Modulation identification method based on scoring system

Through a scoring system-based method, combining power spectrum, bandwidth and center frequency, combined with instantaneous and higher-order feature parameters, the identification and confidence evaluation of signal modulation type is realized, and the modulation recognition method in the prior art is not compatible and extended. The problem of sexuality improves the accuracy and robustness of signal analysis.

CN119094293APending Publication Date: 2024-12-06卢彦卿
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411177911.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing signal modulation type recognition methods cannot be compatible, and lack scalability, making it difficult to adapt to the larger range of modulation recognition types, resulting in difficulty in signal analysis and demodulation in non-cooperative communication.

Method used

Using a modulation recognition method based on the scoring system, digital downconversion and resampling are performed by calculating the power spectrum, bandwidth and central frequency point of the signal. Combining instantaneous and higher-order feature parameters, the modulation type is scored using 13-dimensional feature attributes to achieve Identification and confidence assessment of modulation type.

Benefits of technology

It achieves compatibility with different analysis angles and methods, has good scalability, can accurately identify multiple modulation types, and improves the accuracy and robustness of signal analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119094293A_ABST
    Figure CN119094293A_ABST
Patent Text Reader

Abstract

The invention discloses a modulation identification method based on a scoring system, and aims to solve the problems that various existing classic modulation type identification methods are bound with an identification type set, the various methods are incompatible and keen, and mutual compatibility optimization and identification type expansion are not easy to further carry out in a classic judgment tree mode. According to the method, a scoring system is introduced, various modulation identification analysis methods are integrated, a scoring mode is adopted, various analysis methods of different dimensions are linearized into one-dimensional score values, each group of input data is converted into scores under different modulation types through calculation, and the modulation type corresponding to the highest score is the identified modulation type. Meanwhile, the confidence degree of the recognition result is given according to the score distribution condition of the modulation type set.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of wireless communication, signal analysis and signal recognition, and in particular to signal modulation type recognition. Background Art

[0002] Signal modulation type identification has always been a key and difficult issue in the field of signal analysis and identification, especially in some non-cooperative communications. The accuracy of modulation identification directly determines the accuracy of signal analysis and whether the signal can be correctly demodulated.

[0003] At present, there are many types of modulation type recognition methods proposed in the industry, and there is no consensus. They all analyze and summarize the modulation recognition methods from different analysis angles, but are limited to their own modulation recognition sets. There is no unified standard method that is compatible with each other. The latest modulation type recognition based on deep learning is still in the theoretical research stage.

[0004] Among the existing recognition methods, the more classic ones generally include the decision tree method based on instantaneous parameter calculation, the high-order cumulant eigenvalue matching method, and the characteristic spectrum analysis method. The main problem that these methods cannot be compatible with each other is that they all adopt a "tree-type" recognition method, that is, a certain modulation type cannot exist on two or more "branch" decision paths, which can only be applied to a smaller modulation recognition range. A large range of modulation recognition types conflicts with the unique decision path method of this "branch". However, these analysis methods are proven, classic, reasonable, and meaningful modulation recognition methods in practice.

[0005] Therefore, in actual signal recognition, it is urgent to introduce a modulation type recognition method that is compatible and has strong scalability. Summary of the invention

[0006] The purpose of the present invention is to provide a compatible modulation type identification method that integrates different analysis angles and theoretical methods. At the same time, the method has good scalability, and the method and modulation identification type can be easily expanded in practical applications.

[0007] In order to achieve the above object, the present invention adopts the following technical scheme:

[0008] A modulation recognition method based on a scoring system, characterized by comprising the following steps:

[0009] (1) Let the original sampled data be a(n), where n = 1, 2, 3, ..., and the sampling rate of the corresponding input signal be f sFirst, the signal power spectrum of the first frame data is calculated with 4096 sampling chips as one frame; then after shifting 2048 chips, 4096 chips are taken to calculate the power spectrum of the second frame data, that is, the sampling data of the first frame and the second frame have The frame data are overlapped to make the spectrum curve flatter and smoother; and similarly, the power spectra of the 3rd, 4th, ..., 64th frames of input signals are calculated.

