JPH06214972A - Signal processor - Google Patents

Abstract

PURPOSE:To attain accurate recognition by correcting the position of a signal waveform by correcting positional deviation on the time axis of the signal waveform in a signal processor handling an input signal as a pattern. CONSTITUTION:The identification object signal of an identification object signal file 27 segmented by a signal segmenting device 4 is regarded as the output of a system which can be expressed by an AR model, a linear prediction coefficient estimating device 6 estimates the coefficient of the AR model so as to identify the system, an AR model output generator 7 determines the inpulse response of the identified system and a neural network 10 executes learning and identification based on inpulse response.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal processing device,
In particular, the present invention relates to a signal processing device suitable for cutting out an identification target signal from a time-series signal and performing identification using the cut out identification target signal.

[0002]

2. Description of the Related Art Generally, when cutting out an identification target signal from a time series signal, it is difficult to accurately cut out only the identification target signal. In many cases, the cut-out point of the identification target signal changes by cutting out to a part that is not the identification target signal, or by cutting off part of the identification target signal,
The displacement of the identification target signal occurs. As an identification result,
Since it is necessary to obtain a correct identification result even if the position is displaced, it is necessary to adopt an identification method capable of identifying even if the position is displaced.

Therefore, conventionally, a signal to be discriminated is subjected to Fourier transform to obtain a power spectrum or a linear prediction coefficient to be mapped to a space unrelated to the positional deviation for discrimination or position discrimination. The identification was performed so as to consider all the shifted signals.

Such a technique is disclosed in JP-A-3-11189.
No. 8, JP-A-3-237574, and JP-A-1-2778.
No. 99 and the document “Learning of misalignment pattern by error back-propagation method” (Kenji Yasu, Shuichi Kuroki, Kiyotoshi Matsuoka, IEICE Transactions, D-11, Vol. J74-D-
II, No. 1 pp. 27-35, January 1991)
Etc.

FIG. 10 is a block diagram showing an example of a conventional signal processing apparatus, and particularly exemplifies a configuration of an identification apparatus for performing identification by a linear prediction coefficient. FIG. 11 is a block diagram illustrating the configuration of the neural network in FIG. These configurations are inferred from the reference “image recognition device” (Chinori Nomune: Japanese Patent Application No. 2-31523).

In FIG. 10, 1 is a learning data file, 2 is an identification data file, 3 is a teacher data file containing teacher data indicating each identification result, and 4 is time series data such as learning data and identification data. From the identification target signal file 27, a signal extraction device for extracting the identification target signal from the identification target signal file 27 storing the identification signal extracted by the signal extraction device 4, Prediction coefficient estimation device, 10 is a neural network for performing discrimination, 1
1 is an identification result determination means for determining the identification result by the neural network 10, 12 is an identification result file for storing identification result data, and 9 is a learning / identification changeover switch.

On the other hand, in FIG. 11, reference numeral 60 is a neural network element, 61 is a connecting line having a weight, and 64 is a connecting line.
Is an input layer, 63 is an intermediate layer, and 62 is an output layer.

The operation of the above-described structure will be described below.

First, the learning / identification changeover switch 9 is switched to the learning side, and the neural network 10 is learned by the linear prediction coefficient obtained by the linear prediction coefficient estimation device 6. By this learning, the neural network 10 changes the weight of the connecting line 61 and outputs the learning data close to the desired teacher data.

Next, the learning / identification changeover switch 9 is switched to the identification side, and the neural network 10 identifies the linear prediction coefficient obtained by the linear prediction coefficient estimation device 6. The neural network 10 executes an operation according to the changed weight and outputs an output.

The discrimination result judging means 11 compares the output from the output layer 62 of the neural network 10 with the contents of the teacher data file 3 and outputs the discrimination result indicated by the teacher data which is considered to be the closest.

[0012]

Since the conventional signal processing apparatus is configured as described above, the classification is performed when there is no clear difference between the linear prediction coefficient and the power spectrum obtained from different identification target signals. There is a problem that it becomes difficult. That is, when considering a feature space having linear prediction coefficients on each axis, the linear prediction coefficient obtained from the discrimination target signal for which different discrimination results are desired to be obtained is the linear prediction coefficient obtained from the discrimination target signal for which the same result is desired. Compared to, it is difficult to identify unless they are separated in this feature space. For example, regarding the information on the time axis such as the pulse width when the signal to be identified is a pulse, the difference in the feature space does not appear clearly in the linear prediction coefficient and the power spectrum, which makes it difficult to identify. in this way,
The conventional signal processing device has a problem of identification failure due to the fact that the difference between the linear prediction coefficient and the power spectrum of the identification target signal is not apparent.

The present invention solves the problems of the prior art as described above, and even if the identification target signal cut out from the time-series signal has a positional deviation, the positional deviation is corrected, and the feature space for easy identification is provided. It is an object of the present invention to obtain a signal processing device that maps a signal to be identified to and enables reliable identification.

[0014]

In order to achieve the above object, as the signal processing device according to claim 1, there is provided a signal cutting-out means for cutting out a discrimination target signal from among discrimination signals.
Signal position correction means for performing system identification by adding white noise to the linear prediction model of the identification target signal and for correcting the positional deviation of the identification target signal according to the impulse response of the system, and positional deviation correction by the signal position correction means And a discriminating means for discriminating a predetermined signal from the discriminating signal by performing learning based on the discriminated signal.

In order to achieve the above object, as a signal processing device according to a second aspect of the present invention, a signal cutting-out means for cutting out an identification target signal from an identification signal and white noise in a linear prediction model of the identification target signal. In addition, the system position is corrected by the signal position correction unit, which corrects the position error of the identification target signal based on the system impulse response and normalizes the signal level to remove the influence of noise. And a discriminating means for discriminating a predetermined signal from the discriminating signal by learning based on the signal whose signal level is corrected and whose signal level is normalized.

