Characteristics analysis and processing method of mesoscale time-shortwave signals based on multimodal fusion

Abstract

The present invention discloses a method for analyzing and processing the characteristics of mesoscale time-wave signals based on multimodal fusion, which belongs to the technical field of wireless communication signal processing. It solves the problems of dynamics, enhanced signal-time coupling, and characteristic instability caused by dynamic characteristics in existing feature extraction methods of mesoscale time-wave signals. The method for analyzing and processing the characteristics of mesoscale time-wave signals based on multimodal fusion of the present invention includes extracting statistical features in the time domain and frequency domain, extracting time-frequency distribution graph features, and extracting one-dimensional time series dynamic features to form a multi-level feature analysis framework, and utilizes the complementarity between the time domain and frequency domain statistical features, the time-frequency distribution graph features, and the one-dimensional time series dynamic features to fuse the above-mentioned extracted multiple modal features, effectively improving the description ability of signal features, and at the same time improving the accuracy of modulation recognition of mesoscale time-wave signals.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication signal processing, and in particular to a method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion. Background Art

[0002] Modulation recognition (MR) is a critical component of wireless communication systems. It identifies the modulation scheme of a received signal with limited or no prior knowledge. Whether a primary task in electronic reconnaissance or enabling link adaptation in civilian communications, modulation recognition provides fundamental support for subsequent signal processing.

[0003] Shortwave communications have attracted widespread attention due to their unique propagation characteristics. However, due to the complex channel environment caused by ionospheric reflection, shortwave signals exhibit characteristics such as multipath, fading, and Doppler shift, making their feature extraction more difficult. Traditional shortwave signal feature extraction methods mainly rely on signal processing and statistical analysis techniques. These methods extract signal features from different angles such as the time domain and frequency domain, providing a basis for signal analysis and identification. However, these methods still have the following problems for feature extraction of mesoscale shortwave signals:

[0004] (1) The dynamic changes of the channel characteristics of the medium-scale time shortwave signal introduce time-varying characteristics, which causes the channel conditions to change randomly. The dynamic characteristics of the amplitude change and spectrum broadening affect its stability, resulting in increased signal uncertainty.

[0005] (2) The time-varying characteristics of the mesoscale shortwave channel make the signal characteristics interdependent between time segments, resulting in enhanced signal-time coupling.

[0006] (3) Due to the weakening of inter-modal coupling in medium-scale time-shortwave signals, the performance differences of different modal characteristics increase under channel changes, making it difficult to fuse inter-modal information; and due to the random changes in signal amplitude and phase, channel fading and noise superposition lead to the difficulty of stably representing the signal with a single modal feature. Summary of the Invention

[0007] In response to the shortcomings and deficiencies in the prior art, the present invention provides a method for analyzing and processing the characteristics of mesoscale time-wave signals based on multimodal fusion, which can effectively capture the dynamic changes of signals in time series, accurately describe and compensate for the dynamic characteristics of signals, significantly enhance the stability and robustness of modal characteristics, and significantly improve the accuracy of signal modulation recognition.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: The method for analyzing and processing the characteristics of mesoscale time-frequency shortwave signals based on multimodal fusion provided by the present invention adopts a comprehensive characterization of the static properties and dynamic change characteristics of the signal to mine the multimodal characteristics of the signal at multiple levels, specifically including extracting statistical features in the time and frequency domains, deep learning features of the time-frequency distribution diagram, and one-dimensional time series dynamic features, to form a multi-level feature analysis framework;

[0009] The complementarity of the above-mentioned different modal features is used to fuse the extracted multimodal features, thereby improving the ability to describe signal characteristics, improving the accuracy of modulation recognition of medium-scale time shortwave signals, and improving the performance of the recognition system under complex channel conditions.

[0010] Preferably, the method specifically includes the following steps:

[0011] S1. Dataset Generation: First, modulated signal simulation is performed, including amplitude modulation signals, frequency modulation signals, phase modulation signals, and orthogonal amplitude modulation signals, from which shortwave signals are obtained. The log-normal distribution is used to describe medium-term and long-term channel variations. Medium-term and long-term channel variation models are introduced to generate a mesoscale time shortwave signal dataset.

[0012] S2. Multimodal feature extraction: Extract statistical features in the time and frequency domains, extract time-frequency distribution features, and extract one-dimensional time series dynamic features, as follows:

[0013] The time-domain and frequency-domain statistical features describe the overall distribution of the signal and are used to distinguish the average amplitude and spectrum distribution of the above-mentioned different modulation methods. The time-frequency distribution features effectively capture modulated signals with sudden changes or local characteristics through the joint representation of time and frequency. The one-dimensional time series dynamic features are used to capture the evolution pattern and dependence of the signal in time, effectively distinguish the above-mentioned different phase modulation methods, and learn the interaction between amplitude and phase changes in orthogonal modulation signals.

[0014] The mesoscale time shortwave signal dataset generated in step S1 is used to extract the multidimensional characteristics of the mesoscale time shortwave signal using the above-mentioned multiple modes, forming a multi-level feature analysis framework to comprehensively characterize the static properties and dynamic change characteristics of the shortwave signal;

[0015] S3, multimodal feature fusion: The multimodal features extracted in step S2 are fused by utilizing the complementarity between the time-domain and frequency-domain statistical features, the time-frequency distribution features, and the one-dimensional time series dynamic features to improve the ability to describe signal features and the accuracy of modulation recognition of mesoscale time-shortwave signals. The specific steps are as follows:

[0016] The three modal features of time domain and frequency domain statistical features, time-frequency distribution diagram features and one-dimensional time series dynamic features complement each other according to their respective characteristics, and together constitute a complete representation of the modulation modes of amplitude modulated signals, frequency modulated signals, phase modulated signals and orthogonal amplitude modulated signals, effectively improving the accuracy of signal modulation recognition.

[0017] Preferably, in step S1, the amplitude modulation signal includes 2ASK, 4ASK, and 8ASK; the frequency modulation signal includes 2FSK and 4FSK; the phase modulation signal includes BPSK and QPSK; and the quadrature amplitude modulation signal includes 16QAM.

[0018] The medium-term and long-term channel variation models are introduced, wherein the mesoscale effect is obtained by means of the LTV and ITV models, and its time constant and standard deviation are obtained according to the measured data. The LTV and ITV models are then combined to obtain a signal-to-noise ratio (SNR) sequence. N data points are selected from the generated signal-to-noise ratio (SNR) sequence, and the N SNR points are sampled at intervals of 1 second. The Watterson model is then combined to ensure that each signal-to-noise ratio (SNR) value data point corresponds to a specific time period, ensuring that the signal is affected by different SNRs at different time points, thereby simulating the real channel environment and ultimately generating a complete data set containing N seconds of mesoscale time shortwave signals.

