JP6515670B2 - Sleep depth estimation device, sleep depth estimation method, and program - Google Patents

Description

  The present invention relates to a sleep depth estimation device, a sleep depth estimation method, and a program.

  2. Description of the Related Art In recent years, in order to determine a subject's state of health, such as the state of sleep, a technique has been developed that detects vibrations of the subject's breathing and body movement without contact and estimates sleep depth based on the detected vibration information. . For example, in Patent Document 1 below, the subject's sleep depth is estimated by measuring the subject's vibration through an air pad provided in the bedding used by the subject, and calculating the subject's cycle of vibration from the measured signal. Technology is disclosed. Further, in Patent Document 2 below, with respect to an IQ signal (I: In-phase, Q: Quadrature-phase) indicating the vibration of a subject acquired by a Doppler sensor, from the norm of the signal on a two-dimensional plane called IQ plane. A technique is disclosed that calculates the period of vibration by specifying the boundary of the waveform for each period of vibration, and determines the sleep depth of the subject based on the calculated period of vibration.

Unexamined-Japanese-Patent No. 2006-263032 JP, 2014-014708, A

  However, in Patent Document 1 described above, since the fine fluctuation included in the vibration waveform is reduced by the low pass filter, the waveform of the vibration including the fine fluctuation is not accurately extracted. Therefore, it is difficult to capture changes in the sleep state including disturbance of the breathing cycle. Further, in Patent Document 2 described above, when noise is superimposed on an IQ signal indicating vibration, or when vibration is continuous and occurs continuously, a stop period of vibration that is a division of the period of vibration is detected. As this is less likely to occur, the accuracy of the estimated period of vibration is reduced, and the accuracy of the estimated sleep depth is also reduced accordingly. Therefore, it is desirable to accurately estimate the period of vibration including minute fluctuations and to improve the estimation accuracy of the sleep depth of the subject.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a new and improved sleep depth estimation capable of estimating the sleep depth of a subject with higher accuracy. It is in providing an apparatus.

  To solve the above problems, according to one aspect of the present invention, there is provided a signal conversion unit for converting a beat signal detected by a Doppler sensor into a one-dimensional signal according to the vibration of a living body, and a frequency of the one-dimensional signal. A frequency estimation unit for estimation and a reference signal having a section length corresponding to a period corresponding to the frequency of the estimated one-dimensional signal are cut out from the one-dimensional signal, and the phases of the reference signal and a sine wave having the frequency A reference time estimation unit that estimates a reference time of the one-dimensional signal by calculating a relational number, a vibration information specification unit that specifies vibration information of the beat signal using the reference time, and the specified vibration A sleep depth estimation device is provided, comprising: a sleep depth estimation unit that estimates the sleep depth of the living body based on information.

  The reference time estimation unit may estimate the reference time based on the phase of the sine wave when the correlation coefficient between the reference signal and the sine wave is maximum.

  The reference time estimation unit extracts a plurality of reference times estimated in the past in a section of the reference signal used when estimating the reference time, and the reference is estimated based on a distribution of the plurality of reference times extracted. The time may be specified.

  The vibration information identification unit may identify the scalar amount of the beat signal represented by the two-dimensional vector at the reference time as the amplitude of the beat signal.

  The vibration information identification unit may identify the amplitude of the one-dimensional signal at the reference time as the amplitude of the beat signal.

  The vibration information identification unit may identify an interval between two consecutive reference times as a cycle of the beat signal.

  The sleep depth estimation unit may estimate the sleep depth of the living body based on vibration information of body movement of the living body.

  The sleep depth estimation unit may change the estimation condition of the sleep depth of the living body according to the duration of the sleep of the living body.

  The sleep depth estimation apparatus may further include a filter unit that performs a filtering process to reduce components of frequencies higher than a cutoff frequency from the beat signal.

  The cut-off frequency of the filter unit may be 1.5 Hz or more.

  The vibration of the living body may include vibration due to breathing of the living body.

  In order to solve the above problem, according to another aspect of the present invention, there is a step of converting a beat signal detected by a Doppler sensor into a one-dimensional signal according to the vibration of a living body, and a frequency of the one-dimensional signal. Estimating a phase of a reference signal having a section length corresponding to a period corresponding to the frequency of the estimated one-dimensional signal from the one-dimensional signal, and phase relationship between the reference signal and a sine wave having the frequency Calculating a number to estimate a reference time of the reference signal; identifying vibration information including an amplitude and a period of the beat signal using the reference time; and identifying the vibration information specified. And estimating the sleep depth of the living body.

  Further, in order to solve the above problems, according to another aspect of the present invention, there is provided a computer comprising: a signal conversion unit for converting a beat signal detected by a Doppler sensor into a one-dimensional signal according to a vibration of a living body; A frequency estimation unit for estimating a frequency of a one-dimensional signal, and a reference signal having a section length corresponding to a period corresponding to the estimated frequency of the one-dimensional signal are cut out from the one-dimensional signal, and the reference signal and the frequency are A reference time estimation unit that estimates a reference time of the reference signal by calculating a correlation coefficient with the sine wave, and a vibration that specifies vibration information including the amplitude and period of the beat signal using the reference time. A program is provided to function as an information identification unit and a sleep depth estimation unit that estimates the sleep depth of the living body based on the identified vibration information.

  As described above, according to the present invention, it is possible to estimate the sleep depth of the subject with higher accuracy.

It is a figure showing an outline of a sleep depth estimating system concerning one embodiment of the present invention. It is a block diagram showing an example of composition of a sleep depth estimating device concerning one embodiment of the present invention. It is a figure which shows the example which compares the waveform of the one-dimensional signal converted from the beat signal filtered by the filter in which a pass band differs. It is a figure which shows an example of the locus | trajectory in IQ plane of a beat signal. It is a figure which shows the example which implements the conversion process by a 2nd conversion part about the locus | trajectory in IQ plane of a beat signal. It is a figure which shows an example of the estimation method of the reference | standard time by the reference | standard time estimation part which concerns on one Embodiment of this invention. It is a figure which shows an example of the identification method of the reference | standard time by the reference | standard time estimation part which concerns on one Embodiment of this invention. It is a figure which shows an example of the flowchart which shows the determination flow of the sleep depth by the depth determination part which concerns on one Embodiment of this invention. It is a figure which shows an example of the graph which output the presumed result of the sleep depth. It is a flowchart which shows the operation example of the sleep depth estimation apparatus which concerns on one Embodiment of this invention. It is a block diagram showing an example of hardware constitutions of a sleep depth estimating device concerning an embodiment of the present invention.

  The present invention will now be described more fully with reference to the accompanying drawings, in which exemplary embodiments of the invention are shown. In the present specification and the drawings, components having substantially the same functional configuration will be assigned the same reference numerals and redundant description will be omitted.

<1. Example of configuration of sleep depth estimation system>
FIG. 1 is a diagram showing an overview of a sleep depth estimation system 1 according to an embodiment of the present invention. Referring to FIG. 1, the sleep depth estimation system 1 includes a Doppler sensor 2 and a sleep depth estimation device 10.

  As shown in FIG. 1, the Doppler sensor 2 is installed, for example, on the ceiling of a room and emits radiation waves such as light, electromagnetic waves or sound waves toward the indoors of the room which is a detection area. In 1, the reflected wave reflected by the person P) is received. At this time, the frequency of the reflected wave changes from the frequency of the radiation wave by the Doppler effect caused by the movement of the object such as the vibration. The Doppler sensor 2 generates a beat signal having a frequency that is the difference between the frequency of the radiation wave and the frequency of the reflected wave. The Doppler sensor 2 may be equipped with a quadrature detection method, in which case the Doppler sensor 2 generates two types of beat signals, a cosine wave component (I component) and a sine wave component (Q component). The Doppler sensor 2 outputs the generated beat signal to the sleep depth estimation device 10.

  The Doppler sensor 2 can be installed at any position as long as the vibration of the object can be detected. Further, in the example shown in FIG. 1, the Doppler sensor 2 is configured such that the transmitter of the radiation wave and the receiver of the reflected wave are integrated, but the Doppler sensor 2 separates the transmitter and the receiver. It may be realized by the above configuration. Moreover, the radiation wave which the Doppler sensor 2 radiates may be a wave of any frequency band as long as it is possible to cause the Doppler effect by the vibration of the object. For example, when the vibration of respiration is detected by a Doppler sensor, a wave having a wavelength larger than that of displacement of the vibration of respiration, such as the 2.4 GHz to 10.5 GHz band, is often used. However, in the present embodiment, a quasi-millimeter wave such as a 24 GHz band or a millimeter wave band is used to detect a wide range of displacement of vibration such as respiration and to detect minute fluctuations included in the waveform. Is preferred.

  The sleep depth estimation device 10 is a device that identifies vibration information of the subject (for example, the person P) from the beat signal output from the Doppler sensor 2 and estimates the sleep depth of the subject based on the identified vibration information. The sleep depth estimation device 10 may be realized by, for example, one or more information processing devices on a network. More specifically, the sleep depth estimation device 10 may be realized by a server, a PC (Personal Computer), or the like. Here, in the present specification, vibration information means, for example, an amplitude or a cycle obtained from a vibration waveform of an object. For example, the sleep depth estimation apparatus 10 identifies vibration information of the person P from beat signals obtained from vibrations of the person P, such as breathing and body movement, and estimates the sleep depth of the person P using the identified vibration information. It is possible. For example, the sleep depth estimation device 10 can transmit the estimated sleep depth to an external device or the like by wire or wirelessly via a communication unit (not shown). Thereby, for example, in the external device, it is possible to analyze the sleep depth. By analysis of the sleep depth, for example, it becomes possible to grasp the sleep state of the person P, stress during sleep, and the like. In addition, about the analysis using a sleep depth, it is also possible to carry out in the sleep depth estimation apparatus 10 main body instead of the above external devices.