In order to achieve the above object, as a signal processing device according to a third aspect of the present invention, there is provided a signal processing device for cutting out a discrimination target signal from discrimination signals, and a system based on a linear prediction model of the discrimination target signal. A signal position correction unit that performs identification and identification, adds a signal generated in consideration of the pulse width of the identification target signal, and corrects the position shift of the identification target signal, and the position shift is corrected by the signal position correction unit The present invention provides a signal processing device including identification means for learning based on a signal and identifying a predetermined signal from the identification signal.

In order to achieve the above object, as a signal processing device according to a fourth aspect, a signal cutting-out means for cutting out an identification target signal from an identification signal, and Fourier-transforming the identification target signal to perform spectrum estimation. , A signal position correction unit that corrects the position shift of the identification target signal by correcting the phase component and performing an inverse Fourier transform, and the identification signal that performs learning based on the signal that has been position shift corrected by the signal position correction unit. And a discrimination means for discriminating a predetermined signal from the signal processing device.

In order to achieve the above object, as a signal processing device according to a fifth aspect of the present invention, a signal cutting-out means for cutting out an identification target signal from an identification signal and a Fourier transform of the identification target signal for spectrum estimation. , A signal position correcting unit that detects a position shift of the identification target signal from a phase of a relatively low frequency, and performs a position shift correction of the identification target signal based on the detection result, and a position shift correction by the signal position correction unit. And a discriminating means for discriminating a predetermined signal from the discriminating signal by learning based on the signal.

In order to achieve the above object, as a signal processing device according to a sixth aspect of the present invention, a signal cutting-out means for cutting out an identification object signal from an identification signal, and a wavelet transform of the identification object signal, and the result of the conversion. The position deviation of the identification target signal is detected on the basis of the detection result, and the signal position correction means for correcting the position deviation of the identification target signal based on the detection result, and the learning based on the signal subjected to the position deviation correction by the signal position correction means. And a discriminating means for discriminating a predetermined signal from the discriminating signal.

In order to achieve the above object, as a signal processing device according to claim 7, a signal cutting-out means for cutting out an identification object signal from an identification signal, and Haar transforming the identification object signal, and a result of the conversion. The position deviation of the identification target signal is detected on the basis of the detection result, and the signal position correction means for correcting the position deviation of the identification target signal based on the detection result, and the learning based on the signal position correction corrected by the signal position correction means. And a discriminating means for discriminating a predetermined signal from the discriminating signal.

[0021]

In the above means, the signal processing apparatus according to claim 1 performs system identification by adding white noise to the linear prediction model of the identification target signal cut out from the identification signal, and at the same time, the impulse response of the system. Displacement correction of the identification target signal is performed by using, and learning is performed based on the signal subjected to the displacement correction to identify a predetermined signal from the identification signal.At this time, the signal displacement due to the impulse response obtained without displacement Is eliminated and is converted to a signal on the time axis, so that information on the time axis can be easily identified.

In the above-mentioned means, the signal processing device according to the second aspect adds the white noise to the linear prediction model of the identification target signal cut out from the identification signal to identify the system, and also the impulse response of the system. Corrects the position deviation of the signal to be identified by using, and normalizes the signal level to remove the influence of noise, and performs the learning based on the signal that has corrected the position error and the signal level, and identifies the predetermined signal from the identification signal. However, at this time, the amplification and attenuation of the impulse response level caused by the estimation error of the linear prediction model due to the noise included in the identification signal is corrected.

In the above means, the signal processing apparatus according to claim 3 performs system identification and identification based on a linear prediction model of the identification target signal cut out from the identification signal, and the pulse width of the identification target signal. By adding the signal generated in consideration of, the positional deviation correction of the identification target signal is performed, and learning is performed based on the signal subjected to the positional deviation correction to identify a predetermined signal from the identification signal, but in this case, the identification target signal is In the case of a pulse waveform, the output of the system is obtained without displacement of the signal, so that the displacement of the signal is eliminated and the signal on the time axis is restored, so that the information on the time axis can be easily identified.

In the above means, the signal processing device according to the fourth aspect is characterized in that the identification target signal cut out from the identification signal is Fourier-transformed to estimate the spectrum, and the phase component is corrected to perform the inverse Fourier transform. Displacement correction of the identification target signal is performed by using, and learning is performed based on the signal subjected to the displacement correction to identify a predetermined signal from the identification signal. At this time, the phase is corrected and a signal without displacement is obtained. , The information on the time axis can be easily identified.

In the above means, the signal processing apparatus according to claim 5 Fourier-transforms the identification target signal cut out from the identification signal to estimate the spectrum, and determines the identification target signal from the phase of a relatively low frequency. The positional deviation is detected, the positional deviation of the identification target signal is corrected based on the detection result, and the learning is performed based on the position deviation corrected signal to identify a predetermined signal from the identification signal. Is a pulse waveform, the positional deviation is corrected with the waveform of the signal to be identified as it is, so that accurate recognition can be performed and information on the time axis can be easily identified.

In the above means, the signal processing device according to the sixth aspect performs wavelet transform on the identification target signal cut out from the identification signal, and detects the positional deviation of the identification target signal based on the conversion result. The displacement of the identification target signal is corrected based on this detection result, and learning is performed based on the signal subjected to the displacement correction to identify a predetermined signal from the identification signal. At that time, the displacement of the identification target signal remains unchanged. Is corrected, the subsequent accurate recognition can be performed and the information on the time axis can be easily identified.