[0019] Preferably, the method includes step S2.1 of extracting statistical features in the time domain and frequency domain, including extracting statistical features in the time domain and extracting statistical features in the frequency domain;

[0020] The time domain statistical features include mean value, skewness, and kurtosis factor; the frequency domain statistical features include frequency domain amplitude mean value, centroid frequency, and frequency standard deviation;

[0021] The six time-domain and frequency-domain statistical features extracted above are normalized to make the multimodal features more compatible and consistent during the fusion process. The specific calculation formula is:

[0022]

[0023] Among them, x norm is the normalized standard feature; x is the original feature; μ is the average value of each feature; α is the standard deviation of each feature.

[0024] Preferably, (1) the average value is an indicator of the overall DC offset or base value of the signal, which can distinguish whether the DC component of the signal has a significant offset and reflect the overall DC offset degree of the modulated signal:

[0025] The trend of average value changes caused by different modulation depths used on different signals can clearly distinguish 2ASK, 4ASK, 8ASK, 2FSK, and 16QAM. Specifically, the amplitude of the 2ASK, 4ASK, and 8ASK amplitude modulation signals determines the change in their average values. Different frequency components in the 2FSK frequency modulation signal cause the average value to change. The average value of the 16QAM orthogonal amplitude modulation signal reflects the distribution characteristics of the constellation diagram points.

[0026] The average value is used to distinguish the above different signal modulation methods. The specific calculation formula is:

[0027]

[0028] Where S i is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample;

[0029] (2) Skewness is used to measure the degree of asymmetry of the signal waveform. For FM signals, skewness is used to describe whether there is an asymmetric amplitude distribution characteristic that deviates from the mean in the signal. Because the spectrum shows slight asymmetry during the switching process between different frequencies, the 2FSK FM signal can be clearly distinguished. Therefore, skewness is used to distinguish FM signals from shortwave signals. The specific calculation formula is:

[0030]

[0031] Where S i is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; ρ t is the standard deviation, and the formula is as follows:

[0032]

[0033] ρ t 3 is the cube of the standard deviation;

[0034] (3) The kurtosis factor describes the sharpness of the signal amplitude and is an important feature in time domain analysis. It is especially effective for detecting impulse signals, pulse signals, and high-peak modulation signals.

[0035] When sudden amplitude changes occur in the AM signal and rapid jumps between different frequencies occur in the FM signal, the kurtosis factor will increase significantly. Signals with high kurtosis contain more spikes or high-energy transient components, which can clearly distinguish 2ASK, 4ASK, 2FSK, and 4FSK. Specifically, sudden amplitude changes in 2ASK and 4ASK AM signals lead to increased kurtosis; rapid jumps between different frequencies in 2FSK and 4FSK signals increase kurtosis.

[0036] The specific calculation formula is:

[0037]

[0038] Where S i is the amplitude of the signal’s sampling data points; N is the number of sampling data points for each sample.

[0039] Preferably, (4) the frequency domain amplitude average value reflects the central trend of the signal spectrum amplitude distribution, describes the energy concentration in the spectrum, and is used to distinguish different modulation modes;

[0040] The frequency domain amplitude average value can clearly distinguish 2ASK, 4ASK, 8ASK, 2FSK, 4FSK, BPSK, QPSK, and 16QAM, as follows: the spectrum amplitude distribution of 2ASK, 4ASK, and 8ASK amplitude modulation signals has significant characteristics; the spectrum energy of 2FSK and 4FSK keying signals is concentrated at a specific frequency position; BPSK, QPSK, and 16QAM phase modulation or composite modulation signals have different energy distributions in the frequency domain, so the frequency domain amplitude average value can be used as a basis for identification;

[0041] The specific calculation formula is:

[0042]

[0043] Where k is the spatial frequency of the wave; s(k) represents the frequency domain amplitude at the kth frequency point;

[0044] (5) The center of gravity frequency describes the dominant frequency position of the signal spectrum and is an important indicator for measuring spectrum distribution. The center of gravity frequency positions of different signals are significantly different, and 2FSK, 4FSK, BPSK, and QPSK can be clearly distinguished as follows: The different frequency components of 2FSK and 4FSK frequency keying signals determine the position of their center of gravity frequencies; the spectrum distribution center of BPSK and QPSK phase modulated signals is closely related to their modulation methods. The specific calculation formula is:

[0045]

[0046] Where k is the spatial frequency; s(k) represents the frequency domain amplitude at the kth frequency point; f k is the frequency value of the kth frequency point;

[0047] (6) Frequency standard deviation describes the degree of dispersion or concentration of the spectrum. It is one of the key features to distinguish signals of different modulation modes. It can clearly distinguish 2FSK, 4FSK, BPSK, and QPSK. Specifically, the frequency hopping signals of 2FSK and 4FSK will cause the frequency standard deviation to change. The bandwidth of BPSK and QPSK modulation signals will directly affect the frequency standard deviation. The specific calculation formula is:

[0048]

[0049] Where k is the spatial frequency; S i is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; s(k) represents the frequency domain amplitude at the kth frequency point; f k is the frequency value of the kth frequency point; S2 is the centroid frequency of the kth point.

[0050] Preferably, the method further includes step S2.2 of extracting time-frequency distribution features, specifically using a ResNet152 network to extract features of two time-frequency distribution features: SPWVD (smoothed pseudo Wigner-Ville distribution) time-frequency distribution features and BJD (Born-Jordan distribution) time-frequency distribution features;

[0051] The SPWVD time-frequency distribution feature captures local features through high resolution, which is used to observe the changes of signals within a specific time range and improve the recognition ability of signal features.

[0052] The specific calculation formula is:

[0053] SPWVD x (t,f)=∫∫x(t-v+τ / 2)x * (tv-τ / 2)·h(τ)g(v)e -j2πfτ dvdτ

[0054] In the formula, SPWVD x (t,f) is the time-frequency result of SPWVD time-frequency distribution characteristics, τ is the frequency domain window length; v is the time domain window length; h(τ) is the frequency domain smoothing window; g(v) is the time domain smoothing window; x * represents the complex conjugate of x; x(t) is the analytical signal of r(t), which is expressed as:

[0055] x(t)=r(t)+jH[r(t)]

[0056] Where H[·] represents the Hilbert transform; r(t) is the received signal;

[0057] The BJD time-frequency distribution feature is used to provide a global time-frequency description of the signal, which can accurately concentrate the signal energy around its true time and frequency position, making the signal's spectral characteristics more obvious. The specific calculation formula is:

[0058]

[0059] Where, BJD x(t,f) is the time-frequency result in the BJD time-frequency distribution feature; τ is the frequency domain window length, v is the time domain window length; x * represents the complex conjugate of x;

[0060] where φ(t,τ) is:

[0061] Preferably, step S2.2 further includes fusing the BJD time-frequency distribution features and the SPWVD time-frequency distribution features, a total of 2048-dimensional features, by using the JS divergence (Jensen–Shannon divergence), as follows:

[0062] First, the features obtained by SPWVD and BJD are converted into probability distributions p and q using the softmax function, and then the JS divergence between them is calculated to capture details while not losing the overall signal; the calculation formula is as follows:

[0063]

[0064] Where KL(P||Q) is the KL divergence (Kullback-Leibler divergence). Let P and Q be the two probability distributions. Assuming they are continuous random variables, their corresponding probability density functions are p(x) and q(x). p is the feature obtained by SPWVD converted to a probability distribution using the softmax function; q is the feature obtained by BJD converted to a probability distribution using the softmax function.