[0010] (2) Calculate the mean of the 64 consecutive frames of power spectrum obtained in the previous step to achieve noise reduction and obtain the spectrum mean. The spectrum mean is further used to estimate the bandwidth and center frequency of the input signal: Here, the N-dB bandwidth estimation method is used, and the default N is 10dB, that is, first find the maximum value P of the mean spectrum max , and the location Index max{P} ; Then set the threshold value 10dB below the maximum value as threshold = P max -10dB, the length of the horizontal axis corresponding to the part above the threshold in the calculated spectrum mean sequence is the signal 10-dB bandwidth BW, and the center position of the horizontal axis above the threshold is the center frequency position f of the signal. c .

[0011] (3) Find the center frequency f of the input signal c After the bandwidth BW is calculated, the signal is digitally down-converted to move the input signal spectrum to the position of zero intermediate frequency, and the signal is resampled at a resampling rate of 8 times the bandwidth oversampling, that is Since the signal bandwidth is estimated and may have a decimal part, the resampling rate is 8 times the signal bandwidth and rounded up.

[0012] (4) Assume that the new sampled data {x(m)|m=1,2,3,…} is obtained after the above 8-fold bandwidth oversampling, and then the resampled data is processed frame by frame with a frame length of 32768. Then the i-th frame data can be expressed as {x i (j)|j=1,2,3,…,32768}.

[0013] (5) The i-th frame data is first normalized, that is, the amplitude range is normalized to between -1 and +1. The normalization calculation expression is: That is, the maximum amplitude in the i-th frame data is extracted, and each code chip in the frame data is divided by the maximum amplitude.

[0014] (6) Normalized data Calculate the five instantaneous characteristic parameters, namely the maximum value of the zero-centered normalized instantaneous amplitude spectrum γ i,max , the standard deviation of the absolute value of the zero-centered normalized instantaneous amplitude σ i,a , the standard deviation of the absolute value of the zero-centered normalized instantaneous frequency σi,f , envelope variation coefficient R i , envelope kurtosis coefficient K i,42 .

[0015] The calculation formulas of instantaneous characteristic parameters are as follows:

[0016] The maximum value of the zero-centered normalized instantaneous amplitude spectrum That is, first calculate the amplitude sequence of each frame of data by taking the modulus, then calculate the amplitude spectrum of the amplitude sequence by FFT, and then extract the maximum value. i,max It reflects the change of signal envelope. The smaller the corresponding value is, the greater the possibility of constant envelope.

[0017] Zero-centered normalized instantaneous amplitude absolute value standard deviation That is, the amplitude value of each frame of data is first calculated by taking the modulus, and then the standard deviation of the amplitude value sequence in a statistical sense is calculated.

[0018] Zero-centered normalized instantaneous frequency absolute value standard deviation That is, first transform the time domain signal into the frequency domain through FFT calculation, and obtain the frequency domain sequence {f i (n)|n=1,2,3,…,32768}, and then calculate the absolute value sequence of each frame of data in the frequency domain by taking the modulus, and then calculate the standard deviation of the absolute value sequence in the frequency domain in a statistical sense.

[0019] Envelope variation coefficient In the calculation formula That is, first calculate the sequence envelope by taking the modulus, and then calculate the mean value avrg of the envelope i,1 , and then subtract the mean avrg from each complex element of the envelope sequence i,1 , the obtained complex sequence is used to find the square sequence of the amplitude, and then the square sequence of the amplitude is averaged to get avrg i,2 , and finally calculate avrg i,2 and (avrg i,1 ) 2 The ratio of is the envelope variation coefficient, which reflects the change of the signal envelope.

[0020] Envelope Kurtosis in That is, the envelope change coefficient R of the previous characteristic parameter i The calculation is similar, except that the fourth power of the difference modulus of the complex sequence minus the envelope mean is further calculated, and then the fourth power mean avrg is further calculated. i,4 and the quadratic mean avrg i,2The envelope peak coefficient is obtained by calculating the ratio of the signal amplitude to the signal amplitude. The envelope peak coefficient reflects the compactness of the signal amplitude. For example, the value under ASK modulation is smaller than that under AM modulation.