In the above-mentioned means, the signal processing device according to claim 7 performs Haar transformation on the discrimination target signal cut out from the discrimination signals, and detects the positional deviation of the discrimination target signal based on the conversion result. Positional deviation correction of the identification target signal is performed based on this detection result, learning is performed based on the signal subjected to the positional deviation correction, and a predetermined signal is identified from the identification signal. At that time, if the identification target signal is a pulse waveform, pulse The Haar transform value changes greatly at the pulse waveform start time and end time in order to perform the Haar transform of the bite.Therefore, by detecting this time, the signal position is corrected with the waveform of the identification target signal as it is. As a result, a signal without positional deviation is obtained, and since the signal on the time axis remains, it is easy to identify the information on the time axis.

[0028]

Embodiments of the present invention will be described below with reference to the drawings.

Embodiment 1 FIG. 1 is a block diagram of a signal processing device according to a first embodiment of the present invention. In the figure, reference numeral 5 is a signal position correction device that corrects the positional deviation of the identification target signal using an AR model, and is composed of a linear prediction coefficient estimation device 6 and an AR model output generation device 7. The signal position correction device 5 is provided with the extraction signal from the identification target signal file 27, and outputs the position deviation correction signal to the position deviation correction signal file 28. An impulse data file 8 is connected to the AR model output generation device 7 of the signal position correction device 5.

FIG. 2 shows the AR used in the configuration of FIG.
Model is an explanatory view of, in FIG., 21 is AR model consisting of a system of linear prediction coefficients a _1, a ₂ ··· a _m and the coupling coefficient z ^-1, 22 is an input data file for the AR model 21, Reference numeral 22 is an output data file from the AR model 21.

FIG. 3 is an explanatory diagram of signal position shift correction using the AR model 21 of FIG. In the figure, 26 is A
A system represented by an R model, 25 is an identification / response switch, and 23 is a white noise data file. By the way, the identification / response switch 25 is switched to connect the white noise data file 23 and the identification target signal file 27 to the system 26 represented by the AR model at the time of identification, and at the time of response, the impulse data file 8 and the positional deviation correction are performed. The signal file 28 is switched to connect to the system 26 represented by the AR model.

The operation of the above arrangement will be described below.

The learning / identification changeover switch 9 is switched to the learning side, and the signal to be identified is cut out from the learning data file 1 by the signal cutting-out device 4. This cutout is
The threshold value is lowered so that the signal to be discriminated is not always included, but a portion other than the signal to be discriminated is also included in the margin.

Next, the cut-out identification target signal is regarded as an output from a certain system driven by white noise, the linear prediction coefficient estimator 6 estimates the linear prediction coefficient ak of the AR model 21, and the system is identified. I do.

FIG. 2 shows a system represented by an AR model 21. Thus, in the AR model 21, the output x (n) at a certain time is

[Equation 1] Thus, the transfer function H (z) of the system represented by the AR model 21 can be represented by a linear combination of previous output values.

[Equation 2] Is. In addition, in this system identification, in FIG. 3, the identification / response selector switch 25 is switched to the identification side, and the AR
This corresponds to obtaining the system 26 represented by the model.

Next, the impulse data file 8 is input to the AR model output generator 7 and the impulse response of the obtained system is obtained. This corresponds to switching the identification / response changeover switch 25 to the response side in FIG. 3, inputting the impulse data file 8 and deriving the positional deviation correction signal file 28. This impulse response is different from the input such as white noise and impulse, but is considered to be a signal output from the same system as the signal to be discriminated, and is therefore considered to include the same characteristics as the signal to be discriminated. Further, since it is an impulse response, there is no displacement of the signal.

As described above, the impulse response from the position shift correction signal file 28 including the position shift correction signal obtained by the signal position correction device 5 and the teacher data extracted from the teacher data file 3 are input to the neural network 10. Then, let's learn this.

Next, the learning / identification change-over switch 9 is switched to identification, the identification target signal is cut out from the identification data file 2, the same processing as in the case of learning is performed, and the obtained impulse response is output to the neural network 10.
Enter and identify.

Next, the discrimination result discriminated by the neural network 10 is compared by the discrimination result judging means 11 by comparing the discrimination result outputted from the neural network 10 and the teacher data file 3 taught to the input data file 20. It is obtained and stored in the identification result file 12.

Through the above-described operation, the identification target signal stored in the identification data file 2 can be accurately cut out and identified.

As a result, the identification performance can be improved when the identification is performed with the information that is difficult to identify with the linear prediction coefficient and the power spectrum.

Embodiment 2 FIG. 4 is a partial block diagram of a signal processing apparatus according to a second embodiment of the present invention. In the figure, 30 is a signal level normalization device for normalizing the signal level of the output of the AR model output generation device 7. Incidentally, the configuration of this embodiment is exactly the same as the configuration of FIG. 1 except the configuration of the signal position correction device 5.

In the above configuration, when the identification target signal from the identification target signal file 27 is regarded as the output of the system driven by white noise and the system identification is performed by the AR model, the identification target signal is stationary. The condition is that there is a certain condition, but in most cases, it is not stationary in the actual signal.

When the linear prediction coefficient ak is obtained from a stationary signal, almost the same value can be obtained even if there is a positional deviation of the signal, but when the linear prediction coefficient ak is obtained from a non-stationary signal, the value varies. Occurs. As described above, the impulse response obtained from the varied linear prediction coefficient ak has its level amplified or attenuated, and causes the deterioration of the discrimination performance when the neural network 10 or the template matching is used as the discrimination means. Becomes

Therefore, the level of the impulse response from the AR model output generation device 7 is normalized by the signal level normalization device 30 and output to the positional deviation correction signal file 28, thereby eliminating the influence of noise included in the identification signal. Therefore, the accuracy of the subsequent identification can be improved.

