WEIGHT SENSING

Abstract

A weight sensor may include a weighing platform and a load cell coupled to the platform to sense a weight applied to the platform, the load cell may include a deformable plate with one or more strain gauges arranged to provide an electrical signal representing the weight applied to the platform, and a base supporting the load cell, wherein the deformable plate is movably mounted to the base at only three contact points, the contact points allowing lateral movement of the plate when the plate deforms in response to a weight applied to the platform. The weight sensor makes it possible to monitor the weight and weight shifting of two people sharing the bed. The weight sensor is self-centering when a load is applied off-center to the platform, which is beneficial when used underneath a bed, e.g., under a bed leg or other support member which may not be aligned.

Claims

1. A weight sensing system for a bed having more than one occupant, the system comprising: a first set of weight sensors arranged under a first side of a bed; a second set of weight sensors arranged under a second side of a bed; and a circuit connecting the weight sensors that is arranged to measure a difference between the weight(s) sensed by the first and second sets of weight sensors.

2. A weight sensing system according to claim 1, wherein the circuit connecting the weight sensors comprises a data hub arranged to record differential weight data.

3. A weight sensing system according to claim 2, wherein the data hub is also arranged to record absolute weight data relating to a change in weight.

4. A weight sensing system according to claim 2, wherein the data hub is arranged to continuously collect weight data sensed by the first and second sets of weight sensors.

5. A system according to claim 2, wherein the data hub comprises a memory arranged to store only the recorded weight data.

6. A weight sensing system according to claim 1, further comprising a wireless data connection.

7. Use of a system according to claim 1 to monitor load changes of a bed while one or more occupants are sleeping.

8. Use of a system according to claim 1 in ballistocardiography.

9. A weight sensing system as claimed in claim 1, wherein the circuit is further arranged to output a first occupant weight and a second occupant weight based on the measured difference.

10. A weight sensing system as claimed in claim 1, wherein the circuit is further arranged to measure a sum of at least the weights sensed by the first set of weight sensors and the weights sensed by the second set of weight sensors.

11. A weight sensing system as claimed in claim 10, wherein the system is arranged to detect a change in weight by detecting a change in the measured sum with a magnitude that is greater than a first threshold.

12. A weight sensing system as claimed in claim 11, wherein the detected change in weight is associated with a user based at least on the measured difference between weights sensed by the first and second sets of weight sensors.

13. A weight sensing system as claimed in claim 11, wherein the detected change in weight is associated with a user based at least on a comparison between the detected change in weight and a database of users' current weights and/or users' weight histories.

14. A weight sensing system as claimed in claim 11, wherein the detected change in weight is associated with a user based at least on a user input.

15. A weight sensing system as claimed in claim 11, wherein the system is arranged to perform a frequency analysis on the measured sum signal to extract heart rate information and/or breathing rate information therefrom.

16. A weight sensing system as claimed in claim 15, wherein the heart rate information and/or breathing rate information is associated with a user based at least on a comparison between the information and a database of users' heart rate and/or breathing rate information and/or users' heart rate and/or breathing rate histories.

17. A weight sensing system as claimed in claim 11, wherein the system is arranged to perform a sleep analysis on at least one of the measured sum signal and the measured difference using a sleep model.

18. A weight sensing system as claimed in claim 17, wherein the sleep analysis is associated with a user based at least on a comparison between the analysis and a database of users' sleep data and/or sleep histories.

19. A method of sensing the weights of more than one occupant in a bed, comprising: obtaining weight data from a first set of weight sensors arranged under a first side of a bed; obtaining weight data from a second set of weight sensors arranged under a second side of the bed; measuring a difference between the weight data from the first set of weight sensors and the weight data from the second set of weight sensors; measuring a sum of at least the weights sensed by the first set of weight sensors and the weights sensed by the second set of weight sensors; detecting a change in weight by detecting a change in the measured sum with a magnitude that is greater than a first threshold; and associating the detected change in weight with one of a plurality of users.

20. A method as claimed in claim 19, wherein associating the detected change in weight with a user comprises at least one of: the measured difference between weights sensed by the first and second sets of weight sensors, a comparison between the detected change in weight and a database of users' current weights and/or users' weight histories, and a user input.

