A component waveform extraction system, a component waveform extraction method, and a program for operating a computer as a component waveform extraction system; and a signal analysis system, a signal analysis method, and a program for operating a computer as a signal analysis system.

Abstract

To provide a signal waveform analysis system and the like capable of acquiring a targeted component waveform included in a measured signal waveform and quantitatively analyzing each feature of the acquired component waveform.SOLUTION: A component waveform extraction system comprises: an analysis unit 10 which calculates a wavelet coefficient group of a signal waveform and obtains a coefficient group of a coefficient aggregate, on the basis of a wavelet coefficient aggregate; and a component waveform extraction unit 20 which calculates the adaptation of a component waveform relative to the coefficient group of the coefficient aggregate, detects the component waveform on the basis of the adaptation, waveform-reproduces the coefficient group corresponding to the component waveform by inverse wavelet transformation, and thereby obtains the component waveform included in the signal waveform.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a component waveform extraction system and method for analyzing mixed signal waveforms consisting of irregularly occurring single waves, and to a signal analysis system and method for analyzing component waveforms of such signal waveforms. [Background technology]

[0002] Sometimes, the goal is to understand the behavior of a component waveform, which is formed by the superposition of multiple single waves with poor regularity, such as periodicity, contained within a given signal waveform. Examples of signal waveforms targeted for such analysis include silent voice recognition (so-called lip-syncing), heart sounds, electrical waveforms from electrocardiograms, electromyographic waveforms that capture muscle activity, and, in engineering terms, abnormal waveforms such as noise, vibration, or equipment malfunction detection generated from equipment or various environments.

[0003] The act of trying to obtain information from observed signal waveforms is extremely common. In this case, if the signal waveform is composed of or contains smaller, finite-length unit waveforms (component waveforms), it is very natural to want to capture the information that the signal waveform shows with higher accuracy by examining the behavior of the component waveforms. However, the more complex the situation becomes—for example, if the shape of the component waveforms is not simple, if the time distribution of the frequency components is uneven, if there is a range of allowable changes in the component waveforms, if various waveforms other than the component waveforms are included, if there is no periodicity or regularity in the appearance of the component waveforms, or if multiple component waveforms with time differences overlap—the more difficult it becomes to detect the component waveforms and capture their behavior.

[0004] To grasp the behavior of such component waveforms, there are methods that use multi-channel measurement data with different measurement conditions according to specific rules and estimate it using an analytical model. However, there are many challenges such as the cost of multi-channel measurement, the accuracy of the model, and the computational cost, so it is generally necessary to compromise by obtaining the characteristics of the signal waveform using abstract features. Examples of signal waveforms that can be analyzed in this way include the activity and sounds of various muscles, electromyographic waveforms such as electrocardiogram waveforms and heart sounds that are difficult to capture as a simple sine wave synthesis, as well as operating sounds and vibrations generated by equipment, and abnormal waveforms that suggest malfunctions, etc.

[0005] These waveforms contain various other waveforms besides the component waveform to be analyzed, and in order to perform the analysis, it is first necessary to extract the relevant component waveform from among them. Furthermore, this component waveform, along with the signal waveform, is not simple, especially since its frequency is subject to time fluctuations, making it difficult to understand its behavior while taking into account frequency fluctuations over time.

[0006] Here, we will explain the analysis of such signal waveforms using electromyography (EMG) measurement and analysis techniques as a concrete example. A single muscle is composed of numerous muscle fibers. However, the activity of each muscle fiber is not independent; rather, the activity is managed by a group of muscle fibers innervated by a single nerve cell (hereinafter also referred to as a "motor unit"), and muscle activity is achieved when each motor unit operates in overlapping sequence with a time lag. Assuming that the muscle fibers constituting a single motor unit operate simultaneously, to capture detailed muscle activity, it is sufficient to capture the activity of the motor units, and the motor unit action potential waveform, which is the superposition of potential changes generated in each muscle fiber, serves as the tool for this purpose.

[0007] Electromyography (EMG) measurement can be broadly divided into two main methods: needle EMG and surface EMG, which measure the movement of muscles that make up the human skeleton. Needle EMG involves inserting electrode needles into the muscle, allowing for fairly direct measurement of the action potential of the target motor unit. However, it is painful and difficult to apply during exercise, making it impractical for casual use. Furthermore, the data obtained is only local information near the insertion site, and is insufficient to capture the activity of the entire muscle. In contrast, surface EMG is a non-invasive measurement method that simply involves attaching electrodes to the skin surface, making it easy to handle and capable of capturing motor unit activity over a wide area. However, it measures the superposition of action potentials of multiple motor units.

[0008] To capture detailed muscle activity, it is desirable to obtain such motor unit activity, and capturing the resulting changes in electrical potential is the most efficient method. Furthermore, obtaining motor unit activity from surface electromyography, which contains a wealth of information, would be extremely useful. However, conventional techniques, as mentioned above, mostly rely on multi-channel signals and models for estimation, and it is difficult to say that they directly acquire differences in motor unit types or activity timing.

[0009] Regarding techniques related to the analysis of muscle activity, the following documents have been disclosed: Non-Patent Document 1 concerns general information related to muscle tissue and muscle activity. Non-Patent Document 2 is an attempt at evaluation by measurement at multiple poles, and basically involves relative comparison during the same type of exercise. Non-Patent Document 3 states that "there are no research reports on the relationship between power, muscle fatigue, and muscle activity from surface electromyography." [Prior art documents] [Patent Documents]

[0010] [Patent Document 1] Japanese Patent Publication No. 2018-047230 [Patent Document 2] Japanese Patent Publication No. 2002-272692 [Patent Document 3] Japanese Patent Publication No. 2007-202612 [Patent Document 4] Japanese Patent Publication No. 2016-063995 [Patent Document 5] WO2004 / 023996 publication [Patent Document 6] Special Publication No. 05-032057 [Patent Document 7] Japanese Patent Publication No. 2013-244027 [Non-patent literature]

[0011] [Non-Patent Document 1] Tetsuo Fukunaga (ed.), "Dictionary of Muscle Science: Structure, Function, and Movement," Asakura Shoten (2002). [Non-Patent Document 2] Rei Shirai, Kazuyuki Mito, "Evaluation of Muscle Fatigue During Dynamic Exercise Using Surface Electromyography with Multi-Point Electrodes," IEICE Technical Report Vol. 114 No. 79, MBE2014-16, pp. 19-22 (2014). [Non-Patent Document 3] Tatsushi Tokuyasu, Kodai Matsuno, and Shinpei Matsudaira, "Quantitative Evaluation of Muscle Fatigue by Pattern Entropy Analysis of Surface Electromyography," Proceedings of the 2012 Robotics and Mechatronics Conference of the Japan Society of Mechanical Engineers, 2A2-B04 (2012). [Non-Patent Document 4] Hidetoshi Nagai, Computational Complexity in Continuous Estimation of Muscular Activity based on Redundant Wavelet Coefficients of Surface EMG, Proc. of Life Engineering Symposium 2015, pp. 329-334, 2015. [Non-Patent Document 5] Hidetoshi Nagai, "Shift Selection Method for Wavelet Coefficient Set for Feature Extraction in Surface Electromyography," IEICE Technical Report, vol. 119, no. 391, MBE2019-78, pp. 51-56 (2020) [Overview of the Initiative] [Problems that the invention aims to solve]

[0012] Waveforms that cannot be expected to be generated completely discretely as single pulses at a fixed frequency include, for example, the motor unit activity waveforms included in the aforementioned electromyography. In such waveforms, the signal components of the single pulses contained within the waveform exist with temporal and frequency spread.

[0013] Therefore, as disclosed in Patent Documents 1 to 3, when displaying the continuous wavelet transform results in two or three dimensions, it is difficult to determine which parts are part of the single-wave signal component and which parts are noise components.

[0014] When performing signal analysis on a computer, discrete data obtained through sampling is used, and wavelet analysis on discrete data is called discrete wavelet analysis. When simply using the results of a typical multi-resolution analysis in discrete wavelet analysis, it becomes extremely difficult to identify single waves on the time-frequency plane due to constraints on the division of the time-frequency plane.

[0015] As described in Patent Document 4, redundant discrete wavelet analysis can avoid the problems of multi-resolution analysis, but problems similar to those of continuous wavelet transform remain. In this regard, there are methods that use multi-channel signals and models or special measurement conditions to enable the detection and separation of elemental signals from a single-channel signal (for example, motor unit action potentials, which are constituent elements, from a single-channel surface electromyographic signal) (see Patent Documents 5 and 6).

[0016] However, these methods require complex calculations, and the accuracy of elemental signal estimation decreases when signals containing elements that do not fit the model's conditions, such as noise, are mixed in. A more serious problem is that because multi-channel feature information is used during estimation, analysis based on the obtained elemental signals cannot be performed from a multi-channel perspective. If extraction and separation from a single-channel signal can be achieved, technologies aiming for analysis utilizing multiple channels (for example, Patent Document 7) can be further developed.

[0017] Regarding the aforementioned non-patent documents, Non-Patent Document 4 is a fundamental paper on algorithms for redundant discrete wavelet analysis. Non-Patent Document 5 is a paper that attempts to generalize the practice of shifting the wavelet coefficient extraction time. While it is a fundamental paper for the current wavelet coefficient set, it does not clearly discuss the relationship between the coefficient set and component waveforms, and has not reached the point of concluding that the coefficient set leads to component waveform extraction.

[0018] This invention has been made in view of the above circumstances, and aims to provide a signal waveform analysis method, system, and program that can acquire target component waveforms included in a measured signal waveform and quantitatively analyze the characteristics of each of these acquired component waveforms. [Means for solving the problem]

[0019] The inventors of this invention have conducted extensive research to solve the above problems and have found that the following invention is suitable for the above purpose, leading to the present invention. That is, the present invention relates to the following invention.

[0020] <1> An analysis unit that calculates a set of wavelet coefficients of a signal waveform and obtains a set of coefficients based on the wavelet coefficient set, The system includes a component waveform extraction unit that calculates the degree of fit of the component waveform to the set of coefficients, detects the component waveform based on the degree of fit, and obtains the component waveform included in the signal waveform by reconstructing the coefficient group corresponding to the component waveform using an inverse wavelet transform. Component waveform extraction system. <2> An analysis unit that calculates a set of wavelet coefficients of a signal waveform and obtains a set of coefficients based on the wavelet coefficient set, The system includes a feature analysis unit that calculates the degree of fit between the coefficient set and the component waveform, detects the component waveform based on the degree of fit, and analyzes feature quantities based on the characteristics of the component waveform. Signal analysis system. <3> The degree of fit is obtained based on the fundamental characteristics of the component waveforms in the coefficient set. <1> The component waveform extraction system described above. <4> The aforementioned basic characteristics are fundamental vectors that represent the characteristics of the component waveforms, The degree of fit is obtained by the basis vector and the vector based on the set of coefficients of the coefficient set. <3> The component waveform extraction system described above. <5> For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. <1> , <3> , and, <4> A component waveform extraction system as described in any of the following. <6> Using the aforementioned degree of fit and an adjustment filter weighted based on the fundamental characteristics of the component waveforms, the multiple waveforms and component waveforms included in the coefficient set are separated and adjusted into a wavelet coefficient group corresponding to the component waveforms. <1> , and, <3> ~ <5> A component waveform extraction system as described in any of the following. <7> The adjustment filter has a distribution type extraction filter that distributes the set of coefficients of the coefficient set to the set of coefficients of the coefficient set corresponding to each of the multiple component waveforms detected based on the goodness of fit. <6> The component waveform extraction system described above. <8> The adjustment filter applies a derivation-type extraction filter that derives a group of wavelet coefficients corresponding to the component waveform detected based on the goodness of fit from the coefficient set. <6> or <7> The component waveform extraction system described above. <9> The system includes a feature analysis unit that analyzes feature quantities based on the characteristics of the component waveforms detected by the component waveform extraction unit, The goodness of fit is corrected by feature quantities that indicate the intensity and / or properties of the component waveforms. <1> , and, <3> ~ <8> A component waveform extraction system as described in any of the following. <10> The feature quantity indicating the intensity is a signal intensity feature quantity indicating the intensity of the signal of the component waveform, and the feature quantity indicating the property is a frequency feature quantity based on the distribution of frequency components contained in the component waveform. <9> The component waveform extraction system described above. <11> The feature analysis unit obtains multiple component waveforms included in the signal waveform as a single block based on the degree of fit. <9> or <10> The component waveform extraction system described above. <12> The analysis unit calculates wavelet coefficient sets for each of the multiple sub-waveform components included in the signal waveform and obtains a set of coefficients based on each wavelet coefficient set, and the component waveform extraction unit calculates the degree of fit of the component waveform to each of the coefficient sets. <9> or <10> The component waveform extraction system described above. <13> The analysis unit calculates a wavelet coefficient set by concatenating multiple sub-waveform components included in the signal waveform into a single component waveform, and obtains a set of coefficients based on the wavelet coefficient set. <9> or <10> The component waveform extraction system described above. <14> The degree of fit is obtained based on the fundamental characteristics of the component waveforms in the coefficient set. <2> The signal analysis system described above. <15> The aforementioned basic characteristics are fundamental vectors that represent the characteristics of the component waveforms, The degree of fit is obtained by the basis vector and the vector based on the set of coefficients of the coefficient set. <14> The signal analysis system described above. <16> For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. <2> , <14> , and, <15> A signal analysis system as described in any of the following. <17> Using the aforementioned degree of fit and an adjustment filter weighted based on the fundamental characteristics of the component waveforms, the multiple waveforms and component waveforms included in the coefficient set are separated and adjusted into a wavelet coefficient group corresponding to the component waveforms. <2> , and, <14> ~ <16> A signal analysis system as described in any of the following. <18> The adjustment filter has a distribution type extraction filter that distributes the set of coefficients of the coefficient set to the set of coefficients of the coefficient set corresponding to each of the multiple component waveforms detected based on the goodness of fit. <17> The signal analysis system described above. <19> The adjustment filter applies a derivation-type extraction filter that derives a group of wavelet coefficients corresponding to the component waveform detected based on the goodness of fit from the coefficient set. <17> or <18> The signal analysis system described above. <20> The goodness of fit is corrected using the feature quantities that indicate the intensity and / or properties of the component waveforms obtained by the feature analysis unit. <2> , and, <14> ~ <19> A signal analysis system as described in any of the following. <21> The feature quantity indicating the intensity is a signal intensity feature quantity indicating the intensity of the signal of the component waveform, and the feature quantity indicating the property is a frequency feature quantity based on the distribution of frequency components contained in the component waveform. <20> The signal analysis system described above. <22> The feature analysis unit obtains multiple component waveforms included in the signal waveform as a single block based on the degree of fit. <20> or <21> The signal analysis system described above. <23> The feature analysis unit includes a block variation analysis unit that analyzes the average variation and the amount of variation change of the component waveforms within the block, and analyzes the coefficient of variation and degree of variation of the block based on the average variation and the amount of variation change. <22> The signal analysis system described above. <24> The block variation analysis unit estimates the activity level of the component waveform in response to block variation based on the degree of variation of the block. <23> The signal analysis system described above. <25> The feature analysis unit includes a block output characteristic analysis unit that analyzes the activity coefficient and activity level of the block based on the variability of the block and the signal intensity features of the component waveforms constituting the block. <22> ~ <24> A signal analysis system as described in any of the following. <26> The feature analysis unit includes a block output fluctuation characteristic analysis unit and a convergence degree analysis unit that analyze the activity reserve of blocks based on the occurrence rate and characteristics of the blocks. <22> ~ <25> A signal analysis system as described in any of the following. <27> The block output characteristic analysis unit obtains component waveforms exhibiting different characteristics from among the multiple component waveforms included in the signal waveform as blocks, and analyzes the activity level based on the degree of variation of the block based on the different characteristics and the signal intensity features of the component waveforms constituting the block. <25> The signal analysis system described above. <28> The feature analysis unit obtains component waveforms exhibiting different characteristics from among the multiple component waveforms included in the signal waveform as blocks, and analyzes the activity capacity based on the frequency of occurrence of these blocks and the characteristics of the blocks. <25> The signal analysis system described above. <29> The component waveform extraction system allows The steps include: calculating the wavelet coefficient set of the signal waveform, and obtaining a set of coefficients based on the wavelet coefficient set; The process includes the steps of: calculating the degree of fit of the component waveform to the set of coefficients; detecting the component waveform based on the degree of fit; and obtaining the component waveform included in the signal waveform by reconstructing the coefficient group corresponding to the component waveform using an inverse wavelet transform. Component waveform extraction method. <30> The signal analysis system, The steps include: calculating the wavelet coefficient set of the signal waveform, and obtaining a set of coefficients based on the wavelet coefficient set; The method comprises the steps of calculating the degree of fit between the coefficient set and the component waveform, detecting the component waveform based on the degree of fit, and analyzing feature quantities based on the characteristics of the component waveform. Signal analysis method. <31> Computers, An analysis unit that calculates a set of wavelet coefficients of a signal waveform and obtains a set of coefficients based on the wavelet coefficient set, The system includes a component waveform extraction unit that calculates the degree of fit of the component waveform to the set of coefficients, detects the component waveform based on the degree of fit, and obtains the component waveform included in the signal waveform by reconstructing the coefficient group corresponding to the component waveform using an inverse wavelet transform. A program that operates as a component waveform extraction system. <32> Computers, An analysis unit that calculates a set of wavelet coefficients of a signal waveform and obtains a set of coefficients based on the wavelet coefficient set, The system includes a feature analysis unit that calculates the degree of fit between the coefficient set and the component waveform, detects the component waveform based on the degree of fit, and analyzes feature quantities based on the characteristics of the component waveform. A program that operates as a signal analysis system. [Effects of the Invention]

[0021] According to the present invention, it is possible to acquire target component waveforms included in the measured signal waveform, and to quantitatively analyze and evaluate the characteristics of each of these acquired component waveforms.