[0065] JS divergence represents the similarity of two time-frequency distribution features. The larger the divergence value, the less similar the two time-frequency distribution features are, and the smaller the weight assigned; the smaller the divergence value, the more similar the two time-frequency distribution features are, and the larger the weight assigned; each shortwave signal sample is assigned different weights according to the size of the divergence between the multimodal features, and the above multimodal features are weighted averaged according to the corresponding weights to obtain a fused high-dimensional feature vector.

[0066] Preferably, step S2.3 is included to extract the dynamic features of the one-dimensional time series, specifically including using a gated recurrent unit (GRU) to capture the dynamic changes and trends in the one-dimensional time series signal, and obtain the characteristics of the signal so that it can better selectively remember which information should be retained and which information should be forgotten. This helps to avoid the common gradient vanishing or gradient exploding problems in recurrent neural networks (RNNs) and is used to process dynamic signals with long-range dependencies. The gated recurrent unit GRU controls the flow of information by introducing an update gate and a reset gate. The specific calculation formula is as follows:

[0067] zt =σ(W z ·[h t-1 ,x t ])

[0068] r t =σ(W r ·[h t-1 ,x t ])

[0069]

[0070] Where x t Enter information for the current moment; h t-1 is the hidden state at the previous moment; h t is the hidden state passed to the next moment; is the candidate hidden state; r t is the reset gate, z t is the update gate, σ is the sigmoid function, which converts the data into a value in the range of 0-1. The tanh function converts the data into a value in the range of [-1, 1].

[0071] Preferably, in step S3, feature splicing and fully connected layer processing is to splice the time-domain and frequency-domain statistical features, the time-frequency distribution map features and the one-dimensional time series features by dimension to form a high-dimensional feature vector. During the splicing process, all features can be processed in a unified feature space;

[0072] The concatenated feature vector is processed by the fully connected layer. The fully connected layer uses the weight matrix to perform linear transformation and nonlinear activation on the input features, adopts ReLU as the activation function, combines and re-encodes each dimension of the input features through calculation, and learns the nonlinear relationship between different features, thereby improving the performance of the recognition system under complex channel conditions.

[0073] The method for analyzing and processing medium-scale shortwave signal characteristics based on multimodal fusion provided by the present invention has the following beneficial effects:

[0074] (1) The method for analyzing and processing the characteristics of medium-scale shortwave signals based on multimodal fusion of the present invention, by extracting time-frequency distribution characteristics, time-domain and frequency-domain statistical characteristics, and one-dimensional time series dynamic characteristics, mines the multidimensional characteristics of the signal from multiple levels, comprehensively characterizes the static properties and dynamic change characteristics of the signal, and solves the difficult problems of dynamics, enhanced signal-time coupling, and characteristic instability existing in medium-scale time shortwave signals. The method also makes full use of the complementarity between time-frequency distribution characteristics, time-domain and frequency-domain statistical characteristics, and one-dimensional time series dynamic characteristics to perform multimodal fusion, so that it has the high resolution of the time-frequency distribution diagram, the time series modeling capability of the gated loop, and the robustness of the statistical characteristics, forming a multi-dimensional, multi-perspective signal analysis model, improving the overall performance of the signal, enhancing the characteristic stability and robustness, while effectively improving the recognition accuracy of the signal modulation, and improving the performance of the recognition system under complex channel conditions.

[0075] (2) By extracting the mean value, skewness, kurtosis factor, frequency domain amplitude mean value, and center of gravity frequency from the time domain and frequency domain statistical features, a comprehensive representation of the medium-scale shortwave signal in the time domain and frequency domain is provided, which can better understand the changes of the signal in a dynamic environment; and normalize it so that different modal features have better compatibility and consistency in the fusion process, thereby improving the effect of feature fusion.

[0076] (3) Extracting time-frequency distribution features: The BJD time-frequency distribution features provide a global time-frequency description of the signal, while the SPWVD time-frequency distribution features can capture local features with high resolution. The two are fused through the JS divergence to further optimize the clarity of the time-frequency representation. The two work together to more comprehensively analyze the dynamic behavior of the signal, capturing details while maintaining the overall signal. Furthermore, a neural network is used to integrate statistical features in the time and frequency domains to dynamically adjust the signal processing strategy to adapt to changing channel conditions, effectively solving the problem of instability caused by dynamic characteristics.

[0077] (4) Extracting One-Dimensional Time Series Features: A gated recurrent unit (GRU) is used to capture dynamic changes and trends in one-dimensional time series signals. GRU is a recurrent neural network suitable for time series data. It can capture dynamic changes and trends in one-dimensional time series signals, improve the accuracy and reliability of feature extraction, and process dynamic signals with long-range dependencies. By introducing update gates and reset gates to control the flow of information, it enables efficient and selective retention or forgetting of information, helping to avoid the common gradient vanishing or gradient exploding problems in recurrent neural networks (RNNs), thereby better processing long sequence data. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 is a flow chart of the method in Example 1 of the present invention;

[0079] Figure 2 This is a flow chart of the signal-to-noise ratio (SNR) sequence within the mesoscale time in Example 2;

[0080] Figure 3 This is a simulation diagram of the signal-to-noise ratio (SNR) sequence within the mesoscale time in Example 2;

[0081] Figure 4 This is a simulation diagram comparing the short-time shortwave signal and the medium-scale shortwave signal in this embodiment 2.

