Adaptive identification system, adaptive identification device and adaptive identification method

Abstract

In an adaptive identification system, an adaptive identification device and an adaptive identification method, the same identification accuracy as that of an approximate steepest descent algorithm can be obtained with less computational effort than that of an approximate steepest descent algorithm. The adaptive identification system uses an adaptive filter to identify the characteristics of a transmission system, and comprises: a signal generation unit, which generates a periodic identification input signal that includes frequency components that are integer multiples of a fundamental frequency and satisfies a continuous excitation condition; a setting unit, which sets a moving average time to the fundamental period of the identification input signal; and an adaptive algorithm execution unit, which uses a moving average and a diagonalization matrix to update the coefficients of the adaptive filter, wherein the moving average is a value obtained by moving average the mutual correlation vector of the identification input signal vector and the observation signal using the moving average time, and the diagonalization matrix is ​​a matrix obtained by diagonalizing a matrix obtained by moving average the autocorrelation matrix of the identification input signal vector using the moving average time.

Description

Technical Field

[0001] The invention relates to an adaptive identification system, an adaptive identification device and an adaptive identification method. Background Art

[0002] For example, when audio signals are processed for the purpose of original sound reproduction, stereophonic sound, noise reduction, etc., the characteristics of the sound transmission system from the speaker to the microphone set at the control point (ideally, the listener's ear) are used.

[0003] In addition, as one of the representative methods for obtaining the characteristics of the sound transmission system, there is a technology called adaptive identification. This adaptive identification is a technology for modeling the sound transmission system by an adaptive filter, and the LMS (Least Mean Square) algorithm is widely used as an adaptive algorithm (for example, refer to Patent Document 1).

[0004] Patent Document 1: Japanese Patent Application Publication No. 2007-281860

[0005] When adaptive identification is performed in an actual environment, the identification accuracy may be reduced due to interference such as noise. In the case of the LMS algorithm, the reduction in identification accuracy can be avoided by setting the step size small, but if the step size is reduced, the adaptive speed decreases, so there is a problem that it takes time to obtain the final identification result.

[0006] In order to solve such a problem, for example, an adaptive algorithm (hereinafter referred to as an approximate steepest descent algorithm) can be used to approximate the calculation of the expected value in the steepest descent algorithm by a finite number of time averages to shorten the time until the final identification result is obtained. However, in this method, there is a problem that the amount of calculation of the adaptive identification system is greatly increased compared to the LMS algorithm. Summary of the invention

[0007] One embodiment of the present invention has been made in view of the above-mentioned problem, and in an adaptive identification system, it is possible to obtain identification accuracy equivalent to that of the approximate steepest descent algorithm with less computational effort than the approximate steepest descent algorithm.

[0008] In order to solve the above-mentioned problem, an adaptive identification system according to one embodiment of the present invention is an adaptive identification system that uses an adaptive filter to identify the characteristics of a transmission system, and comprises: a signal generating unit that generates a periodic identification input signal that includes frequency components that are integer multiples of a basic frequency and satisfies a continuous excitation condition; a setting unit that sets a moving average time to the basic period of the above-mentioned identification input signal; and an adaptive algorithm executing unit that uses a moving average value and a diagonalization matrix to update the coefficients of the above-mentioned adaptive filter, wherein the above-mentioned moving average value is a value obtained by moving average the cross-correlation vector of an observed signal (the identification input signal on which the above-mentioned transmission system is superimposed including interference) and the above-mentioned identification input signal vector using the above-mentioned moving average time, and the above-mentioned diagonalization matrix is ​​a matrix obtained by diagonalizing a matrix obtained by moving average the autocorrelation matrix of the above-mentioned identification input signal vector using the above-mentioned moving average time.

[0009] Effects of the Invention

[0010] According to an embodiment of the present invention, in an adaptive identification system, it is possible to obtain identification accuracy equivalent to that of an approximate steepest descent algorithm with less computational complexity than that of an approximate steepest descent algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1A and Figure 1B This is a diagram showing an example of a system configuration of an adaptive identification system according to an embodiment.

[0012] Figure 2 This is a diagram showing an example of a functional configuration of a control unit according to one embodiment.

[0013] Figure 3 This is a diagram showing an image of a signal waveform of an identification input signal according to one embodiment.

