A method for separating and identifying co-frequency signals based on real-time spectrum analysis

Abstract

The present invention discloses a method for separating and identifying same-frequency signals based on real-time spectrum analysis, including obtaining radio frequency signals in space and preprocessing them, converting the preprocessed radio frequency signals into intermediate frequency analog signals; performing sampling and digital filtering processing based on the intermediate frequency analog signals to obtain digital intermediate frequency signals; using a delay control module to perform adaptive time slicing reorganization processing on the digital intermediate frequency signals to obtain a reorganized signal sequence; converting the reorganized signal sequence into spectrum statistical information; converting the spectrum statistical information into radio frequency-amplitude spectrum statistics, and using a threshold segmentation module to perform dynamic multi-dimensional threshold segmentation on the radio frequency-amplitude spectrum statistics to obtain signal classification results. The present invention can effectively distinguish between background signals and foreground signals, so that foreground signals with lower amplitudes are not affected by background signals and can be directly observed, thereby improving the anti-interference ability of spectrum equipment in complex electromagnetic signal environments.

Description

Technical Field

[0001] The invention belongs to the field of radar monitoring, and in particular to a same-frequency signal separation and identification method based on real-time spectrum analysis. Background Art

[0002] With the rapid development of electronic information technology, various wireless communication devices and radio frequency application systems are becoming increasingly popular, making the use of electromagnetic spectrum resources more intensive. In large-scale event venues, urban central areas, industrial parks and other scenarios, spectrum usage is highly complex and dynamically changing. As a key means of electromagnetic signal monitoring, spectrum analysis technology can analyze and intuitively display the signal's carrier frequency, amplitude, bandwidth and other characteristics, and plays an important role in spectrum resource management, signal quality assessment, interference identification and other aspects. However, with the continuous innovation of wireless technology and the diversification of application scenarios, traditional spectrum analysis methods are facing unprecedented challenges, and there is an urgent need to develop more advanced signal separation and identification technologies to adapt to the increasingly complex electromagnetic environment.

[0003] At present, spectrum analysis technology has achieved success in specific application scenarios, but there are still many technical bottlenecks when facing highly complex and dynamically changing electromagnetic environments. In practical applications, spectrum analysis technology faces a series of specific and thorny problems. First, in a high-density signal environment, frequency domain overlap is common, which makes it difficult for traditional energy detection-based methods to effectively separate signals of the same frequency or close frequency. In particular, when the amplitude of the background signal exceeds the target signal, mainstream spectrum analysis equipment can often only observe signals with larger amplitudes, while the weak signals of real interest are "submerged". Secondly, the existence of non-stationary signals (such as frequency hopping and spread spectrum signals) makes it difficult for traditional fixed time window analysis methods to take into account both time and frequency resolution at the same time, resulting in insufficient signal feature extraction. Furthermore, in a multipath propagation environment, the delay spread and Doppler spread of the signal will cause the ambiguity of the spectrum analysis results and affect the accurate estimation of signal parameters. In addition, for signals using advanced modulation technologies (such as OFDM and MIMO), their complex time-frequency characteristics bring great challenges to signal separation and modulation recognition. Finally, in real-time monitoring scenarios, how to quickly process massive amounts of spectrum data with limited computing resources and achieve timely capture and analysis of transient signals is still an urgent problem to be solved. These technical difficulties seriously restrict the application of spectrum analysis technology in complex electromagnetic environments, and it is urgent to develop new signal processing algorithms and system architectures to overcome these limitations. Summary of the invention

[0004] The purpose of the invention is to provide a method for separating and identifying same-frequency signals based on real-time spectrum analysis to solve the above-mentioned problems existing in the prior art.

[0005] Technical solution: a method for separating and identifying same-frequency signals based on real-time spectrum analysis, comprising the following steps:

[0006] S1, obtaining a radio frequency signal in the space and preprocessing it, and converting the preprocessed radio frequency signal into an intermediate frequency analog signal;

[0007] S2. Based on the intermediate frequency analog signal, sampling and digital filtering are performed to obtain a digital intermediate frequency signal; the delay control module is used to perform adaptive time slicing reorganization on the digital intermediate frequency signal to obtain a reorganized signal sequence; and the reorganized signal sequence is converted into spectrum statistical information;

[0008] S3. Convert the spectrum statistics information into RF-amplitude spectrum statistics, and use a threshold segmentation module to perform dynamic multi-dimensional threshold segmentation on the RF-amplitude spectrum statistics to obtain a signal classification result.

[0009] Beneficial effect: The present invention proposes a method for real-time spectrum analysis and separation of same-frequency signals, which can effectively distinguish between continuous wave signals and pulse wave signals without basically losing the original information of the signals; and then distinguish between background signals and foreground signals through dynamic multi-dimensional threshold segmentation, so that foreground signals with lower amplitudes are not affected by background signals and can be directly observed, thereby improving the anti-interference ability of spectrum equipment in complex electromagnetic signal environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flow chart of the present invention.

[0011] Figure 2 The figure is a principle block diagram of a real-time spectrum analysis system according to an embodiment of the present invention.

[0012] Figure 3 This is a flow chart of step S1 of the present invention.

[0013] Figure 4 This is a flow chart of step S2 of the present invention.

[0014] Figure 5 This is a flow chart of step S3 of the present invention. DETAILED DESCRIPTION

[0015] like Figure 1 and Figure 2 As shown, the present application proposes a method for separating and identifying same-frequency signals based on real-time spectrum analysis, comprising the following steps:

[0016] S1, obtaining a radio frequency signal in the space and preprocessing it, and converting the preprocessed radio frequency signal into an intermediate frequency analog signal;

[0017] S2. Based on the intermediate frequency analog signal, sampling and digital filtering are performed to obtain a digital intermediate frequency signal; the delay control module is used to perform adaptive time slicing reorganization on the digital intermediate frequency signal to obtain a reorganized signal sequence; and the reorganized signal sequence is converted into spectrum statistical information;

[0018] S3. Convert the spectrum statistics information into RF-amplitude spectrum statistics, and use a threshold segmentation module to perform dynamic multi-dimensional threshold segmentation on the RF-amplitude spectrum statistics to obtain a signal classification result.

[0019] In this embodiment, the signal receiving module receives the radio frequency signal in the space and converts it into an intermediate frequency analog signal IF. The intermediate frequency analog signal IF is converted into spectrum statistical information through the spectrum generating module. Then, the spectrum analyzing module analyzes the real-time spectrum information generated by the spectrum generating module, displays the analysis result on the user terminal, and controls the operation of the remaining modules according to user instructions.

[0020] like Figure 3 As shown, according to one aspect of the present application, step S1 is further:

[0021] S11, receiving a radio frequency signal in space through an antenna receiving module, and limiting the amplitude of the radio frequency signal through an amplitude limiting protection module to obtain a radio frequency signal after amplitude limiting;

[0022] S12, using a radio frequency amplification module to amplify the limited radio frequency signal to obtain an amplified radio frequency signal; based on the amplified radio frequency signal, using a radio frequency filtering module to filter out radio frequency signals outside a pre-specified frequency band to obtain an interference-free radio frequency signal;

[0023] S13, obtaining a first local oscillator signal in a primary mixing module, up-mixing the first local oscillator signal with the interference-eliminating radio frequency signal to obtain a primary mixing signal; using a mixing filter module to filter the primary mixing signal to filter out interference from the first local oscillator signal and its harmonics, and outputting the filtered primary mixing signal;

[0024] S14, obtaining a second local oscillator signal in a secondary mixing module, down-mixing the second local oscillator signal with the filtered primary mixing signal to obtain an intermediate frequency analog signal; using an intermediate frequency filtering module to filter the intermediate frequency analog signal to filter out interference from the second local oscillator signal and its harmonics, and obtaining a filtered intermediate frequency analog signal.