21. A method as claimed in claim 19, further comprising: analysing the measured sum data and/or the measured difference data to extract at least one of heart rate, breathing rate and an indicator of sleep quality.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Some preferred embodiments of the present invention will now be described, by way of example only, and with reference to the accompanying drawings, in which:

[0053] FIG. 1 is an exploded perspective view of a first embodiment of a sensor pad, with the platform and load cell shown as half sections;

[0054] FIG. 2 is a perspective view of an alternative load cell;

[0055] FIG. 3 is a half section of a pad fitted with the load cell of FIG. 2;

[0056] FIG. 4 is an exploded perspective view of another embodiment of a sensor pad with a plain load cell, with the platform and platform cover shown as half sections;

[0057] FIG. 5 is a half section of the sensor pad of FIG. 4 showing a bed leg with an attached wheel resting on the pad;

[0058] FIG. 6 is a half section of the sensor pad of FIG. 4 showing a slightly misaligned bed leg sitting on the pad;

[0059] FIG. 7 shows a half section of the sensor pad of FIG. 1 with a grossly misaligned bed leg causing the platform to tilt until it touches the load cell;

[0060] FIG. 8 shows a plan view of a six beam load cell that may be used in a modified sensor pad;

[0061] FIG. 9 shows the six beam load cell of FIG. 8 when it is under load;

[0062] FIG. 10 shows the absolute weight data and changes sensed for a bed over one night;

[0063] FIG. 11 shows the daily weight recorded for two occupants of the same bed over a period of five months; and

[0064] FIG. 12 shows ballistocardiography measurements from weight sensors under a bed.

[0065] FIG. 13 is a graph of the total weight signal for a bed with multiple occupants over a 24 hour period;

[0066] FIG. 14 shows a finite state machine for a weight monitoring system;

[0067] FIG. 15 shows the weight difference signal for a bed with multiple occupants over a period of about an hour;

[0068] FIG. 16 is a scatter plot showing the density of individual impulses for a portion of the time period shown in FIG. 15; and

[0069] FIG. 17 shows signals contributions for each of two individuals together with the combined total signal.

DETAILED DESCRIPTION

[0070] Referring to FIG. 1 there is shown an exploded perspective view of a sensor pad 1 consisting of an upper platform 2 mounted over a load cell 3 and a base 4. The base 4 has a bottom 19, side wall 18, an optional bearing surface 20 with optional mounting pads 21 and a cable cut-out 22. An optionally truncated triangular plate 12 (shown as a half section) with an apex bump 15 at each of three apexes 13 on a bottom surface of the plate 12, one or more strain gauges 14 mounted to either the top, bottom or both surfaces of plate 12, a central bump 17 on a top surface of the plate 12 and optional bump recesses 16 to aid the formation of the bumps 15, 17, can be placed into the base 4 so that the three apex bumps 15 rest on the bearing surface 20, or optionally directly on the bottom 19, such that the load cell 3 sits in a stable manner by virtue of having three points of contact.

[0071] The platform 2 consists of a top disc 7 with an optional upper recess 8, and an optional thickened section 9 can be placed on top of load cell 3 so that platform 2 rests on central bump 17. Any vertical force due to a load resting generally centrally on platform 2 will therefore be transmitted through the central bump 17 and then equally distributed via each of the three apex bumps 15 to the base 4. If base 4 is placed onto the floor and the leg of a bed is placed centrally on top of platform 2, the vertical force due to a component of the weight of the bed (and any occupant) will be transmitted through load cell 3 causing it to deform. The deformation can be detected using one or more strain gauges 14 forming an arm/arms of a Wheatstone bridge, a standard way to detect deformation in load cells.

[0072] An optional resilient bottom pad 23 can be placed between base 4 and the ground and an optional resilient top pad 5 can be placed between platform 2 and a bed leg (not shown). A pad recess 6 in top pad 5 is provided so that if a wheel is attached to the leg of a bed, the wheel can be stabilised by being placed into the pad recess 5. A cable 24 and cable relief bush 25 allow excitation and sense wires for the strain gauge/gauges to enter the pad 1 and resilient adhesive pads 11 (or any other connection means such as Velcro or magnets) can be positioned between platform 2 and base 4 of the load cell 3 so that the pad 1 does not come apart when handled. Whilst apex bumps 15 and central bump 17 are shown as rounded domes, the important requirement is that they have point contact and are able to slide relative to the surfaces they touch. In this manner any minor lateral motion due to a bed leg flexing side ways or any spreading of the apex bumps 15 due to deformation of the load cell 3 does not result in complex lateral deformation of the load cell 3. Only vertical motion can be stored in the load cell 3 as it deforms.

