A large bandwidth signal classification method based on multi-dimensional variable matrix operation

Abstract

The purpose of the present invention is to provide a large-bandwidth signal classification method based on multi-dimensional variable matrix operations, which can solve the problem of limited recognition and classification efficiency in accurately and quickly detecting and identifying multiple target signals in a large bandwidth under traditional large-bandwidth spectrum monitoring. The method has the following characteristics: through channelized segmentation and efficient matrix structure operations, a hierarchical filtering method is adopted to improve the efficiency of the narrowband channelization function of large-bandwidth signals; by constructing a variable-length feature matrix template and adopting a halving and fine search strategy, the efficiency of parallel feature sequence template matching of thousands of narrowband signals is improved; for different practical application scenarios, the feature matrix dimension can be dynamically adjusted to achieve dynamic expansion of the matching length and type of the template to meet the requirements of different data processing time and bandwidth. The method has strong flexibility, low complexity, high target signal recognition and classification performance, and is particularly suitable for high-speed parallel signal recognition and classification systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital signal processing, and in particular relates to a large-bandwidth signal classification method based on multi-dimensional variable matrix operations. Background Art

[0002] The airborne radio electromagnetic signal environment is dense, including signals of various frequency bands such as shortwave, ultrashortwave, and microwave. The communication frequency band is wide and the electromagnetic environment is complex. Various signals such as broadcast signals, civil signals, and industrial noise are mixed together. The target signal is hidden in the crowded and complex electromagnetic environment. In the actual electromagnetic environment, the ability to directly and quickly detect and identify the target radio signal in a wide bandwidth environment is of great significance for subsequent signal processing.

[0003] The mainstream method of traditional large-bandwidth signal detection and identification is to search for large-bandwidth spectrum energy and then analyze the signals one by one; or to use dynamic thresholds, signal background iteration and other improved methods as the threshold for detecting signals, especially for weak signals, to improve the probability of signal detection. However, these methods all have the problem of first detecting the presence and location of the signal under broadband conditions, and then filtering and analyzing the signal at that location before completing signal identification and classification. These serial processing mechanisms all have the same defect: the processing bandwidth is relatively narrow, making it difficult to directly identify and classify signals under broadband conditions. In terms of processing mechanism and speed, it is difficult to meet the requirements for identifying target signals in large-bandwidth complex electromagnetic environments, and it is easy to produce false alarms and missed alarms.

[0004] To sum up, the signal processing system under large communication bandwidth needs to be able to process a large amount of data in a very short time. The existing technology cannot meet the requirements of accurate and rapid detection and identification of target signals under large communication bandwidth. It is necessary to design a high-speed parallel computing algorithm to put forward higher requirements on the accuracy and real-time performance of high-speed signal processing design to meet the requirements of real-time and rapid detection and identification of target signals under large communication bandwidth. Summary of the Invention

[0005] The purpose of the present invention is to provide a large bandwidth signal classification method based on multi-dimensional variable matrix operation, which can solve the problem of limited recognition speed in the traditional large bandwidth signal classification and recognition method for accurately and quickly detecting and recognizing multiple target signals.

[0006] In a first aspect, an embodiment of the present invention discloses a large-bandwidth signal classification method based on multi-dimensional variable matrix operations, comprising:

[0007] S1, obtain the discrete signal to be classified, the sampling rate Fs, the total number of filter channels D and the number of two-stage filter channels; the two-stage filter channels include the number of first-stage filter channels NUM1 and the number of second-stage filter channels NUM2; the length of the discrete signal to be classified is Len;

[0008] S2, obtaining a feature template matrix; the dimension of the feature template matrix is M×N, N represents the category of the feature signal, and M represents the length of the feature signal; the column vector of the feature template matrix is a feature signal;

[0009] S3, classifying the discrete signal to be classified based on the feature template matrix to obtain a category value of the discrete signal to be classified;

[0010] The classifying process of the discrete signal to be classified based on the feature template matrix to obtain the category value of the discrete signal to be classified includes:

[0011] S31, based on the sampling rate Fs, the total number of filter channels D and the number of two-stage filter channels, a first-stage filter matrix and a second-stage filter matrix are designed;

[0012] S32, using a first-level filter matrix to process the discrete signal to be classified to obtain a first filter output matrix;

[0013] S33, using a secondary filter matrix to process the first filter output matrix to obtain a two-dimensional data matrix;

[0014] S34, performing matching calculation processing on the two-dimensional data matrix based on the feature template matrix to obtain a matching quantity set;

[0015] S35 , performing category extraction processing on the matching amount set to obtain a category value of the discrete signal to be classified.

[0016] The first-stage filter matrix and the second-stage filter matrix are designed based on the sampling rate Fs, the total number of filter channels D, and the number of two-stage filter channels, including:

[0017] S311, constructing a first-level filter matrix according to the sampling rate Fs and the number of first-level filter channels NUM1;

[0018] S312: construct a secondary filter matrix according to the sampling rate Fs and the number of secondary filter channels NUM2.

[0019] The first-level filter matrix is constructed according to the sampling rate Fs and the number of first-level filter channels NUM1, including:

[0020] The effective bandwidth corresponding to the sampling rate Fs is evenly divided into NUM1 adjacent sub-bands;

[0021] Set the first-level filter order constant len_filter1 to len_filter1=2D;

[0022] For the frequency band range of each sub-band, use the digital filter design method to design a filter vector with an order of len_filter1 / NUM1 for each sub-band;

[0023] The first-level filter matrix is constructed using the filter vectors of all sub-bands.