[0021] (7) Then normalize the data Calculate the characteristic parameters of high-order cumulants {C 42}, the fourth-order cumulant calculation formula is Among them, M i,42 is the fourth-order moment, M i,20 and M i,21 is the second-order moment, and the corresponding calculation formulas are: Where conj(·) represents the conjugate calculation of complex numbers.

[0022] (8) Then normalize the data Calculate seven major spectrum characteristic parameters, namely the number of effective peaks of the primary spectrum S i,1 , the number of effective peaks of the envelope spectrum S i,a , effective peak number of delay spectrum S i,b , the number of effective peaks of the secondary spectrum S i,2 , the effective peak number of the cubic spectrum S i,3 , the effective peak number of the quartic spectrum S i,4 , the number of effective peaks of the octave spectrum S i,8 .

[0023] The calculation formulas of spectral characteristic parameters are as follows:

[0024] The formula for calculating the first spectrum is: That is, the primary spectrum is the power spectrum of the signal. The normalized data is directly transformed by FFT, then modulo, and then converted into a dB-valued primary spectrum sequence. Then, the threshold is set to 10 dB above the noise floor, and the number of effective peaks above the threshold in the primary spectrum sequence is counted. i,1 .

[0025] The envelope spectrum calculation formula is: That is, first calculate the amplitude of the signal, then perform FFT transformation on the amplitude sequence, then take the modulus, and then convert it into a dB-valued envelope spectrum sequence. Then, set the threshold to be 10 dB above the noise floor, and count the number of effective peaks S in the envelope spectrum sequence that are above the threshold. i,a .

[0026] The delay spectrum calculation formula is: That is, the resampled data is delayed by 1 chip and then multiplied with itself, and then FFT is performed, and then modulo is taken, and then converted into a delay spectrum sequence with dB value. Then, the threshold is set to 10dB above the noise floor, and the number of effective peaks above the threshold in the delay spectrum sequence is counted. i,bIn practice, the delay spectrum of a general MPSK modulated signal generally has spectral lines, and the position of the spectral lines corresponds to the symbol rate of the MPSK modulated signal. The exception is that the delay spectrum of an OQPSK modulated signal does not have spectral lines, which can be used to identify the OQPSK modulation type.

[0027] The calculation formula of the secondary spectrum is: That is, first calculate the quadratic power of the resampled complex sequence, then perform FFT transformation, then take the modulus, and then convert it into a dB-valued secondary spectrum sequence. Then, set the threshold of 10 dB above the noise floor, and count the number of effective peaks S in the secondary spectrum sequence that are above the threshold. i,2 .

[0028] The cubic spectrum is calculated as That is, first calculate the cube of the resampled complex sequence, then perform FFT transformation, then take the modulus, and then convert it into a dB-valued cubic spectrum sequence. Then, set the threshold of 10 dB above the noise floor, and count the number of effective peaks S in the cubic spectrum sequence that are above the threshold. i,3 .

[0029] The quartic spectrum is calculated as That is, first calculate the fourth power of the resampled complex sequence, then perform FFT transformation, then take the modulus, and then convert it into a dB-valued fourth spectrum sequence. Then, set the threshold of 10dB above the noise floor and count the number of effective peaks S above the threshold in the fourth spectrum sequence. i,4 .

[0030] The octave spectrum is calculated as That is, first calculate the eighth power of the resampled complex sequence, then perform FFT transformation, then take the modulus, and then convert it into an eighth spectrum sequence with dB value. Then, set the threshold of 10dB above the noise floor, and count the number of effective peaks S above the threshold in the eighth spectrum sequence. i,8 For high-order modulation methods such as 8PSK, the frequency multiplication calculation of the time domain data corresponds to the price reduction processing of the spectrum, which is conducive to the identification of high-order modulation methods.

[0031] (9) After the above calculation process, for each frame of resampled data All of the 13-dimensional feature attributes Features = {γ max ,σ a ,σ f ,R,K 42 ,C 42 ,S1,S a ,S b ,S2,S3,S4,S8}, for each eigenvalue, the modulation type set

[0032] In the above method, 1 point can be given to one or several modulation types that meet the conditions, and no point will be given to modulation types that do not meet the conditions.