[0025] In another embodiment of the present application, a signal receiving module is used to receive a radio frequency signal RF in space and convert it into an intermediate frequency analog signal IF, including: an antenna feed receiving module, a limiting protection module, a radio frequency amplification module, a radio frequency filtering module, a primary mixing module, a mixing filtering module, a secondary mixing module and an intermediate frequency filtering module;

[0026] The antenna receiving module is used to receive the radio frequency signal RF in the space, and the output is sequentially passed through the limiting protection module, the radio frequency amplification module, the radio frequency filtering module, the primary mixing module, the mixing filtering module, the secondary mixing module, and the intermediate frequency filtering module;

[0027] A limiting protection module is used to limit the amplitude of the received radio frequency signal RF to protect subsequent circuits;

[0028] The RF amplification module is used to amplify the RF signal after limiting to compensate for the power loss of subsequent processing;

[0029] RF filter module, used to filter out RF signals outside the specified frequency band to reduce interference;

[0030] A primary mixing module is used to generate a local oscillator signal LO1, and use the local oscillator LO1 to perform up-mixing with the radio frequency signal RF to obtain a primary mixing signal RF′=RF+LO1;

[0031] The mixing filter module is used to filter the primary mixing signal RF′ to remove the interference of the local oscillator signal and its harmonics;

[0032] The secondary mixing module is used to generate a local oscillator signal LO2, and down-mix the primary mixing signal RF′ to obtain an intermediate frequency signal IF=RF′-LO2;

[0033] The intermediate frequency filtering module is used to filter the intermediate frequency analog signal IF, filter out the interference of the local oscillator signal and its harmonics, and send it to the spectrum generation module.

[0034] This embodiment achieves high-sensitivity reception and effective protection of weak signals in complex electromagnetic environments through multi-stage signal reception and processing. The antenna feed receiving module improves the system's ability to capture spatial RF signals, especially the reception performance of low signal-to-noise ratio signals. The introduction of the limiting protection module effectively prevents strong signals from damaging subsequent circuits and improves the anti-interference ability and reliability of the system. The construction of the RF amplification module not only compensates for signal transmission losses, but also improves the dynamic range of the system through low-noise amplification technology, so that weak signals can still maintain a good signal-to-noise ratio in subsequent processing. The out-of-band suppression characteristics of the RF filter module effectively reduce adjacent frequency interference and improve the purity of the signal. The architecture construction of primary and secondary mixing, combined with precise local oscillator signal control, realizes efficient down-conversion of broadband signals while minimizing the impact of mirror frequencies. The cascaded application of mixing filtering and intermediate frequency filtering modules not only removes the spurious signals generated during the mixing process, but also further optimizes the spectral characteristics of the signal.

[0035] This embodiment can effectively capture and retain the key features of the target signal even in a complex electromagnetic environment, laying a solid foundation for subsequent spectrum analysis. In particular, for the separation and identification of co-frequency signals, high-quality signal reception and preprocessing directly determine the accuracy and reliability of subsequent analysis. Through sophisticated frequency planning and filtering construction, the system can effectively suppress the influence of co-frequency interference signals while maintaining high sensitivity, thereby improving the distinguishability of the target signal. This high-quality signal preprocessing capability provides a clear and reliable data source for subsequent spectrum generation and analysis, and improves the success rate of separation and identification of co-frequency signals, especially when the signal strengths are similar or there is strong interference.

[0036] like Figure 4 As shown, according to one aspect of the present application, step S2 is further:

[0037] S21, using a high-speed ADC sampling module to sample and digitally filter the intermediate frequency analog signal to obtain a digital intermediate frequency signal;

[0038] S22, using a delay control module to perform adaptive time slicing reorganization processing on the digital intermediate frequency signal to obtain a reorganized signal sequence;

[0039] S23, based on the reorganized signal sequence, using an adaptive polynomial window function generated by a time domain window function modulation module to obtain a windowed signal sequence;

[0040] S24, using a fast Fourier transform module to perform a fractional-order short-time Fourier transform on the windowed signal sequence to obtain a time-frequency distribution matrix;

[0041] S25, using a spectrum superposition module to perform multi-scale spectrum fusion on the time-frequency distribution matrix to obtain a high-resolution spectrum graph;

[0042] S26. Use the pipeline spectrum statistics and forwarding module to perform binarization processing and superposition statistics on the high-resolution spectrum map to obtain spectrum statistical information.

[0043] In one embodiment of the present application, the spectrum generation module includes: a high-speed ADC sampling module, a delay control module, a time domain window function modulation module, a fast Fourier transform (FFT) module, a spectrum superposition module, and a pipeline spectrum statistics and forwarding module;

[0044] The high-speed ADC sampling module is used to sample and digitally filter the intermediate frequency analog signal IF output by the signal receiving module to obtain a digital intermediate frequency signal IF′, and the output is sequentially passed through the delay control module, the time domain window function modulation module, the fast Fourier transform (FFT) module, the spectrum superposition module, and the pipeline spectrum statistics and forwarding module;

[0045] The delay control module is used to copy the digital intermediate frequency signal IF′ output by the high-speed ADC sampling module into n channels, and perform delay processing on the n channels of signals respectively. The delay lengths are 0, ΔT, 2ΔT, 3ΔT, …, (n-1)ΔT in sequence. The n channels of signals obtained after the delay are recorded as IF1′, IF2′, IF3′, …, IFn′, and the n channels of signals are output in parallel to the time domain window function modulation module;

[0046] A time domain window function modulation module is used to generate a Hanning window function every nΔT in the time domain to form a window function sequence, and use the window function sequence to modulate the signals IF1′, IF2′, IF3′, … IFn′ in sequence and then output them to the fast Fourier transform (FFT) module;

[0047] A fast Fourier transform (FFT) module is used to perform fast Fourier transform (FFT) on the n intermediate frequency signals modulated by the window function sequence according to the set time slice length, to form n intermediate frequency FFT real-time stream spectrums IF_FFT1, IF_FFT2, IF_FFT3, ... IF_FFTn, and output them to the spectrum superposition module;

[0048] The spectrum superposition module is used to perform superposition statistics on the n-channel intermediate frequency FFT real-time streaming spectrum IF_FFT1, IF_FFT2, IF_FFT3, ... IF_FFTn within a single time slice length, and output the spectrum statistics results within each time slice length to form a real-time streaming spectrum;

[0049] The pipeline spectrum statistics and forwarding module is used to binarize the generated FFT real-time pipeline spectrum and perform superposition statistics on the binary pipeline spectrum within the length of N time slices. The spectrum statistics information IF_Sum(x,y) is sent to the spectrum analysis module in the form of a two-dimensional array.

[0050] This embodiment improves the accuracy and resolution of spectrum generation, and provides key support for the separation and identification of co-frequency signals. The introduction of the high-speed ADC sampling module realizes the high-precision digitization of the intermediate frequency signal, and provides a high-quality digital signal foundation for subsequent processing. The adaptive time-slicing reorganization processing technology effectively captures the transient characteristics and long-term change trends of the signal by dynamically adjusting the time resolution of signal analysis. It is particularly suitable for processing co-frequency signals with complex time-varying characteristics, and can separate the originally aliased signal components in the time domain. The adaptive polynomial window function generation technology reduces spectrum leakage and sidelobe interference by optimizing the window function shape in real time, and improves the accuracy of spectrum estimation. This is particularly important for separating co-frequency signals with similar frequencies, because it enhances the system's ability to distinguish small frequency differences. Through the fractional short-time Fourier transform (FSTFT) technology, the limitations of the traditional integer-order Fourier transform are broken, and a more flexible time-frequency analysis capability is provided. It is particularly suitable for analyzing signals with nonlinear frequency modulation characteristics, and can more accurately capture and characterize subtle differences in co-frequency signals. Multi-scale spectrum fusion technology generates a high dynamic range and high resolution spectrum diagram by intelligently combining the analysis results of different time and frequency resolutions. While maintaining high frequency resolution, it can also effectively show the time-varying characteristics of the signal. It is particularly effective for separating the same-frequency signals with similar frequencies but non-co-varying patterns. The construction of the stream spectrum statistics and forwarding module realizes the real-time accumulation and transmission of spectrum information, providing a continuous and stable data stream for subsequent spectrum analysis.