[0024] The secondary filter matrix is constructed based on the sampling rate Fs and the number of secondary filter channels NUM2, including:

[0025] The effective bandwidth corresponding to the sampling rate Fs is evenly divided into NUM2 adjacent sub-bands;

[0026] Set the secondary filter order constant len_filter2 to len_filter2=4D;

[0027] For the frequency band range of each sub-band, use the digital filter design method to design a filter vector with an order of len_filter2 / NUM2 for each sub-band;

[0028] The filter vectors of all sub-bands are used to construct a two-level filter matrix.

[0029] The method of processing the discrete signal to be classified by using a first-level filter matrix to obtain a first filter output matrix includes:

[0030] Representing the discrete signal to be classified as a digital matrix matching the dimension of the first filter output matrix;

[0031] Using a first-level filter matrix, parallel filtering is performed on the column vectors of the digital matrix to obtain a first output matrix;

[0032] Perform FFT processing on each column vector of the first output matrix to obtain a first filtered output matrix.

[0033] The method of processing the first filter output matrix using a secondary filter matrix to obtain a two-dimensional data matrix includes:

[0034] Using a two-stage filter matrix, parallel filtering is performed on the column vectors of the first filter output matrix to obtain a second output matrix;

[0035] Perform FFT processing on each column vector of the second output matrix to obtain a two-dimensional data matrix.

[0036] The matching calculation process is performed on the two-dimensional data matrix based on the feature template matrix to obtain a matching quantity set, including:

[0037] S341, performing dimension expansion processing on the two-dimensional data matrix and the feature template matrix respectively to obtain corresponding four-dimensional data matrix and four-dimensional feature template matrix;

[0038] S342, performing a first matching calculation process on the four-dimensional data matrix and the four-dimensional feature template matrix to obtain a first matching four-dimensional matrix;

[0039] S343, performing a second matching calculation process on the four-dimensional data matrix and the four-dimensional feature template matrix to obtain a second matching four-dimensional matrix;

[0040] S344: Construct a matching quantity set using the first matching four-dimensional matrix and the second matching four-dimensional matrix.

[0041] The expression for the first matching calculation process is:

[0042] c=|(FFT(a*conj(b))) 2

[0043] Where c is the first matching four-dimensional matrix, conj is the conjugate calculation, a is the four-dimensional data matrix, and b is the four-dimensional feature template matrix;

[0044] The expression for the second matching calculation process is:

[0045]

[0046] Among them, S a is the normalized four-dimensional data matrix, S b is the normalized four-dimensional feature template matrix, S ab is the cross-correlation four-dimensional matrix, and coff is the first matching four-dimensional matrix.

[0047] The performing category extraction processing on the matching amount set to obtain the category value of the discrete signal to be classified includes:

[0048] For each value of the third dimension P1 of the first matching four-dimensional matrix, obtain a corresponding three-dimensional matrix, count the total number of elements in the three-dimensional matrix that are greater than a preset first discrimination value, and use the total number as the first matching value corresponding to the value of the third dimension P1;

[0049] Constructing a first matching vector using first matching values corresponding to all values of the third dimension P1 of the first matching four-dimensional matrix;

[0050] For each value of the third dimension P2 of the second matching four-dimensional matrix, obtain a corresponding three-dimensional matrix, count the total number of elements in the three-dimensional matrix that are greater than a preset second discrimination value, and use the total number as the second matching value corresponding to the value of the third dimension P2;

[0051] Constructing a corresponding second matching vector using the second matching values corresponding to all values of the third dimension P2 of the second matching four-dimensional matrix;

[0052] The first matching vector and the second matching vector are jointly calculated to obtain a category value of the discrete signal to be classified.

[0053] The beneficial effects of the present invention are:

[0054] The present invention utilizes a hierarchical channelized efficient matrix parallel operation method to realize the segmentation function of large-bandwidth signals, wherein the hierarchical channelization can effectively reduce the demand for high-order filters and improve the subsequent calculation efficiency; constructs multiple groups of characteristic matrix templates and an efficient matrix matching method to realize the characteristic sequence template matching of thousands of narrowband signals in parallel; the characteristic matrix template can be dynamically adjusted according to demand to improve flexibility and realize efficient matching function, and adopts a halving and fine search strategy to improve calculation efficiency; the matrix dimension is variable, and for different actual application scenarios, the signal data matrix dimension is adjusted to meet the requirements of different signal sampling time lengths and bandwidths. At the same time, the characteristic template matrix can also be dynamically added and changed to realize the dynamic expansion of the matching length and type of the template.

[0055] In order to solve the problem of multiple signal classification and identification under broadband of traditional serial processing system, the present invention directly eliminates the signal search process, performs channelization segmentation on large-bandwidth signals, divides the large-bandwidth signals into thousands of narrowband signals, and performs target signal identification and classification on thousands of narrowband signals in parallel. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Schematic diagram of the even arrangement of complex signal channels according to the present invention;

[0057] Figure 2 Schematic diagram of the complex signal channelized receiving structure of the present invention;

[0058] Figure 3 Schematic diagram of matrix parallel batch processing model;

[0059] Figure 4 It is a feature sequence template, efficient matrix data organization and operation structure;

[0060] Figure 5 This is a flow chart of a large bandwidth signal classification method based on multi-dimensional variable matrix operations proposed by the present invention;

[0061] Figure 6 This is the real frequency domain diagram of the large bandwidth signal;

[0062] Figure 7 This is the DDC spectrum diagram after the large bandwidth signal is segmented;

[0063] Figure 8 This is the actual classification effect diagram of large bandwidth signals. DETAILED DESCRIPTION

[0064] In order to better understand the content of the present invention, an embodiment is given here.