[0033] The 13 characteristic attributes are used in turn to score the types that meet the conditions in the modulation type set, so that the 13-dimensional characteristic attributes are linearized into a one-dimensional score value. The modulation type with the highest score is the modulation type identified by the scoring system. The full score is 13 points. Even if one of the characteristic values ​​occasionally does not meet the theoretical range, the corresponding modulation type has the highest score. There is no decision path requirement, so it has a certain degree of robustness.

[0034] Finally, it is determined whether there is one and only one highest score among all the modulation type scores. If there is only one highest score, the modulation recognition is successful, and the modulation type corresponding to the highest score is the recognition result. The confidence of the recognition result is calculated, that is, the higher the score, the higher the confidence, and the smaller the interference type score, the higher the confidence. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The figure is a flow chart of a modulation recognition method based on a scoring system.

[0036] Figure 2 This is an example of a scoring table for 13-dimensional feature attributes under 14 known modulation methods.

[0037] Figure 3 This is a schematic diagram for 10dB bandwidth calculation.

[0038] Figure 4 Schematic diagram of the seven-dimensional spectrum characteristics of BPSK modulated signal.

[0039] Figure 5 Schematic diagram of the seven-dimensional spectrum characteristics of 8PSK modulation signal.

[0040] Figure 6 Schematic diagram of the seven-dimensional spectrum characteristics of 2FSK modulation signal.

[0041] Figure 7 Schematic diagram of the seven-dimensional spectrum characteristics of MSK modulated signals.

[0042] Figure 8 Schematic diagram of the seven-dimensional spectrum characteristics of 8FSK modulation signal.

[0043] Specific real-time method

[0044] The following is an explanation of the 8PSK modulation type identification of high-order modulation in conjunction with the accompanying drawings.

[0045] For the reliability of the system, a signal source is used to generate data, and an 8PSK modulated signal with a certain frequency, bandwidth, and signal power is randomly generated without loss of generality. Then, a signal acquisition device is used to obtain sampled data as input at a fixed sampling rate of 25MHZ, and then the modulation recognition scoring system begins to process the data as follows, including the following steps:

[0046] Step 1: Calculate the power spectrum by FFT with 4096 chips as one frame of data; then shift the half-frame data to correspond to 2048 chips, extract 4096 chips as one frame of data, and calculate the power spectrum of the second frame of data by FFT; and so on, calculate the power spectrum of the 3rd, 4th, ..., 64th frames of data.

[0047] Step 2: Average the corresponding positions of the power spectra of the first 64 frames calculated in step 1 to obtain the spectrum mean. Then further analyze the spectrum mean to calculate the 10dB bandwidth of the sampled signal and the center frequency of the signal. The 10dB bandwidth here refers to the maximum value of the power spectrum minus 10dB as the threshold, and the spectrum mean is traversed. The frequency band range occupied by the part above the threshold is taken as the bandwidth of the signal. Figure 3 To illustrate, the spectrum mean of the sampled signal has been calculated through steps 1 and 2. At this time, the maximum value of the spectrum mean is 48.48dB, so 47.48dB is used as the threshold value. The entire spectrum mean sequence is traversed, and the points in the range of [1597 to 2137] of the spectrum mean are all higher than the mean, so the signal bandwidth is And the signal center frequency can be obtained That is, the corresponding frequency deviation is 0.

[0048] Step 3: Since the frequency deviation is 0, the input data does not need to be spectrally shifted. 8 times bandwidth oversampling is used, so only the input data needs to be resampled. At the same time, in order to simplify the calculation, the sampling is rounded up in MHZ, so the resampling rate is used. Resample the input data.

[0049] Step 4: The resampled data, i.e., the 9 MHz sampling data, is processed frame by frame with 32768 chips as one frame, the modulation type is identified frame by frame, and the modulation identification result and confidence are output frame by frame.

[0050] For each frame of data after resampling, normalization processing is first required, so that each code piece of the current frame data is divided by the maximum code piece modulus within the frame, that is, the input data size is normalized to between [-1, +1].

[0051] Step 5: After normalizing each frame of data, the five major instantaneous characteristic parameters are calculated, including: the maximum value of the zero-centered normalized instantaneous amplitude spectrum γ max , the standard deviation of the absolute value of the zero-centered normalized instantaneous amplitude σa , the standard deviation of the absolute value of the zero-centered normalized instantaneous frequency σ f , envelope variation coefficient R, envelope peak coefficient K 42 .

[0052] As mentioned above, the five instantaneous characteristic parameters {γ max ,σ a ,σ f ,R,K 42} are all important parameters that have been verified and used in classical modulation recognition. Zero center refers to the zero intermediate frequency without frequency deviation. Zero center and normalization have been processed in steps 3 and 4, so the following calculations can be performed directly according to the calculation formula.

[0053] That is, first find the modulus of each code chip of the current frame data, then perform FFT transformation on the modulus sequence, then find the square of the modulus of the data transformed to the frequency domain, then extract the maximum value and divide it by the length of FFT 32768, and finally get γ max,8PSK =0.35.

[0054] That is, first find the modulus of each code piece of the current frame data, then calculate the mean of the modulus, and square the mean; then find the square of each code piece of the current frame data, and then find the mean of the modulus; then subtract the square of the mean from the mean of the modulus to get the variance value, and finally take the square root of the variance value to get the standard deviation, and get σ a,8PSK =1.22.

[0055] That is, first transform the current frame data into the frequency domain through FFT, and then calculate the standard deviation in the frequency domain. Similarly, first calculate the square of the mean of the frequency domain data, then calculate the mean of the square of the frequency domain data. The difference between the two is the variance value, and finally take the square root to get the standard deviation, and we can get σ f,8PSK =1.01.

[0056] in That is, we still need to calculate the average value of the modulus of the current frame data first, avrg i,1 , then subtract the mean from each code piece in the current frame, and then square the modulus to get avrg i,2 Finally, find the ratio of the two. Since avrg i,2 The square is taken, so the ratio avrg i,1 We also need to take the square, and we can get R i,8PSK =0.07.

[0057] in That is avrg i,1 ,avrg i,2It has been calculated in the previous step, so we only need to further subtract the mean avrg from each code piece in the current frame. i,1 Then, calculate the fourth power of the modulus and get K 42,8PSK =0.49.

[0058] Step 6: After normalizing each frame of data, start calculating the fourth-order cumulant. According to the fourth-order cumulant calculation formula You need to calculate the fourth-order moment of the sequence first and two second-order moments That is, first calculate the fourth power of each code chip modulus of the current data frame, then find the average to get the fourth-order moment; then calculate the mean of the square of each code chip and the mean of each code chip multiplied by its own conjugate to get two second-order moments; finally, calculate the fourth-order moment C of the current frame data according to the fourth-order cumulant formula 42,8PSK =0.7.

[0059] Step 7: After normalizing each frame of data, start calculating the seven major spectrum characteristic parameters, including: the number of peaks above the threshold in the primary spectrum S1, the number of peaks above the threshold in the envelope spectrum S a , the number of peaks above the threshold in the delay spectrum S b , the number of peaks above the threshold in the secondary spectrum is S2, the number of peaks above the threshold in the cubic spectrum is S3, the number of peaks above the threshold in the quartic spectrum is S4, and the number of peaks above the threshold in the octal spectrum is S8. The gradually increasing high-order spectrum mainly completes the modulation order reduction processing, and the specific calculation is as follows:

[0060] Primary spectrum That is, the primary spectrum is the power spectrum of the current frame data. First, complete the data power spectrum calculation, and then find the number of peaks higher than the threshold from the power spectrum curve. The threshold defaults to 10dB. Figure 5 As shown in the first sub-figure, the number of primary spectrum peaks S of 8PSK modulation 1,8PSK =0.

[0061] Envelope spectrum That is, first calculate the amplitude of the current frame data, then perform FFT transformation on the amplitude, calculate the power spectrum of the amplitude sequence, and then find the number of peaks above the threshold from the power spectrum curve. The threshold defaults to 10dB. Figure 5 As shown in the second sub-figure, the envelope spectrum peak number S of 8PSK modulation a,8PSK =5.