[0014] Figure 4 Detailed description of the invention is a flowchart showing an example of processing of an adaptive identification system according to an embodiment.

[0015] Figure 5 This is a diagram showing an example of the amount of calculation of an adaptive identification system according to one embodiment.

[0016] Figure 6 This is a diagram for explaining a conventional adaptive identification system.

[0017] Figure 7 This is a diagram for explaining an example of a conventional adaptive algorithm.

[0018] Fig. 8A and Figure 8B FIG. 1 is a diagram showing an example of a system configuration of a conventional adaptive identification system.

[0019] Fig. 9 FIG. 1 is a diagram showing an example of the amount of calculation of a conventional adaptive identification system.

[0020] Description of symbols

[0021] 100 Adaptive Identification System

[0022] 110 Adaptive Identification Device

[0023] 111 Adaptive Filter

[0024] 112 Adaptive algorithm execution unit

[0025] 115 Storage

[0026] 201 Signal Generation Department

[0027] 202 Setting Department

[0028] 203 Computing Department DETAILED DESCRIPTION

[0029] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0030] <About Adaptive Identification System>

[0031] Before describing the adaptive identification system according to the present embodiment, a conventional adaptive identification system will be briefly described.

[0032] Figure 6 1 is a diagram for explaining a conventional adaptive identification system. The adaptive identification system 1 is a system for identifying the characteristics of a transmission system 11 of, for example, sound, using an adaptive identification device 10. For example, the adaptive identification device 10 adaptively identifies the characteristics of the transmission system 11 based on an identification input signal x(n) and an output signal d(n) output from the transmission system 11 by a predetermined adaptive algorithm.

[0033] However, when adaptive identification is performed in an actual environment, interference such as noise v(n) is mixed into the observation signal y(n) input to the adaptive identification device 10, which causes a problem that the identification accuracy of the system decreases.

[0034] Figure 7 1 is a diagram for explaining an example of a conventional adaptive algorithm. In a conventional adaptive identification system 1, the LMS (Least Mean Square) algorithm is widely used as an adaptive algorithm.

[0035] The LMS algorithm is an adaptive algorithm based on the steepest descent method, which is a classical numerical analytical method used to find the minimum value of the function f(w) of the vector w. Figure 7 , which represents the formula of the adaptive algorithm based on the steepest descent method (steepest descent algorithm).

[0036] In the formula of the steepest descent algorithm, the operation E[] represents an expected value operation for finding the average value of infinite trials. Therefore, it is difficult to implement the formula of the steepest descent algorithm in the adaptive identification device 10 as an adaptive algorithm.

[0037] Therefore, as an algorithm that can be implemented in the adaptive identification device 10 , an expression of the LMS algorithm in which the expected value calculations 20 and 21 of the steepest descent algorithm are replaced with instantaneous values ​​22 and 23 is widely used.

[0038] As another algorithm that can be implemented in the adaptive identification device 10 , there is known an adaptive algorithm (hereinafter referred to as an approximate steepest descent algorithm) that approximates the expected value calculations 20 and 21 of the steepest descent algorithm by a finite number of time averages 24 and 25 .

[0039] (Using LMS's adaptive identification system)

[0040] Fig. 8A FIG. 1 shows an example of a system configuration of an adaptive identification system 1 using an LMS algorithm. Fig. 8A In the example of FIG. 1 , the adaptive identification device 10 includes an adaptive filter 31 and an adaptive algorithm execution unit 32. The adaptive algorithm execution unit 32 will Figure 7 The transfer function w(n) of the adaptive filter 31 is updated using the formula of the LMS algorithm described in .

[0041] In such an adaptive identification system 1 using the LMS algorithm, when there is no noise v(n) as interference and the transfer system 11 can be completely simulated by the transfer function w(n) of the adaptive filter 31, the adaptive filter 31 converges to the characteristics of the transfer system 11. That is, w(∞)=h.

[0042] However, when there is noise v(n) as interference, the adaptive filter 31 is affected by it and does not converge to the optimal solution but vibrates around the optimal solution. And the larger the noise v(n) as interference is, the larger the vibration is.

[0043] In such a case, in the adaptive identification system 1 using the LMS algorithm, the vibration can be suppressed to a smaller value by setting the step size μ smaller. On the other hand, if the step size μ is set smaller, the adaptive speed decreases, so there is a problem that more time is required to obtain the final result.

