Quantized perceptual wide linear minimum error mode adaptive filter

Abstract

This invention relates to the field of digital filter technology and discloses a quantization-aware wide-linearity minimum error modulus adaptive filter. It utilizes quantization parameters to perform non-uniform low-bit quantization on the input signal and the desired signal to obtain the corresponding quantized values; a linearity factor is constructed based on the quantization parameters; a normalized correlation matrix between the input signal and the linearized error signal is constructed based on the non-circularity of the input signal, and a compensation matrix is ​​constructed in conjunction with the linearity factor; the quantized values ​​of the input signal at the current and historical times are obtained to construct an augmented input signal vector, and its product with the compensation matrix is ​​calculated to construct a corrected input vector; wide-linearity filtering is performed on the corrected input vector to obtain the output signal, and the error signal is calculated in conjunction with the quantized value of the desired signal; the weight vector at the next time step is updated using the corrected input vector and the error signal. This invention significantly reduces the cost and energy consumption of the analog-to-digital converter by introducing non-uniform low-bit quantization and a bias compensation mechanism, while maintaining a low steady-state error.

Description

Technical Field

[0001] This invention relates to the field of digital filter technology, and in particular to a quantization-sensing wide linear minimum error mode adaptive filter. Background Technology

[0002] System identification is an important branch of adaptive signal processing. Many problems, such as traditional adaptive channel equalization, adaptive noise cancellation, adaptive echo cancellation, and active noise control, can be reduced to system identification problems. Wide-linear adaptive filters are a type of complex system identification model. Their characteristic is that the filter simultaneously filters both the complex input signal vector and its conjugate vector. This fully utilizes the second-order non-circular characteristics of complex signals, overcoming the limitation of strictly linear adaptive filters that can only handle circular complex signals. Therefore, it shows great promise for applications in stereo echo cancellers.

[0003] Wide linear least mean square (WL-LMS) adaptive filters are commonly used wide linear adaptive filters. Based on the minimum mean square error criterion, they exhibit good convergence performance in noisy environments that approximately conform to a Gaussian model. However, their performance deteriorates sharply or even diverges when severe non-Gaussian noise, such as impulse noise, is present, rendering them unrobust. To address this, some robust wide linear adaptive filters have been proposed, such as the wide linear maximum correlation entropy adaptive filter. These filters introduce nonlinear processing of the error signal, which can suppress the destructive effect of impulse noise. However, the computational cost of the nonlinear function is relatively high, hindering many low-cost, miniaturized applications. Research in real-number adaptive filters shows that using the minimum absolute error criterion can improve the effectiveness of impulse noise resistance with a smaller computational cost.

[0004] Most current adaptive filters assume that the input and desired signals are accurate. However, in real-world applications, high-precision signals cannot always be obtained, especially for mobile and miniaturized devices. High-precision analog-to-digital converters (ADCs) consume significant energy and increase equipment costs. Furthermore, as precision decreases, quantization errors affect the convergence results of adaptive filters. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the convergence deviation problem and the poor filtering robustness of the wide linear minimum mean square filter in the prior art when processing low bit quantized signals.

[0006] To address the aforementioned technical problems, this invention provides a quantization-sensing wide-linearity minimum error modulus adaptive filter, comprising:

[0007] Using quantization parameters, non-uniform low-bit quantization is performed on the input signal and the desired signal to obtain the quantized values ​​of the input signal and the desired signal.

[0008] Based on the non-circularity of the input signal and the preset identity matrix, a normalized correlation matrix of the input signal and a normalized correlation matrix of the linearized error signal are constructed.

[0009] Based on the quantization parameters, a linear factor is constructed; based on the linear factor, the normalized correlation matrix of the input signal, and the normalized correlation matrix of the linearized error signal, a compensation matrix is ​​constructed.

[0010] Get Time and Quantized values ​​of the input signal from multiple consecutive time steps prior to time step [time] are used to construct [the following]. The augmented input signal vector at time t;

[0011] calculate The product of the augmented input signal vector and the compensation matrix at time step [time] is obtained. The corrected input vector at time step; calculation The product of the augmented weight vector and the corrected input vector at time step 1 is obtained. The output signal at time; calculate the quantized value of the desired signal and The difference between the output signals at time points is obtained. Error signal at time;

[0012] based on The corrected input vector and error signal at time step, for Update the augmented weight vector at time step [time] to obtain [the data]. The augmented weight vector at time step.

[0013] Preferably, the acquisition of quantization parameters includes:

[0014] Obtain a sequence of auxiliary real Gaussian random variables with zero mean and unit variance;

[0015] Using the Lloyd-max algorithm, obtain the sequence of auxiliary real Gaussian random variables at a preset quantization bit depth. Quantization parameters below;

[0016] The quantization parameters include a quantization threshold sequence. With quantized label sequence .

[0017] Preferably, using quantization parameters, non-uniform low-bit quantization is performed on the input signal and the desired signal to obtain the quantized values ​​of the input signal and the desired signal, including:

[0018] The real and imaginary parts of the input signal or the desired signal are quantized separately to obtain the quantized values ​​of the real part and the imaginary part of the input signal or the desired signal.

[0019] The real and imaginary quantized values ​​of the input or desired signal are rewritten into complex number form to obtain the quantized value of the input signal or the quantized value of the desired signal.

[0020] The quantization of the real or imaginary part is expressed as: if Then obtain the corresponding quantized value. ; Indicates the real part or the imaginary part. express Standard deviation; Represents the quantization threshold sequence of the th One quantization threshold, ; Indicates the quantization label sequence of the 1st generation. A quantitative label, .

[0021] Preferably, based on the non-circularity of the input signal and a preset identity matrix, a normalized correlation matrix of the input signal is constructed, including:

[0022] Based on the variance of the input signal With pseudovariance Calculate the non-circularity of the input signal , is represented as: ;

[0023] Based on the non-circularity of the input signal and the preset identity matrix Construct the normalized correlation matrix of the input signal. , is represented as:

[0024] ;

[0025] in, The conjugate of the non-circularity of the input signal. express The identity matrix, This indicates the length of the standard input signal vector.

[0026] Preferably, based on the non-circularity of the input signal and a preset identity matrix, a normalized correlation matrix for the linearized error signal is constructed, including:

[0027] Based on the variance of the input signal With pseudovariance Calculate the non-circularity of the input signal , is represented as: ;

[0028] Non-circularity of the input signal Perform the real part operation to obtain Construct the normalized correlation matrix of the linearized error signal. , is represented as:

[0029] ;

[0030] in, This indicates the operation of taking the real part of a complex number.

[0031] Preferably, a linear factor is constructed based on the quantization parameters, expressed as:

[0032] ;

[0033] in, Indicates a linear factor. The quantization threshold sequence representing the quantization parameter is the first... One quantization threshold, The quantization label sequence representing the quantization parameter is the first... A quantitative label, , Indicates the preset quantization bit depth. This indicates exponentiation.

[0034] Preferably, a compensation matrix is ​​constructed based on the linearity factor, the normalized correlation matrix of the input signal, and the normalized correlation matrix of the linearized error signal, as follows:

[0035] ;

[0036] in, This represents the normalized correlation matrix of the input signal. This represents the normalized correlation matrix of the linearized error signal. This indicates the conjugate transpose operation.

[0037] Preferably, obtain Time and Quantized values ​​of the input signal from multiple consecutive time steps prior to time step [time] are used to construct [the following]. The augmented input signal vector at time t includes:

[0038] Get Quantized value of the input signal at time 1 and the preceding consecutive Quantized value of the input signal at time 1 , build Standard input signal vector at time 1 , is represented as: ;

[0039] Will Standard input signal vector at time 1 Its conjugate vector splicing, constructing a length of of Augmented input signal vector at time step ;

[0040] in, This indicates the transpose operation. This indicates the conjugate operation. This indicates the conjugate transpose operation.

