: Based on the amplitude quality determination result of each time frame in the first amplitude spectrum, the time frames in the first amplitude spectrum are divided into a set of amplitude qualified frames and a set of amplitude abnormal frames. Traverse each amplitude abnormal frame in the set of amplitude abnormal frames, and perform the following operation on the current amplitude abnormal frame: use the verification results of the current amplitude abnormal frame that does not meet the required dimensions in the amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to perform correction processing on the current amplitude abnormal frame to obtain the corrected amplitude frame. The amplitude-qualified frames and the corrected amplitude frames in the amplitude-qualified frame set are recombined to obtain the corrected first amplitude spectrum.

5. The method as described in claim 4, characterized in that, Based on the amplitude quality assessment result of each time frame in the first amplitude spectrum, the time frames in the first amplitude spectrum are divided into a set of amplitude-qualified frames and a set of amplitude-abnormal frames, including: Iterate through the amplitude quality determination results for each time frame in the first amplitude spectrum; The time frames in the first amplitude spectrum whose amplitude quality determination result is qualified are included in the amplitude qualified frame set. The time frames in the first amplitude spectrum whose amplitude quality determination result is abnormal are classified into the amplitude abnormal frame set.

6. The method as described in claim 4, characterized in that, Using the verification results of the current amplitude anomaly frame failing to meet the requirements in the amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics, the current amplitude anomaly frame is corrected to obtain a corrected amplitude frame, including: Based on the verification results of the current amplitude anomaly frame in the amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics, the dimensions that do not meet the requirements are determined; When the unmet requirement dimension includes one of the following dimensions: amplitude non-negativity, transmit power dynamic range limitation, inter-frame amplitude change gradient, and frequency band edge attenuation smoothness, the current amplitude abnormal frame is corrected according to the correction strategy corresponding to the unmet requirement dimension. When the dimension that does not meet the requirements is time-frequency energy concentration or there are two or more dimensions that do not meet the requirements, the current amplitude abnormal frame is corrected according to the amplitude qualified frame that is closest to the current amplitude abnormal frame in time.

7. The method as described in claim 6, characterized in that, When the unmet requirement dimension includes one of the following dimensions: amplitude non-negativity, transmit power dynamic range limitation, inter-frame amplitude variation gradient, and frequency band edge attenuation smoothness, the current amplitude abnormal frame is corrected according to the correction strategy corresponding to the unmet requirement dimension, including: If the dimension that does not meet the requirement is only amplitude non-negativity, then set all negative elements in the current amplitude abnormal frame to zero; If the dimension that does not meet the requirements is only the dynamic range limitation of the transmit power, then all out-of-limit elements in the current amplitude abnormal frame are truncated to the preset transmit power upper limit; If the dimension that is not met is the gradient of inter-frame amplitude change or the smoothness of frequency band edge attenuation, then the amplitude-qualified frames adjacent to the current amplitude-abnormal frame shall be used as a reference: If there are amplitude-qualified frames before and after the current amplitude-abnormal frame, then perform temporal linear interpolation on the current amplitude-abnormal frame; If there is a frame with acceptable amplitude on only one side, then perform unidirectional time-domain copying or frequency-domain edge smoothing filtering on the current frame with abnormal amplitude to obtain a corrected amplitude frame.

8. The method as described in claim 6, characterized in that, When the dimension that fails to meet the requirement is time-frequency energy concentration, or when there are two or more dimensions that fail to meet the requirement, the current amplitude abnormal frame is corrected based on the amplitude qualified frame that is closest to the current amplitude abnormal frame in time, including: If the set of amplitude qualified frames is not empty, then select the amplitude qualified frame that is closest in time to the current amplitude abnormal frame from the set of amplitude qualified frames as the amplitude template, and use the amplitude template to replace the current amplitude abnormal frame; If the set of amplitude-qualified frames is empty, then the original data of the current amplitude-abnormal frame is retained.

9. The method as described in claim 1, characterized in that, The phase spectrum morphology discrimination indicators include: phase continuity deviation, phase jump frequency, and phase time-frequency stability; The phase spectrum physical characteristics include: phase continuity constraint and global monotonicity of carrier phase.

