JPH0594543A - Method and device for analyzing physical variable - Google Patents

Abstract

PURPOSE:To simplify constitution and to make it possible to analyze the physical variable of a physical source based upon a measured value even in the case of a small number of times of learning. CONSTITUTION:Plural physical formula computing units 11 to 1m execute the operation of physical formulas whose main parts are determined based upon a physical variable in accordance with known information, a sigma unit 2 accumulatively adds the outputs of the units 11 to 1m, an error computing element 3 calculates a difference between the added value and a practically measured value, and correction parts 11a to 1ma in the units 11 to 1m correct respective variables included in the physical formulas so as to reduce the differences. After repeating a series of processing only by the necessary number of times, an information collecting unit 4 collects corrected variables and outputs the collected value as a physical variable analyzing result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a physical quantity analysis device, and in the case where a governing law of a system including a physical source is known as a predetermined arithmetic expression, a physical quantity measurement value obtained at a plurality of points apart from the physical source and The present invention relates to a device that calculates a physical quantity of a physical source based on observation conditions at the time of measurement.

[0002]

2. Description of the Related Art Conventionally, a superconducting quantum interference device (Supercon
ducting Quantum Interference Device, hereafter SQU
An apparatus has been proposed that analyzes a magnetic field source (a type of physical source) inside a living body by arranging a plurality of SQUID magnetometers each using an ID (abbreviated as ID) in the state of being close to the living body.

This device uses a supercomputer to perform the following processing. That is: a) Scatter m current pieces using random numbers in the search space by a plurality of SQUID magnetometers. Here, the current element i
Since the input parameters of are the position information P (x, y, z) and the current vector I (X, Y, Z), xi, yi,
zi, Xi, Yi, Zi (where i = 1, 2, ...
The 6m parameters of m) are determined using random numbers. b) Total estimation error E by the estimation error calculation process described later.
To calculate. c) Repeat the following d) to g). d) The current element is arbitrarily selected, and the parameters and the total estimation error of the corresponding current element k are saved. That is, position information Ps (xs, ys, zs) = Pk (xk, y
k, zk) current vector Is (Xs, Ys, Zs) = Ik (Xk,
Yk, Zk) Perform processing of total estimation error Es = E. e) The parameter of the current element k is changed by a minute amount using random numbers. That is, the minute change amount of each component is represented by Δx, Δ
If y, Δz, ΔX, ΔY, and ΔZ, Pk (xk, yk, zk) = Pk (xk + Δx, yk +
Δy, zk + Δz) Ik (Xk, Yk, Zk) = Ik (Xk + ΔX, Yk +
ΔY, Zk + ΔZ) processing is performed. f) Total estimation error E by the estimation error calculation process described later.
To calculate. g) Compare the saved total estimation error Es with the total estimation error E calculated in f), and if the total estimation error Es is smaller,
The information saved in d) is restored. That is, position information Pk (xk, yk, zk) = Ps (xs, y
s, zs) current vector Ik (Xk, Yk, Zk) = Is (Xs,
Ys, Zs) Perform processing of total estimation error E = Es.

The estimation error calculation process is as follows. I. From the parameters of each current element, each measurement point j (j = 1,
2. Calculate the magnetic field at N). That is, 1) The following processes 2) and 3) are performed for all measurement points j. 2) The following 3) process is performed for all the current pieces i. 3) Using the Biot-Savart law, the magnetic field Beji (BXeji, BYeji, B) generated by the current element i at the measurement point j
Zeji) is calculated. 4) The magnetic field Bej created by the m current elements at the measurement point j is calculated based on the following equation.

[0005]

[Equation 1]

The calculation of is performed. II. Measured value Bj (BXj, BY at each measurement point j
j, BZj) and the estimated value Ej based on all current elements, the estimated error Ej is calculated, and the total estimated error E is calculated. That is, 5) The following 6) processing is performed for all measurement points j. 6) The estimation error Ej at each measurement point j is calculated based on the following equation. Ej = (BXj-BXej) ² + (BYj-BYej) ² + (B
Zj-BZej) ² 7) Calculate the total estimation error E based on the following equation.

[0007]

[Equation 2]

[0008]

When the magnetic field source is analyzed by using the above apparatus, the parameters of the current element k are changed in small increments so that the total estimation error E is reduced, so that it is finally correct. It seems that the analysis results can be obtained. However, when the initial state of the current element k is set as shown in FIG. 17A, when each of the current elements k reaches the state shown in FIG. 17B when the processing is performed 2400 times. FIG. 17 shows a case where the processing is changed and the processing is performed 3600 times.
Each current element k is only changed to the state shown in (C), and a final solution cannot be obtained. Also, FIG.
7 (B) and 7 (C) are compared, the state of the current element k does not change so much, and there is an inconvenience that a final solution cannot be obtained even if the number of times of processing is increased. Also 3600
It takes about 20 minutes even if a super computer is used to perform the processing once, and there is a disadvantage that it cannot be put to practical use at all.

The occurrence of such an inconvenience is that there is no guarantee that the total estimation error E will be small even if a single process is performed, and that I. 1), 2), 3) and II. The reason is that the parallel processing can be performed only on the portions 5) and 6) and the other processing cannot be performed in parallel. Therefore, even if a parallel processor is used, the speedup of the overall calculation cannot be achieved. Found.

In recent years, research on neural networks has progressed, and it is considered to apply neural networks to the analysis of the magnetic field source. Here, the neural network is roughly classified into a hierarchical perceptron (see FIG. 18) and a pop field model (see FIG. 21). As shown in FIG. 18, the hierarchical perceptron includes an input layer that receives an input pattern, an intermediate layer that includes at least one layer, and an output layer that outputs an output pattern. The neuron elements are connected to each other. Then, a backpropagation rule is used as a learning rule of the hierarchical perceptron. However, if the problem to be dealt with in the hierarchical perceptron becomes complicated, a multi-layered structure is required, or an increase in the number of neuron elements for forming one intermediate layer is required, and the total number of neuron elements is significantly increased. Therefore, there is an inconvenience that the number of weights that need to be determined by learning increases remarkably, and the calculation load for converging the solution becomes enormous. Specifically, FIG.
As shown in (1), normally, by performing learning about 50 times for one pattern, the error sharply decreases and the value converges to the first convergence value. However, if learning is further continued, the error may change abruptly several times. FIG. 20 is a diagram showing a change in error with respect to each pattern, and the error distribution is still concealed even in a portion where there is almost no overall error (see broken line in FIG. 20) (see solid line in FIG. 20). Attempts to reduce the overall error will continue. Therefore, if the desired accuracy cannot be obtained with the first convergence value, a huge amount of calculation is required until the convergence of the next error, and sufficient accuracy can be obtained with the next convergence value. There is no guarantee.

As is clear from the above, it is practically almost impossible to use the hierarchical perceptron for the magnetic field source analysis because of the restriction on the learning process of the physical phenomenon. As shown in FIG. 21, the Popfield model has a configuration in which each neuron model is combined with all other neuron models, and like the hierarchical perceptron, the input layer, the intermediate layer, and the output layer are classified. Not done. All neuron models can function as an input layer, an output layer, or an intermediate layer. Here, assuming that each neuron model is a threshold element model, the state change of the neuron model i is modeled by any one of the equations (3).

[0012]

[Equation 3]

However, Ui is the activity or output value of the neuron model i, hi is the threshold value of the neuron model i, Wij is the weight, and i ≠ j. Then, the function shown by the equation 4 defined by Popfield as the evaluation function for determining the weight of the neuron model and the learning rule is calculated by each change if each neuron model asynchronously changes the internal state based on the equation 3. Four
Is reduced, and the activity or output value of the neuron model converges when it reaches the minimum or minimum value.

[0014]

[Equation 4]

However, αi is a positive constant, and i ≠ j in the first term of the above equation. In addition, when performing analysis using the Popfield model, the following preparatory work from 1) to 4) is essential. That is, 1) The concrete objective function given for analysis is converted into the functional form of Equation 4. 2) Specifically, the independent variables forming the objective function are converted into the activity or output value Ui of the neuron model, and the conversion rule of the activity or output value Ui as given by Equation 3 is determined. 3) Determine an input pattern to each Ui, that is, an initial value of each Ui so that it can finally converge to a certain pattern. 4) Using the learning rules obtained in 1) and 2), the same information processing is repeated until the output pattern converges or the function En becomes the minimum.

Of these preparatory works, the works 1) and 2) are extremely difficult, and there is a high possibility that the functions cannot be converted well, and a remarkably large amount of work is required. Also in the preparatory work of 3), depending on how the initial value is given, it may or may not converge. Therefore, it is difficult to set an initial value with high convergence and depending on how the initial value is given. There are cases where different solutions are obtained. Further, in the preparation work of 4), it takes a considerable time for the non-equilibrium state caused by the change of the internal state of any neuron model to propagate to the new stable state. The time required to obtain the convergence value by performing information processing asynchronously may become very long.

As is apparent from the above, it is practically almost impossible to use the pop field model for magnetic field source analysis. In addition, although the above has described only the case of applying to the magnetic field source analysis, the governing law of the corresponding system such as the pressure source and the temperature source is expressed by a mathematical formula, and the physical source for which linear additivity is satisfied is analyzed. The same disadvantage occurs when applied.

[0018]

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and provides a new physical quantity analysis device capable of easily and rapidly achieving analysis of a physical source based on measured values obtained at a plurality of points. The purpose is to do.

[0019]

In order to achieve the above object, the physical quantity analysis method according to claim 1 is such that a physical quantity measurable at an arbitrary location apart from each physical source is a physical quantity of the physical source and an observation condition. It is possible to calculate based on a predetermined arithmetic expression including, and when the physical addable property of the physical quantity measurable at an arbitrary location apart from the plurality of physical sources is satisfied, the physical quantity of each physical source is determined from the physical source. It is a method of analysis based on the physical quantity measured at a location, and it calculates a plurality of physical formulas that are determined according to the type of physical quantity based on known information, and obtains by cumulatively adding the calculation results of physical formulas. Calculate the difference between the measured value and the measured physical quantity,
Correct the multiple variables included in each physical formula based on the calculated differences, repeat the above series of processes until the differences are sufficiently small, and then use the corrected variables included in each physical formula as the physical quantity analysis results. This is the method of output.

In the physical quantity analyzing device according to the second aspect, the physical quantity measurable at an arbitrary position apart from each physical source can be calculated based on a predetermined arithmetic expression including the physical quantity of the physical source and the observation condition, and A device for calculating the physical quantity of each physical source based on the physical quantity measured at a predetermined location apart from the physical source when linear additivity is established in the physical quantity measurable at any location apart from the multiple physical sources. A physical formula calculation unit that performs a calculation based on the above calculation formula in a number larger than the number of causes of the physical quantity to be analyzed, and a cumulative addition unit that cumulatively adds the calculation results output from each physical formula calculation unit, An error calculator that calculates the error by inputting the cumulative addition result output from the cumulative addition means and the physical quantity measurement value and feeds back the calculated error to each physical formula calculation means. When, and a correction result collecting means for outputting a physical quantity analysis result of the physical source and collect results cause correction based on the calculation error of the physical quantity is performed in each physical formula operating means.

In the physical source analyzer of claim 3, the measurable physical quantity at an arbitrary position apart from each physical source is the quantity of the fluorescent label bound in the vicinity of the optical waveguide by the antigen-antibody reaction, The predetermined formula is an empirical formula that regulates the intensity of the fluorescence emitted from the optical waveguide in response to the introduction of pumping light that propagates while being totally reflected in the optical waveguide. It is a predetermined number that is not less than the number of unknowns contained in the equation, and the correction result collecting means outputs at least the corrected unknowns corresponding to the immunofluorescence and the non-specific adsorption fluorescence as the immunoassay result.

In the physical source analyzing apparatus according to a fourth aspect, a physical quantity measurable at an arbitrary position apart from each physical source is an amount of a substance which is generated or lost in the presence of an enzyme, and a predetermined calculation is performed. The formula is a formula that regulates the intensity of the electric signal output from the base electrode supporting the enzyme-immobilized membrane depending on whether the test solution is spotted directly or indirectly on the enzyme-immobilized membrane. The number of means is a predetermined number that is not less than the number of unknowns included in the formula, and the correction result collecting means outputs at least the corrected unknowns corresponding to the concentration of the substance that performs the enzyme reaction as the concentration measurement result of the corresponding substance. To do.

In the physical quantity analyzing device according to the fifth aspect, the physical quantity measurable at an arbitrary location apart from each physical source can be calculated based on a predetermined arithmetic expression including the physical quantity of the physical source and the observation condition, and This is the case when linear additivity is established for a measurable physical quantity at an arbitrary location away from multiple physical sources, and the physical quantity to be analyzed is the physical property of the physical source. A device for analyzing based on a physical quantity measured at a predetermined location apart from, a transmitting means for radiating a wave to a physical source and a wave reflected from the physical source to receive a physical quantity measurement value. A plurality of wave receiving means to be obtained, and a plurality of physical formula calculation means for performing calculation based on a physical formula corresponding to physical properties,
A cumulative addition means for cumulatively adding the calculation results output from each physical formula calculation means, and an error calculation for calculating an error by inputting the cumulative addition result output from the cumulative addition means and the physical quantity measurement value obtained by the wave receiving means Means, a correction means for correcting the value that defines the cause of the physical quantity in each physical formula calculation means based on the calculation error, and the result corrected by the correction means is collected and output as a physical quantity analysis result of the physical source. And a correction result collecting means.

According to a sixth aspect of the present invention, there is provided a line spectrum noise removing apparatus for removing line spectrum noise from a measurement signal containing line spectrum noise, the number of which corresponds to the type of line spectrum noise. Physical formula calculation means for performing calculation based on a physical formula corresponding to line spectrum noise, cumulative addition means for cumulatively adding calculation results output from each physical formula calculation means, and cumulative addition result output from cumulative addition means And an error calculating means for calculating an error with the measured spectrum signal as an input, and a correcting means for correcting the value defining the cause of the line spectrum in each physical formula calculating means based on the calculated error.

[0025]

According to the physical quantity analysis method of claim 1, the physical quantity at an arbitrary location apart from the physical source can be calculated on the basis of a predetermined arithmetic expression, and the physical quantity of the physical source for which linear additivity is satisfied can be calculated. When performing analysis based on physical quantities measured at a prescribed location away from, calculation of multiple arithmetic expressions (multiple arithmetic expressions with different constant values) that are determined based on the type of physical quantity based on known information The
The estimated value corresponding to the measured value is obtained by cumulatively adding the calculation results of the calculation formula. Then, the difference between the obtained estimated value and the measured physical quantity is calculated, and the variable included in each arithmetic expression is corrected based on the calculated difference, thereby changing the variable to reduce the difference. Then, the variable is further changed to further reduce the difference by repeating the series of processes based on the changed variable. Then, by repeating the above series of processes until the difference becomes sufficiently small, the estimated value can be approximated to the measured value with high accuracy, and thus the corrected variable included in each arithmetic expression at this point is used as the physical quantity analysis result. By outputting it, the analysis result of the physical quantity can be obtained.

In this method, it is only necessary to change the constant, which is included in the known physical formula and should be determined in accordance with the physical quantity, in accordance with the difference between the cumulative addition result and the measurement result. It can be simplified and the analysis time can be greatly shortened. That is, conventional neural
It is also possible to use a net and analyze the physical quantity using the measurement result as a teacher signal, but in this case the learning time is remarkably lengthened because the arithmetic expression itself must be determined by learning. Learning Real
Of course, this cannot be applied when time is required. Moreover, it cannot be applied to the analysis of physical quantities when the physical laws to be handled are complicated. Further, since the arithmetic expression itself is determined by learning, there is no guarantee that the physical quantity analysis result with the desired accuracy can be obtained. Furthermore, it is possible to obtain an arithmetic expression for analyzing the physical quantity from the measurement result based on the known physical formula and analyze the physical quantity,
With the increase in the number of physical sources, the arithmetic formula becomes significantly complicated,
Alternatively, there is a disadvantage that an appropriate arithmetic expression may not be obtained. In addition, this method can be applied only to a specific system, and if the system changes, the arithmetic expression must be reset, so that there is a disadvantage that the versatility is poor. On the other hand, according to the invention of claim 1, the known physical formula or the like is applied as it is to perform the calculation.
Since the calculation formula can be set easily and the calculation result can be obtained accurately, and only the constants that are included in the physical formula and should be determined corresponding to the physical quantity are corrected by learning, the required time can be shortened remarkably.

According to the physical quantity analyzing device of the second aspect, the physical quantity at an arbitrary position apart from the physical source can be calculated based on a predetermined arithmetic expression, and the physical quantity of the physical source satisfying the linear additivity can be calculated. When calculating on the basis of the physical quantity measured at a predetermined location away from, the physical formula calculation means of a number equal to or greater than the number of causes of the physical quantity to be analyzed performs the calculation based on the above calculation formula, and each physical formula calculation The calculation result output from the means is cumulatively added by the cumulative addition means, the cumulative addition result output from the cumulative addition means and the physical quantity measurement value are input, the error is calculated by the error calculation means, and the calculated error is calculated by each physical formula. By feeding back to the calculating means, the correction in each physical formula calculating means is performed. Then, the correction result collecting means collects the result of the correction based on the calculation error of the cause of the physical quantity in each physical formula calculating means, and outputs it as the physical quantity analysis result of the physical source.

More specifically, a physical quantity at an arbitrary location apart from the physical source can be calculated based on a predetermined arithmetic expression, and at the observation position, a physical source that establishes linear additivity of the physical quantity produced by each physical source is established. When calculating the physical quantity based on the physical quantity measured at a predetermined location away from the physical source, there is a large amount of known information such as the measurement time and the measurement position. Each known information is supplied to the physical formula calculating means in a state where the variable included in is set to an arbitrary value to obtain a corresponding calculation result, and the calculation result output from each physical formula calculating means is supplied to the cumulative addition means. Then, a cumulative addition value, that is, a value corresponding to the physical quantity measurement value is obtained. The cumulative addition value obtained in this case is highly likely to be inconsistent with the physical quantity measurement value because the variable is arbitrarily set. However, the error calculating means calculates the cumulative addition value from the actual physical quantity measurement value. To calculate the error between them, and feed back the calculated error to each physical formula calculation means, and correct the set value of the variable corresponding to the error to reduce each error. The variables of the instrument can be changed.

Therefore, by repeating the above-mentioned series of processing a required number of times, the error between the two can be significantly reduced,
At this point, the physical quantity of the physical source can be obtained by collecting and outputting the variables set in each physical formula calculating means by the correction result collecting means. In addition,
In this case, it may seem that the final analysis result will be a local minimum, but since all the variables of each physical formula calculation means are changed based on the calculated error, Variables are changed so that fluctuations are reduced in the entire system, and it is possible to finally obtain accurate analysis results that are not local minimums.

Further, the necessary calculation is only the calculation in each physical formula calculation means, the cumulative addition in the cumulative addition means, the error calculation in the error calculation means, and the correction calculation of the variable by the correction means based on the error fed back. The calculation amount can be significantly reduced as compared with the method. According to the physical quantity analysis device of claim 3, a constant value is obtained by supplying known information (for example, time) obtained after introducing the excitation light to the optical waveguide and starting the immunoassay to a plurality of physical formula calculating means. Perform a calculation of a plurality of empirical formulas different from each other, and cumulatively add the calculation results of the empirical formula by the cumulative addition means to obtain an estimated fluorescence intensity corresponding to the measured fluorescence intensity. Then, the difference between the obtained estimated fluorescence intensity and the measured fluorescence intensity is calculated by an error calculation means, and the calculated error is fed back to each physical formula calculation means to obtain a constant value in each physical formula calculation means. Make corrections.
Then, the above series of processing is repeated until the error calculated by the error calculating means becomes sufficiently small, and at the time when the error becomes sufficiently small, at least the immunofluorescence and the non-specific adsorption fluorescence are corrected by the correction result collecting means. The corresponding corrected constant number is output as the immunoassay result.

Therefore, conventionally, the accuracy of immunoassay is lowered due to the influence of fluorescence or the like caused by the optical waveguide, and
Although it took a long time to obtain the immunoassay result, according to the present invention, it is possible to obtain the immunoassay result with high accuracy in a short time using only the data obtained in the initial stage of the immunoassay. .. According to the physical source analyzer of claim 4, by supplying known information (for example, time) obtained after spotting the test solution onto the enzyme-immobilized membrane to a plurality of physical formula calculation means, the constant value can be calculated. A plurality of mutually different formulas are calculated, and the calculation results of the formulas are cumulatively added by the cumulative addition means and correspond to the measured electric signal corresponding to the amount of the substance produced or lost as a result of the enzymatic reaction of the substance. Obtain an estimated electrical signal. Then, the difference between the obtained estimated electric signal and the measured electric signal is calculated by the error calculating means, and the calculated error is fed back to each physical formula calculating means to correct the constant value in each physical formula calculating means. Let me do it. Then, the series of processes described above is repeated until the error calculated by the error calculating means becomes sufficiently small, and at the time when the error becomes sufficiently small, the correction result collecting means corresponds to at least the concentration of the substance that performs the enzyme reaction. The corrected constant number is output as the result of measuring the concentration of the substance that carries out the enzyme reaction.

Therefore, conventionally, the accuracy of the concentration measurement is lowered due to the influence of the film thickness, the mounting state of the film on the base electrode, etc., and if waiting until the electric signal is stabilized, it takes a long time until the concentration measurement result is obtained. It took a while,
According to the present invention, the concentration measurement result can be obtained with high accuracy in a short time regardless of whether the concentration of the target substance is high or low, using only the data obtained at the initial stage of the concentration measurement.

According to the physical quantity analyzing device of the fifth aspect, the physical quantity at an arbitrary position distant from the physical source can be calculated based on a predetermined arithmetic expression, and is a physical quantity of the physical source for which linear additivity is established, And the physical quantity to be analyzed is a physical property that the physical source has, and when analyzing this physical quantity based on the physical quantity measured at a predetermined location away from the physical source, the physical source is determined by the transmitting means. A wave is radiated to the wave, and the wave reflected from the physical source is received by a plurality of wave receiving means to obtain a physical quantity measurement value. Then, by the physical formula calculation means of a number larger than the number of causes of the physical quantity to be analyzed, the calculation based on the above calculation formula is performed,
The calculation result output from each physical formula calculation means is cumulatively added by the cumulative addition means, the error is calculated by the error calculation means with the cumulative addition result output from the cumulative addition means and the physical quantity measurement value as input, and the calculated error Based on the above, the correction means causes the correction in each physical formula calculation means. Then, the correction result collecting means collects the result of the correction based on the calculation error of the cause of the physical quantity in each physical formula calculating means, and outputs it as the physical quantity analysis result of the physical source.

Therefore, when the physical quantity to be analyzed is an amount that the physical source does not radiate to the outside positively, for example, when the infrared absorption rate, ultrasonic reflectance, etc. are the physical quantity to be analyzed. In addition, by radiating a corresponding wave from the transmitting means and reflecting it by the physical source, it is possible to equivalently generate a state in which the physical source itself radiates some physical quantity to the outside. Highly accurate physical quantity analysis results can be obtained with a very short learning time.

In the line spectrum noise elimination device of the sixth aspect, the physical formula calculation means of the number corresponding to the type of line spectrum noise performs the calculation based on the physical formula corresponding to each line spectrum noise. The calculation results output from the physical formula calculation means are cumulatively added by the cumulative addition means. Then, the cumulative addition result output from the cumulative addition means and the measured spectrum signal are supplied to the error calculation means to calculate the error of the cumulative addition result with respect to the measurement result, and based on the calculated error, each physical formula is calculated by the correction means. The value that defines the cause of the line spectrum in the calculation means is corrected.

After the correction of the value defining the cause of the line spectrum is completed, only the correction by the correction means is interrupted to output a signal obtained by removing the line spectrum noise from the spectrum signal from the error calculation means. Therefore, it is not necessary to collect the values corrected by the respective correction means, the structure can be further simplified as a whole, and a high quality observation signal in which only the line spectrum noise is removed can be obtained.

[0037]

Embodiments will now be described in detail with reference to the accompanying drawings showing embodiments. FIG. 1 is a block diagram showing an embodiment of a physical quantity analyzing apparatus of the present invention. A plurality of physical formula calculation units 11, 12, ... 1 m and a physical formula calculation unit 1 are shown.
1, 12, ... Calculation results g1, g output from 1 m
2, ... Sigma unit 2 that cumulatively adds gm,
Cumulative addition result Oj output from sigma unit 2
(T) and the physical quantity measurement value Sj (t) as a teacher pattern
And an error calculator 3 that calculates the difference between the two, and correction units 11a, 12a that correct the variables estimated in the physical formula calculation unit based on the calculated difference.
... 1 ma and physics formula calculation unit 11, 12, ...
.. The information collection unit 4 that collects the value of the variable estimated at 1 m and outputs it as the analysis result. The physical formula calculation units 11, 12, ...
In response to the supply of known information such as the time t and the measurement position, 1 m performs an operation based on the physical formula set in each physical formula operation unit based on the known information, and from the error calculator 3 Output estimation error dj (t)
In response to the supply of {= Sj (t) -Oj (t)}, the variables included in the physical formula are corrected to reduce the estimation error. Further, the number of physical formula calculation units is set to a number larger than the number of causes of the physical quantity to be analyzed. Furthermore, the physical formula calculation units may be controlled to operate synchronously or asynchronously.

The operation of the physical quantity analyzing device having the above configuration is as follows. The physical quantity Oj (t) to be analyzed is the number 5
Suppose that is given by.

[0039]

[Equation 5]

That is, the physical quantity Oj (t) is a function gi (t, ai1, ai2, ai3, ... aiL) having a time t and L unknowns ai1, ai2, ai3 ,.
Suppose it is a linear sum of m. However, the function gi may include known information in its formula. It should be noted that such assumptions are by no means unnatural, for example, pressure,
This is a hypothesis that holds when analyzing temperature, sound waves, electromagnetic waves, electric fields, magnetic fields, light waves, gravity, particle beams, and the like.

Under this assumption, the known information such as the time t, the measurement position, and the like are provided to m physical formula operation units 11 and 1.
2, ... 1m to supply functions g1, g2, ...
..Calculation of gm and sigma of all calculated function values
Cumulative addition value Oj (t) by supplying to unit 2
Can be obtained. However, since the unknowns are set appropriately at the beginning, the cumulative addition value Oj obtained
(T) is different from the actual measured value Sj (t). Therefore, the error calculator 3 calculates the difference between the actual measured value Sj (t) and the cumulative addition value Oj (t), and the calculated difference is used as the estimated error dj (t) to calculate the physical formula calculation unit 11,1.
2, ... 1 m of correction units 11a, 12a ,.
And the unknown number of each physical formula calculation unit is changed so that the estimation error dj (t) becomes small.

By repeating the above series of processing, the estimation error dj
Since (t) becomes small and the estimation error dj (t) finally becomes almost 0, at this point, the unknown values of the physical formula calculation units 11, 12, ... 1 m are collected by the information collecting unit 4. By outputting, the analysis result regarding the physical quantity of the physical source can be obtained. Further, if the estimation error evaluation function Ej (t) is defined by the following equation, the equation 6 can be obtained. Ej (t) = (1/2) {Sj (t) -Oj (t)} ²

[0043]

[Equation 6]

If the unknowns are corrected in each physical formula calculation unit based on the steepest descent method, the unknowns with the smallest estimation error evaluation function value can be estimated based on the equation 7. However, εk is a learning gain (correction gain) of the unknown aik.

[0045]

[Equation 7]

Further, the relational expressions of Expressions 5 to 8 are obtained.

[0047]

[Equation 8]

Then, by substituting the equation 8 into the equation 7, the equation 9 is obtained.

[0049]

[Equation 9]

Therefore, the physical formula g with a physical model
When the physical quantity represented by j (t, ai1, ai2, ai3, ... aiL) and estimated by Equation 5 can be calculated, the accuracy of estimation of unknowns is improved by performing the processing of Equation 9, and It is possible to obtain an accurate physical quantity. FIG. 2 is a diagram showing an example of the estimation error evaluation function,
An unknown number estimation operation that minimizes the estimation error evaluation function value will be described with Table 1 showing the state of the estimation error evaluation function.
However, Δaik is a correction value for the unknown number aik.

[0051]

[Table 1]

As described above, the unknown value may be estimated so that the value of the estimation error evaluation function becomes small. Therefore, it suffices to pay attention to the sign of the inclination of the estimation error evaluation function Ej (t). When the value is positive, the correction value Δaik may be negative, and when the inclination is negative, the correction value Δ may be positive. Further, since the minimum point is included in FIG. 2, it seems that the unknown number aik corresponding to the minimum point can be obtained when the above-mentioned unknown number is estimated. However, since the estimation process is not performed for only one unknown number but for all unknowns, the estimation error evaluation function itself changes every time the estimation process is performed. Finally, the unknowns that minimize the estimation error evaluation function value can be obtained. Therefore, the analysis of the physical source can be achieved by collecting and outputting the finally obtained unknown number by the information collecting unit 4.

[0053]

SPECIFIC EXAMPLE 1 FIG. 3 is a schematic view showing an example of an immunoassay device using fluorescence. When performing an immunoassay using this immune device, a reaction tank 32 is provided on one side of the optical waveguide 31, and, for example, an antigen 33 is fixed on the boundary surface between the reaction tank 32 and the optical waveguide 31, The test solution 34 is injected into the reaction tank 32 in a state where the excitation light that travels while being totally reflected inside 31 is introduced to cause an antigen-antibody reaction in an amount corresponding to the degree of immunity, and then labeled with a fluorescent substance. Labeled antibody 35
Is injected to excite only the labeled antibody 35 constrained near the boundary surface by the evanescent wave component of the excitation light. The excited fluorescence propagates while being totally reflected inside the optical waveguide 31, and is emitted from the incident end surface 31a of the excitation light. Then, the emitted fluorescence is converted into a beam splitter 3
It is separated from the excitation light by 6 and guided to an optical sensor 37 such as a photomultiplier tube to obtain a measurement signal.

FIG. 4 is a schematic diagram showing the time change of the measurement signal obtained by the above immunoassay device.
Until the injection, the offset due to the fluorescence emitted by the optical waveguide 31 itself is obtained as a measurement signal, and the labeled antibody 3
After the injection of 5, the labeled antibody 35 reacts with the antigen 38 that has undergone the antigen-antibody reaction by the antigen-antibody reaction and is bound to the vicinity of the boundary surface, and the amount of the bound labeled antibody 35 is limited. Since it gradually increases, the measurement signal exponentially increases with the offset as a reference, and finally reaches a predetermined value determined based on the degree of immune reaction.

Therefore, the degree of immunity cannot be accurately measured unless not only the measurement signal finally obtained but also the offset is accurately obtained. In consideration of this point, it is possible to estimate the offset by using the least squares method (first-order regression) of the linear approximation. Cannot be achieved.

However, the amount of immune reaction can be estimated with high accuracy by using the physical source analyzer having the configuration of FIG. That is, it has been experimentally confirmed that the measurement signal corresponding to the emitted fluorescence can be calculated based on Equation 10. However,
A is the sum of the immunofluorescence and non-specific adsorption fluorescence, B is the numerical value depending on the immune antibody concentration, C is the reaction start time, and D is the offset of the optical waveguide 31.

[0057]

[Equation 10]

In this case, since there are four unknowns to be estimated, only four physics formula calculation units are required, and each physics formula calculation unit should perform the process based on the equation (10). Good. Also, each unknown A,
Arbitrary initial values are given to B, C, and D, respectively. It is sufficient to supply the time as a common measurement condition to the physical source analysis apparatus thus initialized, and also to supply the measurement signal obtained corresponding to the supplied time as the teacher signal. The unknowns A, B, C, D can be estimated with high accuracy by performing the processing based on the flowchart shown in FIG. That is, the degree of immune reaction can be estimated with high accuracy.

More specifically, the measured signal is expressed by
Since it is experimentally known that the value can be calculated by 0, the partial differential value of the measurement signal is given by Expression 11.

[0060]

[Equation 11]

As a result, the estimation rules for the unknowns A, B, C, D are as shown in Eq.

[0062]

[Equation 12]

Therefore, in step SP1, an initial value is given to each of the unknowns A, B, C, D, and step S
At P2, an arbitrary time t after the start of the immune reaction and the corresponding measurement signal Sj (t) are extracted, and at step SP3, each physical formula calculation unit 11, 12, based on the current estimated unknowns A, B, C, D. The calculation by 13 and 14 and the cumulative addition by the sigma unit 2 are performed, and the difference Sj between the measurement signal and the cumulative added value is calculated in step SP4.
(T) -Oj (t) is calculated, and in step SP5, it is determined whether or not the calculated difference is equal to or smaller than a preset desired estimation error. Then, when it is determined that the calculated difference is larger than the desired estimation error, the process of Formula 11 is performed in step SP6 to calculate the partial differential value of the cumulative addition value, and the process of Formula 12 is performed in step SP7. Is performed to estimate the unknowns A, B, C, D, and the process of step SP2 is performed again. Step SP5 above
If it is determined that the calculated difference is less than or equal to the desired estimation error set in advance, the series of processes is ended.

The unknowns A, B, C and D can be estimated with high accuracy by executing the series of processes described above. In addition, it is possible to accurately estimate the unknown number using the initial data.
Furthermore, since the process for estimation is simple, the required time can be significantly shortened. Further, estimation of an unknown number in a chemical reaction based on different arithmetic expressions and empirical expressions can be easily dealt with by only changing the data processing part, that is, changing the program. Of course, it is possible to estimate other parameters.

[0065]

SPECIFIC EXAMPLE 2 FIG. 6 is a schematic view showing a main part of a glucose concentration measuring apparatus utilizing an enzymatic reaction. When the glucose concentration is measured using this glucose concentration measuring device, for example, a hydrogen peroxide selective permeable membrane 52, a glucose oxidase (hereinafter, referred to as GOD) is formed on the surface of a base electrode 51 having a platinum electrode 51a and a silver electrode 51b. (Hereinafter, abbreviated as ") is fixed in this order and a diffusion limiting film 54 that limits the diffusion of glucose to some extent in this order, and the test solution is dropped on the diffusion limiting film 54. Diffusion of the glucose in the test solution that has been dropped is limited to some extent by the diffusion limiting film 54, and glucose having a concentration lower than the glucose concentration in the test solution reaches the GOD-immobilized film 53 and is converted in the presence of GOD. Reaction 1 is performed.

[0066]

[Chemical 1]

Hydrogen peroxide generated as a result of this reaction is guided to the surface of the base electrode 51 through the hydrogen peroxide selective permeable film 52, and the base electrode 51 outputs a measurement signal corresponding to the amount of hydrogen peroxide. It Therefore, the glucose concentration can be measured based on the measurement signal. Further, as is clear from Chemical formula 1, since the amount of glucose that can react, that is, the measurement limit of glucose concentration is determined by the amount of oxygen in the test solution, the test solution is dropped directly onto the GOD-immobilized membrane 53. Instead, the diffusion limiting film 54 is interposed to limit the diffusion amount of glucose and raise the measurement limit of glucose concentration.

FIG. 7 is a schematic diagram showing the temporal change of the measurement signal obtained by the above-mentioned glucose concentration measuring device. Until the test solution is dropped, the measurement signal is gradually affected by the preceding measurement. After decreasing the amount of the test solution dropped,
Since the above reaction is performed, the measurement signal increases exponentially and finally reaches a predetermined value determined based on the glucose concentration. However, FIG. 7 shows a state in which the amount of oxygen in the test solution is sufficient.

If the glucose concentration is measured after waiting until the measurement signal does not change, the required time becomes extremely long. Therefore, the maximum value of the time derivative of the measurement signal is obtained to measure the glucose concentration. Is common. Further, since the obtained measurement signal is affected not only by the glucose concentration but also by the thickness of the film laminated on the base electrode 51, the lamination conditions, etc., generally, the known glucose concentration of the known glucose concentration is changed every time the membrane is replaced. The solution is used for calibration.

However, the glucose concentration can be estimated easily and highly accurately by using the physical source analyzer having the configuration shown in FIG. That is, it has been confirmed that the measurement signal can be calculated based on Equation 13 from which the diffusion second equation can be derived.
Here, A is the product of the glucose concentration and the amount determined by the film quality, B is the numerical value related to the rising speed of the reaction, and C is
Indicates the reaction start time, and D indicates the value of the measurement signal at the start of the reaction. It is known that B is large when the film thickness is thin, and B is large when the glucose concentration is high. When B is large, the reaction converges quickly.

[0071]

[Equation 13]

In this case, since there are four unknowns to be estimated, only four physics formula calculation units are required, and if each physics formula calculation unit performs the process based on the equation (13). Good. Also, each unknown A,
Arbitrary initial values are given to B, C, and D, respectively. It is sufficient to supply the time as a common measurement condition to the physical source analysis apparatus thus initialized, and also to supply the measurement signal obtained corresponding to the supplied time as the teacher signal. The unknowns A, B, C, D can be estimated with high accuracy by performing the same processing as the flowchart shown in FIG. That is, the glucose concentration can be estimated with high accuracy.

To explain in more detail, the measured signal is
Since it can be calculated by 3, the partial differential value of the measurement signal is given by Expression 14.

[0074]

[Equation 14]

As a result, the estimation rules for the unknowns A, B, C, D are as shown in equation 15.

[0076]

[Equation 15]

Therefore, in step SP1, initial values are given to the unknowns A, B, C, D, and step S
At P2, an arbitrary time t after the start of the reaction of Chemical formula 1 and the corresponding measurement signal Sj (t) are extracted, and at step SP3, each physical formula calculation unit 11, based on the current estimated unknowns A, B, C, D, The calculation by 12, 13, and 14 and the cumulative addition by the sigma unit 2 are performed, and the difference Sj between the measurement signal and the cumulative added value is calculated in step SP4.
(T) -Oj (t) is calculated, and in step SP5, it is determined whether or not the calculated difference is equal to or smaller than a preset desired estimation error. Then, when it is determined that the calculated difference is larger than the desired estimation error, the process of formula 14 is performed in step SP6 to calculate the partial differential value of the cumulative addition value, and the process of formula 15 is performed in step SP7. Is performed to estimate the unknowns A, B, C, D, and the process of step SP2 is performed again. Step SP5 above
If it is determined that the calculated difference is less than or equal to the desired estimation error set in advance, the series of processes is ended.

The unknowns A, B, C and D can be estimated with high accuracy by executing the series of processes described above. In addition, it is possible to accurately estimate the unknown number using the initial data.
Furthermore, since the process for estimation is simple, the required time can be significantly shortened. Further, estimation of an unknown number in a chemical reaction based on different arithmetic expressions and empirical expressions can be easily dealt with by only changing the data processing part, that is, changing the program. Of course, it is possible to estimate other parameters. Further, the unknown number A is the unknown number A1 corresponding to the glucose concentration and the unknown number A corresponding to the amount determined by the film quality.
It is possible to estimate the unknowns A1 and A2 separately by dividing them into two. In this case, since the glucose concentration can be estimated directly, calibration is not necessary.

Although only the immunoassay and the glucose concentration measurement have been described above, it goes without saying that they can be applied to the estimation of various parameters in other chemical reactions.

[0080]

[Embodiment 2] FIG. 8 is a schematic block diagram showing the configuration of a magnetic field source analysis apparatus using the apparatus of FIG. 6. The difference from FIG. 6 is that N magnetic field sensors MS are arranged to detect a magnetic field. N
Multiplexers MX1 and MX2 whose measurement values and observation conditions are controlled by the control circuit C, respectively.
It is only selected by and supplied to the error calculator 3 as a teacher signal.

Here, of the N magnetic field sensors, the measured value by the jth magnetic field sensor j is Bj, and the magnetic field sensor j
Let Bej be the estimated value at the measurement point j according to Eq.
Is expressed by the following equation. Ej = (1/2) (Bj−Bej) ^{2 If} Bej is formed by a magnetic field generated by m current elements, position information, current vector and measurement point of each current element i. The estimated value Bej can be obtained based on the position information of j and the like. Current element i is at measurement point j
If the magnetic field created at is Beji, then Equation 16 holds. That is, it has linear additivity.

[0082]

[Equation 16]

Further, the input parameter of the current element i is set to W
Shown by ik (k = 1, 2, ... L) {for example, L = 6 when the input parameter includes position information Pi (xi, yi, zi) and current vector Ii (Xi, Yi, Zi) And Wi1 = xi, Wi2 = yi, Wi3 =
zi, Wi4 = Xi, Wi5 = Yi, Wi6 = Zi}, and if the change amount of the input parameter Wik is represented by ΔWik, in order to lead the estimation error Ej to the minimum by using the steepest descent method, the equation 17 is I have to be satisfied.

[0084]

[Equation 17]

Considering these relationships, the estimation error Ej
Equation 18 is obtained as an operation for correcting the input parameter Wik for making the value smaller.

[0086]

[Equation 18]

However, εk is a learning gain, i = 1, 2,
3, ... M, k = 1, 2, 3, ... L. Here, of the N magnetic field sensors, the measurement value by the j-th magnetic field sensor j is Bzj, and the magnetic field component in the z direction formed at the measurement point j by the current element i having the current vector parallel to the xy plane is If Bzeji is used, it is known that the magnetic field component Bzeji is given by the following equation. Bzeji = (μ0 · Mi / 4π) {(yj-yi) c
osθi- (xj-xi) sinθi} / {(xj-x
i) ² + (yj-yi) ² + (zj-zi) ² } ^3/2 where the unknowns xi, yi, zi are the current segments i, respectively.
X, y, z coordinates, the unknown θi is the angle that the current element i makes with the x axis on the xy plane, the unknown Mi is the moment of the current element i, and the unknowns xj, yj, zj are the x of the measurement point j, respectively.
These are the y and z coordinates.

Therefore, by substituting each unknown value into the equation 18, the magnetic field source analysis by the steepest descent method can be performed. That is, Equation 19 is obtained for each unknown.

[0089]

[Formula 19]

Where εx, εy, εz, εT and ε
M is a small positive value, which is a learning gain. Further, the correction processing of the unknown number based on the equation 19 in each physical formula calculation unit is performed synchronously. The number 19
Since each partial differential term included in (1) can be analytically obtained based on the equation (20), the unknown number correction processing based on the equation (19) can be easily achieved. However, in order to simplify the formula, X = xj-xi, Y = yj-yi, Z = zj-z
i, A = X ² + Y ² + Z ² .

[0091]

[Equation 20]

By performing the magnetic field source analysis using the apparatus shown in FIG. 8, it is possible to obtain an unknown number with sufficiently high accuracy with a relatively small number of learnings, and the obtained unknown number is collected by the information collecting unit 4. It can be output as an analysis result.
Further, even though the learning gain for correcting the unknowns is set to a sufficiently small value, when the difference between the measured value Bzj and the estimated value Bzej is large, each unknown can be corrected largely, so that the convergence does not occur. The number of times of learning required for can be reduced. Further, when the difference is small, each unknown value can be corrected by a small amount, so that the accuracy of the finally obtained unknown value can be sufficiently increased. Further, when the difference between the measured value Bzj and the estimated value Bzej is small, and at least one of the partial differential terms included in the mathematical expression 19 has a finite value in the local minima (when it is not a true solution). , It is possible to correct each unknown and get out of the state of local minima. Although the estimation error temporarily increases, each unknown can be corrected so that the error becomes small again.
Therefore, simulated annealing used in other methods
You don't have to. It should be noted that although this embodiment has been described only when applied to the analysis of a magnetic field source, it can be similarly applied to any system in which the physical source radiates some physical quantity (heat, electromagnetic wave, etc.).

[0093]

[Embodiment 3] FIG. 9 is a schematic block diagram showing the configuration of an acoustic exploration apparatus using the apparatus of FIG. 6. The difference from FIG.
One of the n wave detectors is arranged by arranging m wave-receiving sensors i that transmit sound waves from one wave transmitter 7 and receive reflected waves from n reflection points. Of the measured value and its observation condition of the multiplexer M controlled by the control circuit C, respectively.
It is only that it is selected by X1 and MX2 and is supplied to the error calculator 3 as a teacher signal.

Here, assuming that the wave transmitter 7 is a minute point sound source and transmits a short pulse whose time waveform is S (t), as shown in FIG. 5, the reflection point k (k = 1, 2, ...
sound waves reflected by n) are received by the sensor i (i = 1,
2, ... M), the velocity potential φij observed at time j by the wave receiving sensor i is represented by the following equation 21 when the sound potential is sufficiently small. It is known to be determined by

[0095]

[Equation 21]

In the wave receiving sensor i, the sound wave path that produces the velocity potential ψijk is common from the wave transmitter 7 to the reflection point k, and different from the reflection point k to each wave reception sensor i. If the distance from the wave transmitter 7 to the reflection point k is rsk and the distance from the reflection point k to the receiving sensor i is rRki, the velocity potential is attenuated in inverse proportion to the distance. can get. ψijk = αk · S {t− (rsk + rRki) / c}
* (1 / rsk) * (1 / rRki) where c is the speed of sound and αk is a constant proportional to the reflectance at the reflection point k. Further, it is assumed that the conversion between the time t and the time j is performed by t = Δt · j based on the sampling interval Δt. Further, in order to unify the coordinate systems, the coordinates of the transmitter 7 are (xs, ys, zs) and the coordinates of the reflection point k are (x
k, yk, zk), and the coordinates of the receiving sensor i are (xi, y
i, zi), rsk = {(xs-xk) ² + (ys-yk) ² + (zs
-Zk) ² } ^1/2 rRki = {(xi-xk) ² + (yi-yk) ² + (z
i-zk) ² } ^1/2 , and ψijk is four unknowns α
A function {ψijk = ψ (α that has k, xk, yk, zk
k, xk, yk, zk)}. Therefore, it can be applied to the apparatus of FIG. The input pattern is j,
xi, yi, zi, and the teacher pattern is the actual measurement value Φij of φij.

As is clear from the above, the learning rule of the unknown number in each physical formula calculation unit of FIG. 4 may be defined as shown in Formula 22, and the acoustic survey with high accuracy can be achieved with a relatively small number of learning times.

[0098]

[Equation 22]

However, εa, εx, εy, εz are learning gains and are positive constants. The partial differential term of Expression 22 can be calculated by mathematically modifying the expression, but it can also be calculated based on a numerical calculation method. In the case of performing the calculation based on the numerical calculation method, the calculation can be easily performed by the formula 23.

[0100]

[Equation 23]

Although this embodiment has been described only in the case of being applied to acoustic exploration, it can be similarly applied to the case of performing an analysis (infrared absorption characteristic analysis, etc.) based on a physical quantity not radiated by a physical source. ..

[0102]

[Embodiment 4] FIG. 11 is a block diagram showing a line spectrum removing apparatus for removing only the line spectral noise when the frequency band of the signal to be measured and the frequency band of the line spectral noise overlap. The only difference from the device of FIG. 6 is that the information collecting unit 4 is omitted.

The removal of such line spectral noise is performed in order to improve the analysis accuracy of the data processing in the subsequent stage, but conventionally, an analog filter or an adaptive filter is generally used. Of these, the band elimination filter, which is a type of analog filter, has a simple configuration and is inexpensive, but it attenuates not only the line spectrum component of the noise source but also the spectrum component of the measurement target signal, Further, there is a disadvantage that not only the amplitude but also the phase is shifted in the vicinity of the center frequency, which inevitably lowers the analysis accuracy of the measurement target signal. Further, if an adaptive filter is adopted, only the line spectrum component of the noise source can be removed and the analysis accuracy of the measurement target signal can be improved, but a reference channel that measures only the line spectrum component of the noise source is essential. Therefore, there is an inconvenience that the configuration becomes complicated. It is not applicable when it is impossible to measure only the line spectrum component of the noise source.

The line spectrum removing apparatus shown in FIG. 11 solves the above-mentioned inconvenience, and the number of physical formula calculation units 11, 1 is equal to the number of line spectra to be removed.
2, ... The calculation result output from 1 m is supplied to the sigma unit 2, and the cumulative addition result O (t) output from the sigma unit 2 and the physical quantity measurement value S (t) as the teacher pattern are output. The difference d between the two is supplied to the error calculator 3.
(T) is calculated, and the calculated difference d (t) is calculated using the physical formula calculation units 11, 12, ...
a, ... 1ma is fed back and is output as a line spectrum removal signal. The number and frequency of line spectra to be removed are measured in advance. Further, in the correction units 11a, 12a, ... 1ma,
For example, the unknown number included in the physical formula is corrected so that the square of the difference d (t) becomes the minimum value.

The operation of the line spectrum removing device having the above configuration is as follows. Assuming that the frequency fi of the line spectrum, which is a disturbing factor, can be measured by a conventionally known measuring means, the corresponding line spectrum is g (t, Ai, θi) = Ai · sin (2πfit + θ)
It can be expressed in i). Therefore, the cumulative addition result O (t)
Becomes the number 24.

[0106]

[Equation 24]

The difference d (t) is d (t) = S (t)-
It becomes O (t). Furthermore, the partial differential value of the cumulative addition result O (t) is given by Equation 25.

[0108]

[Equation 25]

From the above equations, the equations for correcting the unknowns Ai, θi are: Ai = Ai + εA · d (t) · sin (2πfit + θ)
i) θi = θi + εT · d (t) · Ai · cos (2πfi
t + θi). ΕA and εT are positive constants. Therefore, each physical formula calculation unit 11, 12, ...
The unknowns Ai, θi can be accurately estimated by performing the calculation based on the difference d (t) in the correction units 11a, 12a, ... If the unknowns Ai, θi can be accurately estimated, the corresponding line spectrum noise is output from each of the physical formula calculation units 11, 12, ... 1 m, and the error calculation unit 3 performs the cumulative addition after the cumulative addition unit 2. Since it is subtracted from S (t), the difference d (t) from which only the line spectrum noise component is removed is output.

The following program is a program for estimating the amplitude Ai and the phase θi of one line spectrum. FOR j = 0 TO 20000 Sj = FNF (j, A0, f0, θ0) Oj = FNF (j, Ai, fi, θi) dj = Sj−Oj Ai = Ai + εA * dj * Oj / Ai IF Ai <0 THEN Ai = -Ai: θi = θ
i + π θi = θi-INT (θi / 2π) * 2π IF θi <0 THEN θi = θi + 2π Oj = FNF (j, Ai, fi, θi) D0 = FNG (j, Ai, fi, θi) dj = Sj-Oj θi = θi + εT * dj * D0 θi = θi-INT (θi / 2π) * 2π IF θi <0 THEN θi = θi + 2π NEXT j FNF (j, A, f, θ) = A * sin (2π * f)
* J / 1000 + θ), FNG (j, A, f, θ) = A
* Cos (2π * f * j / 1000 + θ), εA = 0.
003 and εT = 0.000003 are defined respectively.

In the above program, the amplitude of the line spectrum is 48.0000, the phase is 0.9767, and the frequency is 9
When the simulation was performed by setting various values to 0.000000 and performing various simulations, the same estimated value {amplitude Ai
Was 48.0006 or 47.9994, the phase θi was 0.9767, the estimation error dj was about ± SE-4, and the evaluation function value Ej was about 3E-7. It may seem like there are two answers to Ai here,
This calculation is performed in the 32-bit single precision floating point format, and the precision is only about 5.5 decimal digits.

The amplitude of the line spectrum is 72.000.
0, phase 1.3525, frequency 18.000000
As a result, when various initial values were given and the simulation was performed, a total of 10,000 times of the processing was repeated and the results shown in FIG.
The same estimated value {amplitude Ai is 71.998
7, phase θi is 1.3525, estimation error dj is 1.2E
-3, and the evaluation function value Ej was about 1.53E-6}. That is, an accuracy of about 5.5 digits in decimal was obtained.

The following program is a program for estimating the amplitude Ai and the phase θi of a plurality of line spectra. FOR j = 0 TO 10000 Sj = 0 FOR k = 1 TO k1 Sj = Sj + FNF (j, A0 (k), f0 (k), θ
0 (k)) NEXT k FOR k = 1 TO k1 Oj = 0 FOR kL = 1 TO k1 Oj = Oj + FNF (j, Ai (kL), fi (k)
L), θi (kL)) NEXT kL dj = Sj−Oj dA = FNFA (j, Ai (k), fi (k), θi
(K)) Ai (k) = Ai (k) + εA * dj * dA IF Ai (k) <0 THEN Ai (k) = − Ai
(K): θi (k) = θi (k) + π θi (k) = θi (k) -INT (θi (k) / 2π)
* 2 π IF θi (k) <0 THEN θi (k) = θi
(K) + 2π Oj = 0 FOR kL = 1 TO k1 Oj = Oj + FNF (j, Ai (kL), fi (k
L), θi (kL)) NEXT kL dj = Sj−Oj dT = FNFT (j, Ai (k), fi (k), θi
(K)) θi (k) = θi (k) + εT * dj * dT θi (k) = θi (k) -INT (θi (k) / 2π)
* 2 π IF θi (k) <0 THEN θi (k) = θi
(K) + 2π NEXT k NEXT j Note that FNF (j, A, f, θ) = A * sin (2π * f
* J / 1000 + θ), FNFA (j, A, f, θ) =
sin (2π * f * j / 1000 + θ), FNFT
(J, A, f, θ) = A * cos (2π * f * j / 10
00 + θ), εA = 0.005, εT = 0.0000
5, k0 and k1 are defined as the numbers of unknowns.

In the above program, k0 = 7, k1 =
Set to 7, the amplitude and phase of 7 line spectra,
The frequencies are 69.0000, 2.1046 and 3 respectively.
4.000000, 75.0000, 2.66461,
9.000000, 86.0000, 5.0893,4
5.0000, 29.0000, 5.9306, 67.
000000, 90.0000, 0.8073, 20.
000000, 37.0000, 3.4211, 90.
000000, 77.0000, 1.6288, 46.
When the simulation is performed by setting an initial value of 000000 and giving an arbitrary initial value, the amplitude and the phase are 69.0000 and 2.104, respectively, corresponding to each line spectrum.
6, 75.0003, 2.6461, 85.9998,
5.0894, 29.0000, 5.9308, 90.
0000, 0.8073, 36.9999, 3.421
It was estimated to be 1, 77.0002, 1.6288. Further, in these cases, the number of estimation processes was 10,000, the estimation error dj was -4.8828E-4, and the evaluation function value Ej was 2.38419E-7. That is, each unknown value could be estimated with a precision of 6 digits, which is the limit of the 32-bit single precision floating point format. In addition, the estimation process of these seven line spectra is shown in FIG.

In this embodiment, the cumulative addition value O
If (t) is defined as the equation 26, the DC offset can be canceled.

[0116]

[Equation 26]

However, as shown in FIG. 15, A0 = A0 + ε
It is necessary to add a physics formula calculation unit 10 for calculating an unknown value by calculating A · d (t).

[0118]

[Embodiment 5] FIG. 16 is a block diagram showing a magnetic field measuring apparatus. The difference from the embodiment of FIG.
1, the electrocardiogram measurement unit 62, and the obtained electrocardiogram (hereinafter, E
An R wave extraction unit 63 that extracts an R wave based on CG), and a time range output unit 64 that obtains the time ranges of the P wave and the T wave before and after the extracted R wave as a center. The magnetic field (hereinafter, abbreviated as MCG) is supplied to the trigger unit 65 for triggering the R wave of the ECG and the error calculator 3 of the MCG corresponding to the time range of the P wave and the T wave. The only difference is that it further has a prohibited magnetic field limiting unit 66.

The R wave extraction unit 63 uses the Q in the ECG.
Paying attention to the fact that the shape of the RS group is a pulse wave of negative positive and negative,
A rectangular window composed of the first negative wave, the positive wave, and the second negative wave set to a predetermined time ratio (t1: t2: t3) (see FIG. 17).
Cross-correlation with the QRS complex using the reference), and the time corresponding to the maximum value of the cross-correlation is the central time of the QRS complex, that is, R
The time of the wave

As the electrocardiographic field measuring unit 61, the electrocardiogram measuring unit 62 and the trigger unit 65, conventionally known ones can be adopted, and detailed description thereof will be omitted. The operation of the magnetocardiographic field measuring device having the above configuration is as follows. As a result of the inventor's intensive research on the magnetic field, a spectrum of the same frequency as the frequency component of the line spectrum caused by commercial power supply, refrigerator noise, etc. appears at the time of appearance of the R wave or T wave of the ECG, It was found that it acts as a disturbance on the estimation process of line spectrum noise in the magnetic field measurement. Then, when the entire range of the magnetic field is supplied to the error calculator 3 as a teacher signal and the unknown quantity estimation process is performed in each physical formula calculation unit, it is caused by the spectrum appearing at the appearance time of the R wave or T wave. We have found that the time required for the estimation process is significantly longer.

This embodiment was made on the basis of the above-described findings, and the sampling of the magnetic field corresponding to the region from the beginning of the P wave to the end of the T wave is excluded from the unknown value estimation process to improve the convergence. , Remarkably reduce the time required for the estimation process. More specifically, the relationship between the MCG and the ECG in the ideal state that does not include the spectrum as the disturbance is as illustrated in FIGS. 13A and 13B, and the spectrum noise of 29 Hz and 60 Hz is included in the MCG in the ideal state. The waveform when the is mixed is as shown in FIG.

Then, the line spectrum removing apparatus having the configuration of FIG. 11 was used to estimate the unknowns by using the entire range of the waveform of FIG. 18 as the teacher signal, and based on 2500 samples of the teacher signal shown in FIG. Once, twice, three times,
Spectral noise can be largely removed by performing the estimation process four times and five times (FIGS. 19B, 19C, and 19D).
(E) and (F)), and an MCG waveform (see FIG. 14 (F)) quite close to the waveform of FIG. 19 (A) could be obtained.
However, in the waveform of FIG. 19 (F) also, spectral noise remains to some extent, which becomes a cause of lowering the accuracy in the subsequent data processing. Therefore, in order to further reduce the spectral noise, it is necessary to increase the number of estimation processes, and as a result, the required time is greatly increased.

On the other hand, in the case of performing the unknown number estimation process using the magnetocardiographic apparatus having the configuration of FIG.
The time range output unit 6 based on the peak time of the R wave extracted by the R wave extraction unit 63 in the region including the P wave to the T wave of the ECG shown in (B) (see the middle region R1 in FIG. 18B).
By 4, the time range from the beginning of the P wave to the end of the T wave is obtained. Then, the MCG obtained by the cardiac magnetic field measurement unit 61
In addition, the trigger unit 65 triggers in response to the R wave of the ECG, and the magnetocardiographic field limiting unit 66 prohibits the supply of the MGC corresponding to the time range of the P wave and the T wave to the error calculator 3.

In this way, when the unknowns were estimated using the limited magnetic field in the limited range as a teacher signal, once based on 2500 samples of the limited magnetic field in the limited range,
Spectral noise can be largely removed by performing the estimation process twice, three times, four times and five times (FIG. 20 (A)).
(B), (C), (D), (E)), and MCG waveforms (see FIG. 20 (E)) very close to the waveforms of FIG. 20 (A) could be obtained. Further, comparing the waveforms of FIG. 19 and FIG. 20, the waveform of FIG.
It can be seen that it is almost the same as the waveform in the middle of (F) (the estimation processing result for 4.5 times). As is clear from this result, when it is sufficient to achieve the removal of the spectrum noise of the identification degree with the line spectrum removing apparatus having the configuration of FIG. 11, the number of estimation processes can be greatly reduced, and thus the estimation processing time is reduced. Can be greatly shortened. Further, if the estimation processing is performed the same number of times as in the line spectrum removing apparatus having the configuration of FIG. 11, the spectrum noise removing effect can be greatly enhanced, and the accuracy in the subsequent data processing can be significantly enhanced.

Further, the work of collecting only the spectral noise and producing the template is not required, and the workability can be remarkably improved.

[0126]

[Embodiment 6] FIG. 21 is a block diagram showing the configuration of a spectrum analyzer for analyzing a desired frequency component. From the physical formula operation units 11, 12, ... 1 m in the number corresponding to the number of frequencies to be analyzed. The output calculation result is supplied to the sigma unit 2 to perform cumulative addition, and the cumulative addition result O (t) and the measurement data S to be analyzed are obtained.
(T) is supplied to the error calculator 3 to calculate the estimated error d (t), and the estimated error d (t) is used as information for correcting the unknown number in the physical formula calculation units 11, 12, ... Feedback is made to the correction units 11a, 12a, ... 1 ma. Then, when the unknown number estimation processing converges, the information collecting unit 4 extracts the converged unknown number from the physical formula calculation units 11, 12, ... 1 m and outputs it as a frequency component analysis result. Further, in the correction units 11a, 12a, ... 1ma, for example, the unknown number included in the physical formula is corrected so that the square of the disturbance difference d (t) becomes the minimum value.

Conventionally, as a method capable of performing highly accurate analysis of frequency components, a fast Fourier transform (hereinafter referred to as F
FT) is known. However, since FFT is a method based on the sampling theorem,
Usually, it is essential to provide a low-pass filter called an anti-aliasing filter in order to remove unnecessary harmonics included in the measurement signal, which is a disadvantage in that the configuration becomes complicated. Also, if there is no guarantee that the data in the sampling interval is cyclically continuous, the accuracy of the frequency analysis result will drop significantly,
There is also the inconvenience that the types of applicable signals are limited. In order to eliminate such an inconvenience, it has been proposed to use a window function such as Hamming and Hanning. However, if the window function is used at the time of the calculation of the inverse filter, the waveform is distorted over the entire sampling interval after the calculation, and the analysis accuracy becomes There is a new inconvenience of decreasing. Further, since only the output of the frequency interval determined based on the sampling interval can be obtained, there is also a disadvantage that the number of samples must be increased inevitably when wide band analysis is required. Furthermore, even if the frequency axis is a logarithmic scale, sampling must be performed at equal intervals, so that a remarkably large amount of memory is required, and the number of samples must also be 2 ^N. is there.

This embodiment is designed to eliminate these disadvantages, does not require an anti-aliasing filter, can use sample data at unequal intervals, and can set the analysis frequency arbitrarily. can do. More specifically, Expression 27 can be used as a model of the component of the analysis target signal having the frequency fi.

[0129]

[Equation 27]

However, Ai and Bi are constants indicating the amplitude, respectively. Further, the relationship of Expression 28 is obtained from the above equation.

[0131]

[Equation 28]

Substituting these equations 28 into equation 13 yields equation 2
9 is obtained.

[0133]

[Equation 29]

Therefore, the sigma unit 2 calculates O (t), the error calculator 3 calculates the estimated error d (t), and each physical formula calculation unit calculates the estimated error d (t). Based on each unknown A
i, Bi, C may be corrected and a physical formula may be calculated based on the corrected unknowns. When the estimation of the unknowns Ai, Bi, C converges, the information collecting unit 4 converges each. By collecting the unknowns, it is possible to accurately analyze an arbitrary frequency component included in the analysis target signal.

[0135]

As described above, according to the first aspect of the present invention, the influence due to the physical quantity of the physical source is known, and in a system where linear additivity is established, the physical quantity can be easily and accurately based on the measured value. It has a unique effect of being able to analyze. According to the second aspect of the present invention, the influence of the physical quantity of the physical source is known, and the physical quantity can be accurately analyzed in the system in which the linear additivity is established for the influence of the plurality of physical sources. The unique effect is that it is possible to easily deal with the increase or decrease in the estimated number of physical sources to be analyzed.

The invention of claim 3 has a unique effect that an immunoassay result can be obtained with high accuracy in a short time by using only data obtained in the initial stage of the immunoassay. The invention of claim 4 has a unique effect that the concentration measurement result can be obtained with high accuracy in a short time regardless of whether the concentration of the target substance is high or low, using only the data obtained at the initial stage of the concentration measurement.

According to the fifth aspect of the present invention, the influence due to the physical properties of the physical source is known, and in the system where the linear additivity is established with respect to the influences due to the plurality of physical sources, the physical properties are accurately determined. It is possible to perform an analysis, and it is possible to easily deal with an increase or decrease in the estimated number of physical sources to be analyzed. According to the invention of claim 6, it is necessary to accurately analyze the line spectrum noise and obtain a high quality observation signal in which only the line spectrum noise is removed, and only the analysis result of the line spectrum noise needs to be collected. Since there is no such a case, it has a unique effect that the configuration can be simplified.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a physical quantity analysis device of the present invention.

FIG. 2 is a diagram showing an example of an estimation error evaluation function.

FIG. 3 is a schematic diagram showing an example of an immunoassay device using fluorescence.

FIG. 4 is a schematic diagram showing a temporal change of a measurement signal obtained by the immunoassay device of FIG.

5 is a flowchart illustrating an immunoassay process using the physical quantity analysis device of FIG.

FIG. 6 is a schematic view showing an example of a glucose concentration measuring device.

7 is a schematic diagram showing a temporal change of a measurement signal obtained by the glucose concentration measuring device of FIG.

FIG. 8 is a schematic block diagram showing the configuration of a magnetic field source analysis apparatus using the apparatus of FIG.

FIG. 9 is a schematic block diagram showing a configuration of an acoustic exploration apparatus using the apparatus of FIG.

FIG. 10 is a diagram schematically showing a relationship among a transmission point, a reflection point, and a reception point.

FIG. 11 is a block diagram showing a line spectrum removing device.

FIG. 12 is a diagram illustrating a convergence state of one line spectrum noise corresponding to a change in initial value.

FIG. 13 is a diagram illustrating a convergence state of a plurality of line spectrum noises.

FIG. 14 is a diagram illustrating a convergence state of a plurality of line spectrum noises.

FIG. 15 is a block diagram showing a line spectrum removing device having a DC offset removing function.

FIG. 16 is a block diagram showing a magnetic field measuring apparatus.

FIG. 17 is a diagram showing an example of a rectangular window for extracting a QRS complex in EGC.

FIG. 18 is a diagram showing waveforms of MGC and EGC.

FIG. 19 shows an MG corresponding to an increase in the number of times of learning in the case of performing a cardiac magnetic field measurement using the entire range of MGC as a teacher signal.
It is a figure which shows C waveform.

20 is a diagram showing an MGC waveform corresponding to an increase in the number of times of learning when a cardiac magnetic field is measured by the cardiac magnetic field measuring device of FIG.

FIG. 21 is a block diagram showing a configuration of a spectrum analyzer that analyzes a desired frequency component.

FIG. 22 is a diagram showing a magnetic field source analysis result by a conventional method.

FIG. 23 is a diagram schematically showing a configuration of a hierarchical perceptron.

FIG. 24 is a diagram showing a relationship between the number of times of learning and an error in the hierarchical perceptron.

FIG. 25 is a diagram showing a change in an error with respect to each pattern corresponding to the number of times of learning in the hierarchical perceptron and an error as a whole.

FIG. 26 is a diagram schematically showing a Popfield model.

[Explanation of symbols]

11, 12 ... 1m Physical formula calculation unit 11a, 12a, ... 1ma Correction unit 2 Sigma unit 3 Error calculation unit 4 Information collection unit 7 Transmitter i Receiving sensor 31 Optical waveguide 35 Labeled antibody 51 Base Electrode 53 GOD immobilization membrane

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Office reference number FI technical display location G01R 33/035 ZAA 8203-2G

Claims (6)

[Claims] 1. A physical quantity measurable at an arbitrary location apart from each physical source can be calculated based on a predetermined arithmetic expression including the physical quantity of the physical source and observation conditions, and the physical quantity can be separated from a plurality of physical sources. A method of analyzing the physical quantity of the physical source based on the physical quantity measured at a predetermined location apart from the physical source when the linear additivity is satisfied in the measurable physical quantity at an arbitrary location, and the physical quantity is based on known information. Calculates the difference between the value obtained by cumulatively adding the calculation results of each calculation formula and the measured physical quantity, and calculates each calculation formula based on the calculated difference. It corrects multiple variables included in, and repeats the above series of processing until the difference becomes sufficiently small, and then outputs the corrected variables included in each physical formula as the physical quantity analysis result. Physical analysis method. 2. A physical quantity measurable at an arbitrary position apart from each physical source can be calculated based on a predetermined arithmetic expression including the physical quantity of the physical source and observation conditions, and an arbitrary distance from a plurality of physical sources. This is a device that analyzes the physical quantity of each physical source based on the physical quantity measured at a predetermined location away from the physical source when the linear additivity is established for the measurable physical quantity at the location Physical formula calculation means (11) (12) ... (1
m) and each physics formula calculation means (11) (12) ...
The cumulative addition means (2) for cumulatively adding the calculation result output from (1 m), and the error calculation means () for calculating the error by inputting the cumulative addition result output from the cumulative addition means (2) and the physical quantity measurement value. 3) and the correction means (11a) (12) for correcting the cause of the physical quantity in each physical formula calculation means (11) (12) ... (1m) based on the calculation error.
a) ... (1 ma) and correction means (11 a) (12
a) A physical quantity analyzing device comprising: a correction result collecting means (4) for collecting the results of the correction by (1 ma) and outputting the results as a physical quantity analysis result of the physical source. 3. A physical quantity measurable at an arbitrary location apart from each physical source is the quantity of the fluorescent label (35) bound in the vicinity of the optical waveguide (31) by an antigen-antibody reaction, and is a predetermined quantity. The calculation formula is an empirical formula that defines the intensity of the fluorescence emitted from the optical waveguide (31) in response to the introduction of the excitation light that propagates while being totally reflected in the optical waveguide (31). (11) (12) ・ ・ ・ (1
The number of m) is a predetermined number not less than the number of unknowns included in the empirical formula, and the correction result collecting means (4) immunoassays the corrected unknowns corresponding to at least the immunofluorescence and the non-specific adsorption fluorescence. The physical quantity analysis device according to claim 2, which outputs the result. 4. A physical quantity measurable at an arbitrary location apart from each physical source is an amount of a substance produced or lost in the presence of an enzyme, and a predetermined arithmetic expression is an enzyme-immobilized membrane ( 53) is a formula that defines the intensity of the electric signal output from the base electrode (51) supporting the enzyme-immobilized membrane (53) depending on whether the test solution is spotted directly or indirectly on the 53) physical formula. Computing means (11) (12)
..The number of (1 m) is not less than the number of unknowns included in the formula, and the correction result collecting means (4) corresponds to at least the corrected unknown corresponding to the concentration of the substance that carries out the enzyme reaction. The physical quantity analysis device according to claim 2, which is output as a concentration measurement result of the substance. 5. A physical quantity measurable at an arbitrary location apart from each physical source can be calculated based on a predetermined arithmetic expression including the physical quantity of the physical source and observation conditions, and an arbitrary location separated from a plurality of physical sources. When the physical quantity measurable at a location has linear additivity, and the physical quantity to be analyzed is the physical property of the physical source, and this physical quantity is measured at a predetermined location away from the physical source. A device for analyzing based on a physical quantity, the transmitting means (7) for radiating a wave to a physical source, and a plurality of receiving means for receiving a wave reflected from the physical source to obtain a physical quantity measurement value. Wave means (i) and a plurality of physical formula calculation means (11) (12) for performing calculation based on a physical formula corresponding to physical properties
.. (1m) and each physical formula calculation means (11) (12)
... Cumulative addition means (2) for cumulatively adding the calculation result output from (1 m), the cumulative addition result output from the cumulative addition means (2) and the physical quantity measurement value obtained by the wave receiving means (i) Error calculation means (3) for calculating an error by inputting and and each physical formula calculation means (1
1) (12) ... Correction means (11a) (12a) for correcting the value that defines the cause of the physical quantity in (1m)
... (1 ma) and correction means (11 a) (12 a)
.. A physical quantity analysis device, comprising: a correction result collecting means (4) for collecting the result corrected by (1 ma) and outputting the result as the physical quantity analysis result of the physical source. 6. An apparatus for removing line spectrum noise from a measurement signal containing line spectrum noise, comprising:
Physical formula calculation means (11) (12) ... (1 m) for performing calculation based on the physical formula corresponding to the line spectrum noise in the number corresponding to the type of line spectrum noise.
And each physics formula calculation means (11) (12) ... (1
m)) cumulative addition means (2) for cumulatively adding the calculation results output, and error calculation means for calculating an error by inputting the cumulative addition results output from the cumulative addition means (2) and the measured spectrum signal. (3) and correction means (11a) (12a) for correcting the value that defines the cause of the line spectrum in each physical formula calculation means based on the calculation error.
.. (1 ma) is included, The line spectrum noise removal apparatus characterized by the above-mentioned.