[0041] Preferably, calculation The product of the augmented weight vector and the corrected input vector at time step 1 is obtained. The output signal at time t is represented as:

[0042] ;

[0043] in, express The output signal at time, express Augmented weight vector at time step The conjugate transpose of; express The corrected input vector at time step is expressed as follows: , Represents the compensation matrix. express The augmented input signal vector at time t.

[0044] Preferably, based on The corrected input vector and error signal at time step, for Update the augmented weight vector at time step [time] to obtain [the data]. The augmented weight vector at time step 1 is expressed as:

[0045] ;

[0046] in, express The augmented weight vector at time step 1. Indicates the step size parameter; express The error signal at time t is expressed as: , express The expected quantized value of the signal at time; This represents the modulo operation for complex numbers. express . conjugate.

[0047] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0048] The quantization-sensing wide-linear minimum error modulus adaptive filter described in this invention introduces the minimum error modulus criterion into a wide-linear adaptive filter and uses deviation compensation to reduce the convergence deviation caused by low-bit quantization, achieving robust adaptive filtering of the output signal of the system to be estimated. This invention utilizes a wide-linear model to design an augmented complex adaptive filter, enabling dual-channel echo cancellation with a single filter. Simultaneously, by selecting a cost function based on the minimum error modulus, this invention achieves robust filtering results with relatively low computational complexity even when the desired signal is superimposed with impulse interference. Furthermore, the deviation compensation mechanism significantly reduces the convergence deviation caused by quantization errors, allowing the filter to effectively filter low-bit quantized signals, thereby reducing the cost of the analog-to-digital converter and system power consumption, and effectively improving the robustness of the adaptive filter in impulse noise environments. Attached Figure Description

[0049] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0050] Figure 1 This is a flowchart of the steps of the quantization-sensing wide linear minimum error mode adaptive filter provided by the present invention;

[0051] Figure 2 This is a structural block diagram of the coarse quantization system identification model;

[0052] Figure 3 This is a comparison chart of the normalized mean square deviation curves of the adaptive filtering system in the system identification scenario. Detailed Implementation

[0053] 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.

[0054] Reference Figure 1 The flowchart shown illustrates the steps of the quantization-sensing wide linear minimum error modulus adaptive filter provided by this invention. The specific steps include:

[0055] S101: Using quantization parameters, perform non-uniform low-bit quantization on the input signal and the desired signal to obtain the quantized values ​​of the input signal and the desired signal.

[0056] S102: Based on the non-circularity of the input signal and the preset identity matrix, construct the normalized correlation matrix of the input signal and the normalized correlation matrix of the linearized error signal, including:

[0057] S102-1: Based on the variance of the input signal With pseudovariance Calculate the non-circularity of the input signal , is represented as: ;

[0058] S102-2: Non-circularity of input signal and preset identity matrix Construct the normalized correlation matrix of the input signal. , is represented as:

[0059] ;

[0060] S102-3: Non-circularity of the input signal Perform the real part operation to obtain Construct the normalized correlation matrix of the linearized error signal. , is represented as:

[0061] ;

[0062] in, The conjugate of the non-circularity of the input signal. express The identity matrix, This represents the length of the standard input signal vector; This indicates the operation of taking the real part of a complex number;

[0063] S103: Construct a linear factor based on quantization parameters; construct a compensation matrix based on the linear factor, the normalized correlation matrix of the input signal, and the normalized correlation matrix of the linearized error signal.

[0064] linear factor , is represented as: ; The quantization threshold sequence representing the quantization parameter is the first... One quantization threshold, The quantization label sequence representing the quantization parameter is the first... A quantitative label, , Indicates the preset quantization bit depth. Indicates exponentiation;

[0065] Compensation matrix , is represented as: ; This represents the normalized correlation matrix of the input signal. This represents the normalized correlation matrix of the linearized error signal. This represents the conjugate transpose operation;

[0066] S104: Acquisition Time and Quantized values ​​of the input signal from multiple consecutive time steps prior to time step [time] are used to construct [the following]. The augmented input signal vector at time t includes:

[0067] S104-1: Obtain Quantized value of the input signal at time 1 and the preceding consecutive Quantized value of the input signal at time 1 , build Standard input signal vector at time 1 , is represented as: ;