10. The method as described in claim 9, characterized in that, The phase quality determination result for each time frame in the first phase spectrum is obtained by verifying each time frame using phase spectrum morphology discrimination index and phase spectrum physical characteristics, including: For each time frame in the first phase spectrum, perform the following operation on the current time frame: Calculate the average value of the absolute value of the phase difference between adjacent time sampling points in the current time frame in the first phase spectrum, compare the average value with a preset continuity deviation upper limit, and determine that the current time frame in the first phase spectrum meets the requirements in the phase continuity deviation dimension when the average value is less than or equal to the preset continuity deviation upper limit. The proportion of sampling points in the first phase spectrum where the absolute value of the phase difference exceeds a preset transition threshold to the total number of sampling points is counted. This proportion is compared with a preset upper limit of transition frequency. When the proportion is less than or equal to the preset upper limit of transition frequency, it is determined that the current time frame in the first phase spectrum meets the requirements in the dimension of phase transition frequency. The standard deviation of the phase value is calculated within the time-frequency local window of the current time frame in the first phase spectrum. The standard deviation is compared with a preset stability upper limit. When the standard deviation is less than or equal to the preset stability upper limit, it is determined that the current time frame in the first phase spectrum meets the requirements in the phase time-frequency stability dimension. Check whether there is a case in the first phase spectrum where the absolute value of the phase difference between any adjacent time sampling points in the current time frame is greater than the preset continuity hard limit. If not, it is determined that the current time frame in the first phase spectrum meets the requirements in the phase continuity constraint dimension. Extract the phase sequence of the carrier frequency in the current time frame of the first phase spectrum, and fit a trend line of the phase sequence changing with time. If the derivative sign of the trend line remains consistent throughout the entire time interval, it is determined that the current time frame in the first phase spectrum meets the requirements in the global monotonicity dimension of the carrier phase. If the current time frame in the first phase spectrum meets the requirements in all the above dimensions, then the phase quality judgment result of the current time frame in the first phase spectrum is determined to be qualified; otherwise, the phase quality judgment result of the current time frame in the first phase spectrum is determined to be abnormal.

11. The method as described in claim 9, characterized in that, Based on the phase quality determination result of each time frame in the first phase spectrum, the first phase spectrum is corrected, including: Based on the phase quality determination result of each time frame in the first phase spectrum, the time frames in the first phase spectrum are divided into a set of phase qualified frames and a set of phase abnormal frames. Traverse each phase anomalous frame in the set of phase anomalous frames, and perform the following operation on the current phase anomalous frame: use the verification results of the current phase anomalous frame that does not meet the required dimensions in the phase spectrum morphology discrimination index and phase spectrum physical characteristics to perform correction processing on the phase anomalous frame to obtain the corrected phase frame. The phase-qualified frames and the corrected phase frames in the set of phase-qualified frames are recombined to obtain the corrected first phase spectrum.

12. The method as described in claim 11, characterized in that, Based on the phase quality determination result of each time frame in the first phase spectrum, the time frames in the first phase spectrum are divided into a set of phase-qualified frames and a set of phase-abnormal frames, including: The phase quality determination results of each time frame in the first phase spectrum are traversed. The time frames in the first phase spectrum where the phase quality determination result is qualified are included in the phase qualified frame set. The time frames in the first phase spectrum where the phase quality determination result is abnormal are included in the phase abnormal frame set.

13. The method as described in claim 11, characterized in that, Using the verification results of the current phase anomaly frame failing to meet the required dimensions in the phase spectrum morphology discrimination index and phase spectrum physical properties, the phase anomaly frame is corrected to obtain a corrected phase frame, including: Based on the verification results of the current phase anomaly frame in the phase spectrum morphology discrimination index and phase spectrum physical characteristics, the dimensions that do not meet the requirements are determined; When the unmet requirement dimension includes one of the following dimensions: phase jump frequency, phase continuity deviation, phase continuity constraint, and global monotonicity of carrier phase, the current phase abnormal frame is corrected according to the correction strategy corresponding to the unmet requirement dimension. When the dimension that does not meet the requirements is phase time-frequency stability or there are two or more dimensions that do not meet the requirements, the current phase abnormal frame is corrected according to the phase qualified frame that is closest to the current phase abnormal frame in time.