[0082] Figure 5 This is the average value simulation diagram in this embodiment 2;

[0083] Figure 6 This is the skewness simulation diagram in Example 2;

[0084] Figure 7 This is a simulation diagram of the kurtosis factor in Example 2;

[0085] Figure 8 This is a simulation diagram of the frequency domain amplitude average value in this embodiment 2;

[0086] Figure 9 This is a simulation diagram of the center of gravity frequency in Example 2;

[0087] Figure 10 This is a frequency standard deviation simulation diagram in Example 2;

[0088] Figure 11 This is a structural diagram of the multimodal fusion network model in Example 2;

[0089] Figure 12 This is a comparison chart of the accuracy curve and loss curve of the time-frequency distribution features and the statistical features extracted after multimodal fusion in Example 3;

[0090] Figure 13 This is a comparison chart of the accuracy curve and loss curve of the one-dimensional time series feature and the statistical feature after extracting multimodal fusion in this embodiment 3;

[0091] Figure 14 Schematic diagram of the confusion matrix after the fusion of the three modalities in Example 3. DETAILED DESCRIPTION

[0092] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0093] Example 1

[0094] The present invention provides a method for analyzing and processing the characteristics of mesoscale time-frequency shortwave signals based on multimodal fusion, which includes extracting statistical features in the time and frequency domains, extracting time-frequency distribution diagram features, and extracting one-dimensional time series dynamic features to form a multi-level feature analysis framework. The complementarity between different modal features is utilized to fuse the above-mentioned extracted multiple modal features, thereby improving the ability to describe signal characteristics and the accuracy of modulation recognition of mesoscale time-frequency shortwave signals.

[0095] like Figure 1 As shown, the present invention performs feature analysis on mesoscale time-wave signals and finds that the current feature extraction of mesoscale time-wave signals mainly has the following difficulties: First, the dynamic changes in the channel characteristics in the mesoscale time-wave signals introduce time-varying characteristics, causing the channel conditions to change randomly, resulting in the dynamic characteristics of drastic changes in amplitude and spectrum broadening, which affects its stability and increases the uncertainty of the signal. Secondly, the time-varying characteristics of the mesoscale time-wave channel make the signal characteristics interdependent between time segments, resulting in enhanced signal-time coupling. The traditional assumption that time segments are independent is no longer applicable, which brings challenges to signal processing and feature extraction. Thirdly, due to the weakening of inter-modal coupling of mesoscale time-wave signals, the performance differences of different modal features increase under channel changes, making it difficult to fuse inter-modal information, and the superposition of channel fading and noise due to random changes in signal amplitude, phase, etc., making it difficult for a single modal feature to stably characterize the signal.

[0096] The present invention addresses the above difficult problems. First, it extracts statistical features in the time domain and frequency domain to provide a comprehensive representation of the signal in the time domain and frequency domain, so as to better understand the changes in the signal in a dynamic environment. The time-frequency distribution features are extracted to accurately describe and compensate for the dynamic characteristics of the signal. At the same time, the signal processing strategy is dynamically adjusted by combining deep learning technology with the attention mechanism and fusing the extracted time domain and frequency domain statistical features to adapt to the ever-changing channel conditions. The dynamic behavior of the signal can be analyzed more comprehensively, while capturing details without losing the overall signal characteristics. It is used to solve the problem of instability caused by dynamic characteristics. Secondly, by extracting one-dimensional time series features and using time series modeling technology, the dynamic changes and trends in the one-dimensional time series signal can be captured, and the characteristics of the signal can be obtained to process dynamic signals with long-range dependencies, thereby improving the accuracy and reliability of feature extraction and effectively solving the problem of enhanced signal-time coupling due to time-varying characteristics. Finally, multimodal fusion technology is used to perform multimodal feature fusion by utilizing the complementarity between time-frequency distribution features, time-domain and frequency-domain statistical features, and one-dimensional time series dynamic features. The multimodal model not only combines the high-resolution characteristics of the time-frequency distribution graph and the time series modeling capabilities of the gated loop, but also utilizes the robustness of statistical features to construct a multi-dimensional, multi-perspective signal analysis, improve the overall signal performance, enhance feature stability and robustness, and is used to solve the problem of feature instability caused by dynamic channels.

[0097] Example 2

[0098] like Figure 2-Figure 11 As shown, the method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion of the present invention specifically includes the following steps:

[0099] S1. Dataset generation: Perform modulation signal simulation, including amplitude modulation signals 2ASK, 4ASK, 8ASK, frequency modulation signals 2FSK, 4FSK, phase modulation signals BPSK, QPSK and orthogonal amplitude modulation signals 16QAM, etc., to obtain shortwave signals;

[0100] like Figure 2 As shown in the figure, based on the above shortwave signals, the dynamic change characteristics of the ionospheric channel are fully considered, the log-normal distribution is used to describe the medium-term and long-term channel changes, and the medium-term and long-term channel change models are introduced. Among them, the mesoscale effect is achieved with the help of LTV and ITV models. According to the measured data: the time constants are 160 seconds and 6.5 seconds, the standard deviation of LTV is 3.493dB, and the standard deviation of ITV is 3.889dB. It is realized by combining Alpha filters and high-pass filters, and then the signal-to-noise ratio SNR sequence is obtained.

[0101] like Figure 3As shown in the figure, 160 data points are selected from the generated SNR sequence, and 160 SNR points are sampled at intervals of 1 second. Combined with the Watterson model, each data point value in the signal-to-noise ratio (SNR) corresponds to a specific time period, ensuring that the signal is affected by different data point values ​​at different time points. This simulates the real channel environment and ultimately generates a mesoscale time shortwave signal dataset with a duration of 160 seconds.

[0102] like Figure 4 As shown, the spectrum of the output signal of the mesoscale channel model generated by the present invention exhibits more obvious spectrum broadening than that of the Watterson model, resulting in energy dispersion. Furthermore, in terms of amplitude variation, the amplitude variation of the mesoscale time shortwave signal is more dramatic than that of the short-time shortwave signal. This accurately reflects the changes in channel quality caused by various fading over a long period of time.

[0103] S2. Multimodal feature extraction: including extracting statistical features in the time and frequency domains, extracting time-frequency distribution features, and extracting one-dimensional time series dynamic features; using the mesoscale time shortwave signal dataset generated in step S1, the multi-dimensional characteristics of the mesoscale time shortwave signal are extracted using the above-mentioned multiple modes to form a multi-level feature analysis framework to comprehensively characterize the static properties and dynamic change characteristics of the shortwave signal.

[0104] S2.1 Extraction of time domain and frequency domain statistical features. A total of six statistical features were selected. The time domain statistical features include mean value, skewness, and kurtosis factor; the frequency domain statistical features include frequency domain amplitude mean value, centroid frequency, and frequency standard deviation.

[0105] like Figure 5 As shown, the average value is an indicator of the overall DC offset of the signal or the base value of the signal, which can reflect the overall DC offset degree of the modulated signal. For some baseband signals or amplitude modulated signals, different modulation depths may cause changes in the average value. As can be seen from the figure, the amplitude of the amplitude modulated signals such as 2ASK, 4ASK, and 8ASK determines the change in the average value; the different frequency components of the frequency modulated signals represented by 2FSK and 4FSK will cause the average value to change; the average value of the orthogonal amplitude modulation represented by 16QAM reflects the distribution characteristics of the constellation diagram points. Therefore, the average value can more clearly distinguish 2ASK, 4ASK, 8ASK, 2FSK, and 16QAM. The specific calculation formula is:

[0106]

[0107] Where S i is the amplitude of the signal’s sampling data points; N is the number of sampling data points for each sample.