[0068] S104-2: Will Standard input signal vector at time 1 Its conjugate vector splicing, constructing a length of of Augmented input signal vector at time step ;

[0069] in, This indicates the transpose operation. This indicates the conjugate operation. This represents the conjugate transpose operation;

[0070] S105: Calculation The product of the augmented input signal vector and the compensation matrix at time step [time] is obtained. The corrected input vector at time step; calculation The product of the augmented weight vector and the corrected input vector at time step 1 is obtained. The output signal at time; calculate the quantized value of the desired signal and The difference between the output signals at time points is obtained. Error signal at time;

[0071] S106: Based on The corrected input vector and error signal at time step, for Update the augmented weight vector at time step [time] to obtain [the data]. Augmented weight vector at time step , is represented as:

[0072] ;

[0073] in, Indicates the step size parameter; express The error signal at time t is expressed as: , express The expected quantized value of the signal at time; This represents the modulo operation for complex numbers. express . conjugate.

[0074] Specifically, in step S101, the acquisition of the input signal quantization value and the desired signal quantization value includes:

[0075] S101-1: Acquisition of quantization parameters, including:

[0076] Obtain a sequence of auxiliary real Gaussian random variables with zero mean and unit variance;

[0077] Using the Lloyd-max algorithm, obtain the sequence of auxiliary real Gaussian random variables at a preset quantization bit depth. Quantization parameters below;

[0078] The quantization parameters include a quantization threshold sequence. With quantized label sequence ;

[0079] S101-2: Non-uniform low-bit quantization, including:

[0080] The real and imaginary parts of the input signal or the desired signal are quantized separately to obtain the quantized values ​​of the real part and the imaginary part of the input signal or the desired signal.

[0081] The real and imaginary quantized values ​​of the input or desired signal are rewritten into complex number form to obtain the quantized value of the input signal or the quantized value of the desired signal.

[0082] The quantization of the real or imaginary part is expressed as: if Then obtain the corresponding quantized value. ; Indicates the real part or the imaginary part. express Standard deviation; Represents the quantization threshold sequence of the th One quantization threshold, ; Indicates the quantization label sequence of the 1st generation. A quantitative label, .

[0083] This invention introduces quantization parameters to perform low-bit quantization on the input signal and the desired signal, enabling the adaptive filter to process the low-bit quantized input signal and the desired signal with second-order non-circular characteristics, thereby reducing the cost of the analog-to-digital converter and the system power consumption.

[0084] Specifically, step S105 includes:

[0085] S105-1: Calculation The product of the augmented input signal vector and the compensation matrix at time step [time] is obtained. Corrected input vector at time step , is represented as: , Represents the compensation matrix. express The augmented input signal vector at time t;

[0086] S105-2: Calculation The product of the augmented weight vector and the corrected input vector at time step 1 is obtained. Output signal at time , is represented as: ; express Augmented weight vector at time step The conjugate transpose of;

[0087] S105-3: Calculate the quantization value of the desired signal and Output signal at time The difference, obtain Error signal at time , is represented as: .

[0088] The quantization-sensing wide-linear minimum error modulus adaptive filter described in this invention introduces the minimum error modulus criterion into a wide-linear adaptive filter and uses deviation compensation to reduce the convergence deviation caused by low-bit quantization, achieving robust adaptive filtering of the output signal of the system to be estimated. This invention utilizes a wide-linear model to design an augmented complex adaptive filter, enabling dual-channel echo cancellation with a single filter. Simultaneously, by selecting a cost function based on the minimum error modulus, this invention achieves robust filtering results with relatively low computational complexity even when the desired signal is superimposed with impulse interference. Furthermore, the deviation compensation mechanism significantly reduces the convergence deviation caused by quantization errors, allowing the filter to effectively filter low-bit quantized signals, thereby reducing the cost of the analog-to-digital converter and system power consumption, and effectively improving the robustness of the adaptive filter in impulse noise environments.