14. The method as described in claim 13, characterized in that, When the unmet requirement dimension includes one of the following dimensions: phase jump frequency, phase continuity deviation, phase continuity constraint, and carrier phase global monotonicity, the current phase abnormal frame is corrected according to the correction strategy corresponding to the unmet requirement dimension, including: If the dimension that does not meet the requirements is only the phase jump frequency, then local phase unwrapping and three-sampling-point median filtering are performed on the phase jump points in the current phase abnormal frame. If the dimension that does not meet the requirements is only the phase continuity deviation, then perform temporal sliding window averaging filtering or low-pass phase smoothing filtering on the current phase abnormal frame. If the dimension that does not meet the requirements is only the phase continuity constraint, then global phase unwrapping is performed on the current phase abnormal frame, and the phase difference between adjacent sampling points is forced to be less than or equal to the preset continuity hard limit. If the dimension that does not meet the requirements is only the global monotonicity of the carrier phase, then the carrier phase trajectories of the two nearest qualified phase frames before and after the current phase abnormal frame are extracted, the monotonic phase trend line at the current moment is obtained by fitting the extracted carrier phase trajectory, and the phase of all frequency points of the current phase abnormal frame is offset and aligned along the monotonic phase trend line.

15. The method as described in claim 13, characterized in that, When the dimension that fails to meet the requirement is phase-time frequency stability, or when there are two or more dimensions that fail to meet the requirement, the current phase-abnormal frame is corrected based on the phase-qualified frame that is closest in time to the current phase-abnormal frame, including: If there are phase-qualified frames before and after the current phase-abnormal frame, then time-domain phase interpolation is performed on the current phase-abnormal frame based on the two most recent phase-qualified frames. If a phase-qualified frame exists on only one side, then the nearest phase-qualified frame on that side is copied as a replacement for the current phase-abnormal frame. If the set of qualified phase frames is empty, then the original data of the current abnormal phase frame is retained.

16. A satellite laser communication signal noise joint suppression device, characterized in that, include: The first transformation unit is configured to perform a short-time Fourier transform on the time-domain signal of satellite laser communication to obtain a first amplitude spectrum and a first phase spectrum. The first correction unit is configured to verify each time frame of the first amplitude spectrum using amplitude spectrum morphology discrimination index and amplitude spectrum physical characteristics to obtain amplitude quality judgment result of each time frame of the first amplitude spectrum; and to correct the first amplitude spectrum based on the amplitude quality judgment result of each time frame of the first amplitude spectrum. The second correction unit is configured to verify each time frame of the first phase spectrum using phase spectrum morphology discrimination index and phase spectrum physical characteristics to obtain the phase quality judgment result of each time frame of the first phase spectrum; and to correct the first phase spectrum based on the phase quality judgment result of each time frame of the first phase spectrum. The denoising unit is configured to input the corrected first amplitude spectrum and the corrected first phase spectrum to the two input terminals of a pre-established denoising model, and output the second amplitude spectrum and the second phase spectrum from the output terminal of the denoising model. The second transformation unit is configured to perform inverse short-time Fourier transform on the second amplitude spectrum and the second phase spectrum to obtain the denoised satellite laser communication time domain signal; The denoising model includes: a first encoder, a second encoder, a fusion module, a phase-aware Transformer module, a first decoder, and a second decoder. The input terminals of the first encoder and the second encoder constitute the input terminals of the denoising model, and are used to extract features from the corrected first amplitude spectrum and the corrected first phase spectrum, respectively, to obtain a first feature map and a second feature map. The fusion module is used to concatenate the first feature map and the second feature map to obtain a first fused feature tensor. The phase-aware Transformer module is used to obtain a second fused feature tensor based on the first fused feature tensor. The output terminals of the first decoder and the second decoder constitute the output terminals of the denoising model, and are used to generate the second amplitude spectrum and the second phase spectrum based on the second fused feature tensor, respectively.

17. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 15.

18. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor of a computer device, it implements the method according to any one of claims 1 to 15.

19. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor of a computer device, it implements the method according to any one of claims 1 to 15.

Images