[0044] (Adaptive identification system using approximate steepest descent algorithm)

[0045] Figure 8B FIG. 1 shows an example of a system configuration of an adaptive identification system 1 using an approximate steepest descent algorithm. Figure 8B In the example of FIG. 1 , the adaptive identification device 10 includes an adaptive filter 33 and an adaptive algorithm execution unit 34. The adaptive algorithm execution unit 34 will Figure 7 The approximate steepest descent algorithm described in is used as an adaptive update formula to update the transfer function w(n) of the adaptive filter 31.

[0046] In such an adaptive identification system 1 using the approximate steepest descent algorithm, even if the step size μ is not set small, the influence of the noise v(n) as interference can be suppressed by time averaging, so the time until the final result is obtained can be shortened. However, in the adaptive identification system 1 using the approximate steepest descent algorithm, there is a problem that more calculations are required compared to the adaptive identification system 1 using the LMS algorithm.

[0047] Fig. 9 : is a diagram showing an example of the amount of calculation of the adaptive identification system. Here, for ease of explanation, the number of calculations of the LMS algorithm and the number of calculations of the approximate steepest descent algorithm are shown when the number of taps of the adaptive filter is 2.

[0048] like Fig. 9 As shown, in the adaptive identification system 1 using the LMS algorithm, the number of calculations required for one adaptation is 2 multiplications for gradient estimation, 2 multiplications for step size adjustment, and 2 additions for adaptive update.

[0049] On the other hand, in the adaptive identification system 1 using the approximate steepest descent algorithm, the number of calculations required for one adaptation is 9 multiplications and 19 additions for gradient estimation, 2 multiplications for step size adjustment, and 2 additions for adaptive update. In addition, as an example, the calculation amount when the moving average addition process is a ring buffer method and three addition processes (data value addition, subtraction, pointer increment) are performed is shown here.

[0050] Furthermore, the adaptive filter actually used often has a tap number of 128 or more, for example. In such a case, the amount of calculation of the adaptive identification system 1 using the approximate steepest descent algorithm becomes even greater.

[0051] Therefore, in this embodiment, an adaptive identification system is described which realizes identification accuracy equivalent to that of the approximate steepest descent algorithm with a smaller amount of calculation than the approximate steepest descent algorithm.

[0052] <System Structure>

[0053] Figure 1A and Figure 1B This is a diagram showing an example of a system configuration of an adaptive identification system according to an embodiment.

[0054] Figure 1A An example of a system configuration of the adaptive identification system 100 according to the present embodiment is shown. Figure 1A The adaptive identification device 110 shown in FIG. Figure 8B In addition to the structure of the conventional adaptive identification device 10 shown, it further includes a control unit 114 and a storage unit 115. Specifically, the adaptive identification device 110 includes an adaptive filter 111, an adaptive algorithm execution unit 112, a subtractor 113, a control unit 114, a storage unit 115, and the like.

[0055] In addition, in the text of the specification, since characters representing vectors and matrices cannot be represented by bold characters, vectors and matrices are represented by standard characters in this text. On the other hand, in numerical expressions, vectors and matrices are represented by bold characters. Thus, for example, "vector x" in this text corresponds to the bold characters x in the numerical expression. The same is true for matrices and other characters.

[0056] The adaptive filter 111 is implemented by, for example, a DSP (Digital Signal Processor) included in the adaptive identification device 110 , and is a FIR (Finite Impulse Response) filter or the like whose coefficient vector w(n) can be changed under control from the adaptive algorithm execution unit 112 .

[0057] The adaptive algorithm execution unit 112 is realized by, for example, a DSP included in the adaptive identification device 110 , and updates the coefficient vector w(n) of the adaptive filter 111 using an adaptive update equation described later based on the identification input signal x(n) and the observation signal y(n).

[0058] The subtractor 113 is implemented by, for example, a DSP included in the adaptive identification device 110, and outputs an error signal e(n) based on the observation signal y(n) and the output signal of the adaptive filter 111. In the present embodiment, the error signal e(n) is not used directly for adaptive updating, but is observed to observe the degree of adaptive progress of how close the coefficient vector w(n) of the adaptive filter 111 is to the transfer system h11 to be obtained.

[0059] The control unit 114 is implemented by, for example, a program executed by a computer included in the adaptive identification device 110 or a DSP included in the adaptive identification device 110, and has the following functions: Figure 2 The functional structure of the control unit 114 will be described later.

