CN113114597B - Four-input signal separation method and system based on JADE algorithm - Google Patents
Four-input signal separation method and system based on JADE algorithm Download PDFInfo
- Publication number
- CN113114597B CN113114597B CN202110322268.7A CN202110322268A CN113114597B CN 113114597 B CN113114597 B CN 113114597B CN 202110322268 A CN202110322268 A CN 202110322268A CN 113114597 B CN113114597 B CN 113114597B
- Authority
- CN
- China
- Prior art keywords
- matrix
- dimensional
- unit
- rotation
- input signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03828—Arrangements for spectral shaping; Arrangements for providing signals with specified spectral properties
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Power Engineering (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a four-input signal separation method and a system based on a JADE algorithm, which adopt a pipeline design, have the idea of pulse array processing, and can realize the parallelized JADE algorithm, so that the real system can separate signals by linear aliasing of at most four paths of signals, and the system has the characteristics of high calculation speed, high precision and low hardware resource overhead.
Description
Technical Field
The invention belongs to the technical field of signal separation, and particularly relates to a four-input signal separation method and system based on a JADE algorithm.
Background
In the case where there is little information available for the source signal and the transmission channel, the source signal is extracted or recovered only by summing the observed mixed signals, a type of signal processing method known as blind signal processing. In the real-world situation, the blind separation has a strong engineering value when the respective separations are independent from each other, and the process is called independent component analysis. Independent component analysis under the assumption that source signal statistics are independent, an objective function is established, and an observation signal is decomposed into a plurality of independent components through an optimization algorithm. Independent component analysis is currently practiced in demanding areas, including biomedical, image processing, and communication systems.
Because each algorithm of independent component analysis involves a large amount of iteration, multiply-accumulate and matrix operation, the hardware implementation complexity of the algorithm is high. However, the field-editable gate array (FPGA) chip has extremely strong real-time performance and parallel processing capability, and the complex independent component analysis algorithm is realized through FPGA hardware, so that the requirement on data real-time processing in various fields can be met, the limitation of algorithm software can be eliminated, and the integrated design of a special chip is facilitated.
Disclosure of Invention
Aiming at the defects in the prior art, the four-input signal separation method and the four-input signal separation system based on the JADE algorithm solve the problems that the real-time performance is difficult to guarantee and is limited by algorithm software in the existing signal separation process.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a four-input signal separation method based on a JADE algorithm comprises the following steps:
s1, preprocessing the original mixed signal to be separated to obtain whitened data;
s2, calculating a corresponding four-dimensional cumulant matrix based on the whitened data;
s3, carrying out iterative computation on the four-dimensional cumulant matrix through a JADE algorithm to obtain a solution separation matrix;
and S4, performing matrix multiplication calculation on the solution separation matrix and the original mixed signal to be separated to obtain four paths of signals after solution separation.
Further, the step S1 is specifically:
sequentially carrying out mean value removing processing, covariance matrix calculation, eigenvalue decomposition, whitening matrix calculation and matrix multiplication calculation on the original mixed signal to be separated;
the formula of the mean value removing processing is as follows:
X′=X-E(X)
wherein, X' is the signal data after mean value removal, X is the input signal data, E (X) is the mean value of the input signal data;
the covariance matrix S obtained by calculation is:
wherein cov (-) is a covariance operation, E (-) is a mean operation function,subscript i is the sample point number of the signal component, x1i,x2iIs divided into a first signal component and a second signal componentQuantity, T is the number of signal sampling points;
the eigenvalue decomposition formula is:
D=UTSU=diag(λ1,λ2,…,λn)
wherein D is an eigenvalue matrix obtained by eigenvalue decomposition, U is an eigenvector matrix, and U is [ U ]1,u2,…,un],u1,u2,…,unIs the 1 st to n th eigenvector, diag (·) is a diagonal matrix, λ1,λ2,…,λnThe 1 st to n th characteristic values are marked with T as a transposition operator;
the whitening matrix W is:
W=D1/2*UT
whitened data Z obtained by matrix multiplication4*nComprises the following steps:
Z4*n=W4*4*X′4*n
in the formula, W4*4Is a whitening matrix, i.e. W, X'4*nIs the original mixed signal to be separated.
