Data processing method and device of filter, equipment, storage medium and filter

Abstract

The invention discloses a data processing method and device of a filter, equipment, a storage medium and the filter. The method comprises the following steps: in a block sparse system, determining a filter estimation value of a filter at the current moment; determining a loss function according to the filter estimated value at the current moment; and respectively solving partial derivatives of the filter estimation value at the next moment and the Lagrange multiplier in the Lagrange multiplier function by adopting a stochastic gradient method, and enabling a partial derivative function to be equal to zero so as to determine the optimal filter estimation value of the filter at the next moment. According to the method, a stochastic gradient method is adopted, partial derivatives of a filter estimation value and a Lagrange multiplier at the next moment in a Lagrange multiplier function are solved, the partial derivative function is made to be equal to zero, the optimal filter estimation value of the filter at the next moment is determined, the p norm parameter p used in the loss function approaches zero, and the optimal filter estimation value of the filter at the next moment is determined. In a scene of a block sparse system, the convergence speed of the filter can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of filters, and in particular to a filter data processing method, device, equipment, storage medium and filter. Background Art

[0002] In the related art, a system is considered sparse if its impulse response vector consists mainly of zero or near-zero coefficients, except for a few coefficients with significant values. Many systems are block-sparse systems, where non-zero coefficients are clustered in clusters in a structured manner, and filters converge slowly in such block-sparse systems. Summary of the invention

[0003] The present invention provides a filter data processing method, device, equipment, storage medium and filter, which can solve the technical problem of slow convergence speed in related technologies. The technical solution is as follows:

[0004] On the one hand, a filter data processing method is provided, the method comprising:

[0005] In a block sparse system, determining a filter estimate of the filter at a current time instant;

[0006] Determine a loss function according to the filter estimation value at the current moment;

[0007] Wherein, the p-norm parameter p used in the loss function approaches zero;

[0008] Determine a Lagrange multiplier function according to the loss function and the linear regression objective function;

[0009] The stochastic gradient method is used to respectively obtain partial derivatives of the filter estimation value and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and the partial derivative function is set equal to zero to determine the optimal filter estimation value of the filter at the next moment.

[0010] In one embodiment, determining the linear regression objective function includes:

[0011] In the block sparse system, determining the unknown impulse response vector, input vector and noise signal of the filter at the current moment;

[0012] A linear regression objective function is determined according to the unknown impulse response vector, the input vector and the noise signal at the current moment.

[0013] In one implementation, determining the loss function according to the filter estimation value at the current moment specifically includes:

[0014] Determine a filter estimate value difference between a filter estimate value at a next moment and a filter estimate value at a current moment;

[0015] Determine a mixed norm of the filter estimation value difference; the first auxiliary parameter of the mixed norm is 2, and the second auxiliary parameter is the inverse matrix of the step size matrix G(n);

[0016] Determine a first product of the inverse matrix of the step size matrix G(n) and the estimated value at the next moment;

[0017] determining a p-norm of the first product and a second product of the p-norm of the first product and a predetermined coefficient;

[0018] A sum of the mixed norm and the second product is determined as the loss function.

[0019] In one implementation, the stochastic gradient method is used to obtain the filter estimate at the next moment and the Lagrange multiplier in the Lagrange multiplier function respectively to obtain the filter estimate at the next moment, including:

[0020] In the Lagrange multiplier function, a partial derivative of the filter estimate at the next moment is calculated to obtain a first partial derivative calculation formula, and the first partial derivative calculation formula is set to zero;

[0021] In the Lagrange multiplier method function, a partial derivative of the Lagrange multiplier is obtained to obtain a second partial derivative calculation formula, and the second partial derivative calculation formula is set to be zero;

[0022] Determine a first calculation formula for a filter estimation value at a next moment according to the first partial derivative calculation formula being equal to zero;

[0023] In the first calculation formula of the filter estimation value at the next moment, the filter estimation value at the next moment is set equal to the filter estimation value at the current moment, and simplified to obtain the second calculation formula of the filter estimation value at the next moment;

[0024] According to the second partial derivative calculation formula being equal to zero, determining the calculation formula of the linear regression objective function;

[0025] Determine a calculation formula for the Lagrange multiplier according to the second calculation formula for the filter estimation value at the next moment and the constraint formula of the linear regression objective function;

[0026] Determine a third calculation formula for the filter estimation value at the next moment according to the calculation formula of the Lagrange multiplier and the first calculation formula for the filter estimation value at the next moment;

[0027] The filter estimation value at the next moment calculated by the third calculation formula is used as the filter estimation value at the next moment.

[0028] In one implementation, determining a Lagrange multiplier function according to the loss function and the linear regression objective function includes:

[0029] Multiplying the second product in the loss function by a Lagrange multiplier to obtain a first Lagrangian transition product;

[0030] Calculate a linear regression model difference between the linear regression model function and a reference standard value of the linear regression function;

[0031] Multiplying the linear regression model difference with the Lagrange multiplier to obtain a second Lagrange transition product;

[0032] The first Lagrangian transition product, the second Lagrangian transition product and the second-order norm of the filter estimation value difference in the loss function are summed to obtain a Lagrangian multiplier function.

[0033] In one embodiment, the step size matrix G(n) is a diagonal matrix, and each element in the diagonal matrix is ​​g i (n)1 M ;

[0034] G(n)=diag{g1(n)1 M ,…,g N (n)1 M};

[0035]

[0036] ρ is the guarantee update parameter;

[0037] q is the parameter to prevent freezing;

[0038] is the i-th block;

[0039] k i is the metric value of the i-th block;

[0040] k j is the metric value of the jth block;

[0041] g i (n) is the i-th element on the diagonal of the step size matrix;

[0042] 1M is a row vector of length M containing all 1s;

[0043] N is the number of blocks that are evenly divided in the block-sparse system.

[0044] On the other hand, a filter data processing device is provided, the device comprising:

[0045] A first determination module, used to determine a filter estimation value of the filter at a current moment in a block sparse system;

[0046] A second determination module, used to determine a loss function according to the filter estimation value at the current moment;

[0047] Wherein, the p-norm parameter p used in the loss function approaches zero;

[0048] A third determination module is used to determine a Lagrange multiplier function according to the loss function and the linear regression objective function;

[0049] The processing module is used to adopt the stochastic gradient method to respectively obtain partial derivatives of the filter estimation value and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and set the partial derivative function equal to zero to determine the optimal filter estimation value of the filter at the next moment.

[0050] In some embodiments, the third determination module is further used to determine the unknown impulse response vector, input vector and noise signal of the filter at the current moment in the block sparse system;

[0051] A linear regression objective function is determined according to the unknown impulse response vector, the input vector and the noise signal at the current moment.

[0052] In some embodiments, the second determination module is further used to determine a filter estimation value difference between a filter estimation value at a next moment and a filter estimation value at a current moment;

[0053] Determine a mixed norm of the filter estimation value difference; the first auxiliary parameter of the mixed norm is 2, and the second auxiliary parameter is the inverse matrix of the step size matrix G(n);

[0054] Determine a first product of the inverse matrix of the step size matrix G(n) and the estimated value at the next moment;

[0055] determining a p-norm of the first product and a second product of the p-norm of the first product and a predetermined coefficient;

[0056] A sum of the mixed norm and the second product is determined as the loss function.

[0057] In some embodiments, the processing module is further used to calculate the partial derivative of the filter estimate at the next moment in the Lagrange multiplier function to obtain a first partial derivative calculation formula, and set the first partial derivative calculation formula equal to zero.

[0058] In the Lagrange multiplier method function, a partial derivative of the Lagrange multiplier is obtained to obtain a second partial derivative calculation formula, and the second partial derivative calculation formula is set to be zero;

[0059] Determine a first calculation formula for a filter estimation value at a next moment according to the first partial derivative calculation formula being equal to zero;

[0060] In the first calculation formula of the filter estimation value at the next moment, the filter estimation value at the next moment is set equal to the filter estimation value at the current moment, and simplified to obtain the second calculation formula of the filter estimation value at the next moment;

[0061] According to the second partial derivative calculation formula being equal to zero, determining the calculation formula of the linear regression objective function;

[0062] Determine a calculation formula for the Lagrange multiplier according to the second calculation formula for the filter estimation value at the next moment and the constraint formula of the linear regression objective function;

[0063] Determine a third calculation formula for the filter estimation value at the next moment according to the calculation formula of the Lagrange multiplier and the first calculation formula for the filter estimation value at the next moment;

[0064] The filter estimation value at the next moment calculated by the third calculation formula is used as the filter estimation value at the next moment.

[0065] In some embodiments, the third determination module is further used to multiply the second product in the loss function by a Lagrangian multiplier to obtain a first Lagrangian transition product;

[0066] Calculate a linear regression model difference between the linear regression model function and a reference standard value of the linear regression function;

[0067] Multiplying the linear regression model difference with the Lagrange multiplier to obtain a second Lagrange transition product;

[0068] The first Lagrangian transition product, the second Lagrangian transition product and the second-order norm of the filter estimation value difference in the loss function are summed to obtain a Lagrangian multiplier function.

[0069] On the other hand, a computer device is provided, which includes a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the filter data processing method described above.

[0070] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored in the storage medium, and when the computer program is executed by a processor, the steps of the filter data processing method described above are implemented.

[0071] On the other hand, a filter is provided, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the filter data processing method described above are implemented.

[0072] The technical solution provided by the present invention can at least bring the following beneficial effects:

[0073] The stochastic gradient method is used to calculate the partial derivatives of the filter estimate and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and the partial derivative function is set equal to zero to determine the optimal filter estimate at the next moment of the filter. The p-norm parameter p used in the above loss function is set close to zero. In the scenario of block sparse systems, the convergence speed of the filter can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0075] Figure 1 This is a principle block diagram of an adaptive algorithm filter provided by an embodiment of the present invention;

[0076] Figure 2 is a flow chart of a data processing method of a filter provided by an embodiment of the present invention;

[0077] Figure 3 is a flow chart of a data processing method of a filter provided by an embodiment of the present invention;

[0078] Figure 4 is a flow chart of a data processing method of a filter provided by an embodiment of the present invention;

[0079] Figure 5 is a flow chart of a data processing method of a filter provided by an embodiment of the present invention;

[0080] Figure 6 is a structural diagram of a data processing device for a filter provided by an embodiment of the present invention;

[0081] Figure 7 It is a hardware architecture diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0082] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0083] See attached Figure 1 The schematic diagram of an adaptive algorithm filter is shown in Figure 1. The adaptive filter is mainly composed of two parts: one is the filter structure, such as the finite impulse response (FIR) filter structure; the other is the adaptive algorithm, such as the least mean square (LMS) algorithm. It can process the signal in real time and continuously optimize the filtering performance. The input of the filter is x(n). After passing through the filter, the output of the filter y(n) is used as negative feedback and added to the linear regression model d(n) to obtain the error e(n). An adaptive filter is a filter that can automatically adjust its filter coefficients according to the input signal and the expected signal.

[0084] It is used for channel equalization in the field of communications, which can compensate for channel distortion and improve signal transmission quality. In the echo cancellation scenario, it can effectively remove echoes in voice communications, such as in hands-free calls or video conferences on mobile phones. In the scenario of block sparse systems, non-zero coefficients are clustered in clusters in a structured manner. In such block sparse systems, the filter converges slowly. In order to improve the convergence speed of the filter, in this application, the Lp-like norm constraint is introduced into the block-sparse proportional normalized minimum mean square error algorithm (BS-PNLMS). A block sparse L2p norm improved proportional normalized minimum mean square error algorithm (BS-L2p-like-PNLMS) is proposed. The convergence rate of the mixed norm type block sparse adaptive filter is further improved, the sparsity of the unknown system is better utilized, and the performance of the algorithm is improved.

[0085] The specific implementation of the above concept is described below.

[0086] Please refer to Figure 2 , an embodiment of the present invention provides a filter data processing method, the method comprising the following steps:

[0087] In step S102, in a block sparse system, a filter estimation value of the filter at the current moment is determined.

[0088] In this embodiment, the filter estimation value at the current moment is represented by w(n), where n represents the current moment; w(n) is an L×1 vector. w(n) is divided into I groups, each group containing M=L / I coefficients. The mixed l2, p-like class norm of w(n) is represented by ||w(n)|| (2，p-like) .

[0089]

[0090] For any 0 <p<2,w [i] (n) is an M×1 vector whose mth element is the ((i-1)·M+m)th element in w(n).

[0091] In step S104, a loss function is determined according to the filter estimation value at the current moment.

[0092] In step S106, a Lagrange multiplier function is determined according to the above loss function and the linear regression objective function.

[0093] In step S108, the stochastic gradient method is used to obtain partial derivatives of the filter estimate at the next moment and the Lagrange multiplier in the Lagrange multiplier function, and the partial derivative function is set equal to zero to determine the optimal filter estimate of the filter at the next moment.

[0094] In this embodiment, the stochastic gradient method is used to calculate the partial derivatives of the filter estimate value and the Lagrange multiplier in the above-mentioned Lagrange multiplier function at the next moment, and the partial derivative function is set equal to zero. The purpose is to determine the extreme point in the local area, which is the optimal filter estimate value of the filter at the next moment.

[0095] The above-mentioned technical solution of the present application adopts the stochastic gradient method to respectively calculate the partial derivatives of the filter estimate value and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and set the partial derivative function equal to zero, determine the optimal filter estimate value of the filter at the next moment, and set the p-norm parameter p used in the above-mentioned loss function to approach zero. In the scenario of block sparse system, the convergence speed of the filter can be improved.

[0096] In some embodiments, in step S106, determining the above linear regression objective function may further include the following steps:

[0097] In the block sparse system described above, the unknown impulse response vector of the filter at the current moment, the input vector and the noise signal are determined.

[0098] The linear regression objective function is determined according to the above-mentioned unknown impulse response vector, input vector and noise signal at the current moment.

[0099] See attached Figure 3 In the above step S104, determining the loss function according to the filter estimation value at the current moment may further include the following steps:

[0100] In step S302, the difference between the filter estimation value w(n+1) at the next moment and the filter estimation value w(n) at the current moment is determined.

[0101] In step S304, the mixed norm of the difference is determined.

[0102] The first auxiliary parameter of the above mixed norm is 2, and the second auxiliary parameter is the inverse matrix G of the step matrix G(n) -1 (n).

[0103] In step S306, the inverse matrix G of the step size matrix G(n) is determined. -1 (n) and the first product of the estimated value w(n+1) at the next moment mentioned above.

[0104] In step S308, a p-norm of the first product is determined, wherein a parameter p in the p-norm tends to zero; and a second product of the p-norm of the first product and a predetermined coefficient γ.

[0105] In step S310, the sum of the mixed norm and the second product is determined as the loss function.

[0106] Exemplarily, the loss function is determined using the following formula:

[0107]

[0108] std(n)=x T (n)w(n+1);

[0109] in,

[0110] Among them, w(n+1) is the filter estimate at the next moment;

[0111] G(n) is the step size matrix;

[0112] γ is the loss function coefficient;

[0113] std(n) is the reference standard value of the linear regression function;

[0114] x(n) is the pulse input vector, whose size is L×1.

[0115] x T (n) is the transposed matrix of the pulse input vector;

[0116] X is an arbitrary matrix;

[0117] A is the auxiliary parameter matrix in the norm;

[0118] X T is the transposed matrix of matrix X;

[0119] α is an arbitrary vector;

[0120] N is the number of elements in the α vector;

[0121] α i is the i-th element in the α vector.

[0122] Among them, the norm parameter p in each matrix element in the step size matrix G(n) in the above loss function approaches zero. Making the norm parameter p in each matrix element in the step size matrix G(n) in the above loss function approach zero can improve the convergence speed of the filter.

[0123] See attached Figure 4 In step S106, determining the Lagrange multiplier function according to the above loss function and the linear regression objective function may further include the following steps:

[0124] In step S402, the second product in the loss function is multiplied by the Lagrangian multiplier λ to obtain a first Lagrangian transition product.

[0125] In step S404, a linear regression model difference between the linear regression model function and a reference standard value of the linear regression function is calculated.

[0126] In step S406, the linear regression model difference is multiplied by the Lagrangian multiplier λ to obtain a second Lagrangian transition product.

[0127] In step S408, the first Lagrangian transition product, the second Lagrangian transition product and the second-order norm of the filter estimation value difference in the loss function are summed to obtain a Lagrangian multiplier function.

[0128] Exemplarily, a specific calculation formula is used below to express it:

[0129] J(n)=(w(n+1)-w(n)) T G -1 (n)(w(n+1)-w(n))+λ(d (n)

[0130] +γ‖G -1 (n)w(n+1)‖ p-l -x T (n)w(n+1)).