[0073] FIG. 2 shows an alternative construction for an optionally truncated triangular ball load cell 30 in which a simple plate 31 with holes has apex holes 32 at each of its apexes 13 and corresponding apex balls 33 that are too large to pass through, and a central hole 35 with a corresponding central ball 34 that is too large to pass through. The balls 33, 34 may be steel ball bearings. The advantage of this alternative construction is that the ball load cell 30 is simpler to manufacture and the flat profile makes the fitting of strain gauges 14 simpler on a production line. The balls in FIG. 2 act in the same manner as the bumps in FIG. 1, ensuring that any vertical force passes from the platform 2, via the central ball 34, through the deformed plate 31 with holes and equally through the three apex balls 33 into the base 4. Any lateral force does not lead to deformation as the balls 33, 34 are free to slide on their contact surfaces with the base 4 and platform 2.

[0074] FIG. 3 shows a pad 1 fitted with a ball load cell 30 and the apex balls 33 can be seen resting on bottom 19 of base 4 which is shown without bearing surface 20. Top disc 7, which is shown without thickened section 9, of platform 2 can also be seen resting on central ball 34.

[0075] FIG. 4 shows another embodiment in which a plain sensor pad 48 contains an optionally truncated triangular plain load cell 40 consisting of a plain plate 47 mounted with one or more strain gauges 14. The plain load cell 40 does not require any balls and holes or integrated bumps as it simply rests on a support surface 49 within plain base 50. A plain platform 51 has a central platform bump 41 on a bottom surface that rests against the centre of plain plate 47 and transmits any force acting on the plain platform 51 into plain load cell 40 which will deform due to the stable support provided at each plain plate apex 52 by support surface 49. An optional central platform recess 42 provides a location point for a wheel that may be affixed to the bottom of a bed leg (not shown) and a resilient top cover 43 may be placed to cover the plain platform 51 and provide an optional top cover recess 45 to match the central platform recess 42 and a top cover pad 44 to help align more conventional bed legs to the centre of the plain pad 48. A top cover return 46 is provided to wrap around plain base side wall 53 and seal plain pad 48 against dirt ingress. By using a thin and flexible material such as silicone rubber for top cover 43, the resistance to deformation will have a negligible effect on the deformation of the load cell.

[0076] Note that strain gauge wiring is not shown in FIGS. 1 to 4.

[0077] FIG. 5 shows a bed leg 61 with a leg wheel 60 resting in top cover recess 45 of plain pad 48 to keep it centrally located.

[0078] FIG. 6 shows a bed leg 61 resting on top cover pad 44 of resilient top cover 43. Even though the bed leg 61 is misaligned, the vertical force acting through it will be transferred vertically into plain plate 47 due to the point contact of the domed face of central platform bump 41. Tipping of platform 51 is avoided due to the area of contact between top cover pad 44 and bed leg 61.

[0079] FIG. 7 shows the sensor pad 1 of FIG. 1 with a grossly misaligned bed leg 61 resting on top pad 5 such that it causes top disc 7 to pivot on central bump 17 until it comes into contact with load cell 3 at contact point 65. Under such a condition the vertical load being transmitted by bed leg 61 is no longer passing only through the central bump 17. This will result in load cell 3 not deforming sufficiently and an incorrect measurement of the weight carried by the bed leg being derived from the strain gauges. By attaching peripheral contacts to each of the top disc 7 and the load cell 3, it is possible to detect such contact occurring when the sensing wires touch each other and are connected to a circuit that can alert the user or log a fault condition. If the top disc 7 and load cell 3 are constructed from a conductive material such as aluminium or steel, an insulator (not shown) may be placed between them at the central bump 17 and sensor wires attached to each component so that unwanted contact can be detected and the user alerted without the need for peripheral contacts. Contact detection can also be arranged so that contact is detected between the platform and the base or the platform and a dedicated contact ring.

[0080] It will be seen that the different arrangement of load cells and pads have a common theme of a triangular load cell that is supported at its three apexes (which may be truncated) and centrally loaded via a domed contact point. In this way the load acting through the platform is transferred into the load cell causing it to deform. The central domed contact point ensures that the vertical component of any force acting on the platform is the dominant force transmitted to the load cell with lateral forces being negligible.

[0081] As the load cell deforms due to the vertical load, strain gauges on the top surface of the plate will be compressed and strain gauges on the bottom surface of the plate will be stretched. By connecting one or more strain gauges to the cable and/or to each other to provide well established arrangements such as quarter, half or full Wheatstone bridges, a change in deformation can be converted to a change in resistance and voltage across the Wheatstone bridge which can be calibrated to correspond to the load being applied to the pad. For a full bridge arrangement two pairs of opposed strain gauges would be connected to each other and then to a pair of sensing wires and a pair of excitation wires, as commonly found in the field of load cells.

[0082] FIG. 8 shows a further embodiment in which a six beam load cell 66 that has six beams 67 each mounted with a pair of strain gauges 14. Three beam bumps 68 on the underside of the six beam plate 71 are mounted on three tongues 69 arranged so that if a load is placed anywhere between the three beam bumps 68 each pair of beams 67 will bend in an s shaped fashion as shown in FIG. 9 so that on each beam there will be one strain gauge 14 in compression and one in tension. The strain gauges 14 can be wired together in a conventional manner to allow the two pairs of strain gauges 14 at each apex to act in Wheatstone bridges and allow the deformation of the beams 67 to be converted into a measure of the force being applied. In this case the strain gauges 14 can be connected so that each pair of beams 67 acts as an independent load cell and the total load measured by the three load cells is equal to the load acting through the bed leg. For this reason a central bump is not required to ensure that equal loads passes through each beam bump. Instead the platform can be connected to the central area 71 of the six beam load cell (shown with hatched lines) by adhesive or other means.

[0083] A weight sensor as described hereinabove, or other variants within the scope of the present invention, may be used in a weight sensing system for a bed having one or more occupants. Such a system may comprise multiple weight sensors, for example distributed equally underneath a bed. In one example, such a system has been deployed to monitor weight changes for a bed having two occupants. From FIG. 10 it can be seen that a weight measurement graph can be generated from recording changes to the weight signal e.g. when someone gets into or out of the bed. By recording the changes in weight, the system is effectively self-zeroing and can eliminate effects e.g. due to changing the bed sheets. Furthermore, by only recording and storing changes in the weight signal, the system can keep the amount of stored data to a minimum even if the weight sensor(s) collect a weight reading every second.

[0084] From FIG. 10 it can be seen how the weight changes when one occupant leaves the bed for a short time. In the morning there is a weight change reduction when one person gets up, then the second person is seen to stay in bed for a further period of time. The system is therefore able to separate the absolute weight signals from the two occupants. FIG. 11 shows the absolute weights measured for a husband and pregnant wife sharing the same bed for five months during the pregnancy.

[0085] Finally, FIG. 12 shows how small changes in weight measured by a bed monitoring system can be used to monitor breathing and heart rate.

From Multiple Sensors to Two Data Streams

[0086] There may be any number of sensors measuring the force between some part of the bed and the floor, such that the bed is free-standing and not connected e.g. to a wall. A typical arrangement might be with four sensors, one on each leg of a bed. Another might be with a fifth sensor if the bed has a fifth leg in the middle as is common with cheaper beds. Another possibility is that a bed with rails has three sensors under each rail with one at each end and one in the middle. Whatever the arrangement sensors will always fit into three categories depending on their position with respect to the midline of the bed running from head to toe: (1) centred, (2) left of centre, and (3) right of centre.

[0087] Each sensor may be individually calibrated so that its raw signal output can be converted to a force output. This calibration may be done at the factory, for example using a known mass. The calibration may depend on the location of intended use so that a sensor calibrated for left side, top has a different calibration from a sensor calibrated for right side, centre for example. Once calibrated, the sensors must then be installed in the correct locations according to their calibrations. In alternative examples the sensors may be calibrated at the time of installation so that each sensor is calibrated for the position in which it is installed. The first stage of signal analysis is that each individual signal has the calibration for that sensor applied to it. From here onwards any change in the signal from any sensor represents the same change in force. This can be done either by an electronic circuit or using software. There may be a temperature sensor on the hub measuring the ambient temperature (outside the hub, which may generate heat), and the temperature may be used to adjust the calibration accordingly.

[0088] The first of two data streams is the total force on all sensors. This is calculated by simply adding all of the calibrated signals together. This can be done either by an electronic circuit or using software. This gives the total force between the bed and the floor. The signal may be sampled at high frequency, e.g. 60 hz. The resulting signal will provide the magnitude of any change in force on the bed.

[0089] The second of the two data streams is the left/right position of the force. This is calculated by taking the difference between all category 2 and category 3 sensors (left or right of centre); category 1 sensors do not contribute to this signal. This can be done either by an electronic circuit or using software. The signal may be sampled at high frequency, e.g. 60 Hz. The resulting signal will move one way (e.g. up) if there is a change of force to the right side of the bed and the other way (e.g. down) if it is applied to the other side. The amplitude of the signal change (as a fraction of the first signal) indicates how far to the left or right of the midline of the bed the change in force has taken place.

[0090] It should be noted that there are alternative configurations giving two data streams with the same information content, e.g. if there are no category 1 sensors then one data stream for category 2 and another for category 3 sensors is equivalent in information to the data streams described above. These two streams may be added and differenced to create the two signals discussed above. However for the purpose of describing the system, we will refer to the two data streams as defined above from here onwards.

[0091] Data streams are compressed for storage and transfer purposes in a 2-stage process. In the first stage, only changes in signal are recorded and timestamped, so if there are hours of no change, e.g. from an unoccupied bed during the day, then nothing is recorded. The second stage uses standard lossless data compression methods such as gzip.

Measurement of Occupant Weight

[0092] The first data stream gives the total force between the bed and the floor. Although possible, it is not usually necessary to calculate the absolute value, e.g. in kg, of the weight of the bed and everything on it. Instead, when there is a change in force representing an impulse above a certain threshold that then stabilises, the difference in force before and after can be used to determine the change in weight around a specific time point. For example if a person gets into bed, or indeed out of bed, the weight of that person (being above the threshold) will be recorded.

[0093] FIG. 13 is a graph showing the total weight signal on the y-axis and time on the x-axis for two people in a 24 hours period from midnight to midnight. Two different sizes of jump in the signal can be observed, since the two people are different weights. Both people are in bed at midnight; the lighter person goes to the toilet in the middle of the night; they both get up in the morning, but the lighter one goes back to bed for a lie-in before finally getting up; the lighter person goes to bed first with the heavier one joining just before midnight. The heavier person has two weight recordings and the lighter person has six during this 24 hour period.

Matching Weight to Occupant

[0094] Matching a weight recording to a person can be done in a number of ways. Three ways are discussed below. Of course any combination of the three together can be used too.

[0095] The first method uses the second signal (i.e. the difference between left side and right side sensors) to determine which side of the bed a weight measurement from the first signal (the sum of all sensors) is coming from. If it is known who sleeps on that side of the bed then the weight measurement can be matched to that person.

[0096] The second method uses the long-term weight of people. By tracking a person's weight over many days it is possible to model their average weight, the daily variance and also the second derivative (how weight is changing over time). Modeling each person (i.e. each user of the bed) allows a weight measurement to be matched to the correct model if it falls within the bounds of expectation. In the simplest form, if one person is 60 kg and the other is 75 kg, it should be easy to match a weight recording of e.g. 60.5 kg. The model can match more complex situations, e.g. if one person is gaining weight over time and the other person is losing weight over time, if their weights cross (i.e. are the same for a short period) then the model can still match weight recordings correctly after their weights subsequently diverge again.

[0097] The third method is by user self-identification, either via a button on the hub or by accessing a website. This button can be pushed before or after somebody gets into or out of bed to identify them to the system. Users can go to an online system where they can manually allocate orphan weight recordings to a person.

[0098] Methods can be combined and used together e.g. using a finite state machine. FIG. 14 shows such a finite state machine 100. An initial state Null state 110 transitions to the Empty Bed state 120 upon start up. The finite state machine 100 then transitions between the Empty Bed state 120, One Person state 130 and Full Bed state 140. The transitions X, Y and Z represent respectively a low weight measurement, an intermediate weight measurement and a high weight measurement. These may by weight bands based on thresholds. For example transition Y may be triggered as a result of a weight measurement above a first threshold (large enough that it likely represents the weight of a person in the bed) and below a second threshold (small enough that it is unlikely to represent the combined weight of two people in the bed).

Measuring Heart Rate and Breathing Rate

[0099] Heart rate (HR) and breathing (BR) produce periodic forces which can be seen in the data streams.

[0100] FIG. 12 shows high resolution data shown on a very fine scale from the first signal (sum of all sensor outputs) annotated with periodicity associated with heart rate and breathing (17 bpm in the case of breathing rate and 65 bpm in the case of heart rate in this example). The extraction of frequency information from signals with perfect periodicity is done using a Fourier Transform, however human signals are imperfect and changing. Thus various spectral analysis mathematical techniques for extracting frequency magnitude from changing signals of imprecise periodicity are applied to extract the breathing and heart rate from the signal, e.g. Lomb periodogram. Principles taken from ballistocardiography can be applied to both heart rate and breathing. These techniques are known in their respective fields and are not described further here.

[0101] The second signal can be used to determine which side of the bed corresponds to which frequency and thus match a breathing rate and heart rate to a side of the bed. Matching a person to the side of the bed (and thus the BR and HR to a person) can be done if it is known which side of the bed a person is on.

[0102] In the case where two people have different weights, matching a person to the side of the bed can equally well be done by matching the weight recording to a person when they get into or out of bed using the second method described in the weight section above, then using the second signal (left/right) to determine which side of the bed each person is on. This can also be done using the previously described method of user-self-identification. It can also be done by using long-term knowledge of breathing and heart rate variability, which could in turn be used for weight matching using the second signal, but usually long-term knowledge of weight will help to match people for the purposes of measuring HR and BR more accurately than vice versa.

Measuring Sleep

[0103] Whether occupants of a bed are awake or asleep, and whether they are in deep or light sleep and which stage of sleep they are in, can all make use of information in the signals from the device. The same signals are used for determining HR and BR, except that instead of applying spectral analysis to the signals, a complex sleep model is used. Sleep models which can be used for this are not described here.

[0104] The key point with this device is that matching sleep analysis outputs to people (i.e. multiple bed occupants) can be done using exactly the same methods as are employed for HR and BR discussed above.

Separating Forces from Multiple Occupants in a Bed

[0105] In fact the methods for matching signals to people can be used for any application beyond HR, BR and sleep analysis that makes use of fluctuations in the forces between the bed and the floor. Effectively the forces created by one person in a shared bed with multiple occupants can be split off and isolated from the rest based on the distance from the midline of the bed.

[0106] With multiple occupants, each person generates a force at a given distance left or right of the midline of the bed. Every fluctuation in the first signal can be mapped to a position on the left/right axis of the bed. If one person moves, the signal will map to their position, if a second person moves then it will map to their position. If both people move at the same time, then it will map to somewhere between the two people, however using software and algorithms to analyse the density of positions reported over a period of e.g. one or more minutes, the median densities can reveal the individual positions of each person. Given the positions calculated over a time of minutes, the share of force at any one specific point in time (e.g. several milliseconds) can be mapped to the person at the expected position. Thus software can more or less accurately separate the raw force signals for each occupant of the bed for separate processing as follows.

[0107] Step 1. Determine periods of time when all occupants are at a fixed position with respect to the midline of the bed.

[0108] For example, FIG. 15 shows the second signal over a period of about an hour for a bed with two occupants. On six occasions, one or both occupants shift their position with respect to the midline, settling the average position to a new point between left/right. Time is shown on the x-axis and the second signal (left/right difference signal) is shown on the y-axis.

[0109] Step 2. For each period of fixed position determine the left/right location of small movements or impulses.

[0110] FIG. 16 shows what the density of individual impulses could be on a scatter plot zoomed in on a roughly 20 minute period of FIG. 15 from the second signal. Centres of density can be mapped to two fractional positions between the left/right sensors to locate the individual positions relative to the midline. These two fractional positions are indicated by the horizontal lines at fractional positions around 0.24 and 0.8.

[0111] Step 3. Use the knowledge of position to separate contributions to the trace for the first signal into components coming from two separate individuals over fine timescales. E.g. the first signal is allocated to each person relative to the fractional position of the second signal between the two known locations either side of the midline determined in step 2 for a longer period of time.

[0112] FIG. 17 shows the first signal (top line) as the total of the contribution from the person on the left side (middle line) and the person on the right side (bottom line). The top line contains features of both the middle and bottom lines, but the part of the signal caused by each person is separated out in the middle and bottom lines.