[0089] Based on the above embodiments, in this embodiment of the invention, filtering is performed using the quantization-sensing wide linear minimum error modulus adaptive filter provided by the present invention. Specific steps include:

[0090] S201: Calculate the optimal quantization parameters using the Lloyd-Max algorithm for the input signal. With expected signal Perform non-uniform low-bit quantization to obtain the quantized value of the input signal. and the expected signal quantization value ,include:

[0091] S201-1: Generate a sequence of auxiliary real Gaussian random variables with zero mean and unit variance, and calculate its quantization bit depth using the Lloyd-max algorithm. The optimal quantization parameters, including the quantization threshold. and quantification labels ;

[0092] S201-2: For the input signal and expected signal Quantize the real and imaginary parts separately:

[0093] set up express or The standard deviation of the real or imaginary part is . ,if Then its quantized value is obtained. ;

[0094] The quantized real and imaginary parts are then rewritten as complex numbers to obtain the quantized value of the input signal. and the expected signal quantization value ;

[0095] S202: Obtain the continuous sequence at time n and before time n. The low-bit quantized values ​​of the input signal at each time step are used to construct the augmented input signal vector. ,include:

[0096] S202-1: Obtain Time and preceding consecutive Quantized value of the input signal at time 1 Construct the standard input signal vector ;

[0097] S202-2: Convert the standard input signal vector Its conjugate vector splicing, constructing a length of augmented input signal vector ;

[0098] in, This indicates the transpose operation. This indicates the conjugate operation. This represents the conjugate transpose operation;

[0099] S203: Non-circularity based on input signal Construct the normalized correlation matrix of the input signal. Normalized correlation matrix of linearized error signal , respectively represented as:

[0100] ;

[0101] ;

[0102] in, This indicates the operation of taking the real part of a complex number. Indicates the input signal Non-circularity, and They represent variance and pseudo-variance express The identity matrix;

[0103] S204: Calculate the linear factor based on the quantization parameters, and then construct the compensation matrix. ,include:

[0104] S204-1: Based on the quantization threshold and quantization label, a linear factor is constructed, expressed as:

[0105] ;

[0106] S204-2: Based on the linearity factor, the normalized correlation matrix of the input signal, and the normalized correlation matrix of the linearized error signal, a compensation matrix is ​​constructed, expressed as:

[0107] ;

[0108] in, Indicates exponentiation;

[0109] S205: Based on the compensation matrix and augmented input signal vector The product of these two elements is used to construct the corrected input vector. , is represented as: ;

[0110] S206: Perform wide linear filtering on the corrected input vector to obtain the output signal;

[0111] That is, the output signal is calculated based on the inner product of the augmented weight vector and the corrected input vector. , is represented as: ;

[0112] in, express The augmented weight vector of the time-wide linear adaptive filter;

[0113] S207: Calculate the error signal using the difference between the coarsely quantized desired signal value and the output signal. , is represented as: ;

[0114] S208: Utilizing the corrected input vector Sum of error signals Update the weight vector at the next time step. , is represented as:

[0115] ;

[0116] in, Indicates the step size parameter. Represents the modulo operation for complex numbers.

[0117] The quantization-sensing wide-linear minimum error modulus adaptive filter of this invention enables the adaptive filter to process low-bit quantized input signals and desired signals with second-order non-circular characteristics through deviation compensation, thereby reducing the power consumption and hardware cost of the device. In addition, the weight update formula derived based on the minimum error modulus criterion achieves the adaptive filter's ability to resist impulse noise interference with low computational complexity.

[0118] To demonstrate the effectiveness of this invention, this embodiment employs computer experiments to verify the performance of the Quantization-Aware Wide Linear Minimum Error Mode (QA-WL-LEM) adaptive filter provided by this invention. The experiment targets a system identification application scenario, estimating an unknown system in an environment containing impulse noise interference, and comparing the results with those of the Quantization-Aware Wide Linear Minimum Mean Square (QA-WL-LMS) adaptive filter. The Quantization-Aware Wide Linear Minimum Mean Square (QA-WL-LMS) adaptive filter is a result of extending the existing Quantization-Aware Minimum Mean Square (QA-LMS) adaptive filter method to a wide linear scenario.

[0119] Reference Figure 2 The diagram shown is a structural block diagram of the coarse-quantization system identification model. In this embodiment, the noise signal is Gaussian noise plus impulse noise; the normalized mean square deviation (NMSD) is used as the performance measure in the system identification scene experiment. The unit is dB, where Indicates taking the logarithm. These are the weights of the actual system.

[0120] In the experiment, the weight vector of the unknown system was randomly generated, and the input signal used had a variance of 1 and a non-circularity of [value missing]. The zero-mean Gaussian white signal, using the noise signal It contains a mean of zero and a variance of . Gaussian white noise and a pulse noise ,Right now The noise-free desired signal and Gaussian noise are preserved. The signal-to-noise ratio in dB remains consistent with the impulse noise. Signal-to-noise ratio in dB. 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.

[0121] 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 NMSD curves are obtained by averaging 100 Monte Carlo experiments. The step size parameter for each method is QA-WL-LMS(…). ), QA-WL-LEM ( ).Depend on Figure 3 As can be seen, the QA-WL-LEM adaptive filtering system of this application can maintain a lower steady-state error and has good anti-pulse interference capability and deviation compensation effect.

[0122] The quantization-sensing wide-linear minimum error modulus adaptive filter described in this invention introduces the minimum error modulus criterion into the wide-linear adaptive filter and uses deviation compensation to reduce the convergence deviation caused by low-bit quantization, achieving robust adaptive filtering of the output signal of the system to be estimated. By introducing low-bit non-uniform quantization and deviation compensation mechanisms, this invention significantly reduces the cost of the analog-to-digital converter and system energy consumption while maintaining a low steady-state error, and effectively improves the robustness of the adaptive filter in impulse noise environments. It is suitable for parameter estimation scenarios such as the Internet of Things (IoT) where energy efficiency and anti-interference performance are critical. This invention improves the robustness of the adaptive filter in environments containing impulse noise with relatively low computational complexity by minimizing the cost function based on the error modulus. Simultaneously, based on a wide-linear framework, this invention can fully utilize the second-order non-circular characteristics of complex signals to improve filtering capability and achieve better filtering results.

[0123] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0125] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0126] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0127] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A quantized perceptually wide linear minimum error modeling adaptive filter characterized by, The method comprises the following steps: quantizing the input signal and the expected signal by using a quantization parameter to obtain quantized values of the input signal and the expected signal; constructing an input signal normalized correlation matrix and a linearized error signal normalized correlation matrix based on the non-circularity of the input signal and a preset unit matrix; Based on the quantization parameter, a linear factor is constructed and expressed as: ; wherein, represents the linear factor, represents the qth quantization threshold in a quantization threshold sequence of the quantization parameter, represents the qth quantization label in a quantization label sequence of the quantization parameter, represents the qth quantization threshold in a quantization threshold sequence of the quantization parameter, represents the qth quantization label in a quantization label sequence of the quantization parameter, , represents a preset quantization bit number, represents an exponential operation; Based on the linear factor, the input signal normalized correlation matrix and the linearized error signal normalized correlation matrix, a compensation matrix is constructed, denoted as: ; wherein, denotes the input signal normalized correlation matrix, denotes the linearized error signal normalized correlation matrix, denotes the conjugate transpose operation; Get Time and Quantized values ​​of the input signal from multiple consecutive time steps prior to time step [time] are used to construct [the following]. The augmented input signal vector at time t; Computing the product of the augmented input signal vector at the time instant and the compensation matrix, obtaining the modified input vector at the time instant; computing the product of the augmented weight vector at the time instant and the modified input vector, obtaining the output signal at the time instant; computing the difference between the desired signal quantization value and the output signal at the time instant, obtaining the error signal at the time instant; Based on The correction input vector of the moment is updated with the error signal, and the augmented weight vector of the moment is obtained The augmented weight vector of the moment is updated with the error signal, and the augmented weight vector of the moment is obtained The augmented weight vector of the moment is updated with the error signal, and the augmented weight vector of the moment is obtained 2. The quantized perceptual wide linear minimum error modeling adaptive filter according to claim 1, characterized in that, the quantization parameter is obtained by: obtaining a sequence of auxiliary real Gaussian random variables with zero mean and unit variance; The Lloyd-max algorithm is used to obtain quantization parameters of the auxiliary real Gaussian random variable sequence under a preset quantization bit number . The quantization parameter includes a quantization threshold sequence With the quantization label sequence .

3. The quantized perceptual wide linear minimum error modeling adaptive filter according to claim 2, characterized in that, quantizing the input signal and the expected signal by using a quantization parameter to obtain quantized values of the input signal and the expected signal, comprising: quantizing the real part and the imaginary part of the input signal or the expected signal respectively to obtain quantized values of the real part and the imaginary part of the input signal or the expected signal; rewriting the quantized values of the real part and the imaginary part of the input signal or the expected signal into complex form to obtain quantized values of the input signal or the expected signal; The quantization of the real or imaginary part is expressed as: if Then obtain the corresponding quantized value. ; Indicates the real part or the imaginary part. express Standard deviation; Represents the quantization threshold sequence of the th One quantization threshold, ; Indicates the quantization label sequence of the 1st generation. A quantitative label, .

4. The quantized perceptual wide linear minimum error modeling adaptive filter according to claim 1, wherein, constructing an input signal normalized correlation matrix based on the non-circularity of the input signal and a preset unit matrix, comprising: Based on the variance of the input signal With the pseudo-variance , the non-circularity of the input signal is calculated , is expressed as: ; Non-circularity based on input signal and preset unit matrix , constructing an input signal normalized correlation matrix , is expressed as: ； wherein denotes the conjugate of the non-circularity of the input signal, denotes the identity matrix, denotes the length of the standard input signal vector.

5. The quantized perceptual wide linear minimum error modeling adaptive filter according to claim 1, wherein, constructing a linearized error signal normalized correlation matrix based on the non-circularity of the input signal and a preset unit matrix, comprising: Based on the variance of the input signal With the pseudo-variance , the non-circularity of the input signal is calculated , is expressed as: ; Non-circularity of an input signal a real part operation is performed to obtain , a linearized error signal normalized correlation matrix is constructed is expressed as: ； wherein represents taking the real part of a complex number.

6. The quantized perceptual wide-linear minimum error modeling adaptive filter according to claim 1, wherein, Get Time and Quantized values ​​of the input signal from multiple consecutive time steps prior to time step [time] are used to construct [the following]. The augmented input signal vector at time t includes: acquiring quantized values of the input signal at the time , and the preceding quantized values of the input signal at the time , to construct a standard input signal vector at the time , denoted as: ; The standard input signal vector at time instant and its conjugate vector splicing, constructing a length of augmented input signal vector at time instant ; wherein, denotes a transpose operation, denotes a conjugate operation, denotes a conjugate transpose operation.

7. The quantized perceptual wide-linear minimum error modeling adaptive filter according to claim 1, wherein, Computing the product of the augmented weight vector at the time instant and the correction input vector, obtaining the output signal at the time instant, denoted as: ； wherein denotes the output signal at time instant denotes the augmented weight vector at time instant the conjugate transpose of denotes the modified input vector at time instant , denotes the compensation matrix denotes the augmented input signal vector at time instant 8. The quantized perceptual wide linear minimum error modeling adaptive filter according to claim 7, characterized in that, Based on The modified input vector at the time t is updated by the error signal, and the augmented weight vector at the time t is updated, to obtain The augmented weight vector at the time t is updated by the error signal, and the augmented weight vector at the time t is updated, to obtain The augmented weight vector at the time t is updated by the error signal, and the augmented weight vector at the time t is updated, to obtain ； wherein denotes augmented weight vector at time instant denotes a step size parameter; denotes error signal at time instant , denotes expected signal quantization value at time instant denotes the modulo operation on complex numbers, denotes the conjugate of

Images