[0022] According to the present invention, the behavior of a waveform consisting of component elements whose frequency fluctuates over time can be evaluated in detail using a signal waveform acquired by a simple single-channel measurement system. This can also be applied to understanding the behavior of component waveforms in real time. [Brief explanation of the drawing]

[0023] [Figure 1] This diagram illustrates the evaluation of the characteristics of component waveforms. [Figure 2] This diagram illustrates the basis vectors projected onto a hypersphere. [Figure 3] This is a diagram to explain the analysis based on distance. [Figure 4] This figure shows the analysis flow of the component waveform extraction system according to an embodiment of the present invention. [Figure 5] This figure shows the analysis flow of a signal waveform analysis system according to an embodiment of the present invention. [Figure 6] This is a diagram illustrating an experiment conducted using the present invention. [Figure 7] This is a diagram illustrating an experiment conducted using the present invention. [Figure 8] This is a diagram illustrating an experiment conducted using the present invention. [Figure 9] This is a diagram illustrating an experiment conducted using the present invention. [Figure 10] This is a diagram illustrating an experiment conducted using the present invention. [Figure 11] This is a diagram illustrating an experiment conducted using the present invention. [Figure 12] This is a diagram illustrating an experiment conducted using the present invention. [Figure 13] This is a diagram illustrating an experiment conducted using the present invention. [Figure 14] This is a diagram illustrating an experiment conducted using the present invention. [Figure 15] This is a diagram illustrating an experiment conducted using the present invention. [Figure 16] This is a diagram illustrating an experiment conducted using the present invention. [Figure 17] This is a diagram illustrating an experiment conducted using the present invention. [Figure 18] This is a diagram illustrating an experiment conducted using the present invention. [Figure 19] This is a diagram illustrating an experiment conducted using the present invention. [Figure 20] This is a diagram illustrating an experiment conducted using the present invention. [Figure 21] This is a diagram illustrating an experiment conducted using the present invention. [Figure 22] This is a diagram illustrating an experiment conducted using the present invention. [Figure 23] This is a diagram illustrating an experiment conducted using the present invention. [Figure 24] This is a diagram illustrating an experiment conducted using the present invention. [Figure 25] This is a diagram illustrating an experiment conducted using the present invention. [Figure 26] This is a diagram illustrating an experiment conducted using the present invention. [Figure 27] This is a diagram illustrating an experiment conducted using the present invention. [Figure 28]This is a diagram illustrating an experiment conducted using the present invention. [Figure 29] This is a diagram illustrating an experiment conducted using the present invention. [Figure 30] This is a diagram illustrating an experiment conducted using the present invention. [Figure 31] This is a diagram illustrating an experiment conducted using the present invention. [Figure 32] This is a diagram illustrating an experiment conducted using the present invention. [Figure 33] This is a diagram illustrating an experiment conducted using the present invention. [Figure 34] This is a diagram illustrating an experiment conducted using the present invention. [Figure 35] This is a diagram illustrating an experiment conducted using the present invention. [Figure 36] This is a diagram illustrating an experiment conducted using the present invention. [Figure 37] This is a diagram illustrating an experiment conducted using the present invention. [Figure 38] This is a diagram illustrating an experiment conducted using the present invention. [Figure 39] This is a diagram illustrating an experiment conducted using the present invention. [Figure 40] This is a diagram illustrating an experiment conducted using the present invention. [Figure 41] This is a diagram illustrating an experiment conducted using the present invention. [Figure 42] This is a diagram illustrating an experiment conducted using the present invention. [Figure 43] This is a diagram illustrating an experiment conducted using the present invention. [Modes for carrying out the invention]

[0024] The embodiments of the present invention will be described in detail below, but the description of the constituent elements described below is just one example (representative example) of an embodiment of the present invention, and the present invention is not limited to the following unless its gist is changed. In this specification, when the expression "~" is used, it is used to mean an expression that includes the numbers before and after it.

[0025] [System and method of the present invention] The present invention comprises a component waveform extraction system that extracts component waveforms from a signal waveform to be analyzed, and a signal waveform analysis system capable of quantitatively analyzing the characteristics of the extracted component waveforms. The component waveform extraction system performs redundant wavelet analysis on the measured signal waveform to determine the waveform characteristics present at each time point in the signal waveform. Then, based on the characteristics of a target component waveform that have been pre-set, the system extracts the component waveforms along with a degree of fit indicating that the waveform is a component waveform. The signal waveform analysis system calculates quantitative values ​​such as the magnitude of the extracted component waveforms, enabling a quantitative evaluation of the behavior of these component waveforms within the signal waveform.

[0026] [Analysis target] The present invention can be applied to various waveforms and component waveforms for analysis. For example, it can be applied to silent voice recognition (so-called lip-syncing), electromyography waveforms that capture heart sounds, electrical waveforms from electrocardiograms, muscle activity, etc., or, in engineering terms, noise and vibration waveforms generated from equipment and various environments, or waveforms intended for fault detection of equipment, etc. In this invention, electromyography waveforms are described as the waveform to be analyzed, and motor unit action potential waveforms are described as the component waveforms, as typical examples.

[0027] For example, when using the present invention for electromyography waveform measurement, while previous inventions only visualize and qualitatively understand the temporal fluctuations of fast-twitch / slow-twitch muscle activity from the frequency distribution of the motor units that make up fast-twitch / slow-twitch muscle fibers, the present invention enables the extraction and quantitative evaluation of target component waveforms. Therefore, it is possible to quantify the activity of fast-twitch / slow-twitch muscle fibers that are the target of those component waveforms, and evaluate the state of each muscle activity, the degree of fatigue, or the influence of each muscle activity when viewed as a whole muscle fiber.

[0028] First, let's explain the component waveform extraction system. The component waveform extraction system has an analysis unit that analyzes the signal waveform of the target to be analyzed, measured by various sensors, and a component waveform extraction unit that extracts component waveforms from the signal waveform of the target to be analyzed based on the various information obtained by the analysis unit. Each of these will be explained below.

[0029] [Analysis Department] In the analysis unit, redundant wavelet analysis is performed on the signal waveforms measured and acquired by various sensors. From the set of wavelet coefficients of the waveform present at each time point in the signal waveform calculated by this analysis (hereinafter referred to as the "wavelet coefficient set"), a set of wavelet coefficients corresponding to the component waveform to be extracted (hereinafter referred to as the "coefficient set of coefficients") is obtained.

[0030] In redundant wavelet analysis, the system responds sensitively to signal changes with shapes similar to the wavelet shape being used. Therefore, when analyzing a given waveform, wavelet coefficients corresponding to the various frequency components that make up this waveform are output for each time period in which these components exist. Furthermore, by inversely transforming these wavelet coefficients, the waveforms of these frequency components can be reproduced. Thus, since wavelet coefficients have a strong relationship with the frequency components of the waveform being analyzed, the set of coefficients obtained by the analysis unit can be said to represent the characteristics of the component waveforms present in the signal waveform.

[0031] To calculate the set of coefficients for the coefficient set, first, redundant wavelet analysis is performed on the measured and acquired signal waveform to calculate the wavelet coefficient set for each waveform present at each time point in the signal waveform. Next, a wavelet coefficient set corresponding to the component waveform is established, and based on this wavelet coefficient set, the coefficient set for the coefficient set corresponding to the component waveform is extracted from the wavelet coefficient sets of each waveform mentioned above.

[0032] Here, we will explain the wavelet coefficient set corresponding to the component waveform. As mentioned above, wavelet coefficients have a strong relationship with the frequency components of the waveform being analyzed, and this relationship is further related to the time period in which the frequency components exist. Therefore, there are wavelet coefficient sets that are distributed regularly on the time-frequency plane. The wavelet coefficient set defines the relationship between the time position and the frequency band in the time-frequency domain for those wavelet coefficients. Considering this, the wavelet coefficient set corresponding to the component waveform (hereinafter also referred to as the "wavelet coefficient set") defines the relationship between the time position and the frequency band in the time-frequency domain when the component waveform is represented by wavelet coefficients.

[0033] Furthermore, the component waveforms that we intend to capture with the wavelet coefficient set have a time width, just like the wavelet coefficient set itself, and the reference point for the wavelet coefficient set is also the time selected from the time width in which the component waveforms exist. As a result, if we extract the wavelet coefficients of each waveform present at each time point of the signal waveform calculated above, for example, when represented on the time-frequency plane, based on the wavelet coefficient set with that time as the reference point, we can extract the wavelet coefficient group corresponding to the component waveform present at that time. In other words, if we adjust and arrange all the wavelet coefficients corresponding to the wavelet coefficient set for each frequency band based on the reference point, as mentioned above, each frequency component included in the component waveform will be output as a group consisting of a frequency band and its wavelet coefficient, i.e., as a set of coefficients in the coefficient set, corresponding to the time in which the component waveforms exist, without each component being output at different times.

[0034] Any method can be used to define the wavelet coefficient set, but for example, it can be defined as follows. (1) Set based on the result of wavelet transforming the ideal shape of the component waveform. (2) Extraction from the wavelet transform results of the test signal data samples (3) Fine-tuning from the basic wavelet coefficient set so that the inverse wavelet transform result approaches the component waveform.

[0035] Regarding (1), waveforms based on various shapes, such as shapes that can be estimated from various information such as experimental results, shapes that are generally known or in use, or waveforms created based on similar waveforms, can be used as component waveforms, and wavelet coefficients can be determined by redundant wavelet analysis to set the wavelet coefficient set. Regarding (2), a test is intentionally performed in which the target component waveform is output, and the signal waveform acquired in this test is subjected to redundant wavelet analysis to set the wavelet coefficient set.

[0036] (3) is set by fine-tuning a certain wavelet coefficient set and bringing it closer to the component waveform using an inverse wavelet transform. Specifically, for example, as in the previous invention, in the time-frequency plane, a wavelet coefficient set is set in which the time axis end positions of all the regions of each constituent frequency band are aligned to either the left or the right for a given wavelet coefficient set.

[0037] Next, based on this wavelet coefficient set, the end of the lowest frequency band is used as a reference, and the end of the adjacent high-frequency band is adjusted. Similarly, the adjusted band and the adjacent high-frequency band are sequentially adjusted to set the wavelet coefficient set for the target component waveform. When setting this wavelet coefficient set, it is crucial to adjust it so that when the wavelet coefficient set formed by fine-tuning the adjustment magnitude is plotted on the time-frequency plane, the coefficients are aligned perpendicular to the time axis, and the result of the inverse transform of the coefficients is close to the component waveform. The creation of the signal waveform by inverse transform can be done by placing each coefficient in the coefficient set at its original time position and then summing the waveforms obtained by inverse transforming each coefficient.

[0038] Examples of wavelet coefficient set patterns include T+S and HS. T and H represent one pattern of wavelet coefficient set, where the positions of the time axis ends (larger side with respect to the time axis) and time axis leads (smaller side with respect to the time axis) of each frequency band region are aligned. S represents the amount of shift in each frequency band region. For example, in the case of T+S, based on the pattern of T, the time axis end (larger side with respect to the time axis) of the high-frequency band adjacent to the lowest frequency band of that pattern is shifted by S relative to the time width of that high-frequency band, and so on, with respect to the time axis end of the lowest frequency band of that pattern as the reference point. This shift is then carried out sequentially between adjacent bands in an even higher frequency band.

[0039] For example, in the case of T+S, the shift amount S can be set to T+1 / 2, T+1 / 4, or T+0; in the case of HS, it can be set to H-1 / 2, H-1 / 4, or H+0; or S is not limited to these values, but can be varied in increments of 1 / 32 based on 14 / 32, etc. The shift amount S can be adjusted appropriately according to the characteristics of the component waveform to be extracted.

[0040] On the other hand, in the case of component waveforms in the signal to be analyzed that include a single main peak frequency component, such as the wavelet with “Daubechies N=2”, a wavelet coefficient set with a standard end alignment (T) and an adjustment shift amount of 0 or greater is useful. By adjusting the shift amount, the rise and fall rates can be balanced, enabling stable analysis of the component waveform.

[0041] Figure 4 shows the analysis flow of the component waveform extraction system according to an embodiment of the present invention. The wavelet coefficient set and wavelet coefficient group of the component waveform may be stored in a memory unit (not shown) set in the analysis unit 10, and read from this memory unit when analysis is performed by the analysis unit 10. Alternatively, the wavelet coefficient group can be obtained by the analysis unit 10 and then used in real time by the component waveform extraction unit 20 described later, without being stored in the memory unit. The wavelet coefficient set can also be set in advance; in this case, for example, it may be stored in the memory unit mentioned above and read from the memory unit during analysis.

[0042] The above example shows how to obtain the coefficient set of the coefficient set using the analysis unit 10. However, the analysis unit 10 may also be equipped with a calculation unit (hereinafter referred to as the "coefficient group calculation unit") that calculates the coefficient set of the coefficient set, and the analysis may be performed using this unit. In this case, redundant wavelet analysis is performed on the measured and acquired signal waveform, and the wavelet coefficients of each waveform present at each time point in the signal waveform are calculated. Next, the coefficient group calculation unit extracts and obtains the coefficient set of the coefficient set corresponding to the component waveform from the wavelet coefficient groups of each waveform based on the wavelet coefficient set. This reduces the computational load on the analysis unit 10 and improves the computational processing efficiency. Furthermore, by providing the coefficient group calculation unit, it is possible to efficiently obtain visualization information of the behavior of the component waveform based on this analysis result. Moreover, by connecting to the signal waveform analysis system described later and using the results, it becomes possible to perform useful analyses, such as analyses that take into account the conditions of the surrounding waveforms other than the component waveform.

[0043] The sensors used to acquire signal waveforms are not particularly limited, as long as they can acquire the signal to be analyzed. For example, for biological signals, in addition to the surface electromyography sensors used to measure surface electromyography mentioned above, muscle sound sensors, electrocardiogram sensors, pulse wave sensors, blood pressure sensors, etc., can be used. For noise and vibration, examples include sound pressure sensors, intensity sensors, acceleration sensors, strain sensors, etc.

[0044] [Component waveform extraction section] Next, the component waveform extraction unit 20 will be described below. The component waveform extraction unit 20 calculates a goodness of fit that indicates the likelihood that the extracted waveform is a component waveform, and extracts the component waveform based on that goodness of fit. The goodness of fit is calculated based on the basic characteristics of the component waveform.

[0045] [Basic characteristics] The basic characteristics are the features that a component waveform can possess, such as the shape and magnitude of the component waveform and its temporal change within the signal waveform, based on the wavelet coefficient set. Specifically, they are set by the frequency characteristics of the wavelet coefficients, which show the relationship between the wavelet coefficient values ​​that constitute the component waveform and their frequency and time. For example, frequency characteristics corresponding to a typical shape of the component waveform, or frequency characteristics corresponding to various waveform shapes in the process of change, considering that the component waveform changes over time or due to some external factor, can be set. Furthermore, these basic characteristics can be set as appropriate according to the purpose of analysis and required accuracy. As mentioned above, wavelet coefficients have a strong relationship with the frequency components of the waveform being analyzed, so it is possible to grasp the shape, magnitude, and temporal change of each frequency component constituting the component waveform based on the distribution of wavelet coefficients that form the wavelet coefficient set of the component waveform. By comparing these basic characteristics, the validity of considering the set of wavelet coefficients obtained by the analysis unit 10 in relation to each time point as component waveforms can be determined. Furthermore, the temporal characteristics of the component waveforms within the signal waveform can be grasped from the magnitude of the wavelet coefficient values ​​and the temporal changes in the frequency distribution of the wavelet coefficients when the set of coefficients is considered to be component waveforms.