[0131] Where J(n) is the Lagrange multiplier function.

[0132] λ is the Lagrange multiplier.

[0133] In some embodiments, see Appendix Figure 5 In step S108, the random gradient method is used to obtain the filter estimation value at the next moment and the Lagrange multiplier in the Lagrange multiplier function respectively to obtain the filter estimation value at the next moment, which can further include the following steps:

[0134] In step S502, the partial derivative of the filter estimation value at the next moment in the Lagrange multiplier function is calculated to obtain a first partial derivative calculation formula, and the first partial derivative calculation formula is set to zero.

[0135] In this embodiment, the following calculation formula is obtained:

[0136]

[0137] In step S504, in the Lagrange multiplier method function, the partial derivative of the Lagrange multiplier is calculated to obtain a second partial derivative calculation formula, and the second partial derivative calculation formula is set to be zero.

[0138] In this embodiment, the following calculation formula is obtained:

[0139]

[0140] In step S506, based on the fact that the first partial derivative calculation formula is equal to zero, a first calculation formula for the filter estimation value at the next moment is determined.

[0141] In this embodiment,

[0142]

[0143] in,

[0144] In step S508, in the first calculation formula of the filter estimation value at the next moment, the filter estimation value at the next moment is set equal to the filter estimation value at the current moment, and simplified to obtain the second calculation formula of the filter estimation value at the next moment.

[0145] In this embodiment, it is assumed that the difference between the filter coefficients at two moments is small, and the filter estimation value w(n+1) at the next moment is approximately equal to the filter estimation value w(n) at the current moment.

[0146]

[0147] w(n+1)=w(n)-γF(n)+λG(n)x(n);

[0148] In step S510, based on the second partial derivative calculation formula being equal to zero, the calculation formula of the linear regression objective function is determined.

[0149] In this embodiment, the calculation formula of the above linear regression objective function is as follows:

[0150] d(n)=x T (n)w(n+1)=x T (n)(w(n)-γF(n)+λG(n)x(n))=x T (n)w(n)-

[0151] γx T (n)F(n)+λx T (n)G(n)x(n).

[0152] In step S512, a calculation formula for the Lagrange multiplier is determined according to the second calculation formula for the filter estimation value at the next moment and the constraint formula of the linear regression objective function.

[0153] In this embodiment, the calculation formula of the Lagrange multiplier is as follows:

[0154]

[0155] In step S514, a third calculation formula for the filter estimation value at the next moment is determined according to the calculation formula for the Lagrange multiplier and the first calculation formula for the filter estimation value at the next moment.

[0156]

[0157] Among them, μ is the step size;

[0158] eps is a minimum value used to prevent the denominator from being zero.

[0159] The filter estimation value at the next moment calculated by the third calculation formula is used as the filter estimation value at the next moment.

[0160] In this embodiment, the filter estimation value at the next moment calculated by using the third calculation formula is a local optimal solution.

[0161] In some embodiments, the second auxiliary parameter is a step size matrix G(n); the step size matrix G(n) is a diagonal matrix, and each element in the diagonal matrix is ​​g i (n)1 M .

[0162] G(n)=diag{g1(n)1 M ,…,g N (n)1M};

[0163]

[0164] ρ is the guarantee update parameter;

[0165] q is the parameter to prevent freezing;

[0166] is the i-th block;

[0167] k i is the metric value of the i-th block;

[0168] k j is the metric value of the jth block;

[0169] g i (n) is the i-th element on the diagonal of the step size matrix;

[0170] 1M is a row vector of length M containing all 1s;

[0171] N is the number of blocks that are evenly divided in the block-sparse system.

[0172] Please refer to Figure 6 , an embodiment of the present invention provides a filter data processing device 6, the device 6 comprising:

[0173] A first determination module 61 is used to determine a filter estimation value of a filter at a current moment in a block sparse system;

[0174] A second determination module 62, configured to determine a loss function according to the filter estimation value at the current moment;

[0175] Wherein, the p-norm parameter p used in the loss function approaches zero;

[0176] A third determination module 63, used to determine a Lagrange multiplier function according to the loss function and the linear regression objective function;

[0177] The processing module 64 is used to use the stochastic gradient method to respectively obtain partial derivatives of the filter estimate and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and set the partial derivative function equal to zero to determine the optimal filter estimate of the filter at the next moment.