[0051] This embodiment improves the system's ability to separate co-frequency signals. In particular, the system shows significant advantages when processing co-frequency signals with complex modulation, fast frequency hopping or low signal-to-noise ratio. Through sophisticated time-frequency analysis and multi-dimensional signal feature extraction, even highly overlapping signals in the spectrum can be effectively distinguished. This high-precision spectrum generation capability not only improves the success rate of co-frequency signal separation, but also provides a more reliable data basis for subsequent signal identification and parameter estimation, and enhances the system's signal analysis capabilities in complex electromagnetic environments.

[0052] like Figure 5 As shown, according to one aspect of the present application, step S3 is further:

[0053] S31, using a frequency correction module to convert spectrum statistics information into radio frequency spectrum statistics; using a power correction module to convert radio frequency spectrum statistics into amplitude value information to obtain radio frequency-amplitude spectrum statistics;

[0054] S32, using a threshold segmentation module to perform dynamic multi-dimensional threshold segmentation on the radio frequency-amplitude spectrum statistics to obtain a classified signal spectrum, i.e., a signal classification result;

[0055] S33, using a spectrum peak detection module to perform peak detection on the classified signal spectrum, and parse out the frequency and amplitude value information corresponding to the signal peak point;

[0056] S34, using an image display module to perform RGB color display on the radio frequency-amplitude spectrum statistics, signal classification results, and frequency and amplitude value information corresponding to the signal peak point.

[0057] According to one aspect of the present application, it also includes:

[0058] S35, using the work control module to control the work of each module according to the monitoring task parameters set by the user, and determine whether there is a new monitoring task after the monitoring is completed.

[0059] In one embodiment of the present application, the spectrum analysis module includes: a frequency correction module, a power correction module, a threshold segmentation module, a spectrum peak search module, an image display module and a work control module;

[0060] The frequency correction module is used to convert the received spectrum statistical information IF_Sum(x,y) into radio frequency spectrum statistics RF_Sum(x,y) according to the microwave frequency conversion formula RF=IF+LO, and its output passes through the power correction module, threshold segmentation module, spectrum peak search module and image display module in sequence.

[0061] The power correction module is used to obtain the corresponding relationship between the amplitude and the spectrum code value of the received RF spectrum statistics result RF_Sum(x,y) through power calibration, and convert the code value information into power amplitude value information according to this corresponding relationship to obtain the RF-amplitude spectrum statistics RF_dB_Sum(x,y). The RF-amplitude spectrum statistics RF_dB_Sum(x,y) are sent to the threshold segmentation module and the spectrum peak search module respectively;

[0062] The threshold segmentation module is used to perform adaptive threshold image segmentation on the RF-amplitude spectrum after the RF conversion and amplitude conversion (RF_dB_Sum(x,y)), use the Otsu binarization method (OTSU) algorithm to separate the background signal spectrum and the foreground signal spectrum, and send the segmented spectrum to the spectrum peak search module;

[0063] The spectrum peak search module is used to detect the signal peak points of the RF-amplitude spectrum statistics RF_dB_Sum(x,y) and the background signal spectrum and foreground signal spectrum obtained after segmentation, and analyze the frequency and amplitude value information corresponding to the signal peak point, and send the analysis information to the image display module;

[0064] The image display module is used to display the RF-amplitude spectrum statistics RF_dB_Sum(x,y) and its corresponding background signal spectrum and foreground signal spectrum in RGB color on the user terminal, and display the peak frequency and peak amplitude value information corresponding to each spectrum diagram.

[0065] The work control module is used to allow the user to set monitoring task parameters such as monitoring frequency band and monitoring duration, control the aforementioned modules to work according to the monitoring task parameters set by the user, and determine whether a new monitoring task is received after a single monitoring duration ends. If so, control each module to start a new work process according to the new monitoring parameters, otherwise the process ends.

[0066] This embodiment uses spectrum analysis technology to achieve high-precision separation and identification of co-frequency signals, and improves the signal processing capability of the system in a complex electromagnetic environment. The introduction of the frequency correction module ensures the accuracy of spectrum analysis, and improves the accuracy of frequency measurement by compensating for frequency offset and nonlinearity in the system. This is crucial for distinguishing co-frequency signals with extremely close frequencies. The application of the power correction module not only improves the accuracy of amplitude measurement, but also achieves power balance between different frequency bands, so that the system can accurately evaluate the relative strength of each signal component. This accurate power estimation capability is particularly valuable for separating co-frequency signals with similar intensities. Dynamic multi-dimensional threshold segmentation technology improves the system's adaptability to complex signal environments. By comprehensively considering multiple characteristic dimensions of the signal (such as amplitude, frequency, bandwidth, modulation characteristics, etc.), co-frequency signals can be more accurately distinguished and classified. Especially when processing highly similar signal features, this multi-dimensional analysis method shows significant advantages. Not only can it separate signals with similar amplitudes, but it can also distinguish signals with similar spectrum characteristics but different modulation modes. The optimization of the spectrum peak search module improves the system's detection capability for weak signals and closely spaced signals. By adopting adaptive threshold and intelligent peak recognition algorithm, the system can accurately locate and characterize each signal component in complex background noise, and even weak signals partially masked by strong signals can be effectively detected. The enhanced function of the image display module provides intuitive and rich visualization for signal analysis. Through multi-dimensional color coding and dynamic display technology, complex signal characteristics and time-varying behaviors can be clearly presented, improving the operator's perception and understanding of the signal environment. The intelligent construction of the working control module realizes the adaptive adjustment of system parameters, so that the entire signal processing process can be automatically optimized according to the real-time monitored signal environment. This adaptive capability improves the system's response speed and processing efficiency when facing a rapidly changing electromagnetic environment.

[0067] This embodiment improves the system's ability to separate and identify co-frequency signals in complex electromagnetic environments. It can effectively handle challenges that are difficult to address with traditional methods, such as signal detection under extremely low signal-to-noise ratio conditions, separation of highly overlapping signals, and tracking of fast time-varying signals. The improvement of these capabilities not only enhances the application effect of the system in the fields of electronic reconnaissance, spectrum monitoring, and communication interference, but also provides strong technical support for spectrum resource management and signal intelligent identification in high-density signal environments.

[0068] According to one aspect of the present application, step S22 is further:

[0069] S221, obtaining a digital intermediate frequency signal output by a high-speed ADC sampling module, performing a wavelet transform on the digital intermediate frequency signal, and obtaining a wavelet coefficient matrix at different scales;

[0070] S222, based on the wavelet coefficient matrix, calculating the local energy distribution at each time point and scale; based on the local energy distribution, using a sliding window to perform smoothing processing to obtain a two-dimensional energy distribution diagram reflecting the local energy change of the signal;

[0071] S223, based on the two-dimensional energy distribution diagram, using a dynamic programming algorithm to determine the optimal time slice point to form a predetermined number of time slices; performing a short-time Fourier transform on each time slice, calculating its characteristic parameters, and forming a time slice feature set; wherein the characteristic parameters include instantaneous frequency, frequency bandwidth, and phase continuity;

[0072] S224. Based on the time-slice feature set, a feature graph is constructed, and a graph-theoretic clustering algorithm is used to calculate a clustering result; based on the clustering result, the time slices of the digital intermediate frequency signal are reorganized to obtain a reorganized signal sequence.

[0073] In one embodiment of the present application, an adaptive time slice reorganization module is constructed. The module dynamically divides the input signal into time slices of different lengths, and then recombines these time slices according to signal characteristics. Specifically, the following steps are included:

[0074] a) Perform multi-scale analysis on the input signal using wavelet transform;

[0075] The digital intermediate frequency signal is input into a preset wavelet transformer, and an appropriate wavelet basis function (such as Daubechies wavelet or Morlet wavelet) is selected to perform multi-scale decomposition to obtain multi-scale wavelet coefficients, that is, wavelet coefficient matrices at different scales.