[0060] The storage unit 115 is realized by, for example, a memory or a storage device included in the adaptive identification device 110 , and stores various information and data described below.

[0061] Figure 1B FIG. 2 shows another example of the system configuration of the adaptive identification system 100 according to the present embodiment. Figure 1B As shown, the control unit 114 and the storage unit 115 of the adaptive identification system 100 may also be a control device 120 external to the adaptive identification device 110. For example, the control device 120 has a computer structure including a CPU (Central Processing Unit), a memory, a storage device, a communication interface, etc., and the control unit 114, the storage unit 115, etc. are implemented by the CPU executing a prescribed program.

[0062] <Functional Structure of Control Unit>

[0063] Figure 2 1 is a diagram showing an example of a functional configuration of a control unit according to an embodiment. The control unit 114 includes, for example, a signal generation unit 201, a setting unit 202, and a calculation unit 203.

[0064] The signal generation unit 201 generates the identification input signal x(n) of the adaptive identification system 100. The signal generation unit 201 may generate the identification input signal x(n) within the signal generation unit 201 or may control the DSP included in the adaptive identification device 110 to generate the identification input signal x(n).

[0065] (Input signal for identification)

[0066] Here, the identification input signal vector x(n) generated by the signal generation unit 201 will be described.

[0067] As described above, the adaptive update formula of the approximate steepest descent algorithm is expressed by the following formula.

[0068] [Formula 1]

[0069]

[0070] In this approximate steepest descent algorithm, the autocorrelation matrix x(n)x(n) of the input signal vector x(n) for identification is calculated. TThe moving average of the autocorrelation matrix and the moving average of the cross-correlation vector x(n)y(n) between the identification input signal vector x(n) and the observation signal y(n) are calculated, and the product of the moving average of the autocorrelation matrix and the current adaptive filter coefficient vector w(n) is calculated. In addition, in the adaptive update formula, the adaptive filter coefficient vector at the next sampling is calculated by adding the moving average of the cross-correlation vector to the product of the moving average of the autocorrelation matrix and the value of the current adaptive filter coefficient vector.

[0071] Here, the autocorrelation matrix is ​​represented by the following equation.

[0072] [Formula 2]

[0073]

[0074] Since the matrix calculates the moving average of each element, the amount of calculation (computation) of the moving average becomes particularly large. Therefore, if the autocorrelation matrix can be made into a simple form, the amount of calculation can be greatly reduced.

[0075] Therefore, in order to simplify the autocorrelation matrix, the signal generation unit 201 generates a periodic identification input signal x(n) that includes frequency components that are integer multiples of the fundamental frequency and satisfies the PE (Persistently Exciting) condition.

[0076] For example, the signal generation unit 201 combines a plurality of sinusoidal signals of integer multiples of the fundamental frequency to generate a periodic signal containing only frequency components of integer multiples of the fundamental frequency as the identification input signal x(n). At this time, in order to satisfy the order of the PE property, the signal generation unit 201 adds sinusoidal signals of a number of frequencies corresponding to the number of taps of the adaptive filter 111 and having a coherent frequency under the condition of no interference to generate the identification input signal x(n).

[0077] Thus, the signal generation unit 201 can generate a periodic identification input signal x(n) that includes a frequency that is an integer multiple of the fundamental frequency and satisfies the PE property condition.

[0078] In addition, for the periodic identification input signal x(n) that satisfies the PE condition, by setting the basic period T of the basic frequency of the identification input signal x(n) as the time interval for calculating the moving average and performing the moving average of the autocorrelation matrix, a matrix of regularization constants can be obtained.

[0079] In the application to the sound system, it is preferable that the signal generating unit 201 changes the phases of the plurality of added sinusoidal signals so that the signal waveform of the identification input signal x(n) does not become a pulse shape, thereby generating the identification input signal x(n). For example, the signal generating unit 201 changes the phases of the plurality of sinusoidal signals randomly.

[0080] exist Figure 3 2 shows an image of a signal waveform of the identification input signal x(n) generated by the signal generation unit 201 .