Further, the four-dimensional cumulative quantity matrix Q in the step S2z(M) is:
in the formula, Qz(M) is a four-dimensional cumulative metric matrix, Z is Z4*nFor whitened 4-channel signal data, z ═ z1,z2,z3,z4]TM is a set of basis matrices in a 4 x 4 dimensional linear space, K is a four dimensional joint accumulation quantity, KijklIs the four-dimensional cumulative amount of the i, j, k, l four components in z, mklFor the elements of k rows and l columns of the matrix M, the indices i, j are QzAnd (M) element labels, k and l are element labels of the matrix M, when the input signal is 4-way, the number of corresponding four-dimensional cumulative quantity matrixes is 10, and the size of each four-dimensional cumulative quantity matrix is 4 × 4.
Further, the step S3 is specifically:
s31, executing the iterative computation, and setting each three-wheel parallel rotation loop as one iterative computation;
s32, dividing the second-order primary-secondary matrixes corresponding to the current four-dimensional cumulative quantity matrix into two groups, and calculating the rotation angle required by the current round circulation for each group of second-order primary-secondary matrixes;
s33, Given rotation is carried out on elements in the current four-dimensional cumulative quantity matrix according to the calculated rotation angle, and the rotated results are stored in the four-dimensional cumulative quantity matrix again according to a set rule;
s34, based on the calculated rotation angle, carrying out parallel unilateral Given rotation on the elements in the current unmixed matrix, and storing the elements in the unmixed matrix again according to a set rule;
s35, repeating the steps S32-S34 until three cycles are completed;
s36, judging whether the current rotation angle is smaller than a set threshold value, and entering the step S37;
if yes, go to step S37;
if not, returning to the step S31;
and S37, satisfying the requirement of iterative computation, and obtaining the current unmixing matrix, namely the solution separation matrix.
Further, in step S32, determining a second-order prime/sub-type matrix corresponding to the 1 st, 2 nd row and 1 st, 2 nd column of the four-dimensional cumulative metric matrix, and a second-order prime/sub-type matrix corresponding to the 3 rd, 4 th row and 3 rd, 4 th column thereof, dividing the second-order prime/sub-type matrices of all the four-dimensional cumulative metric matrices into two groups, and calculating corresponding rotation angles respectively;
wherein the kth second-order principal component matrix of the same groupSubscript k 1, 2.., 10, ak,bk,ck,dkAre all elements in the kth second-order principal component matrix;
the rotation angle theta is as follows:
wherein, a, b, c and d are rotation angle calculation parameters obtained based on the grouped second-order principle type matrix, and b is equal to c, a, b, c and dWherein G ═ G1,g2,...,gk]Vector g ofk=[ak-bk,bk+ck]T;
In step S33, the method for Given rotation of the current four-dimensional cumulative metric matrix specifically includes:
and sequentially carrying out left rotation and right rotation on the 10 four-dimensional cumulative quantity matrixes based on the calculated rotation angle, wherein the rotation angles when the left rotation and the right rotation are carried out are respectively the rotation angles calculated based on the two groups of second-order primary-secondary matrixes.
A four-input signal separation system based on the JADE algorithm, comprising:
the data preprocessing module is used for preprocessing the original mixed signal to be separated to obtain whitened data and inputting the whitened data into the JADE algorithm module;
the JADE algorithm module is used for performing de-separation processing on the whitened data to obtain four paths of signals after de-separation;
the RAM memory is used for storing the original mixed signals to be separated;
the register is used for storing a de-mixing matrix required by de-separation processing;
and the logic control module is used for regulating and controlling the working state of each module and realizing logic control during mutual communication among the modules.
Further, the data preprocessing module comprises a mean value removing unit, a covariance matrix calculating unit, an eigenvalue decomposition unit, a whitening matrix calculating unit and a first matrix multiplication unit which are connected in sequence;
the output end of the RAM memory is connected with the input end of the first matrix multiplication unit;
the characteristic value decomposition unit is an iterative calculation unit which is based on a jacobi method, adopts a parallel pipeline technology and is designed by applying a parallel pulsation array structure;
the first matrix multiplication unit adopts a parallel pipeline design, and calculation results are sequentially output in parallel under a set clock beat.
Further, the JADE algorithm module comprises a four-dimensional cumulant matrix calculation unit, a JADE algorithm iteration unit and a second matrix multiplication unit which are connected in sequence;
the output end of the RAM unit is also connected with the input end of the second matrix multiplication unit;
and the output end of the register is connected with the JADE algorithm iteration unit.
Further, the JADE algorithm module is designed by adopting a parallel pipeline technology;
and the JADE algorithm iteration unit performs parallel processing on the four-dimensional cumulant matrix output by the first matrix multiplication unit and performs twice parallel Givens rotations.
The invention has the beneficial effects that:
(1) the signal separation system provided by the invention can realize the separation of input signals with the maximum thought, adopts a pipeline design, has a pulse array processing thought, and can realize a parallelized JADE algorithm, so that the realization framework can perform signal separation on linear aliasing of signals with the maximum four paths.
(2) The method has the characteristics of high calculation speed, high precision, low hardware resource overhead and strong portability, and can be used for various hardware platforms needing signal classification.
Drawings
Fig. 1 is a flow chart of a four-input signal separation method based on a JADE algorithm provided by the invention.
Fig. 2 is a schematic diagram illustrating a rule for storing the rotated four-dimensional cumulative metric matrix again according to the present invention.
Fig. 3 is a block diagram of a four-input signal separation system based on the JADE algorithm provided by the present invention.
Fig. 4 is a block diagram of a data preprocessing module-based architecture provided by the present invention.
Fig. 5 is a block diagram of a JADE algorithm module structure provided by the present invention.
Fig. 6 is a schematic diagram of a JADE algorithm iteration unit provided by the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
Example 1:
as shown in fig. 1, a four-input signal separation method based on the JADE algorithm includes the following steps:
s1, preprocessing the original mixed signal to be separated to obtain whitened data;
s2, calculating a corresponding four-dimensional cumulant matrix based on the whitened data;
s3, carrying out iterative computation on the four-dimensional cumulant matrix through a JADE algorithm to obtain a solution separation matrix;
and S4, performing matrix multiplication calculation on the solution separation matrix and the original mixed signal to be separated to obtain four paths of signals after solution separation.
Step S1 of this embodiment specifically includes:
sequentially carrying out mean value removing processing, covariance matrix calculation, eigenvalue decomposition, whitening matrix calculation and matrix multiplication calculation on the original mixed signal to be separated;
the formula of the mean value removing processing is as follows:
X′=X-E(X)
wherein, X' is the signal data after mean value removal, X is the input signal data, E (X) is the mean value of the input signal data;
the covariance matrix S obtained by calculation is:
wherein cov (-) is a covariance operation, E (-) is a mean operation function,subscript i is the sample point number of the signal component, x1i,x2iRespectively a first signal component and a second signal component, and T is the number of signal sampling points;
in this embodiment, for the feature decomposition process, a parallel systolic array design is adopted, and a Jacobi method idea is applied, so that the eigenvalue and the eigenvector of the input matrix can be calculated after a certain number of iterations, specifically, the eigenvalue decomposition formula is:
D=UTSU=diag(λ1,λ2,…,λn)
wherein D is an eigenvalue matrix obtained by eigenvalue decomposition, U is an eigenvector matrix, and U is [ U ]1,u2,…,un],u1,u2,…,unIs the 1 st to n th eigenvector, diag (·) is a diagonal matrix, λ1,λ2,…,λnThe 1 st to n th characteristic values are marked with T as a transposition operator;
the whitening matrix W is:
W=D1/2*UT
whitened data Z obtained by matrix multiplication4*nComprises the following steps:
Z4*n=W4*4*X′4*n
in the formula, W4*4Is a whitening matrix, i.e. W, X'4*nIs the original mixed signal to be separated.
The matrix multiplication process in the embodiment adopts a parallel pipeline design, the calculation results are sequentially and parallelly output after a plurality of clock beats and enter the next process,
the four-dimensional cumulative quantity matrix Q in step S2 of the present embodimentz(M) is:
in the formula, Qz(M) is a four-dimensional cumulative metric matrix, Z is Z4*nFor whitened 4-channel signal data, z ═ z1,z2,z3,z4]TM is a set of basis matrices in a 4 x 4 dimensional linear space, K is a four dimensional joint accumulation quantity, KijklIs the four-dimensional cumulative amount of the i, j, k, l four components in z, mklFor the elements of k rows and l columns of the matrix M, the indices i, j are QzAnd (M) element labels, k and l are element labels of the matrix M, when the input signal is 4-way, the number of corresponding four-dimensional cumulative quantity matrixes is 10, and the size of each four-dimensional cumulative quantity matrix is 4 × 4.
Step S3 of this embodiment specifically includes:
s31, executing the iterative computation, and setting each three-wheel parallel rotation loop as one iterative computation;
s32, dividing the second-order primary-secondary matrixes corresponding to the current four-dimensional cumulative quantity matrix into two groups, and calculating the rotation angle required by the current round circulation for each group of second-order primary-secondary matrixes;
s33, Given rotation is carried out on elements in the current four-dimensional cumulative quantity matrix according to the calculated rotation angle, and the rotated results are stored in the four-dimensional cumulative quantity matrix again according to a set rule in sequence to complete a cycle;
s34, based on the calculated rotation angle, carrying out parallel unilateral Given rotation on the elements in the current unmixed matrix, and storing the elements in the unmixed matrix again according to a set rule;
s35, repeating the steps S32-S34 until three cycles are completed;
s36, judging whether the current rotation angle is smaller than a set threshold value, and entering the step S37;
if yes, go to step S37;
if not, returning to the step S31;
and S37, satisfying the requirement of iterative computation, and obtaining the current unmixing matrix, namely the solution separation matrix.
In the step S32, determining the second-order primary-secondary matrix corresponding to the 1 st, 2 nd row, 1 st, 2 nd column of the four-dimensional cumulative metric matrix, and the second-order primary-secondary matrix corresponding to the 3 rd, 4 th row, 3 rd, 4 th column thereof, dividing all the second-order primary-secondary matrices of the four-dimensional cumulative metric matrix into two groups, and calculating the corresponding rotation angles respectively;
wherein the kth second-order principal component matrix of the same groupSubscript k 1, 2.., 10, ak,bk,ck,dkAre all elements in the kth second-order principal component matrix;
the rotation angle theta is:
wherein, a, b, c and d are rotation angle calculation parameters obtained based on the grouped second-order principle type matrix, and b is equal to c, a, b, c and dWherein G ═ G1,g2,...,gk]Vector g ofk=[ak-bk,bk+ck]T;
In step S33, the method for Given rotation of the current four-dimensional cumulative metric matrix specifically includes:
sequentially carrying out left rotation and right rotation on two sides of the 10 four-dimensional cumulative quantity matrixes based on the calculated rotation angle, wherein the rotation angles when the left rotation and the right rotation are carried out are respectively the rotation angles calculated based on the two groups of second-order primary-secondary matrixes;
specifically, for 10 four-dimensional cumulative quantity matrixes, the four-dimensional cumulative quantity matrixes rotate one at a time for 10 times in total, wherein the rotation is bilateral rotation, the rotation angle of the left hand is different from the rotation angle of the right hand, namely the rotation angles are two rotation angles obtained based on two different groups of second-order principal-type matrixes, and the sequence is left hand rotation and then right hand rotation; the above rotation formula is expressed as:
In step S33, after one sub-matrix is rotated, the result is stored in the four-dimensional cumulative quantity matrix again according to the set rule, and the next sub-matrix is rotated until all 10 sub-matrices are rotated, and one cycle is completed;
in step S33, the rule set during rotation is: for the column scheduling, column 1 is fixed, column 4 is changed to the position of column 2, and column 2 and column 3 are moved back one column in turn; for row scheduling, after the column scheduling is finished, the first row is not moved, the 4 th row is changed to the 2 nd row, and the 2 nd row and the 3 rd row are sequentially moved downward by one row. In this embodiment, taking a four-dimensional cumulative metric matrix as an example, the element movement rule of each matrix is the same, and the corresponding setting rule is shown in fig. 2.
In step S34, the setting rule for storing the rotated unmixing matrix is matched with the setting rule in step S33.
The threshold set in step S36 is set by a user, and can be changed according to the requirement, and the smaller the threshold is, the more the number of iterations of the whole process is.
After the step S36 returns to the step S31, the subsequent four-dimensional cumulative metric matrix is the corresponding four-dimensional cumulative metric matrix when the set threshold is not satisfied, and the four-dimensional cumulative metric matrix is not obtained in the step S2 at the beginning; and after the primary solution separation matrix is determined, the former solution mixing matrix is reset to be the unit matrix, and initialization preparation is made for next solution separation work of the system.
Example 2:
as shown in fig. 3, the four-input signal separation system based on the four-input signal separation method based on the JADE algorithm in embodiment 1 includes:
the data preprocessing module is used for preprocessing the original mixed signal to be separated to obtain whitened data and inputting the whitened data into the JADE algorithm module;
the JADE algorithm module is used for performing de-separation processing on the whitened data to obtain four paths of signals after de-separation;
the RAM memory is used for storing the original mixed signals to be separated;
the register is used for storing a de-mixing matrix required by de-separation processing;
and the logic control module is used for regulating and controlling the working state of each module and realizing logic control during mutual communication among the modules.
And for the logic control module which is used as the brain of the whole system, a plurality of control signals are arranged between each sub-module and the logic control module, the logic control module judges the current working stage according to the module working state fed back in the control signals, controls when the control signals change, instructs the corresponding module to work next, and controls when the sub-modules receive the control signals given by the logic control, so that the sub-modules change correspondingly.
Specifically, as shown in fig. 4, the data preprocessing module includes a mean value removing unit, a covariance matrix calculating unit, an eigenvalue decomposition unit, a whitening matrix calculating unit, and a first matrix multiplication unit, which are connected in sequence;
the output end of the RAM memory is connected with the input end of the first matrix multiplication unit;
the characteristic value decomposition unit is an iterative calculation unit which is based on a jacobi method, adopts a parallel pipeline technology and is designed by applying a parallel pulsation array structure;
the first matrix multiplication unit adopts a parallel pipeline design, and calculation results are sequentially output in parallel under a set clock beat.
In this embodiment, the input signal is sequentially subjected to the above-mentioned several units of calculation to obtain a whitening matrix, the data stored in the RAM memory and the whitening matrix are sent to the first matrix multiplication unit together, and the processed whitening data are obtained and sequentially output in parallel at a set clock beat. The data in the first matrix multiplication unit is input into a subsequent calculation network in a parallel mode, the whole network outputs results in a pipeline mode according to beats, each calculation unit independently calculates partial results according to the data received by an upstream neighbor unit, stores the results in an internal register temporarily and transmits the results to a downstream neighbor unit.
As shown in fig. 5, the JADE algorithm module includes a four-dimensional cumulant matrix calculation unit, a JADE algorithm iteration unit, and a second matrix multiplication unit, which are connected in sequence;
the output end of the RAM unit is also connected with the input end of the second matrix multiplication unit;
the output end of the register is connected with the JADE algorithm iteration unit.
As shown in fig. 6, a schematic diagram of the implementation of iterative computation by the JADE algorithm module in the present embodiment is shown.
Specifically, the JADE algorithm module in the embodiment is designed by adopting a parallel pipeline technology;
and the JADE algorithm iteration unit performs parallel processing on the four-dimensional cumulant matrix output by the first matrix multiplication unit and performs twice parallel Givens rotations.
Claims (5)
1. A four-input signal separation method based on a JADE algorithm is characterized by comprising the following steps:
s1, preprocessing the original mixed signal to be separated to obtain whitened data;
s2, calculating a corresponding four-dimensional cumulant matrix based on the whitened data;
s3, carrying out iterative computation on the four-dimensional cumulant matrix through a JADE algorithm to obtain a solution separation matrix;
s4, performing matrix multiplication calculation on the solution separation matrix and the original mixed signal to be separated to obtain four paths of signals after solution separation;
the step S1 specifically includes:
sequentially carrying out mean value removing processing, covariance matrix calculation, eigenvalue decomposition, whitening matrix calculation and matrix multiplication calculation on the original mixed signal to be separated;
the formula of the mean value removing processing is as follows:
X′=X-E(X)
wherein, X' is the signal data after mean value removal, X is the input signal data, E (X) is the mean value of the input signal data;
the covariance matrix S obtained by calculation is:
wherein cov (-) is a covariance operation, E (-) is a mean operation function,subscript i is the sample point number of the signal component, x1i,x2iRespectively a first signal component and a second signal component, and T is the number of signal sampling points;
the eigenvalue decomposition formula is:
D=UTSU=diag(λ1,λ2,…,λn)
wherein D is an eigenvalue matrix obtained by eigenvalue decomposition, U is an eigenvector matrix, and U is [ U ]1,u2,…,un],u1,u2,…,unIs the 1 st to n th eigenvector, diag (·) is a diagonal matrix, λ1,λ2,…,λnThe 1 st to n th characteristic values are marked with T as a transposition operator;
the whitening matrix W is:
W=D1/2*UT
whitened data Z obtained by matrix multiplication4*nComprises the following steps:
Z4*n=W4*4*X′4*n
in the formula, W4*4Is a whitening matrix, i.e. W, X'4*nIs the original mixed signal to be separated;
the four-dimensional cumulative quantity matrix Q in the step S2z(M) is:
in the formula, Qz(M) is a four-dimensional cumulative metric matrix, Z is Z4*nFor whitened 4-channel signal data, z ═ z1,z2,z3,z4]TM is a set of basis matrices in a 4 x 4 dimensional linear space, K is a four dimensional joint accumulation quantity, KijklIs the four-dimensional cumulative amount of the i, j, k, l four components in z, mklFor the elements of k rows and l columns of the matrix M, the indices i, j are QzThe element label of (M), k, l are the element labels of matrix M, when the input signal is 4 ways, the number of the corresponding four-dimensional accumulative metric matrix is 10, and the size of each four-dimensional accumulative metric matrix is 4 multiplied by 4;
the step S3 specifically includes:
s31, executing the iterative computation, and setting each three-wheel parallel rotation loop as one iterative computation;
s32, dividing the second-order primary-secondary matrixes corresponding to the current four-dimensional cumulative quantity matrix into two groups, and calculating the rotation angle required by the current round circulation for each group of second-order primary-secondary matrixes;
s33, Given rotation is carried out on elements in the current four-dimensional cumulative quantity matrix according to the calculated rotation angle, and the rotated results are stored in the four-dimensional cumulative quantity matrix again according to a set rule;
s34, based on the calculated rotation angle, carrying out parallel unilateral Given rotation on the elements in the current unmixed matrix, and storing the elements in the unmixed matrix again according to a set rule;
s35, repeating the steps S32-S34 until three cycles are completed;
s36, judging whether the current rotation angle is smaller than a set threshold value, and entering the step S37;
if yes, go to step S37;
if not, returning to the step S31;
s37, satisfying the iterative computation requirement, obtaining the current unmixing matrix, namely the solution separation matrix;
in step S32, determining a second-order primary-secondary matrix corresponding to the 1 st, 2 nd row, 1 st, 2 nd column of the four-dimensional cumulative metric matrix, and a second-order primary-secondary matrix corresponding to the 3 rd, 4 th row, 3 rd, 4 th column thereof, dividing all the second-order primary-secondary matrices of the four-dimensional cumulative metric matrix into two groups, and calculating corresponding rotation angles respectively;
wherein the kth second-order principal component matrix of the same groupSubscript k 1, 2.., 10, ak,bk,ck,dkAre all elements in the kth second-order principal component matrix;
the rotation angle theta is as follows:
wherein, a, b, c and d are rotation angle calculation parameters obtained based on the grouped second-order principle type matrix, and b is equal to c, a, b, c and dWherein G ═ G1,g2,...,gk]Vector g ofk=[ak-bk,bk+ck]T;
In step S33, the method for Given rotation of the current four-dimensional cumulative metric matrix specifically includes:
and sequentially carrying out left rotation and right rotation on the 10 four-dimensional cumulative quantity matrixes based on the calculated rotation angle, wherein the rotation angles when the left rotation and the right rotation are carried out are respectively the rotation angles calculated based on the two groups of second-order primary-secondary matrixes.
2. A four-input signal separation system based on the four-input signal separation method based on the JADE algorithm of claim 1, comprising:
the data preprocessing module is used for preprocessing the original mixed signal to be separated to obtain whitened data and inputting the whitened data into the JADE algorithm module;
the JADE algorithm module is used for performing de-separation processing on the whitened data to obtain four paths of signals after de-separation;
the RAM memory is used for storing the original mixed signals to be separated;
the register is used for storing a de-mixing matrix required by de-separation processing;
and the logic control module is used for regulating and controlling the working state of each module and realizing logic control during mutual communication among the modules.
3. The JADE algorithm-based four-input signal separation system of claim 2, wherein the data preprocessing module comprises a mean removing unit, a covariance matrix calculation unit, an eigenvalue decomposition unit, a whitening matrix calculation unit and a first matrix multiplication unit which are connected in sequence;
the output end of the RAM memory is connected with the input end of the first matrix multiplication unit;
the characteristic value decomposition unit is an iterative calculation unit which is based on a jacobi method, adopts a parallel pipeline technology and is designed by applying a parallel pulsation array structure;
the first matrix multiplication unit adopts a parallel pipeline design, and calculation results are sequentially output in parallel under a set clock beat.
4. The JADE algorithm-based four-input signal separation system according to claim 3, wherein the JADE algorithm module comprises a four-dimensional cumulative metric matrix calculation unit, a JADE algorithm iteration unit and a second matrix multiplication unit which are connected in sequence;
the output end of the RAM unit is also connected with the input end of the second matrix multiplication unit;
and the output end of the register is connected with the JADE algorithm iteration unit.
5. The JADE algorithm-based four-input signal separation method according to claim 4, wherein the JADE algorithm module is designed by adopting a parallel pipeline technology;
and the JADE algorithm iteration unit performs parallel processing on the four-dimensional cumulant matrix output by the first matrix multiplication unit and performs twice parallel Givens rotations.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110322268.7A CN113114597B (en) | 2021-03-25 | 2021-03-25 | Four-input signal separation method and system based on JADE algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110322268.7A CN113114597B (en) | 2021-03-25 | 2021-03-25 | Four-input signal separation method and system based on JADE algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113114597A CN113114597A (en) | 2021-07-13 |
CN113114597B true CN113114597B (en) | 2022-04-29 |
Family
ID=76712320
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110322268.7A Active CN113114597B (en) | 2021-03-25 | 2021-03-25 | Four-input signal separation method and system based on JADE algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113114597B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009083615A1 (en) * | 2008-01-03 | 2009-07-09 | Commissariat A L'energie Atomique | Method for separating mixed signals into a plurality of component signals |
CN103163505A (en) * | 2013-01-31 | 2013-06-19 | 西安电子科技大学 | Time-varying narrow-band interference suppression method based on joint approximate diagonalization of eigen-matrices (JADE) |
CN108282424A (en) * | 2018-01-26 | 2018-07-13 | 大连理工大学 | The tetradic Joint diagonalization algorithm of blind source separating is closed for four data set associatives |
CN111106866A (en) * | 2019-12-13 | 2020-05-05 | 南京理工大学 | Satellite-borne AIS/ADS-B collision signal separation method based on hessian matrix pre-estimation |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101086730A (en) * | 2007-07-13 | 2007-12-12 | 上海大学 | Convolution mixed blind separation frequency domain method based on non continuous smoothness |
CN102215048A (en) * | 2011-07-27 | 2011-10-12 | 中国人民解放军总参谋部第六十三研究所 | Receiving method and receiving device of spread spectrum signals |
-
2021
- 2021-03-25 CN CN202110322268.7A patent/CN113114597B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009083615A1 (en) * | 2008-01-03 | 2009-07-09 | Commissariat A L'energie Atomique | Method for separating mixed signals into a plurality of component signals |
CN103163505A (en) * | 2013-01-31 | 2013-06-19 | 西安电子科技大学 | Time-varying narrow-band interference suppression method based on joint approximate diagonalization of eigen-matrices (JADE) |
CN108282424A (en) * | 2018-01-26 | 2018-07-13 | 大连理工大学 | The tetradic Joint diagonalization algorithm of blind source separating is closed for four data set associatives |
CN111106866A (en) * | 2019-12-13 | 2020-05-05 | 南京理工大学 | Satellite-borne AIS/ADS-B collision signal separation method based on hessian matrix pre-estimation |
Non-Patent Citations (2)
Title |
---|
JADE 盲信号分离算法失效问题研究;刘伟群,阳春华,易叶青;《计算技术与自动化》;20050930;全文 * |
一种基于JADE与小波去噪的自适应盲分离方法;石文斌,王大鸣,仵国锋,崔维嘉;《信息工程大学学报》;20120229;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN113114597A (en) | 2021-07-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xing et al. | End-to-end learning for joint image demosaicing, denoising and super-resolution | |
CN108108809B (en) | Hardware architecture for reasoning and accelerating convolutional neural network and working method thereof | |
CN110263925B (en) | Hardware acceleration implementation device for convolutional neural network forward prediction based on FPGA | |
US20210357735A1 (en) | Split accumulator for convolutional neural network accelerator | |
JP7007488B2 (en) | Hardware-based pooling system and method | |
Oja | The nonlinear PCA learning rule and signal separation: Mathematical analysis | |
CN113034391B (en) | Multi-mode fusion underwater image enhancement method, system and application | |
CN111652360B (en) | Convolution operation device based on pulsation array | |
CN113822284B (en) | RGBD image semantic segmentation method based on boundary attention | |
CN108765282B (en) | Real-time super-resolution method and system based on FPGA | |
Miao et al. | Quaternion-based bilinear factor matrix norm minimization for color image inpainting | |
CN114663593B (en) | Three-dimensional human body posture estimation method, device, equipment and storage medium | |
CN115222581A (en) | Image generation method, model training method, related device and electronic equipment | |
CN114781629A (en) | Hardware accelerator of convolutional neural network based on parallel multiplexing and parallel multiplexing method | |
CN113114597B (en) | Four-input signal separation method and system based on JADE algorithm | |
CN111882028B (en) | Convolution operation device for convolution neural network | |
CN102624653B (en) | Extensible QR decomposition method based on pipeline working mode | |
CN109190755B (en) | Matrix conversion device and method for neural network | |
Li et al. | YOLOSA: Object detection based on 2D local feature superimposed self-attention | |
CN103177447A (en) | Hyperspectral image anomaly detection system based on FPGA (Field Programmable Gata Array) | |
CN115578325A (en) | Image anomaly detection method based on channel attention registration network | |
WO2011161202A2 (en) | A parallel multi-pipeline systolic array for complex singular value decomposition on a multi-processor device | |
CN108776994B (en) | Roesser model based on true three-dimensional display system and implementation method thereof | |
CN113657587A (en) | FPGA-based deformable convolution acceleration method and device | |
CN113486718A (en) | Fingertip detection method based on deep multitask learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |