A wide linear adaptive filter against impulse noise interference

Abstract

The application relates to the technical field of adaptive filter, and discloses a wide linear adaptive filter, which comprises tensor product decomposition of a system weight vector of the wide linear adaptive filter at a moment, so as to obtain two adaptive weight vectors; an input matrix is constructed based on sample values of an input signal, and an estimated signal at the moment is calculated based on the product of the input matrix at the moment and the two adaptive weight vectors; a difference value between an expected signal at the moment and the estimated signal is calculated, an estimated error signal is obtained, a gain vector at the moment is updated by using a minimum error module criterion, the two adaptive weight vectors at the moment are updated by using the estimated error signal, and the system weight vector of the wide linear adaptive filter at the moment is reconstructed. The application realizes a recursive adaptive filter algorithm resisting impulse noise interference with low calculation complexity by using weight value decomposition and selection of a robust cost function.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of adaptive filter technology, in particular to a wide linear adaptive filter with anti-pulse noise interference. BACKGROUND

[0002] System identification is an important branch of adaptive signal processing, and many problems such as traditional adaptive channel equalization, adaptive noise cancellation, adaptive echo cancellation, and active noise control can be attributed to system identification. A wide linear adaptive filter is a complex system identification model, which has the characteristics of filtering both the complex input signal vector and its conjugate vector, and can fully utilize the second-order non-circular characteristics of complex signals to overcome the limitation of strict linear model adaptive filter that can only process circular complex signals. The non-circularity of complex signals includes the inequality of energy of real and imaginary components or the correlation of the two components.

[0003] The commonly used wide linear adaptive filter includes a wide linear least mean square (WL-LMS) adaptive filter and a wide linear recursive least square (WL-RLS) adaptive filter. The former updates the filter weight using the random gradient descent method only according to the input signal and the expected signal at the current time, has small calculation amount but slow convergence speed, while the latter updates the filter weight according to all known input and expected signals at all times, has fast convergence but large calculation amount. When the input is a colored signal, the convergence speed of WL-LMS will be further slowed down, while the WL-RLS adaptive filter is not affected and has a significant advantage in convergence performance.

[0004] However, the WL-RLS adaptive filter is based on the least mean square error criterion, and has good convergence performance in a noise environment that approximately conforms to the Gaussian model, but when there is severe non-Gaussian noise such as pulse noise, the performance will sharply decrease, and even diverge, and it is not robust. In view of this situation, some robust recursive adaptive filters are proposed, such as recursive maximum correlation entropy adaptive filter. These filters introduce nonlinear processing of the error signal, which can suppress the destructive effect of pulse noise, but the calculation amount of the nonlinear function is large, which becomes an obstacle in many low-cost and miniaturized application scenarios.

[0005] In the WL-RLS adaptive filter, the convergence accuracy and the tracking speed are contradictory, if the accuracy requirement is higher, the memory of the past information needs to be increased, but this will lead to the tracking speed of the time-varying system to be reduced, and the time delay of the adaptive filter to be enlarged. Moreover, when the impulse response of the unknown system is long, the calculation complexity of the WL-RLS adaptive filter in updating the inverse correlation matrix will be obviously increased, which becomes a thorny problem. The recursive least square adaptive filter will encounter the problems of high calculation complexity, slow tracking speed and poor filtering robustness when coping with the unknown system with long impulse response.

[0006] In summary, the existing wide linear adaptive filter needs to process the signal vector and its conjugate vector at the same time, the high-dimensional matrix operation in the complex domain leads to the increase of the calculation complexity, the reduction of the filtering efficiency, and the inability to effectively track the update of the system weight vector and the inverse correlation matrix in the time-varying system; and the impulse noise affects the system update, leading to poor system robustness and filtering distortion. SUMMARY

[0007] Therefore, the technical problem to be solved by the present application is to overcome the problem that the prior art cannot balance the calculation complexity and the filtering robustness, leading to slow filtering speed and poor filtering effect of the filter.

[0008] To solve the above technical problems, the present application provides a wide linear adaptive filter resistant to impulse noise interference, comprising:

[0009] The wide linear adaptive filter The system weight vector at the moment is decomposed by tensor product to obtain The first adaptive weight vector and the second adaptive weight vector at the moment;

[0010] The sampling values of the input signals at the moment and the continuous multiple moments before the moment are obtained, and a standard input signal vector is constructed; the standard input signal vector and its conjugate vector are spliced to obtain an augmented input signal vector, and The input matrix at the moment is constructed.

[0011] Based on the product of the input matrix at the moment and the first adaptive weight vector and the second adaptive weight vector, the The estimated signal at the moment is calculated as the output signal of the wide linear adaptive filter The moment. The difference between the expected signal and the estimated signal at the moment is calculated to obtain

[0012] The estimated error signal at the moment. Based on the product of the input matrix at the moment and the first adaptive weight vector and the second adaptive weight vector, the The moment.

[0013] ​​The estimated error signal at time step is updated using the minimum error modulus criterion. Gain vector at time step;

[0014] based on The gain vector at time step and the estimation error signal, for The first adaptive weight vector and the second adaptive weight vector are updated at time step [time] to obtain [the desired result]. Using the first and second adaptive weight vectors at time points, reconstruct the wide linear adaptive filter. The system weight vector at time t.

[0015] Preferably, for wide linear adaptive filters The system weight vector at time t is obtained by tensor integration. The first adaptive weight vector and the second adaptive weight vector at time t are represented as follows:

[0016] The first adaptive weight vector at time 1 , represented as: ;

[0017] The second adaptive weight vector at time 1 , represented as: ;

[0018] in, and They represent The first adaptive weight vector at time 1 With the second adaptive weight vector The first in Sub-weight vectors , This indicates the order of the tensor integral solution. The length of each sub-weight vector in the first adaptive weight vector is The length of each sub-weight vector in the second adaptive weight vector is , , The length of the weight vector of the standard augmented complex adaptive filter; This indicates the transpose operation.

[0019] Preferably, based on The estimated error signal at time step is updated using the minimum error modulus criterion. The gain vector at time step includes:

[0020] Based on the The length of each sub-weight vector in the adaptive weight vector is calculated. forgetting factor corresponding to the adaptive weight vector , is denoted as: ; , denotes the first adaptive weight vector, denotes the second adaptive weight vector; denotes a preset hyperparameter, denotes the first adaptive weight vector, denotes the length of each sub-weight vector in the adaptive weight vector, denotes the decomposition order of the tensor product;

[0021] Based on the input matrix at the moment and the adaptive weight vector, the first adaptive weight vector is constructed The regression quantity at the moment corresponding to the adaptive weight vector The regression quantity at the moment corresponding to the first adaptive weight vector is denoted as: , includes:

[0022] The regression quantity at the moment corresponding to the second adaptive weight vector is denoted as:

[0023] ; denotes the transpose operation;

[0024] The regression quantity at the moment corresponding to the second adaptive weight vector is denoted as:

[0025] ; denotes the conjugate transpose operation, denotes the conjugate transpose of the input matrix at the moment;

[0026] Based on the regression quantity at the moment corresponding to the first adaptive weight vector, and the product of the inverse correlation matrix of the regression quantity, the intermediate variable at the moment corresponding to the first adaptive weight vector is obtained The intermediate variable at the moment corresponding to the first adaptive weight vector is denoted as: ; denotes the inverse correlation matrix of the regression quantity at the moment; The gain vector at the moment corresponding to the first adaptive weight vector is calculated based on the intermediate variable at the moment corresponding to the first adaptive weight vector, the forgetting factor, the estimation error signal and the regression quantity The gain vector at the moment corresponding to the first adaptive weight vector is denoted as:

[0027] The gain vector at the moment corresponding to the first adaptive weight vector is calculated based on the intermediate variable at the moment corresponding to the first adaptive weight vector, the forgetting factor, the estimation error signal and the regression quantity The gain vector at the moment corresponding to the first adaptive weight vector is denoted as: The gain vector at the moment corresponding to the first adaptive weight vector is calculated based on the intermediate variable at the moment corresponding to the first adaptive weight vector, the forgetting factor, the estimation error signal and the regression quantity The gain vector at the moment corresponding to the first adaptive weight vector is denoted as:​​​​​​​​​

[0028] ; express The estimation error signal at time.

[0029] Preferably, based on The gain vector at time step and the estimation error signal, for The first adaptive weight vector and the second adaptive weight vector are updated at time step [time] to obtain [the desired result]. The first adaptive weight vector and the second adaptive weight vector at time step include:

[0030] calculate The product of the gain vector at time step and the estimation error signal, and... The adaptive weight vectors at each time step are summed to obtain... The adaptive weight vector at time step 1 is represented as:

[0031] The first adaptive weight vector at time 1 ;

[0032] The second adaptive weight vector at time 1 ;

[0033] in, This indicates the conjugate operation.

[0034] Preferably, a reconstructed wide linear adaptive filter The system weight vector at time t is represented as:

[0035] ;

[0036] in, express The weight vector of the wide linear adaptive filter at time t; and They represent The first adaptive weight vector at time 1 With the second adaptive weight vector The first in Sub-weight vectors; This represents the Kronecker product operation.

[0037] Preferably, based on the first The adaptive weight vector corresponding to Regression at time conjugate transpose ,and inverse correlation matrix and the The adaptive weight vector corresponding to The forgetting factor of time and gain vector Update to obtain the first The adaptive weight vector corresponding to Regression at time inverse correlation matrix , represented as:

[0038] .

[0039] Preferably, The construction of the input matrix at each time step includes:

[0040] Get Time and preceding consecutive Sample value of the input signal at each moment Construct the standard input signal vector ;

[0041] standard input signal vector Its conjugate vector splice to obtain a length of augmented input signal vector ;

[0042] Divide the augmented input signal vector into equal parts A length of The sub-input vector is used to obtain the equally divided augmented input signal vector, which is represented as: ;

[0043] Based on the equally divided augmented input signal vector, construct Input matrix at time step ;

[0044] in, This indicates the conjugate operation. This indicates the conjugate transpose operation.

[0045] Preferably, the product of the input matrix and the first adaptive weight vector and the second adaptive weight vector is used to calculate... The estimated signal at time t is expressed as:

[0046] ;

[0047] in, express The estimated signal at time, express The conjugate transpose of .

[0048] Preferably, calculation The difference between the expected signal and the estimated signal at the time instant is obtained The estimation error signal at the time instant is represented as:

[0049]

[0050] Wherein, The expected signal is represented as: The estimation error signal at the time instant, The expected signal is represented as:

[0051] Preferably, when applied to a stereo video communication echo cancellation system, the method comprises:

[0052] The input signal is a complex signal constructed by taking two near-end loudspeaker signals in the stereo video communication echo cancellation system as the real part and the imaginary part, respectively;

[0053] The expected signal is a complex signal constructed by taking two near-end microphone signals in the stereo video communication echo cancellation system as the real part and the imaginary part, respectively.

[0054] The above technical solutions of the present application have the following beneficial effects compared with the prior art:

[0055] The anti-impulse noise interference wide-linearity adaptive filter provided by the present application can effectively reduce the calculation amount of updating the inverse correlation matrix and improve the tracking ability of the filter, thereby improving the convergence speed and calculation efficiency of the system. Meanwhile, the minimum error module criterion is introduced into the recursive adaptive algorithm, and the tensor product decomposition method is used to reduce the time delay and calculation complexity of the algorithm, so that robust adaptive filtering of the output signal of the system to be estimated is realized. Meanwhile, the cost function selected by the present application is based on the minimum error module at all time instants, so that in the case that the expected signal is superimposed with impulse interference, a filter result with faster convergence speed and better robust performance can be obtained.

[0056] When applied to a stereo video communication echo cancellation system, the present application uses a wide-linearity model to design an augmented complex adaptive filter, which can realize the function of double-channel echo cancellation with a single filter, can fully utilize the second-order non-circular characteristics of the complex signal to improve the filtering ability, and can achieve better filtering effect and improve the communication quality. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific embodiments of the present application and in conjunction with the drawings, in which:

[0058] Figure 1 is a working principle diagram of the anti-impulse noise interference wide-linearity adaptive filter of the present application; ​

[0059] Figure 2 This is a system block diagram of the system identification model;

[0060] Figure 3 This is a comparison chart of the normalized mean square deviation (NMSD) curves of the adaptive filtering system in the system identification scenario. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0062] Reference Figure 1 The diagram shows the working principle of the wide linear adaptive filter for resisting impulse noise interference of the present invention. Its working process includes:

[0063] S101: For wide-linear adaptive filters The system weight vector at time t is obtained by tensor integration. The first adaptive weight vector and the second adaptive weight vector at time t;

[0064] S102: Acquisition The standard input signal vector is constructed by sampling the input signal at time t and several consecutive time points prior to time t. The standard input signal vector is then concatenated with its conjugate vector to obtain the augmented input signal vector, and this is used to construct the augmented input signal vector. The input matrix at time step;

[0065] S103: Based on The product of the input matrix at time step 1 with the first adaptive weight vector and the second adaptive weight vector is calculated. Estimated signal at time As a wide linear adaptive filter The output signal at time t is represented as:

[0066] ;

[0067] in, express The conjugate transpose of ;

[0068] S104: Calculation Expected signal at time The difference between the estimated signal and the actual signal is obtained. Time estimation error signal , is represented as: ;

[0069] S105: Based on The estimated error signal at time step is updated using the minimum error modulus criterion. Gain vector at time step;

[0070] S106: Based on The gain vector at time step and the estimation error signal, for The first adaptive weight vector and the second adaptive weight vector are updated at time step [time] to obtain [the desired result]. Using the first and second adaptive weight vectors at time points, reconstruct the wide linear adaptive filter. The system weight vector at time t.

[0071] In this embodiment of the invention, the wide linear adaptive filter is divided into two parts, one part being based on The input signal at time is filtered to obtain The estimated signal output at time; another part is based on The estimated signal and the expected signal at time t, the estimation error signal are calculated, and then... The system weight vector of the wide linear adaptive filter at time step 1 is updated to obtain... A time-varying wide-linear adaptive filter continuously filters the input signal and outputs a new signal.

[0072] Specifically, in step S101, the tensor integral solves the wide-width linear adaptive filter. The system weight vector at time t is obtained The first adaptive weight vector and the second adaptive weight vector at time t are represented as follows:

[0073] The first adaptive weight vector at time 1 , represented as: ;

[0074] The second adaptive weight vector at time 1 , represented as: ;

[0075] in, and They represent The first adaptive weight vector at time 1 With the second adaptive weight vector The first in Sub-weight vectors , This indicates the order of the tensor integral solution. The length of each sub-weight vector in the first adaptive weight vector is The length of each sub-weight vector in the second adaptive weight vector is , , is the length of the standard augmented complex adaptive filter weight vector; denotes the transpose operation.

[0076] The tensor product decomposition method can improve the tracking speed of the adaptive filter for a low-rank system, reduce the time delay, and decompose the autocorrelation matrix in the adaptive algorithm to be smaller, thereby saving the calculation amount. The low-rank here refers to the optimal weight of the adaptive filter which can be approximately arranged into a low-rank matrix. Therefore, the wide linear adaptive filter against impulse noise interference decomposes the long weight vector into a series of shorter weight vectors in the form of a sum of tensor products, reduces the calculation complexity of the adaptive filter under acceptable precision loss, and improves the tracking speed.

[0077] Specifically, in step S102, the input matrix at the moment is constructed .

[0078] S102-1: Obtain the sampling values of the input signal at the moment and the continuous moments before the moment , and construct the standard input signal vector . S102-2: Concatenate the standard input signal vector and its conjugate vector

[0079] to obtain the augmented input signal vector with a length of ; wherein, denotes the conjugate operation, denotes the conjugate transpose operation; S102-3: Divide the augmented input signal vector into sub-input vectors with a length of , obtain the divided augmented input signal vector, denoted as:

[0080] . S102-4: Construct the input matrix based on the divided augmented input signal vector.

[0081] Specifically, in step S105, the gain vector at the moment is updated, including: S105-1: Based on the length of each sub-weight vector in the adaptive weight vector, calculate the forgetting factor corresponding to the adaptive weight vector

[0082] , denoted as:

[0083] S105-2: Update the gain vector at the moment based on the forgetting factor corresponding to the adaptive weight vector , denoted as: .​​​; , denotes the first adaptive weight vector, denotes the second adaptive weight vector; denotes a preset hyperparameter, denotes the first adaptive weight vector, denotes the length of each sub-weight vector in the adaptive weight vector, denotes the decomposition order of the tensor product;

[0084] S105-2: based on the input matrix at the moment and the adaptive weight vector, construct the regression quantity at the moment corresponding to the first adaptive weight vector S105-3: based on the regression quantity at the moment corresponding to the first adaptive weight vector, and the product of the inverse correlation matrix of the regression quantity, obtain the intermediate variable at the moment corresponding to the first adaptive weight vector S105-4: based on the intermediate variable at the moment corresponding to the first adaptive weight vector, the forgetting factor, the estimation error signal and the regression quantity, calculate the gain vector at the moment corresponding to the first adaptive weight vector

[0085] the regression quantity at the moment corresponding to the first adaptive weight vector is expressed as:

[0086] ; denotes the transpose operation;

[0087] the regression quantity at the moment corresponding to the second adaptive weight vector is expressed as:

[0088] ; denotes the conjugate transpose operation, denotes the conjugate transpose of the input matrix at the moment

[0089] S105-3: based on the regression quantity at the moment corresponding to the first adaptive weight vector, and the product of the inverse correlation matrix of the regression quantity, obtain the intermediate variable at the moment corresponding to the first adaptive weight vector S105-4: based on the intermediate variable at the moment corresponding to the first adaptive weight vector, the forgetting factor, the estimation error signal and the regression quantity, calculate the gain vector at the moment corresponding to the first adaptive weight vector ; denotes the inverse correlation matrix of the regression quantity at the moment

[0090] S105-4: based on the intermediate variable at the moment corresponding to the first adaptive weight vector, the forgetting factor, the estimation error signal and the regression quantity, calculate the gain vector at the moment corresponding to the first adaptive weight vector

[0091] ​​​​​​​​​​​​​​​​​ ; express The estimation error signal at time.

[0092] Among them, based on the first The adaptive weight vector corresponding to Regression at time conjugate transpose ,and inverse correlation matrix and the The adaptive weight vector corresponding to The forgetting factor of time and gain vector Update to obtain the first The adaptive weight vector corresponding to Regression at time inverse correlation matrix , represented as: .

[0093] Specifically, in step S106, the calculation is performed. The product of the gain vector at time step and the estimation error signal, and... The adaptive weight vectors at each time step are summed to obtain... The adaptive weight vector at time step is expressed as:

[0094] The first adaptive weight vector at time 1 ;

[0095] The second adaptive weight vector at time 1 .

[0096] Specifically, based on The adaptive weight vector at time step is used to reconstruct the wide linear adaptive filter. The system weight vector at time t is represented as:

[0097] ;

[0098] in, express The weight vector of the wide linear adaptive filter at time t; and They represent The first adaptive weight vector at time 1 With the second adaptive weight vector The first in Sub-weight vectors; This represents the Kronecker product operation.

[0099] The wide-linear adaptive filter for resisting impulse noise interference described in this invention utilizes tensor integral solutions to decompose the traditional high-dimensional system weight vector into two sets of low-dimensional adaptive weight vectors. This effectively reduces the computational cost of updating the inverse correlation matrix and improves the filter's tracking capability, thereby increasing the system's convergence speed and computational efficiency. Simultaneously, the minimum error modulus criterion is introduced into the recursive adaptive algorithm, and the tensor integral solution method is used to reduce the algorithm's time delay and computational complexity, achieving robust adaptive filtering of the output signal of the system to be estimated. Furthermore, the cost function selected in this invention is based on the minimum error modulus at all times, enabling faster convergence and better robust filtering results even when the desired signal is superimposed with impulse interference.

[0100] Based on the above embodiments, in this embodiment of the invention, when the wide linear adaptive filter for resisting impulse noise interference provided by this embodiment of the invention is applied to a stereo video communication echo cancellation system, it includes:

[0101] The input signal is a complex signal constructed by using the two near-end speaker signals from the stereo video communication echo cancellation system as the real and imaginary parts, respectively.

[0102] The desired signal is a complex signal constructed using the real and imaginary parts of two near-end microphone signals from a stereo video communication echo cancellation system, respectively.

[0103] At this point, echo cancellation in stereo video communication specifically includes:

[0104] Using the two near-end speaker signals from the stereo video communication echo cancellation system as the real and imaginary parts respectively, the input signals of the wide linear adaptive filter are constructed in real time.

[0105] Using the two near-end microphone signals from a stereo video communication echo cancellation system as the real and imaginary parts respectively, the desired signal of a wide linear adaptive filter is constructed in real time.

[0106] The input signal is input in real time into the wide linear adaptive filter for resisting impulse noise interference as described above to obtain... The estimated signal at time;

[0107] based on The estimated signal and the expected signal at time are calculated. The estimation error signal at time t, as The output of a clean signal after echo cancellation at any given moment;

[0108] Using the minimum error modulus criterion, obtain Gain vector at time step;

[0109] based on The gain vector at time step and the estimation error signal are used to update and obtain the wide linear adaptive filter. The system weight vector at time t;

[0110] based on The system weight vector at each time step, along with the real-time input signal and the desired signal, outputs a clean signal in real time and updates the system weight vector of the wide linear adaptive filter at the next time step.

[0111] When applied to a stereo video communication echo cancellation system, this invention utilizes a wide linear model to design an augmented complex adaptive filter. This filter can achieve dual-channel echo cancellation with a single filter and fully leverages the second-order non-circular characteristics of complex signals to improve filtering capabilities, resulting in better filtering effects and improved communication quality.

[0112] Based on the above embodiments, in this embodiment of the invention, the step of identifying a wide linear system in a scenario where the desired signal contains impulse noise, using the wide linear adaptive filter for resisting impulse noise provided by the present invention, specifically includes:

[0113] S201: Based on the system weight vector of a wide linear adaptive filter, construct two sets of adaptive weight vectors after the tensor integral solution. and ;

[0114] Sub-weight vector and The lengths are respectively and ,satisfy , The length of the weight vector of the standard augmented complex adaptive filter; Let the order of the tensor integral solution be . ;

[0115] S202: By Time and preceding consecutive The sampled value of the complex input signal input to the system to be estimated at each time step. The standard input signal vector is composed of The standard input signal vector is concatenated with its conjugate vector. Then the length is of Divided into equal parts A length of The sub-input vector, i.e. Finally, the input matrix is ​​constructed using the individual sub-input vectors. ; superscript Indicates the transpose operation;

[0116] S203: By Adaptive filter is calculated by two adaptive weight vectors and input matrix at time Estimated signal at time , is expressed as: ;

[0117] S204: Calculate the difference between the expected signal and the estimated signal at time , obtain the estimated error signal at time , is expressed as: ;

[0118] , wherein, represents the estimated error signal at time , represents the expected signal at time , represents the estimated signal at time ;

[0119] S205: Update the adaptive weight vector and the inverse correlation matrix based on the minimum error norm criterion, obtain the tensor product decomposition weight vector at time , including:

[0120] S205-1: Calculate the regression quantity about and at time , , , and the forgetting factor , , are expressed as:

[0121] ;

[0122] ;

[0123] ; ;

[0124] , wherein, is a hyperparameter for balancing the convergence speed and the steady-state error, , and are the vector lengths of and , respectively;

[0125] Based on the minimum error norm criterion, the application derives the update formula of the tensor product decomposition weight vector by using the recursive optimization method, has a fast convergence speed, and has good robustness in the environment containing impulse noise;

[0126] S205-2: Calculate the intermediate variable based on the inverse correlation matrix and the regression quantity ​、 , is denoted as:

[0127] ; ;

[0128] wherein, and are the inverse correlation matrices of and after the th iteration, respectively;

[0129] S205-3: Calculate the gain vector at time , denoted as:

[0130] ;

[0131] ;

[0132] wherein, is the estimation error signal, and are two forgetting factors, and are the regressors for updating and , respectively, and are two intermediate variables;

[0133] S205-4: Update the adaptive weight vector at time , denoted as:

[0134] ; ;

[0135] S205-5: Update the inverse correlation matrix at time , denoted as:

[0136] ; ;

[0137] In the iteration process, the initial values of the weight vector and are set as: , , wherein, denotes concatenating all columns of a matrix into a column vector, denotes a unit matrix of size , semicolon denotes concatenating two matrices up and down, denotes a zero matrix of size ; the initial value of the inverse correlation matrix , Set as , ; is a constant When the environmental noise is large, taking a smaller value is beneficial to the stability in the initial stage of convergence.

[0138] S206: by Adaptive filter two groups of adaptive weight vectors at time and , reconstruct System weight vector at time , expressed as: , wherein Indicates the Kronecker product operation;

[0139] S207: by Input signal matrix at time and Weight vector at time, calculate the estimated signal of adaptive filter At time, expressed as: .

[0140] The wide linear adaptive filter of the application can introduce the minimum error modulus criterion into the recursive adaptive algorithm, and use the tensor product decomposition method to reduce the time delay of the algorithm and reduce the calculation complexity, so as to realize the robust adaptive filtering of the estimated system output signal. The wide linear model is used to design the augmented complex adaptive filter, so that the double-channel echo cancellation function can be realized by a single filter. At the same time, the cost function selected by the application is based on the minimum error modulus at all times, so that in the case of superimposed pulse interference on the expected signal, the filter result with faster convergence speed and better robust performance can be obtained. The tensor product decomposition method used by the application can effectively reduce the calculation amount of updating the inverse correlation matrix, and improve the tracking ability of the filter.

[0141] In order to prove the effectiveness of the application, the computer experiment method is used in this embodiment to verify the performance of the wide linear recursive sign (NKP-WL-RSA) adaptive filter based on tensor product decomposition provided by the application. The unknown system is estimated in the system identification application scene under the environment containing pulse noise interference, and the experimental results of the wide linear recursive least square (WL-RLS) adaptive filter, the tensor product decomposition wide linear recursive least square (NKP-WL-RLS) adaptive filter and the wide linear recursive sign (WL-RSA) adaptive filter are compared.

[0142] Referring to Figure 2 , it is a system block diagram of a system identification model. The noise signal Gaussian noise plus impulse noise was added; the normalized mean square deviation (NMSD) was used as the performance measure in the system's scene identification experiment. The unit is dB, where Indicates taking the logarithm. These are the weights of the actual system.

[0143] In the experiment, the input signal was generated by passing a Gaussian white signal through a first-order autoregressive system with a pole of 0.9, and the noise signal used was... It contains a mean of zero and a variance of . Gaussian white noise and a pulse noise ,Right now Gaussian noise maintains a 30dB signal-to-noise ratio (SNR) with the desired noiseless signal, while impulse noise maintains a -10dB SNR. (Impulse noise) Produced by Bernoulli Gaussian process, i.e. ,in It is a Bernoulli process, and the probability of it taking the value 0 is 0.9 and the probability of it taking the value 1 is 0.1. Gaussian white noise with zero mean.

[0144] Reference Figure 3 The figure shows a comparison of the normalized mean square deviation (NMSD) curves of the adaptive filtering system in the system identification scenario; the parameters of each method are WL-RLS ( , ), NKP-WL-RLS ( , , WL-RSA , ), the NKP-WL-RSA implementation of this application ( , , ).Depend on Figure 3 As can be seen, the NKP-WL-RSA adaptive filtering system of this application has good anti-pulse convergence performance while maintaining low computational complexity, and can achieve faster convergence and tracking speed.

[0145] The wide linear anti-impulse noise interference adaptive filter provided by the application can effectively reduce the calculation amount of updating the inverse correlation matrix, improve the tracking ability of the filter, and further improve the convergence speed and calculation efficiency of the system; meanwhile, the minimum error module criterion is introduced into the recursive adaptive algorithm, and the tensor product decomposition method is used to reduce the time delay and calculation complexity of the algorithm, so that the robust adaptive filtering of the estimated system output signal is realized.

[0146] Those skilled in the art will understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0147] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The functions specified in one or more flows and / or blocks.

[0148] These computer program instructions can also be stored in a computer readable storage medium that can guide the computer or other programmable data processing devices to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The functions specified in one or more flows and / or blocks.

[0149] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, so that the instructions executed on the computer or other programmable data processing devices provide processes for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0150] Obviously, the above embodiments are only examples for the purpose of clear illustration, and are not limitations to the embodiments. Based on the above description, other different forms of changes or variations can also be made by those skilled in the art. Here, it is not necessary and also impossible to enumerate all the embodiments. The obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A wide linear adaptive filter resistant to impulse noise interference, characterized in that, Comprise: Wide linear adaptive filter The system weight vector at time t is decomposed into a tensor product as The first and second adaptive weight vectors at time t are expressed as first adaptive weight vector at time instant , is represented as: ; second adaptive weight vector at time instant is denoted as: ; wherein with respectively a first adaptive weight vector and a second adaptive weight vector , a first adaptive weight vector , denotes the decomposition order of the tensor product ; the length of each subweight vector in the first adaptive weight vector , the length of each subweight vector in the second adaptive weight vector , , is the length of the standard augmented complex adaptive filter weight vector; denotes the transposition operation; Acquisition The sampling values of the input signals at the time and the continuous multiple times before the time are obtained to construct a standard input signal vector; the standard input signal vector is spliced with a conjugate vector thereof to obtain an augmented input signal vector, and an input matrix at the time is constructed ​ based on input matrix at time instant a product of the first adaptive weight vector and the second adaptive weight vector, to calculate estimated signal at time instant as a wide-linear adaptive filter output signal at time instant denotes the conjugate transpose vector of Computing the difference between the desired signal and the estimated signal at the time instant, obtaining an estimation error signal at the time instant; Based on the estimation error signal of the time instant, the gain vector of the time instant is updated by using a minimum error norm criterion, comprising: the estimation error signal of the time instant, the gain vector of the time instant is updated by using a minimum error norm criterion, comprising: Based on the The length of each sub-weight vector in the adaptive weight vector is calculated. Forgetting factor corresponding to adaptive weight vector , represented as: ; , Time represents the first adaptive weight vector. Time represents the second adaptive weight vector; This indicates the preset hyperparameters. Indicates the first The length of each sub-weight vector in the adaptive weight vector; Based on the input matrix and the adaptive weight vector at the moment, the first regression quantity at the moment is constructed corresponding to the adaptive weight vector the regression quantity at the moment , comprising: The first adaptive weight vector corresponds to The regression quantity at the time instant t is expressed as: ; denotes a transpose operation; The second adaptive weight vector corresponds to The regression quantity at the time instant t is expressed as: ; denotes a conjugate transpose operation, denotes input matrix at time instant the conjugate transpose of Based on the first The adaptive weight vector corresponds to The regression quantity at the moment t, and the product of the inverse correlation matrix of the regression quantity, obtain the first The adaptive weight vector corresponds to The intermediate variable at the moment t , expressed as: ; The inverse correlation matrix of the regression quantity At the moment t ; Based on the first The gain vector corresponding to the adaptive weight vector at the time point The intermediate variable, the forgetting factor, the estimation error signal and the regression quantity at the time point The gain vector corresponding to the adaptive weight vector at the time point The gain vector corresponding to the adaptive weight vector at the time point , is expressed as: ; indicates an estimation error signal of the time instant; based on updating the first adaptive weight vector and the second adaptive weight vector at the time instant based on the gain vector at the time instant and the estimation error signal, updating the first adaptive weight vector and the second adaptive weight vector at the time instant based on the gain vector at the time instant and the estimation error signal, updating the first adaptive weight vector and the second adaptive weight vector at the time instant based on the gain vector at the time instant and the estimation error signal, updating the first adaptive weight vector and the second adaptive weight vector at the time instant based on the gain vector at the time instant and the estimation error signal.

2. The wide-linear, anti-impulse noise interference adaptive filter of claim 1, wherein, Based on the gain vector at the moment and the estimation error signal, the first adaptive weight vector and the second adaptive weight vector at the moment are updated, and the first adaptive weight vector and the second adaptive weight vector at the moment are obtained the gain vector at the moment and the estimation error signal, the first adaptive weight vector and the second adaptive weight vector at the moment are updated, and the first adaptive weight vector and the second adaptive weight vector at the moment are obtained the gain vector at the moment and the estimation error signal, the first adaptive weight vector and the second adaptive weight vector at the moment are updated, and the first adaptive weight vector and the second adaptive weight vector at the moment are obtained Computing the product of the gain vector at time n and the estimation error signal, and adding the product to the adaptive weight vector at time n-1, to obtain the adaptive weight vector at time n, denoted as the product of the gain vector at time n and the estimation error signal, and adding the product to the adaptive weight vector at time n-1, to obtain the adaptive weight vector at time n, denoted as the product of the gain vector at time n and the estimation error signal, and adding the product to the adaptive weight vector at time first adaptive weight vector at time instant ; second adaptive weight vector at time instant ; wherein represents a conjugation operation.

3. The wide-linear, anti-impulse noise interference adaptive filter of claim 2, wherein, Reconfiguring wide linear adaptive filters The system weight vector at time instant n is denoted as: ； wherein denotes a wide linear adaptive filter a system weight vector at time instant and denote a first adaptive weight vector at time instant and a second adaptive weight vector the thsubweight vector in denotes a Kronecker product operation.

4. The wide linear, anti-impulse noise interference adaptive filter of claim 1, wherein, Based on the The adaptive weight vector corresponding to Regression at time conjugate transpose ,and inverse correlation matrix and the The adaptive weight vector corresponding to The forgetting factor of time and gain vector Update to obtain the first The adaptive weight vector corresponding to Regression at time inverse correlation matrix , represented as: 。 5. The wide linear, anti-impulse noise interference adaptive filter of claim 1, wherein, Construction of the input matrix of the moment, including: acquiring successive sample values of the input signal at the time instant and before , constructing a standard input signal vector ; concatenating the standard input signal vector with its conjugate vector to obtain an augmented input signal vector of length ; The augmented input signal vector is equally divided into sub-input vectors with length The equally divided augmented input signal vector is denoted as ; Based on the divided augmented input signal vector, the following equation is constructed input matrix at time t ; wherein denotes a conjugate operation, denotes a conjugate transpose operation.

6. The wide linear, anti-impulse noise interference adaptive filter of claim 1, wherein, Computing the difference between the desired signal and the estimated signal at the time instant, obtaining an estimation error signal at the time instant, denoted as: ； wherein represents an estimation error signal of the time instant, represents an expected signal of the time instant.

7. The wide linear, anti-impulse noise interference adaptive filter of claim 1, wherein, When applied to a stereo video communication echo cancellation system, comprising: The input signal is a complex signal constructed by taking two near-end loudspeaker signals in the stereo video communication echo cancellation system as the real part and the imaginary part, respectively; The desired signal is a complex signal constructed by taking two near-end microphone signals in the stereo video communication echo cancellation system as the real part and the imaginary part, respectively.

Images