[0108] like Figure 6As shown in the figure, skewness can measure the degree of asymmetry of the signal waveform. Since the switching between different frequencies in the FM signal makes the spectrum slightly asymmetric, skewness is used to describe whether there is an asymmetric amplitude distribution characteristic that deviates from the mean in the signal, distinguishing the FM signal 2FSK. The specific calculation formula is:

[0109]

[0110] Where S i is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; ρ t is the standard deviation, and the formula is as follows:

[0111]

[0112] ρ t 3 is the cube of the standard deviation;

[0113] like Figure 7 As shown in the figure, the kurtosis factor can describe the sharpness of the signal amplitude and is an important feature in time domain analysis. It is especially effective for detecting impact signals, pulse signals, and high-peak modulation signals. Signals with high kurtosis usually contain more spikes or high-energy instantaneous components. When there are sudden amplitude changes in the amplitude modulated signal and signals that jump rapidly between different frequencies in the frequency modulated signal, the kurtosis factor will increase significantly. As can be seen in the figure, the sudden amplitude changes in the amplitude modulated signals such as 2ASK and 4ASK lead to increased kurtosis. The rapid jump signals between different frequencies in the frequency modulated signals such as 2FSK and 4FSK will increase the kurtosis. Therefore, the kurtosis factor can more clearly distinguish 2ASK, 4ASK, 2FSK, and 4FSK. The specific calculation formula is:

[0114]

[0115] Where S i is the amplitude of the signal’s sampling data points; N is the number of sampling data points for each sample.

[0116] like Figure 8 As shown, the frequency domain amplitude average value reflects the central trend of the signal spectrum amplitude distribution and can describe the degree of energy concentration in the spectrum. As can be seen from the figure, 2ASK, 4ASK, 8ASK: The spectrum amplitude distribution of the amplitude modulated signal has significant characteristics, with a peak of the same amplitude on each side of the carrier peak; the spectrum energy of frequency modulated signals such as 2FSK and 4FSK is concentrated at a specific frequency position; BPSK, QPSK, 16QAM and other phase modulation or composite modulation signals have different energy distribution in the frequency domain. Therefore, the frequency domain amplitude average value is used to identify the above different signal modulation methods. The specific calculation formula is:

[0117]

[0118] Where k is the spatial frequency of the wave; s(k) represents the frequency domain amplitude at the kth frequency point.

[0119] like Figure 9 As shown in the figure, the center of gravity frequency can describe the dominant frequency position of the signal spectrum and is an important indicator for measuring the spectrum distribution. As can be seen in the figure, the different frequency components in the 2FSK and 4FSK frequency modulation signals determine their center of gravity frequency position; the spectrum center distribution of BPSK and QPSK phase modulation signals is the same as their modulation method. The center of gravity frequency is used to clearly distinguish the above 2FSK, 4FSK, BPSK, and QPSK signal modulation methods. The specific calculation formula is:

[0120]

[0121] Where k is the spatial frequency; s(k) represents the frequency domain amplitude at the kth frequency point; f k is the frequency value of the kth frequency point;

[0122] like Figure 10 As shown in the figure, the frequency standard deviation can describe the distribution width of the spectrum and is one of the key features to distinguish signals with different modulation modes. As can be seen in the figure, the frequency standard deviation of 2FSK and 4FSK frequency modulation signals changes with the change of different frequency hopping signals. The frequency standard deviation of BPSK and QPSK phase modulation signals changes with the change of bandwidth. The frequency standard deviation is used to clearly distinguish the above 2FSK, 4FSK, BPSK, and QPSK signal modulation modes. The specific calculation formula is:

[0123]

[0124] Where k is the spatial frequency; S i is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; s(k) represents the frequency domain amplitude at the kth frequency point; f k is the frequency value of the kth frequency point; S2 is the centroid frequency of the kth point.

[0125] The six time-domain and frequency-domain statistical features extracted above are normalized to make different modal features more compatible and consistent during the fusion process, thereby improving the effect of feature fusion.

[0126] The specific calculation formula is:

[0127]

[0128] Among them, x normis the normalized standard feature; x is the original feature; μ is the average value of each feature; α is the standard deviation of each feature.

[0129] S2.2 extracts the time-frequency distribution features, including SPWVD time-frequency distribution features and BJD time-frequency distribution features; the SPWVD time-frequency distribution features can provide good time and frequency resolution at the same time, making it excellent in capturing the rapid frequency changes of the signal. This is especially important for frequency-modulated signals 2FSK and 4FSK, because these signals have more frequent frequency jumps on the time axis. Moreover, SPWVD is particularly suitable for observing changes in signals within a specific time range. Amplitude-modulated signals 2ASK and 4ASK have sudden amplitude changes at certain moments, and SPWVD can clearly display these changes and improve the ability to identify signal features; to improve the ability to identify signal features, the calculation formula is:

[0130] SPWVD x (t,f)=∫∫x(t-v+τ / 2)x * (tv-τ / 2)·h(τ)g(v)e -j2 π fτ dvdτ

[0131] In the formula, SPWVD x (t,f) is the time-frequency result of SPWVD time-frequency distribution characteristics, τ is the frequency domain window length; v is the time domain window length; h(τ) is the frequency domain smoothing window; g(v) is the time domain smoothing window; x * represents the complex conjugate of x; x(t) is the analytical signal of r(t), which is expressed as:

[0132] x(t)=r(t)+jH[r(t)]

[0133] Where H[·] represents the Hilbert transform; r(t) is the received signal;

[0134] The BJD time-frequency distribution feature can better concentrate the energy of the signal around its true time and frequency position, making the spectral characteristics of the signal more obvious. This is especially important for phase-modulated signals BPSK and QPSK, because the spectral characteristics of these signals are more complex. Moreover, BJD can more accurately reflect the spectral characteristics and phase changes of the phase-modulated signal, which helps to better understand the modulation method of the signal. For example, in the QPSK signal, BJD can effectively display the distribution of its four phase states in the time and frequency domain, which is beneficial to the signal differentiation in the actual demodulation process. For complex modulation methods such as 16QAM, BJD can provide a clearer signal feature representation, which is used to provide a global time-frequency description of the signal, and can accurately concentrate the signal energy around its true time and frequency position, making the spectral characteristics of the signal more obvious. The specific calculation formula is:

[0135] BJD x (t,f)=∫∫x(v+τ / 2)x * (v-τ / 2)·φ(tv,τ)e -j2πfτ dvdτ

[0136] Where, BJD x (t,f) is the time-frequency result in the BJD time-frequency distribution feature; τ is the frequency domain window length, v is the time domain window length; x * represents the complex conjugate of x;