[0178] In some embodiments, the third determination module 63 is further used to determine the unknown impulse response vector, input vector and noise signal of the filter at the current moment in the block sparse system;

[0179] A linear regression objective function is determined according to the unknown impulse response vector, the input vector and the noise signal at the current moment.

[0180] In some embodiments, the second determination module 62 is further used to determine a filter estimation value difference between a filter estimation value at a next moment and a filter estimation value at a current moment;

[0181] Determine a mixed norm of the filter estimation value difference; the first auxiliary parameter of the mixed norm is 2, and the second auxiliary parameter is the inverse matrix of the step size matrix G(n);

[0182] Determine a first product of the inverse matrix of the step size matrix G(n) and the estimated value at the next moment;

[0183] Determining a p-norm of the first product and a second product of the p-norm of the first product and a predetermined coefficient; wherein a parameter p in the p-norm tends to zero;

[0184] A sum of the mixed norm and the second product is determined as the loss function.

[0185] In some embodiments, the processing module 64 is further used to calculate the partial derivative of the filter estimate at the next moment in the Lagrange multiplier function to obtain a first partial derivative calculation formula, and set the first partial derivative calculation formula equal to zero.

[0186] In the Lagrange multiplier method function, a partial derivative of the Lagrange multiplier is obtained to obtain a second partial derivative calculation formula, and the second partial derivative calculation formula is set to be zero;

[0187] Determine a first calculation formula for a filter estimation value at a next moment according to the first partial derivative calculation formula being equal to zero;

[0188] In the first calculation formula of the filter estimation value at the next moment, the filter estimation value at the next moment is set equal to the filter estimation value at the current moment, and simplified to obtain the second calculation formula of the filter estimation value at the next moment;

[0189] According to the second partial derivative calculation formula being equal to zero, determining the calculation formula of the linear regression objective function;

[0190] Determine a calculation formula for the Lagrange multiplier according to the second calculation formula for the filter estimation value at the next moment and the constraint formula of the linear regression objective function;

[0191] Determine a third calculation formula for the filter estimation value at the next moment according to the calculation formula of the Lagrange multiplier and the first calculation formula for the filter estimation value at the next moment;

[0192] The filter estimation value at the next moment calculated by the third calculation formula is used as the filter estimation value at the next moment.

[0193] In some embodiments, the third determination module 63 is further configured to multiply the second product in the loss function by a Lagrangian multiplier to obtain a first Lagrangian transition product;

[0194] Calculate a linear regression model difference between the linear regression model function and a reference standard value of the linear regression function;

[0195] Multiplying the linear regression model difference with the Lagrange multiplier to obtain a second Lagrange transition product;

[0196] The first Lagrangian transition product, the second Lagrangian transition product and the second-order norm of the filter estimation value difference in the loss function are summed to obtain a Lagrangian multiplier function.

[0197] It should be noted that: the data processing device of the filter provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the data processing device of the filter provided in the above embodiment and the data processing method embodiment of the filter belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0198] The embodiment of the present application also provides a computer device, please refer to Figure 7 The computer device includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the filter data processing method provided by the above-mentioned method embodiments.

[0199] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the data processing method of the filter provided in the above-mentioned method embodiments.

[0200] An embodiment of the present application further provides a filter, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the filter data processing method described in any of the above embodiments.

[0201] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0202] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0203] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0204] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A filter data processing method, characterized in that: The method comprises: In a block sparse system, determining a filter estimate of the filter at a current time instant; Determine a loss function according to the filter estimation value at the current moment; Wherein, the p-norm parameter p used in the loss function approaches zero; Determine a Lagrange multiplier function according to the loss function and the linear regression objective function; The stochastic gradient method is used to respectively obtain partial derivatives of the filter estimation value and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and the partial derivative function is set equal to zero to determine the optimal filter estimation value of the filter at the next moment.

2. The filter data processing method according to claim 1, characterized in that: Determining the linear regression objective function includes: In the block sparse system, determining the unknown impulse response vector, input vector and noise signal of the filter at the current moment; A linear regression objective function is determined according to the unknown impulse response vector, the input vector and the noise signal at the current moment.

3. The filter data processing method according to claim 1, characterized in that: The step of determining the loss function according to the filter estimation value at the current moment specifically includes: Determine a filter estimate value difference between a filter estimate value at a next moment and a filter estimate value at a current moment; determining a mixed norm of the filter estimate difference; The first auxiliary parameter of the mixed norm is 2, and the second auxiliary parameter is the inverse matrix of the step size matrix G(n); Determine a first product of the inverse matrix of the step size matrix G(n) and the estimated value at the next moment; Determining a p-norm of the first product and a second product of the p-norm of the first product and a predetermined coefficient; wherein a parameter p in the p-norm tends to zero; A sum of the mixed norm and the second product is determined as the loss function.

4. The filter data processing method according to claim 2, characterized in that: The stochastic gradient method is used to obtain partial derivatives of the filter estimation value and the Lagrange multiplier at the next moment in the Lagrange multiplier function, respectively, to obtain the filter estimation value at the next moment, including: In the Lagrange multiplier function, a partial derivative of the filter estimate at the next moment is calculated to obtain a first partial derivative calculation formula, and the first partial derivative calculation formula is set to zero; In the Lagrange multiplier method function, a partial derivative of the Lagrange multiplier is obtained to obtain a second partial derivative calculation formula, and the second partial derivative calculation formula is set to be zero; Determine a first calculation formula for a filter estimation value at a next moment according to the first partial derivative calculation formula being equal to zero; In the first calculation formula of the filter estimation value at the next moment, the filter estimation value at the next moment is set equal to the filter estimation value at the current moment, and simplified to obtain the second calculation formula of the filter estimation value at the next moment; According to the second partial derivative calculation formula being equal to zero, determining the calculation formula of the linear regression objective function; Determine a calculation formula for the Lagrange multiplier according to the second calculation formula for the filter estimation value at the next moment and the constraint formula of the linear regression objective function; Determine a third calculation formula for the filter estimation value at the next moment according to the calculation formula of the Lagrange multiplier and the first calculation formula for the filter estimation value at the next moment; The filter estimation value at the next moment calculated by the third calculation formula is used as the filter estimation value at the next moment.

5. The filter data processing method according to claim 3, characterized in that: Determining a Lagrange multiplier function according to the loss function and the linear regression objective function includes: Multiplying the second product in the loss function by a Lagrange multiplier to obtain a first Lagrangian transition product; Calculate a linear regression model difference between the linear regression model function and a reference standard value of the linear regression function; Multiplying the linear regression model difference with the Lagrange multiplier to obtain a second Lagrange transition product; The first Lagrangian transition product, the second Lagrangian transition product and the second-order norm of the filter estimation value difference in the loss function are summed to obtain a Lagrangian multiplier function.

6. The filter data processing method according to claim 1, characterized in that: The step size matrix G(n) is a diagonal matrix, and each element in the diagonal matrix is ​​g i (n)1 M ; G(n)=diag{g1(n)1 M ,…,g N (n)1 M }; ρ is the guarantee update parameter; q is the parameter to prevent freezing; is the i-th block; k i is the metric value of the i-th block; k j is the metric value of the jth block; g i (n) is the i-th element on the diagonal of the step size matrix; 1M is a row vector of length M containing all 1s; N is the number of blocks that are evenly divided in the block-sparse system.

7. A filter data processing device, characterized in that: The device comprises: A first determination module, used to determine a filter estimation value of the filter at a current moment in a block sparse system; A second determination module, used to determine a loss function according to the filter estimation value at the current moment; Wherein, the p-norm parameter p used in the loss function approaches zero; A third determination module is used to determine a Lagrange multiplier function according to the loss function and the linear regression objective function; The processing module is used to adopt the stochastic gradient method to respectively obtain partial derivatives of the filter estimation value and the Lagrange multiplier in the Lagrange multiplier function at the next moment, and set the partial derivative function equal to zero to determine the optimal filter estimation value of the filter at the next moment.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the filter data processing method described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the filter data processing method according to any one of claims 1 to 6 are implemented.

10. A filter, characterized in that: The invention comprises a computer program, which, when executed by a processor, implements the steps of the filter data processing method according to any one of claims 1 to 6.

Images