[0046] As a basic characteristic used to calculate the degree of fit, as will be described later, the characteristics of the component waveform can be vectorized. This vector can be constructed, for example, based on the wavelet coefficient set of the component waveform, with each frequency band (hereinafter referred to as "frequency level") as a dimension, and the wavelet coefficient corresponding to that frequency level as the component element of the vector. Furthermore, since the waveform shape of the component waveform changes over time within the signal waveform, it is similarly vectorized based on the frequency characteristics of the wavelet coefficients corresponding to the waveform that shows a typical state in the process of the temporal change of the component waveform, and these vectors are each set as unit vectors to be set as the basis vectors. Similarly, the wavelet coefficient set analyzed for the waveform at a certain time can also be vectorized, and the degree to which it fits as a component waveform by comparison with the basis vector, and at what point in the process of change of the component waveform that the waveform occurred, can be analyzed and evaluated.

[0047] Specific examples of the above-mentioned fundamental vectors can be shown based on the wavelet coefficients of the waveforms for fast-twitch muscle (non-fatigued), fast-twitch muscle (fatigued), slow-twitch muscle (non-fatigued), and slow-twitch muscle (fatigued), which will be discussed later. Fast-twitch muscle (non-fatigued) and fast-twitch muscle (fatigued), or slow-twitch muscle (non-fatigued) and slow-twitch muscle (fatigued), are representative states in the process of change when fast-twitch or slow-twitch muscle is used as a component waveform. When extracting the waveforms of motor units such as fast-twitch and slow-twitch muscle, including their changing waveforms, four fundamental vectors can be set, each with a frequency level as its dimension, based on the frequency characteristics of these wavelet coefficients. Thus, it is possible to set not just one but multiple fundamental vectors, depending on the properties of the component waveforms.

[0048] [Method for extracting goodness-of-fit and component waveforms] Next, we will explain the method for extracting goodness-of-fit and component waveforms.

[0049] Since the waveform shape of a component waveform changes over time within the signal waveform, the direction of acceptable change in the shape of the component waveform (for example, the change before and after muscle fatigue as described above) must be taken into consideration during extraction. When the features of a component waveform are vectorized (for simplicity, we will discuss this in two dimensions), if these features can change between A and B, which are defined as the change range of a certain component waveform, then it is considered that the features of the sample data will be evaluated as having equivalent "component waveform characteristics" when they lie on the path from A to B.

[0050] Figure 1 is a diagram illustrating the evaluation of the characteristics of component waveforms. For example, the characteristicness of component waveforms in Figure 1 should be evaluated as V=W>X>Y>Z. However, if paths A and B are part of the change process, and the extension of these paths can also be considered a reasonable change in the component waveform, then it could be argued that the evaluation should be V=W=Z>X>Y.

[0051] Furthermore, if we consider the above two-dimensional content in a multi-dimensional space and represent the characteristics of the component waveforms as vectors, we can consider that the characteristics of the waveform shape of such component waveforms appear in the direction of the vector, and the strength of the signal appears in the length of the vector.

[0052] Therefore, to first capture only the characteristics of the waveform shape without considering the strength of the signal, we will consider the basis vectors projected onto a hypersphere. It is simple to use the vectorized characteristics of the component waveforms as the basis vectors, convert them into unit vectors, and project them onto a unit hypersphere with radius 1. The values ​​of each dimension are assumed to be greater than or equal to 0. Figure 2 shows a plane cut out from the hypersphere, passing through the origin 0 and points A and B on the hypersphere.

[0053] Furthermore, if we consider A and B as basis vectors, and C as a vector that exhibits a certain characteristic of the waveform whose "component waveform-likeness" is evaluated by analyzing the signal waveform, and examine the relationship between points A and B on arc AB and basis vectors A and B, the angle between the vectors of points A and B and the basis vectors is ∠AAB = ∠ABB, which is the maximum value of ∠ACB with respect to any point C on the hypersphere (where the values ​​of each dimension are greater than or equal to 0). Also, if we define point C as being between A and B when ∠ABC ≤ 90° and ∠BAC ≤ 90°, then the angle ∠ACB between the vector of point C and basis vectors A and B becomes smaller as point C moves away from arc AB.

[0054] Thus, the characteristics of the waveform shape of the component waveform are expressed in the direction of the vector, and the relationship between the waveform being evaluated and these characteristics can be examined by the distance from the basis vector. Furthermore, this distance-based examination can be performed based on the angle between the vector representing the characteristics of the waveform being evaluated and the basis vector. Note that, as shown in Figure 3, if point γ is outside the b side of ab, the evaluation is performed using ∠abγ instead of ∠aγb.

[0055] In this invention, the characteristics of the waveform present at a given time in the signal waveform to be analyzed are represented by the wavelet coefficient values ​​of each frequency level in the wavelet coefficient group corresponding to that time. The component waveform extraction unit 20 sets each frequency level as a coordinate axis in a multidimensional space, and sets a basis vector based on the wavelet coefficient set of the component waveform and a vector based on the coefficient group of the coefficient set corresponding to the component waveform in the signal waveform, which has been analyzed by the analysis unit 10. Based on the above concept in relation to the basis vector, the "component waveform-likeness" can be evaluated.

[0056] In other words, the "component waveform-likeness" based on the above concept can be calculated as an evaluation value called "goodness of fit," which indicates whether the set of coefficients in a set of coefficients at a certain time point constitutes a component waveform. In this concept, we showed that the distance relationship with the tolerance range (a and b in Figure 3) can be evaluated based on the angle formed by the vectors, but it is also possible to evaluate it using the distance from each endpoint of the tolerance range (Euclidean distance, opening angle, etc., can be appropriately selected) and the ratio of the distances.

[0057] Once the goodness of fit is calculated, the time at which the goodness of fit is greatest is estimated to be the time at which the component waveform exists. By performing an inverse wavelet transform on the wavelet coefficient group corresponding to the vector W at that time, the component waveform is extracted.

[0058] Since component waveforms exist with a temporal width, even at times slightly deviated from the time defined as the existence time of the component waveform, it is expected that the component waveform will have a high degree of fit. Taking this into consideration, in order to capture the component waveform, the optimal existence time and shape of the component waveform (obtained by the inverse wavelet transform) can be obtained not only based on the degree of fit value, but also on the change in the degree of fit, or both the degree of fit value and the change in the degree of fit.

[0059] While the above evaluation of component waveform similarity is based solely on geometric features, it is also possible to evaluate other features, such as signal intensity (magnitude), in addition to their geometric characteristics. In this case, as described later, the evaluation can be performed by multiplying these other feature parameters with the degree of fit, enabling the extraction of component waveforms with higher accuracy in line with the analysis objectives.

[0060] Furthermore, if the shape of the component waveforms is fixed and the dispersion of the observed waveforms is probabilistic, the "component waveform-likeness" of the component waveforms in the sampled signal waveform can also be evaluated using simple distances such as the Euclidean distance or Mahalanobis distance from the basis vectors mentioned above.

[0061] [Fitness calculation unit] Regarding the calculation of the degree of fit, it is also possible to provide a degree of fit calculation unit (not shown) separately from the component waveform extraction unit 20. Specifically, in the degree of fit calculation unit, a basis vector is set based on the wavelet coefficient set of the component waveform and the coefficient set analyzed by the analysis unit 10, in the same manner as above, and the degree of fit is calculated based on this basis vector. By setting up a degree of fit calculation unit, the computational load on the component waveform extraction unit can be reduced, and for example, if the aforementioned "component waveform-likeness" is based on something other than geometric features, the degree of fit can be treated separately as being based on geometric features, and the evaluation of component waveforms can be performed efficiently.

[0062] [filter] To more efficiently calculate / evaluate the degree of fit and extract component waveforms, a fitting filter and an adjustment filter can be set and used. The fitting filter can be used with the coefficient set analyzed by the analysis unit 10, while the adjustment filter can be used with the coefficient set after the degree of fit of a given waveform has been calculated, in conjunction with that degree of fit. Both filters are described below.

[0063] [Suitable filter (extraction filter)] The frequency components of a component waveform do not necessarily exist in all frequency bands obtained through the analysis of the signal waveform. Furthermore, even in frequency bands where component waveforms exist, there may be bands where other signal waveforms are mixed in or bands with a low signal-to-noise ratio. In such cases, the influence of these factors must be considered when evaluating the fit to the component waveform and when extracting it. When determining the fundamental vector of a component waveform, only the frequency bands contained in that component waveform are required.

[0064] Therefore, in order to ensure that each frequency band used for evaluating the degree of fit and extracting component waveforms is appropriate, a fitting filter can be set that defines a weight for each frequency band. The fitting filter assigns a weight of 0 to 1 to each frequency band; a weight of 0 is assigned to bands that do not contain the frequency components of the component waveform, and a value in the range of 0 to 1 (for example, an intermediate value) is assigned to bands where the inclusion of unintended signal components is expected and whose influence should be suppressed, according to the influence. The fitting filter is used by multiplying it by the set of coefficients of the coefficient set that is the target of the degree of fit evaluation before calculating the degree of fit.

[0065] As an example of applying a fitting filter, if the component waveforms are targeted to fast-twitch and slow-twitch muscle fibers as mentioned above, in an electromyographic waveform signal where fast-twitch muscle component waveforms, which remain in the relatively high-frequency band, and slow-twitch muscle component waveforms, which exist in the lower-frequency band, are mixed, a fitting filter can be set to suppress the influence of the low-frequency band when calculating the degree of fit to extract the fast-twitch muscle component waveform. Similarly, a fitting filter can also be set to extract the slow-twitch muscle component.

[0066] The set of coefficients used for extracting and analyzing component waveforms can be the values ​​obtained after applying this fitted filter. In this case, for example, when analyzing and extracting multiple types of component waveforms, such as the fast-twitch and slow-twitch muscle components of an electromyographic waveform signal, different fitted filters are set for each. Therefore, even if the original set of coefficients is the same, the sets of coefficients used in each analysis will be different.

[0067] In particular, considering that fast-twitch muscle components tend to remain in the relatively high-frequency range, a matching filter can be set to reduce wavelet coefficient values ​​in frequency bands outside the range in which the component exists when comparing the basis vector with the coefficient set. In this case, for example, each element of the wavelet coefficient set of the component waveform may be weighted with a value of 1 or 0, or a weight that slopes using real numbers may be used. It is also possible to perform the same processing separately for fast-twitch and slow-twitch muscle fibers.

[0068] This approach is not limited to electromyography waveforms; it is also possible to treat cases where multiple types of component waveforms with different frequency bands are mixed (including cases where a single type of component waveform is being analyzed by separating it by frequency band) in the same way as the relationship between fast-twitch and slow-twitch muscle fibers described above. The simultaneity of their occurrence can also be analyzed based on the goodness-of-fit evaluation values ​​described later, for example, by using the product of each goodness-of-fit.

[0069] [Adjustment filter] When the signal to be analyzed consists of multiple component waveforms, separating and extracting them requires distributing the wavelet coefficient values ​​of each frequency band as elements of each component waveform. For this purpose, adjustment filters can be set for each type of component waveform. These adjustment filters are set considering the weights of each frequency band in the wavelet coefficient set of the component waveforms, and include both distributive and derivational adjustment filters.

[0070] A distribution-type adjustment filter is used when multiple component waveforms overlap in a certain frequency band, resulting in the wavelet coefficient values ​​of that frequency band being a composite of the wavelet coefficient values ​​of those component waveforms. In such cases, the wavelet coefficient values ​​are distributed to each component waveform. A derivation-type adjustment filter, on the other hand, is used to interpolate the set of coefficients from which the goodness of fit has been calculated, so that the frequency distribution of the component waveforms is reflected in the ranges that were not used in the goodness of fit evaluation, such as the ranges that were removed by the fitting filter.

[0071] Furthermore, the distributive tuning filter assigns weight values ​​to each frequency band of the wavelet coefficient set and is used in combination with the goodness of fit. Specifically, for each frequency band, the product of the goodness of fit and the value of the distributive tuning filter is calculated for each component waveform, and the wavelet coefficient values ​​for that frequency band are distributed to each component waveform according to their ratio. When analyzing the component waveforms, these distributed wavelet coefficient values ​​are then used.

[0072] Let W be a coefficient group of a coefficient set for which a fitness has been calculated, and w be the level-i wavelet coefficient of W -i Furthermore, let DF be the value of the distributed adjustment filter for each component waveform at level-i -i set as above. Next, calculate the product of the fitness calculated for the plurality of component waveforms to be distributed and the distributed adjustment filter of each component waveform, and distribute the wavelet coefficients to each component waveform included in W according to the ratio of the products.

[0073] For example, taking myoelectric waveform analysis as an example, let fDF be the value of the distribution filter for fast-twitch muscle -i and sDF be the value of the distribution filter for slow-twitch muscle -i Furthermore, when fMR is the fitness as a fast-twitch motor unit action potential waveform and sMR is the fitness as a slow-twitch motor unit action potential waveform, fw, which is the level-i wavelet coefficient value of fW, the coefficient group of the coefficient set for fast-twitch muscle, and sw, which is the level-i wavelet coefficient value of sW, the coefficient group of the coefficient set for slow-twitch muscle -i and sw -i can be given by the following equations. fw -i =w -i *(fMR*fDF -i / (fMR*fDF -i +sMR*sDF -i )) sw -i =w -i *(sMR*sDF -i / (fMR*fDF -i +sMR*sDF -i ))

[0074] Hereinafter, when simply described as coefficient group W of a coefficient set, unless otherwise explicitly specified, W refers to fW when targeting a fast-twitch motor unit action potential waveform, and refers to sW when targeting a slow-twitch motor unit action potential waveform.

[0075] This section describes a derivation-type adjustment filter. In the signal waveform to be analyzed, it is assumed that there may be signals or states that affect the accuracy of the analysis, such as the inclusion of signals different from the component waveform in a part of the frequency band of the component waveform, or the complex superposition of waveforms. In this invention, such frequency bands that may affect the accuracy of the analysis can be removed using a fitting filter for analysis. However, it is possible that elements of the component waveform to be analyzed were originally included in these frequency bands, and in order to further improve the accuracy of the analysis, it may be better to reflect the information of the removed frequency bands. Therefore, the derivation-type adjustment filter uses the results of the coefficient set from which the degree of fit has been calculated by applying the fitting filter (including the coefficient set obtained by applying the distribution-type adjustment filter) to estimate and interpolate the wavelet coefficient values ​​that have been removed by the fitting filter.

[0076] The derived adjustment filter is based on a wavelet coefficient set and reflects the frequency distribution of wavelet coefficient values ​​that component waveforms may possess, including the process of change, according to the purpose of analysis and the accuracy to be ensured. Specifically, it is weighted for each frequency band based on the frequency characteristics of the wavelet coefficient values ​​and set as the wavelet coefficient values ​​that the component waveforms are estimated to possess. Regarding the interpolation method using the derived adjustment filter, in order to evaluate the intensity of the component waveform that serves as the basis for derivation, the wavelet coefficient values ​​of the coefficient set to which the fitted filter is applied (hereinafter referred to as "fitted filter applied coefficient values") are calculated. Next, the fitted filter is similarly applied to the derived adjustment filter, and the result of this application is considered as the coefficient set of coefficients to obtain wavelet coefficient values ​​for evaluating the magnitude of the values ​​(hereinafter referred to as "derived adjustment filter coefficient values"). Then, the ratio of the fitted filter applied coefficient values ​​and the derived adjustment filter coefficient values ​​and the degree of fit are multiplied by the value of the wavelet coefficient values ​​of the derived adjustment filter to interpolate. In other words, the derived adjustment filter coefficient values ​​are the wavelet coefficient values ​​in the range where the fitted filter has high values. By calculating the ratio with these values, the proportion of possible values ​​in the signal waveform to the wavelet coefficient values ​​of the derived adjustment filter is determined. Furthermore, by multiplying this by the degree of fit, which indicates the degree to which the waveform is actually likely to exist, the wavelet coefficient values ​​that the component waveform is estimated to inherently possess are derived.

[0077] Specifically, the wavelet coefficient values ​​in the set of coefficients calculated by the analysis unit 10 are w1, ..., w n The suitable filters are MF1, ..., MF n (However, 0.0 ≤ MF) i Let ≤ 1.0. Apply this fitting filter and the resulting goodness of fit be denoted as MR, and the derived extraction filters are EF1, ..., EF n Let's assume that EF i You may set a negative value for this if necessary.

[0078] First, the square root of the sum of the squares of the wavelet coefficients after applying the fitted filter is S. wIt is calculated using the following formula. S w =sqrt(Σ((w i * nearby i ) 2 )) w i * nearby i This is the fitting filter application coefficient value, S w This corresponds to the magnitude of the coefficient value used in the goodness-of-fit evaluation. Furthermore, the square root of the sum of the squares of the wavelet coefficient values ​​of the derived adjustment filter after applying the fitted filter is S. E It is calculated using the following formula. S E =sqrt(Σ((EF i * nearby i ) 2 )) EF i * nearby i This is the derived adjustment filter coefficient value, S E This corresponds to the magnitude of the derived adjustment filter within the range used for goodness-of-fit evaluation. Since the derived extraction filter is set to reflect the wavelet coefficient distribution of the component waveform, it corresponds to the magnitude of the coefficient values ​​in the portion that would be used for goodness-of-fit evaluation, assuming that a component waveform with the magnitude of the derived extraction filter exists.

[0079] At this time, the wavelet coefficient values ​​of the component waveforms to be extracted are ew1, ..., ew n Therefore, ew1,...,ew n MR and the above S w S E It is calculated using the following formula. ew i =(w i * nearby i )+sqrt(MR*S w / S E )*EF i *(1-MF i ) According to this formula, for example, in parts where the fitting filter is 1, that is, where the original value was used in the goodness-of-fit evaluation, the coefficient values ​​of the coefficient set are used as they are, and in parts where the fitting filter is 0, that is, where the portion was deleted and not used in the goodness-of-fit evaluation, the values ​​are estimated based on the values ​​of the derived adjustment filter that reflect the frequency characteristics of the component waveform.

[0080] The value of the fitting filter is like the confidence level of the coefficient values ​​in that part. To suppress its impact on the analysis, the fitting filter can be used to exclude wavelet coefficients in the coefficient set from the goodness-of-fit evaluation or to reduce their value. Equation ew above i w i It is designed by weighting the direct value of the component waveform with estimated values ​​from the frequency characteristics of the component waveform. Furthermore, MR can be interpreted as reflecting how much of the coefficient values ​​used in the goodness-of-fit evaluation can be trusted as component waveform values, and is therefore used as part of the weighting of how much of the supplementary coefficient values ​​to trust and add.

[0081] Next, based on the analysis flow shown in Figure 4, the processing flow for extracting component waveforms in the component waveform extraction system will be explained. First, the analysis unit 10 performs redundant wavelet analysis on the signal waveform to be analyzed, measured and acquired by various sensors, and calculates a wavelet coefficient set. Based on this wavelet coefficient set, a coefficient set is extracted from this wavelet coefficient set. Next, the component waveform extraction unit 20 calculates the degree of fit for the waveform included in the coefficient set extracted by the analysis unit 10, based on the basic characteristics of the component waveform. At that time, based on a predetermined criterion, a wavelet coefficient set corresponding to the component waveform is extracted from the degree of fit of the component waveform and the coefficient set with that degree of fit. The waveform of the component waveform is calculated and estimated by performing an inverse wavelet transform on the wavelet coefficients in that wavelet coefficient set.

[0082] Regarding the fitting filter shown in Figure 4, it is possible to apply the fitting filter as appropriate before the degree of fit of the component waveform and the wavelet coefficient group corresponding to the component waveform are extracted. This filter filters out frequency bands unnecessary for analysis from the coefficient set, allowing for the efficient extraction of the wavelet coefficient group corresponding to the component waveform.

[0083] Regarding the adjustment filter, it is possible to apply a filter to the wavelet coefficient group corresponding to the extracted component waveform, based on the fundamental characteristics of the component waveform, in order to interpolate and make the extracted waveform closer to the component waveform.

[0084] In the following, the calculation of the degree of fit using fundamental vectors and the extraction of component waveforms according to the analysis flow described above will be specifically explained using an electromyographic waveform as an example. The set of wavelet coefficients of the component waveform, which is the basis of the fundamental vectors, and the values ​​of each level (frequency band) of the wavelet coefficient group are all assumed to be positive. However, if it is appropriate for a particular level to be negative due to the characteristics of the component waveform, it is possible to treat it in the same way as if it were assumed to be positive by reversing the positive / negative treatment of that level (for example, by inverting the positive / negative of the wavelet coefficient value, or by only using negative values ​​during half-wave rectification).

[0085] Furthermore, although we do not assume here that the value of a specific level changes between positive and negative as a direction of change in the component waveform, it is also possible to apply this method by performing reversible processing by adding (and multiplying by a constant if necessary) an offset value (reversing it back in the case of inverse wavelet transform before use).

[0086] The set of wavelet coefficients obtained by redundant wavelet analysis on a signal waveform acquired at a certain time is half-wave rectified to obtain only positive values, and this set of coefficients is denoted as W. However, W is not assumed to be a zero vector. The case where W is a zero vector will be discussed later. Furthermore, S is denoted as W converted into a unit vector.

[0087] Furthermore, if we denote the basis vectors at low fatigue as L and the basis vectors at high fatigue as H, then the endpoints of each vector will be located on a hypersphere with radius 1. From now on, unless ambiguous, the endpoints of each basis vector will be indicated by the same symbol.

[0088] The angle between unit vectors with endpoints S, L, and H is θ. SL θ SH θ LH Let's assume that the length of the line connecting the endpoints of the vectors is |SL|=sqrt(2*(1-cos(θ SL ))) |SH|=sqrt(2*(1-cos(θ SH ))) |LH|=sqrt(2*(1-cos(θ LH ))) Each cosine value is obtained by the dot product of unit vectors.

[0089] ∠LSH=φ S ∠SLH=φ L ∠SHL=φ H In that case, the following applies: cos(φ S )=(1+cos(θ LH )-cos(θ SL )-cos(θ SH )) / (2*sqrt((1-cos(θ SL ))*(1-cos(θ SH ))) cos(φ L )=(1+cos(θ SH )-cos(θ SL )-cos(θ LH )) / (2*sqrt((1-cos(θ SL ))*(1-cos(θ LH ))) cos(φ H )=(1+cos(θ SL )-cos(θ SH )-cos(θ LH )) / (2*sqrt((1-cos(θ SL ))*(1-cos(θ LH ))) Note that the case where each denominator becomes 0 is excluded by setting the conditions described later, so this does not cause any problem.

[0090] If there is a high probability that the coefficient group of the coefficient set corresponds to the motor unit activity of fast-twitch muscles (or slow-twitch muscles), it can be inferred that S exists on or close to the isolated LH. φ S reaches its maximum when S is on the isolated LH, and decreases as S moves away from it, and accordingly cos(φ S ) increases. These values can be used to evaluate the high suitability as motor unit activity. Note that in the above description, when cos(φ S ) is minimal, cos(φ S )=-sqrt((1+cos(θ LH )) / 2).

[0091] Furthermore, since it can be considered that the suitability is high even when S is located on the extension line instead of between the isolated LH, the matching rate of S is defined as MR S , and MR S (where 0≦MR S ≦1) can be given as follows, and the matching degree can be obtained.

[0092] (1) When 1-cos(θ SL )<ε (where ε is an appropriately small value) S and L are considered identical, let |SL|=1, |SH|=0, MR S =1.0.

[0093] (2) When 1-cos(θ SH )<ε (where ε is an appropriately small value) S and H are considered identical, let |SL|=0, |SH|=1, MR S =1.0.

[0094] (3) When cos(φ L )≧0 and cos(φ H )≧0 S is in a situation existing between L and H. mincos=-sqrt((1+cos(θLH )) / 2) MR S =(1-cos(φ S )) / (1-mincos)

[0095] (4) When cos(φL)<0 S is in a situation existing outside the L side of arc LH. mincos=-sqrt((1+cos(θ SH )) / 2) MR S =(1-cos(φ L )) / (1-mincos)

[0096] (5)cos(φ H ) When <0 S is in a situation existing outside the H side of arc LH. mincos=-sqrt((1+cos(θ SL ))) / 2) MR S =(1-cos(φ H )) / (1-mincos)

[0097] Furthermore, when L and H can be regarded as the endpoints of allowable changes, it is more appropriate to evaluate the goodness of fit in the above (4) and (5) lower than in the case of the above (3). In that case, the same value of mincos as in the above (3) is used as mincos in the above (4) and (5). In addition, the goodness of fit MR has been defined on the premise that W is not a zero vector, but when W is a zero vector, MR S =0.

[0098] Next, after MR S is calculated, the time at which the value becomes large is the time estimated as the existence time of motor unit activity, and the wavelet coefficient group corresponding to the vector W at that time is subjected to inverse wavelet transform to extract the motor unit action potential waveform.

[0099] The embodiments described above are examples of preferred forms for carrying out the present invention, but are not limited thereto. Various modifications and additions can be made without departing from the spirit of the present invention. Examples of such modifications and additions are shown below.

[0100] In the above-mentioned goodness-of-fit evaluation, it is also possible to evaluate the behavior of the component waveforms in a manner consistent with the purpose of the analysis, such as the situation within their range of change or the direction in which they are changing. For this, a fundamental vector corresponding to the specific one of interest is set based on the change characteristics of the component waveforms, and the goodness-of-fit is evaluated based on the distance from the vector of the waveform being evaluated to that fundamental vector. For example, using the electromyography waveform analysis described above as an example, to evaluate whether S is leaning towards low fatigue or high fatigue using |SL| and |SH|, the goodness-of-fit MR is used. S Low fatigue side MRL S and high fatigue side MRH S The tasks are then divided as follows. It can also be used in some form of fatigue-related assessment. MRL S =MR S *|SL| 2 / (|SL| 2 +|SH| 2 ) MRH S =MR S *|SH| 2 / (|SL| 2 +|SH| 2 )

[0101] Regarding the evaluation of the degree of fit, it is possible to set a threshold by, for example, examining the change in the degree of fit or fit intensity in a test signal waveform beforehand. This allows for efficient analysis according to the purpose of the analysis.

[0102] Multiple fundamental vectors V1, ..., V whose component waveform change characteristics are continuous max (However, for i=1, ..., max-1, basis vector pairs V i and V i+1If defined as (all angles formed by are equal), then the basic vector pair (V) used to calculate the goodness of fit of vector X based on the waveform being evaluated. i , V i+1 Select the following in order of priority:

[0103] (i) If there is a pair of basic vectors that satisfies (1) or (2) above, select that pair. Excluding the endpoints of a continuous set of basis vectors, there are two pairs that satisfy the condition, but the practical result is the same regardless of which pair is chosen.

[0104] (ii) If no basic vector pair satisfies condition (i), select the basic vector that has the highest goodness of fit among the pairs that satisfy (3) above.

[0105] (iii) If no basis vector pair satisfies conditions (i) and (ii) (as in (4) or (5) above), select the basis vector pair that minimizes mincos.

[0106] [Extraction of component waveforms based on matching intensity] Regarding evaluation using goodness of fit, as mentioned above, we considered capturing the waveform shape features without considering signal strength. However, there may be cases where it is necessary to also consider intensity when extracting component waveforms. Therefore, when extracting such component waveforms, we can extract a goodness of fit that indicates the characteristic of a component waveform and a group of wavelet coefficients with that goodness of fit from the coefficient set, based on the basic characteristics of the component waveform. We can then multiply the wavelet coefficient group by the feature information indicating the intensity of the waveform corresponding to that wavelet coefficient group and use the corrected goodness of fit. Here, intensity does not simply mean that the signal intensity of the component waveform is large, but includes all features that have the property that the component waveform is strong, or that the value becomes large when a signal waveform simply exists. Specifically, there are signal intensity features such as the root mean square of the coefficients in the coefficient set, and frequency features that show the property that a strong component waveform has a bias in the distribution of the magnitude of each frequency component of that component waveform. Details of signal intensity features and frequency features will be explained later.

[0107] Furthermore, although the above describes calculating the fit strength from the set of coefficients, in this system it is important to ultimately identify the component waveform. Therefore, when calculating the fit strength, other methods such as the wavelet coefficients after applying the fitting filter or the wavelet coefficients after applying the adjustment filter may also be used as needed. Depending on the characteristics of the target, the goodness of fit may be used directly without multiplying by these feature information. Since there is no difference in the subsequent procedures regardless of which method is chosen, for simplicity, the expression "fit strength" will be used from now on unless there is a need to specifically distinguish between the goodness of fit and the product of each of the above feature quantities. The closer the wavelet coefficients are to the component waveform, the higher the fit strength.

[0108] Redundant wavelet analysis is performed on the signal to be analyzed, and the wavelet coefficient set for each time point is obtained from the results. Then, based on the obtained wavelet coefficient set, the fit intensity for each time point is determined, and based on that fit intensity, the existence time of the component waveform and the wavelet coefficient set are extracted and inversely transformed to obtain the component waveform.

[0109] Thus, the present invention provides a method for evaluating the goodness of fit of wavelet coefficient groups for detecting and extracting component waveforms with a normal direction of change, using a simple process that avoids increased computational costs. Furthermore, since the present invention significantly improves computational efficiency, real-time processing is possible. Once the goodness of fit is obtained, the time of existence of the component waveform can be estimated based on that goodness of fit, and the component waveform can be obtained by inverse transforming the wavelet coefficient group extracted at that time.

[0110] Next, the signal waveform analysis system according to the present invention will be described. Figure 5 is a diagram showing the analysis flow of the signal waveform analysis system according to an embodiment of the present invention. The signal waveform analysis system includes a feature information analysis unit 30 that analyzes the characteristic information of the component waveforms in the signal waveform to be analyzed, based on the information obtained by the analysis unit 10 of the component waveform extraction system. The feature information analysis unit 30 will be described below. In addition, it is possible to provide the analysis unit 10 of the component waveform extraction system in the signal waveform analysis system so that it can operate independently of the component waveform extraction system.

[0111] In the feature information analysis unit 30, the goodness of fit is calculated for the set of coefficients obtained by the analysis unit 10 of the component waveform extraction system. Based on this goodness of fit, a group of wavelet coefficients and a goodness of fit corresponding to the component waveform are extracted, and the feature information of the component waveform in the signal waveform under analysis is analyzed based on this wavelet coefficient group. Furthermore, regarding the analysis of this feature information, if the group of wavelet coefficients corresponding to the component waveform is extracted based on the fit strength, a group of wavelet coefficients that have characteristics of the component waveform and their goodness of fit are first extracted from the set of coefficients based on the goodness of fit, and the analysis can be performed based on the value obtained by multiplying the feature information of that wavelet coefficient group (the feature information will be described later) by the goodness of fit. This feature information indicates the intensity and / or properties of the component waveform. For example, it includes a signal intensity feature to indicate the intensity of the component waveform and a frequency feature to indicate the properties of the component waveform. In addition, although not shown in the figure, each feature is analyzed by the intensity feature analysis unit and the frequency feature analysis unit.

[0112] Furthermore, the feature information analysis unit 30 can also be installed in a component waveform extraction system. In this case, the feature information analysis unit 30 can determine the fit strength based on the signal intensity feature quantity or frequency feature quantity and the degree of fit of those features, and extract component waveforms that take the feature quantities into account.

[0113] The intensity feature analysis unit provided in the feature information analysis unit 30 can calculate signal intensity features. Signal intensity features indicate the intensity (magnitude) of the component waveform to be extracted, and specifically are expressed as the intensity (WSP; Wavelet-Set Power) of the wavelet coefficient group corresponding to the component waveform, estimated based on the goodness of fit, and the wavelet coefficients of each frequency level k that constitute this wavelet coefficient group W are w k (k is -1 to -L) MAX When ), WSP(W) is given by the following equation. WSP(W) = sqrt(Σ k (w k ) 2 )

[0114] The signal strength feature can also be used to analyze component waveforms by multiplying it by the goodness of fit. This is particularly useful when analyzing component waveforms while considering not only the waveform shape features such as frequency characteristics but also the signal strength. As mentioned earlier, the goodness of fit is a value calculated based on the waveform shape features of the component waveform, so the goodness of fit can be set as MR, and this goodness of fit can be multiplied by WSP, which represents the signal strength, to obtain the signal strength feature, i.e., MR*WSP.

[0115] Furthermore, as mentioned earlier, MR*WSP, which is the product of the goodness of fit and the signal intensity feature, allows for the extraction of component waveforms while also considering the signal intensity as the goodness of fit.

[0116] In electromyography (EMG) waveform signal analysis, MR*WSP, obtained by multiplying WSP by the goodness-of-fit MR, can be used as a signal intensity feature from the perspective of motor unit activity. In this case, MR*WSP is a feature that reflects the sum of action potentials of motor units that cannot be separated in terms of time resolution, and tends to have a larger value as more motor units are activated synchronously to achieve strong muscle activity.

[0117] The frequency feature analysis unit provided in the feature information analysis unit 30 can calculate frequency features. As frequency features, the wavelet centroid (CoB) is determined for the wavelet coefficient group W corresponding to the component waveform, which is estimated based on the goodness of fit. The wavelet centroid is the value obtained by dividing the sum of the moment amounts of each wavelet coefficient with respect to a certain reference frequency by the sum of each wavelet coefficient in a coefficient group assembled from wavelet coefficients, and it indicates the frequency position of the center of this coefficient group. Specifically, the level range of the wavelet coefficient group W corresponding to the component waveform is -1, ..., -Lmax, and the wavelet coefficient of each level is w -1 ...w -Lmax In this case, CoB(W) is given by the following: (1)Σ k |w -k When |=0, CoB(W)=0.0 (2)Σ k |w-k When |!= 0, CoB(W) = (Σ k (|w -k |*2 (Lmax-k) ) / Σ k |w -k |)-1

[0118] In addition, since CoB indicates instantaneous frequency characteristics with a shorter time response than the analysis time width (window width) handled by windowed Fourier transform, the temporal change in the value may be significant. Therefore, for example, to address issues such as that fluctuations in the high frequency band are more noticeable than fluctuations in the medium to low frequency bands and handling over a wide dynamic range is required, evaluation can also be performed using a characteristic value that expresses the frequency band to be handled on a logarithmic axis. For example, as will be described later, logarithmic wavelet centroid (LCoB) can be used.

