Blind signal detection method and system based on ZS unsupervised clustering, and storage medium

Abstract

The invention discloses a blind signal detection method and system based on ZS unsupervised clustering, and a storage medium, relates to the technical field of signal detection, and aims to solve the problems that various types of signals cannot be detected at the same time and windowing coefficients and thresholds need to be set depending on experience or statistics in an existing method. Comprising the following steps of: 1, generating a candidate set: collecting power spectrum data, generating a preliminary candidate set by adopting a Z-Score filter and a Mean filter, and obtaining the candidate set by adopting differential encapsulation for the preliminary candidate set; the left boundary and the right boundary of the candidate set are expanded through a Start expander and an End expander according to the boundary slope, when the Start expander and the End expander meet slope fluctuation, a buffer area is set, continuous judgment is conducted in the buffer area, and if the slope characteristic identical to that of the corresponding boundary exists, the candidate set is expanded; otherwise, ending expansion; and merging the expanded candidate set to obtain a complete signal, namely a frequency band of the blind signal, thereby realizing blind signal detection.

Description

Technical Field

[0001] The present invention relates to the field of signal detection technology, and in particular to a blind signal detection method, system and storage medium based on ZS unsupervised clustering. Background Art

[0002] Traditional frequency domain signal detection methods mainly include spectrum analysis, matched filter detection algorithm, cyclostationary feature detection algorithm, adaptive threshold detection algorithm based on power spectrum, etc. The spectrum analysis method is to perform Fourier transform on the received data to obtain the power spectrum of the signal. According to the spectrum amplitude of the signal in the power spectrum, it can be determined whether there is a signal, and further detect the signal's carrier frequency, bandwidth and other parameters. The disadvantage is that for application scenarios such as short signal duration and low radar duty cycle, this method cannot obtain the spectrum peak of the frequency band where the signal is located, which will lead to signal leakage. "Burst Detection of Frequency-Hopping and Time-Hopping Signals Based on Interval Autocorrelation" discloses a matched filter detection method, which is composed of maximizing the signal-to-noise ratio to complete spectrum perception. The method has high detection accuracy and short detection time. The disadvantage is that it requires a deeper understanding of the prior information of the detection signal, such as the modulation mode, bandwidth and working frequency band of the signal, which is not suitable for blind signal detection. "Research on Spectrum Sensing Technology Based on Artificial Intelligence" discloses a cyclostationary feature detection algorithm, which is composed of using the periodicity inside the signal to detect the signal. The signals used today have unique periodic properties after artificial modulation and other operations, but noise does not have this property. The algorithm can detect signals in poor channel environments or when the signal-to-noise ratio is low. The disadvantage is that the computational complexity is high and it cannot detect signals without periodic characteristics. Saarnisaari H et al. disclosed a signal detection method based on window analysis, which is composed of setting different window sizes, extracting the features of the signal in the window, and thus detecting the signal. The disadvantage is that the window size setting of the method will affect the detection accuracy, and it is impossible to set appropriate window coefficients for different signals in the application scenario of blind signals. Summary of the invention

[0003] The technical problems to be solved by the present invention are:

[0004] Existing methods cannot detect multiple types of signals at the same time, and need to rely on experience or statistics to set windowing coefficients and thresholds.

[0005] The present invention adopts the following technical solutions to solve the above technical problems:

[0006] The present invention provides a blind signal detection method based on ZS unsupervised clustering, comprising the following steps:

[0007] Step 1: Generation of candidate sets: Collect power spectrum data, use Z-Score filter and Mean filter to generate preliminary candidate sets, and use differential encapsulation to obtain candidate sets for the preliminary candidate sets;

[0008] Step 2: Expansion of candidate set: expand the left and right boundaries of the candidate set according to the boundary slope, set a buffer, continuously judge the boundary characteristics in the buffer, merge the expanded candidate set, and obtain a complete signal.

[0009] Furthermore, the Z-Score filter and the Mean filter are used in step 1 to generate a preliminary candidate set, including the following process:

[0010] Power spectrum pxx n Perform Z-Score calculation, the calculation formula is as follows:

[0011]

[0012] Where: pxx i is the power value of the i-th data, μ and σ represent the mean and variance respectively;

[0013] Get the Z dispersion array:

[0014] Z=[z 1 , z 2 , z 3 , z 4 , z 5 …z n ]

[0015] By selecting the filter value ε for filtering and filtering out redundant data through the Mean filter, a preliminary candidate set is obtained.