  Here, the beat signal obtained from the vibration of the person P may include signals resulting from various movements of the person P. For example, not only the vibration of the respiration of the person P, but also the vibration of the motion or heart rate of the person P may be included in the beat signal. Furthermore, the beat signal generated by the Doppler sensor 2 may include noise due to the surrounding environment. Therefore, for example, when the vibration of respiration is to be detected, it is desirable to detect only the waveform information of respiration from the beat signal.

  For example, in Japanese Patent Application Laid-Open No. 2006-263032, in order to detect the period of respiration, the vibration signal is subjected to filtering with a band pass filter of 0.15 to 0.30 Hz corresponding to the frequency band of respiration. It is done. However, in this case, when fine fluctuation occurs in one cycle of respiration, the fluctuation is reduced by the band pass filter, and it is difficult to extract a waveform that represents the vibration accompanied by the fine fluctuation. Become. Therefore, it is difficult to detect a change in sleep state including a disorder in the breathing cycle.

  Further, in Japanese Patent Application Laid-Open No. 2014-014708, the cycle of breathing is calculated by calculating the norm of the beat signal on the IQ plane and detecting the time when the norm becomes substantially zero as switching between suction and discharge of breath. It is estimated. However, for example, in the case where many noises are superimposed on the beat signal, or when the breathing operation is continuous and continuous, it is difficult to detect the time when the norm becomes substantially zero. Therefore, there is a possibility that the cycle of breathing may be incorrectly estimated. Accordingly, the change in sleep depth of the subject may be erroneously estimated.

  Therefore, based on the above-mentioned circumstances, the sleep depth estimation device 10 has been created. The sleep depth estimation apparatus 10 according to an embodiment of the present invention can estimate the vibration state of the subject's respiration accompanied by fine fluctuations with high accuracy, and thus can further enhance the sleep depth estimation accuracy. Hereinafter, the configuration of the sleep depth estimation device 10 according to an embodiment will be described in more detail.

<2. Configuration Example of Sleep Depth Estimation Device>
FIG. 2 is a block diagram showing a configuration example of the sleep depth estimation device 10 according to an embodiment of the present invention. Referring to FIG. 2, the sleep depth estimation apparatus 10 includes a respiration estimation unit 100, a body movement estimation unit 200, and a sleep depth estimation unit 300. The illustrated sleep depth estimation apparatus 10 estimates the vibration information of the subject's respiration and the vibration of the subject's body motion with minute fluctuations, and estimates the sleep depth of the subject based on the estimated vibration information. be able to. The functions of each configuration will be described below.

[2-1. Breathing estimation unit]
Breath estimation unit 100 includes beat signal acquisition unit 110, filter unit 120, signal conversion unit 130, frequency estimation unit 140, reference time estimation unit 150, and vibration information identification unit 160. The respiration estimation unit 100 converts the acquired beat signal into a one-dimensional signal, estimates the frequency of the converted one-dimensional signal, and calculates a reference position in the one-dimensional signal for calculating the period of the beat signal from the estimated frequency. presume. Then, the respiration estimation unit 100 specifies the oscillation information including the period or amplitude of the respiration vibration using the estimated reference position. The identified vibration information is output to the sleep depth estimation unit 300. The configuration of each function included in the respiration estimation unit 100 will be described below.

(Beat signal acquisition unit)
The beat signal acquisition unit 110 acquires a beat signal D (t) output from the Doppler sensor 2. The beat signal D (t) output from the Doppler sensor 2 has two components of I component and Q component, and the amplitude is A (t), the wavelength is λ, the Doppler sensor 2 at time t and the target object (for example, Assuming that the distance to the person P) shown in FIG. 1 is d (t), the initial phase is φ ₀ , the direct current component is O, and the noise component is w, the beat signal D (t) is Expressed in

  The beat signal acquisition unit 110 outputs the acquired beat signal D (t) to the filter unit 120. Further, the beat signal acquisition unit 110 may store the acquired beat signal D (t) in a storage unit (not shown) in time series. The fundamental frequency of breathing during sleep is said to be about 0.15 to 0.30 Hz. However, the signal representing the vibration of actual respiration is not a sine wave, but has a complex waveform. Therefore, it is preferable that the sampling frequency is at least a frequency component of 1.5 Hz or more, which is at least five times greater than 0.30 Hz, which is the fundamental frequency of respiration, in order to accurately extract the respiration cycle including minute fluctuations. . More specifically, the sampling frequency is more preferably 3 Hz or more. By the way, in order to suppress aliasing, it is necessary to apply a low pass filter for obtaining frequency components below the Nyquist frequency before sampling of the signal. This is because if the components above the Nyquist frequency are not removed to some extent from the signal, the distortion of the waveform makes it difficult to estimate properly. In any case, the sampling frequency needs to be 1.5 Hz or more in order to extract the vibration waveform of respiration. And in order to extract a breathing cycle more correctly, it is preferable to set sampling frequency to about 3 Hz-20 Hz. If the frequency is higher than this, the accuracy of estimation of the respiration cycle may be improved, but the noise level superimposed on the signal increases as the band spreads. Therefore, even if the sampling frequency is set higher than 20 Hz, the demerit due to the increase in the processing amount is considered to be greater than the improvement of the estimation accuracy of the respiration cycle. However, if frequency components that are effective for estimation of the respiratory cycle can be sampled, setting the sampling frequency high has no influence on the estimation accuracy of the cycle other than the increase in processing amount. In the present embodiment, the sampling frequency is 20 Hz.

(Filter part)
The filter unit 120 performs a filtering process to reduce or remove a low frequency component such as a direct current component O included in the beat signal D (t) acquired by the beat signal acquisition unit 110 and a high frequency component such as a noise component w. The beat signal D (t) subjected to the filter processing is output to the signal conversion unit 130. For example, the filter unit 120 may include a high pass filter that reduces or removes the DC component O included in the beat signal D (t). When the amplitude of the beat signal D (t) is weak, if the beat signal D (t) is amplified by an amplifier (not shown) while the direct current component O is included, the direct current component O is also amplified. Then, it is difficult to extract only the waveform of a signal indicating a feature contributing to the vibration of respiration among the beat signals D (t). Therefore, by reducing or removing the DC component O by the high-pass filter, it is possible to appropriately amplify the waveform of the beat signal D (t) that indicates the feature contributing to the vibration of respiration. The high pass filter as described above may be realized by, for example, an analog filter such as a filter circuit provided in the sleep depth estimation device 10 or may be realized by a digital filter such as an IIR (Infinite Impulse Response) filter. It is also good. Further, the filter unit 120 may reduce or remove the DC component O by a filter circuit that arranges a subtraction circuit so as to identify and remove the DC component O.

  The filter unit 120 may also include a low pass filter that reduces or removes the noise component w included in the beat signal D (t). The low pass filter may be, for example, a low pass filter having a cutoff frequency of 1.5 Hz or more. In general, the fundamental frequency of breathing during sleep is said to be about 0.15 to 0.30 Hz, as described above. However, the subject's breathing fluctuates depending on the subject's sleep state. For example, when the subject's sleep is shallow or when the subject's breathing is unstable due to fever etc., fluctuations in breathing may occur. Vibrations due to respiratory fluctuations have frequency components higher than the fundamental frequency of respiration. In order to extract such fluctuation of respiration accurately, it is preferable to set the cutoff frequency of the low pass filter to a value (for example, 5 to 10 times or more) higher than the fundamental frequency of respiration.

  FIG. 3 is a diagram showing an example of comparing waveforms of one-dimensional signals r (t) converted from beat signals D (t) filtered by filters having different passbands. The waveform W0 is a waveform of a one-dimensional signal r (t) converted from the unfiltered beat signal D (t). A waveform W1 is a waveform of a one-dimensional signal r (t) converted from a beat signal D (t) filtered by a band pass filter whose passband is 0.15 to 0.30 Hz. The waveform W2 is a waveform of a one-dimensional signal r (t) converted from the beat signal D (t) that has been filtered by a low pass filter having a passband of 3.0 Hz or less. The phases of the signals shown by the waveforms W0 to W2 are the same.

  When three waveforms in the region R1 are compared, the waveform W2 shows a shape substantially the same as the waveform W0, while the waveform W1 shows a shape different from the waveform W0. Further, with respect to the position of time T1 indicating the lower limit peak position of waveform W0, the time indicating the lower limit peak position of waveform W2 is substantially the same as time T1, while the time indicating the lower limit peak position of waveform W1 is It can be seen that there is a divergence from time T1. That is, the waveform W1 of the one-dimensional signal r (t) converted from the beat signal D (t) filtered by the band pass filter having a pass band of 0.15 to 0.30 Hz becomes a simple sine wave. It can be said that the waveform is close and does not accurately represent the vibration of actual breathing. In this case, for example, it is difficult to estimate, using the waveform W1, the REM sleep state in which disturbance in the vibration of breathing may occur. On the other hand, the waveform W2 of the one-dimensional signal r (t) converted from the beat signal D (t) filtered by the low pass filter having a passband of 3.0 Hz or less is a respiration in which noise is removed from the waveform W0. Shows the waveform of the vibration. Therefore, the low-pass filter having a cutoff frequency of 3 Hz, which is a value of 1.5 Hz or more described above, can extract the vibration of respiration including a minute fluctuation component while removing noise. However, since the waveform W2 extracted from the low pass filter as described above is not a simple sine waveform, it is difficult to estimate the period of the respiratory vibration by the method disclosed in, for example, Japanese Patent Laid-Open No. 2006-263032. is there. Therefore, in the present invention, the period of the vibration of the respiration including the component of the fine fluctuation is estimated by the reference time estimation unit 150 or the like described later.

  In addition, as above-mentioned, 1.5 Hz or more is preferable and, as for the cutoff frequency of the above low pass filters which implement | achieve the filter part 120, it is preferable to set it as 3 Hz more specifically, The cut-off frequency may exceed 3 Hz if it is desired to extract a vibration including such fluctuation. For example, the above-mentioned cutoff frequency may be set between 10 Hz and 20 Hz. Thus, it is possible to more accurately estimate the period of a signal that indicates a vibration including a minute fluctuation included in respiration. However, if the cut-off frequency exceeds 20 Hz, the vibration of the respiration may contain not only fine fluctuations but also many noises. Therefore, it is preferable that the cutoff frequency of the low pass filter which realizes the filter unit 120 be 20 Hz or less.

(Signal converter)
The signal conversion unit 130 converts the beat signal D (t) output from the filter unit 120 into a one-dimensional signal r (t). For example, the signal conversion unit 130 converts the beat signal D (t) into a one-dimensional first candidate signal R ₁ (t) based on the distribution of the beat signal D (t) in the IQ plane. You may have. Further, the signal conversion unit 130 converts the beat signal D (t) into a one-dimensional second candidate signal R ₂ (t) based on a temporal change in position of the beat signal D (t) on the IQ plane, It may have a converter. Furthermore, when the signal conversion unit 130 has the above-described two conversion units, the signal conversion unit 130 uses one or more of the first candidate signal R ₁ (t) and the second candidate signal R ₂ (t) to generate a one-dimensional signal r ( It may have a signal determination unit that determines t). Hereinafter, each functional unit described above will be described.

First Converter The first converter calculates, for example, the inner product of a two-dimensional vector representing the beat signal D (t) and an eigenvector corresponding to the maximum eigenvalue of the covariance matrix of the beat signal D (t). Thus, the beat signal D (t) may be converted into the first candidate signal R ₁ (t). That is, the first conversion unit can obtain the first candidate signal R ₁ (t) obtained by projecting the beat signal D (t) in the principal component direction corresponding to the maximum eigenvalue. Specifically, the first candidate signal R ₁ (t) may be calculated by Equation 2 below.

  Here, b (t) is a two-dimensional vector representing the I and Q components of the beat signal D (t), and p (t) is an eigenvector corresponding to the maximum eigenvalue of the covariance matrix of the beat signal D (t) It is. The origin of the two-dimensional vector representing the beat signal D (t) may be estimated based on, for example, the distribution of the beat signal D (t). For example, the midpoint between the maximum position and the minimum position from the origin of the IQ plane of the trajectory of the beat signal D (t) in the past several seconds to several tens of seconds may be the origin. Further, the locus of the beat signal D (t) may be approximated as a circle or an ellipse by the least squares method or the like, and the central point of the approximated figure may be the origin. Further, since the beat signal D (t) processed by the high pass filter in the filter unit 120 is close to the origin of the IQ plane, the origin of the IQ plane may be the origin of a two-dimensional vector.

FIG. 4 is a view showing an example of a locus on the IQ plane of the beat signal D (t). Referring to the graph G1 shown in FIG. 4, a locus Tr1 of the beat signal D (t) is shown on the IQ plane. For example, when the displacement of the vibration of respiration is relatively small, such as about 1/10 of the wavelength of the radiation wave emitted from the Doppler sensor, the beat signal D (t) has a flat eight-character shape such as a locus Tr1. It is often a distribution. In such a case, for example, when the principal component analysis is performed on the trajectory Tr1, the maximum unique value corresponding to the first principal component indicated by the axis Pc1 indicates a large value as compared with other principal components. The first conversion unit can obtain the first candidate signal R ₁ (t) by projecting the beat signal D (t) onto the first principal component.

Note that the first conversion unit according to the present embodiment converts the beat signal D (t) into the first candidate signal R ₁ (t) using the eigenvalues and the eigenvectors obtained by performing principal component analysis or the like on the trajectory Tr1. Do. However, for example, with regard to the distribution of a flat eight-character like the locus Tr1, if the axial direction of the flat direction can be extracted and the movement of the signal in the axial direction can be extracted, the first conversion unit The beat signal D (t) may be converted into the first candidate signal R ₁ (t) using a method other than principal component analysis.

-Second conversion unit The second conversion unit is, for example, the distance (amplitude equivalent to the distance to the position of the beat signal D (t) from the center of the distribution estimated based on the distribution of the beat signal D (t) in the IQ plane. By converting the beat signal D (t) into a second candidate signal R ₂ (t) by calculating a product of the rotation angle of the beat signal D (t) with respect to the center of the distribution. May be For example, R ₂ (t) may be calculated using a function using a continuous function (in the present embodiment, a sigmoid function) in Equation 3 below.

  Here, Amp (t) is the amplitude of the beat signal D (t), θ (t) is the angle on the IQ plane of the beat signal D (t), and θ ′ (t) is the angle θ (t) It is an amount of change in angle of unit time, or a time derivative value of angle θ (t). Also, a, b and c are all constants and can be freely set.

  FIG. 5 is a diagram showing an example in which the conversion process by the second conversion unit is performed on the trajectory of the beat signal D (t) on the IQ plane. Referring to the graph G2 shown in FIG. 5, the amplitude Amp (t) and the angle θ (t) are shown for the beat signal D (t) on the locus Tr2. The product of the amplitude Amp (t) and the change amount θ ′ (t) of the angle θ (t) corresponds to the area velocity of the region Av1 shown in FIG. This area velocity corresponds to the size of the respiratory vibration per unit time. In addition, since the change amount θ ′ (t) takes a positive or negative value depending on the direction of breathing vibration (breathing and exhaling), the direction of breathing vibration can be expressed. Therefore, the product of the amplitude Amp (t) and the change amount θ ′ (t) makes it possible to accurately express the state of respiratory movement.

  The amplitude Amp (t) corresponds to the distance from the center of the distribution estimated based on the distribution in the IQ plane of the beat signal D (t) to the position of the beat signal D (t), and the angle θ (T) corresponds to the rotation angle of the beat signal D (t) with respect to the center of the distribution. Here, the center of the distribution indicates, for example, the center point C1 shown in FIG. Here, the center of the distribution of the beat signal D (t) may be estimated based on the distribution of the beat signal D (t). For example, in the distribution of the beat signal D (t) in the past several seconds to several tens of seconds, an intermediate position between the maximum position and the minimum position from the origin of the IQ plane or a position corresponding to the center of the distribution is the beat signal D (t It may be the center of the distribution of Further, the locus of the beat signal D (t) may be approximated as a circle or an ellipse by the least squares method or the like, and the central point of the approximated figure may be the center of the distribution. Further, since the beat signal D (t) processed by the high pass filter in the filter unit 120 is distributed so as to be close to the origin of the IQ plane, the origin of the IQ plane is the center of the distribution of the beat signal D (t) It is also good.

-Signal Determination Unit For example, the signal determination unit may determine one of the first candidate signal R ₁ (t) and the second candidate signal R ₂ (t) as the one-dimensional signal r (t). Further, the signal determination unit may determine a signal obtained by integrating the first candidate signal R ₁ (t) and the second candidate signal R ₂ (t) as a one-dimensional signal r (t).

In the locus Tr1 of the beat signal D (t) in the case where the displacement of the vibration of the respiration shown in FIG. 4 is small, the variation θ ′ (t) of the angle θ (t) from the center of the distribution is shown in FIG. As compared with the case of the trajectory Tr2 of the beat signal D (t), the accuracy is lowered due to the influence of noise and quantization error. That is, when the vibration of respiration is small, the first candidate signal R ₁ (t) can more accurately represent the waveform of the vibration of respiration than the second candidate signal R ₂ (t). On the other hand, in the locus Tr2 of the beat signal D (t) in the case where the displacement of the vibration of breathing shown in FIG. 5 is large, the second candidate signal R ₂ (t) is more than the first candidate signal R ₁ (t). Can accurately represent the waveform of the respiratory vibration. This means that although the wave of the first candidate signal R ₁ (t) projected in the principal component direction includes vibration of a cycle (high frequency) shorter than the respiration cycle, the second candidate signal R ₂ (t) is This is because it is obtained by tracking the sequential change of the beat signal D (t), and thus does not include high frequency components caused by the distribution of the beat signal D (t). Therefore, when the displacement of the vibration of respiration is large, the second candidate signal R ₂ (t) can more accurately represent the waveform of the vibration of respiration than the first candidate signal R ₁ (t).

The signal determination unit may compare each of the first candidate signal R ₁ (t) and the second candidate signal R ₂ (t), and determine the one-dimensional signal r (t) to be output based on the comparison result. . The signal determination unit may calculate, for example, the evaluation value s as a parameter for determining which candidate signal to use preferentially. The evaluation value s may be determined from, for example, the comparison result of the frequency of each candidate signal or an index value such as the deviation of the change amount of the angle of the beat signal D (t). For example, the signal determination unit may perform multidimensional mapping on the index values and the like as described above, and calculate the evaluation value s based on the obtained distribution. Further, the signal determination unit may obtain an output result obtained by inputting the index value as described above as the evaluation value s using a data processing method such as machine learning.

Also, for example, when one candidate signal is selected one time ago as a one-dimensional signal r (t) and then switched to the other candidate signal, or when combining both candidate signals, both are selected. It is necessary to match the phase of the candidate signal of. For example, the phase phi ₂ and the phase phi ₁ of the first candidate signal _R 1 (t) a second candidate signal _R 2 (t) is not always necessarily coincide. Therefore, the signal determination unit aligns the phase of one candidate signal with the phase of the other candidate signal. Thereby, in the switching process of the signal of the one-dimensional signal r (t) or the integration process of both candidate signals, it is possible to prevent the disturbance of the waveform due to the mismatch of the phase of each candidate signal. In the present embodiment, processing is performed to match the phase of the first candidate signal R ₁ (t) with the phase of the second candidate signal R ₂ (t). For example, the phase matching process is performed by converting one candidate signal into an analysis signal by Hilbert transform and calculating the phase difference between the two signals obtained by integrating the analysis signal and the other candidate signal. Is possible.

Furthermore, the signal determination unit may perform a process of integrating both candidate signals when the above-described phase matching process is performed. For example, the signal determination unit generates a one-dimensional signal as the weighted sum of the first candidate signal R _1a (t) and the second candidate signal R ₂ (t) calculated using the evaluation value s as described above as a weight. It can be determined as r (t).

  As described above, the signal conversion unit 130 can select an appropriate conversion method regardless of the magnitude of the displacement of the respiration vibration. The one-dimensional signal r (t) converted from the beat signal D (t) by the signal conversion unit 130 is output to the frequency estimation unit 140.

(Frequency estimation unit)
The frequency estimation unit 140 estimates the frequency of the one-dimensional signal r (t) output from the signal conversion unit 130. For example, in order to adjust to the large fluctuation of the breathing cycle, the frequency estimation unit 140 generates the one-dimensional signal r (t) based on the phase difference between the one-dimensional signal r (t) and the reference signal and the time change of the phase difference. You may have a 1st frequency estimation part which estimates a frequency. Further, the frequency estimation unit 140 cuts out a comparison signal having a section length corresponding to the period corresponding to the estimated frequency estimated by the first frequency estimation section from the one-dimensional signal r (t), and compares the comparison signal with the above section length. The second frequency estimation unit may estimate a frequency of the one-dimensional signal r (t) by calculating a correlation coefficient with the one-dimensional signal r (t). Hereinafter, the frequency of the one-dimensional signal r (t) estimated by the first frequency estimation unit is referred to as a first estimated frequency, and the frequency of the one-dimensional signal r (t) estimated by the second frequency estimation unit is referred to as a second It is called an estimated frequency.

First Frequency Estimating Unit The first frequency estimating unit integrates, for example, the one-dimensional signal r (t _k ) at time t _k and the reference signal, and the low frequency component of the integrated signal is a low-pass filter etc. The phase difference ψ (t _k ) between the one-dimensional signal r (t _k ) and the reference signal is calculated. Here, as a reference signal, a sine wave or cosine wave having a phase and its time change as a parameter is used to facilitate estimation of the period based on the phase change. In that case, the reference signal is a sine wave or cosine wave signal having a frequency f _A (t _{k -1} ) estimated at time t _{k -1} . Then, the first frequency estimation unit calculates the phase difference ψ (t _k), the time t _k-1 phase difference calculated in [psi the change in the (t _k-1). Time variation of the phase difference _{ψ (t k) -ψ (t} k-1) , assuming that produced by a change in the frequency of the one-dimensional signal r (t), the first frequency estimation unit, the reference signal the frequency _{f a (t k-1)} , by adding a frequency corresponding to the time variation of the phase difference _{ψ (t k) -ψ (t} k-1), the first estimated frequency _f a (t _k) Output

In addition, when the time change of the phase difference is equal to or more than a predetermined threshold value, the feedback value of the frequency is easily estimated, so that the first estimated frequency f _A (t _k ) does not converge with vibration. There is. Therefore, the value of the frequency to be added to the frequency f _A (t _k-1 ) of the reference signal is less than 1 at the frequency corresponding to the time variation of phase difference ψ (t _k ) −ψ (t _k−1 ) It may be a value obtained by integrating coefficients. This allows f _A (t _k ) to converge in a relatively short time.

Further, when the calculated first estimated frequency f _A (t _k ) does not exist in a predetermined frequency band, the frequency f _A (t _{k -1} ) of the reference signal is set as the first estimated frequency f _A (t _k ). You may output it. Here, for example, in the present embodiment, the predetermined frequency band means a frequency band corresponding to the cycle of respiratory motion.

-Second frequency estimation unit The second frequency estimation unit performs the correlation analysis of the input one-dimensional signal r (t) to obtain the first estimated frequency f _A obtained by the first frequency estimation unit. t _k) to estimate a highly accurate second estimated frequency f _{B (t k)} to local changes more. First, the second frequency estimation unit, from a one-dimensional signal r (t), cut out a section from time t _k -T _A until time t _k, handled as a comparison signal this. T _A is a period corresponding to the first frequency f _A (t _k-1 ) estimated above. Next, the second frequency estimation unit calculates a correlation coefficient using the cut out comparison signal. The calculation range of the correlation coefficient preferably includes the range of t _k -2T _A to t _k -T _A. The second frequency estimator, in calculating the above range, the one-dimensional signal r (t) having a period length T _A, the correlation coefficient and the comparison signal R (tau) is calculated time lag tau _max which maximizes . At this time, the second frequency estimation unit may store the correlation coefficient R (τ _max ) when τ = τ _max in a storage unit (not shown). The second frequency estimation unit calculates a second estimated frequency f _B (t _k ) from the time lag τ _max when the correlation coefficient calculated in the correlation coefficient calculation range is maximum. For example, the second estimated frequency f _B (t _k ) may be the inverse of the time lag τ _max .

-Determination part The frequency estimation part 140 may have a determination part which determines which frequency is used about two estimated frequency mentioned above. For example, the determination unit may determine which frequency to use as the estimated frequency f _E (t) based on whether the difference between the two estimated frequencies is within a predetermined threshold. Further, based on the value of the time lag τ _max calculated by the second frequency estimation unit, it is determined which one of the first estimated frequency f _A (t _k ) and the second estimated frequency f _B (t _k ) to use You may

As mentioned above, although the frequency estimation part 140 is difficult to follow irregular fluctuation | variation with the 1st frequency estimation part which can estimate an approximate frequency, it is difficult to detect a local change of a waveform. By determining which of the second frequency estimation unit to be estimated is to be used by the determination unit, it is possible to estimate the frequency of the waveform including many fine fluctuations with high accuracy for each breath. The estimated frequency f _E (t) estimated by the frequency estimation unit 140 is output to the reference time estimation unit 150.

(Reference time estimation unit)
The reference time estimation unit 150 estimates the reference time of the one-dimensional signal r (t) based on the estimated frequency f _E (t), and determines the estimated reference time as the beat signal D (t) or the one-dimensional signal r (t). And the vibration information identification unit 160. Specifically, the reference time estimation unit 150 extracts a reference signal having a section length corresponding to a period corresponding to the estimated frequency f _E (t) from the one-dimensional signal r (t), and generates the reference signal and the estimated frequency f _E The reference time of the one-dimensional signal r (t) is estimated by calculating the correlation coefficient with the sine wave having (t). Here, in the present specification, the reference time has a common feature such as the time at which the peak of the waveform is indicated among the similar waveforms of the waveforms for one period of the one-dimensional signal r (t). Means the time of day Further, the reference time may be, for example, a time corresponding to a position indicating the minimum value of the waveform, or may be a time when r (t) = 0.

FIG. 6 is a diagram showing an example of a method of estimating a reference time by the reference time estimation unit 150 according to an embodiment of the present invention. A one-dimensional signal OS1 and a sine wave RS1 are shown in a graph G3 shown in FIG. Here, the reference time estimating unit 150 estimates the reference time RT1 one-dimensional signal r _{(t k)} at time _{t k.} First, the reference time estimation unit 150 cuts out from the one-dimensional signal OS1 a reference signal BS1 (broken line portion) having a section length corresponding to the period T _E (t _k ) corresponding to the estimated frequency f _E (t _k ). At this time, the end of the section is t _k . Next, the reference time estimation unit 150 calculates the correlation coefficient between the sine wave RS1 having the estimated frequency f _E (t _k ) and the reference signal BS1 in the above section. For example, the reference time estimation unit 150 sets an arbitrary initial phase of the sine wave RS1 at time t _k in the above section as φ ₀ (t _k ), changes the phase at time t _k , and generates the sine wave RS1 and the reference signal BS1. The phase of the sine wave RS1 at which the correlation coefficient is maximized is identified. The phase of the sine wave RS1 at this time can be easily identified, for example, by converting the sine wave RS1 into an analysis signal. _Assuming that the phase specified here is φ _Rmax (t _k ), the reference time RT1 is t _k −T _E (φ (t _k ) / 2π). However, it is _{φ (t k) = φ Rmax} (t k) -φ 0 (t k). For example, in the case of the initial phase φ ₀ (t _k ) = 0, as shown in the graph G3, the time indicating the phase corresponding to the peak of the sine wave RS1 is the reference time RT1. Thereby, the reference time RT1 of the one-dimensional signal r (t) can be obtained. The reference time estimation unit 150 may store the estimated reference time RT1 in a storage unit (not shown).

Further, the reference time estimating unit 150 extracts a plurality of reference time estimated in the past in the section T _E of the reference signal as described above was used in estimating the reference time, the extracted plurality of reference time The reference time may be specified based on the distribution. For example, the reference time estimation unit 150, a reference time that is estimated in the past within the scope of the section T _E to the time t _k embodying the estimation of the reference time and the end point is extracted from the storage unit (not shown), each reference time You may obtain the distribution of The distribution of the extracted reference times may be expressed in a histogram divided by unit intervals.

FIG. 7 is a diagram showing an example of a method of specifying a reference time by the reference time estimation unit 150 according to an embodiment of the present invention. The graph G4 shown in FIG. 7 shows a one-dimensional signal OS1. Here, the reference time RT3 estimated at the time t _k-1 is deviated from the reference time RT2 estimated at t _k , t _k-2 , and t _k-3 . That is, the reference time estimated by the reference time estimation unit 150 is not necessarily accurate, and it may be considered that an error is included. Therefore, the reference time estimation unit 150 may statistically specify the reference time from the distribution of the plurality of estimated reference times.

Specifically, first, the reference time estimation unit 150, the information of the estimated reference time in the interval T _E of the reference signal and ending the t _k, extracted from the storage unit (not shown). For example, the reference time estimation unit _150, t k time estimation process in the interval from -T _E until _{t k} is performed _{t k, t k-1,} t k-2, t k-3, in.. Information on each reference time is extracted from the storage unit, and a histogram H1 is created based on each extracted reference time. Then, the reference time estimation unit 150 may specify the reference time at time t _k based on the created histogram H1. For example, the reference time estimation unit 150 may specify a reference time indicating the mode M1 of the histogram H1 as a reference time at time t _k . In addition, the reference time estimation unit 150 calculates the average value of the reference times estimated in the section T _E , and obtains the result by multiplying the Gaussian function having an arbitrary dispersion value and the histogram H1 with the average value as the center. the reference time in the mode of the new histogram that is, may be specified as the reference time at time t _k. Further, the reference time estimation unit 150 may specify the reference time at the time t _k using a value representing the probability distribution such as an average value and an intermediate value without being limited to the mode value.

(Vibration information identification unit)
The vibration information identification unit 160 identifies vibration information of the beat signal D (t) (or one-dimensional signal r (t)) using the estimated reference time. Here, the vibration information in the present specification includes, for example, the period and the amplitude of the beat signal D (t).

  For example, the vibration information identification unit 160 identifies the interval between two consecutive reference times estimated by the reference time estimation unit 150 as the period of the beat signal D (t). Thereby, it is possible to specify the cycle of the vibration of respiration for each cycle.

  Also, the vibration information identification unit 160 may specify the amplitude of the signal representing the vibration of the respiration at the reference time estimated by the reference time estimation unit 150 as the amplitude of the beat signal D (t). For example, the vibration information identification unit 160 may identify the scalar amount of the beat signal D (t) represented by the two-dimensional vector at the reference time as the amplitude of the beat signal D (t). Here, the scalar quantity may be, for example, the value of the amplitude Amp (t) corresponding to the distance between the center C1 and the beat signal D (t) shown in FIG.

  The vibration information identification unit 160 may also identify the amplitude of the one-dimensional signal r (t) at the reference time as the amplitude of the beat signal D (t). For example, when the beat signal D (t) is not held in the processing after the signal conversion unit 130 and the information of the one-dimensional signal r (t) is held, the vibration information identification unit 160 generates the one-dimensional signal r (t The amplitude of the beat signal D (t) may be specified from the amplitude of. At this time, the vibration information identification unit 160 converts the one-dimensional signal r (t) into an analysis signal by Hilbert transform, and multiplies the amplitude of the beat signal D (t) by the product of the analysis signal and the conjugate signal of the analysis signal. It may be specified. Thereby, it is possible to specify vibration information related to the amplitude from the one-dimensional signal r (t). Also, the vibration information identification unit 160 may logarithmically convert the identified amplitude. Thereby, it is possible to reduce the influence of the change of the amplitude (S / N ratio, quantization error, etc.) due to the change of the distance between the Doppler sensor and the subject.

  The respiration estimation unit 100 outputs the vibration information identified by the vibration information identification unit 160 to the sleep depth estimation unit 300. Here, since the identified vibration information is information that changes in each breathing cycle, for example, the vibration information is output to the sleep depth estimation unit 300 at a frequency lower than the sampling frequency (3 Hz or more) of the Doppler sensor 2 May be In the present embodiment, the respiration estimation unit 100 outputs vibration information to the sleep depth estimation unit 300 at a frequency of 1 Hz.

  As described above, the respiration estimation unit 100 can estimate the vibration information (period and amplitude) related to the vibration of the subject's respiration from the beat signal D (t) output from the Doppler sensor 2. In particular, the respiration estimation unit 100 can estimate the period of the respiration vibration more accurately by estimating the reference position of the waveform representing the respiration vibration including the minute fluctuation.

[2-2. Motion estimation unit]
The body movement estimation unit 200 estimates vibration information related to the body movement of the subject from the beat signal D (t) output from the Doppler sensor 2, and outputs the estimated vibration information to the sleep depth estimation unit 300. Here, the body movement means, for example, the subject's turning over or the movement of the limbs. Also, vibration information related to body movement includes, for example, an amplitude amount of body movement.

  The body movement estimation unit 200 first acquires the beat signal D (t) output from the Doppler sensor 2. The body movement estimation unit 200 may acquire the beat signal D (t) from the beat signal acquisition unit 110 which has been acquired by the respiration estimation unit 100.

  Next, the body motion estimation unit 200 performs a filtering process on the acquired beat signal D (t) with a high pass filter or a band pass filter. For example, when the Doppler sensor 2 detects a body movement such as turning over during sleep using a 24 GHz radiation wave, the body movement estimation unit 200 performs a body movement by using a band pass filter with a pass band of 5 to 20 Hz. Can be detected. In this case, the body movement estimation unit 200 may extract the norm of the beat signal D (t) that has been subjected to the filtering process by the band pass filter as the amplitude amount of the body movement.

  In addition, the body motion estimation unit 200 may perform subsampling processing lower than the upper limit cutoff frequency of the pass band of the band pass filter on the beat signal D (t) processed by the band pass filter. For example, the body motion estimation unit 200 may perform 5 Hz subsampling processing on the beat signal D (t) that has been subjected to the filter processing using a band pass filter whose upper limit cutoff frequency is 20 Hz. In this case, frequency components larger than the Nyquist frequency of 2.5 Hz are folded back as components of 2.5 Hz or less. However, since the body motion estimation unit 200 extracts only the amplitude of the beat signal D (t), it is not affected by the noise generated by the aliasing. Therefore, the body movement estimation unit 200 can reduce the load of the estimation process of body movement vibration information by performing the subsampling process using a lower subsampling frequency.

  The body movement estimating unit 200 may estimate the integrated value of the extracted amplitude per unit time as the amount of amplitude. Further, the body movement estimation unit 200 may estimate a value obtained by logarithmically converting the above integrated value as an amplitude amount. By performing the above-described logarithmic transformation, the body movement estimating unit 200 can emphasize a minute difference in body movement. Although the strength of the signal changes in inverse proportion to the square of the distance between the sensor and the subject, by performing logarithmic conversion, the strength of the signal changes linearly with the distance. Thereby, the body movement estimation unit 200 can reduce the influence of the distance between the sensor and the subject on the amplitude amount. For example, when the sleep depth estimation system 1 is applied to a house between 8 mats, the distance between the subject and the sensor is approximately 2.5 to 5 m, and therefore the rate of attenuation of the signal intensity is suppressed to about 50% by logarithmic conversion. It is possible.

  The vibration information including the amplitude amount estimated by the body movement estimation unit 200 is output to the sleep depth estimation unit 300. Here, although the estimated vibration information is a time-varying value, for example, the body movement estimating unit 200 may output the vibration information to the sleep depth estimating unit 300 at a frequency higher than the frequency of the vibration of respiration. . More specifically, the body movement estimation unit 200 may output vibration information to the sleep depth estimation unit 300 at a frequency of 1 Hz, which is the output frequency of the respiration estimation unit 100. At this time, the timing of the output by the body movement estimation unit 200 may be synchronized with the timing of the output by the respiration estimation unit 100. As a result, the sleep depth estimation unit 300 described later can estimate the sleep depth of the subject based on the respiration and body movement vibration information estimated at the same timing.

[2-3. Sleep depth estimation unit]
The sleep depth estimation unit 300 includes a feature quantity calculation unit 310, a model selection unit 320, and a depth determination unit 330. The sleep depth estimation unit 300 uses at least one of the vibration information of the subject's respiration acquired from the respiration estimation unit 100 or the vibration information of the body motion of the subject acquired from the body movement estimation unit 200 to obtain the sleep depth of the subject. presume. In this case, the sleep depth is estimated for each predetermined section, and the size of the predetermined section is not particularly limited, but in the present embodiment, Rechtschaffen, A. et al. and Kales, A. “R & K method, an international sleep depth criterion described in the Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stage of Human Subjects, Washington Public Health Service Government Printing Office, Washington, DC (1968) "For 30 seconds on". This 30-second interval is called an epoch.

(About sleep depth)
First, the sleep depth estimated by the sleep depth estimation unit 300 will be described. The depth of sleep means the depth of sleep classified according to the state of sleep of the living body. The sleep stages can be distinguished, for example, according to the R & K method described above, into awake, rem sleep, and non-rem sleep states. Furthermore, the state of non-REM sleep can be distinguished, for example, into the state of light sleep and deep sleep. In the present embodiment, the sleep depth is estimated in four stages, but in the other embodiments, the number of sleep depth stages is not particularly limited. For example, in another embodiment, the sleep depth may be two stages of awakening and sleep, or may be further subdivided non-REM sleep. Depending on the depth of sleep, the activity state of the subject's brain during sleep, the activity of the autonomic nerve, and the tonic state or relaxation state of the skeletal muscle change. These changes appear in features such as, for example, the magnitude of the subject's motion, the speed and fluctuation of the respiration cycle, or the fluctuation of the respiration amplitude. Therefore, the sleep depth estimation unit 300 can estimate the sleep depth of the subject by analyzing the above features. The configuration of each function included in the sleep depth estimation unit 300 will be described below.

(Feature amount calculation unit)
The feature amount calculation unit 310 calculates a feature amount used to estimate the sleep depth of the subject. Here, the feature amount may include, for example, the following feature amounts.

-Feature quantity bm related to body movement
The feature amount bm related to body movement is a feature amount indicating the size of body movement such as turning over of the subject. bm is used to determine the state of sleep or awakening and the state of REM sleep or non-REM sleep. For example, when the value of bm is low, the subject is considered to be in a sleep state because the motion is small. The feature amount calculation unit 310 calculates bm from the amplitude amount included in the vibration information of the body movement obtained one per second. Specifically, the feature amount calculation unit 310 may calculate an average value of 30 amplitude amounts included in one epoch as bm in the epoch.