[0119] In this case, for example, normalizing such that the range of possible values (defined by the level range of the wavelet coefficient set) is 0 to 1 is suitable, for example, when comparing a plurality of types of signals having different frequency fluctuation range characteristics on the same scale (such as comparing what proportional position the signal is within a specified frequency range).

[0120] Therefore, when wavelet coefficients from level -1 to -Lmax are obtained, the wavelet coefficients in the partial interval are Ws = w -a , ···, w -b (that is, defined from level -a to -b; 1≦a<Lmax, a≦b≦Lmax), the logarithmic wavelet centroid (LCoB(Ws)) whose value ranges from 0 to 1 is given below (provided that in Σ k the range of k is a≦k≦b). (1) When Σ k |w -k |= 0, LCoB(Ws) = 0.0 (2) When Σ k |w -k |!= 0, LCoB(Ws) = log2(Σ k (|w -k |*2 (b-k) ) / Σ k |w -k|) / (ba) The relationship between CoB(Ws) and LCoB(Ws) is LCoB(Ws) = log2(CoB(Ws)+1) / (ba).

[0121] Since CoB and LCoB essentially represent the same thing, we will refer to both as CoB from now on unless there is a specific need to distinguish between them. Users can substitute CoB and LCoB as needed.

[0122] Figure 5 shows the basic analysis flow based on the fitting intensity in the signal waveform analysis system. First, the analysis unit 10 performs a redundant wavelet transform on the signal to be analyzed and calculates a set of wavelet coefficients. Then, it applies a set of wavelet coefficients to this wavelet coefficient set to obtain a set of coefficients. Next, the component waveform extraction unit 20 uses a fitting filter to apply a degree of fit obtained based on the basic characteristics of the component waveform to the set of coefficients obtained in the analysis unit 10 (circled number 1 in Figure 5). Then, it uses an adjustment filter to obtain a set of wavelet coefficients corresponding to the component waveform (circled number 2 in Figure 5). Furthermore, if the fitting intensity is used, a wavelet set corresponding to the component waveform, taking magnitude into consideration, is obtained, and the characteristic information of the component waveform (feature quantities based on the characteristics of the component waveform) is analyzed (Case B in Figure 5). If the fitting intensity is not used, the characteristic information of the component waveform is analyzed using the wavelet coefficient set corresponding to the component waveform obtained using the adjustment filter (Case A in Figure 5).

[0123] [Analysis by block] When analyzing the behavior of component waveforms in a signal waveform to be analyzed, it is conceivable that multiple component waveforms may overlap because the time interval between consecutively occurring component waveforms is shorter than the time width of the component waveforms, or, for example, as in octave band analysis of vibration or noise analysis, it may be better to analyze multiple component waveforms as a single unit. Therefore, the feature information analysis unit 30 is provided with a block analysis unit (not shown), which, when the signal waveform to be analyzed contains multiple component waveforms, can treat them as a block rather than separating them, and analyze them as the average characteristics of the component waveforms contained in the block. The block analysis unit sets blocks based on the degree of fit or fit strength within a predetermined range. For example, it is possible to set a threshold as the predetermined range and set a block to include waveforms whose degree of fit or fit strength is equal to or greater than the threshold.

[0124] For example, let the threshold be ΔA, and extract continuous intervals where the goodness-of-fit MR or fitting strength exceeds this threshold as blocks of component waveforms. The continuous values ​​of the goodness-of-fit or fitting strength are s0, s1, ..., s N , s N+1 Therefore, the threshold is exceeded between s0 and s1, s N and s N+1 If the value falls below the threshold within a certain range, that is, if the value exceeds the threshold consecutively, that range is extracted as a block, and the width of that block is defined as N+1. If the fit strength is used, the fit strength is given by, for example, MR*WSP or MR*CoB.

[0125] Furthermore, changes in the goodness-of-fit or fit intensity during the time interval in which a component waveform exists take the shape of a peak waveform at the time the component waveform exists, and if multiple component waveforms are contained within a block, they appear as multiple peaks within the block. Therefore, it is also possible to estimate the number of component waveforms contained within this block (hereinafter referred to as the "estimated number of component waveforms") by the number of changes in the goodness-of-fit or fit intensity that appear as such peak waveforms.

[0126] The change in the values ​​of fitness or fitness strength within a block d0, d1, ..., dN (However, d i =s i+1 -s i ) is calculated. At the beginning and end of the block, the threshold is crossed, so d0 > 0, d N <0. This change in magnitude allows us to estimate the number of component waveforms contained in the block by truncating the decimal part of ((number of inversions + 1) / 2) and converting it to an integer. Furthermore, if the value of the change fluctuates with small increases and decreases, and simply counting the number of positive and negative changes would result in a very large number of reversals, it is possible to set a threshold range and efficiently calculate the number of changes while ensuring accuracy or for analytical purposes. For example, it is effective to define a threshold ΔZ and treat the range from -ΔZ to +ΔZ as the zero zone (value 0), and not count any change in sign unless it exceeds this range. These subtle fluctuations could be caused by noise, slight time differences in the arrival of synchronously changing component waveforms, or time differences due to differences in the time and location of generation of the component waveforms. It is useful to take these factors into account when conducting analysis.

[0127] Next, the block analysis unit can analyze the block's variation characteristics based on the magnitude of the temporal change in the component waveforms within the block (hereinafter referred to as "block variation characteristics"), the magnitude of the output from the block (hereinafter referred to as "block output characteristics"), and the degree of convergence of component waveforms based on the block's output characteristics.

[0128] Block variation characteristics can be analyzed by a block variation analysis unit (not shown) located in the block analysis unit. Block variation characteristics represent the average behavioral changes caused by multiple component waveforms contained within a block, and are given by the block variation degree and the block variation evaluation value. The block variation degree indicates the time change of feature quantities expressed in units of the block, based on the magnitude and frequency characteristics of each component waveform contained within the block, and is expressed as the rate of change over time of a value based on the average behavioral characteristics of the block and the amount of characteristic change per component waveform in the block. The block variation evaluation value is expressed as the average value of the block variation degree of blocks contained within a predetermined time range.

[0129] The average behavioral characteristics of a block and the amount of characteristic change per component waveform of the block can be determined based on the individual features of the component waveforms contained within the block, and the rate of change over time can be calculated by dividing by the width of the block. Hereinafter, a detailed explanation is given based on the case in which block variability and block variability evaluation values ​​are used in electromyography waveform analysis as one example.

[0130] It is known that as muscle fatigue progresses, the signal transmission velocity of muscle fibers decreases. If the transmission velocity slows down, it can be inferred that the rate of change in potential at the potential measurement site will also slow down, and the time change in the value of the fitted intensity will also decrease. However, when multiple motor unit activities are fused, it is difficult to appropriately evaluate the rate of change of individual motor units. In such cases, block variation characteristics can be used for evaluation. Specifically, first, the amount of change in the characteristic quantity of each motor unit action potential waveform within a block is divided by the estimated number of motor unit action potential waveforms, that is, the estimated number of component waveforms, to calculate the amount of characteristic change per component waveform.

[0131] It is also known that as muscle fatigue progresses, the distribution of frequency components shifts toward lower frequencies. In terms of characteristics based on wavelet coefficient values, this appears as a decrease in the value of CoB or LCoB, that is, a decrease in frequency feature values, which in turn manifests as a decrease in the frequency feature value of the motor unit action potential waveforms contained within a block. Therefore, as the frequency feature value of a block, an average value is calculated by multiplying the frequency feature value of each motor unit action potential waveform by its fitness.

[0132] From these points, it is considered that how active motor unit activities are (whether they are active in a low-fatigue state) can be grasped based on the latter base value and the former value, which can be described as the intensity of individual activities. Thus, this is defined as the motor unit activity coefficient per block. Furthermore, since this motor unit activity coefficient has the property of decreasing in value as motor unit fatigue progresses, it is possible to analyze the activity level of motor units, that is, the fluctuation characteristics of blocks.

[0133] As described above, when the muscle is in a low-fatigue state, the component waveforms are accompanied by temporal changes in frequency distribution, such as increases and decreases in frequency components up to higher frequencies. When in a high-fatigue state, the temporal changes in frequency component distribution are limited to changes up to lower frequencies than in the low-fatigue state. It is also known that when a muscle is in a low-fatigue state, the muscle's signal transmission speed is high, and when in a high-fatigue state, the signal transmission speed decreases. If the signal transmission speed is high, the temporal change in the frequency feature value is large, and if the speed is low, the temporal change in the frequency feature value becomes gradual. In view of this, the amount of characteristic change per component waveform, that is, the evaluation of the amount of change per motor unit, is defined as sqrt((Σ k (dd k 2 )) / CMU) and incorporated into the evaluation of motor unit activity. Here, CMU is the estimated number of motor unit action potential waveforms contained in a block, and dd k is MR*CoB samples in the block (ss i , 1<ss i <N, N+1; feature change amounts dd0, dd1, ..., dd between values of block width (BW) N (where dd i = ss i+1 - ss i) . At the beginning and end of the block, the threshold for the block setting is crossed, so dd0>0, dd N The result is <0. Furthermore, since the motor unit activation coefficient is evaluated based on the motor unit action potential waveform being analyzed, the frequency features used are those obtained by multiplying the frequency features by the goodness of fit (MR), which indicates that the motor unit action potential waveform contained in that block is the waveform being analyzed, i.e., the waveform is likely to be a component waveform.

[0134] Therefore, the motor unit activation coefficient defined at the block level is denoted as MUActCoef(t) and defined as follows: (1) t is outside the block: MUActCoef(t) = 0 (2) t is inside the block: MUActCoef(t) = (AVE C +sqrt((Σ k (dd k 2 )) / CMU))*(CMU / BW) For t belonging to the same block, MUActCoef(t) will have the same value.

[0135] AVE C This is the average value of MR*CoB values, AVE C =sqrt(Σ k (ss k ) / BW=sqrt(Σ k (ss k It is given by ) / (N+1) (where k=1,...,N). In this way, the block variation characteristic can be expressed by dividing the average value of the feature quantities of each component waveform within the block, which is the basis of the block's features, and a value based on the amount of feature change per component waveform of the block, which indicates the change in the block's features, by the block width, which is the time width of the analysis target, and thereby obtaining information about the variation of the block.

[0136] Furthermore, muscle fatigue does not progress uniformly in all motor units that make up a muscle; rather, it is natural to assume that low-fatigue motor units and fatigued motor units are active together. Since MUActCoef evaluates the average activity level of motor unit activity groups fused into a single block, even when fatigue is progressing, low-fatigue and high-fatigue blocks will appear together.

[0137] Therefore, the evaluation of muscle conditions such as muscle fatigue must be based on the average characteristics of multiple blocks within a specific time interval, and this can be done using block variation evaluation values. Specifically, a predetermined time range to be evaluated is set, the average of the MUActCoef values ​​at each time point within that range is calculated, and this is expressed as the Muscular Activity Evaluation Value (MActEval) for evaluation. The predetermined time range can be set as appropriate depending on the purpose of the analysis and the required analytical accuracy, etc. For example, in the embodiment described later, 100~200 [ms] is used.

[0138] There are various values ​​for the mean, such as the arithmetic mean, root mean square (RMS), and harmonic mean, and they can be appropriately selected and used depending on the purpose of the analysis and the required analytical accuracy.

[0139] The muscle activity level evaluation value tends to decrease as overall muscle fatigue (average fatigue of the motor units that make up the muscle) progresses. When evaluating using the motor unit activity coefficient, it is possible to directly capture and understand the fatigue of motor units. When evaluating using the muscle activity level evaluation value within a predetermined time range, it becomes possible to capture the overall behavior of the blocks within that predetermined time range. For example, as muscle fatigue progresses, the need for individual motor unit activities to complement each other and synchronize increases, and the tendency for blocks within this predetermined time range to become sparser increases.

[0140] When calculating the average of muscle activity evaluation values, there are two approaches: one that evaluates based only on the state during activity (primarily based on the state of the motor unit), targeting only the blocked portion within a predetermined time range; and another that includes the tendency for blocks to become sparser (a comprehensive muscle activity evaluation that includes the need for synchronization of motor unit activity), including the portion outside the block as a value of 0. Depending on the purpose and situation of comprehensively evaluating muscle activity, one or both of the above methods can be used in combination. Furthermore, to directly address the tendency for blocks to become sparser, it is also possible to evaluate and combine the proportion of time occupied by blocks within a predetermined time range. Note that when evaluating muscle activity in combination with the muscle activity reserve coefficient described later, the tendency for blocks to become sparser is a stronger evaluation element in the muscle activity reserve coefficient, so it is better to average only the blocked portion in the muscle activity evaluation value.

[0141] Furthermore, it is well known that the intensity (amplitude) of electromyographic waveform signals is strongly related to muscle activity intensity (force exerted). However, it is also well known that the amplitude increases not only when the force exerted increases, but also as muscle fatigue progresses, even at the same force exerted. Therefore, it is conceivable that it may be difficult to evaluate muscle fatigue using the signal intensity of electromyographic waveforms during exercise, where it is difficult to calculate the changing force exerted.

[0142] In this case, since the motor unit activation coefficient (MUActCoef) is calculated almost entirely based on frequency characteristics, it is less susceptible to the influence of signal intensity that fluctuates with force exerted, and the stability of the muscle activity evaluation value (MActEval) based on this motor unit activation coefficient is good even when the load (force exerted) fluctuates greatly.

[0143] Furthermore, because evaluation can be performed even in short intervals within a predetermined time range, it has a high ability to track frequent changes in muscle activity during exercise. These functional characteristics make it possible to perform various analyses in real time.

[0144] [Differentiation] The following are variations on the block variation evaluation value. The block variation evaluation value shows the average characteristics of a block within a specific time interval. However, when attempting to analyze changes in these characteristics in detail, it may be difficult to capture them with graphs or other methods due to the minute nature of the changes. Therefore, by using the block variation evaluation value as the value and performing a power calculation using the following formula, the block variation evaluation value can be numerically expanded to analyze its characteristics more clearly. Z=Y*((X (scale*(base-value)) ) ― zbase) Z; Scaling of block variation evaluation value Real numbers that control the range of the Y;Z values. zbase; A real number that controls the zero position in the range of Z values. X; a positive real number that serves as the base for the expansion. scale; a non-zero real number that determines the degree of magnification. base; value that represents the center of the magnification (magnification 1.0). Unless there are special circumstances, it is preferable to set the value of X to 2, the value of Y to 1, and zbase to 0, and then set the degree of increase in the value of Z (increase in slope) according to the value of (base-value) using scale. For other values, X and scale can be set by considering the degree of increase in the value of Z (increase in slope) within the range of values ​​that (base-value) can take, and Y and zbase can be set by considering the range of Z values ​​when X and scale are defined. Here, we will describe in detail below one example of using the above magnified values ​​in electromyography waveform analysis.

[0145] While the aforementioned muscle activity evaluation values ​​are sufficiently sensitive to changes due to fatigue, these changes may not be clearly visible. Furthermore, fluctuations in values ​​exist due to variations in the degree of fatigue of the motor units being used. Therefore, even if the values ​​are presented as they are in a graph, it may be difficult to visually perceive the differences resulting from these changes.

[0146] Since the muscle activity evaluation value can be considered to simultaneously indicate "how active" and "how fatigued," the present invention defines a mathematical formula to amplify the change due to fatigue (relatively reducing the change due to activity) and uses the transformed value for analysis. Furthermore, since this transformed value represents the change due to fatigue, it can be defined as the muscle fatigue change coefficient (FatigueCoef) in electromyography waveform analysis to analyze muscle fatigue. This coefficient takes a value close to 0 when the muscle fatigue level is low (high activity level) and shows a larger value as the fatigue level increases.

[0147] Various conversion formulas can be devised to emphasize the difference, and there are no particular constraints, and the essence of the value does not change, so this conversion formula does not take us outside the scope of the present invention. However, as an example, in the present invention, the muscle activity evaluation value is 2 (scale*(base-value)) This can be used. Muscle activity evaluation values ​​tend to decrease as fatigue progresses, but the numerical difference due to this change is small, so the role of scale is to amplify the value by raising it to a power in order to make the difference easier to understand. Base and scale function as sensitivity adjustments when performing analysis based on muscle activity evaluation values. That is, base is a value that serves as the boundary for whether or not the activity falls under normal conditions for the analysis (e.g., no fatigue or low activity), and scale is a value that is determined by how finely you want to detect even small amounts of boundary crossings.

[0148] If the base value is set to a negative value for the scale, representing an approximate evaluation of muscle activity when the muscles are highly fatigued, then the lower the muscle activity, the higher the value. Conversely, if the base value is set to a positive value for the scale, representing an approximate evaluation of muscle activity when the muscles are low-fatigued, then the higher the value, the more advanced the muscle fatigue. For example, to analyze how far muscle fatigue has progressed, the base value can be set (scale set to a positive value) based on the value measured when the electrodes were attached (when the muscles were low-fatigued).

[0149] In the embodiment described later, the muscle activity evaluation value at low fatigue (a lower value within the range judged as low fatigue is recommended) is taken as the reference reference value, and the muscle fatigue evaluation coefficient (FatigueCoef) is obtained by setting base=reference / 2 and scale=1 / (0.2*reference).

[0150] [Block Output Characteristics] Next, the block output characteristics will be described. The block output characteristics can be analyzed by a block output characteristic analysis unit (not shown) provided in the block analysis unit. The block output characteristics indicate the magnitude caused by the behavior of component waveforms included in a block, are represented by a block output intensity and a block output estimated value, and are calculated based on the feature quantities of these component waveforms. First, the block output intensity will be described.

[0151] A plurality of component waveforms are superimposed and exist in a block. Under such a situation, the magnitude output from the block is generally evaluated based on the amplitude of each component waveform. In this case, since it is considered that each component waveform fluctuates while its signal transmission speed changes, further improvement in analysis accuracy for the magnitude can be expected when frequency changes of the component waveforms caused by the change in the signal transmission speed are also taken into consideration. That is, when fluctuations in amplitude accompanying frequency changes of each component waveform in the block are also taken into consideration, the magnitude caused by the block can be evaluated more accurately.

[0152] Therefore, the degree of change in magnitude caused by a block is defined as the block output intensity, which can be expressed based on the block fluctuation degree that indicates temporal fluctuation characteristics incorporating the signal strength feature quantity (WSP) indicating the magnitude (intensity) of each component waveform in the block and the frequency characteristics of the block. Specifically, it is set by multiplying the signal strength feature (WSP) of each component waveform in the block multiplied by the matching degree (MR) by the block fluctuation degree. Here, as an example of the block output intensity, a detailed description will be given for the case where it is used in myoelectric waveform analysis.

[0153] When estimating the force that can be generated by a muscle (hereinafter referred to as "force exerted") based on electromyographic waveforms, it is common to use the rectified mean (ARV) or root mean square (RMS) of measured values ​​within a predetermined time range. On the other hand, since the increase or decrease in amplitude is also affected by muscle fatigue, it is necessary to consider the progression of muscle fatigue, and there is a need for a method that has low dependence on amplitude and can capture the degree of muscle fatigue with high sensitivity.

[0154] Therefore, the block output characteristic analysis unit can estimate and evaluate muscle force using a value (hereinafter referred to as the "force evaluation coefficient") based on the motor unit activation coefficient, which is the block variability, and the signal intensity feature (WSP) of each motor unit activity waveform, which is the component waveform within the block, as parameters for estimating muscle force. Muscle force does not appear only at the moment when the motor unit activity waveform, which is the component waveform, exists; rather, there is a change in force that depends on the process of muscle contraction during the time interval in which the motor unit activity waveform exists. The instantaneous value of this change is approximated by multiplying the WSP at each time by the goodness of fit (MR). Specifically, the force evaluation coefficient (PowerCoef(t)) can be expressed by the following equation, where the motor unit activation coefficient, signal intensity feature, and goodness of fit at time t are MUActCoef(t), WSP(t), and MR(t), respectively. PowerCoef(t)=MUActCoef(t)*sqrt(MR(t)*WSP(t)) As mentioned above, MUActCoef(t) is the same value for t within a single block.

[0155] The motor unit activation coefficient captures the activity efficiency from the perspective of the motor unit, which forms the basis of muscle activity. The signal intensity feature (WSP) represents the magnitude and power (intensity) of the activity of motor units (muscle fibers) that are simultaneously active. The performance evaluation coefficient is the product of these two, i.e., "efficiency × quantity" based on the motor unit, and can be expressed as a feature that serves as the standard for evaluating performance.

[0156] Next, we will explain the estimated block output. The estimated block output is expressed as the average value of the block output degree of blocks within a predetermined time range. Since the block output degree is the instantaneous value of the size (intensity) of a block that occurs at a certain time, based on this average value of the block output degree, it is possible to understand the changes in block output, etc., due to the behavior of the component waveforms within the block, within this predetermined time range. In addition to the usual arithmetic mean, there are various values ​​such as moving average, root mean square (RMS), and harmonic mean, and these can be appropriately selected and used depending on the purpose of the analysis and the required analytical accuracy.

[0157] Here, we will explain the block power estimate using the example of its application in electromyography waveform analysis. The power evaluation coefficient mentioned above is the instantaneous value when motor unit activity is excited. On the other hand, when it is necessary to evaluate the duration of muscle activity excited over a continuous period of time, such as when continuously applying a load to capture the activity status of a muscle, the block power estimate can be used for analysis. In this case, the block power estimate can be expressed as the power estimate value (PowerEst), corresponding to the power evaluation coefficient mentioned above. The power estimate can be calculated, for example, by a moving average of the power evaluation coefficient. Since the power evaluation coefficient has high responsiveness to frequent changes in muscle activity during exercise, it is possible to evaluate muscle power in real time and accurately. The relationship between the power estimate value and the magnitude of the load has far better linearity than the RMS of the amplitude of the electromyography waveform signal, even when there are changes in the muscle fatigue state.

[0158] For example, regarding the estimation of the force exerted in muscle activity by fast-twitch and slow-twitch muscle fibers, the force exerted by each can be estimated by calculating the moving average of the force evaluation coefficients for fast-twitch and slow-twitch muscle fibers, and then summing them up to estimate the overall force exerted by the muscle.

[0159] Next, we will explain the component waveform dispersion. Regarding the behavior of component waveforms in the signal being analyzed, the degree of dispersion of component waveforms—how much the appearance of multiple component waveforms is dispersed or synchronized (hereinafter referred to as "component waveform dispersion")—is also considered an important characteristic. For example, if the appearance of multiple component waveforms is concentrated in close succession, they are likely to form a single block, with gaps appearing between blocks. Furthermore, if the synchronization increases and the way the component waveforms forming the blocks appear becomes more uniform, the number of blocks per unit time will decrease due to the concentration, and the variability in the geometric characteristics of each block is also expected to decrease.

[0160] The block output characteristics described above allow us to estimate and understand the rate of change in the size of the output from a block over time, as well as the average size of each block within a predetermined time range. On the other hand, it would be extremely useful if we could capture changes in the size and number of these blocks as they change over time, such as changes in the fusion of blocks, or changes in the distribution of block appearances, such as the time intervals between blocks within a predetermined time range.

[0161] Therefore, a convergence / dispersion analysis unit (not shown) is provided in the block analysis unit to quantitatively calculate the convergence / dispersion of component waveforms. Under these conditions, it is possible to understand the fluctuations and changes in the output by the blocks over a predetermined time and evaluate the trends of the component waveforms in the signal waveform.

[0162] [Component waveform convergence] Regarding the component waveform clumping coefficient (ClumpingCoef), as described above, as variations in block size and other factors progress, the following changes in ssrr values ​​over a predetermined time are predicted. However, N is the number of blocks (or gaps between blocks) within the predetermined time, and x1, ..., x are the target values ​​for each block (or gap between blocks). N When this is the case, ave=( Σ i (x i )) / N and ssrr = Σ i (((x i-ave) / ave) 2 ) (1) With respect to the time width of a block and the time width of the gap between blocks, the sum of squares of the residuals (ssrr1) obtained by ratio the average time width of all blocks within a predetermined time (ave1) to the mean value, and the sum of squares of the residuals (ssrr2) obtained by ratio the average time width of all gap between blocks within a predetermined time (ave2) to the mean value. (2) With respect to the average value of a block, the sum of squares of the residuals (ssrr3) obtained by taking the ratio of the residuals to the average value of all blocks within a given time (ave3) and the average value. (3) With respect to the number of component waveforms estimated within a block, the sum of squares of the residuals relative to the mean (ave4) of all blocks within a predetermined time period, as a ratio to the mean (ssrr4) (4) With respect to the sum of squares of the inter-sample changes within a block, the sum of squares of the residuals relative to the mean (ave5) of all blocks over a predetermined time period, as a ratio to the mean (ssrr5)

[0163] Alternatively, the following changes in component waveform convergence / convergence values ​​are predicted. These can be selected as needed during analysis. (1) Number of blocks (p1), and number of gaps between blocks (p2) (2) With respect to the average value of a block, the sum of squared residuals with respect to the average of all blocks within a given time period (p3) (3) Regarding the number of component waveforms estimated within a block, the sum of squares of residuals relative to the mean of all blocks within a given time (p4) (4) With respect to the sum of squares of the inter-sample changes within a block, the sum of squares of the residuals relative to the mean of all blocks over a given time period (p5)

[0164] Since the number and duration of blocks and the gaps between blocks are not necessarily independent relationships, it is recommended to include both ssrr1 and ssrr2, and p1 and p2, in the evaluation. However, this does not prohibit including only one of them in the evaluation. Although p1 and p2 essentially refer to the same thing, both are included as elements to reduce the possibility that the counting situation at the boundary of the analysis interval of a given time width may have a negative impact. If two features with weak independence are included independently, their influence will become unequally large. Therefore, for ssrr1 and ssrr2, we will treat sqrt(ssrr1*ssrr2), and for p1 and p2, considering that the value is 1 or greater when it is possible to evaluate the convergence of component waveforms, we will treat sqrt((p1-1)*(p2-1)) as one of the features to be incorporated into the evaluation formula. The ssrr1 and ssrr2 pair is suitable when the difference in the degree of convergence of component waveforms tends to manifest as differences in block shape, while the p1 and p2 pair is suitable when the difference in the degree of convergence tends to manifest as an increase or decrease in the number of blocks within a given time interval.

[0165] There is a strong correlation between ssrr3 and p3, ssrr4 and p4, and ssrr5 and p5. The former (ssrr) deals with the degree of variation as a relative quantity, while the latter (p) deals with the degree of variation as an absolute quantity. If the difference in the degree of convergence of component waveforms tends to manifest as excess variation regardless of the difference in mean values, then choosing the absolute quantity is more appropriate. On the other hand, if the amount of variation tends to decrease as the mean value decreases, then choosing the relative quantity is more appropriate. Therefore, depending on the characteristics of the component waveform being targeted, you may replace the elements described as ssrr3~5 with p3~p5 in the following text. Of course, you may also treat them as different types of features based on their differences and include both in the calculation formula.

[0166] When analyzing a signal while moving through analysis intervals of a predetermined time width, blocks or gaps between blocks typically straddle the boundaries of the analysis intervals. In this case, it is inappropriate to extract only the portion of the analysis interval from the blocks or gaps located at the boundary and include it in the calculation. When blocks or gaps between blocks straddle a boundary, the system should choose whether to include or exclude the entire block or gap. As a result of this boundary processing, the total number of samples (winN) involved in the calculation at each predetermined time interval will vary.

[0167] Using these numerical values, the component waveform clumping degree (Clumping Coef) is given by the following formula. ( (sqrt(ssrr1 * ssrr2) * ssrr3 * ssrr4 * ssrr5) (1 / 4) ) / winN This formula is obtained by dividing the geometric mean of the ssrr of (1) to (4) mentioned above (wherein for (1), it is the geometric mean of two ssrr) by the time width of a predetermined period. It should be noted that the use of all of ssrr1 to ssrr5 is not mandatory; only a part may be selected, and the value obtained by dividing the geometric mean thereof by winN may be used. When only one of ssrr1 and ssrr2 is used, the selected ssrr is directly used instead of the geometric mean thereof. In addition, when only a part of (1) to (4) mentioned above is selected, or when the number of selected items increases as a result of introducing both ssrr and p, the geometric mean of the selected items is used as the numerator of the formula for the component waveform clumping degree.

[0168] When there are k types of component waveforms in the signal to be analyzed and it is desired to obtain the overall clumping degree of all types of component waveforms, the processing is performed as follows. (1) After superimposing the blocks of all component waveforms, ssrr1 and ssrr2, or p1 and p2 are obtained. (2) For each of ssrr3 to ssrr5, the ssrr of k types of component waveforms are defined as ssrr1,...,ssrr k , the ssrr for obtaining the overall clumping degree is (Π i (ssrr i )) (1 / k) given by.

[0169] When component waveforms in a signal to be analyzed appear discretely, the more stable the shape of the component waveforms and the more periodic the appearance thereof, the smaller the value of Clumping Coef. When a plurality of component waveforms appear with temporal overlap, the more concentrated the appearance thereof, the smaller the value of Clumping Coef. This makes it possible to quantify and evaluate one aspect of the appearance trend of component waveforms in the signal to be analyzed. The component waveform clumping coef can be used as is, but if the component waveform being targeted has specific characteristics, it can be used in combination with those characteristics. One example of this component waveform clumping coef is described below, which is applied to electromyography waveform analysis while incorporating characteristics specific to motor unit activity.

[0170] When the load on a muscle increases, many motor units are activated simultaneously to increase the force they generate. Since the number of motor units that make up a muscle is finite, when the number of simultaneous activations increases, there are not enough motor units to distribute the motor unit activity widely over time in order to stabilize the muscle's force, resulting in a concentration of motor unit activity.

[0171] This situation is often observed when muscle fatigue progresses, but it is not a direct result of muscle fatigue itself, but rather a result of the relative increase in load due to a decrease in the output of individual motor units caused by fatigue.

[0172] The concentration of motor unit activity, when viewed in terms of the degree of fitness, manifests as a decrease in the number of blocks generated within a given time and a homogenization of the block shape features. In other words, a situation where the number of blocks decreases and the variability of the various feature quantities of the blocks is lost can be considered a state of activity limit, where there is no room to create differences in motor unit activity that would affect these features. Based on the magnitude of this variability, a feature quantity used to evaluate how much margin a muscle has over the limits of motor unit activity is called the Tolerance Coefficient of Muscular Activity (ToleranceCoef). In motor unit activity, discretization of blocks similar to that seen in concentrated motor unit activity occurs even when muscle activity is low. When considering activity limits, it is necessary not to equate these two. Therefore, considering that the width of blocks tends to increase in cases of discretization due to high load or high fatigue compared to discretization due to low activity, the activity capacity coefficient (ToleranceCoef) is given by the following formula. ToleranceCoef=(ssrr2 / ssrr1)*ClumpingCoef Furthermore, the reciprocal of (ssrr1 / ssrr2), which is used as an adjustment term for the difference between muscle fatigue and low activity in this formula, tends to decrease in value as muscle fatigue progresses and concentration increases. Therefore, if we are only considering the evaluation of the degree of convergence of component waveforms, it is possible to introduce (ssrr1 / ssrr2) as an evaluation parameter.

[0173] The activity reserve coefficient will be large when there is a large variation in the blocks, for example when the required force can be obtained with only a subset of motor units and the force can be stabilized by distributing the excitation timing. Conversely, it will be small when there is a small variation in the blocks, for example when the required force cannot be obtained without concentrating the activity of motor units. Furthermore, in situations where the activity level of individual motor units is sufficiently high, such as during low fatigue, or during low activity, there is more leeway in how motor unit activity is concentrated, so fluctuations in the activity capacity coefficient tend to be large. However, when fatigue progresses and the activity level of motor units decreases, or during high activity, the leeway in how concentration is concentrated also decreases, and fluctuations in the activity capacity coefficient tend to be smaller. Therefore, when evaluating based on the activity capacity coefficient, it is desirable to look not only at the value over a short period of time, but also at trends such as the range of fluctuations in the value over a longer period of time.

[0174] Since the activity reserve coefficient cannot be used when there is no muscle activity, and the evaluation results tend to be inaccurate when muscle activity is minimal, it is preferable to use it when there is a certain level of muscle activity or higher.

[0175] The component waveform convergence index and the activity reserve coefficient, which are extensions of this index to motor unit activity, deal with the distribution of component waveforms in the signal being analyzed, but do not include the effect of activity summation of component waveforms caused by differences in distribution. If you wish to evaluate the summation effect, you can combine the component waveform convergence index, etc., with features that handle the summation effect you want to evaluate, such as signal intensity. An example of this in motor unit activity is described below.

[0176] The activity capacity coefficient can be described as a numerical representation of how much leeway there is to concentrate motor unit activity to gain further power from the current state within a given time, but it does not directly evaluate the leeway in terms of force. To deal with the magnitude of force, it is also necessary to consider the activity level of the motor unit within the same given time. When the estimated value of the leeway in terms of force is taken as the activity capacity estimate value (ToleranceEst), then ToleranceEst = (ToleranceCoef R It can be given by )*ave(MActCoef). Note that R is the weighting for the change in ToleranceCoef, and R=1 / 2 is particularly preferable. Also, ave(MActCoef) is the average value of MActCoef over a predetermined time period for calculating ToleranceCoef. As the average value, the root mean square can be used instead of a simple arithmetic mean over the predetermined time period.

[0177] The activity capacity estimate (ToleranceEst) indicates the magnitude of available force based on its value, but like the power output estimate (PowerEst), it does not directly represent the magnitude of the force. While this is sufficient for tracking relative changes in capacity, it is necessary to adjust for the difference in scale when comparing or adding it to the power output estimate, which is strongly related to evaluating muscle activity. Therefore, we introduce a constant scale factor (sf) as an adjustment term to create the activity capacity adjusted estimate SToleranceEst = sf * ToleranceEst, and use this when comparing it with PowerEst. The apparent result of the comparison between exerted force and reserve force changes drastically depending on the value given for sf. If there were a true limit value of muscle exertion force that could be accurately measured, it could be used as a standard for obtaining the optimal solution for sf. However, it is almost impossible to expect to obtain such an ideal standard value, and because the "correct answer" is unknown, it is not easy to determine an appropriate value for sf. One method for determining sf is to use data obtained by making several round trips of exertion force from a weak force to a sufficiently large force in a low-fatigue state, excluding outliers in ToleranceEst, to find the approximate peak value peakT and the approximate peak value peakP, excluding outliers in PowerEst, and set sf = peakP / peakT as the set value. The obtained set value can be fine-tuned as needed. Furthermore, while 100% MVC may not represent the true limit of performance, it would be beneficial to measure 100% MVC and then adjust the upper limit or average value of PowerEst+SToleranceEst during low-fatigue activity to be relatively stable.

[0178] While evaluation based on the activity capacity coefficient is strongly related to muscle fatigue, it offers a different perspective. Due to the difficulty in setting the activity capacity coefficient and the large temporal fluctuations (range) of the value, the estimated activity capacity does not precisely provide a specific value for the capacity. However, it does have a tendency for capacity to decrease when exertion is high and to recover when exertion decreases, and for the estimated capacity to decrease and the range of fluctuation to decrease as muscle fatigue progresses. Furthermore, the greater the load of continuous exercise for the subject, the faster the rate of deterioration of capacity from the perspective of the estimated activity capacity tends to be, and the temporal changes in characteristics such as the value and range of fluctuation are often linear in the process from the start of deterioration to approaching the limit of exercise. These properties are considered to be extremely useful for managing the load during exercise and determining the limit point.

[0179] In this invention, considering the ease of understanding the concept of the direction of change within the normal range of component waveforms, we will explain using motor unit action potential waveforms, which are component waveforms of electromyography (EMG) waveform signals that can be measured non-invasively, as one example. However, the scope of application of the method of this invention is not limited to EMG waveform signals.

[0180] In addition, the present invention can be applied not only to waveforms but also to data sets in problem domains where normal / abnormal discrimination is required, by making the data feature vectors under the condition that all vector components are greater than or equal to 0. Furthermore, the method of the present invention can be applied even when there is directionality in the variation range of the feature vector considered normal.

[0181] This invention provides a method for detecting and extracting component waveforms, and a method for performing real-time behavioral analysis of component waveforms based on these methods. In order to capture the behavior of a signal waveform being analyzed, it is important to understand and evaluate the characteristics of the component waveforms it contains, such as the amount and magnitude of change. However, conventional techniques mostly involve imposing specific conditions or constructing complex models, making it virtually impossible to predict in advance, evaluate in a short time, and perform real-time evaluation.

[0182] The technology described in Japanese Patent Publication No. 2016-063995 is one of the few technologies that aims to make this possible, but its main purpose is to visualize and evaluate phenomena that appear in observed waveforms. This invention further develops this by calculating various characteristic quantities of the component waveform that gives rise to the phenomenon, and by making it possible to evaluate the behavior of the component waveform based on these characteristic quantities, so that a more quantitative evaluation of the phenomenon can be performed.

[0183] According to the present invention, the following effects can be obtained. This technology enables the extraction of target component waveforms from a signal waveform being analyzed, and the quantitative behavioral analysis of those component waveforms, all within a single channel, in real time, with high accuracy and sensitivity—something that was virtually impossible with conventional technology. When this invention is used, particularly in electromyography waveform analysis, it can improve the accuracy of real-time evaluation of muscle fatigue and performance during exercise by enabling highly accurate muscle fatigue assessment. Furthermore, by providing advanced muscle activity information not available in conventional technologies, it contributes to technological advancements in the fields of medicine and sports science. Of course, since the present invention can extract and analyze component waveforms that are the target of analysis from irregular signal waveforms, it can be applied to various waveforms and component waveforms to be analyzed. For example, it can target silent voice recognition (so-called lip-syncing), heart sounds, electrical waveforms from electrocardiograms, electromyographic waveforms that capture muscle activity, or, in engineering terms, noise and vibration waveforms generated from equipment and various environments, or waveforms intended for fault detection of equipment, etc.

[0184] Rather than simply capturing a general overview of muscle activity from non-invasive surface electromyography (EMG) measurements, this technology enables analysis from the perspective of motor unit activity that constitutes muscle activity, while still using the same surface EMG data. This allows for direct investigation of the balance of contributions from fast-twitch and slow-twitch muscle fibers and the progression of muscle fatigue during exercise, contributing to the advancement of basic medicine and sports medicine, as well as helping to refine EMG-based devices such as MYO prosthetics and power assists. However, the application of this technology is not limited to surface EMG. [Examples]

[0185] [Example 1] The present invention will be described in more detail below with reference to examples, but the present invention is not limited to the following examples unless its essence is changed. First, as Example 1, an experiment described later was conducted to confirm the effects of the present invention.

[0186] [Measurement Method] The measurement method used in this experiment is described below. (1) Stand upright, bending your elbows to a 90-degree angle and keeping your forearms horizontal. (2) Measure the surface electromyography of the biceps brachii muscle when a belt is placed at the wrist and weight is applied (the forearm should be kept as horizontal as possible. The time for increasing and decreasing the weight should be approximately 3 seconds each).

[0187] The experimental method used in this experiment is described below. [Experimental Method] (1) The measurement method described above will be performed four times, with a relaxation break of approximately 20 seconds in between. (2) Perform dumbbell lifts for approximately 1 minute to fatigue the biceps brachii muscle to be measured. (3) The measurement method described above is performed five times with a relaxation break of approximately 10 seconds in between. Here, the phase in (1) above is referred to as "low fatigue," the phase in (2) above as "muscle fatigue exercise," and the phase in (3) above as "post-fatigue." In other words, electromyography measurements were taken four times during the low fatigue phase and five times after fatigue, using the measurement method described above.

[0188] Figures 6-15 illustrate the experiments conducted using the present invention. Figure 6 is a graph showing the changes in the weight values ​​during four measurements at low fatigue and five measurements after fatigue. As shown in Figure 6, the weight increases from approximately 3000ms to 4000ms, and then decreases thereafter. Furthermore, the differences in the changes in the weight values ​​for each measurement are sufficiently small.

[0189] Figure 7 is a graph showing the changes in the electromyography (EMG) waveform during each measurement (four measurements during low fatigue and five measurements after fatigue). In other words, it shows graphs of the EMG waveform measured at each timing of (1) to (3) in the experimental method described above. From this figure, no significant changes in the EMG waveform are observed between measurements during the four measurements during low fatigue. The amplitude increases and decreases in response to the increase and decrease in load. On the other hand, immediately after fatigue (bottom right of Figure 7), the EMG waveform changes significantly, and then, as 10-second breaks are taken in between, the changes in the EMG waveform gradually subside. In other words, it can be seen that the subject's arm (biceps brachii) gradually recovers even with only 10-second breaks in between.

[0190] Figure 8 is a graph showing the change in the RMS value (window width 200 ms) of surface electromyography for each measurement. Figure 9 is a graph showing the change in the result of plotting the weighted value on the horizontal axis and the RMS value on the vertical axis. The RMS value is a value that depends on the amplitude of surface electromyography and is an estimation method that is often used in conventional techniques for evaluating force exertion. However, as can be seen from Figures 8 and 9, there is a significant change in the characteristics of the value before and after fatigue. In other words, it can be said that it is stable in the range where the effect of muscle fatigue can be ignored, but when the degree of the effect of muscle fatigue cannot be evaluated, the possibility of all these states cannot be ruled out, so it is necessary to consider the correlation on a distribution that mixes samples of all states, and therefore the accuracy of estimating force exertion from the RMS value becomes extremely low.

[0191] Figure 10 shows an example of analysis using the present invention. The conditions for the analysis are as follows. • Maximum depth of redundant wavelet analysis: Level -11 • Wavelet coefficient set: T+7 / 16 (T+14 / 32) • Set the base vectors for fast-twitch and slow-twitch muscle fibers, as well as the fitted filter and split filter (see top of Figure 15). Under the above conditions, the coefficient set is obtained based on the wavelet coefficient set for the first test under low fatigue and the first test after fatigue, respectively. The goodness of fit for fast-twitch and slow-twitch muscle fibers is calculated, and a partition-type adjustment filter (fast-twitch / slow-twitch) is applied to detect the component waveform (wavelet coefficient set).

[0192] Figure 11 also shows the MR×LCoB of fast-twitch and slow-twitch muscle fibers, calculated similarly for the first test under low fatigue conditions and the first test after fatigue, respectively. A partition-type adjustment filter (fast-twitch / slow-twitch) was applied to graph the MR×LCoB of fast-twitch and slow-twitch muscle fibers. This shows that when muscle fatigue occurs, the high-frequency components decrease, meaning that sharp potential changes in the component waveform are lost, and a decrease in the frequency of large values ​​is observed. Furthermore, when muscle fatigue occurs, the rate of potential change in the component waveform decreases, meaning that the shape of the component waveform becomes blunted, and a calming of time variation is observed. In addition, when muscle fatigue occurs, the rate of change of the potential between bipolar induction electrodes decreases, meaning that the transmission velocity decreases, and an increase in the width of individual peaks is observed.

[0193] Figure 12 shows an example of fitting and adjusting the wavelet coefficient set. The conditions for the analysis are as described above. • Maximum depth of redundant wavelet analysis: Level -11 • Wavelet coefficient set: T+7 / 16 (T+14 / 32) In other words, the base wavelet coefficient set was "T+7 / 16(T+14 / 32)", and the baseline vectors for fast-twitch and slow-twitch muscle fibers, as well as the fitted filter and split filter, were set using the same wavelet coefficient set as described above.

[0194] As a result, when examining the change in the RMS ratio of fast-twitch and slow-twitch muscle fibers (see bottom of Figure 12), we see that it crosses around T+7 / 32 during low fatigue and around T+12 / 32 during fatigue. Therefore, T+13 / 32 was adopted as the optimal wavelet coefficient set.

[0195] Figure 13 shows the results of calculating the fit for fast-twitch and slow-twitch muscle fibers in the first low-fatigue test and the first post-fatigue test, respectively, using this optimal wavelet coefficient set, similar to Figure 11. A distributive adjustment filter (fast-twitch / slow-twitch) was then applied to graph the MR × LCoB of fast-twitch and slow-twitch muscle fibers. Furthermore, the motor unit activation coefficient was calculated.

[0196] Figure 14 shows the changes in the performance evaluation coefficients for each measurement in this experiment. For each measurement, the upper graph shows the performance evaluation coefficient for fast-twitch muscle fibers, and the lower graph shows the performance evaluation coefficient for slow-twitch muscle fibers. This shows that by incorporating the effects of muscle fatigue into the evaluation, the significant changes in the characteristics of the values ​​before and after muscle fatigue, as seen in the amplitude, RMS value, and WRMS value (the root mean square of the coefficient values ​​of the set of wavelet coefficients, which is the same as WSP except that it is divided by a constant term for averaging), were not observed, indicating that the adverse effects of muscle fatigue were reduced.

[0197] Furthermore, Figure 15 shows the changes in the results for each measurement in this experiment, with the horizontal axis representing the weighted value and the vertical axis representing the estimated force (moving average with a window width of 200 ms). This indicates that the correlation with weight (load) is far better than that of evaluation using the conventional RMS value. It is presumed that the higher values ​​observed when the weight increases and the lower values ​​observed when the weight decreases are due to the effect of coordination with antagonistic muscles to maintain the elbow angle.

[0198] [Example 2-1] Next, as Example 2-1, we present an example of setting the wavelet coefficient set using the mechanomyogram (MMG) waveform during electrical stimulation as the subject.

[0199] First, we created the virtual MMG data shown in Figure 16 by referring to the cited literature below. This data is an artificial reproduction of the muscle sound waveform produced by electrical stimulation. Cited document: “Evaluation method for muscles, measuring mechanomyogram induced by electrical muscle stimulation using lead zirconate titanate-based acoustic sensor”, Japanese Journal of Applied Physics 58, SLLD11 (2019)

[0200] Then, redundant wavelet analysis was performed on this virtual MMG data using "Daubechies N=2," and only the positive coefficients were extracted from the results. Figure 17 is a 3D graph of the extracted values, and Figure 18 is a contour graph of the extracted values. However, the curve in Figure 18 (see arrow in Figure 18) was added to show the grouping of sub-waveform components.

[0201] Here, we assume that the MMG waveforms shown in Figures 17 and 18 are the component waveforms of the signal being analyzed. Judging from the results of analyzing the virtual MMG waveform data, it can be seen that the MMG waveform is formed by a combination of two types of sub-component waveforms (the grouped portion shown by the two curves in Figure 18).

[0202] In redundant wavelet analysis, outputting multiple sets of wavelet coefficients simultaneously does not significantly change the computational cost. Therefore, the wavelet coefficient set setting and analysis procedure for extracting such component waveforms can be broadly divided into: (1) For each sub-component waveform, an independent set of coefficients is set and analyzed individually, and then an overall evaluation is performed based on the degree of fit for each corresponding time (for example, the dotted line position in Figure 18). (2) The coefficient sets for each sub-component waveform are concatenated and used as a single wavelet coefficient set, allowing for analysis using the same procedure as for component waveforms with simple properties. It could turn out to be something like that.

[0203] Method (1) above requires more processing effort, but it facilitates analysis when there may be changes in the characteristics of the sub-component waveforms. In contrast, method (2) above has the opposite properties to method (1) above, and is a simpler method when there is no need to worry about deformation of the component waveforms.

[0204] [Example 2-2] Next, a waveform was created by superimposing (combining) MMG waveforms. The time interval at which electrical stimulation was applied was randomly varied, and furthermore, the intensity of the MMG waveforms being added together was also randomly varied.

[0205] The graph shown in Figure 19 is a waveform created by superimposing MMG waveforms. The values ​​on the horizontal axis correspond to the time when electrical stimulation was applied, i.e., time 0 for each MMG waveform.

[0206] Figure 20 is a graph showing the results of redundant wavelet analysis performed on the composite waveform shown in Figure 19 using "Daubechies N=2," displayed as contour lines. The wavelet coefficient set corresponds to H+0. The letters "Hi" and "Lo" are added before the numerical values ​​of the level (frequency band) on the vertical axis; these indicate the correspondence between the frequency bands containing the high-frequency (Hi) and low-frequency (Lo) signal components of the two curves in Figure 18. For example, while both the high-frequency (Hi) and low-frequency (Lo) sides have a level of -6 (see Figure 18), "HiL-6" (see Figure 20) indicates the level of -6 on the high-frequency (Hi) side.

[0207] Furthermore, Figure 21 is a graph obtained by extracting only the positive portion of the wavelet coefficients (positive coefficients) from Figure 20 and displaying them as contour lines. From this figure, it is difficult to find a relationship with the existence time of the component waveform, the MMG waveform (time 0 of the MMG waveform).

[0208] Figure 22 shows an example where wavelet coefficients were extracted along the low-frequency curve for levels -8 and -7, and along the high-frequency curve for levels -6 to -1, creating a set of wavelet coefficients. Here too, only positive coefficients are extracted and displayed as contour lines. While this figure certainly reveals some issues, such as missed data, potential missed data, and potential false positives, it can be said that, compared to the results of simple redundant wavelet analysis, the extraction accuracy can be expected to be considerably higher by applying further processing, such as tuning, in some cases.

[0209] [Examples 2-3] Next, the wavelet coefficient set was configured to follow the curve in Figure 18, and the analysis was performed. However, as can be seen from Figure 22, the low-frequency band components were unreliable due to excessive waveform overlap, so the fit was examined using only the high-frequency band components, and a fitting filter as shown in Figure 23 was set up.

[0210] This fitting filter uses only the high-frequency band to evaluate the component waveform characteristics. However, to obtain the component waveform, it is necessary to estimate and add at least the low-frequency band components that were omitted during the component waveform characteristics evaluation. Therefore, an example of a derived adjustment filter set based on the fundamental vectors for evaluating component waveform characteristics is shown in Figure 24. Note that the wavelet coefficient set here is a simplified set that extracts only the prominent feature elements of the component waveform; therefore, even if you perform an inverse transform on the basis vectors, for example, the shape will differ significantly from that of the component waveform.

[0211] Figure 25 shows a comparison between the component waveform (inv-WT of base wavelet-set) and the waveform obtained by inversely transforming the wavelet coefficient set of the component waveform (pessudo MMG wave). The waveform that extends to tims:700ms is the component waveform (inv-WT of base wavelet-set) (the same applies to Figures 27, 29, and 30). This diagram shows that while some characteristics of the component waveforms are retained, the differences are significant. However, on the other hand, the peaks around 30ms and 300ms are accurately captured.

[0212] Furthermore, as an example of applying a derivation-type adjustment filter, examples of reproducing component waveforms detected at certain times are shown below. Figure 26 is the same figure as Figure 22, but examples of reproducing component waveforms detected at the times marked with circles in Figure 26 (67ms, 152ms, 226ms, 311ms, 499ms, and 525ms) are shown in Figures 27(A) to 27(F) in chronological order.

[0213] The "scale" values ​​in the legends of Figures 27(A) to 27(F) (see the upper right of each graph) represent the weights applied to the amplitude of the component waveforms when they are synthesized at each location, and the component waveform graphs show the results of this application. These examples demonstrate that the amplitudes of the component waveforms synthesized at each location are also captured well.

[0214] Furthermore, by defining the waveform using more wavelet coefficients, it is possible to make the reproduced waveform of the component waveform by the derived adjustment filter more detailed. Figure 28 is an example of a filter that has been extended by adding more wavelet coefficients to the derived adjustment filter shown in Figure 24. The time position of each coefficient is not shown, so this figure alone is insufficient information, but it is shown as a reference example when negative values ​​are also included in the setting.

[0215] Figure 29 shows the result of applying the filter shown in Figure 28 to the basis vectors. Figures 30(A) to 30(F) show the results of applying the filter from Figure 28 to each coefficient set detection result from the previous example. These findings indicate that the refinement of the derived adjustment filter improves the accuracy of component waveform reproduction.

[0216] [Example 3] Furthermore, in Example 3, based on the surface electromyography data obtained using the measurement and experimental methods performed in Example 1, estimated values ​​of fast-twitch and slow-twitch muscle force output and estimated values ​​adjusted for activity reserve were calculated.

[0217] Figure 31 is a graph showing the estimated activity capacity of fast-twitch muscle fibers in each measurement. Figure 32 is a graph showing the estimated activity capacity of slow-twitch muscle fibers in each measurement. Figure 33 is a graph showing the estimated combined activity capacity of fast-twitch and slow-twitch muscle fibers in each measurement. The evaluation of activity capacity is based on an unstable feature, the degree of variability in observed motor unit activity, and therefore the fluctuations in the estimated activity capacity are usually large. However, as shown in the dotted box in Figure 31, for example, when a high load is applied while the muscles are fatigued, there is little room for fluctuation in the degree of activity variability, and the estimated activity capacity shows low values ​​with low variability. On the other hand, as fatigue recovers during rest, the estimated activity capacity gradually becomes able to show high values ​​even under high load, and it can be seen that the waveform after recovery is quite similar to the waveform immediately after the start of the experiment (see, for example, the upper left and upper right graphs in Figure 31).

[0218] Figure 34 is a graph showing a combination of the estimated force output and the activity-reserve-adjusted estimated force output for fast-twitch muscle fibers in each measurement. Figure 35 is a graph showing a combination of the estimated force output and the activity-reserve-adjusted estimated force output for slow-twitch muscle fibers in each measurement. Figure 36 is a graph showing a combination of the overall estimated force output and the activity-reserve-adjusted estimated force output for fast-twitch and slow-twitch muscle fibers in each measurement.

[0219] The upper line graph (for example, (A) in the upper left of Figure 34) shows the sum of the estimated output value and the adjusted activity capacity value, while the lower line graph (for example, (B) in the upper left of Figure 34) shows the estimated output value.

[0220] While this can vary considerably depending on the adjustment of the scaling factor, the sum of the estimated output value and the activity capacity-adjusted estimate tends to fluctuate around a relatively stable value in the long term, although the range of fluctuation is large when muscle fatigue is not progressing. However, as muscle fatigue progresses, the decrease in the activity capacity-adjusted estimate becomes more pronounced as the load increases, and as shown in the dotted box in Figure 34, it can be seen that the amount of muscle activity being performed in a state of limited reserve increases.

[0221] [Example 4] Finally, as Example 4, the following experiment was conducted.

[0222] [Measurement Method] The measurement method used in this experiment is described below. (1) Stand upright and repeatedly move back and forth between a position where your elbows are extended and your arms are straight down, and a position where your elbows are bent at a 90-degree angle and your forearms are horizontal. (2) Measure the surface electromyography of the biceps brachii muscle when the biceps are raised and lowered for approximately 2 seconds each, and the repetitions are performed at a stable speed without using momentum. (3) One trial consists of continuing the exercise until 100 seconds have passed or the person is no longer able to perform the stepping motion. This is repeated three times with a break in between.

[0223] Figure 37 shows the measured surface electromyography waveforms. From top to bottom, the results of the first trial, second trial, and third trial are shown, respectively. From Figure 37, it can be seen that in the second and third trials, the participant was unable to lift the dumbbells and stopped midway.

[0224] Figure 38 is a graph showing the estimated activity capacity of fast-twitch muscle fibers in each measurement. Figure 39 is a graph showing the estimated force output of fast-twitch muscle fibers in each measurement, and the sum of the estimated force output and the activity capacity-adjusted estimated value.

[0225] On the other hand, Figure 40 is a graph showing the estimated activity capacity of slow-twitch muscle fibers in each measurement. Figure 41 is a graph showing the estimated output of slow-twitch muscle fibers in each measurement, and the sum of the output estimate and the activity capacity-adjusted estimate.

[0226] Figure 42 is a graph showing the estimated activity capacity of fast-twitch and slow-twitch muscle fibers combined for each measurement. Figure 43 is a graph showing the estimated combined output of fast-twitch and slow-twitch muscle fibers for each measurement, and the sum of the estimated output and the activity capacity-adjusted estimate.

[0227] As shown in Figures 38, 40, and 42, it can be seen that the estimated activity capacity of fast-twitch muscle fibers, slow-twitch muscle fibers, and the combined activity capacity of fast-twitch and slow-twitch muscle fibers all decrease as muscle fatigue progresses. In the second and third trials, insufficient rest meant that muscle fatigue had not fully recovered, resulting in a state of reduced performance due to muscle fatigue from the start, and exercise began with a low estimated activity capacity. The value further decreased as exercise continued, meaning that it became difficult to exert sufficient force, and the remaining capacity was lost, leading to the discontinuation of exercise. Furthermore, as shown in the sum of the estimated force output and the activity-reserve-adjusted estimated value in Figures 39, 41, and 42, in situations where muscle fatigue has not progressed much, such as at the beginning of the first trial, when the estimated force output decreases, the activity-reserve-adjusted estimated value increases slightly, suppressing the fluctuation in the total value. On the other hand, as muscle fatigue progresses, the total value decreases significantly, indicating a situation where only a very small amount of force can be exerted compared to the start of exercise. [Industrial applicability]

[0228] This invention can be used for analyzing signal waveforms, and for example, in rehabilitation involving muscle strengthening exercises, it can be used to apply appropriate loads for appropriate durations based on the analysis results. It can also be applied in the medical and sports fields, and is therefore industrially useful. [Explanation of symbols]

[0229] 10 Analysis Department 20 component waveform extraction section 30. Feature Information Analysis Department

Claims

1. An analysis unit calculates a set of wavelet coefficients for a signal waveform corresponding to a surface electromyographic waveform formed by the superposition of multiple single waves, and obtains a set of coefficients corresponding to the component waveform of the target based on the set of wavelet coefficients corresponding to the component waveform corresponding to the target motor unit action potential waveform, The system includes a component waveform extraction unit that calculates the degree of fit of the component waveform to the set of coefficients, detects the component waveform based on the degree of fit, and obtains the component waveform included in the signal waveform by reconstructing the set of coefficients corresponding to the component waveform using an inverse wavelet transform, The degree of fit is obtained by a basis vector representing the characteristics of the component waveform of the coefficient group of the coefficient set, and a vector based on the coefficient group of the coefficient set. For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. Using the degree of fit and an adjustment filter weighted based on the basic characteristics of the component waveform, The plurality of waveforms included in the coefficient group of the coefficient set and the component waveform are separated and adjusted into a wavelet coefficient group corresponding to the component waveform. Component waveform extraction system.

2. An analysis unit calculates a set of wavelet coefficients for a signal waveform corresponding to a surface electromyographic waveform formed by the superposition of multiple single waves, and obtains a set of coefficients corresponding to the component waveform of the target based on the set of wavelet coefficients corresponding to the component waveform corresponding to the target motor unit action potential waveform, The system includes a feature analysis unit that calculates the degree of fit between the coefficient set and the component waveform, detects the component waveform based on the degree of fit, and analyzes feature quantities based on the characteristics of the component waveform. The degree of fit is obtained by a basis vector representing the characteristics of the component waveform of the coefficient group of the coefficient set, and a vector based on the coefficient group of the coefficient set. For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. Using the degree of fit and an adjustment filter weighted based on the basic characteristics of the component waveform, The plurality of waveforms included in the coefficient group of the coefficient set and the component waveform are separated and adjusted into a wavelet coefficient group corresponding to the component waveform. Signal analysis system.

3. The component waveform extraction system according to claim 1, wherein the adjustment filter has a distribution type extraction filter that distributes the set of coefficients of the coefficient set to the set of coefficients of the coefficient set corresponding to each of the plurality of component waveforms detected based on the degree of fit.

4. The component waveform extraction system according to claim 3, wherein the adjustment filter applies a derivation type extraction filter that derives a group of wavelet coefficients corresponding to the component waveform detected based on the goodness of fit from the group of coefficients of the coefficient set.

5. The system includes a feature analysis unit that analyzes feature quantities based on the characteristics of the component waveforms detected by the component waveform extraction unit, The component waveform extraction system according to claim 4, wherein the degree of fit is corrected by a feature quantity indicating the intensity and / or properties of the component waveform.

6. The component waveform extraction system according to claim 5, wherein the feature quantity indicating the intensity is a signal intensity feature quantity indicating the intensity of the signal of the component waveform, and the feature quantity indicating the property is a frequency feature quantity based on the distribution of frequency components included in the component waveform.

7. The component waveform extraction system according to claim 5 or 6, wherein the feature analysis unit obtains a plurality of component waveforms included in the signal waveform as a single block based on the degree of fit.

8. The aforementioned analysis unit, For each of the multiple sub-waveform components included in the aforementioned signal waveform, a set of wavelet coefficients is calculated, and a set of coefficients is obtained based on each wavelet coefficient set. The aforementioned component waveform extraction unit is The component waveform extraction system according to claim 5 or 6, which calculates the degree of fit of the component waveform to the coefficient group of each of the coefficient sets.

9. The aforementioned analysis unit, The component waveform extraction system according to claim 5 or 6, wherein a group of sub-waveform components contained in the signal waveform are concatenated to form a single component waveform to calculate a group of wavelet coefficients, and a group of coefficients of the coefficient set is obtained based on the wavelet coefficient set.

10. The signal analysis system according to claim 2, wherein the adjustment filter has a distribution type extraction filter that distributes the set of coefficients of the coefficient set to the set of coefficients of the coefficient set corresponding to each of the plurality of component waveforms detected based on the degree of fit.

11. The signal analysis system according to claim 10, wherein the adjustment filter applies a derivation type extraction filter that derives a group of wavelet coefficients corresponding to the component waveform detected based on the degree of fit from the group of coefficients of the coefficient set.

12. The goodness of fit is corrected using feature quantities that indicate the intensity and / or properties of the component waveforms obtained by the feature analysis unit. The signal analysis system according to claim 11, wherein the feature quantity indicating the intensity is a signal intensity feature quantity indicating the intensity of the signal of the component waveform, and the feature quantity indicating the property is a frequency feature quantity based on the distribution of frequency components included in the component waveform.

13. The signal analysis system according to claim 10 or 11, wherein the feature analysis unit obtains a plurality of component waveforms included in the signal waveform as a single block based on the degree of fit.

14. The signal analysis system according to claim 13, wherein the feature analysis unit analyzes the average variation and the amount of variation change of the feature quantities of the component waveforms represented in block units within the block, and has a block variation analysis unit that analyzes the coefficient of variation and degree of variation of the block based on the average variation and the amount of variation change.

15. The signal analysis system according to claim 14, wherein the block variation analysis unit estimates the activity of the component waveform in response to the variation of the block based on the degree of variation of the block.

16. The component waveform extraction system allows The steps include: calculating a set of wavelet coefficients for a signal waveform corresponding to a surface electromyographic waveform formed by the superposition of multiple single waves, and obtaining a set of coefficients corresponding to the target component waveform based on the set of wavelet coefficients corresponding to the component waveform corresponding to the target motor unit action potential waveform; The step of calculating the degree of fit of the component waveform to the set of coefficients of the coefficient set, detecting the component waveform based on the degree of fit, and obtaining the component waveform included in the signal waveform by reconstructing the coefficient group corresponding to the component waveform using an inverse wavelet transform, In these steps, The degree of fit is obtained by a basis vector representing the characteristics of the component waveform of the coefficient group of the coefficient set, and a vector based on the coefficient group of the coefficient set. For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. Using the degree of fit and an adjustment filter weighted based on the basic characteristics of the component waveform, The plurality of waveforms included in the coefficient group of the coefficient set and the component waveform are separated and adjusted into a wavelet coefficient group corresponding to the component waveform. Component waveform extraction method.

17. The signal analysis system, The steps include: calculating a set of wavelet coefficients for a signal waveform corresponding to a surface electromyographic waveform formed by the superposition of multiple single waves, and obtaining a set of coefficients corresponding to the target component waveform based on the set of wavelet coefficients corresponding to the component waveform corresponding to the target motor unit action potential waveform; The method comprises the steps of calculating the degree of fit between the coefficient set and the component waveform, detecting the component waveform based on the degree of fit, and analyzing feature quantities based on the characteristics of the component waveform. In these steps, The degree of fit is obtained by a basis vector representing the characteristics of the component waveform of the coefficient group of the coefficient set, and a vector based on the coefficient group of the coefficient set. For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. Using the degree of fit and an adjustment filter weighted based on the basic characteristics of the component waveform, The plurality of waveforms included in the coefficient group of the coefficient set and the component waveform are separated and adjusted into a wavelet coefficient group corresponding to the component waveform. Signal analysis method.

18. Computers, An analysis unit calculates a set of wavelet coefficients for a signal waveform corresponding to a surface electromyographic waveform formed by the superposition of multiple single waves, and obtains a set of coefficients corresponding to the component waveform of the target based on the set of wavelet coefficients corresponding to the component waveform corresponding to the target motor unit action potential waveform, The system includes a component waveform extraction unit that calculates the degree of fit of the component waveform to the set of coefficients, detects the component waveform based on the degree of fit, and obtains the component waveform included in the signal waveform by reconstructing the set of coefficients corresponding to the component waveform using an inverse wavelet transform, The degree of fit is obtained by a basis vector representing the characteristics of the component waveform of the coefficient group of the coefficient set, and a vector based on the coefficient group of the coefficient set. For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. Using the degree of fit and an adjustment filter weighted based on the basic characteristics of the component waveform, The plurality of waveforms included in the coefficient group of the coefficient set and the component waveform are separated and adjusted into a wavelet coefficient group corresponding to the component waveform. A program that operates as a component waveform extraction system.

19. Computers, An analysis unit calculates a set of wavelet coefficients for a signal waveform corresponding to a surface electromyographic waveform formed by the superposition of multiple single waves, and obtains a set of coefficients corresponding to the component waveform of the target based on the set of wavelet coefficients corresponding to the component waveform corresponding to the target motor unit action potential waveform, The system includes a feature analysis unit that calculates the degree of fit between the coefficient set and the component waveform, detects the component waveform based on the degree of fit, and analyzes feature quantities based on the characteristics of the component waveform. The degree of fit is obtained by a basis vector representing the characteristics of the component waveform of the coefficient group of the coefficient set, and a vector based on the coefficient group of the coefficient set. For the set of coefficients obtained by applying a fitted filter weighted based on the component waveform to the set of coefficients, the degree of fit with the component waveform is calculated. Using the degree of fit and an adjustment filter weighted based on the basic characteristics of the component waveform, The plurality of waveforms included in the coefficient group of the coefficient set and the component waveform are separated and adjusted into a wavelet coefficient group corresponding to the component waveform. A program that operates as a signal analysis system.

Images