[0016] Furthermore, in step 1, for the preliminary candidate set, differential encapsulation is used to obtain the candidate set, and the specific process is as follows:

[0017] Define the difference array Δ i is the preliminary candidate set array The difference between adjacent elements in is:

[0018]

[0019] The difference array Δ is obtained, which is expressed as follows:

[0020] Δ={Δ 1 , Δ 2 ,…,Δ n}

[0021] If Δ i >θ, then extract the interval in and They respectively represent the start and end of the preliminary candidate set currently being traversed; the candidate set Cx is expressed as:

[0022]

[0023] Where: C k represents the kth interval that meets the conditions, and m is the number of intervals that meet the conditions.

[0024] Furthermore, step one also includes smoothing the collected power spectrum data, specifically:

[0025]

[0026] Where: c j are the coefficients obtained by polynomial fitting, and m is half the size of the window.

[0027] Furthermore, step 2 includes the following process:

[0028] For each candidate set array subarray, first find its starting element variable s and ending element variable e; use the Start expander and End expander to expand the left and right boundaries of the subarray according to the slope respectively, and update the expanded subarray back to the signal candidate set array C_Arrays. After all subarrays are updated, the updated candidate sets are merged into a final complete signal.

[0029] Furthermore, the function implementation process of the Start expander in step 2 is as follows: first, the slope is calculated according to the current position; if the slope is greater than or equal to 0, the current position is added to the candidate set subarray, and then moved to the left; if the slope is less than 0, it is allowed to move to the left, but the maximum number of moves is limited, and a point with a slope greater than or equal to 0 is found in the range, and the candidate set is updated; if no point with a slope greater than or equal to 0 is found, or all frequency points have been traversed, the search is terminated;

[0030] The function implementation process of the End expander is as follows: first, the slope is calculated according to the current position; if the slope is less than or equal to 0, the current position is added to the candidate set subarray and then moves to the right; if the slope is greater than 0, moving to the right is allowed, but the maximum number of moves is limited, a point with a slope less than or equal to 0 is found in the range, and the candidate set is updated; if no point with a slope less than or equal to 0 is found, or all frequency points have been traversed, the search is terminated.

[0031] The present invention provides a blind signal detection system based on ZS unsupervised clustering. The system has a program module corresponding to the steps of the method described in any one of the above technical solutions, and executes the steps in the above blind signal detection method based on ZS unsupervised clustering during operation.

[0032] The present invention provides a computer-readable storage medium, which stores a computer program. The computer program is configured to implement the steps in the blind signal detection method based on ZS unsupervised clustering described in any one of the above technical solutions when called by a processor.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention provides a blind signal detection method, system and storage medium based on ZS unsupervised clustering, adopting ZS (Z-ScoreSlope) unsupervised clustering algorithm, and performing blind signal detection by performing a segmentation strategy of generating candidate sets and expanding candidate sets on power spectrum data. First, a preliminary candidate set is generated by using a preliminary candidate set generator, and then each candidate set is encapsulated by differential, and then the candidate set is expanded by an expander, and a cache buffer is added to assist the expansion of the expander, and finally a complete signal set is generated.

[0035] The method of the present invention can detect the signal range of the actual broadband service signal and narrowband service signal relatively clearly, and can also detect the signal segment with strong interference; it can also detect the narrowband and broadband signals of the mixed service signal well; the method of the present invention performs well in terms of signal frequency band accuracy and recall rate, and has high performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Flow chart of a blind signal detection method based on ZS unsupervised clustering in an embodiment of the present invention;

[0037] Figure 2 is a preliminary candidate set result diagram in an embodiment of the present invention;

[0038] Figure 3 is a candidate set result graph in an embodiment of the present invention;

[0039] Figure 4 is a Savitzky-Golay smoothing result graph in an embodiment of the present invention;

[0040] Figure 5 is a diagram of broadband signal detection results in an embodiment of the present invention;

[0041] Figure 6 is a diagram of narrowband signal detection results in an embodiment of the present invention;

[0042] Figure 7 is a diagram of the mix signal detection result in an embodiment of the present invention;

[0043] Figure 8 It is a statistical comparison diagram of accuracy and false alarm rate in an embodiment of the present invention;

[0044] Fig. 9 Statistical comparison chart of precision, recall and F1 score in the embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementations or embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described implementations or embodiments are only implementations or embodiments of a part of the present invention, not all of them. Based on the implementations or embodiments of the present invention, all other implementations or embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0047] Specific implementation plan 1: Figure 1 As shown, the present invention provides a blind signal detection method based on ZS unsupervised clustering, comprising the following steps:

[0048] Step 1: Generation of candidate sets: Collect power spectrum data, use Z-Score filter and Mean filter to generate preliminary candidate sets, and use differential encapsulation to obtain candidate sets for the preliminary candidate sets;

[0049] Step 2: Expansion of the candidate set: The Start expander and the End expander are used to expand the left and right boundaries of the candidate set according to the boundary slope. When the Start expander and the End expander encounter slope fluctuations, a buffer Cache is set up, and continuous judgment is made in the Cache buffer. If there is a slope characteristic that is the same as the corresponding boundary, the candidate set is expanded; otherwise, the expansion is terminated; the expanded candidate set is merged to obtain a complete signal, that is, the frequency band of the blind signal, thereby realizing blind signal detection.

[0050] Specific implementation scheme 2: The Z-Score filter and Mean filter are used to generate a preliminary candidate set as described in step 1, including the following process:

[0051] Power spectrum pxx n Perform Z-Score calculation, the calculation formula is as follows:

[0052]

[0053] Where: pxx i is the power value of the i-th data; μ and σ represent the mean and variance respectively.

[0054] Get the Z dispersion array:

[0055] Z=[z 1 , z 2 , z 3 , z 4 , z 5 …z n ]

[0056] By selecting the filter value ε, ε∈[1,3] for the first filtering, we can obtain data with a large degree of dispersion, namely the preliminary candidate set. In the generated preliminary candidate set, there will be a situation where the candidate set is smaller than the overall data, so it is necessary to add a Mean filter to filter out these redundant preliminary candidate sets and obtain the filtered preliminary candidate set.

[0057] Specific implementation plan three: In step 1, for the preliminary candidate set, differential encapsulation is used to obtain the candidate set. The specific process is as follows:

[0058] Define the difference array Δ i is the preliminary candidate set array The difference between adjacent elements in is:

[0059]

[0060] The difference array Δ is obtained, which is expressed as follows:

[0061] Δ={Δ 1 , Δ 2 ,…,Δ n}

[0062] If Δ i >θ, then extract the interval in and They respectively indicate the start and end of the preliminary candidate set currently being traversed; candidate set C k It is expressed as:

[0063]

[0064] Where: c k represents the kth interval that meets the conditions, and m is the number of intervals that meet the conditions. The rest of this implementation is the same as the second implementation.

[0065] Specific implementation scheme 4: Step 1 also includes smoothing the collected power spectrum data, specifically:

[0066]

[0067] Where: c j are the coefficients obtained by polynomial fitting, and m is half the size of the window.

[0068] Since the actual business signal noise is large and it is difficult to separate the signal from the noise, the power spectrum is smoothed. Compared with filters such as mean smoothing and local weighted regression smoothing, it can not only reduce the noise but also maintain the characteristics of the signal, such as peak and fluctuation characteristics. The rest of this implementation plan is the same as the specific implementation plan three.

[0069] Specific implementation plan 5: Step 2 includes the following process:

[0070] Candidate set C k Use Start and End expanders to expand the left and right boundaries. For each candidate array subarray, first find its start element variable s and end element variable e; use s_index and e_index to call Start expander and End expander to expand the left and right boundaries of subarray according to the slope, and update the expanded subarray back to the signal candidate array C_Arrays. After all subarrays are updated, use Signal function to merge the updated candidate sets into a final complete signal. The rest of this implementation plan is the same as the specific implementation plan four.

[0071] The candidate set expansion pseudo code is:

[0072]

[0073] Specific implementation scheme six: The function implementation process of the Start expander described in step 2 is as follows: first, the slope is calculated according to the current position; if the slope is greater than or equal to 0, the current position is added to the candidate set subarray, and then moved to the left; if the slope is less than 0, it is allowed to move to the left, but the maximum number of moves is limited, and a point with a slope greater than or equal to 0 is found within the range, and the candidate set is updated; if no point with a slope greater than or equal to 0 is found, or all frequency points have been traversed, the search is terminated;

[0074] The function implementation process of the End expander is as follows: first, the slope is calculated according to the current position; if the slope is less than or equal to 0, the current position is added to the candidate set subarray, and then moved to the right; if the slope is greater than 0, it is allowed to move to the right, but the maximum number of moves is limited, and a point with a slope less than or equal to 0 is found in the range, and the candidate set is updated; if a point with a slope less than or equal to 0 is not found, or all frequency points have been traversed, the search is terminated. The rest of this implementation plan is the same as the specific implementation plan four.

[0075] The End expander pseudo code is:

[0076]

[0077] The blind signal detection method (algorithm) based on ZS unsupervised clustering proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0078] Based on the method proposed in the present invention, a blind signal detection system based on ZS unsupervised clustering is developed using a programming language. The system has a program module corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned blind signal detection method based on ZS unsupervised clustering during operation.

[0079] The computer program of the developed system (software) is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-mentioned blind signal detection method based on ZS unsupervised clustering when called by a processor, that is, the present invention is materialized on a carrier to become a computer program product.

[0080] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0081] The computer programs (also referred to as programs, software, software applications, or codes) of the present invention include machine instructions for programmable processors, and these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or device (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0082] The beneficial effects of the present invention will be described below in conjunction with specific embodiments.

[0083] In order to verify the practicability of the method of the present invention in engineering, all experimental data use actual business signals, as shown in Table 1. Such actual business signals have complex characteristics such as large noise and interference, uncertain signal components, and varying signal bandwidths.

[0084] Table 1 Data

[0085]

[0086] Example 1

[0087] Detecting a blind signal of a wireless communication system includes the following steps:

[0088] Step 1: Assume a broadband wireless communication system where the receiver uses a MIMO antenna array to receive signals from multiple transmitters. Each transmitter sends a BPSK modulated signal with a bandwidth of 5MHz and a carrier frequency of 2GHz. The receiver has an array of 8 antennas with a sampling frequency of 20MHz.

[0089] Step 2: Perform 2048-point FFT processing on the received signal.

[0090] Step 3: Use the SGolay (Savitzky-Golay) filter to smooth the power spectrum data.

[0091] Step 4: Use the Z-Score filter to obtain a preliminary candidate set, and then use the Mean filter to generate a preliminary candidate set.

[0092] Step 5: The preliminary candidate set is differentiated to obtain the candidate set.

[0093] Step 6: Use the Start and End expanders to expand the left and right boundaries of the candidate set to obtain the signal set.

[0094] Step 7: Detect the number of signals and frequency range, such as Figure 5 The broadband signal detection graph results.

[0095] This embodiment detects the actual broadband service signal, which is a wide spectrum signal with a wide frequency range, usually covering multiple frequency bands, and may cause continuous signal fluctuations and interference, and the detection effect is often poor using general detection methods. After detection in this embodiment, the signal range can be detected more clearly, and signal segments with strong interference can also be detected.

[0096] Example 2

[0097] Detecting targets in a radar-free system includes the following steps:

[0098] Step 1: Set up a pulse radar system with an array antenna at the receiving end to receive the target echo signal. The pulse signal reflected by the target has a frequency of 10GHz, a pulse width of 1μs, and a repetition period of 5ms. The array has 4 antenna units, a sampling frequency of 50MHz, and down-conversion.

[0099] Step 2: Perform 1024-point FFT processing on the received signal.

[0100] Step 3: Use the SGolay (Savitzky-Golay) filter to smooth the power spectrum data.

[0101] Step 4: Use the Z-Score filter to obtain a preliminary candidate set, and then use the Mean filter to generate a preliminary candidate set.

[0102] Step 5: The preliminary candidate set is differentiated to obtain the candidate set.

[0103] Step 6: Use the Start and End expanders to expand the left and right boundaries of the candidate set to obtain the signal set.

[0104] Step 7: Detect the number of signals and frequency range, such as Figure 6 The narrowband signal detection diagram results.

[0105] This embodiment detects the actual narrowband service signal. The energy of the local frequency bandwidth of the actual narrowband service signal may also be submerged by noise, making it difficult to identify. It is difficult to accurately detect the signal segment using general detection methods. After experiments, this embodiment can detect the signal range more clearly and accurately capture such narrowband signals.

[0106] Example 3

[0107] Detecting blind signals in a satellite communication link includes the following steps:

[0108] Step 1: In a low-orbit satellite communication system, the satellite signal is received, and the goal is to detect the signal transmitted by the ground station. The satellite communication signal is modulated by QPSK and the carrier frequency is 12GHz. The receiving array has 6 antenna elements and the sampling frequency is 100MHz.

[0109] Step 2: Perform 2048-point FFT processing on the received signal.

[0110] Step 3: Use the SGolay (Savitzky-Golay) filter to smooth the power spectrum data, such as Figure 4 shown.

[0111] Step 4: Use the Z-Score filter to get Figure 2The preliminary candidate set shown is then generated using the Mean filter.

[0112] Step 5: The preliminary candidate set is differentiated to obtain the candidate set, such as Figure 3 shown.

[0113] Step 6: Use the Start and End expanders to expand the left and right boundaries of the candidate set to obtain the signal set.

[0114] Step 7: Detect the number of signals and frequency range, such as Figure 7 The mix signal detection diagram results.

[0115] This embodiment detects the actual mixed service signal. The actual mixed service signal contains both narrowband signals and broadband signals, which makes signal detection more difficult and puts higher requirements on the robustness of the method. The method of the present invention can better detect narrowband and broadband signals for such mixed signals.

[0116] The method of the present invention is compared with the existing single-threshold detection method and the dual-threshold detection method, and the service signals of Examples 1-3 are detected using the single-threshold detection method and the dual-threshold detection method respectively.

[0117] The single threshold detection method is suitable for detecting narrowband signals. It compares the signal power spectrum with the noise power spectrum, and combines the set threshold and the signal-to-noise ratio (SNR) wall of signal detection to determine whether the signal exists. The power of the signal is calculated as:

[0118]

[0119] Where: P(f) represents the frequency at f 1 to f 2 The signal power between .

[0120] Select a fixed threshold γ, which can be selected based on the statistical characteristics of the system noise or experience; then the signal detection is as follows:

[0121]

[0122] The dual threshold detection method uses the gradient information of the Welch power spectrum estimation to compare with the adaptive dual threshold, and completes the signal detection and signal frequency band range positioning by detecting the change of the power spectrum gradient value. It is suitable for detecting broadband signals.

[0123] The determination of the adaptive double threshold first calculates the mean of the input data global and standard deviation σ global , use these two values ​​to calculate the adaptive upper threshold UT global and lower threshold LTglobal :

[0124] UT global =mean global +kσ global

[0125] LT global =mean global -kσ global

[0126] Use a sliding window to traverse the input data and calculate the local mean within the window local and the local standard deviation σ local , calculate the local upper threshold UT local and the local lower threshold LT local :

[0127] UT local =mean local +kσ local

[0128] LT local =mean local -kσ local

[0129] The k in the formula is a coefficient greater than 0 and can be set based on experience.

[0130] In order to ensure the rationality of the threshold and to suppress the excessive change of the threshold to a certain extent, the local upper threshold UT local With adaptive upper threshold UT global For comparison, if the local upper threshold UT local Below the adaptive upper threshold UT global , then use the adaptive upper threshold UT global ; Use the local lower threshold LT local With adaptive lower threshold LT global For comparison, if the local lower threshold LT local If the adaptive lower threshold is higher than the maximum adaptive lower threshold, the adaptive lower threshold LT is used. global :

[0131] ut=max(UT global , U.T. local )

[0132] lt=max(LT global , L T local )

[0133] Get the adaptive upper and lower threshold array:

[0134] UT=[ut 1 ,ut2 ,ut 3 ,ut 4 ,ut 5 …ut n ]

[0135] LT=[lt 1 ,lt 2 ,lt 3 ,lt 4 ,lt 5 …lt n ]

[0136] Where: ut n Represents the adaptive upper threshold obtained each time the window slides; lt n Represents the adaptive lower threshold obtained each time the window slides.

[0137] Find the gradient of the power spectrum:

[0138]

[0139] The judgment criterion for power spectrum signal detection is that when a gradient value crosses the upper threshold and the next gradient value also crosses the upper threshold or a gradient value crosses the lower threshold and the next gradient value also crosses the lower threshold, the current area is also noise; when a gradient value crosses the upper threshold and the next gradient value crosses the lower threshold, the current area is a signal.

[0140] The accuracy, false alarm rate, precision, recall rate, F1 score and time complexity are used as evaluation criteria to compare the method of the present invention with the single threshold detection method and the double threshold detection method. The average of the evaluation criteria results of each method in Examples 1-3 is compared. The results are as follows: Figure 8 and 9 As shown, it can be seen that in terms of signal detection number accuracy and false alarm rate, the accuracy of the single threshold and double threshold detection methods is not high and the false alarm is too large, and they cannot achieve good detection effects in practical applications. The method of the present invention has a better effect on the number of signals. In terms of signal frequency band accuracy, recall rate, and F1 score, the method of the present invention performs best, with high accuracy and recall rate, and the highest F1 score, indicating that it performs well in balancing accuracy and recall rate. The double threshold detection method is better than the single threshold detection method in terms of accuracy and recall rate, and the F1 score is also improved, but there is still a large gap compared with the method of the present invention. The performance of the single threshold detection method is the worst, especially the accuracy and F1 score are low, showing its shortcomings in accuracy and positive class capture ability. In terms of time complexity, the single threshold detection method is 23%, the double threshold detection method is 26%, and the method of the present invention is 24. The time complexity of the three methods is not much different.

[0141] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A blind signal detection method based on ZS unsupervised clustering, characterized in that: The steps include: Step 1: Generation of candidate sets: Collect power spectrum data, use Z-Score filter and Mean filter to generate preliminary candidate sets, and use differential encapsulation to obtain candidate sets for the preliminary candidate sets; Step 2: Expansion of candidate set: expand the left and right boundaries of the candidate set according to the boundary slope, set a buffer, continuously judge the boundary characteristics in the buffer, merge the expanded candidate set, and obtain a complete signal.

2. The blind signal detection method based on ZS unsupervised clustering according to claim 1, characterized in that: In step 1, the Z-Score filter and Mean filter are used to generate a preliminary candidate set, including the following process: Power spectrum pxx n Perform Z-Score calculation, the calculation formula is as follows: Where: pxx i is the power value of the i-th data, μ and σ represent the mean and variance respectively; Get the Z dispersion array: Z=[z1,z2,z3,z4,z5…z n ] By selecting the filter value ε for filtering and filtering out redundant data through the Mean filter, a preliminary candidate set is obtained.

3. The blind signal detection method based on ZS unsupervised clustering according to claim 2, characterized in that: In step 1, for the preliminary candidate set, differential encapsulation is used to obtain the candidate set. The specific process is as follows: Define the difference array Δ i is the preliminary candidate set array The difference between adjacent elements in is: The difference array Δ is obtained, which is expressed as follows: Δ={Δ1, Δ2,…, Δ n } If Δ i >θ, then extract the interval in and They respectively indicate the start and end of the preliminary candidate set currently being traversed; candidate set C k It is expressed as: Where: C k represents the kth interval that meets the conditions, and m is the number of intervals that meet the conditions.

4. The blind signal detection method based on ZS unsupervised clustering according to claim 3, characterized in that: Step 1 also includes smoothing the collected power spectrum data, specifically: Where: c i are the coefficients obtained by polynomial fitting, and m is half the size of the window.

5. The blind signal detection method based on ZS unsupervised clustering according to claim 4, characterized in that: Step 2 includes the following process: For each candidate set array subarray, first find its starting element variable s and ending element variable e; use the Start expander and End expander to expand the left and right boundaries of the subarray according to the slope respectively, and update the expanded subarray back to the signal candidate set array C_Arrays. After all subarrays are updated, the updated candidate sets are merged into a final complete signal.

6. The blind signal detection method based on ZS unsupervised clustering according to claim 5, characterized in that: The functional implementation process of the Start expander described in step 2 is as follows: first, the slope is calculated according to the current position; if the slope is greater than or equal to 0, the current position is added to the candidate set subarray and then moved to the left; if the slope is less than 0, it is allowed to move to the left, but the maximum number of moves is limited, and a point with a slope greater than or equal to 0 is found in the range, and the candidate set is updated; if no point with a slope greater than or equal to 0 is found, or all frequency points have been traversed, the search ends; The function implementation process of the End expander is as follows: first, the slope is calculated according to the current position; if the slope is less than or equal to 0, the current position is added to the candidate set subarray and then moves to the right; if the slope is greater than 0, moving to the right is allowed, but the maximum number of moves is limited, a point with a slope less than or equal to 0 is found in the range, and the candidate set is updated; if no point with a slope less than or equal to 0 is found, or all frequency points have been traversed, the search is terminated.

7. A blind signal detection system based on ZS unsupervised clustering, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 6, and executes the steps of the blind signal detection method based on ZS unsupervised clustering when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps in the blind signal detection method based on ZS unsupervised clustering according to any one of claims 1 to 6 when called by a processor.

Images