[0076] b) Dynamically determine the length of the time slice based on the local energy distribution of the wavelet coefficients;

[0077] Analyze the wavelet coefficient matrix and calculate the local energy distribution at each time point and scale: square each element in the wavelet coefficient matrix, and then use a sliding window to smooth it on the time axis to obtain a two-dimensional energy distribution diagram that reflects the local energy changes of the signal. Based on the energy distribution diagram, use a dynamic programming algorithm to determine the optimal time slice point and obtain a series of time slices. Set the energy change threshold, and calculate the cumulative energy change point by point from the starting point of the time axis of the energy distribution diagram. When the cumulative energy change exceeds the preset threshold, mark the point as a slice point. Repeat this process until the entire time axis is traversed, and finally obtain a series of time slices of unequal lengths.

[0078] c) Extract features from each time slice, such as instantaneous frequency, phase continuity, etc.;

[0079] Perform short-time Fourier transform on each time slice and calculate its characteristic parameters such as instantaneous frequency, frequency bandwidth, phase continuity, etc. These parameters are combined into a feature vector corresponding to each time slice to form a time slice feature set.

[0080] d) Construct a time slice reorganization algorithm based on graph theory to cluster time slices with similar characteristics;

[0081] Each time slice is regarded as a node in the graph, and the edge weights between nodes are calculated based on the similarity of feature vectors. Using graph theory clustering methods such as spectral clustering or Louvain algorithm, community discovery is performed on the constructed graph to obtain the clustering results of the time slices. According to the labels of the clustering results, the time slices belonging to the same category are sequentially connected. For the transition between different categories, the overlap-addition method is used for smoothing, and finally the recombined signal sequence is obtained.

[0082] e) The recombinant signal sequence serves as input for subsequent processing.

[0083] This embodiment can analyze the characteristics of the signal at different scales through wavelet transform. This multi-scale analysis method can effectively capture the changes of the signal in different frequency ranges and provide a more comprehensive description of the signal characteristics. This is particularly important for the processing of complex signals because it can reveal details that cannot be captured by single-scale analysis. Based on the wavelet coefficient matrix, the local energy distribution is calculated, and the sliding window is smoothed to generate a two-dimensional energy distribution diagram, which can effectively reflect the local energy changes of the signal in time and frequency, and help to identify the important characteristics and change trends in the signal. The optimal time slicing point is determined using a dynamic programming algorithm to ensure the rationality and accuracy of the time slicing. The characteristic parameters of each time slice, such as instantaneous frequency, frequency bandwidth and phase continuity, are calculated by short-time Fourier transform to form a time slice feature set. These characteristic parameters can fully describe the characteristics of the signal in each time slice, providing a solid foundation for subsequent feature map construction and clustering. A feature map is constructed based on the time slice feature set, and the clustering result is calculated using a graph theory clustering algorithm. By reorganizing the time slices of the digital intermediate frequency signal, a reorganized signal sequence can be generated. Not only can similar time slices be effectively clustered, but the overall quality and consistency of the signal can also be improved through reorganization. The recombinant signal sequence finally generated has the characteristics of high resolution and high dynamic range. This embodiment fully utilizes the multi-scale analysis capability of wavelet transform and the optimization capability of dynamic programming during the processing process to ensure the consistency and integrity of the signal at different scales and time slices.

[0084] Through multi-scale analysis, local energy distribution calculation, dynamic programming optimization of time slicing, graph theory clustering and signal recombination, the digital intermediate frequency signal output by the high-speed ADC sampling module is efficiently processed and reorganized, generating a high-resolution and high dynamic range signal sequence.

[0085] According to one aspect of the present application, step S23 is further:

[0086] S231, obtaining the reorganized signal sequence output by the delay control module, performing short-time Fourier transform on the signal sequence to obtain a time-frequency diagram matrix; based on the time-frequency diagram matrix, calculating the energy distribution and frequency variation characteristics to form a signal feature descriptor;

[0087] S232, based on the signal feature descriptor, construct an optimization objective function to obtain an optimization problem model, including minimizing spectrum leakage, maximizing the sidelobe suppression ratio and minimizing the main lobe width; use an evolutionary algorithm to solve the optimization problem model to obtain optimal polynomial coefficients;

[0088] S233, constructing an adaptive window function based on the optimal polynomial coefficients to obtain a window function sequence; performing a point-by-point multiplication operation on the window function sequence and the reorganized signal sequence to obtain a windowed signal sequence.

[0089] In one embodiment of the present application, an adaptive polynomial window function generator is constructed. This module dynamically generates an optimal window function according to the characteristics of the signal. The specific steps are as follows:

[0090] a) Perform short-time Fourier transform (STFT) on the input signal to obtain a time-frequency diagram;

[0091] The signal sequence is divided into several overlapping short time segments, discrete Fourier transform is applied to each segment, and the results are combined to form a two-dimensional time-frequency map matrix.

[0092] b) Analyze the energy distribution and frequency variation characteristics in the time-frequency diagram;

[0093] The time-frequency graph matrix is ​​normalized and the energy value of each time-frequency point is calculated. The instantaneous rate of change of frequency is calculated using the central moment method, and combined with the energy distribution, a feature descriptor vector is formed to describe the time-frequency characteristics of the signal.

[0094] c) constructing an optimization algorithm that generates a set of polynomial coefficients with the goal of minimizing spectral leakage and sidelobe suppression;

[0095] Based on the signal feature descriptor, an optimization objective function is constructed to obtain the optimization problem model. A multi-objective function is constructed, including minimizing spectrum leakage, maximizing sidelobe suppression ratio, minimizing mainlobe width and other objectives. These objectives are combined into an objective function in the form of a weighted sum, where the weights are dynamically adjusted according to the signal feature descriptor. An improved differential evolution algorithm is used to initialize a set of candidate polynomial coefficients. Through the iterative optimization process, the candidate solutions are continuously updated and evaluated until convergence or the preset number of iterations is reached, and finally the optimal set of polynomial coefficients is obtained.

[0096] d) constructing a window function using this set of coefficients;

[0097] Using the Chebyshev polynomial expansion method, the optimal coefficients are substituted into the polynomial expression to generate a continuous window function curve. The curve is discretized and sampled to obtain a window function sequence that matches the length of the input signal.

[0098] e) Apply this adaptively generated window function to the original recombined signal sequence to obtain the windowed signal sequence. Perform point-by-point multiplication of the original signal sequence and the window function sequence to generate the windowed signal sequence. Smooth the beginning and end of the sequence to eliminate possible discontinuities.

[0099] This embodiment can convert the reorganized signal sequence into a time-frequency diagram matrix through short-time Fourier transform (STFT). This time-frequency analysis method can effectively reveal the changing characteristics of the signal in time and frequency, and provide detailed energy distribution and frequency change information. Based on the time-frequency diagram matrix, the energy distribution and frequency change characteristics are calculated to form signal feature descriptors. These feature descriptors can comprehensively describe the time-frequency characteristics of the signal, providing a solid foundation for subsequent optimization. Based on the signal feature descriptor, an optimization objective function is constructed, covering multiple objectives such as minimizing spectrum leakage, maximizing sidelobe suppression ratio, and minimizing main lobe width. These objective functions can effectively measure the quality and performance of the signal and ensure the directionality and effectiveness of the optimization process. By constructing an optimization problem model, the objectives and constraints of signal optimization can be systematically described, providing clear guidance for the solution of the evolutionary algorithm. Using the evolutionary algorithm to solve the optimization problem model can effectively search for the optimal solution. The evolutionary algorithm has a global search capability, can avoid falling into the local optimum, and ensure that the global optimal polynomial coefficients are obtained. Through the iterative optimization of the evolutionary algorithm, the optimal solution is gradually approached to ensure the accuracy and reliability of the optimization results. Based on the optimal polynomial coefficients, an adaptive window function is constructed to generate a window function sequence. The adaptive window function can be dynamically adjusted according to the characteristics of the signal to improve the flexibility and adaptability of signal processing. By performing point-by-point multiplication operations on the window function sequence and the reorganized signal sequence, the windowed signal sequence is obtained, which can effectively suppress spectrum leakage and improve the spectrum quality of the signal. The resulting windowed signal sequence has the characteristics of high resolution and high dynamic range.