Embodiment 3 FIG. 5 is a partial block diagram of a signal processing apparatus according to a third embodiment of the present invention. In the figure, 31 is a pulse width measuring device that measures the pulse width in order to extract the impulse from the identification target signal, and 32 is the phase characteristic of the pulse having the pulse width obtained by the pulse width measuring device 31. Create a signal with flat phase characteristics and flat amplitude characteristics
It is an input signal creation device that creates an input signal to the model output creation device 7. Incidentally, the configuration of this embodiment is exactly the same as the configuration of FIG. 1 except the configuration of the signal position correction device 5.

The operation of the above arrangement will be described below.

Fourier transform F (ω) of a signal f (t)
Is

[Equation 3] Then, F ₀ (w) is called an amplitude characteristic and P (ω) is called a phase characteristic. However, j is an imaginary unit.

The pulse width measuring device 31 and the input signal generating device 32 process the discrimination target signal through the above-mentioned Fourier transform and generate an input signal to the AR model output generating device 7.

In the case of the first embodiment, the impulse is input to the identified system, but in this embodiment, the signal generated from the identification target signal is input. That is, the signal to be identified is set to the pulse width measuring device 3
1 to measure the pulse width and input signal generator 32
Obtain the phase characteristics of the pulse with the pulse width obtained in
A signal having a flat amplitude characteristic with the obtained phase characteristic is generated and given to the identified AR model output generation device 7. Then, the AR model output generation device 7 generates an AR model and stores it in the positional deviation correction signal file 28.

As a result, the waveform having a shape closer to the original waveform than the impulse response obtained in the case of the first embodiment can be obtained, and the discrimination performance can be improved.

Embodiment 4 FIG. 6 is a partial block diagram of a signal processing apparatus according to a fourth embodiment of the present invention. In the figure, reference numeral 40 denotes a Fourier transform device that performs a Fourier transform on the identification target signal from the identification target signal file 27 to obtain the Fourier transform F (ω) of the identification target signal, and 41 denotes the Fourier transform F (obtained by the Fourier transform device 40. a phase correction device that obtains a Fourier transform F ′ (ω) by performing phase correction on all phase characteristics P (ω) of ω), 42
Is an inverse Fourier transform device that inverse-Fourier-transforms the phase-corrected signal F ′ (ω) to generate a signal without positional deviation and gives it to the positional deviation correction signal file 28. By the way,
The configuration of this embodiment is exactly the same as that of FIG. 1 except the configuration of the signal position correction device 5.

In the configuration described above, the Fourier transform device 40 of the signal position correction device 5 performs the Fourier transform on the identification target signal from the identification target signal file 27 to obtain F (ω). Next, all the phase characteristics F (ω) of the Fourier transform F (ω) obtained by the phase correction device 41 are

[Equation 4] Then, correct the phase,

[Equation 5] Create a signal. The inverse Fourier transform device 42 performs an inverse Fourier transform on this signal to create a signal without positional deviation. A signal obtained in this way, which has no positional deviation and whose waveform is almost unchanged, is used as the positional deviation correction signal file 28.
, It is possible to realize highly accurate identification.

Embodiment 5 FIG. 7 is a partial block diagram of a signal processing apparatus according to a fourth embodiment of the present invention. In the figure, 43 is an FFT device that FFTs the identification target signal from the identification target signal file 27 to obtain the first harmonic, 44 is a signal position deviation calculation device that obtains the position deviation of the identification target signal from the output of the FFT device 43, Reference numeral 45 denotes a position shift correction device that corrects the position shift of the identification target signal based on the position shift calculation result from the signal position shift calculation device 44. Incidentally, the configuration of this embodiment is exactly the same as the configuration of FIG. 1 except the configuration of the signal position correction device 5.

The operation of the configuration described above will be described below.

The Fourier transform F (ω) of the pulse waveform f (ω) of the discrimination target signal is

[Equation 6] And this signal f (t−a) is shifted by the time a.
The Fourier transform of the signal is

[Equation 7] Therefore, if ωa = P (ω) is known, the magnitude a of the positional deviation can be obtained.

This is utilized in this embodiment, and here it is considered that the identification target signal from the identification target signal file 27 is close to a pulse waveform like a radar pulse waveform. Then, the FFT device 43 performs FFT on the identification target signal to obtain the first harmonic. The phase P (ω ₁ ) of the first harmonic is when the signal to be identified has a substantially pulse waveform,

[Equation 8] Is considered to be. However, a is the positional deviation of the signal.

The amplitude of the spectrum of the pulse waveform is maximum at ω = 0 and has the largest value near ω = 0, so that it is unlikely to be affected by noise or collapse of the pulse waveform. Also, ω =
At 0, P (ω) = 0 always holds. Therefore, the first harmonic is used to obtain the magnitude a of the positional deviation.

From the phase characteristic of the first harmonic obtained by the FFT device 43, the magnitude of the positional deviation a is calculated by the signal positional deviation calculating device 44, and this is given to the positional deviation correcting device 45. The misregistration correction device 45 generates a signal obtained by shifting the identification target signal from the identification target signal file 27 by -a in correspondence with the magnitude a of the positional deviation. As a result, it is possible to eliminate the positional deviation without changing the waveform of the identification target signal. Therefore, the discrimination performance can be improved by giving the discrimination target signal whose positional deviation is corrected to the positional deviation correction signal file 28.

Sixth Embodiment FIG. 8 is a partial block diagram of a signal processing device according to a sixth embodiment of the present invention. In the figure, 46 is a wavelet transform device for wavelet-transforming the identification target signal from the identification target signal file 27, and 47 is a signal position detection device for detecting the signal position of the identification target signal based on the conversion result of the wavelet transformation device 46. . Incidentally, the configuration of this embodiment is exactly the same as the configuration of FIG. 1 except the configuration of the signal position correction device 5.