[0081] Figure 3 FIG. 1 is a diagram showing an image of a signal waveform of an identification input signal according to one embodiment. Figure 3 In the example, for example, an example of an identification input signal x(n) is generated by setting the identification model to an 8-tap FIR filter, setting the basic frequency to 10 Hz, and synthesizing cosine waves of 30 Hz, 40 Hz, 50 Hz, 60 Hz, 70 Hz, 80 Hz, 90 Hz, and 100 Hz. Figure 3 In the example of , the signal generation unit 201 generates an identification input signal x(n) by synthesizing a plurality of frequency components (cosine waves) at random phases, thereby generating a periodic signal with a basic period T having no pulse-like peaks.

[0082] Here, back Figure 2 , the functional structure of the control unit 114 is further described.

[0083] The setting unit 202 sets the moving average time of the adaptive identification device 110 (moving average time of the adaptive algorithm) to the fundamental period T of the identification input signal x(n). For example, when the sampling frequency of the adaptive identification device 110 is 1600 Hz and the fundamental frequency of the identification input signal x(n) is 10 Hz, the setting unit 202 sets the moving average number M to M=160. Thus, the moving average time of the adaptive identification device 110 is set to the fundamental period of the fundamental frequency 10 Hz, that is, 0.1 s.

[0084] Thus, the autocorrelation matrix x(n)x(n) of the identification input signal vector x(n) is T The matrix obtained by performing the moving average is a matrix of canonical constants as represented by the following equation, for example.

[0085] [Formula 3]

[0086]

[0087] In order to reduce the amount of calculation, it is preferable that the calculation unit 203 calculates in advance the autocorrelation matrix x(n)x(n) of the identification input signal vector x(n) TThe matrix subjected to moving average is subjected to eigenvalue decomposition, and a diagonalized matrix subjected to diagonalization is calculated and stored in the storage unit 115 or the like.

[0088] <Adaptive Update>

[0089] Next, the adaptive update formula of the adaptive algorithm according to the present embodiment will be described.

[0090] For the adaptive update formula of the above approximate steepest descent algorithm, if the step size calculation and gradient calculation are transformed separately, it can be expressed as

[0091] [Formula 4]

[0092]

[0093] In addition, in the present embodiment, by using the identification input signal x(n) prepared to satisfy the above-mentioned PE property condition, it is possible to obtain the autocorrelation matrix x(n)x(n) of the identification input signal vector x(n) T The matrix obtained after moving average is normalized. In addition, the autocorrelation matrix after normalization can be diagonalized as follows by eigenvalue decomposition.

[0094] [Formula 5]

[0095]

[0096] Here, L is the number of taps of the identification model.

[0097] Therefore, the adaptive update formula of the adaptive algorithm of this embodiment can be expressed as

[0098] [Formula 6]

[0099]

[0100] Furthermore, if both sides of the equation are multiplied by the regular matrix P from the left, it becomes

[0101] [Formula 7]

[0102]

[0103] Furthermore, if we consider the area where the linear transformation based on the regular matrix P is performed, and assume that w'(n) = Pw(n), x'(n) = Px(n-1), then the formula can be expressed as

[0104] [Formula 8]

[0105]

[0106] Thus, according to the present embodiment, the autocorrelation matrix x(n)x(n) of the identification input signal vector x(n) is T The adaptive filter coefficient vector after constantization and linear transformation using the regularization matrix P is obtained by multiplication with the diagonalization matrix D composed of eigenvalues. Therefore, according to this embodiment, the identification accuracy equivalent to that of the approximate steepest descent algorithm can be achieved with less computational effort than the approximate steepest descent algorithm.

[0107] In addition, since the coefficient vector w'(n) of the adaptive filter converges in the area after linear transformation by the regular matrix P, when obtaining the actual impulse response,

[0108] [Formula 9]

[0109] w′(∞)=Pw opt Multiply both sides from the left by the inverse matrix P －1 And perform the inverse transformation to become

[0110] [Formula 10]

[0111] P -1 w′(∞)=w opt The optimal value w can be found opt .

[0112] <Processing Flow>

[0113] Next, the flow of processing of the adaptive identification method according to the present embodiment will be described.

[0114] Figure 4 1 is a flowchart showing an example of processing of an adaptive identification system according to an embodiment. This processing shows the overall flow of an adaptive identification method executed by the adaptive identification system 100 according to this embodiment.

[0115] In step S401 , the signal generation unit 201 of the control unit 114 generates a periodic identification input signal x(n) that combines frequency components that are integer multiples of the fundamental frequency so as to satisfy the PE property condition as described above.

[0116] For example, the signal generator 201 combines sinusoidal signals of frequencies that are integer multiples of a fundamental frequency (eg, 10 Hz) by the number of taps of the adaptive filter 111 with respect to the identification input signal x(n), thereby generating a periodic identification input signal x(n).

[0117] In addition, the signal generating unit 201 makes the phases of the plurality of sinusoidal signals different from each other so that the identification input signal x(n) does not become a pulse. For example, the signal generating unit 201 randomly changes the phases of the plurality of sinusoidal signals to generate Figure 3The periodic identification is shown using the input signal x(n).

[0118] Preferably, the signal generation unit 201 determines the identification input signal x(n) and obtains the matrix D and the matrix P. For example, when the number of taps of the adaptive filter 111 is 2,

[0119] [Formula 11]

[0120]

[0121]

[0122] Furthermore, the signal generation unit 201 considers the linear transformation based on the matrix P and prepares the identification input signal after the linear transformation.

[0123] [Formula 12]

[0124]

[0125]

[0126] The identification input signal after the linear transformation is stored in the storage unit 115 in the following format, for example.

[0127] [Formula 13]

[0128] x′0(n)=p 00 x0(n)+p 01 x1(n)

[0129] x′1(n)=p 10 x0(n)+p 11 x1(n)

[0130] Preferably, the signal generating unit 201 generates the identification input signal x(n) based on the identification input signal pre-stored in the storage unit 115, for example, Figure 1A As shown, the signal is input to the transmission system 11 , the adaptive filter 111 , and the adaptive algorithm execution unit 112 .

[0131] In step S402 , the setting unit 202 of the control unit 114 sets the moving average time of the adaptive identification device 110 (moving average time of the adaptive algorithm) to the fundamental period T of the identification input signal x(n) generated in step S201 .

[0132] For example, when the sampling frequency of the adaptive identification device 110 is 1600 Hz and the fundamental frequency of the identification input signal x(n) is 10 Hz, the moving average time of the adaptive identification device 110 is set to the fundamental period T of the fundamental frequency 10 Hz, that is, 0.1 s.

[0133] In step S403, the calculation unit 203 of the control unit 114 calculates the autocorrelation matrix x(n)x(n) for the identification input signal vector x(n) generated in step S401. T The matrix after moving average is diagonalized.

[0134] Note that this process may be performed before the adaptive update process of step S404. For example, the calculation unit 203 may calculate the diagonalization matrix and store it in the storage unit 115 in advance when generating the identification input signal x(n) and storing it in the storage unit 115.

[0135] In step S404, the adaptive algorithm execution unit 112 updates the coefficients of the adaptive filter 111 using the moving average of the moving average of the cross-correlation vector x(n)y(n) between the identification input signal vector x(n) and the observation signal y(n) and the diagonalization matrix calculated by the calculation unit 203. For example, when the number of taps of the adaptive filter 111 is 2,

[0136] [Formula 14]

[0137]

[0138]

[0139] As an adaptive update formula, an adaptive update process is performed.

[0140] Here, for example, assuming that the moving average addition process is performed by a ring buffer method and three addition processes (data value addition, subtraction, pointer increment) are performed, the number of calculations (amount of operations) required for one adaptation is as follows.

[0141] Gradient estimation: 6 multiplications, 8 additions

[0142] Step size adjustment: multiplication 2 times

[0143] Adaptive update: 2 additions

[0144] Figure 5 1 is a diagram showing an example of the amount of calculation of an adaptive identification system according to an embodiment. The diagram shows the amount of calculation of the approximate steepest descent algorithm and the amount of calculation of the present embodiment when the number of taps of the adaptive filter 111 is 2 taps, 128 taps, or L taps (L is an integer greater than 2).

[0145] like Figure 5 As shown, when the number of taps of the adaptive filter 111 is 2, the amount of calculation of the adaptive algorithm according to the present embodiment is smaller than that of the approximate steepest descent algorithm.

[0146] Furthermore, in the actual adaptive identification system 100, the number of taps of the adaptive filter 111 needs to be at least 128. Figure 5 As shown, it can be seen that the calculation amount of the adaptive algorithm related to this embodiment is significantly reduced compared with the calculation amount of the approximate steepest descent algorithm. In addition, since the adaptive algorithm related to this embodiment is based on the approximate steepest descent algorithm, it can obtain the same identification accuracy as the approximate steepest descent algorithm.