-A feature cv related to the amplitude of respiration
The feature amount cv related to the amplitude of respiration is a feature amount indicating fluctuation of the size of suction and discharge in respiration. cv is used to determine the state of sleep or awakening and the state of light sleep or deep sleep. For example, when the value of cv is low, the subject is considered to be in a sleep state because breathing is stable. The feature amount calculation unit 310 calculates cv from the amplitude included in the vibration information of respiration acquired one per second. Specifically, the feature amount calculation unit 310 calculates an average value and a standard deviation of 30 amplitudes included in one epoch, and calculates a value obtained by dividing the standard deviation by the average value as cv. Good.

-A feature quantity resp related to the length of the breathing cycle
The feature quantity resp related to the length of the breathing cycle is a feature quantity that indicates the length of the breathing cycle. Resp is used to determine the state of light sleep or deep sleep. For example, when the value of resp is low, it is considered that the subject is in light sleep in which non-REM sleep is considered to be a shallow sleep because the breathing cycle is short. The feature amount calculation unit 310 calculates resp from the period included in the vibration information of the respiration obtained one per second. Specifically, the feature amount calculation unit 310 may calculate an average value of 30 cycles included in one epoch, and may calculate a value obtained by dividing the average value by the coefficient α as resp. Here, the coefficient α may be a value that changes according to the subject's breathing cycle, or may be a fixed value. For example, the coefficient α may be an average value of the respiration frequency during sleep of the subject. This makes it possible to suppress the variation in the value of resp due to individual differences. When the coefficient α is used as a fixed value, the coefficient α may be a value of about 0.2 to 0.5.

-Feature quantity resp_ave related to fluctuation of the respiration cycle
The feature amount resp_ave regarding the fluctuation of the respiration cycle is a feature amount indicating the magnitude of the fluctuation of the respiration cycle. resp_ave is used to determine the state of REM sleep or non-REM sleep. For example, when the value of resp_ave is low, the subject is considered to be in a non-REM sleep state because the fluctuation of the breathing cycle is small. The feature amount calculator 310 calculates the feature amount by comparing resp calculated in a certain epoch with resp calculated one epoch before. Specifically, first, the feature quantity calculation unit 310 calculates the square of the difference between resp calculated in a certain epoch and resp calculated one epoch before, and the above calculated in a predetermined number of consecutive epochs. The average value of the squares of the differences of may be calculated as resp_ave. The above-mentioned predetermined number may be, for example, 10 in order to capture fluctuations in the breathing cycle for several minutes in the past.

  The above four feature quantities are output to the depth determination unit 330. Note that, for example, although the average value is calculated using the arithmetic average for bm, cv, resp, and resp_ave, for example, the calculated average value may be a moving average or an exponential smoothing average. Also, instead of the average value, a median, a mode or the like may be used. Also, instead of the standard deviation used to calculate cv, the maximum and minimum differences between the amplitude values, the difference between the third quartile point of the amplitude values and the first quartile point may be used. In addition, resp may be a value obtained by the reciprocal of the average power frequency, which is a weighted average weighted with the power and amplitude of the frequency spectrum for a plurality of periods included in one epoch. In addition, resp_ave may be a value obtained from weighted dispersion in which the power and amplitude of the frequency spectrum are weighted for a plurality of periods included in one epoch.

  On the other hand, the two feature quantities described below are used to determine in the model selection unit 320 a model used to determine the sleep depth in the depth determination unit 330. Here, the state of REM sleep, the state of light sleep, and the state of deep sleep are collectively referred to as a sleep state.

-Sleep duration slp_et
The sleep continuation time slp_et is a feature value indicating the duration of the sleep state of the subject in a predetermined past period. slp_et refers to a time during which a sleep state continues, starting from a point estimated to be a sleep state from an awake state (or a state in which the depth of sleep of a subject is not estimated). The predetermined period may be determined based on the sleep state of the subject or may be a fixed value. For example, the predetermined period may be 4 hours = 14400 seconds. Here, for example, when the sleep state is interrupted and changed to the awake state, slp_et may be the time from the starting point to the time when it changes to the awake state. In addition, when the sleep state is changed again within the first predetermined time after the change to the awake state, the time during the awake state may be added to slp_et as the sleep continuation time. The first predetermined time is not particularly limited, but is preferably such a time that the wakefulness does not affect the sleep state immediately after that, and may be, for example, one hour. In addition, when the awake state is changed to the sleep state and then changed to the awake state again within the second predetermined time, the value of slp_et may be reset to 0. The second predetermined time is not particularly limited, but is preferably a time necessary to determine a temporary sleep state due to sleepiness, and may be, for example, 10 minutes. The first predetermined time and the second predetermined time can be freely changed according to the subject's sleep characteristics, the subject's health condition, and the like.

-Sleep depth integration time sum_rem, sum_light, sum_deep
The sleep depth integration times sum_rem, sum_light, and sum_deep (hereinafter collectively referred to as “sums” when collectively referring to the three sleep depth integration times) are the subject's rem sleep, light sleep, and deep sleep that occurred in a predetermined period in the past. It is a feature that indicates the integrated time of each sleep state. The value of each of sums may be determined based on each sleep state determined in 1 epoch unit. For example, if it is estimated that rem sleep is in an epoch, 30 seconds are added to sum_rem.

  The feature quantity calculation unit 310 updates slp_et and sums by receiving feedback of the sleep depth specification result from the depth determination unit 330. Then, the feature quantity calculation unit 310 outputs the updated slp_et or sums to the model selection unit 320. In addition, although it mentions in detail later, when learning about a model used in depth distinction part 330 as a step before specifying processing of sleep depth by depth distinction part 330, feature quantity calculation part 310 PSG (PolySomnoGraphy: Slp_et and sums may be updated based on the correct value of the sleep depth obtained by other sleep depth measurement methods such as sleep polygraph inspection). By outputting the updated slp_et or sums to the model selection unit 320, learning can also be performed on other models selected based on slp_et or sums. In addition, slp_et and sums may be updated by using in parallel two methods of feedback of the sleep depth identification result and the sleep depth correct value. In this case, even if the values of slp_et and sums updated by two methods deviate between the two methods, even if the values obtained by one method are reflected in the values obtained by the other method Good. In addition, when the values obtained between the two methods deviate, the difference between the two values may be reflected on the model selection unit 320. The model selection unit 320 can optimize the value of each parameter included in the discriminant model described later using the above difference.

  Although the units of slp_et and sums are not particularly limited, in the present embodiment, the units of slp_et and sums are seconds.

(Model selection section)
The model selection unit 320 changes the estimation condition of the sleep depth in accordance with the duration of sleep of the subject and the integration time in each sleep depth. For example, the model selection unit 320 may determine the discrimination model used in the process of identifying the sleep depth by the depth discrimination unit 330 based on at least one of slp_et or sums. The discriminant model has a discriminant matrix M and a correction vector b, which are used to obtain bm, cv, resp, and resp_ave as independent variables and obtain objective variables for estimating the sleep depth of each stage. For the parameters of the discriminant matrix M and the correction vector b, for example, a method such as supervised learning or semi-supervised learning that uses the correct value of the sleep depth obtained by another sleep depth measurement method such as PSG described above as teacher data May be determined. Also, the parameters of the discriminant matrix M and the correction vector b may be optimized by supervised learning after classification based on unsupervised learning is performed. The supervised learning algorithm used to optimize each parameter may be a known algorithm.

  Here, it is said that the tendency of changes in respiration and body movement based on the depth of sleep depends on the sleep duration time. Therefore, the model selection unit 320 selects a discrimination model considered to be optimal based on slp_et indicating the sleep continuation time and the sleep depth integration time sums at each sleep depth, and outputs the selected discrimination model to the depth discrimination unit 330. For example, the model selection unit 320 may include three discriminant models in accordance with the initial, middle and later stages of sleep. Specifically, when the indices of the discrimination model used in the early stage of sleep, the discrimination model used in the middle stage of sleep, and the discrimination model used in the late stage of sleep are 1, 2 and 3, respectively, Model 1 and Model 2 , And the discriminant matrix M and the correction vector b used in the model 3 are expressed by the following equations 4 to 6.

  The model selection unit 320 may select a discriminant model by, for example, comparing slp_et with a predetermined threshold. For example, in the case of slp_et <7000, the model selection unit 320 selects model 1; in the case of 7000 ≦ slp_et <14000, the model selection unit 320 selects model 2; in the case of 14000 ≦ slp_et, the model selection unit 320 is model 3 may be selected. Thereby, it is possible to select a discrimination model according to the sleep duration time.

  Further, the model selection unit 320 may select the discrimination model by comparing sums with a predetermined threshold. For example, if sum_rem + sum_light + sum_deep <7000, the model selection unit 320 may select model 1. In addition, if sum_rem + sum_light + sum_deep ≧ 7000 and (sum_rem + sum_light) × 0.75 <sum_deep, the model selection unit 320 may select Model 2. Further, in the case of sum_rem + sum_light + sum_deep ≧ 7000 and (sum_rem + sum_light) × 0.75 ≧ sum_deep, the model selection unit 320 may select the model 3. Thereby, it is possible to select a discriminant model according to the integration time of each sleep depth. That is, it is possible to select a discrimination model according to the depth of sleep of the subject.

  The model selection unit 320 outputs the selected model to the depth determination unit 330. The values of the parameters included in the discriminant matrix M and the correction vector b of the discriminant models shown in Equations 4 to 6 are merely examples of the values used in this embodiment, and the values of the parameters are as described above. It can be suitably changed using an algorithm such as supervised learning. In the present embodiment, the number of discriminant models is three, but in the other embodiments, the number of discriminant models is not particularly limited. Further, in the present embodiment, the discrimination model is selected according to the sleep duration time or the integration time of each sleep depth, but for example, the model is selected according to the subject's health condition, type of disease, age group, gender or the like. May be built. Also, the parameters of the discriminant model may be determined for each subject. Further, in the present embodiment, the model selection unit 320 selects the discrimination model using one of slp_et and sums, but may select the discrimination model based on both slp_et and sums. For example, the model selection unit 320 may use machine learning such as a neural network to determine a discriminant model to be selected.

(Depth determination unit)
The depth determination unit 330 determines the sleep depth of the subject using the feature amount output from the feature amount calculation unit 310 and the determination model selected by the model selection unit 320. For example, the depth discrimination unit 330 performs a linear discriminant analysis by using each of the feature amounts of bm, cv, resp, and resp_ave as independent variables and using the discriminant matrix M and the correction vector b included in the discriminant model. You may determine the sleep depth of

First, the depth determination unit 330 defines a feature vector v _t including the above four feature quantities as in the following Formula 7.

However, subscript t represents time. Further, a discrimination vector d _t used to determine the sleep depth is defined as Expression 8 below.

Here, W _t and S _t are variables that contribute to the discrimination between the awake state and the sleep state, and R _t and NR _t are variables that contribute to the discrimination between the rem sleep and the non rem sleep, and D _t and L _t is a variable that contributes to the discrimination between deep sleep and light sleep. Assuming that the number indicating the index of the model selected in the model selection unit 320 is model_idx, the discriminant vector d _t is calculated using Equation 9 below.

Next, the depth determination unit 330 determines the sleep depth of the subject based on the calculated determination vector d _t . Specifically, the depth determination unit 330 determines the four levels of sleep depth of the subject by comparing the values of the variables included in the determination vector d _t .

FIG. 8 is a diagram showing an example of a flowchart showing a determination flow of the sleep depth by the depth determination unit 330 according to the embodiment of the present invention. First, the depth determination unit 330 determines whether the subject is awake or sleep (S401). Or specifically, the depth determination unit 330, and contributes variables W _t awake, by comparing the contributing variables S _t sleep state, whether the subject is awake or sleep state To determine Here, W _t and S _t are variables calculated using the feature amount bm related to body movement and the feature amount cv related to the amplitude of respiration. For example, it is known that in the awake state, the subject's motion and respiration increase because the awake state is clearly different from the sleep state in the subject's state. Therefore, when the amplitude of body movement is large and the fluctuation of the amplitude of respiration is large, the depth determination unit 330 can determine that the subject is in an awake state. Here, if W _t > S _t (YES), the depth determination unit 330 determines that the subject is in the awake state (S402). On the other hand, if W _t ≦ S _t in step S401 (NO), the depth judging unit 330 proceeds to the next step S403, assuming that the subject is in the sleep state.

Next, the depth determination unit 330 determines whether the subject is in the REM sleep state or in the non-REM sleep state (S403). Specifically, the depth determination unit 330 compares the variable R _t contributing to the rem sleep state with the variable NR _t contributing to the non rem sleep state to determine whether the subject is in rem sleep state or non rem sleep state. Determine if it is a state. Here, R _t and NR _t are variables calculated using the feature amount bm related to body movement and the feature amount resp_ave related to fluctuation of the breathing cycle. For example, it is known that in the REM sleep state, the subject's autonomic nerve is disturbed more than in the non-REM sleep state, so that in the REM sleep state, the subject's body movement increases and the breathing interval is disturbed. Therefore, the depth determination unit 330 can determine that the subject is in the REM sleep state when the amplitude of the body motion and the fluctuation of the breathing cycle are large. Here, if R _t > NR _t (YES), the depth determining unit 330 determines that the subject is in the rem sleep state (S 404). On the other hand, if R _t ≦ NR _t in step S403 (NO), the depth judging unit 330 proceeds to the next step S405 on the assumption that the subject is in the non-REM sleep state.

Next, the depth determination unit 330 determines whether the subject is in a light sleep state or a deep sleep state (S405). Specifically, the depth discrimination unit 330 compares the variable L _t contributing to the light sleep state with the D _t contributing to the deep sleep state to determine whether the subject is in the light sleep state or the deep sleep state. Determine if it is. Here, L _t and D _t are variables calculated using the feature amount cv relating to the respiration amplitude and the feature amount resp relating to the length of the respiration cycle. For example, since respiration is stable in deep sleep, fluctuations in respiration cycle and amplitude become smaller. Therefore, when the fluctuation of the breathing cycle and amplitude is small, the depth judging unit 330 can judge that the subject is in a deep sleep state. Here, if L _t > D _t (YES), the depth determination unit 330 determines that the subject is in a light sleep state (S406). On the other hand, if L _t ≦ D _t in step S405 (NO), the depth determination unit 330 determines that the subject is in the deep sleep state (S407). As described above, the depth discrimination unit 330 uses the feature quantities obtained from the subject's breathing and body movement calculated by the feature quantity calculation unit 310, and the model selected by the model selection unit 320. Sleep states can be determined in time series.

  FIG. 9 is a diagram illustrating an example of a graph in which the estimation result of the sleep depth is output. FIG. 9 shows a graph Ph1 showing time-series changes in the sleep depth estimation result. The graph Ph1 shows the change in the sleep depth of the subject at four stages of the awake state, the REM sleep state, the light sleep state, and the deep sleep state. The sleep depth estimation unit 300 may output the sleep state of the subject estimated by the depth determination unit 330, for example, every one epoch (30 seconds). In addition, the sleep depth estimation unit 300 may store the sleep state of the subject estimated in each epoch in a storage unit (not shown) and output it later.

  In the present embodiment, the sleep depth estimation unit 300 estimates the four sleep depths, but the number of sleep depths is not particularly limited. For example, the sleep depth estimation unit 300 may estimate a two-stage sleep depth of awake state and a sleep state, or a total of six sleep depths of awake state, a REM sleep state, and four non-REM sleep states. It may be estimated. In this case, the sleep depth estimation unit 300 may set the number of variables of the discriminant vector used to estimate the sleep depth in accordance with the number of stages and the like. This makes it possible to estimate multi-step sleep depths.

  With the configuration of the sleep depth estimation apparatus 10 as described above, vibration information of the subject's breathing and body movement accompanied by fine fluctuations is estimated from the beat signal D (t) acquired by the Doppler sensor 2, and the vibration obtained further Based on the information, it is possible to estimate the sleep depth of the subject. In particular, since the sleep depth estimation device 10 can estimate the fluctuation of the breathing cycle that changes due to the disorder of breathing observed in the sleep state such as, for example, the REM sleep state, the sleep depth of the subject can be estimated more accurately. It becomes.

<3. Operation example of sleep depth estimation device>
Next, an operation example of the sleep depth estimation apparatus 10 according to an embodiment of the present invention will be described. FIG. 10 is a flowchart showing an operation example of the sleep depth estimation apparatus 10 according to an embodiment of the present invention.

  First, the beat signal acquisition unit 110 of the respiration estimation unit 100 and the body movement estimation unit 200 acquire the beat signal D (t) output from the Doppler sensor 2 (S501). The respiration estimation unit 100 estimates vibration information such as the respiration amplitude and respiration cycle of the subject from the acquired beat signal D (t) (S502). Further, the body movement estimation unit 200 estimates vibration information on body movement such as the amplitude amount of the subject from the acquired beat signal D (t) (S503). In steps S501 to S503, when the process of estimating vibration information related to respiration and body movement is continued, these processes are repeatedly performed.

  Next, the feature amount calculation unit 310 calculates a feature amount used to estimate the sleep depth from vibration information on respiration and body movement (S504). Subsequently, the model selection unit 320 selects a discriminant model used to estimate the sleep depth based on the sleep duration time and the integration time of each sleep depth (S505). Then, the depth estimation unit determines the sleep depth of the subject using the feature amount and the discrimination model described above (S506).

<4. Hardware configuration example>
Heretofore, an operation example of the sleep depth estimation device 10 according to an embodiment of the present invention has been described. The information processing of the sleep depth estimation device 10 described above is realized by the cooperation of software and the sleep depth estimation device 10. Hereinafter, the hardware configuration of the sleep depth estimation apparatus 10 according to the embodiment of the present invention will be described.

  FIG. 11 is a block diagram showing an example of the hardware configuration of the sleep depth estimation device 10 according to the embodiment of the present invention. Referring to FIG. 11, the sleep depth estimation apparatus 10 includes a central processing unit (CPU) 901, a read only memory (ROM) 902, a random access memory (RAM) 903, and a host bus 904. In addition, the sleep depth estimation apparatus 10 includes a bridge 905, an external bus 906, an interface 907, an input device 908, an output device 909, a storage device 910, a drive 911, and a network interface 912.

  The CPU 901 functions as an arithmetic processing unit and a control unit, and controls the overall operation in the sleep depth estimation apparatus 10 according to various programs. Also, the CPU 901 may be a microprocessor. In addition, CPU901 controls the whole operation | movement in the sleep depth estimation apparatus 10, or its one part. For example, the CPU 901 stores a program used by the CPU 901, calculation parameters, and the like. The RAM 903 temporarily stores programs used in the execution of the CPU 901, parameters and the like that appropriately change in the execution. These are mutually connected by a host bus 904 configured of a CPU bus and the like.

  The host bus 904 is connected to an external bus 906 such as a peripheral component interconnect / interface (PCI) bus via the bridge 905. Note that the host bus 904, the bridge 905, and the external bus 906 do not necessarily have to be separately configured, and these functions may be implemented on one bus.

  The input device 908 is an input control circuit such as a mouse, a keyboard, a touch panel, a button, a microphone, an input unit for inputting information such as a switch and a lever, and an input control circuit which generates an input signal based on an input by the user. And so on. The user of the sleep depth estimation apparatus 10 can input various data to the sleep depth estimation apparatus 10 or instruct processing operations by operating the input device 908.

  The output device 909 includes, for example, display devices such as a CRT display device, a liquid crystal display (LCD) device, an OLED device and a lamp. Furthermore, the output device 909 includes an audio output device such as a speaker and headphones. The output device 909 outputs, for example, the reproduced content. Specifically, the display device displays various information such as reproduced video data as text or image. On the other hand, the audio output device converts the reproduced audio data, the text data displayed on the display device, etc. into audio and outputs it.

  The storage device 910 is a device for storing data in the sleep depth estimation device 10 according to the embodiment of the present invention. The storage device 910 may include a storage medium, a recording device that records data in the storage medium, a reading device that reads data from the storage medium, and a deletion device that deletes data stored in the storage medium. The storage device is configured of, for example, a hard disc drive (HDD) or a solid state drive (SSD). The storage device 910 stores programs executed by the CPU 901 and various data. The storage device 910 implements the function of a storage unit (not shown).

  The drive 911 is a reader / writer for a storage medium, and is built in or externally attached to the sleep depth estimation device 10. The drive 911 reads out information recorded in a removable storage medium 96 such as a mounted magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, and outputs the read information to the RAM 903. The drive 911 can also write information to the removable storage medium 96.

  The network interface 912 is, for example, a communication interface configured of a communication device or the like for connecting to another device. Also, the network interface 912 may be a wireless local area network (LAN) compatible communication device, a long term evolution (LTE) compatible communication device, or a Bluetooth communication device. The network interface 912 may also be a wire communication device that performs wired communication. The network interface 912 implements the function of a communication unit (not shown).

  The example of the hardware configuration of the sleep depth estimation apparatus 10 has been described above. Each of the components described above may be configured using a general-purpose member, or may be configured by hardware specialized for the function of each component. Such configuration may be changed as appropriate depending on the level of technology to be implemented.

<5. Summary>
Up to this point, one embodiment of the present invention has been described with reference to FIGS. According to one embodiment of the present invention, the sleep depth estimation apparatus 10 estimates the vibration information of the respiration of the subject and the vibration information of the body movement from the acquired beat signal, and uses the estimated vibration information to calculate the subject. Estimate your sleep depth. In particular, the sleep depth estimation apparatus 10 can estimate the period of the vibration of the respiration with high accuracy by calculating the correlation with the sine wave with respect to the vibration with a minute fluctuation included in the respiration of the subject. As a result, the sleep depth estimation device 10 can estimate the sleep depth of each step having different characteristics with respect to the vibration of the respiration, such as REM sleep that causes disturbance of the respiration cycle.

  Although the preferred embodiments of the present invention have been described in detail with reference to the accompanying drawings, the present invention is not limited to such examples. It is obvious that those skilled in the art to which the present invention belongs can conceive of various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also fall within the technical scope of the present invention.

  For example, in the above embodiment, the sleep depth estimation apparatus 10 performs an operation of acquiring beat signals output from the Doppler sensor 2 in real time and analyzing them in parallel, but the present invention is not limited to such an example. . For example, the sleep depth estimation apparatus according to another embodiment may store the signal acquired once in the storage unit, and when it is desired to perform the estimation process, the signal may be extracted again from the storage unit and analyzed. This makes it possible to analyze beat signals acquired in the past collectively.

  Moreover, in the said embodiment, although the test subject assumed that it exists in the area which can be detected by the Doppler sensor 2, this invention is not limited to this example. For example, a Doppler sensor may be installed so that the Doppler sensor can detect the subject only when the subject is in a position such as a chair or bed. This makes it possible to limit the subjects to be analyzed. In addition, the sleep depth estimation apparatus may combine means for detecting a large activity such as the presence or absence of the subject or walking by the subject. This makes it possible to perform the estimation process only when the subject is in a limited state such as sleeping.

  In addition, the sleep depth estimation device 10 and the sleep depth estimation system 1 may be configured to be incorporated into another device, system, or the like. For example, the sleep depth estimation device or the Doppler sensor may be incorporated in a home appliance such as a television, an air conditioner, a medical care device, and a bedline such as a bed. Furthermore, the sleep depth estimation apparatus may use information related to the estimated sleep depth for control of the other apparatus as described above. For example, the sleep depth estimation apparatus may adjust the temperature of the air conditioner of the air conditioner, the amount of air, or the like according to the sleep depth of the subject, or may control the power supply of the television.

  Moreover, in the said embodiment, although the sleep depth estimation apparatus 10 estimated the test subject's sleep depth, this invention is not limited to the example which concerns. For example, the sleep depth estimation device may estimate mental stress of a subject. In this case, for example, the sleep depth estimation apparatus may estimate the mental stress of the subject using the beat signal by performing discriminative learning using teacher's mental stress estimated using other means as teacher data. It becomes possible. The other means described above include, for example, a test of amylase contained in saliva and a means using the frequency ratio of heart beat (ratio of high frequency component and low frequency component). In addition, the sleep depth estimation device may estimate sleep depths not only for humans but also for living bodies such as animals.

  Moreover, in the said embodiment, although the Doppler sensor 2 and the sleep depth estimation apparatus 10 comprised the sleep depth estimation system 1, this invention is not limited to the example which concerns. For example, a sleep depth estimation device or a sleep depth estimation system in which the Doppler sensor 2 and the sleep depth estimation device 10 are integrated may be provided.

  In addition, each step in the processing of the sleep depth estimation apparatus 10 in the present specification does not necessarily have to be processed chronologically in the order described as the flowchart. For example, each step in the processing of the sleep depth estimation device 10 may be processed in an order different from the order described as the flowchart or may be processed in parallel.

  In addition, a computer program for causing hardware such as a CPU, a ROM, and a RAM built in the sleep depth estimation device 10 to exhibit functions equivalent to the respective configurations of the sleep depth estimation device 10 can be created. There is also provided a storage medium storing the computer program.

DESCRIPTION OF SYMBOLS 1 sleep depth estimation system 2 Doppler sensor 10 sleep depth estimation apparatus 100 respiration estimation unit 110 beat signal acquisition unit 120 filter unit 130 signal conversion unit 140 frequency estimation unit 150 reference time estimation unit 160 vibration information identification unit 200 body movement estimation unit 300 sleep Depth estimation unit 310 Feature quantity calculation unit 320 Model selection unit 330 Depth determination unit

Claims (13)

A signal conversion unit for converting a beat signal detected by the Doppler sensor into a one-dimensional signal according to the vibration of a living body;
A frequency estimation unit that estimates the frequency of the one-dimensional signal;
A reference signal having a section length corresponding to a period corresponding to a frequency of the estimated one-dimensional signal is cut out from the one-dimensional signal, and a correlation coefficient between the reference signal and a sine wave having the frequency is calculated. A reference time estimation unit configured to estimate a reference time of the one-dimensional signal;
A vibration information identification unit that specifies vibration information of the beat signal using the reference time;
A sleep depth estimation unit configured to estimate the sleep depth of the living body based on the identified vibration information;
A sleep depth estimation device comprising:   The sleep depth estimation device according to claim 1, wherein the reference time estimation unit estimates the reference time based on a phase of the sine wave when the correlation coefficient between the reference signal and the sine wave is maximum. .   The reference time estimation unit extracts a plurality of reference times estimated in the past in a section of the reference signal used when estimating the reference time, and the reference is determined based on a distribution of the plurality of reference times extracted. The sleep depth estimation apparatus according to claim 1, wherein a time is specified.   The sleep depth according to any one of claims 1 to 3, wherein the vibration information identification unit identifies a scalar quantity of the beat signal represented by a two-dimensional vector at the reference time as the amplitude of the beat signal. Estimator.   The sleep depth estimation device according to any one of claims 1 to 3, wherein the vibration information identification unit identifies the amplitude of the one-dimensional signal at the reference time as the amplitude of the beat signal.   The sleep depth estimation device according to any one of claims 1 to 5, wherein the vibration information identification unit identifies an interval between two consecutive reference times as a period of the beat signal.   The sleep depth estimation device according to any one of claims 1 to 6, wherein the sleep depth estimation unit estimates the sleep depth of the living body based on vibration information of body movement of the living body.   The sleep depth estimation device according to any one of claims 1 to 7, wherein the sleep depth estimation unit changes an estimation condition of the sleep depth of the living body according to a duration of sleep of the living body.   The sleep depth estimation apparatus according to any one of claims 1 to 8, further comprising a filter unit that performs a filtering process to reduce components of frequencies higher than a cutoff frequency from the beat signal.   The sleep depth estimation device according to claim 9, wherein the cutoff frequency of the filter unit is 1.5 Hz or more.   The sleep depth estimation device according to any one of claims 1 to 10, wherein the vibration of the living body includes a vibration caused by respiration of the living body. Converting the beat signal detected by the Doppler sensor to a one-dimensional signal according to the vibration of the living body;
Estimating the frequency of the one-dimensional signal;
A reference signal having a section length corresponding to a period corresponding to a frequency of the estimated one-dimensional signal is cut out from the one-dimensional signal, and a correlation coefficient between the reference signal and a sine wave having the frequency is calculated. Estimating a reference time of the reference signal;
Identifying vibration information including an amplitude and a period of the beat signal using the reference time;
Estimating the sleep depth of the living body based on the identified vibration information;
Sleep depth estimation methods, including: Computer,
A signal conversion unit for converting a beat signal detected by the Doppler sensor into a one-dimensional signal according to the vibration of a living body;
A frequency estimation unit that estimates the frequency of the one-dimensional signal;
A reference signal having a section length corresponding to a period corresponding to a frequency of the estimated one-dimensional signal is cut out from the one-dimensional signal, and a correlation coefficient between the reference signal and a sine wave having the frequency is calculated. A reference time estimation unit configured to estimate a reference time of the reference signal;
A vibration information identification unit that specifies vibration information including the amplitude and period of the beat signal using the reference time;
A sleep depth estimation unit configured to estimate the sleep depth of the living body based on the identified vibration information;
A program to function as.

Images