The wavelet transform device 46 performs wavelet transform on the identification target signal from the identification target signal file 27. The wavelet transform is one of linear transforms similar to the Fourier transform, and in the short-time Fourier transform using the same window function, both the time resolution and the frequency resolution are constant, but in the wavelet transform, in the high frequency component, time It is characterized by high resolution and high frequency resolution for low frequency components.

The wavelet transform f _w (a, b) is

[Equation 9] Is defined as, and * indicates a complex conjugate. Also, Φ is called an analyzing wavelet, which determines the properties of the transformation,
It is set as follows.

[0063]

[Equation 10] From now on, the analyzing wavelet Φ _{a, b} (t)
It can be seen that is created from the expansion, contraction, and translation of Ψ (t). Also, Ψ is

[Equation 11] If the admissible condition of is satisfied, there is an inverse transform of the wavelet transform. Therefore, while the Fourier transform is always a trigonometric function, the wavelet transform can be used with any function as long as the condition of admissible condition is satisfied, and analysis with a higher degree of freedom can be performed.

From these facts, if a shape similar to the waveform of the discrimination target signal is used for the analyzing wavelet,
It is considered that the wavelet transform reacts sensitively to the waveform of the discrimination target signal.

Therefore, the analyzing wavelet is made similar to the discrimination target signal, the wavelet transform device 46 performs wavelet transform, the signal position detecting device 47 detects the signal position, and the position shift correcting device 45 corrects the signal position. By doing so, it becomes possible to eliminate the positional deviation without changing the waveform of the identification target signal. Therefore, the discrimination performance can be improved by giving the discrimination target signal whose positional deviation is corrected to the positional deviation correction signal file 28.

Seventh Embodiment FIG. 9 is a partial block diagram of a signal processing apparatus according to the seventh embodiment of the present invention. In the figure, reference numeral 48 is a Haar converter for Haar converting the identification target signal from the identification target signal file 27. Incidentally, the configuration of this embodiment is exactly the same as the configuration of FIG. 1 except the configuration of the signal position correction device 5.

The operation of the above arrangement will be described below.

The identification target signal from the identification target signal file 27 is input to the Haar conversion device 48 of the signal position correction device 5. The Haar transform device 48 Haar transforms the identification target signal.

The Haar transform is one of linear transforms similar to the Fourier transform, but since it is an orthogonal system of binary functions, it has a feature that it can be easily handled by a computer. Haar transformation h (n,
m, t) is

[Equation 12] Is defined as

When the discrimination target signal has a pulse waveform, it is considered that Haar transformation is sensitive to the pulse waveform.
Therefore, the Haar conversion device 48 performs the Haar conversion and supplies the signal position detection device 47 with the Haar conversion, whereby the position of the pulse waveform can be accurately detected. By correcting the position shift of the identification target signal by the position shift correction device 45 based on the signal position detected by the signal position detection device 47,
It is possible to eliminate the positional deviation without changing the waveform of the identification target signal. Therefore, the discrimination performance can be improved by giving the discrimination target signal whose positional deviation is corrected to the positional deviation correction signal file 28.

[0071]

As described above, according to the signal processing device of the first to seventh aspects, the discrimination performance can be improved when the discrimination is performed by the information which is difficult to discriminate with the linear prediction coefficient or the power spectrum. Not only that, the influence of noise can be removed, and the positional deviation can be corrected with almost no change in the waveform of the discrimination target signal, so that the discrimination performance can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of a signal processing device according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram of an AR model used in the configuration of FIG.

FIG. 3 is an explanatory diagram of signal position shift correction using the AR model 21 of FIG. 2;

FIG. 4 is a partial block diagram of a signal processing device according to a second embodiment of the present invention.

FIG. 5 is a partial block diagram of a signal processing device according to a third embodiment of the present invention.

FIG. 6 is a partial block diagram of a signal processing device according to a fourth embodiment of the present invention.

FIG. 7 is a partial block diagram of a signal processing device according to a fifth embodiment of the present invention.

FIG. 8 is a partial block diagram of a signal processing device according to a sixth embodiment of the present invention.

FIG. 9 is a partial block diagram of a signal processing device according to a seventh embodiment of the present invention.

FIG. 10 is a block diagram of a conventional signal processing device.

11 is a block diagram illustrating a configuration of a neural network having the configuration of FIG.

[Explanation of symbols]

 1 Learning Data File 2 Identification Data File 3 Teacher Data File 4 Signal Extractor 5 Signal Position Corrector 6 Linear Prediction Coefficient Estimator 7 AR Model Output Generator 8 Impulse Data File 9 Learning / Identification Changeover Switch 10 Neural Network 11 Identification Result judging means 12 Identification result file 20 Input data file 21 AR model 22 Output data file 23 White noise data file 25 Identification / response changeover switch 26 System represented by AR model 27 Identification target signal file 28 Misalignment correction signal file 30 Signal Level normalizer 31 Pulse width measuring device 32 Input signal generator 40 Fourier transform device 41 Phase corrector 42 Inverse Fourier transform device 43 FFT device 44 Signal position deviation calculator 45 Location shift correction device 46 wavelet transform unit 47 signals the position detecting device 48 Haar transform device 60 element 61 coupled line 62 the output layer 63 intermediate layer 64 input layer

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 19, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be amended] Claims

[Correction method] Change

[Correction content]

[Claims]

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0006

[Correction method] Change

[Correction content]