[0137] where φ(t,τ) is:

[0138] By combining the two time-frequency distribution features of SPWVD and BJD, the characteristics of different modulation modes can be analyzed more comprehensively and accurately, providing strong support for signal recognition. The fusion of BJD time-frequency distribution features and SPWVD time-frequency distribution features through JS divergence can further optimize the clarity of the time-frequency representation, making the extraction of signal features more refined. BJD provides a global time-frequency description of the signal, while SPWVD captures local features through high resolution. The combination of the two can more comprehensively analyze the dynamic behavior of the signal, capturing details while also maintaining the overall signal. The calculation formula is as follows:

[0139]

[0140] KL(P||Q) is the KL divergence (Kullback-Leibler divergence). We set two probability distributions, P and Q, as continuous random variables. Their corresponding probability density functions are p(x) and q(x), respectively. p is the feature obtained by SPWVD converted to a probability distribution using the softmax function; q is the feature obtained by BJD converted to a probability distribution using the softmax function.

[0141] JS divergence represents the similarity between two time-frequency distribution features. A larger divergence value indicates a greater dissimilarity between the two time-frequency distribution features, and therefore a smaller weight is assigned. A smaller divergence value indicates a greater similarity between the two time-frequency distribution features, and therefore a larger weight is assigned. Each shortwave signal sample is assigned a different weight based on the divergence between its different modal features. The weighted average of these modal features is then taken according to their corresponding weights to produce a fused 2048-dimensional high-dimensional feature vector.

[0142] S2.3 Extraction of one-dimensional time series dynamic features specifically includes the use of a gated recurrent unit (GRU) with 128 hidden units in the GRU layer to capture dynamic changes and trends in one-dimensional time series signals, obtain signal features, and process dynamic signals with long-range dependencies. The gated recurrent unit (GRU) controls the flow of information by introducing update gates and reset gates, enabling it to better selectively remember which information should be retained and which should be forgotten. This helps avoid the common gradient vanishing or gradient exploding problems in recurrent neural networks (RNNs), thereby better processing long sequence data. The specific calculation formula is as follows:

[0143] z t =σ(W z ·[h t-1 ,x t ])

[0144] r t =σ(W r ·[h t-1 ,x t ])

[0145]

[0146] Where x t Enter information for the current moment; h t-1 is the hidden state at the previous moment; h t is the hidden state passed to the next moment; is the candidate hidden state; r t is the reset gate, z t is the update gate, σ is the sigmoid function, which can convert data into a value in the range of 0-1; the tanh function can convert data into a value in the range of [-1,1].

[0147] AM signals such as 2ASK, 4ASK, and 8ASK are characterized by amplitude variations that occur over time. The GRU learns the patterns and regularities of these amplitude variations, capturing dynamic changes in signal amplitude. This allows it to better identify these dynamic changes, especially when the signal amplitude is affected by channel factors such as fading. Furthermore, through a gating mechanism, the GRU memorizes the start and end times of signal amplitude changes, as well as the duration of different amplitude states. These are key features for identifying AM signals.

[0148] Frequency hopping is a characteristic of 2FSK and 4FSK FM signals. The GRU learns the timing of frequency hopping, the relationship between frequencies, and the duration of each frequency state. Furthermore, the GRU can identify changes in signal frequency at different time points, thereby determining the frequency switching pattern of the modulated signal and helping to distinguish different FM signal modulation methods.

[0149] Phase jumps are characteristic of phase-modulated signals such as BPSK and QPSK. The GRU learns the timing of phase jumps, the relationship between phases, and the duration of phase states. Furthermore, the GRU captures the pattern of signal phase changes, such as the amplitude of the sudden phase change and the interval between jumps, which helps distinguish different modulation methods of phase-modulated signals.

[0150] For quadrature modulation signals like 16QAM, both the amplitude and phase of the signal vary, and these variations are related to each other. The GRU learns the dynamic relationships and interactions between these amplitude and phase variations. Therefore, the GRU can capture the complex amplitude and phase variation patterns in 16QAM signals and distinguish between different modulation states.

[0151] S3. Multimodal feature fusion: The multimodal features extracted in step S2 are fused using the complementarity between time-domain and frequency-domain statistical features, time-frequency distribution features, and one-dimensional time series dynamic features. Time-domain and frequency-domain statistical features describe the overall distribution of the signal and are used to distinguish the average amplitude and spectrum distribution of different modulation modes. Time-frequency distribution features effectively capture modulated signals with sudden changes or local characteristics through a joint representation of time and frequency. One-dimensional time series dynamic features are used to capture the temporal evolution pattern and dependency of the signal, effectively distinguish different phase modulation modes, and learn the interaction between amplitude and phase changes in orthogonal modulation signals.

[0152] The three modal features of time domain and frequency domain statistical features, time-frequency distribution diagram features and one-dimensional time series dynamic features complement each other according to their respective characteristics, and together constitute a complete representation of the modulation modes of amplitude modulation signals 2ASK, 4ASK, 8ASK, frequency modulation signals 2FSK, 4FSK, phase modulation signals BPSK, QPSK and orthogonal amplitude modulation signals 16QAM, effectively improving the accuracy of signal modulation recognition.

[0153] like Figure 11 As shown, feature fusion and concatenation, along with fully connected layer processing, fuse and concatenate time-domain and frequency-domain statistical features, time-frequency distribution features, and one-dimensional time series features by dimension, forming a high-dimensional feature vector of 2182 dimensions. This concatenation process enables all features to be processed in a unified feature space. The concatenated feature vector is processed by a fully connected layer, which uses a weight matrix to perform linear transformation and nonlinear activation on the input features. Using ReLU as the activation function, the fully connected layer computationally combines and recodes each dimension of the input features, learning the nonlinear relationships between different features and thus improving the recognition system's performance under complex channel conditions.

[0154] Example 3

[0155] In order to verify the complementarity between the time-domain and frequency-domain statistical features, the time-frequency distribution graph features and the one-dimensional time series features, the present invention conducts a comparative test on the statistical features after the various modal features extracted above are fused with the multimodal features.

[0156] like Figure 12 As shown in FIG, a comparison curve is made between the accuracy and loss of the time-frequency distribution graph features and the statistical features after extracting multimodal fusion in this embodiment;

[0157] like Figure 13 As shown in FIG, a comparison curve is made between the accuracy and loss of the one-dimensional time series dynamic features and the statistical features extracted after multimodal fusion in this embodiment.