[0062] Delay spectrum That is, the current frame data is delayed by 1 code chip and then multiplied with the corresponding bit of the current frame data, and then the sequence after the point multiplication is transformed by FFT to calculate the delay spectrum. For most low-order modulation methods, the delay spectrum reflects the symbol rate of the modulated signal, that is, the frequency position corresponding to the maximum peak is exactly the symbol rate value of the signal. Then find the number of peaks above the threshold from the delay spectrum curve. The threshold defaults to 10dB. As shown in the attached figure Figure 5 As shown in the third sub-figure, the delay spectrum peak number S of 8PSK modulation b,8PSK =0.

[0063] Secondary spectrum That is, the square of each chip of the current frame data is calculated, and then the obtained square sequence is transformed by FFT, so as to calculate the quadratic spectrum, and then the number of peaks higher than the threshold is found from the quadratic spectrum curve. The threshold defaults to 10dB. Figure 5 As shown in the fourth sub-figure, the secondary spectrum peak number S of 8PSK modulation 2,8PSK =0.

[0064] Cubic Spectrum That is, the cube of each chip of the current frame data is calculated, and then the obtained cubic sequence is transformed by FFT, so as to calculate the cubic spectrum, and then the number of peaks higher than the threshold is found from the cubic spectrum curve. The threshold defaults to 10dB. Figure 5 As shown in the fifth sub-figure, the number of cubic spectrum peaks S of 8PSK modulation 3,8PSK =0.

[0065] Quartic spectrum That is, the fourth power of each chip of the current frame data is calculated, and then the obtained fourth power sequence is transformed by FFT, so as to calculate the fourth spectrum, and then the number of peaks higher than the threshold is found from the fourth spectrum curve. The threshold defaults to 10dB. Figure 5 As shown in the sixth sub-figure, the number of quartic spectrum peaks S of 8PSK modulation 4,8PSK =0.

[0066] Octave That is, calculate the eighth power of each code chip of the current frame data, and then perform FFT transformation on the obtained eighth power sequence to calculate the eighth spectrum, and then find the number of peaks above the threshold from the eighth spectrum curve. The threshold defaults to 10dB. Figure 5 As shown in the seventh sub-figure, the number of octal spectrum peaks S of 8PSK modulation 8,8PSK =5.

[0067] Step 8: After the input data is calculated through the above steps 1 to 7, the 13-dimensional feature attributes are as follows: Features = {γ max ,σ a ,σ f ,R,K42 ,C 42 ,S1,S a ,S b ,S2,S3,S4,S8}={0.35,1.22,1.01,0.49,0.7,0,5,0,0,0,0,0,5}, and then according to the attached Figure 2 The scoring table shown in the figure is used to identify the modulation set

[0068] Each modulation type in the set is scored, and the final score is Score = {4, 3, 5, 7, 10, 10, 10, 13, 3, 6, 8, 8, 7, 8}, where the order of the score values ​​corresponds to the order of the modulation types. The maximum score max{Score} = 13 corresponds to 8PSK in the modulation type set. At this time, there is only one maximum value, so the input data modulation recognition is successful, and the recognition type is 8PSK. Further calculation of the confidence is obtained.

Claims

1. A modulation recognition method based on a scoring system, characterized in that The steps include: (1) Let the input raw sample data be represented as a(n), where n = 1, 2, 3, ..., and its sampling rate be represented as f s Calculate the power spectrum of the original sampled data, and calculate the N-dB bandwidth BW of the input signal and the signal center frequency f c ; (2) Based on the signal center frequency f calculated in the previous step c , perform spectrum shifting on the input sample data, and move the zero intermediate frequency position of the data to f c According to the signal bandwidth BW calculated in the previous step, the zero-IF data is resampled by 8 times bandwidth oversampling, and the resampling rate is set to fs new ,but In the formula For rounding up calculation, the resampled data is recorded as {x(m)|m=1,2,3,…}; (3) {x(m)|m=1,2,3,…} is processed frame by frame with a certain frame length. The frame length can be expressed as frame_len. Suppose the i-th frame is represented by {x i (j)|j=1,2,3,…,32768}, where i=1,2,3,…, and frame_len=32768, and then normalize each frame of data; (4) Assume that the normalized data is expressed as The five instantaneous characteristic parameter values ​​of the signal {γ max ,σ a ,σ f ,R,K 42 }; (5) Recalculate the signal The high-order cumulant characteristic parameters {C 42 }; (6) Calculate the signal again The seven spectral characteristic parameters {S1,S a ,S b ,S2,S3,S4,S8}; (7) Based on the known modulation type decision set P, traverse each characteristic parameter calculated above, and add 1 point by default for the modulation type condition score that meets the characteristic parameter range, and no point is added for the modulation type condition score that does not meet the characteristic parameter range. If a certain characteristic parameter is strongly correlated with a known modulation mode, 2 points can be added if the condition is met; (8) Assume that the size of the known modulation type decision set is T, then traverse the i-th frame data and calculate all the feature parameters Features i ={γ i,max ,σ i,a ,σ i,f ,R i ,K i,42 ,C i,42 ,S i,1 ,S i,a ,S i,b ,S i,2 ,S i,3 ,S i,4 ,S i,8 }, score each modulation mode to be determined in turn, and output the score vector That is, the size of the feature parameter set corresponding to a certain modulation type is F=size(Features)=13; (9) Then the score vector Sorting, the modulation type corresponding to the highest score is the suspected debugging type. When there is only one highest score, the modulation recognition is successful and the confidence level is given; Otherwise, modulation recognition fails, and the suspected modulation modes and corresponding confidence levels are given in parallel.

2. The modulation recognition method based on the scoring system according to claim 1 is characterized in that In step (1), the N-dB bandwidth of the original sampled data is calculated by first dividing the original data into multiple frames, and the interval between two consecutive frames is The overlap of frames; then calculate the spectrum for each frame of data, and then superimpose and average the continuous multiple frame spectra to obtain the spectrum mean sequence. By default, 64 frames of spectrum are used for averaging; then find the maximum value of the spectrum mean, and set the maximum value as P max , then the corresponding threshold value threshold=P max -N, the default N is 10dB, so threshold = P max -10dB; further traverse the spectrum mean sequence, find and calculate the horizontal axis length corresponding to the spectrum mean value higher than the threshold threshold, which is the N-dB bandwidth; and the center point of the horizontal axis is the center frequency position of the signal, and the frequency value is recorded as f c .

3. The modulation recognition method based on the scoring system according to claim 1 is characterized in that In step (3), each i-frame data is normalized, and the calculation formula is:

4. The modulation recognition method based on the scoring system according to claim 1 is characterized in that The five instantaneous modulation parameters {γ max ,σ a ,σ f ,R,K 42 The calculation formulas are as follows: The maximum value of the zero-centered normalized instantaneous amplitude spectral density Zero-centered normalized instantaneous amplitude absolute value standard deviation Zero-centered normalized instantaneous frequency absolute value standard deviation Envelope variation coefficient Envelope Kurtosis in 5. The modulation recognition method based on the scoring system according to claim 1 is characterized in that In step (5), the high-order cumulative characteristic parameters are calculated as follows: in Where conj(·) represents the conjugate calculation of complex numbers.

6. The modulation recognition method based on the scoring system according to claim 1 is characterized in that In step (6), the seven spectral characteristic parameters {S1,S a ,S b , S2, S3, S4, S8} are calculated as follows, where the default peak decision threshold is 10dB: Primary spectrum The number of peaks above the threshold S1; Envelope spectrum The number of peaks above the threshold S a ; Delay spectrum The number of peaks above the threshold S b ; Secondary spectrum The number of peaks above the threshold is S2; Cubic Spectrum The number of peaks above the threshold is S3; Quartic spectrum The number of peaks above the threshold is S4; Octave The number of peaks above the threshold S 8。 7. The modulation recognition method based on the scoring system according to claim 1 is characterized in that The known modulation identification decision set P in step (7) is And the corresponding set size T=size(P)=14.

8. The modulation recognition method based on the scoring system according to claim 1 is characterized in that The confidence calculation formula in step (9) is defined as Wherein, T=size(P)=14 is the size of the known decision set, and F=size(Features)=13 is the size of all the feature attribute sets mentioned above.