In FIG. 10, 1 is a learning data file, 2 is an identification data file, 3 is a teacher data file containing teacher data indicating each identification result, and 4 is time series data such as learning data and identification data. identification identification target signal signal clipping device for cutting out, 27 the identification target signal file for storing an identification signal extracted by the signal cutting-out device 4, 6 from the identification target signal file 27 from
A linear prediction coefficient estimation device that separately estimates a linear prediction coefficient to be used, 10 is a neural network that performs discrimination,
Reference numeral 11 is an identification result determination means for determining an identification result by the neural network 10, 12 is an identification result file for storing identification result data, and 9 is a learning / identification changeover switch.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0014

[Correction method] Change

[Correction content]

[0014]

In order to achieve the above object, as the signal processing device according to claim 1, there is provided a signal cutting-out means for cutting out a discrimination target signal from among discrimination signals.
By obtaining the linear prediction model of the signal to be identified,
The system position identification means for correcting the displacement of the identification target signal based on the impulse response of the system, and the learning based on the signal position displacement corrected by the signal position correction means, and the identification signal The present invention provides a signal processing device including identification means for identifying a predetermined signal.

[Procedure amendment 4]

[Document name to be amended] Statement

[Name of item to be corrected] 0015

[Correction method] Change

[Correction content]

In order to achieve the above object, as a signal processing device according to a second aspect of the present invention, a signal cutting-out means for cutting out a discrimination target signal from a discrimination signal and a linear prediction model of the discrimination target signal are obtained. Signal position correcting means for identifying the system, correcting the displacement of the identification target signal based on the impulse response of the system, and normalizing the signal level to remove the influence of noise and non-stationarity of the signal, and the signal position correction. And a discriminating means for discriminating a predetermined signal from the discriminating signal by performing learning based on a signal whose positional deviation has been corrected by the means and whose signal level has been normalized.

[Procedure Amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction content]

In order to achieve the above object, as a signal processing apparatus according to a third aspect of the present invention, a signal cutting-out means for cutting out a discrimination target signal from a discrimination signal and a linear prediction model of the discrimination target signal are obtained. A signal position correction means for performing system identification and adding a signal generated in consideration of the pulse width of the identification target signal to correct the positional deviation of the identification target signal, and a position by the signal position correction means. It is intended to provide a signal processing device comprising: an identification unit that performs learning based on a signal whose displacement has been corrected to identify a predetermined signal from the identification signal.

[Procedure correction 6]

[Document name to be amended] Statement

[Correction target item name] 0021

[Correction method] Change

[Correction content]

[0021]

In the above means, the signal processing apparatus according to claim 1 identifies the system by obtaining a linear prediction model of the identification target signal cut out from the identification signal and identifies the system by the impulse response of the system. The target signal is offset and the learning is performed based on the offset-corrected signal to identify a predetermined signal from the identification signal.At this time, the offset of the signal is eliminated by the impulse response required without the offset. Since the signal on the time axis is used again, the information on the time axis can be easily identified.

[Procedure Amendment 7]

[Document name to be amended] Statement

[Name of item to be corrected] 0022

[Correction method] Change

[Correction content]

In the above means, the signal processing apparatus according to claim 2 identifies the system by obtaining a linear prediction model of the identification target signal cut out from the identification signal, and performs identification by the impulse response of the system. Corrects the displacement of the target signal and normalizes the signal level to remove the effects of noise and non-stationarity of the signal, and corrects the displacement and performs learning based on the signal with the normalized signal level, and then determines from the identification signal. Identify the signal of
Amplification and attenuation of the impulse response level caused by the estimation error of the linear prediction model due to the noise included in the identification signal and the non-stationarity of the signal are corrected.

[Procedure Amendment 8]

[Document name to be amended] Statement

[Name of item to be corrected] 0023

[Correction method] Change

[Correction content]

In the above means, the signal processing device according to the third aspect of the present invention obtains the linear prediction model of the identification target signal cut out from the identification signal, thereby obtaining the system
Identifying signal generated in consideration of the pulse width of the identification target signal performs adding performs misalignment correction of the identification target signal, positional deviation corrected on the basis of a signal identifying a predetermined signal from the identification signal performs learning However, in this case, when the signal to be identified is a pulse waveform, the system output can be obtained without signal displacement, so there is no signal displacement, and the signal on the time axis is restored. Is also easy to identify.

[Procedure Amendment 9]

[Document name to be amended] Statement

[Name of item to be corrected] 0025

[Correction method] Change

[Correction content]

In the above means, the signal processing apparatus according to claim 5 Fourier-transforms the identification target signal cut out from the identification signal to estimate the spectrum, and determines the identification target signal from the phase of a relatively low frequency. The positional deviation is detected, the positional deviation of the identification target signal is corrected based on the detection result, and the learning is performed based on the position deviation corrected signal to identify a predetermined signal from the identification signal. Is a pulse waveform, the positional deviation is corrected while the waveform of the identification target signal remains unchanged, so that accurate identification can be performed and information on the time axis can be easily identified.

[Procedure Amendment 10]

[Document name to be amended] Statement

[Correction target item name] 0026

[Correction method] Change

[Correction content]

In the above means, the signal processing device according to the sixth aspect performs wavelet transform on the identification target signal cut out from the identification signal, and detects the positional deviation of the identification target signal based on the conversion result. The displacement of the identification target signal is corrected based on this detection result, and learning is performed based on the signal subjected to the displacement correction to identify a predetermined signal from the identification signal. At that time, the displacement of the identification target signal remains unchanged. Is corrected, the subsequent accurate identification can be performed and the information on the time axis can be easily identified.

[Procedure Amendment 11]

[Document name to be amended] Statement

[Name of item to be corrected] 0027

[Correction method] Change

[Correction content]

In the above-mentioned means, the signal processing device according to claim 7 performs Haar transformation on the discrimination target signal cut out from the discrimination signals, and detects the positional deviation of the discrimination target signal based on the conversion result. Positional deviation correction of the identification target signal is performed based on this detection result, learning is performed based on the signal subjected to the positional deviation correction, and a predetermined signal is identified from the identification signal. At that time, if the identification target signal is a pulse waveform, pulse
In order to Haar transform the waveform , the Haar transform value greatly changes at the pulse waveform start time and end time, so
By detecting this time, the signal position is corrected with the waveform of the identification target signal as it is, so a signal with no positional deviation is obtained, and since it is a signal on the time axis, information on the time axis is also identified. It will be easier.

[Procedure Amendment 12]

[Document name to be amended] Statement

[Name of item to be corrected] 0030

[Correction method] Change

[Correction content]

FIG. 2 shows the AR used in the configuration of FIG.
Model is an explanatory view of, in FIG., 21 is AR model consisting of a system of linear prediction coefficients a _1, a ₂ ··· a _m a complex parameter z ^-1, 22 is an input data file for the AR model 21, Reference numeral 22 is an output data file from the AR model 21.

[Procedure Amendment 13]

[Document name to be amended] Statement

[Correction target item name] 0034

[Correction method] Change

[Correction content]

Next, the cut-out identification target signal is regarded as an output from a certain system driven by white noise, the linear prediction coefficient estimation device 6 estimates the linear prediction coefficient a _k of the AR model 21, and the system Identify.

[Procedure Amendment 14]

[Document name to be amended] Statement

[Correction target item name] 0035

[Correction method] Change

[Correction content]

FIG. 2 shows a system represented by an AR model 21. Thus, in the AR model 21, the output x (n) at a certain time is

[Equation 1] Thus, the transfer function H (z) of the system represented by the AR model 21 can be represented by a linear combination of previous output values.

[Equation 2] Is. In addition, in this system identification, in FIG. 3, the identification / response selector switch 25 is switched to the identification side, and the AR
This corresponds to obtaining the system 26 represented by the model.

[Procedure Amendment 15]

[Document name to be amended] Statement

[Correction target item name] 0040

[Correction method] Change

[Correction content]

Through the operations described above, it is possible to perform the positional deviation correction and the discrimination of the discrimination target signal stored in the discrimination data file 2.

[Procedure Amendment 16]

[Document name to be amended] Statement

[Correction target item name] 0044

[Correction method] Change

[Correction content]

When the linear prediction coefficient a _k is obtained from a stationary signal, almost the same value can be obtained even if there is a signal displacement, but when the linear prediction coefficient a _k is obtained from a non-stationary signal, the value is Variation occurs. As described above, the impulse response obtained from the varied linear prediction coefficient a _k has its level amplified or attenuated, which deteriorates the discrimination performance when the neural network 10 or the template matching is used as the discrimination means. Cause.

[Procedure Amendment 17]

[Document name to be amended] Statement

[Name of item to be corrected] 0045

[Correction method] Change

[Correction content]

Therefore, the level of the impulse response from the AR model output generation device 7 is normalized by the signal level normalization device 30 and output to the positional deviation correction signal file 28, so that the noise and signal non- existence included in the identification signal are eliminated. The effect of stationarity can be eliminated and the accuracy of subsequent identification can be improved.

[Procedure 18]

[Document name to be amended] Statement

[Correction target item name] 0046

[Correction method] Change

[Correction content]

Embodiment 3 FIG. 5 is a partial block diagram of a signal processing apparatus according to a third embodiment of the present invention. In the figure, 31 is a signal to be identified
A pulse width measuring device for measuring the pulse width of the pulse width measuring device 32, the phase characteristic of the pulse having the pulse width obtained by the pulse width measuring device 31 is obtained, and a signal having a flat amplitude characteristic with the obtained phase characteristic is created to form an AR model. It is an input signal generation device that generates an input signal to the output generation device 7. Incidentally, the configuration of this embodiment is exactly the same as the configuration of FIG. 1 except the configuration of the signal position correction device 5.

[Procedure Amendment 19]

[Document name to be amended] Statement

[Correction target item name] 0048

[Correction method] Change

[Correction content]

Fourier transform F (ω) of a signal f (t)
Is

[Equation 3] Then, F ₀ (w) is called an amplitude characteristic and φ (ω) is called a phase characteristic. However, j is an imaginary unit.

[Procedure amendment 20]

[Document name to be amended] Statement

[Correction target item name] 0052

[Correction method] Change

[Correction content]

Embodiment 4 FIG. 6 is a partial block diagram of a signal processing apparatus according to a fourth embodiment of the present invention. In the figure, reference numeral 40 denotes a Fourier transform device that performs a Fourier transform on the identification target signal from the identification target signal file 27 to obtain a Fourier transform F (ω) of the identification target signal, and 41 denotes a Fourier transform F (obtained by the Fourier transform device 40. 42. A phase correction device that obtains a Fourier transform F ′ (ω) by phase-correcting all phase characteristics φ (ω) of ω),
Is an inverse Fourier transform device that inverse-Fourier-transforms the phase-corrected signal F ′ (ω) to generate a signal without positional deviation and gives it to the positional deviation correction signal file 28. By the way,
The configuration of this embodiment is exactly the same as that of FIG. 1 except the configuration of the signal position correction device 5.

[Procedure correction 21]

[Document name to be amended] Statement

[Correction target item name] 0053

[Correction method] Change

[Correction content]

In the configuration described above, the Fourier transform device 40 of the signal position correction device 5 performs the Fourier transform on the identification target signal from the identification target signal file 27 to obtain F (ω). Next, all the phase characteristics F (ω) of the Fourier transform F (ω) obtained by the phase correction device 41 are

[Equation 4] Then, correct the phase,

[Equation 5] Create a signal. The inverse Fourier transform device 42 performs an inverse Fourier transform on this signal to create a signal without positional deviation. Amplitude characteristics without displacement obtained in this way
It is possible to realize highly accurate identification by applying the same signals to the positional deviation correction signal file 28.

[Procedure correction 22]

[Document name to be amended] Statement

[Correction target item name] 0056

[Correction method] Change

[Correction content]

The Fourier transform F (ω) of the pulse waveform f (ω) of the discrimination target signal is

[Equation 6] And this signal f (t−a) is shifted by the time a.
The Fourier transform of the signal is

[Equation 7] Because it is, ωa = φ P (ω) if is known, the size a of the displacement is obtained.

[Procedure amendment 23]

[Document name to be amended] Statement

[Name of item to be corrected] 0057

[Correction method] Change

[Correction content]

This is utilized in this embodiment, and here it is considered that the identification target signal from the identification target signal file 27 is close to a pulse waveform like a radar pulse waveform. Then, the FFT device 43 performs FFT on the identification target signal to obtain the first harmonic. The phase φ (ω ₁ ) of the first harmonic is, when the signal to be identified is almost a pulse waveform,

[Equation 8] Is considered to be. However, a is the position shift of the signal.

[Procedure correction 24]

[Document name to be amended] Statement

[Name of item to be corrected] 0058

[Correction method] Change

[Correction content]

The amplitude of the spectrum of the pulse waveform is maximum at ω = 0 and has the largest value near ω = 0, so that it is unlikely to be affected by noise or collapse of the pulse waveform. Also, ω =
At 0, φ (ω) = 0 is always obtained. Therefore, the first harmonic is used to obtain the magnitude a of the positional deviation.

[Procedure correction 25]

[Document name to be amended] Statement

[Correction target item name] 0063

[Correction method] Change

[Correction content]

[0063]

[Equation 10] From now on, the analyzing wavelet Φ _{a, b} (t)
It can be seen that is created from the expansion, contraction, and translation of Ψ (t). Also, Ψ is

[Equation 11] If the admissible condition of is satisfied, there is an inverse transform of the wavelet transform. Therefore, in the Fourier transform, the basis
Is always a trigonometric function, whereas the wavelet transform can be used with any function as long as it satisfies the admissible condition, and more flexible analysis is possible.

[Procedure Amendment 26]

[Document name to be amended] Statement

[Correction target item name] 0071

[Correction method] Change

[Correction content]

[0071]

As described above, according to the signal processing device of the first to seventh aspects, the discrimination performance can be improved when the discrimination is performed by the information which is difficult to discriminate with the linear prediction coefficient or the power spectrum. As well as noise and undefined
Since the influence of normality can be removed and the positional deviation can be corrected with almost no change in the waveform of the discrimination target signal, the discrimination performance can be improved.

Claims (7)

[Claims] 1. A signal clipping means for clipping a discrimination target signal from a discrimination signal, and white noise added to a linear prediction model of the discrimination target signal to identify the system, and the impulse response of the system to identify the discrimination target signal. And a discriminating means for discriminating a predetermined signal from the discriminating signal by learning based on the signal whose positional deviation is corrected by the signal positional correcting means. Signal processing device. 2. A signal clipping means for clipping a discrimination target signal from the discrimination signal, and white noise is added to a linear prediction model of the discrimination target signal to identify the system, and the discrimination target signal is determined by the impulse response of the system. And the signal level correction means for normalizing the signal level to remove the influence of noise, and the signal level correction means for correcting the positional deviation and learning based on the signal level normalized. A signal processing device comprising: an identification unit that identifies a predetermined signal from the identification signal. 3. A signal cutting-out means for cutting out a discrimination target signal from among the discrimination signals, and system identification and discrimination based on a linear prediction model of the discrimination target signal and considering the pulse width of the discrimination target signal. A signal position correction means for adding the generated signal to correct the position deviation of the identification target signal, and learning based on the signal subjected to the position deviation correction by the signal position correction means to identify a predetermined signal from the identification signal. A signal processing device comprising: an identification unit. 4. A signal cutting-out means for cutting out a discrimination target signal from among the discrimination signals, and the discrimination target signal by subjecting the discrimination target signal to Fourier transform for spectral estimation, correcting the phase component and performing an inverse Fourier transform. And a discriminating means for discriminating a predetermined signal from the discriminating signal by learning based on the signal whose positional deviation is corrected by the signal positional correcting means. Signal processing device. 5. A signal cutting-out means for cutting out an identification target signal from the identification signal, and Fourier-transforming the identification target signal to estimate a spectrum, and detect a positional deviation of the identification target signal from the phase of a relatively low frequency. Then, based on the detection result, signal position correction means for correcting the position deviation of the identification target signal, and learning based on the signal whose position deviation has been corrected by the signal position correction means are used to identify a predetermined signal from the identification signal. A signal processing device comprising: 6. A signal cutting-out means for cutting out a discrimination target signal from among the discrimination signals, a wavelet transform of the discrimination target signal, a position shift of the discrimination target signal is detected based on the conversion result, and the detection result is obtained. A signal position correcting means for correcting the positional deviation of the identification target signal based on the signal, and an identifying means for learning based on the signal whose positional deviation is corrected by the signal position correcting means and for identifying a predetermined signal from the identification signal. A signal processing device comprising: 7. A signal cutting-out means for cutting out a discrimination target signal from among the discrimination signals, Haar transforming the discrimination target signal, detecting a positional deviation of the discrimination target signal based on the conversion result, and using this detection result. A signal position correcting means for correcting the positional deviation of the identification target signal based on the signal, and an identifying means for learning based on the signal whose positional deviation is corrected by the signal position correcting means and for identifying a predetermined signal from the identification signal. A signal processing device comprising: