Detailed description of the invention
2. the embodiment of pneumatic mode
2.1 pneumatic mode
Fig. 1 represents the principle of the pneumatic mode of inventor's motion. Below the mattress of home-use bed (or below cotton-padded mattress), pavingIf the airbed of the ethene system of thickness 5mm left and right. The interior pressure of airbed is the pressure identical with atmospheric pressure, and people lies in mattressWhen upper, pulse and stand up etc. that the body of formation is moving to be propagated to the air in airbed by mattress.
The pressure of measuring this air with hypersensitivity pressure sensor (RIMO of P Co., Ltd. S11-M2 processed) changes.
Pressure sensor can be measured the pressure oscillation of 0.2Pa-2Pa, has flat frequency characteristic with 0.1-3kHz[13], the signal of being measured by pressure sensor is transfused to human body signal split circuit.
In human body signal split circuit, by after the bandpass filter of 5-10Hz, implement full-wave rectification, envelope processing alsoOutput. The heart rate number of people in sleep is about 0.8Hz~1.5Hz, but has harmonic wave in the waveform of being measured by pressure sensorComponent, carries out filtering with S/N than 5 relatively high~10Hz.
2.2 problem log
In the present invention, use is as follows with the motion of the pneumatic mode measuring-signal of Fig. 1.
(P1), according to output signal, separate accurately the method for the moving component of pulse component and body.
(P2) definition represents the index of REM sleep.
(P3) definition represents the index of Depth of sleep.
(P4) according to the average originating rate that calculates each sleep stage of the age bracket definition of healthy population, the letter of standard deviationNumber.
(P5) propose to infer the R-K method algorithm of sleep stage in turn.
3. comb filter
Problem (P1) is studied, and inventors, to sensor output data, implement FFT and process, according to the base of pulseThis ripple, harmonic component and other component, separate [9-12] by moving to pulse component and body component. But, stand up etc. followingDuring body is moving, contain near the low frequency component of heart rate number, therefore, the heart rate number of sometimes asking for according to the peaks spectrum of FFT is containing wrong morePoor.
In the present invention, proposing has a kind of method, and it is in order to effectively utilize the harmonic component of pulse component, accuratelySeparate pulse and body moving, used comb filter.
In body signal separates, the signal of output samples with Δ t=0.01sec, saves as the time order shown in Fig. 2Column data. The discrete time at 1min interval is made as to k, k=1,2,, Tib。TibThat tested object couches in bed and to measureTotal time. The discrete time of each Δ t in the 1min of discrete time k is made as to l, l=1,2,, N (=60/ ΔT). The pulse component of discrete time k, l is being made as to xk(l), moving body component is made as to nk(l) time, body signal split circuitOutput yk(l) as (1) formula with the linear of the moving component of pulse component and body with represent.
yk(l)=xk(l)+nk(l)(1)
As long as according to the separable pulse component of (1) formula and the moving component of body, just can infer accurately heart rate number. By pectinationWhen the Dead Time of wave filter is made as T, τ=T/ Δ t in discrete time. In addition, the feedback oscillator of comb filter is made as to g(0≤g≤1)。
At this moment, the moving component of the pulse component of discrete time k, l and body is as the output of the comb filter of (2) formula, (3) formulaGive.
Fig. 3 (a) and (b) represent the frequency-gain characteristic of the comb filter of (2) formula, (3) formula.
(2) wave filter of formula as Fig. 3 (a) there is peak value under the frequency of integral multiple of the DC component that is shown in and 1/T. Anti-mistakeCome, it is 0 that the wave filter of (3) formula gains as shown in Figure 3 (b) under the frequency identical with the crest frequency of Fig. 3 (a). Fig. 3 (a),(b) acuity of the peak value of frequency characteristic changes according to feedback oscillator g, approach at 1 o'clock and become sharp. In (2) formula, (3)In formula, in the situation that T approached with the genuine heart rate cycle, the wave filter of (2) formula makes the amplitude amplification of pulse component, conversely,(3) wave filter of formula makes the decay of pulse component, makes body move aliquot amplification. At this, as shown in Figure 4, be arranged with side by side 78 extremelyDistrict's time T is respectively 0.66s, 0.67s, 0.68s ..., 1.43s the comb filter of (2) formula. The 1st comb filter deadDistrict time T=0.66s, the integral multiple that crest frequency is 1.51Hz, therefore, with the resonance pulse of 90 times/min. Similarly, the 78thThe Dead Time T=1.43s of individual comb filter, the integral multiple that crest frequency is 0.7Hz, therefore, is total to the pulse of 42 times/minShake. In the heart rate number of every 1 point of kind, the resolution of the Dead Time of these comb filter is 60 Δ t/T2, at 90 times/minTime be 1.35 times/min, in the time of 42 times/min, be 0.29 time/min. Therefore, these 78 comb filter groups are 42 times/min~90In the scope of inferior/min, can cover the pulse of 0.82 time/min of average resolution. By observation signal yk(l) this pectination of input in parallelFlora of filters, compares the output from each comb filterAmplitude. Infer output amplitude maximum wherein 'sCombThe Dead Time of shape wave filter, as the heart rate cycle. Then, to the Dead Time T inferring as this heart rate cycle, (3) formulaOutputMake the decay of pulse component, on the contrary, make body move aliquot amplification. Thus, can the moving size of putative aspect.
4. sleep index
4.1 indexs taking REM sleep as feature
Problem (P2) is studied.
The comb filter definition of (2) formula of use is about the index of REM sleep.
REM sleep has following feature [2,14]:
(1) expression and Non-REM1 and the similar E.E.G of awakening.
(2) occurrence frequency of δ ripple, spindle wave reduces.
(3) anxiety of antigravity muscle disappears completely.
(4) there is eye movement rapidly.
(5) Pulse Rate, Respiration Rate increase, and rhythm is irregular. In addition, increased blood pressure.
(6), the in the situation that of adult, occur that with the cycle of average 90~100min REM sleeps.
(7) moving concentrated at the forebody-afterbody of REM sleep.
In R-K method, the judgement of REM sleep stage, as shown in feature (1)~(4), is emphatically E.E.G, jaw muscle electricity, eyeballMotion. In the present invention, according to the feature of pneumatic mode (5), be conceived to that " rhythm of heart rate number becomes and do not advise in the time that REM sleeps".
By the data y of the first half 30s of the 1min data of discrete time kk(1),yk(2),yk(3)···,yk(N/2) defeatedEnter the comb filter group of Fig. 4. At this moment outputAmplitude be maximum, the Dead Time that is estimated as the heart rate cycle is made as. Similarly, the data y of the later half 30s to discrete time kk(N/2+1),yk(N/2+2),yk(N/2+3)···,yk(N),By amplitude xBecoming maximum Dead Time is made as. Thus, the first half in the 1min of this discrete time k and the later half heartThe variation of rate number is usedRepresent. With k=1,2,, TibCarry out same processing, enter by a front and back q dataRow rolling average, asks for the index of the state that represents REM sleep.
From the feature (5) of REM sleep, this index keeps large value in the time that REM sleeps.
4.2 sleep as the index of feature taking awakening, Non-REM
Problem (P3) is studied. The comb filter of (2) formula of use, (3) formula, definition is about the finger of the degree of depth of sleepMark.
Non-REM sleep
(1) proportional with the degree of depth of sleep, the occurrence frequency of δ ripple uprises.
(2), in sleep stage Non-REM2, there is spindle wave.
(3) along with sleep deepens gradually from waking state, the moving size of body diminishes, and occurrence frequency tails off.
(4) Pulse Rate is along with sleep deepens and step-down.
(5) Non-REM1 is sometimes at Non-REM3,4 or after REM sleep occurs, along with large body is moving and occur.
With the situation of REM sleep similarly according to the speciality of pneumatic mode, by the feature of feature (3) " along with sleep deepens,The moving size of body diminishes, and occurrence frequency tails off ", the depth index of definition sleep.
Output with the comb filter of (3) formula of approaching Dead Time T of heart rate cycle, pulse component is declinedSubtract, make body move aliquot amplification. In discrete time k, the mean amplitude of tide of this output is made as. What in shallow sleep, body was moving sends outRaw frequency is many,Become larger value, in deep sleep, become less value.
Sensor output signal level is because of bedding and tested object or sleeping position difference. For by this difference benchmark, incite somebody to actionThe sensor output y of discrete time kk(l) mean amplitude of tide is made as. WithRemoveAnd obtain. In addition,HavingIn the moving situation of body and do not have significantly to change in the moving situation of body, even if almost there is no also there is small change under the moving state of bodyChange. In order to control the moving large variation causing of this body, and small variation is amplified, obtainLogarithm, as tableShow the index of Depth of sleep.
This index becomes little value in the time of deep sleep, becomes large value along with shoaling.
5. inferring of sleep stage
The occurrence rate of different sleep stage of 5.1 ages
Study each sleep stage at its age while determining the actual age of input test object for problem (P4)The function of average occurrence rate and outputting standard deviation. This function to 3 years old~92 years old of bibliography [15-18] 306The occurrence rate of each sleep stage of people, with 4 curve approximations decisions.
The occurrence rate of each sleep stage represents by (6) formula, and standard deviation represents by (7) formula. (6) formula, (7) formula are to each sleepStage represents respectively. Table 1 represents coefficient c0~c4And coefficient of determination R2。
fx=c4a4+c3a3+c2a2+c1a+c0(6)
δfx=c4a4+c3a3+c2a2+c1a+c0(7)
Table 1 (6) formula, coefficient and the coefficient of determination of (7) formula
X represents each sleep stage. The domain of definition of a is set as 306 people's that become the data basis that makes this function ageAmplitude 3~92. (6) formula, (7) are even if formula is set as number of times more than 5 times, and the coefficient of determination is also almost constant, therefore, is set as 4Inferior. Fig. 5 (a) represents the average occurrence rate of the each sleep stage representing by (6) formula.
In addition, Fig. 5 (b) represents the standard deviation of the each sleep stage average originating rate representing by (7) formula.
As shown in Fig. 5 (a), fWake+fREM+fNR1+fNR2+fNR3+fNR4=100[%] set up.
The occurrence rate of sleep is along with the age increases and the occurrence rate of awakening rises, and the occurrence rate of contrary deep sleep reduces.
Inferring of 5.2 sleep stages
Problem (P5) is studied.
Inferred the time of REM sleep by REM sleep index RSI (k), (6) formula, (7) formula.
By actual age substitution (6) formula of tested object, a of (7) formula, ask fREM,δfREM。
The occurrence rate of at this moment REM sleep is set in fREM±δfREMIn scope.
As shown in Figure 6, REM is slept to index RSI (k) by the order sequence declining, at fREM±δfREMScope in, willOccur when maximum body is moving to keep the time band of RSI (k) above to distribute the length of one's sleep as REM. This is to have utilized REMThe feature (7) of sleep.
Secondly, distribute the time of awakening, Non-REM sleep 1,2,3,4 according to SDI (k), (6) formula, (7) formula.
The Depth of sleep index S DI (k) of the time band except REM sleep is pressed to the order sequence of the decline shown in Fig. 7.
With the situation of REM sleep similarly by actual age substitution (6) formula, (7) formula of tested object,
Obtain the occurrence rate of awakening, Non-REM1,2,3,4 each sleep stages.
At this moment the occurrence rate of each sleep stage is set in respectively in following scope:
Awakening: [fWake-δfNR1,fWake+δfWake]
Non-REM1:[fNR1-δfNR2,fNR1+δfNR1]
Non-REM2:[fNR2-δfNR3,fNR2+δfNR2]
Non-REM3:[fNR3-δfNR4,fNR3+δfNR3]
Non-REM4:[fNR4-δfNR3,fNR4+δfNR4]
The occurrence rate of awakening and Non-REM1 is same with REM sleep, and position moving large body in scope is separately setFor its occurrence rate. Non-REM2,3,4 occurrence rate, by SDI (k), by the order sequence declining, the large place that tilts is set asIts occurrence rate, so distributes awakening, Non-REM1,2,3,4 sleep stage.