[0158] Depend on Figure 12-13 It can be seen that after multimodal feature fusion of time-domain and frequency-domain statistical features, time-frequency distribution graph features and one-dimensional time series features, dynamic features can be further extracted. Especially for signal modulation modes such as 2ASK, 4ASK, and 8ASK, by combining the time-domain and frequency-domain statistical features with the features of the time-frequency distribution graph and the resolution characteristics of the one-dimensional time series features, it shows significant advantages in the dynamic identification of signals and can more effectively distinguish these modulation modes. It not only improves the model's ability to analyze complex signals, but also can capture the dynamic characteristics of the signal even when the signal amplitude changes drastically.

[0159] In particular, Figure 12 After fusing the extracted multimodal features, the feature accuracy of the time-frequency distribution map extracted by ResNet152 alone is improved by 22.37%, and the loss is reduced by 0.5. Figure 13 After the multimodal features extracted from the GRU are fused, the performance is improved by 10.5% and the loss is reduced by 0.2 compared to the one-dimensional temporal dynamic features extracted by GRU alone.

[0160] Figure 14 This paper provides a schematic diagram of the confusion matrix obtained by further fusing the three modalities of time-domain and frequency-domain statistical features, time-frequency distribution features, and one-dimensional time series features. The fusion of the extracted multimodal features effectively extracts the dynamic characteristics of mesoscale signals, and the recognition rate of mesoscale shortwave signals can reach over 99.88%. The fusion of statistical features in the time and frequency domains not only improves the model's ability to analyze complex signals, but also demonstrates significant advantages in dynamic signal recognition.

[0161] In summary, the present invention extracts key signal features: time-frequency distribution features, time-domain and frequency-domain statistical features, and one-dimensional time series dynamic features, and utilizes the complementarity between the above modal features for multimodal fusion. It solves the difficult problems of medium-scale time shortwave signals being highly dynamic due to the drastic changes in amplitude and the broadening of the spectrum, the mutual dependence of the signal characteristics of each time segment in the signal leading to enhanced coupling, and the difficulty of a single modal feature to stably characterize the signal characteristics. It improves the ability to describe signal features while effectively improving the recognition accuracy of signal modulation.

[0162] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion, characterized in that: By comprehensively characterizing the static properties and dynamic change characteristics of the signal, the multimodal features of the signal are mined at multiple levels. Specifically, this includes extracting statistical features in the time and frequency domains, deep learning features of time-frequency distribution graphs, and one-dimensional time series dynamic features, forming a multi-level feature analysis framework. The complementarity of different modal features is used to fuse the extracted multimodal features, thereby improving the ability to describe signal features, increasing the accuracy of modulation recognition of mesoscale time-shortwave signals, and improving the performance of the recognition system under complex channel conditions. The specific steps include: S1. Dataset Generation: First, modulated signal simulation is performed, including amplitude modulation signals, frequency modulation signals, phase modulation signals, and orthogonal amplitude modulation signals, from which shortwave signals are obtained. The log-normal distribution is used to describe medium-term and long-term channel variations. Medium-term and long-term channel variation models are introduced to generate a mesoscale time shortwave signal dataset. The amplitude modulation signal includes 2ASK, 4ASK, and 8ASK; the frequency modulation signal includes 2FSK and 4FSK; the phase modulation signal includes BPSK and QPSK; the quadrature amplitude modulation signal includes 16QAM; The medium-term and long-term channel variation models are introduced, wherein the mesoscale effect is obtained by using the LTV and ITV models, and its time constant and standard deviation are obtained based on the measured data. The results are then implemented through a combination of an Alpha filter and a high-pass filter to obtain a signal-to-noise ratio (SNR) sequence. N data points are selected from the generated SNR sequence, and the N SNR points are sampled at intervals of 1 second. In combination with the Watterson model, each SNR value data point corresponds to a specific time period, ensuring that the signal is affected by different SNRs at different time points, thereby simulating a real channel environment and ultimately generating a complete data set containing N seconds of mesoscale time shortwave signals. S2. Multimodal feature extraction: Extract statistical features in the time and frequency domains, extract time-frequency distribution features, and extract one-dimensional time series dynamic features, as follows: Time-domain and frequency-domain statistical features describe the overall distribution of the signal and are used to distinguish the average amplitude and spectrum distribution of different modulation methods. Time-frequency distribution features effectively capture modulated signals with sudden changes or local characteristics through joint representation in time and frequency. One-dimensional time series dynamic features are used to capture the temporal evolution pattern and dependency of the signal, effectively distinguish different phase modulation methods, and learn the interaction between amplitude and phase changes in orthogonal modulation signals. The mesoscale time shortwave signal dataset generated in step S1 is used to extract the multidimensional characteristics of the mesoscale time shortwave signal using multiple modes, forming a multi-level feature analysis framework to comprehensively characterize the static properties and dynamic change characteristics of the shortwave signal; S3, multimodal feature fusion: The multimodal features extracted in step S2 are fused by utilizing the complementarity between the time-domain and frequency-domain statistical features, the time-frequency distribution features, and the one-dimensional time series dynamic features to improve the ability to describe signal features and the accuracy of modulation recognition of mesoscale time-shortwave signals. The specific steps are as follows: The three modal features of time-domain and frequency-domain statistical features, time-frequency distribution diagram features and one-dimensional time series dynamic features complement each other according to their respective characteristics, and together constitute a complete representation of the modulation modes of amplitude modulated signals, frequency modulated signals, phase modulated signals and orthogonal amplitude modulated signals, effectively improving the accuracy of signal modulation recognition.

2. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 1 is characterized in that: The method comprises the steps of extracting time domain and frequency domain statistical features, including extracting time domain statistical features and extracting frequency domain statistical features; The time domain statistical features include mean value, skewness, and kurtosis factor; the frequency domain statistical features include frequency domain amplitude mean value, centroid frequency, and frequency standard deviation; The six extracted statistical features in the time and frequency domains are normalized to make the multimodal features more compatible and consistent during the fusion process. The specific calculation formula is: ； in, is the normalized standard feature; is the original feature; is the mean value of each feature; is the standard deviation of each feature.

3. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 2 is characterized in that: (1) The average value is an indicator of the overall DC offset or base value of the signal, which can distinguish whether the DC component of the signal has a significant offset and reflect the overall DC offset degree of the modulated signal: The trend of average value changes caused by different modulation depths used on different signals can clearly distinguish 2ASK, 4ASK, 8ASK, 2FSK, and 16QAM. Specifically, the amplitude of the 2ASK, 4ASK, and 8ASK amplitude modulation signals determines the change in their average values. Different frequency components in the 2FSK frequency modulation signal cause the average value to change. The average value of the 16QAM orthogonal amplitude modulation signal reflects the distribution characteristics of the constellation diagram points. The average value is used to distinguish the above different signal modulation methods. The specific calculation formula is: ； Where, is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; (2) The skewness is used to measure the degree of asymmetry of the signal waveform. For FM signals, the skewness is used to describe whether there is an asymmetric amplitude distribution characteristic that deviates from the mean in the signal. Because the spectrum shows slight asymmetry during the switching process between different frequencies, the 2FSK FM signal can be clearly distinguished. Therefore, the skewness is used to distinguish the FM signal in the shortwave signal. The specific calculation formula is: ； Where, is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; is the standard deviation, and the formula is as follows: ； is the cube of the standard deviation; (3) The kurtosis factor describes the sharpness of the signal amplitude and is an important feature in time domain analysis. It is particularly effective for detecting impulse signals, pulse signals, and high-peak modulation signals. When sudden amplitude changes occur in the AM signal and rapid jumps between different frequencies occur in the FM signal, the kurtosis factor increases significantly. Signals with high kurtosis contain more spikes or high-energy transient components. It can clearly distinguish 2ASK, 4ASK, 2FSK, and 4FSK. Specifically, sudden amplitude changes in 2ASK and 4ASK AM signals increase kurtosis; rapid jumps between different frequencies in 2FSK and 4FSK increase kurtosis. The specific calculation formula is: ； Where, is the amplitude of the sampling data point of the signal; N is the number of sample data points for each example.

4. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 2 is characterized in that: (4) The frequency domain amplitude average value reflects the central trend of the signal spectrum amplitude distribution, describes the energy concentration in the spectrum, and is used to distinguish different modulation modes; The frequency domain amplitude average value can clearly distinguish 2ASK, 4ASK, 8ASK, 2FSK, 4FSK, BPSK, QPSK, and 16QAM, as follows: the spectrum amplitude distribution of 2ASK, 4ASK, and 8ASK amplitude modulation signals has significant characteristics; the spectrum energy of 2FSK and 4FSK keying signals is concentrated at a specific frequency position; BPSK, QPSK, and 16QAM phase modulation or composite modulation signals have different energy distributions in the frequency domain, so the frequency domain amplitude average value can be used as a basis for identification; The specific calculation formula is: ； Where k is the spatial frequency of the wave; Represents the frequency domain amplitude at the kth frequency point; (5) The center of gravity frequency describes the dominant frequency position of the signal spectrum and is an important indicator for measuring spectrum distribution. The center of gravity frequency positions of different signals are significantly different, and 2FSK, 4FSK, BPSK, and QPSK can be clearly distinguished as follows: The different frequency components of 2FSK and 4FSK frequency keying signals determine the position of their center of gravity frequencies; the spectrum distribution center of BPSK and QPSK phase modulated signals is closely related to their modulation methods, and the specific calculation formula is: ； Where k is the spatial frequency; Represents the frequency domain amplitude at the kth frequency point; is the frequency value of the kth frequency point; (6) The frequency standard deviation describes the degree of dispersion or concentration of the spectrum and is one of the key features for distinguishing signals of different modulation modes. It can clearly distinguish 2FSK, 4FSK, BPSK, and QPSK. Specifically, the frequency hopping signals of 2FSK and 4FSK will cause the frequency standard deviation to change; the bandwidth of BPSK and QPSK modulation signals will directly affect the frequency standard deviation. The specific calculation formula is: ； Where k is the spatial frequency; is the amplitude of the sampling data point of the signal; N is the number of sampling data points for each sample; Represents the frequency domain amplitude at the kth frequency point; is the frequency value of the kth frequency point; is the centroid frequency of the k-th point.

5. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 2 is characterized in that: The step S2.2 is also included to extract the time-frequency distribution features, specifically using the ResNet152 network to extract the features of the two time-frequency distribution features, SPWVD time-frequency distribution features and BJD time-frequency distribution features; The SPWVD time-frequency distribution feature captures local features through high resolution, which is used to observe the changes of signals within a specific time range and improve the recognition ability of signal features. The specific calculation formula is: ； Where, is the time-frequency result in the SPWVD time-frequency distribution characteristics, is the frequency domain window length; is the time domain window length; is the frequency domain smoothing window; is the time domain smoothing window; express The complex conjugate of for The analytical signal is expressed as: ； Where, represents the Hilbert transform; To receive signals; The BJD time-frequency distribution feature is used to provide a global time-frequency description of the signal, which can accurately concentrate the signal energy around its real time and frequency position, making the spectral characteristics of the signal more obvious. The specific calculation formula is: ； Where, is the time-frequency result of BJD time-frequency distribution characteristics; is the frequency domain window length, is the time domain window length; express The complex conjugate of in for: .

6. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 5 is characterized in that: Step S2.2 also includes fusing the BJD time-frequency distribution features and the SPWVD time-frequency distribution features, a total of 2048-dimensional features, by using JS divergence, as follows: First, the features obtained by SPWVD and BJD are converted into probability distributions p and q using the softmax function. Then, the JS divergence between them is calculated to capture details while not losing the overall signal. The calculation formula is as follows: ； ； Where, KL divergence, p is the conversion of the features obtained by SPWVD into probability distribution using softmax function; q is the conversion of the features obtained by BJD into probability distribution using softmax function; JS divergence represents the similarity of two time-frequency distribution features. The larger the divergence value, the less similar the two time-frequency distribution features are, and the smaller the weight assigned; the smaller the divergence value, the more similar the two time-frequency distribution features are, and the larger the weight assigned; each shortwave signal sample is assigned different weights according to the size of the divergence between the multimodal features, and the multimodal features are weighted averaged according to the corresponding weights to obtain a fused high-dimensional feature vector.

7. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 1 is characterized in that: The method includes step S2.3 of extracting the dynamic features of the one-dimensional time series, specifically using a gated recurrent unit (GRU) to capture the dynamic changes and trends in the one-dimensional time series signal, obtain the signal features, and process the dynamic signals with long-range dependencies. The gated recurrent unit (GRU) controls the flow of information by introducing an update gate and a reset gate. The specific calculation formula is as follows: ； ； ； ； Where, Enter information for the current moment; is the hidden state at the previous moment; is the hidden state passed to the next moment; is the candidate hidden state; To reset the gate, To update the gate, is the sigmoid function.

8. The method for analyzing and processing characteristics of mesoscale time-shortwave signals based on multimodal fusion according to claim 1 is characterized in that: In step S3, feature concatenation and fully connected layer processing are to concatenate the time-domain and frequency-domain statistical features, the time-frequency distribution features, and the one-dimensional time series features by dimension to form a high-dimensional feature vector. During the concatenation process, all features can be processed in a unified feature space. The concatenated feature vector is processed by the fully connected layer. The fully connected layer uses the weight matrix to perform linear transformation and nonlinear activation on the input features, adopts ReLU as the activation function, combines and re-encodes each dimension of the input features through calculation, and learns the nonlinear relationship between different features, thereby improving the performance of the recognition system under complex channel conditions.

Images