[0147] Therefore, according to the present embodiment, in the adaptive identification system 100 , it is possible to obtain identification accuracy equivalent to that of the approximate steepest descent algorithm with a smaller amount of computation (amount of calculation) than that of the approximate steepest descent algorithm.

[0148] As mentioned above, although embodiment of this invention was described, this invention is not limited to the said embodiment, Various deformation|transformation and change are possible within the range of the summary of this invention described in a claim.

Claims

1. An adaptive identification system using an adaptive filter to identify the characteristics of a transfer system, wherein: have: A signal generating unit that generates a periodic identification input signal that includes a frequency component that is an integer multiple of a fundamental frequency and satisfies a continuous excitation condition; a setting unit that sets the moving average time to a fundamental period of the fundamental frequency of the identification input signal; and The adaptive algorithm execution unit uses a moving average and a diagonalization matrix to update the coefficients of the above-mentioned adaptive filter. The above-mentioned moving average is a value obtained by moving average the input signal vector for identification and the cross-correlation vector of the observation signal using the above-mentioned moving average time, and the above-mentioned diagonalization matrix is ​​a matrix obtained by diagonalizing a matrix obtained by moving average the autocorrelation matrix of the input signal vector for identification using the above-mentioned moving average time.

2. The adaptive identification system according to claim 1, wherein: The signal generating unit generates the periodic identification input signal including only frequency components that are integral multiples of the fundamental frequency.

3. The adaptive identification system according to claim 1 or 2, wherein: The identification input signal includes the frequency components whose number is different from each other and corresponds to the number of taps of the adaptive filter.

4. The adaptive identification system according to claim 1 or 2, wherein: The identification input signal includes the plurality of frequency components having phases different from each other.

5. The adaptive identification system according to claim 1 or 2, wherein: The signal generating unit randomly changes phases of the plurality of frequency components included in the identification input signal.

6. The adaptive identification system according to claim 1 or 2, wherein: The method includes a calculation unit that calculates in advance the diagonalized matrix obtained by moving average the autocorrelation matrix of the identification input signal vector using the moving average time and diagonalizing the matrix.

7. The adaptive identification system according to claim 1 or 2, wherein: A storage unit is provided, which stores in advance the diagonalized matrix obtained by moving average the autocorrelation matrix of the identification input signal vector using the moving average time and diagonalizing the matrix.

8. The adaptive identification system according to claim 7, wherein: The adaptive algorithm execution unit updates the coefficients of the adaptive filter using a moving average value obtained by moving average the cross-correlation vectors using the moving average time and the diagonalization matrix pre-stored in the storage unit.

9. An adaptive identification device, using an adaptive filter to identify the characteristics of a transmission system, wherein: have: A signal generating unit that generates a periodic identification input signal that includes a frequency component that is an integer multiple of a fundamental frequency and satisfies a continuous excitation condition; a setting unit that sets the moving average time to a fundamental period of the fundamental frequency of the identification input signal; and The adaptive algorithm execution unit uses a moving average and a diagonalization matrix to update the coefficients of the above-mentioned adaptive filter. The above-mentioned moving average is a value obtained by moving average the input signal vector for identification and the cross-correlation vector of the observation signal using the above-mentioned moving average time, and the above-mentioned diagonalization matrix is ​​a matrix obtained by diagonalizing a matrix obtained by moving average the autocorrelation matrix of the input signal vector for identification using the above-mentioned moving average time.

10. An adaptive identification method, which is an adaptive identification method performed by an adaptive identification system for identifying characteristics of a transmission system using an adaptive filter, wherein: generating a periodic identification input signal that contains frequency components that are integer multiples of the fundamental frequency and satisfies a continuous excitation condition; Setting the moving average time to the fundamental period of the fundamental frequency of the identification input signal; The coefficients of the adaptive filter are updated using a moving average and a diagonalization matrix. The moving average is a value obtained by moving average the identification input signal vector and the cross-correlation vector of the observation signal using the moving average time. The diagonalization matrix is ​​a matrix obtained by diagonalizing a matrix obtained by moving average the autocorrelation matrix of the identification input signal vector using the moving average time.

Images