Distribute after each sleep stage, turned back to original time series, as R-K method sleep stage in turn.
6. testing authentication
6.1 tested objects and test environment
Tested object, taking the male sex of health adult 10 (A~J), 22.2 years old mean age as object, is obtained after informed consent,Measure 20 evenings.
Fig. 8 represents Measurement situation.
Adopt the moving measurement of pneumatic mode pulse and body. As the relatively use of pulse, sleep stage, trace by multiple fluctuationDevice (SANYOFIT2500NEC tri-honor) has been measured ECG, E.E.G and eye movement simultaneously.
The checking of the validity of 6.2 comb filter
About the validity of problem (P1), the heart rate number that use comb filter is inferred and the heart rate number of being inferred by FFTCompare checking.
The comb filter group of data input Fig. 4 of the 2nd Dinner (being later made as J-2) of tested object J is inferred to heart rateNumber. This comb filter group's feedback oscillator is g=0.95.
In addition, ask the heart rate number of every 1 minute of J-2 with FFT. The data with reference to from ECG as heart rate number read R-RInterval, asks the average heart rate number of every 1min. Fig. 9 (a) represents the heart rate number of being inferred by comb filter and the heart rate of being asked for by ECGNumber, Fig. 9 (b) represents the heart rate number of being inferred by FFT and the heart rate number of being asked for by ECG. In 10min~100min, filtered by pectinationThe heart rate number that ripple device carries out is inferred inferring compared with result of this side of result and FFT, little with the error of the heart rate number of ECG. Due toThis time is waited sneaking into of low-frequency noise with respiratory components and body are moving, therefore thinks by the pectination that has amplification and arrive the characteristic of harmonic waveWave filter to infer result correct. To the data of whole tested objects, infer heart rate number with comb filter, FFT, ECG. Table2 represent the square mean error of the heart rate number of inferring by the FFT of comb filter and the heart rate number of asking for by ECG.
Inferring in result of comb filter has 17 kinds of situation errors little in 20 kinds of situations. Pushing away of comb filterDetermine in result, the mean value of error is also little than the error of FFT result.
The error of heart rate number of table 2 comb filter and the heart rate of FFT are counted the comparison of error
The checking of the appropriate property of 6.3REM sleep index
The validity of validation problem (P2).
Figure 10 represents that δ ripple, the spindle of the E.E.G that RSI (k) with C-3 and polygraph measure involve eye rapidlyThe incidence of ball motion.
(4) the rolling average number of times of the RSI of formula (k) is q=10. In the time that RSI (k) keeps large value, the variation of heart rate numberGreatly, therefore, compared with the feature (5) of REM sleep, the possibility of REM sleep is high.
The time band of the grey of Figure 10 represents that the incidence of δ ripple, spindle wave is little, and oculomotor occurrence number becomes rapidlyHeight, what REM slept is characterized as (2), (4) feature.
In identical time band, REM sleep index RSI (k) represents large value.
In addition, the position that the value of RSI (k) is large occurred with approximately 100 minute cycle, occurred the feature of (6) of REM sleep. ThisBe to be illustrated in REM sleep, the state of its feature (2), (4) and (5) occurs simultaneously, therefore, can infer REM by index RSI (k)The generation of sleep.
The checking of the appropriate property of 6.4 Depth of sleep indexs
For problem (P3) checking validity.
(3) formula is made as the mean amplitude of tide of (5) formula of g=0.8、Rolling average by 10 data is obtained.
Figure 11 represents SDI (k) and the δ ripple of C-3.
In the time band of the grey of Figure 11, the occurrence rate of δ ripple is high, from Non-REM sleep characteristics (1), is positioned at darkSleep state. In this time band, SDI (k) represents little value.
This is to represent because the feature (1) of Non-REM sleep, the state of (3) occur simultaneously, therefore, dark according to what sleepDegree index S DI (k) can infer the degree of depth of sleep.
Inferring of 6.5 sleep stages
For the validity of (P5), use the algorithm narrated and judge sleep stage in the 5th chapter, and sentence by R-K methodFixed sleep stage compares. Figure 12 represents RSI (k), the SDI (k) of the data of C-3, the sleep stage of inferring and uses R-KThe sleep stage that method determines.
Relatively by, sleep stage that SDI (k) infers and infer with R-K method sleep stage time, as in REM sleep spyLevy approximately 100 minute such cycle of (6) rhythm happening part or towards appearance that direction sleep stage shoals etc. at dawnThe shape of entirety is similar. But, near 100 minutes, 280 minutes, be judged to be the feelings of Non-REM4 with respect to the method proposingCondition is judged to be Non-REM2 in R-K method.
In addition, near 470min, in the method for proposition, be judged to be Wake, but be judged to be REM in R-K method. To with carryingThe sleep stage that the method that goes out is judged, the sleep stage of judging by R-K method respectively with 6 stages (awakening, REM, Non-REM1,(awakening, REM, Non-REM1,2,3/4) [19], 3 stages in 5 stages of 2,3,4), Non-REM3,4 being united two into one (awakening,REM, Non-REM) in compare.
Comparative approach uses TibThe ratio (consistent number/T of the consistent number of the result of determination of two kinds of methodsib)*100[%]Concordance rate, k statistic. Concordance rate when table 3 represents the sleep stage of full tested object to be divided into 6 stages, 5 stages, 3 stageK statistic, separately mean value and standard deviation.
The concordance rate of the sleep stage of the mode of table 3 motion and R-K method and κ statistic
The concordance rate when sleep stage of sleep=537min of a Dinner of Figure 12 was 6 stage is that 52.1%, k statistic is0.25, concordance rate when sleep stage was 5 stage is that 58.7%, k statistic is 0.38, concordance rate when sleep stage was 3 stageBe that 79.5%, k statistic is 0.46.
7. same with the tested object of C-3, the concordance rate of each sleep stage number of the tested object of table 3 and k statistic withNumber of stages tails off and rises.
In the situation that sleep stage was 6 stage, average 36.4% with respect to the concordance rate of document [12], concordance rateMean value rises to 51.6%.
The mean value of k statistic is at this moment 0.29, also comprises the accidental error unanimously causing.
In addition, in the time that sleep stage was 5 stage, the mean value of concordance rate rises 4.6% while being 56.2%, with respect to this, 3In stage, be 77.5%.
This be because the inconsistent rate between Non-REM1, Non-REM2, Non-REM3/4 than Non-REM3 andThe inconsistent rate of the judgement between Non-REM4 is many.
K statistic is 0.39 in the time of 5 stage, is 0.48 in the time of 3 stage, can hold figure and the REM of the overall situation of sleepThe rhythm of sleep, Non-REM sleep.
According to method as above, be used between tested object constraint, nothing invasion and attack sleep period, detect and tested objectThe health bio signal of moving, the breathing of the relevant body of sleep quality, heartbeat and so on, use comb filter to carry out data placeReason, this information is got up with relevant tested object during measure objective and subjective sleep parameter correlation join.
From improve tested object sleep quality viewpoint consider, the sleep that need to measure accurately tested object. ButBe, especially can carry out having implemented the measuring chamber of obtaining for above-mentioned data soundproof, vibrationproof, but in general societyMore difficult economically. Therefore, in the time place be set be the bedroom of general ward, family, that be utilized at present, straightConnect in the vibration proof transmission such as anti-skidding buffer, the rubber rumble of mobile wheelchair of the elastomeric material being installed on pillar, for comingFrom the vibration on floor, prevent vibration transmit and carry out high-quality and measure very difficult.
The present invention, in view of above-mentioned problem, prevents by formation the knot that the external vibration of the pillar that imports bed from floor into transmitsThe vibration proof transfer device of structure, is set as following formation.
For example, motion has a kind of vibration proof transfer device, and it is arranged on the bed without the tested object in sleep in useMultiple sensors, detect the physiology signal of h.d. in bed, use comb filter carry out data processing, extract companionIn dormant pulse and the moving detection system of body, the vibration transfer path from floor to bed vertically andHorizontal direction configuration vibration proof transferring elements, cuts off the vibration transmission of vertical component and horizontal component.
Figure 13 represent vibration proof transfer device the 1st scheme examples of implementation. In actual device, use multiple(being generally 4) buffer cell, but this figure is an one side view.
In this configuration example, in order to prevent, the transmission of floor vibration at position is set, conventionally possessing: for elastic bearingThe air spring 304 of device 1 Z direction (above-below direction), for supporting the vibration proof mechanism of this vertical direction and being fixed on bedConnecting axle 302, fixed air spring 304 on post and keep the urceolus of the movement of horizontal direction in buffer moving range303 and in order to improve the damping characteristics of Z direction of air spring 304 and and the buffer 305 of establishing. This buffer 305 is by solidDue to the bottom of device, and in the time of the high vibrationproof of the buffering for carrying out Z direction, urceolus 303 is fixed on floor 300.
When tested object bunk bed, because one end applies load more than behaviour in service, therefore, become connecting axle 302One end 302a and urceolus 303 butts, air spring 304 can not apply necessary above load. On the other hand, do not connect at bed pillarGround situation under, the other end 302b of the groove of connecting axle 302 and urceolus 303 still butt and can not expand to required scope withOutward, this just makes air spring 304 keep intact.
The body weight W1kg that the pneumatics P0kg/cm**2 of air spring 304 is tested object and the weight W 2kg phase of bed ensembleAdding, is W0kg, if the bed of 4 supportings, even if load is partially also on average 1/4, is setting W0/4kg air spring 304When contact area Scm**2, available pneumatics P=W0/4Skg/cm**2 calculates. Transmit in vibration near as this calculatingThe flexible buffer stretching in path, is set as vertically configuring with horizontal direction, vertical component and level can be dividedThe vibration of amount is transmitted and is suppressed for lower flexible buffer, is generally effective.
The 2nd scheme, is characterized in that, implements vibration proof transferring elements, because spring has kept the capable of expansion and contraction of bed pillar,Via be installed on the pulley of a pillar and with rope from bed main body hanging, and up-down vibration vertically, the vibration of vertical componentBe absorbed being installed in extension spring on rope flexible, the vibration of horizontal component at the rope by hanging as putMovement in be absorbed. The parts that are used for dangling by lengthening, also can suppress the intrinsic vibration number of horizontal direction for suitableLow.
The vibration proof transfer device of the third aspect, is characterized in that, vibration proof transferring elements uses magnet, and it is from floorIn the vibration transfer path of bed, vertically configure contrary magnet or electromagnet with horizontal direction with cordless, anti-The only vibration transmission of vertical component and horizontal component.
The vibration proof transfer device of fourth aspect, is characterized in that, vibration proof transferring elements makes to be provided with floating materialBed main body floats over a pillar or is set directly at adding on floor to be had in the tank of liquid, with tank and not phase of described bed main bodyThe mode connecing is installed holding member, makes liquid between bed main body and tank and absorbs vibration.
The vibration proof transfer device of the 5th aspect, is characterized in that, conducts to a body from the vibration on ground through footpost,Under footpost, installing one has the axis of guide of sliding eye as damper, when be subject to from bed body pressure time, bed is known from experience in guidingIn axle, slide, elastomer (air packing ring) is housed in the axis of guide, the bed horizontal direction of body and the position of vertical direction are moved all and are addedBe pressed in elastomer (air packing ring) upper, elastomer (air packing ring) is equivalent to the Compress Spring of passive vertical telescopic,Between the axis of guide and elastomer, be added with lubriation material.