[0065] Figure 1 Schematic diagram of the even arrangement of complex signal channels according to the present invention; Figure 2 Schematic diagram of the complex signal channelized receiving structure of the present invention; Figure 3 Schematic diagram of matrix parallel batch processing model; Figure 4 It is a feature sequence template, efficient matrix data organization and operation structure; Figure 5 This is a flow chart of a large bandwidth signal classification method based on multi-dimensional variable matrix operations proposed by the present invention; Figure 6 This is the real frequency domain diagram of the large bandwidth signal; Figure 7 This is the DDC spectrum diagram after the large bandwidth signal is segmented; Figure 8 This is the actual classification effect diagram of large bandwidth signals.

[0066] In the first aspect of the embodiments of the present invention, a large-bandwidth signal classification method based on multi-dimensional variable matrix operations is disclosed, which eliminates the signal search process and directly uses an efficient matrix operation structure to perform multiple narrowband DDC divisions on large-bandwidth channels. It also implements a real-time parallel processing mechanism for full-spectrum signals, divides large-bandwidth signals into thousands of narrowband channels, and performs real-time continuous acquisition, division, and parallel analysis and processing to ensure signal continuity without leakage.

[0067] In a first aspect of the embodiments of the present application, a method for classifying large-bandwidth signals based on multi-dimensional variable matrix operations is disclosed, comprising:

[0068] S1, obtain the discrete signal to be classified, the sampling rate Fs, the total number of filter channels D and the number of two-stage filter channels; the two-stage filter channels include the number of first-stage filter channels NUM1 and the number of second-stage filter channels NUM2; the length of the discrete signal to be classified is Len;

[0069] S2, obtaining a feature template matrix; the dimension of the feature template matrix is M×N, N represents the category of the feature signal, and M represents the length of the feature signal; the column vector of the feature template matrix is a feature signal;

[0070] S3, classifying the discrete signal to be classified based on the feature template matrix to obtain a category value of the discrete signal to be classified;

[0071] The classifying process of the discrete signal to be classified based on the feature template matrix to obtain the category value of the discrete signal to be classified includes:

[0072] S31, based on the sampling rate Fs, the total number of filter channels D and the number of two-stage filter channels, a first-stage filter matrix and a second-stage filter matrix are designed;

[0073] S32, using a first-level filter matrix to process the discrete signal to be classified to obtain a first filter output matrix;

[0074] S33, using a secondary filter matrix to process the first filter output matrix to obtain a two-dimensional data matrix;

[0075] S34, performing matching calculation processing on the two-dimensional data matrix based on the feature template matrix to obtain a matching quantity set;

[0076] S35, performing category extraction processing on the matching amount set to obtain the category value of the discrete signal to be classified;

[0077] The first-stage filter matrix and the second-stage filter matrix are designed based on the sampling rate Fs, the total number of filter channels D, and the number of two-stage filter channels, including:

[0078] S311, constructing a first-level filter matrix according to the sampling rate Fs and the number of first-level filter channels NUM1;

[0079] S312, constructing a secondary filter matrix according to the sampling rate Fs and the number of secondary filter channels NUM2;

[0080] The first-level filter matrix is constructed according to the sampling rate Fs and the number of first-level filter channels NUM1, including:

[0081] The effective bandwidth corresponding to the sampling rate Fs is evenly divided into NUM1 adjacent sub-bands;

[0082] Set the first-level filter order constant len_filter1 to len_filter1=2D;

[0083] For the frequency band range of each sub-band, use the digital filter design method to design a filter vector with an order of len_filter1 / NUM1 for each sub-band;

[0084] Using the filter vectors of all sub-bands, a first-level filter matrix is constructed;

[0085] The secondary filter matrix is constructed based on the sampling rate Fs and the number of secondary filter channels NUM2, including:

[0086] The effective bandwidth corresponding to the sampling rate Fs is evenly divided into NUM2 adjacent sub-bands;

[0087] Set the secondary filter order constant len_filter2 to len_filter2=4D;

[0088] For the frequency band range of each sub-band, use the digital filter design method to design a filter vector with an order of len_filter2 / NUM2 for each sub-band;

[0089] Using the filter vectors of all sub-bands, a secondary filter matrix is constructed;

[0090] The method of processing the discrete signal to be classified by using a first-level filter matrix to obtain a first filter output matrix includes:

[0091] Representing the discrete signal to be classified as a digital matrix matching the dimension of the first filter output matrix;

[0092] Using a first-level filter matrix, parallel filtering is performed on the column vectors of the digital matrix to obtain a first output matrix;

[0093] Performing FFT processing on each column vector of the first output matrix to obtain a first filtered output matrix;

[0094] The method of processing the first filter output matrix using a secondary filter matrix to obtain a two-dimensional data matrix includes:

[0095] Using a two-stage filter matrix, parallel filtering is performed on the column vectors of the first filter output matrix to obtain a second output matrix;

[0096] Performing FFT processing on each column vector of the second output matrix to obtain a two-dimensional data matrix;

[0097] The matching calculation process is performed on the two-dimensional data matrix based on the feature template matrix to obtain a matching quantity set, including:

[0098] S341, performing dimension expansion processing on the two-dimensional data matrix and the feature template matrix respectively to obtain corresponding four-dimensional data matrix and four-dimensional feature template matrix;

[0099] S342, performing a first matching calculation process on the four-dimensional data matrix and the four-dimensional feature template matrix to obtain a first matching four-dimensional matrix;

[0100] S343, performing a second matching calculation process on the four-dimensional data matrix and the four-dimensional feature template matrix to obtain a second matching four-dimensional matrix;

[0101] S344: Construct a matching quantity set using the first matching four-dimensional matrix and the second matching four-dimensional matrix.

[0102] The expression for the first matching calculation process is:

[0103] c=|(FFT(a*conj(b))) 2 ,

[0104] Where c is the first matching four-dimensional matrix, conj is the conjugate calculation, a is the four-dimensional data matrix, and b is the four-dimensional feature template matrix;

[0105] The expression for the second matching calculation process is:

[0106]

[0107] Among them, S a is the normalized four-dimensional data matrix, S b is the normalized four-dimensional feature template matrix, S ab is the cross-correlation four-dimensional matrix, and coff is the first matching four-dimensional matrix.

[0108] The step of performing dimension expansion processing on the two-dimensional data matrix and the feature template matrix to obtain corresponding four-dimensional data matrix and four-dimensional feature template matrix includes:

[0109] The two-dimensional data matrix and the feature template matrix are respectively copied in the expanded dimension, and the dimensions of the obtained four-dimensional matrices are D_valid×(Len / D)×N×M; the dimension of the two-dimensional data matrix is D_valid×(Len / D); the elements of the obtained four-dimensional data matrix are a(:,:,m,n)=A, and the elements of the four-dimensional feature template matrix are b(k,l,:,:)=B, k, l, m, n respectively represent the first dimension value, second dimension value, third dimension value and fourth dimension value of the four-dimensional matrix, A represents the two-dimensional data matrix, and B represents the feature template matrix; a(:,:,m,n)=A means that for any combination of m and n within their value range, the value of a(:,:,m,n) is A, and the meaning of b(k,l,:,:)=B is the same as above. In a*conj(b), conj(b) means taking the conjugate of each element in b. a*conj(b) means taking the conjugate of the elements in b with the same subscript in matrices a and b and multiplying them with the elements in a to obtain the corresponding elements with the same subscript in the new four-dimensional matrix. Let c1(k,l,m,n)=a(k,l,m,n)*conj(b(k,l,m,n)), where c1 is the four-dimensional matrix obtained by the a*conj(b) operation. The FFT(a*conj(b)) is to perform FFT calculations on the fourth-dimensional vectors c1(k,l,m,:) of the matrix c1 to obtain new fourth-dimensional vectors. After performing FFT calculations on all fourth-dimensional vectors of the matrix c1, the matrix c1 is updated to obtain the new four-dimensional matrix FFT(a*conj(b)). After taking the modulus value of each element of the four-dimensional matrix FFT(a*conj(b)), we can get the square value of |FFT(a*conj(b))| 2 , that is, the four-dimensional matrix c.

[0110] described The fourth dimension vectors of matrices a and b are averaged to obtain and Then the fourth dimension value n is accumulated to obtain the four-dimensional matrix S a and S b , specifically,

[0111] described The specific expression is:

[0112]

[0113] Among them, S ab is a four-dimensional matrix, S ab (k,l,m,1) is S ab Elements in .

[0114] described It is for S ab 、S a and S b Perform the above operation on each element in the four-dimensional matrix coff to obtain the corresponding element. coff(k,l,m,1) represents the corresponding element in the four-dimensional matrix coff.

[0115] The performing category extraction processing on the matching amount set to obtain the category value of the discrete signal to be classified includes:

[0116] For each value of the third dimension P1 of the first matching four-dimensional matrix, obtain a corresponding three-dimensional matrix, count the total number of elements in the three-dimensional matrix that are greater than a preset first discrimination value, and use the total number as the first matching value corresponding to the value of the third dimension P1;

[0117] Constructing a first matching vector using first matching values corresponding to all values of the third dimension P1 of the first matching four-dimensional matrix;

[0118] For each value of the third dimension P2 of the second matching four-dimensional matrix, obtain a corresponding three-dimensional matrix, count the total number of elements in the three-dimensional matrix that are greater than a preset second discrimination value, and use the total number as the second matching value corresponding to the value of the third dimension P2;

[0119] Constructing a corresponding second matching vector using the second matching values corresponding to all values of the third dimension P2 of the second matching four-dimensional matrix;

[0120] performing joint calculation processing on the first matching vector and the second matching vector to obtain a category value of the discrete signal to be classified;

[0121] The expression for the joint calculation process is:

[0122]

[0123] Where N is the length of the second matching vector, μ i is the i-th item of the first matching vector, v i is the i-th item of the second matching vector, and I is the category value of the discrete signal to be classified.

[0124] For each value of the third dimension P1 of the first matching four-dimensional matrix, a corresponding three-dimensional matrix is obtained. When the value of P1 is y1, the corresponding three-dimensional matrix obtained is c(:,:,y1,:); the situation for the second matching four-dimensional matrix is the same as above.

[0125] The method of using a first-level filter matrix to perform parallel filtering on the column vectors of the digital matrix to obtain a first output matrix includes:

[0126] Using each row vector of the first-level filter matrix, a convolution operation is performed with the column vector of the digital matrix with the same serial number to obtain the column vector of the corresponding serial number in the first output matrix;

[0127] The performing FFT processing on each column vector of the first output matrix to obtain the first filter output matrix is to perform FFT processing on each column vector of the first output matrix to obtain the corresponding column vector, and construct the first filter output matrix using all the column vectors;

[0128] The parallel filtering and FFT processing process using a two-stage filter matrix is the same as the above process.

[0129] For each value of the third dimension P1 of the first matching four-dimensional matrix, a corresponding three-dimensional matrix is obtained, which means that when the third dimension P1 takes values from 1 to N, the corresponding three-dimensional matrix AP2(:,:,P1,:) is obtained, where AP2 represents the first matching four-dimensional matrix, and the processing process of the second matching four-dimensional matrix is the same as the above process.

[0130] Using the filter vectors of all sub-frequency bands, a first-order or second-order filter matrix is constructed, wherein the filter vectors of the sub-frequency bands are used as column vectors to construct a first-order or second-order filter matrix;

[0131] The first discrimination value may be 18;

[0132] The first discrimination value may be 22;

[0133] The parallel filtering is to use the column vectors of the digital matrix as signals, filter each column vector with the column vector of the first or second filter matrix of the corresponding column number, obtain output column vectors, and use all the output column vectors to construct a first output matrix or a second output matrix;

[0134] The digital filter design method is a FIR digital filter design method.

[0135] The frequency band range of the i-th sub-band is [(i-1)×Fs / (2×NUM1), i×Fs / (2×NUM1)], or [(i-1)×Fs / (2×NUM2), i×Fs / (2×NUM2)];

[0136] The effective bandwidth is 1 / 2 of the sampling rate Fs;

[0137] The discrete signal to be classified is obtained by digitally sampling a signal received by a communication receiver;

[0138] In a second aspect of the present invention, a method for classifying a large bandwidth signal based on multi-dimensional variable matrix operations is disclosed. Figure 5 As shown, to implement this method, the following steps are taken:

[0139] Step 51: Determine the parameters of large bandwidth IQ sampling duration Len = 32768 × 1024, sampling rate Fs = 40.96 MHz, effective bandwidth BW = 28.5 MHz, number of DDC channels D = 1024, and number of channels for two-stage filtering: NUM1 = 16 and NUM1 = 64, respectively.

[0140] Step 52: Based on the channel segmentation parameters determined in step 51, design two-stage filters for NUM1 = 16 and NUM1 = 64, respectively. Based on the preset pass-stop band relationship, design a prototype low-pass filter to obtain filter coefficients and orders len_filter1 = 2048 and len_filter2 = 4096.

[0141] Step 53: Data organization is matrixed into 1×Len, channelized into matrix structure, after two-stage filtering, the output segmented DDC data is matrixed into 2D×(Len / D), and the effective bandwidth data is matrixed into D_valid×(Len / D).

[0142] In this embodiment, the input IQ length Len=32768×1024, D=1024, and the large bandwidth IQ data sampling rate is Fs=40.96MHz. Since the two-stage filtering method of the path delay is adopted, the data dimension can be expressed as Len=32768×(16*64). Figure 2 In the channelized efficient matrix data organization and operation processing structure, the first-level filtering implementation steps are as follows: first, when the signal NUM1 = 16, the signal is converted into a two-dimensional matrix using the Reshape function, and the matrix dimension is (Len / NUM1) × NUM1; secondly, the prototype low-pass filter coefficient is decomposed using the polyphase decomposition of the Reshape function, and the decomposed matrix is (Len_filter1 / NUM1) × NUM1; then the Filter function is used to implement the parallel filtering function between the signal dimension (Len / NUM1) × NUM1 and the filter dimension (Len_filter1 / NUM1) × NUM1, and the dimension after the parallel filtering is (Len / NUM1) × NUM1; finally, the FFT function is used to implement the parallel FFT of the (Len / NUM1) × NUM1 result, and the final result (Len / NUM1) × NUM1 is output.

[0143] The second-level filtering is analogous to the previous one. According to the effective bandwidth, the filtering results are organized into a valid data matrix as D_valid×(Len / D), where D_valid=1425, Len / D=32768, and the sampling rate of each channel after segmentation is Fs / D=40.96MHz / 1024=40KHz;

[0144] Step 54: Construct a feature template matrix, M×N, group templates of different lengths, and perform normalization on the feature templates;

[0145] In this embodiment, the lengths of the constructed templates are 128, 64, 32, etc., and the number of templates is also different. Three types of feature template matrices are constructed, namely 128×22, 64×4, and 32×8, and the matrices are normalized at the same time;

[0146] Step 55: Matrix calculation of the feature template matching algorithm. According to the data dimension and the feature template dimension, the matching correlation graph results are organized into a four-dimensional data structure D_valid×O×P×Q. A search strategy combining binary detection and fine detection is adopted to improve the operation efficiency.

[0147] In this embodiment, according to the feature template matching principle, Figure 4 As shown in Figure 1, the data matrix implementation structure of the template matching function after channelized segmentation of the signal is shown. It mainly includes the following parts: matching the input two-dimensional data matrix D_valid×(Len / D)=1024×32768 with three feature sequence templates (dimensions are 128×22, 64×4, and 32×8 respectively). The matrix matching formula is xconvfft=|(FFT(a*conj(b))) 2 , where a and b are the input data matrix and feature template matrix, in, The matching correlation graph results are organized into a four-dimensional data structure D_valid × O × P × Q, where D_valid = 1024: the number of partitioned channels, O: the number of data groups for a single match, which can be set arbitrarily based on actual conditions, P: the number of target templates to match, which can be 22, 4, or 8, and Q: the template length to match, which can be 128, 64, or 32. A search strategy combining binary detection and fine detection is employed to improve computational efficiency. In this embodiment, the binary length is set to 64.

[0148] Step 56: Do you need to change the bandwidth and sampling time?

[0149] In this embodiment, for different practical application scenarios, if it is necessary to adjust the input signal sampling duration and bandwidth requirements, step 51 is executed to readjust and organize the data matrix as required.

[0150] Step 57: Do you want to add or change the feature template matrix?

[0151] In this embodiment, for different actual application scenarios, if it is necessary to dynamically add and change the characteristic template matrix to achieve dynamic expansion of the matching length and type of templates, step 54 is executed to modify the corresponding parameters and reorganize the characteristic template matrix.

[0152] The CPU+GPU heterogeneous parallel processing platform used in this embodiment can match the characteristic sequence of 60 signals in 800ms for 3000 channels divided under 28.5MHz bandwidth. Figure 6 As shown, the actual classification effect of the signal under large bandwidth is as follows Figure 7 The running time is shown in Table 1.

[0153] Table 1 Test results of the algorithm running speed of this embodiment

[0154]

[0155] This invention develops an efficient classification and recognition method suitable for high-bandwidth, high-speed parallel computing platforms. Its efficient matrix operation structure is suitable for data matrixing, channelized operation matrixing, parameter extraction, and matching matrixing. This method is suitable for software developed on parallel platforms and offers the advantages of high flexibility and excellent performance.

[0156] In a third aspect, the present invention discloses a method for classifying large-bandwidth signals based on multi-dimensional variable matrix operations. This method eliminates the signal search process and directly uses an efficient matrix operation structure to perform multiple narrowband DDC segmentation on large-bandwidth channels. A real-time parallel processing mechanism is used for full-spectrum signals. The method divides large-bandwidth signals into thousands of narrowband channels, performs real-time continuous acquisition, segmentation, and parallel analysis and processing to ensure signal continuity without leakage. The method comprises the following steps:

[0157] S1) Constructing an efficient matrix parallel computing structure to implement hierarchical channelization, performing channelization operations on wide-bandwidth signals, and dividing the wide-bandwidth signals into thousands of narrowband signals. The hierarchical channelization strategy can effectively reduce the demand for high-order filters and improve subsequent computational efficiency.

[0158] S2) Constructing variable-length feature matrix templates for different target signals, and performing feature sequence template matching of thousands of narrowband signals in parallel. The feature matrix templates can be dynamically adjusted according to demand to achieve efficient matching function;

[0159] S3) For different practical application scenarios, the signal data matrix dimensions are adjusted to meet the requirements of different signal sampling durations and bandwidths. At the same time, the feature template matrix can be dynamically added and changed to achieve dynamic expansion of the template matching length and type.

[0160] Furthermore, the step S1) includes:

[0161] S11) Input and collect large bandwidth IQ data, sampling rate is Fs, number of sample points is 1×Len, according to Figure 3 The figure shows the even channel arrangement of complex signal channelization. The principle is as follows: Assume that the frequency range of the normalized complex signal is 0 to 2π, and the center frequency of the Kth channel is ω k (k=0,1,...K-1) are: ω k =2πk / K, when the critical extraction K=D, according to the derivation process, the mathematical model of the complex signal channelized receiver based on the channelized structure is obtained as follows Figure 1 As shown in the derivation formula (1). Assume that the filter length is 1×len_filter, set len_filter = MD, M is an integer, and the prototype low-pass filter time domain impulse response h(n) is uniformly divided into D groups at a certain interval of M points. The input original sequence is S(n), where n = mD, S P (m) = S(mD-p), h P (m) = h(mD + p), then the output of the kth channel is:

[0162] y k (m) = DFT[S P (m)*h P (m)] k

[0163] The decimation filter for each channel is no longer a low-pass filter, but a polyphase branch of the filter. This reduces the computational complexity to 1 / D, significantly improving the real-time processing capability of this channelization technology. Furthermore, the FFT algorithm can further accelerate computational speed.

[0164] S12) Based on the algorithm implementation process of step S11), after analysis, it is converted into a channelized algorithm with an efficient matrix implementation structure, which mainly includes the following parts: D-path signal delay, filter coefficient downsampling, filtering, and DFT operation. The figure shows the processing process of channelized matrix data and operations. According to formula (3), the matrix acceleration operation library can be used to implement everything from data arrangement to processing. The multi-dimensional matrix operation function library Reshape, Filter, and FFT can be directly called to complete the channelization operation. For signals in the transition band after filtering, the present invention uses a secondary mixing method to cover the transition band.

[0165] S13) according to the channelized algorithm of step S12) an efficient matrix implementation structure is implemented, such as Figure 2The processing diagram of the channelization algorithm with efficient matrix data organization and operation is shown. The input wideband IQ data length is 1×Len, and the prototype low-pass filter length is 1×Len_filter. First, the D-path delay operation of the signal is performed, and the Reshape function is used to convert the signal into a two-dimensional matrix with a matrix dimension of (Len / D)×D; secondly, the prototype low-pass filter coefficients are decomposed using the polyphase decomposition of the Reshape function, and the decomposed matrix is (Len_filter / D)×D; then the Filter function is used to implement the parallel filtering function between the signal dimension (Len / D)×D and the filter dimension (Len_filter / D)×D, and the dimension after parallel filtering is (Len / D)×D; finally, the FFT function is used to implement the parallel FFT of the (Len / D)×D result, and the final result (Len / D)×D is output.

[0166] In particular, the secondary mixing processing method in step S12) can be used to obtain a signal covering the transition band of the filter.

[0167] S14) The input sampling length is 1×Len signal, and the number of hierarchical multi-phase filtering stages used is set to two. The two-stage filters need to be designed separately. If it is assumed that the number of channels divided at one time is D, then the number of channels divided at two levels must satisfy: D=NUM1×NUM2. The number of channels divided at the first level is NUM1, and the sampling rate of different channels after division is: Fs / NUM1, and the length is (1×Len) / NUM1; the number of channels divided at the second level is NUM2 based on the first level division, and the sampling rate of different channels after division is: Fs / (NUM1×NUM2), and the length is (1×Len) / (NUM1*NUM2).

[0168] S15) Based on the segmentation result of step S13), i.e., the two-dimensional matrix dimension is: 2D×(Len / D), and the large bandwidth effective bandwidth value BW is integrated, the useful signal channels are reselected and combined into a matrix dimension: D_valid×(Len / D).

[0169] Furthermore, the step S2) includes:

[0170] S21) Construct a feature template matrix. Group feature templates of different lengths and perform power normalization operations. The template dimension is: M×N, where M represents the feature template length and N represents the number of feature templates;

[0171] S22) According to the principle of feature template matching, it is converted into an efficient matrix implementation structure. It mainly includes the following parts. As shown in the figure, the input two-dimensional data matrix D_valid×(Len / D) is matched with the matching feature sequence template. The matrix matching formula is xconvfft=(FFT(a*conj(b))) 2 , where a and b are the input data matrix and feature template matrix, in, The matching correlation graph result is organized into a four-dimensional data structure D_valid×O×P×Q, where D_valid is the number of divided channels, O is the number of data groups for single matching, P is the number of target templates to be matched, i.e., N in S21), and Q is the length of the template to be matched, i.e., M in S21);

[0172] S23) constructing multiple sets of feature matrix templates. Thus, efficient matching function can be achieved;

[0173] In particular, the data organization matrix and operation matrix implementation in S22) utilize Figure 3 The matrix parallel batch processing model shown in the figure mainly contains multi-dimensional data types: thread, thread block and grid, which represent the number of threads or thread blocks in the x, y and z directions respectively. Multiple threads form a thread block, and multiple thread blocks form a grid. For example, the matrix batch kernel function is defined as: Kernel_name<<<gridDim,blockDim,SharedMemorySize,stream> >>(args...), blockDim and gridDim define the dimensions of the thread grid and block, and determine the total number of threads launched. Based on the kernel's resource allocation, corresponding computing units are generated. Reasonable allocation of computing units can effectively improve parallel processing speed. This embodiment divides tasks into parallel operations based on data and algorithm structures, organizes data into a multidimensional structure, and maps it to the corresponding grid, thread block, and number of threads for parallel operations. The corresponding signal processing algorithm is adjusted to a multidimensional data operation structure.

[0174] S24) In S22), combined Figure 4As shown, the data of the template matching function after the channelized segmentation signal is matrixed. Since the matching results are organized into a four-dimensional data structure D_valid×O×P×Q, O is the number of data groups for a single match. The optimization method used is: using the correlation between code element transmissions, a search method combining binary and fine detection is adopted, and the binary detection matching results are organized into a four-dimensional data structure D_valid×(O / 2)×P×Q. The amount of calculation is reduced by half. According to the results after the binary detection, not only will D_valid be reduced a lot, but the value of O will also be reduced, effectively improving the search efficiency while ensuring the search accuracy.

[0175] Furthermore, the step S3) includes:

[0176] S31) For different practical application scenarios, if it is necessary to adjust the input signal sampling time and bandwidth requirements, then execute step S1) to readjust and organize the data matrix as required, otherwise execute step S32):

[0177] S32) At the same time, if it is necessary to dynamically add and change the characteristic template matrix to achieve dynamic expansion of the matching length and type of the template, then step S2) is executed to modify the corresponding parameters, otherwise the step ends.

[0178] In summary, the large-bandwidth signal classification method based on multi-dimensional variable matrix operations of the present invention is adapted to the current high-speed parallel computing system and improves the recognition and classification efficiency of detecting and identifying specific target signals under large bandwidth.

[0179] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A large bandwidth signal classification method based on multi-dimensional variable matrix operation, characterized in that: include: S1, obtain the discrete signal to be classified, the sampling rate Fs, the total number of filter channels D and the number of two-stage filter channels; The number of two-stage filtering channels includes the number of primary filtering channels NUM1 and the number of secondary filtering channels NUM2; the length of the discrete signal to be classified is Len; S2, obtain the feature template matrix; The dimension of the feature template matrix is M×N, where N represents the category of the feature signal and M represents the length of the feature signal; The column vector of the feature template matrix is a feature signal; S3, classifying the discrete signal to be classified based on the feature template matrix to obtain a category value of the discrete signal to be classified; The classifying process of the discrete signal to be classified based on the feature template matrix to obtain the category value of the discrete signal to be classified includes: S31, based on the sampling rate Fs, the total number of filter channels D and the number of two-stage filter channels, a first-stage filter matrix and a second-stage filter matrix are designed; S32, using a first-level filter matrix to process the discrete signal to be classified to obtain a first filter output matrix; S33, using a secondary filter matrix to process the first filter output matrix to obtain a two-dimensional data matrix; S34, performing matching calculation processing on the two-dimensional data matrix based on the feature template matrix to obtain a matching quantity set; S35, performing category extraction processing on the matching amount set to obtain the category value of the discrete signal to be classified; The matching calculation process is performed on the two-dimensional data matrix based on the feature template matrix to obtain a matching quantity set, including: S341, performing dimension expansion processing on the two-dimensional data matrix and the feature template matrix respectively to obtain corresponding four-dimensional data matrix and four-dimensional feature template matrix; S342, performing a first matching calculation process on the four-dimensional data matrix and the four-dimensional feature template matrix to obtain a first matching four-dimensional matrix; S343, performing a second matching calculation process on the four-dimensional data matrix and the four-dimensional feature template matrix to obtain a second matching four-dimensional matrix; S344: construct a matching quantity set using the first matching four-dimensional matrix and the second matching four-dimensional matrix; The expression for the first matching calculation process is: c=|(FFT(a*conj(b)))| 2 , Where c is the first matching four-dimensional matrix, conj is the conjugate calculation, a is the four-dimensional data matrix, and b is the four-dimensional feature template matrix; The expression for the second matching calculation process is: Among them, S a is the normalized four-dimensional data matrix, S b is the normalized four-dimensional feature template matrix, S ab is the cross-correlation four-dimensional matrix, and coff is the first matching four-dimensional matrix.

2. The large bandwidth signal classification method based on multi-dimensional variable matrix operation according to claim 1, characterized in that: The first-stage filter matrix and the second-stage filter matrix are designed based on the sampling rate Fs, the total number of filter channels D, and the number of two-stage filter channels, including: S311, constructing a first-level filter matrix according to the sampling rate Fs and the number of first-level filter channels NUM1; S312: construct a secondary filter matrix according to the sampling rate Fs and the number of secondary filter channels NUM2.

3. The large bandwidth signal classification method based on multi-dimensional variable matrix operation according to claim 2, characterized in that: The first-level filter matrix is constructed according to the sampling rate Fs and the number of first-level filter channels NUM1, including: The effective bandwidth corresponding to the sampling rate Fs is evenly divided into NUM1 adjacent sub-bands; Set the first-level filter order constant len_filter1 to len_filter1=2D; For the frequency band range of each sub-band, use the digital filter design method to design a filter vector with an order of len_filter1 / NUM1 for each sub-band; The first-level filter matrix is constructed using the filter vectors of all sub-bands.

4. The large bandwidth signal classification method based on multi-dimensional variable matrix operation according to claim 3, characterized in that: The secondary filter matrix is constructed based on the sampling rate Fs and the number of secondary filter channels NUM2, including: The effective bandwidth corresponding to the sampling rate Fs is evenly divided into NUM2 adjacent sub-bands; Set the secondary filter order constant len_filter2 to len_filter2=4D; For the frequency band range of each sub-band, use the digital filter design method to design a filter vector with an order of len_filter2 / NUM2 for each sub-band; The filter vectors of all sub-bands are used to construct a two-level filter matrix.

5. The large bandwidth signal classification method based on multi-dimensional variable matrix operation according to claim 4, characterized in that: The method of processing the discrete signal to be classified by using a first-level filter matrix to obtain a first filter output matrix includes: Representing the discrete signal to be classified as a digital matrix matching the dimension of the first filter output matrix; Using a first-level filter matrix, parallel filtering is performed on the column vectors of the digital matrix to obtain a first output matrix; Perform FFT processing on each column vector of the first output matrix to obtain a first filtered output matrix.

6. The large bandwidth signal classification method based on multi-dimensional variable matrix operation according to claim 4, characterized in that: The method of processing the first filter output matrix using a secondary filter matrix to obtain a two-dimensional data matrix includes: Using a two-stage filter matrix, parallel filtering is performed on the column vectors of the first filter output matrix to obtain a second output matrix; Perform FFT processing on each column vector of the second output matrix to obtain a two-dimensional data matrix.

7. The large bandwidth signal classification method based on multi-dimensional variable matrix operation according to claim 1, characterized in that: The performing category extraction processing on the matching amount set to obtain the category value of the discrete signal to be classified includes: For each value of the third dimension P1 of the first matching four-dimensional matrix, obtain a corresponding three-dimensional matrix, count the total number of elements in the three-dimensional matrix that are greater than a preset first discrimination value, and use the total number as the first matching value corresponding to the value of the third dimension P1; Constructing a first matching vector using first matching values corresponding to all values of the third dimension P1 of the first matching four-dimensional matrix; For each value of the third dimension P2 of the second matching four-dimensional matrix, obtain a corresponding three-dimensional matrix, count the total number of elements in the three-dimensional matrix that are greater than a preset second discrimination value, and use the total number as the second matching value corresponding to the value of the third dimension P2; Constructing a corresponding second matching vector using the second matching values corresponding to all values of the third dimension P2 of the second matching four-dimensional matrix; performing joint calculation processing on the first matching vector and the second matching vector to obtain a category value of the discrete signal to be classified; The expression for the joint calculation process is: Where N is the length of the second matching vector, μ i is the i-th item of the first matching vector, v i is the i-th item of the second matching vector, and I is the category value of the discrete signal to be classified.

Images