[0100] This embodiment makes full use of the advantages of time-frequency analysis, optimization algorithm and adaptive window function to ensure the consistency and integrity of the signal at different scales and time slices. By optimizing the multi-objective optimization of the objective function, it is possible to achieve a balance in multiple aspects such as spectrum leakage, sidelobe suppression ratio and main lobe width, and generate a high-quality signal sequence. This embodiment achieves efficient optimization processing of the recombined signal sequence and generates a signal sequence with high resolution and high dynamic range.

[0101] According to one aspect of the present application, step S24 is further:

[0102] S241, based on the windowed signal sequence, using the least squares method, calculate the instantaneous frequency; based on the instantaneous frequency, determine the optimal fractional order parameter set by a grid search method;

[0103] S242, constructing a fractional-order Fourier transform kernel function based on the fractional-order parameter set, discretizing the fractional-order Fourier transform kernel function, and generating a transformation matrix;

[0104] S243, performing fractional Fourier transform based on the transformation matrix and the windowed signal sequence to obtain an initial time-frequency distribution; performing adaptive time-frequency focusing on the initial time-frequency distribution to obtain an enhanced time-frequency distribution;

[0105] S244: Perform interference suppression processing on the enhanced time-frequency distribution to obtain a final time-frequency distribution matrix.

[0106] In one embodiment of the present application, a fractional short-time Fourier transform (FSTFT) module is constructed. This module combines the advantages of fractional Fourier transform and short-time analysis. The specific steps are as follows:

[0107] a) Construct an adaptive algorithm to estimate the optimal fractional-order parameters of the signal;

[0108] The windowed signal sequence output by the time domain window function modulation module is obtained, and the optimal fractional order parameters are estimated to obtain the fractional order parameter set. The phase change curve of the signal is fitted using the least squares method, and its derivative is calculated as the instantaneous frequency. Based on the changing characteristics of the instantaneous frequency, the optimal fractional order parameter set is determined by the grid search method.

[0109] b) Implement a fast algorithm for fractional Fourier transform;

[0110] Based on the fractional parameter set, the kernel function expression of the fractional Fourier transform is constructed using the Riemann-Liouville fractional integral definition. The kernel function is discretized to generate a transformation matrix for subsequent fast calculations.

[0111] c) Combining the fractional transform with the short-time analysis window to form FSTFT;

[0112] Using matrix multiplication, the transformation matrix is ​​multiplied by the windowed signal sequence to obtain the fractional Fourier transform result. The result is rearranged into a two-dimensional time-frequency distribution matrix.

[0113] d) Implement the parallel computing architecture of FSTFT to improve the processing speed.

[0114] The optimal time-frequency kernel function is iteratively calculated using the adaptive weighted S method. The kernel function is convolved with the initial time-frequency distribution to obtain the focused time-frequency distribution. The morphological filtering method is used to construct an adaptive structural element, and the time-frequency distribution is opened and closed to suppress noise and interference components. Nonlinear threshold processing is applied to further improve the clarity of the time-frequency distribution.

[0115] This embodiment calculates the instantaneous frequency by the least squares method, and can accurately capture the frequency variation characteristics of the signal. The instantaneous frequency is an important parameter in signal analysis, which can reflect the frequency variation of the signal in the time domain. The grid search method is used to determine the optimal fractional parameter set to ensure the accuracy and effectiveness of the fractional Fourier transform. The grid search method can perform a comprehensive search in the parameter space to find the optimal fractional parameter set, thereby improving the accuracy of the transformation. Based on the optimal fractional parameter set, the fractional Fourier transform kernel function is constructed, and it is discretized to generate a transformation matrix. The fractional Fourier transform is a generalized Fourier transform that can more flexibly process non-stationary signals and provide richer time-frequency information. Through discretization, the generated transformation matrix can be efficiently applied to signal processing to ensure the rapidity and accuracy of the calculation. Based on the transformation matrix and the windowed signal sequence, the fractional Fourier transform is performed to obtain the initial time-frequency distribution. The initial time-frequency distribution provides a preliminary characteristic description of the signal in time and frequency. Adaptive time-frequency focusing of the initial time-frequency distribution can enhance the clarity and resolution of the time-frequency distribution. The adaptive time-frequency focusing method can dynamically adjust the focusing parameters according to the signal characteristics, improving the accuracy and interpretability of the time-frequency distribution. Interference suppression processing of the enhanced time-frequency distribution can effectively remove noise and interference signals and improve the quality of the signal. Interference suppression processing ensures the purity and reliability of the final time-frequency distribution. The final time-frequency distribution matrix has the characteristics of high resolution and high dynamic range, which can fully reflect the time-frequency characteristics of the signal and provide a solid foundation for subsequent signal analysis and processing.

[0116] This embodiment realizes efficient optimization processing and time-frequency analysis of the windowed signal sequence, generates a time-frequency distribution matrix with high resolution and high dynamic range, which not only improves the accuracy and efficiency of signal processing, but also enhances the flexibility and adaptability of signal analysis.

[0117] According to one aspect of the present application, step S25 is further:

[0118] S251, performing multi-scale decomposition based on the time-frequency distribution matrix to generate sub-band image sets with different frequencies and time resolutions;

[0119] S252, extracting local features based on each sub-band image in the sub-band image set to form a feature descriptor set; wherein the local features include texture features, edge information, and inflection point information;

[0120] S253, constructing a spectrum feature tree based on the feature descriptor set; performing feature matching and consistency check on the spectrum feature tree to obtain feature correspondence;

[0121] S254. Based on the feature correspondence, an adaptive weight fusion strategy is constructed to generate a fusion weight matrix; based on the fusion weight matrix and the sub-band image set, weighted fusion is performed to obtain a high-resolution spectrum map.

[0122] In one embodiment of the present application, a multi-scale spectrum fusion module is constructed. The module combines spectrum information of different time scales and frequency resolutions. The specific steps are as follows:

[0123] a) Perform FSTFT analysis on the input signal with different parameter settings to obtain a series of time-frequency diagrams;

[0124] The time-frequency distribution matrix output by the FSTFT module is obtained and multi-resolution decomposition is performed to obtain a series of sub-band images. The time-frequency distribution matrix is ​​decomposed at multiple scales using wavelet packet transform to generate a set of sub-band images with different frequencies and time resolutions.

[0125] b) Construct a multi-resolution analysis algorithm based on image processing to extract the features of each time-frequency graph;

[0126] Based on each sub-band image obtained, the local binary pattern (LBP) operator is used to calculate the texture features, and the Hessian matrix is ​​used to extract the edge and inflection point information, which are combined to form the feature descriptor of each sub-band.

[0127] For each subband image I i , calculate its local binary pattern (LBP) feature. The LBP operator generates a binary pattern by comparing the grayscale values ​​of each pixel with its neighboring pixels: for each pixel p, select a radius R and the number of neighborhood points P. Calculate the difference between the grayscale value of the neighborhood point and the grayscale value of the central pixel p to generate a binary pattern. Convert the binary pattern to a decimal value as the LBP value of the pixel. The generated LBP feature map is denoted as LBP i The Hessian matrix is ​​calculated as follows: For each subband image I i , calculate its Hessian matrix to extract edge and inflection point information. The Hessian matrix consists of the second-order partial derivatives of the image: Calculate the image I i The second-order partial derivative I xx ,I yy and I xy . Form the Hessian matrix H: H=(I xx , I xy , I xy , I yy ). Through the eigenvalue analysis of the Hessian matrix, edge and inflection point information is extracted. Feature descriptor combination: LBP feature map LBP i And Hessian matrix features are combined to form the feature descriptor of each subband, specifically LBPi Concatenate the eigenvalues ​​of the Hessian matrix to form the eigenvector F i The feature vectors of all sub-band images are normalized to obtain the final feature descriptor F.

[0128] c) Use the data structure of the spectral feature tree to organize multi-scale features;

[0129] Construct a hierarchical tree data structure to organize feature descriptors of different scales into parent-child relationships. Use a bottom-up approach to construct a feature tree to obtain a multi-scale feature structure, where low-level nodes represent detailed features and high-level nodes represent generalized features.

[0130] d) Construct a spectrum fusion algorithm based on this feature tree to intelligently merge spectrum information of different scales;

[0131] Construct a multi-scale feature matching algorithm, starting from the top layer of the feature tree and performing feature matching layer by layer. Use a graph matching algorithm to ensure the consistency of cross-scale features and obtain the feature correspondence between different sub-bands. Use a fuzzy inference system to calculate the fusion weight of each sub-band based on the saliency, consistency, and reliability of the features. Generate a weight matrix corresponding to the sub-band image set. Based on the fusion weight matrix and the original sub-band image set, use a wavelet reconstruction algorithm to inverse transform and superimpose the weighted sub-band images to generate a fused high-resolution, high-dynamic range spectrum map.

[0132] Based on the features calculated for each sub-band image (such as LBP features and Hessian matrix features), a feature tree is constructed. Each node of the feature tree represents the feature of a time slice or sub-band image, and multi-scale feature matching is performed: starting from the top layer of the feature tree, feature matching is performed layer by layer. A graph matching algorithm (such as the Hungarian algorithm or the maximum weight matching algorithm) is used to ensure the consistency of cross-scale features and obtain the feature correspondence between different sub-bands. The fuzzy inference system calculates the fusion weight: a fuzzy inference system is constructed to calculate the fusion weight of each sub-band based on the significance, consistency and reliability of the features. The input of the fuzzy inference system is the feature significance, consistency and reliability, and the output is the fusion weight. Generate a weight matrix: Based on the output of the fuzzy inference system, a weight matrix W corresponding to the sub-band image set is generated. i . Weighted subband image: For each subband image I i Perform weighted processing to obtain the weighted sub-band image I i ′:I i ′=W i ⋅I i . Inverse wavelet transform and superposition: weighted subband image I i ′ Perform inverse wavelet transform to restore the original image space. i′′ is superimposed to obtain the fused image I f .

[0133] e) Output the fused high-resolution, high-dynamic-range spectrum.

[0134] This embodiment performs multi-scale decomposition based on the time-frequency distribution matrix to generate a sub-band image set with different frequencies and time resolutions. This multi-scale decomposition method can effectively capture the characteristics of the signal at different frequencies and time scales, and provide a more comprehensive signal description. This is particularly important for the processing of complex signals because it can reveal details that cannot be captured by a single scale analysis. Local features are extracted from the sub-band image set, including texture features, edge information, and inflection point information. These local features can fully describe the details and structure of the sub-band image, providing a solid foundation for subsequent feature matching and fusion. Texture features can reflect the surface characteristics of the image, edge information can reveal the contour and shape of the image, and inflection point information can capture the key points and change points in the image. A spectral feature tree is constructed based on the feature descriptor set, and feature matching and consistency checks are performed to obtain feature correspondence. The spectral feature tree can systematically organize and represent the features of the sub-band image, and feature matching and consistency checks ensure the accurate correspondence of features between different sub-band images, can effectively cluster similar sub-band images, and ensure the accuracy and consistency of feature fusion. Based on the feature correspondence, an adaptive weight fusion strategy is constructed to generate a fusion weight matrix. The adaptive weight fusion strategy can dynamically adjust the weights according to the significance and consistency of the features, thereby improving the quality of the fusion results. Through weighted fusion, the information of different sub-band images can be effectively integrated to generate a high-resolution spectrum map. The high-resolution spectrum map finally generated has the characteristics of high resolution and high dynamic range. This embodiment makes full use of the advantages of multi-scale decomposition, local feature extraction, feature matching and adaptive weight fusion to ensure the consistency and integrity of the signal at different scales and time slices. The high-resolution spectrum map can provide clearer and more detailed signal features, providing a solid foundation for subsequent signal analysis and processing.

[0135] This embodiment realizes efficient processing and fusion of the time-frequency distribution matrix, generates a high-resolution, high-dynamic range spectrum diagram, which not only improves the accuracy and efficiency of signal processing, but also enhances the flexibility and adaptability of signal analysis, and provides an effective solution for the processing of complex signals.

[0136] According to one aspect of the present application, step S32 is further:

[0137] S321, obtaining the RF-amplitude spectrum statistics output by the power correction module, applying wavelet packet transform to perform multi-scale decomposition, and obtaining a multi-resolution representation of the spectrum; based on the multi-resolution representation, calculating the local statistical features of each sub-band to form a feature vector set; wherein the local statistical features include energy entropy, spectrum centroid, spectrum bandwidth, kurtosis and skewness;

[0138] S322, inputting the feature vector set into a pre-trained variational autoencoder network to perform nonlinear dimensionality reduction to obtain a low-dimensional latent space representation; based on the low-dimensional latent space representation, applying a dynamic time warping algorithm to calculate the similarity matrix between the current frame and the historical frame to obtain a temporal consistency score;

[0139] S323. Based on the temporal consistency score and the low-dimensional latent space representation, an adaptive kernel density estimation model is constructed to obtain a multidimensional probability density distribution; based on the multidimensional probability density distribution, an expectation maximization algorithm is applied for clustering to obtain a preliminary signal classification result;

[0140] S324. Based on the preliminary signal classification result, a graph cutting algorithm is applied to perform refinement to obtain an optimized signal classification result.

[0141] In one embodiment of the present application, an appropriate wavelet basis function (such as Daubechies wavelet) is selected to perform a 6-layer wavelet packet decomposition on the radio frequency-amplitude spectrum statistics to generate a multi-scale spectrum representation containing 64 sub-bands. For each sub-band, five statistics such as energy entropy, spectrum centroid, spectrum bandwidth, kurtosis and skewness are calculated. The statistics of the 64 sub-bands are combined into a 320-dimensional feature vector to form a feature vector set. A variational autoencoder comprising 3 encoding layers and 3 decoding layers is used, wherein the number of neurons of the encoder is 256, 128, and 64, respectively, and the decoder is 64, 128, and 256 respectively. The 320-dimensional feature vector is mapped to a 64-dimensional latent space by minimizing the reconstruction error and KL divergence. The sliding window size is set to 10 frames, and the current frame is dynamically time-warped matched with the historical frames in the window. The cumulative distance of the matching path is calculated and normalized to obtain a temporal consistency score in the range of [0,1].

[0142] For each dimension, the bandwidth of the kernel function is dynamically adjusted according to the temporal consistency score. The probability density distribution of data points is estimated in 64-dimensional space using an adaptive Gaussian kernel function. The maximum number of clusters is set to 10, and the Bayesian Information Criterion (BIC) is used to automatically determine the optimal number of clusters. The expectation maximization (EM) algorithm is iteratively executed until convergence or the maximum number of iterations (100) is reached to obtain preliminary signal classification labels. A k-nearest neighbor graph (k=5) is constructed, and the edge weights are based on the weighted sum of the Euclidean distance in the low-dimensional latent space and the temporal consistency score. The graph is segmented using the normalized cut algorithm to minimize the inter-class similarity and maximize the intra-class similarity to obtain the refined classification results.

[0143] The optimized classification results are mapped back to the original spectrum domain and morphological post-processing is applied to obtain the final signal classification mask. The region growing algorithm is performed on the original spectrum domain using the class label as a seed. Opening and closing operations (structural element size is 3x3) are applied to remove noise and fill small holes to obtain a smooth signal classification mask.

[0144] Based on the signal classification mask, the characteristic parameters of each type of signal are extracted and a signal classification result report is generated. For each classification area, parameters such as center frequency, bandwidth, duration, and average power are calculated. The characteristic parameters of each signal are summarized to generate a structured classification result report containing information such as signal type, frequency range, time range, and power level.

[0145] This embodiment performs multi-scale decomposition through wavelet packet transform, and can analyze the spectrum at different frequencies and time resolutions to obtain a multi-resolution representation of the spectrum. It can effectively capture the changes of the signal in different frequency ranges and provide a more comprehensive description of signal features. Based on the multi-resolution representation, the local statistical features of each subband are calculated, including energy entropy, spectrum centroid, spectrum bandwidth, kurtosis and skewness. These features can fully describe the local characteristics of the signal and provide a solid foundation for subsequent dimensionality reduction and clustering. The feature vector set is input into the pre-trained variational autoencoder network for nonlinear dimensionality reduction to obtain a low-dimensional latent space representation. The variational autoencoder can effectively capture the potential structure of high-dimensional data and reduce the data dimension while retaining important features. Based on the low-dimensional latent space representation, the dynamic time warping algorithm is applied to calculate the similarity matrix between the current frame and the historical frame to obtain a temporal consistency score. The dynamic time warping algorithm can handle the nonlinear changes of time series data and ensure the accuracy of similarity calculation. Based on the temporal consistency score and the low-dimensional latent space representation, an adaptive kernel density estimation model is constructed to obtain a multi-dimensional probability density distribution. Adaptive kernel density estimation can dynamically adjust the kernel function according to the distribution of data to improve the accuracy of density estimation. Based on the multidimensional probability density distribution, the expectation maximization algorithm is applied for clustering to obtain preliminary signal classification results. The expectation maximization algorithm can gradually optimize the clustering results in the iterative process and improve the accuracy of classification. Based on the preliminary signal classification results, the graph cutting algorithm is applied for refinement to obtain the optimized signal classification results. The graph cutting algorithm can effectively handle complex classification boundaries and further improve the accuracy and consistency of the classification results. The signal classification results finally generated have the characteristics of high resolution and high dynamic range. This embodiment makes full use of the advantages of multi-scale decomposition, nonlinear dimensionality reduction, dynamic time warping, adaptive kernel density estimation and graph cutting algorithms to ensure the consistency and integrity of the signal at different scales and time slices. It can effectively classify and optimize RF-amplitude spectrum statistical signals, improve the accuracy and efficiency of signal processing, and provide an effective solution for the analysis and processing of complex signals. Efficient processing and classification of RF-amplitude spectrum statistical signals are achieved, and signal classification results with high resolution and high dynamic range are generated.

[0146] This embodiment can effectively handle the problem of separation and identification of co-frequency signals in complex electromagnetic environments. Through adaptive feature extraction and nonlinear dimensionality reduction, the ability to distinguish weak signals and spectral overlapping signals is improved. At the same time, the introduction of time consistency constraints and graph cut optimization enhances the robustness of the algorithm to non-stationary signals. This multi-dimensional, multi-stage processing flow provides a more comprehensive and reliable decision-making basis for signal classification.

[0147] In another embodiment of the present application, a high-resolution spectrum graph output by the spectrum superposition module is obtained, and multi-dimensional features are extracted to obtain a feature vector set. The instantaneous amplitude, phase, frequency, bandwidth and other parameters of each time-frequency point in the spectrum graph are calculated. The time-frequency structural features of the signal are estimated using the local covariance matrix, and combined to form a multi-dimensional feature vector set.

[0148] Apply principal component analysis (PCA) to reduce the dimensionality of the eigenvector set to obtain the eigenspace after dimensionality reduction. Calculate the covariance matrix of the eigenvector set and solve its eigenvalues ​​and eigenvectors. Select several principal components with the largest contribution rate, project the original eigenvectors onto these principal components, and obtain the eigenspace after dimensionality reduction.

[0149] In the reduced-dimensional feature space, the improved DBSCAN algorithm is used for density clustering to obtain the initial clustering results. A DBSCAN variant with adaptive density threshold is constructed to dynamically adjust clustering parameters according to the local density distribution. The data points in the feature space are clustered to obtain the initial signal grouping.

[0150] Based on the initial clustering results, a fuzzy decision tree is constructed to obtain a fuzzy rule set. The ID3 algorithm is used to construct the basic structure of the decision tree, and fuzzy sets and membership functions are introduced at each node. Based on prior knowledge and clustering statistics, the fuzzy sets of each feature dimension are defined to generate a series of fuzzy IF-THEN rules.

[0151] The obtained fuzzy rule set is applied to the initial clustering results, and rule reasoning is performed to obtain the fuzzy classification results. The Mamdani fuzzy inference system is used for each cluster to calculate its membership to each predefined signal category. The centroid method is used for defuzzification to obtain the final category label and confidence of each cluster.

[0152] Post-process the fuzzy classification results, apply morphological operations and connected domain analysis to obtain the final signal classification results. Use morphological opening and closing operations to eliminate isolated points and small areas. Perform connected domain labeling and merge adjacent time-frequency regions belonging to the same signal. Add labels to each detected signal region, including signal type, center frequency, bandwidth and other information. Output the signal classification results to the subsequent processing module for signal identification and parameter estimation.

[0153] This embodiment can fully capture the time-frequency characteristics of the signal by extracting parameters such as the instantaneous amplitude, phase, frequency, and bandwidth of each time-frequency point in the spectrum diagram. The local covariance matrix is ​​used to estimate the time-frequency structural characteristics of the signal, further enriching the feature vector set. Applying principal component analysis (PCA) for dimensionality reduction can effectively reduce the dimension of the feature space, retain the main information, and reduce the computational complexity. By calculating the covariance matrix and solving its eigenvalues ​​and eigenvectors, several principal components with the largest contribution rate are selected, and the original eigenvectors are projected onto these principal components to obtain the reduced-dimensional feature space. In the reduced-dimensional feature space, the improved DBSCAN algorithm is used for density clustering, which can effectively identify dense areas and noise points in the data. A DBSCAN variant with an adaptive density threshold is constructed, and the clustering parameters are dynamically adjusted according to the local density distribution, thereby improving the accuracy and robustness of the clustering results. By clustering the data points in the feature space, the signal can be preliminarily grouped to form an initial clustering result. The rationality and consistency of the signal grouping are ensured. Based on the initial clustering results, a fuzzy decision tree is constructed to generate a fuzzy rule set. The ID3 algorithm is used to construct the basic structure of the decision tree. Fuzzy sets and membership functions are introduced at each node to handle uncertainty and ambiguity. The fuzzy rule set is applied to the initial clustering results, and rule reasoning is performed to obtain the fuzzy classification results. The Mamdani fuzzy inference system is used for each cluster to calculate its membership to each predefined signal category. The centroid method is used for defuzzification to obtain the final category label and confidence of each cluster. The fuzzy classification results are post-processed, and morphological operations and connected domain analysis are applied to further optimize the signal classification results. Morphological opening and closing operations are used to eliminate isolated points and small areas to ensure the consistency and integrity of the classification results. Connected domain labeling is performed to merge adjacent time-frequency regions belonging to the same signal, and labels are added to each detected signal region, including information such as signal type, center frequency, and bandwidth. The accuracy and interpretability of the signal classification results are ensured. The final signal classification results have the characteristics of high resolution and high dynamic range, ensuring the consistency and integrity of the signal at different scales and time slices.

[0154] This embodiment realizes efficient processing and classification of high-resolution spectrum diagrams, and generates high-resolution, high-dynamic range signal classification results, which not only improves the accuracy and efficiency of signal processing, but also enhances the flexibility and adaptability of signal analysis, and provides an effective solution for the processing of complex signals.

[0155] In one embodiment of the present application, a broadband noise FM signal and a narrowband linear FM signal are used as inputs, and a simulation analysis is performed using the method of the present application. The results show that a conventional spectrum analyzer can analyze noise FM signals, but a narrowband linear FM signal with a lower power value cannot be found in the simulation results. When the method of the present application is used, a broadband noise FM signal with a higher power value and a narrowband linear FM signal with a lower power value can both be directly observed on the spectrum, or can be observed separately and independently.

[0156] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for separating and identifying same-frequency signals based on real-time spectrum analysis, characterized in that: The steps include: S1, obtaining a radio frequency signal in the space and preprocessing it, and converting the preprocessed radio frequency signal into an intermediate frequency analog signal; S2, based on the intermediate frequency analog signal, sampling and digital filtering are performed to obtain a digital intermediate frequency signal; Adopting a delay control module to perform adaptive time-slicing reorganization processing on the digital intermediate frequency signal to obtain a reorganized signal sequence; converting the reorganized signal sequence into spectrum statistical information; S3, converting the spectrum statistics information into radio frequency-amplitude spectrum statistics, and using a threshold segmentation module to perform dynamic multi-dimensional threshold segmentation on the radio frequency-amplitude spectrum statistics to obtain a signal classification result; Step S2 is further as follows: S21, using a high-speed ADC sampling module to sample and digitally filter the intermediate frequency analog signal to obtain a digital intermediate frequency signal; S22, using a delay control module to perform adaptive time slicing reorganization processing on the digital intermediate frequency signal to obtain a reorganized signal sequence; S23, based on the reorganized signal sequence, using an adaptive polynomial window function generated by a time domain window function modulation module to obtain a windowed signal sequence; S24, using a fast Fourier transform module to perform a fractional-order short-time Fourier transform on the windowed signal sequence to obtain a time-frequency distribution matrix; S25, using a spectrum superposition module to perform multi-scale spectrum fusion on the time-frequency distribution matrix to obtain a high-resolution spectrum graph; S26, using a pipeline spectrum statistics and forwarding module to perform binarization processing and superposition statistics on the high-resolution spectrum map to obtain spectrum statistical information; Step S3 is further as follows: S31, using a frequency correction module to convert spectrum statistics information into radio frequency spectrum statistics; using a power correction module to convert radio frequency spectrum statistics into amplitude value information to obtain radio frequency-amplitude spectrum statistics; S32, using a threshold segmentation module to perform dynamic multi-dimensional threshold segmentation on the radio frequency-amplitude spectrum statistics to obtain a classified signal spectrum, i.e., a signal classification result; S33, using a spectrum peak detection module to perform peak detection on the classified signal spectrum, and parse out the frequency and amplitude value information corresponding to the signal peak point; S34, using an image display module to perform RGB color display on the radio frequency-amplitude spectrum statistics, signal classification results, and frequency and amplitude value information corresponding to the signal peak point.

2. The same-frequency signal separation and identification method based on real-time spectrum analysis according to claim 1 is characterized in that: Step S1 is further as follows: S11, receiving a radio frequency signal in space through an antenna receiving module, and limiting the amplitude of the radio frequency signal through an amplitude limiting protection module to obtain a radio frequency signal after amplitude limiting; S12, using a radio frequency amplification module to amplify the limited radio frequency signal to obtain an amplified radio frequency signal; based on the amplified radio frequency signal, using a radio frequency filtering module to filter out radio frequency signals outside a pre-specified frequency band to obtain an interference-free radio frequency signal; S13, obtaining a first local oscillator signal in a primary mixing module, and up-mixing the first local oscillator signal with the interference-eliminated radio frequency signal to obtain a primary mixing signal; The primary mixing signal is filtered using a mixing filter module to remove interference from the first local oscillator signal and its harmonics, and the filtered primary mixing signal is output; S14, obtaining a second local oscillator signal in a secondary mixing module, down-mixing the second local oscillator signal with the filtered primary mixing signal to obtain an intermediate frequency analog signal; using an intermediate frequency filtering module to filter the intermediate frequency analog signal to filter out interference from the second local oscillator signal and its harmonics, and obtaining a filtered intermediate frequency analog signal.

3. The same-frequency signal separation and identification method based on real-time spectrum analysis according to claim 2 is characterized in that: Step S22 is further as follows: S221, obtaining a digital intermediate frequency signal output by a high-speed ADC sampling module, performing a wavelet transform on the digital intermediate frequency signal, and obtaining a wavelet coefficient matrix at different scales; S222, based on the wavelet coefficient matrix, calculating the local energy distribution at each time point and scale; based on the local energy distribution, using a sliding window to perform smoothing processing to obtain a two-dimensional energy distribution diagram reflecting the local energy change of the signal; S223, based on the two-dimensional energy distribution diagram, using a dynamic programming algorithm to determine the optimal time slice point to form a predetermined number of time slices; performing a short-time Fourier transform on each time slice, calculating its characteristic parameters, and forming a time slice feature set; wherein the characteristic parameters include instantaneous frequency, frequency bandwidth, and phase continuity; S224. Based on the time-slice feature set, a feature graph is constructed, and a graph-theoretic clustering algorithm is used to calculate a clustering result; based on the clustering result, the time slices of the digital intermediate frequency signal are reorganized to obtain a reorganized signal sequence.

4. The method for separating and identifying same-frequency signals based on real-time spectrum analysis according to claim 2, characterized in that: Step S23 is further as follows: S231, obtaining the reorganized signal sequence output by the delay control module, performing short-time Fourier transform on the signal sequence to obtain a time-frequency diagram matrix; based on the time-frequency diagram matrix, calculating the energy distribution and frequency variation characteristics to form a signal feature descriptor; S232, based on the signal feature descriptor, construct an optimization objective function to obtain an optimization problem model, including minimizing spectrum leakage, maximizing the sidelobe suppression ratio and minimizing the main lobe width; use an evolutionary algorithm to solve the optimization problem model to obtain optimal polynomial coefficients; S233, constructing an adaptive window function based on the optimal polynomial coefficients to obtain a window function sequence; performing a point-by-point multiplication operation on the window function sequence and the reorganized signal sequence to obtain a windowed signal sequence.

5. The method for separating and identifying same-frequency signals based on real-time spectrum analysis according to claim 2, characterized in that: Step S24 is further as follows: S241, based on the windowed signal sequence, using the least squares method, calculate the instantaneous frequency; based on the instantaneous frequency, determine the optimal fractional order parameter set by a grid search method; S242, constructing a fractional-order Fourier transform kernel function based on the fractional-order parameter set, discretizing the fractional-order Fourier transform kernel function, and generating a transformation matrix; S243, performing fractional Fourier transform based on the transformation matrix and the windowed signal sequence to obtain an initial time-frequency distribution; performing adaptive time-frequency focusing on the initial time-frequency distribution to obtain an enhanced time-frequency distribution; S244: Perform interference suppression processing on the enhanced time-frequency distribution to obtain a final time-frequency distribution matrix.

6. The same-frequency signal separation and identification method based on real-time spectrum analysis according to claim 2 is characterized in that: Step S25 is further as follows: S251, performing multi-scale decomposition based on the time-frequency distribution matrix to generate sub-band image sets with different frequencies and time resolutions; S252, extracting local features based on each sub-band image in the sub-band image set to form a feature descriptor set; The local features include texture features, edge information and inflection point information; S253, constructing a spectrum feature tree based on the feature descriptor set; performing feature matching and consistency check on the spectrum feature tree to obtain feature correspondence; S254, constructing an adaptive weight fusion strategy based on the feature correspondence and generating a fusion weight matrix; Based on the fusion weight matrix and the sub-band image set, weighted fusion is performed to obtain a high-resolution spectrum map.

7. The same-frequency signal separation and identification method based on real-time spectrum analysis according to claim 2 is characterized in that: Step S32 is further as follows: S321, obtaining the RF-amplitude spectrum statistics output by the power correction module, applying wavelet packet transform to perform multi-scale decomposition, and obtaining a multi-resolution representation of the spectrum; based on the multi-resolution representation, calculating the local statistical features of each sub-band to form a feature vector set; The local statistical features include energy entropy, spectrum centroid, spectrum bandwidth, kurtosis and skewness; S322, inputting the feature vector set into a pre-trained variational autoencoder network to perform nonlinear dimensionality reduction to obtain a low-dimensional latent space representation; based on the low-dimensional latent space representation, applying a dynamic time warping algorithm to calculate the similarity matrix between the current frame and the historical frame to obtain a temporal consistency score; S323. Based on the temporal consistency score and the low-dimensional latent space representation, an adaptive kernel density estimation model is constructed to obtain a multidimensional probability density distribution; based on the multidimensional probability density distribution, an expectation maximization algorithm is applied for clustering to obtain a preliminary signal classification result; S324. Based on the preliminary signal classification result, a graph cutting algorithm is applied to perform refinement to obtain an optimized signal classification result.

Images