WO2023061565A1

WO2023061565A1 - Apparatus, system and method for determining whether a person is asleep

Info

Publication number: WO2023061565A1
Application number: PCT/EP2021/078238
Authority: WO
Inventors: Heikki Vilho NIEMINEN; Jose Maria PEREZ MACIAS; Olli Pekka SUHONEN
Original assignee: Huawei Technologies Co., Ltd.
Priority date: 2021-10-13
Filing date: 2021-10-13
Publication date: 2023-04-20

Abstract

The present disclosure relates to an apparatus (1) for determining whether a person is asleep. The apparatus (1) is configured to receive a heart signal (S1) and/or a temperature signal (S2). The heart signal (S1) is correlated with a heart rate of the person and with blood volume variations at a measurement location. The temperature signal (S2) is correlated with a temperature of the person. The apparatus (1) is further configured to determine, based on an amplitude of the heart signal (S1) and/or based on a variable regarding an amplitude variation of the temperature signal (S2), whether the person is asleep or not. The present discourse further relates to a system comprising such an apparatus and a method for determining whether a person is asleep.

Description

APPARATUS, SYSTEM AND METHOD FOR DETERMINING WHETHER A PERSON IS ASLEEP

TECHNICAL FIELD

The disclosure relates to an apparatus, system and method for determining whether a person is asleep.

BACKGROUND

The disclosure is in the field of determining whether a person is asleep or not. The disclosure is directed to apparatuses configured to determine whether a person is asleep based on a measurement signal measured at the person.

SUMMARY

The following considerations are made by the inventors:

For determining whether a person is asleep or not, acceleration based sleep scoring algorithms may be used. Such algorithms use an acceleration of the person for computing whether the person is asleep or not. A problem with such acceleration only based sleep determination is that separation of a state, in which the person is inactive (awake and not moving or only little movement), from a state, in which the person is asleep, is often impossible using only acceleration based data. That is, when using such algorithms often a person being inactive (e.g. because of watching a movie from a tablet or browsing in the internet on a mobile) is erroneously determined or erroneously classified by such algorithms as being asleep. In other words, based on the acceleration data of the person being inactive the algorithms may erroneously determine that the person is asleep (which actually is not asleep but only inactive).

Thus, apparatuses for determining whether a person is asleep or not using an acceleration signal measured at the person for the determination, may erroneously determine or classify an inactive state of the person (i.e. person being awake but having reduced movement) as a sleeping state of the person (i.e. the person sleeping/being asleep). A further problem is that short wakeups during night are sometimes not determined (not recognized or identified).

In view of the above, the present disclosure aims to improve a determination of whether a person is asleep or not. An objective may be improving such a determination with regard to distinguishing between a state, in which the person is inactive but still awake (not asleep), and a state, in which the person is asleep (not awake). A further objective may be providing an apparatus for determining whether a person is asleep or not, which is improved with regard to the above description. The objective is achieved by the subject-matter of the enclosed independent claims. Advantageous implementations are further defined in the dependent claims.

A first aspect of the disclosure provides an apparatus for determining whether a person is asleep. The apparatus is configured to receive a heart signal and/or a temperature signal. The heart signal is a signal that is correlated with a heart rate of the person and/or with a blood volume variation of the person. The temperature signal is correlated with a temperature of the person. Further, the apparatus is configured to determine, based on an amplitude of the heart signal and/or based on a variable regarding an amplitude variation of the temperature signal, whether the person is asleep or not.

In other words, the first aspect proposes using an amplitude of a heart signal correlated with the heart rate of the person and/or the blood volume variations for determining whether the person is asleep or not. In addition or alternatively, the first aspect proposes using a variable regarding an amplitude variation of a temperature signal correlated with a temperature of the person for determining whether the person is asleep or not. Determining whether the person is asleep or not may be referred to as determining whether the person is in a sleeping state or not (i.e. whether the sleeping state of the person is present or not). When the person is not asleep the person is awake (i.e. the person is in an awake state). Thus, determining whether the person is asleep or not may be referred to as determining whether the person is asleep or awake.

Skin temperature of the person is controlled by the body’s thermoregulatory mechanism. At sleep onset (i.e. when the person falls asleep) the body’s thermal regulation decreases as body operations decrease and at sleep offset (when the person awakes) the body thermal regulation increases. Typically, when falling asleep, body heat production is lowered by a decrease in the heart rate and there is any increase in distal temperatures of the person as well as an increase in heat loss. Due to these two changes, the core body temperature (CBT) decreases. The CBT may be also referred to as a proximal temperature of the person. Thus, at sleep onset of the person (i.e. the person falling asleep), the CBT decreases, whereas the distal temperatures of the person (e.g. finger temperature) increases. Distal temperature (e.g. temperature at extremities of the person) and proximal temperature (CBT) and their gradient (DPG, i.e. gradient between distal temperature and proximal temperature) may be reliable predictors of sleep latency. The greater the peripheral vasodilation the shorter the time taken to fall asleep.

Therefore using the amplitude of the heart signal and/or a variable related to the amplitude variation of the temperature signal allows determining whether a person is asleep or not. In addition, this allows improving the determination of the sleeping state with regard to distinguishing between an inactive state of the person being awake and a sleeping state of the person being asleep. The heart signal is measured on the person’s body, e.g. at a finger, wrist, hand and/or arm of the person. That is, the blood volume variation may be blood volume variations of the person at e.g. a finger, wrist, hand and/or arm of the person. The terms “heart activity” and “heart operation” may be used as synonyms for “operation of heart”. Thus, the terms “signal indicating heart activity” and “signal indicating heart operation” may be used as synonyms for the term “heart signal”.

The amplitude of the heart signal is the difference between peak and valley values of the heart signal over time. Amplitude variation indicates the degree by which the peak to valley difference of the heart signal varies or changes overtime, e.g. during a time period. The greater the amplitude variation during a time period, the greater the degree of variation or degree of change of the amplitude during the time period and vice versa. The amplitude variation of the heart signal may be referred to as the temporal variation of amplitude of the heart signal.

The apparatus may be configured to receive the heart signal from a sensor configured to generate the heart signal. Such a sensor may be referred to as a cardiac functions sensor. The heart signal indicates the heart rate of the person and/or blood volume variation. The heart signal may comprise a sequence of peaks or pulses.

Using the amplitude of the heart signal is advantageous compared to using heart rate variability (HRV) based on heart rate data of the person for determining whether the person is asleep or not. Computing the HRV is more prone to error compared to computing the amplitude of the heart signal. Namely, for computing the HRV each peak of the heart signal needs to be detected. That is, missing one or more peaks of the heart rate reduces the correctness (or quality) of the computed HRV and, thus, may lead to wrong sleeping state determination based on the HRV. In contrast thereto, the amplitude of the heart signal may be detected in a time scale of minutes. Therefore when computing the amplitude of the heart signal it may an average over several beats of the heart may be taken. Thus missing one or more peaks of the heart signal is unlikely to have a big impact on the computed amplitude.

The amplitude variation of the temperature signal is the variation of amplitude of the temperature signal over time. It indicates the degree by which the amplitude of the temperature signal varies or changes overtime, e.g. during a time period. The greater the amplitude variation of the temperature signal during a time period, the greater the degree of variation or degree of change of the amplitude of the temperature signal during the time period and vice versa. The amplitude variation of the temperature signal may be referred to as the temporal variation of amplitude of the temperature signal.

In some conditions, distal skin temperature of a person rises already when the person is inactive, but still awake (i.e. not asleep). For example, sometimes when the person is inactive in the evening the skin temperature rises already before the person falls asleep. Therefore, using the variable related to the amplitude variation of the temperature signal is advantageous for determining whether the person is asleep or not compared to directly using the temperature of the person (e.g. absolute values of the temperature of the person or actual amplitude of the temperature signal).

The temperature of the person may be a temperature of a skin region (optionally a distal skin region), or a body temperature of the person. In other words, the temperature of the person may be a skin temperature (e.g. distal skin temperature) or a body temperature (e.g. body core temperature or proximal temperature of the person). The temperature of the person may be measured for example at the limbs (may be referred to as extremities), such as at a finger, wrist, hand and/or arm of the person. In this case, the temperature is a distal skin temperature. The apparatus may be configured to receive the temperature signal from a temperature sensor configured to generate the temperature signal. The temperature signal indicates a temperature of the person (e.g. a skin temperature, optionally a distal skin temperature). Thus, the amplitude of the temperature signal at a current time (i.e. a current amplitude) may be equal to or indicate the temperature of the person at the current time (i.e. the current temperature). In other words, the amplitude of the temperature signal may be a temperature value of the person.

The apparatus may be configured to compute, based on the heart signal, the amplitude of the heart signal. The apparatus may be configured to compute, based on the temperature signal, the variable related to the amplitude variation of the temperature signal. Optionally, computing the amplitude of the heart signal may be computing a variable indicating the amplitude of the heart signal. The description with regard to the amplitude of the heart signal herein is correspondingly valid for the variable indicating the amplitude of the heart signal.

The apparatus may comprise or be a processor, microprocessor, controller, microcontroller, field- programmable gate array (FPGA), application specific integrated circuit or any combination thereof configured to perform the functions of the apparatus as described herein. The apparatus may be a computer. The computer may comprise at least one processor and at least one data storage.

The person may be referred to as user or user of the apparatus.

In an implementation form of the first aspect, the heart signal is a photoplethysmographic (PPG) signal.

The apparatus may be configured to receive the PPG signal from a PPG sensor configured to generate the PPG signal. The PPG signal amplitude changes due to changes in cardiac functions of the person. The PPG signal amplitude increases (considerably) at sleep onset (i.e. when the person falls asleep) and decreases at sleep offset (i.e. when the person awakes). In an implementation form of the first aspect, the apparatus is configured to compute or determine the amplitude of the heart signal by computing an average amplitude of the heart signal for subsequent time windows.

In other words, for computing or determining the amplitude of the heart signal the apparatus may be configured to compute for two or more subsequent time windows an average amplitude of the heart signal.

This allows countering noise that may be present in the heart signal (e.g. the PPG signal) due to movement artifacts. That is, movement of the person may have an impact on the amplitude of the heart signal. Since the amplitude of the heart signal should only be influenced by the operation of the heart of the person, the impact of movement of the person on the amplitude of the heart signal represents noise. Computing an average amplitude of the heart signal for subsequent time windows allows filtering the noise due to movement (i.e. movement artifacts) out of the heart signal. The time windows each may have a duration of e.g. one minute (1 min).

In an implementation form of the first aspect, the apparatus is configured to determine that a signal quality of the heart signal during a time window of the time windows is equal to or greater than a threshold for the signal quality in case, during the time window, an amplitude variation (variation of peak and valley difference values) of the heart signal and/or a variation of temporal distance between the peaks of the heart signal is within normal physiological limits for cardiac functions. Further, the apparatus may be configured to compute the average amplitude of the heart signal only for time windows for which the heart signal has a signal quality equal to or greater than the threshold for the signal quality.

In other words, a scoring of the heart signal during the time windows with regard to signal quality may be performed. The lower the signal quality of the heart signal the more noise (movement artifacts) is present in the heart signal and vice versa. By computing the average amplitude of the heart signal only for time windows for which the heart signal has a signal quality equal to or greater than the threshold for the signal quality (i.e. having a sufficient signal quality), the average amplitude of the heart signal may be computed for such time windows. In other words, the amplitude of the heart signal may be computed or determined for times during which the signal quality of the heart signal is sufficient, i.e. during which there is not too much noise (e.g. due to movement artefacts) in the heart signal. This ensures that the amplitude of the heart signal is due to the operation of the heart of the person and not due to movement of the person. As long as the amplitude variation of the heart signal and/or a variation of temporal distance between the peaks of the heart signal is within normal physiological limits for cardiac functions, the amplitude of the heart signal is mainly caused by the heart. That is, the impact of other factors, e.g. movement of the person, on the amplitude of the heart signal is negligible.

Optionally, for determining the signal quality of the heart signal during a time window of the time windows, the apparatus may be configured to determine the number of subsequent peaks of the heart signal and the temporal distance between the subsequent peaks of the heart signal during the time window.

The apparatus may be configured to compare the peaks of the heart signal during a time window. The apparatus may be configured to determine that the signal quality of the heart signal during the time window is below the threshold for the signal quality, in case the peaks vary to each other by a degree that is greater than a threshold for a peak variation. The apparatus may be configured to compare the peak to valley differences (may be referred to as magnitude) of the heart signal during the time window with each other. The apparatus may be configured to determine that the signal quality of the heart signal during the time window is below the threshold for the signal quality, in case the amplitude of the heart signal varies during the time window by a degree that is greater than a threshold for an amplitude variation. In other words, the apparatus may be configured to determine the signal quality of the heart signal during the time window by determining how similar the peaks and valleys of the heart signal are to each other during the time window. In addition or alternatively, the apparatus may be configured to determine a temporal distance (duration) between the peaks of the heart signal during a time window. The apparatus may be configured to determine that the signal quality of the heart signal during the time window is below the threshold for the signal quality, in case the temporal distance between the peaks varies during the time window by a degree that is greater than a threshold for a variation of temporal distance.

In an implementation form of the first aspect, the apparatus is configured to determine that the person falls asleep in case the amplitude of the heart signal increases above a first threshold for the amplitude of the heart signal. Further the apparatus may be configured to determine that the person awakes in case the amplitude of the heart signal decreases below a second threshold for the amplitude of the heart signal.

In other words, the apparatus may be configured to determine that the person falls asleep in case the amplitude of the heart signal is greater than the first threshold for the amplitude of the heart signal, and that the person awakes in case the amplitude of the heart signal is smaller than the second threshold for the amplitude of the heart signal. The first threshold for the amplitude of the heart signal may be equal to or greater than the second threshold for the amplitude of the heart signal. The first threshold and the second threshold for the amplitude of the heart signal (e.g. the PPG signal) may depend on how a cardiac functions sensor is installed at the person (or worn by the person) for measuring the cardiac functions of the person and generating the heart signal. Wearing conditions of the cardiac functions sensor that may change are, for instance, the position on the body (e.g. a position on extremities of the body) at which the sensor is worn and how tightly the sensor is worn. The average amplitude of the heart signal may be determined when the person is awake for a current wearing condition before the apparatus starts determining whether the person is asleep or not. Based on this determined average amplitude of the heart signal, the first threshold and the second threshold for the amplitude of the heart signal may be determined. Optionally, the apparatus may be configured to determine the average amplitude of the heart signal when the person is awake for a current wearing condition before starting sleeping state determination. This may be a setup process performed by the apparatus, before the apparatus starts determining whether the person is asleep or not (e.g. when the person activates or turns on the apparatus for the sleeping state determination). The apparatus may be configured to determine average amplitude of the heart signal, the first threshold and the second threshold for the amplitude of the heart signal. From this, the awake state of the person may be determined based on a heart signal and/or the amplitude of the temperature signal, as outlined later on herein. The apparatus may be configured to store and/or receive the first and second threshold for the amplitude of the heart signal.

In an implementation form of the first aspect, the variable related to the amplitude variation of the temperature signal indicates at least one of the amplitude variation of the temperature signal, a frequency of the amplitude variation of the temperature signal, and a rate of the amplitude variation of the temperature signal.

The frequency of the amplitude variation of the temperature signal indicates the number of times (i.e. how often) the amplitude variation of the temperature signal occurs during a time period. The greater the frequency of the amplitude variation of the temperature signal during a time period, the greater the number of times (i.e. the more often) the amplitude variation of the temperature signal occurs during the time period and vice versa.

The rate of the amplitude variation of the temperature signal indicates how fast the amplitude variation of the temperature signal occurs. The greater the rate of the amplitude variation of the temperature signal during a time period the faster the amplitude variation of the temperature signal occurs during the time period and vice versa. As outlined already above, at sleep onset (i.e. when the person falls asleep) the body thermal regulation decreases as bodily operations decrease and at sleep offset (when the person awakes) the body thermal regulation increases. A decreased (reduced or diminished) body thermal regulation (thermal control) is visible as a decrease in the variable related to the amplitude variation of the temperature signal. This variable may be referred to as time domain variable.

In an implementation form of the first aspect, the apparatus is configured to compute, based on the temperature signal, the variable related to the amplitude variation of the temperature signal by computing at least one of a sum of at least two absolute values of temperature signal derivative over a sliding time window, a Fourier transform of the temperature signal for a time window, and a maximum and minimum of the temperature signal over a sliding time window or a fixed step time window.

In other words, the apparatus may be configured to compute, based on the temperature signal, the variable by computing a sum of at least two absolute values of temperature signal derivative over a sliding time window. In addition or alternatively, the apparatus may be configured to compute, based on the temperature signal, the variable by computing a Fourier transform of the temperature signal for a time window. In addition or alternatively, the apparatus may be configured to compute, based on the temperature signal, the variable by computing a maximum and minimum of the temperature signal over a sliding time window or a fixed step time window.

The term “rolling time window” may be used as a synonym for the term “sliding time window”.

The sum of at least two absolute values (i.e. two or more absolute values) of temperature signal derivative over a sliding time window may indicate the rate of the amplitude variation of the temperature signal and optional the amplitude variation of the temperature signal. The Fourier transform of the temperature signal for a time window may indicate the frequency of the amplitude variation of the temperature signal and optional the amplitude variation of the temperature signal. The maximum and minimum of the temperature signal over a sliding time window or a fixed step time window may indicate the amplitude variation of the temperature signal.

Computing the sum of at least two absolute values of temperature signal derivative over a sliding time window may be performed as follows: Taking the derivative of the temperature signal for a current time window, then taking the absolute value of the temperature signal derivative computed for the current time window. Next, the computed absolute values of temperature signal derivative (computed for two or more current time windows of the sliding time window) are summed up.

In an implementation form of the first aspect, the apparatus is configured to determine that the person falls asleep in case the variable related to the amplitude variation of the temperature signal decreases below a first threshold for the variable. Further, the apparatus may be configured to determine that the person awakes in case the variable related to the amplitude variation of the temperature signal increases above a second threshold for the variable.

In other words, the apparatus may be configured to determine that the person falls asleep in case the variable related to the amplitude variation of the temperature signal is smaller than the first threshold for the variable, and that the person awakes in case the variable related to the amplitude variation of the temperature signal is greater than the second threshold for the variable. The first threshold for the variable may be equal to or smaller than the second threshold for the variable. The apparatus may be configured to store and/or receive the first and second threshold for the variable related to the amplitude variation of the temperature signal.

The first and second threshold for the variable related to the amplitude variation of the temperature signal may be determined based on (or dependent on) a machine learning model that may be used by the apparatus for determining, based on at least the variable, whether the person is asleep or not, as outlined later on herein. Optionally, the apparatus may be configured to determine the first and second threshold for the variable related to the amplitude variation of the temperature signal. For example, in case the machine learning model comprises or is a decision tree model, the first and second threshold value may be dependent on other variables and their thresholds and optionally whether the first and second threshold value are used may be dependent on these other variables and their thresholds.

The apparatus may be configured to determine that the person falls asleep in case at least one of the following conditions is fulfilled: The amplitude variation of the temperature signal decreases below a first threshold for the amplitude variation of the temperature signal. The frequency of the amplitude variation of the temperature signal decreases below a first threshold for the frequency of the amplitude variation of the temperature signal. The rate of the amplitude variation of the temperature signal decreases below a first threshold for the rate of the amplitude variation of the temperature signal.

The apparatus may be configured to determine that the person awakes in case at least one of the following conditions is fulfilled: The amplitude variation of the temperature signal increases above a second threshold for the amplitude variation of the temperature signal. The frequency of the amplitude variation of the temperature signal increases above a second threshold for the frequency of the amplitude variation of the temperature signal. The rate of the amplitude variation of the temperature signal increases above a second threshold for the rate of the amplitude variation of the temperature signal. The aforementioned respective first threshold may be equal to or smaller than the aforementioned respective second threshold. The apparatus may be configured to store and/or receive the respective first and second threshold.

In an implementation form of the first aspect, the apparatus is configured to determine, based on at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal, whether the person is asleep or not, using a trained machine learning model. The trained machine learning model is trained based on training data comprising a plurality of data sets. Each data set of the plurality of data sets comprises a sleep state variable indicating whether the person is asleep or not at a respective time, in association with the at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal at the respective time.

The machine learning model (that is trained based on the training data, i.e. the trained machine learning model) may comprise or be a decision tree model, a random forest model, a neural network model, a deep neural network model or a hidden markov model. Any other machine learning model known in the art may be alternatively used. The data sets of the training data depend on whether the apparatus is configured to determine, using the trained machine learning model, a sleeping state of the person based on the amplitude of the heart signal or based on the variable related to the amplitude variation of the temperature signal or based on both the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal. In case of the aforementioned first alternative, each data set of the plurality of data sets comprises the sleep state variable indicating whether the person is asleep or not at a respective time, in association with the amplitude of the heart signal. In case of the aforementioned second alternative, each data set of the plurality of data sets comprises the sleep state variable in association with the variable related to the amplitude variation of the temperature signal at the respective time. In case of the aforementioned last alternative, each data set of the plurality of data sets comprises the sleep state variable in association with the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal at the respective time.

For example, the training of the machine learning model using the training data may be performed by iteratively performing an optimization algorithm. For example in an iteration of the optimization algorithm at first, the amplitude of the heart signal and/or the variable related to the amplitude variation of the temperature signal of a respective data set of the training data is input to the machine learning model. The output of the machine learning model, computed by the machine learning model based on the amplitude of the heart signal and/or the variable related to the amplitude variation of the temperature signal of the respective data set, is a sleep state variable indicating whether the person is asleep or not. This computed sleep state variable (computed for the respective data set) is compared with the sleep state variable (i.e. ground truth) of the respective data set. Next, the difference between them (i.e. error) may be reduced by adapting the machine learning model. For example, the difference (i.e. error) may be reduced by adapting a weighting of the machine learning model for processing the input data of the respective data set. After adapting the machine learning model a further iteration may be performed with another data set of the plurality of data sets of the training data. The iterative performing of the optimization algorithm may be stopped when the computed error of an iteration is smaller than a threshold for the error.

The sleep state variable may be a classification result indicating whether the person is asleep or not. The sleep state variable of each data set is the ground truth (for the respective time) of the respective data set. The sleep state variable may be a binary variable, wherein a first value of the binary variable (e.g. zero “0” or one “1”) may indicate that the person is sleeping and a second value of the binary variable (e.g. one “1” or zero “0”, respectively) may indicate that the person is not sleeping, i.e. the person is awake. Alternatively, the sleep state variable may be a number of a number range (e.g. a percentage value between 0% and 100%), wherein a number greater than a threshold of the number range (e.g. a threshold percentage) may indicate that the person is asleep and a number smaller than the threshold of the range (e.g. the threshold percentage) may indicate that the person is not asleep but awake. This may be also true for the vice versa case, i.e. number greater than the threshold may indicate that the person is awake and number smaller than the threshold may indicate that the person is asleep.

In an implementation form of the first aspect, the apparatus is configured to receive an acceleration signal correlated with an acceleration of the person. Further, the apparatus may be configured to compute, based on the acceleration signal, a degree of activity of the person. Furthermore, the apparatus may be configured to determine, based on the degree of activity in addition to at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal, whether the person is asleep or not.

The acceleration of the person may be or may comprise an acceleration of one or more body parts of the person. For example, the acceleration of the person may be an acceleration of the body region of the person at which an accelerometer for generating the acceleration signal is installed or worn by the person.

For example, the apparatus may be configured to compute, based on the acceleration signal, a degree of activity of the person using an algorithm. Such an algorithm may be referred to as sleep scoring algorithm or activity algorithm. Examples of such an algorithm comprise the Cole-Kripke algorithm [Cole RJ, Kripke DF, Gruen W, Mullaney DJ, Gillin JC. Automatic sleep/wake identification from wrist activity. Sleep. 1992 Oct;15(5):461-9.], the UCSD scoring algorithm [Jean-Louis G, Kripke DF, Mason WJ, Elliott JA, Youngstedt SD. Sleep estimation from wrist movement quantified by different actigraphic modalities. J Neurosci Methods. 2001 Feb 15;105(2): 185-91. doi: 10.1016/s0165- 0270(00)00364-2. PMID: 11275275.] and sleep algorithm by Sedeh et al [Sadeh A, Sharkey KM, Carskadon MA. Activity-based sleep-wake identification: an empirical test of methodological issues. Sleep. 1994 Apr;I7(3):20I-7.].

For example, the Cole-Kripke algorithm computes first from acceleration changes a degree of activity. The term “activity level” may be used as synonym for the term “degree of activity”. Then it may compute a sum of degree of activity over a time window (e.g. a one minute window). Next, it may compute an adjusted activity value for each time window (e.g. one minute window or one minute epoch) using for example the following formula:

Total Activity = E0 + El • 0.2 + (E-l) • 0.2 + E2 • 0.04 + (E-2) • 0.04.

In the above formula E0 is the degree of activity in a time window of interest, El is the degree of activity of one time window later (e.g. one minute later in case of a one minute window) and E-l is the degree of activity of one time window earlier (e.g. one minute earlier in case of a one minute window), and so on. If the total activity in a given time window (e.g. one minute time window or one minute epoch) is less than or equal to a wake threshold, the time window (may referred to as epoch) is scored as asleep. If the total activity in a given time window is greater than the wake threshold, the time window is scored as awake.

The degree of activity (activity level) may be a percentage value between 0% and 100%, wherein the greater the percentage value the greater the activity of the person (e.g. the more the person is moving) and vice versa. Optionally, computing the degree of activity may be computing a variable indicating the degree of activity. The description with regard to the degree of activity herein is correspondingly valid for the variable indicating the degree of activity.

In an implementation form of the first aspect, the apparatus is configured to determine that the person falls asleep in case the degree of activity decreases below a first threshold for the degree of activity. Further, the apparatus is configured to determine that the person awakes in case the degree of activity increases above a second threshold for the degree of activity.

In other words, the apparatus may be configured to determine that the person falls asleep in case the degree of activity is smaller than the first threshold for the degree of activity, and that the person awakes in case the degree of activity is greater than the second threshold for the degree of activity. The first threshold for the degree of activity may be equal to or smaller than the second threshold for the degree of activity. The apparatus may be configured to store and/or receive the first and second threshold for the degree of activity.

In an implementation form of the first aspect, the apparatus is configured to receive a temperature signal being correlated with a temperature of the person. Further, the apparatus may be configured to determine, based on an amplitude of the temperature signal in addition to at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal, whether the person is asleep or not.

The temperature signal may be the temperature signal optionally receivable by the apparatus, as already described above. The above description with regard to the temperature signal is correspondingly valid.

In an implementation form of the first aspect, the apparatus is configured to determine that the person falls asleep in case the amplitude of the temperature signal increases above a first threshold for the amplitude of the temperature signal. Further, the apparatus may be configured to determine that the person awakes in case the amplitude of the temperature signal decreases below a second threshold for the amplitude of the temperature signal.

In other words, the apparatus may be configured to determine that the person falls asleep in case the amplitude of the temperature signal is greater than the first threshold for the amplitude of the temperature signal, and that the person awakes in case the amplitude of the temperature signal is smaller than the second threshold for the amplitude of the temperature signal. The first threshold for the amplitude of the temperature signal may be equal to or greater than the second threshold for the amplitude of the temperature signal. The apparatus may be configured to store and/or receive the first and second threshold for the amplitude of the temperature signal.

In an implementation form of the first aspect, the apparatus is configured to determine, based on at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal and based on at least one of the degree of activity and the amplitude of the temperature signal, whether the person is asleep or not, using a trained machine learning model. The trained machine learning model is trained based on training data comprising a plurality of data sets. Each data set of the plurality of data sets comprises a sleep state variable indicating whether the person is asleep or not at a respective time, in association with the at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal at the respective time, and in association with the at least one of the degree of activity and the amplitude of the temperature signal at the respective time. The above description with regard to a trained machine learning model may be correspondingly valid. The machine learning model (that is trained based on the training data, i.e. the trained machine learning model) may comprise or be a decision tree model, a random forest model, a neural network model, a deep neural network model or a hidden markov model. Any other machine learning model known in the art may be alternatively used. The data sets of the training data depend on which inputs the apparatus uses for determining whether the person is asleep or not using the trained machine learning model. Each data set of the plurality of data sets comprises the sleep state variable indicating whether the person is asleep or not at a respective time, in association with respective inputs used. As outlined above the respective inputs are or comprise at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal and at least one of the degree of activity and the amplitude of the temperature signal.

The training of the machine learning model may be performed as outlined above, e.g. by iteratively performing an optimization algorithm. The optimization algorithm may be implemented as outlined above.

In order to achieve the apparatus according to the first aspect of the disclosure, some or all of the implementation forms and optional features of the first aspect, as described above, may be combined with each other.

A second aspect of the disclosure provides a system for determining whether a person is asleep. The system comprises the apparatus according to the first aspect as outlined above. Further, the system comprises a cardiac functions sensor configured to generate the heart signal and/or a temperature sensor configured to generate the temperature signal.

The cardiac functions sensor may be referred to as sensor for measuring cardiac functions. Optionally, the cardiac functions sensor may be a sensor system that comprise multiple elementary sensors. The cardiac functions sensor may be configured to measure (e.g. non-invasively measure) the heart rate of the person and blood volume variations for generating the heart signal. Optionally, the temperature sensor may be a sensor system that comprise multiple elementary sensors. The temperature sensor may be configured to measure (e.g. non-invasively measure) the temperature of the person for generating the temperature signal. The system may be referred to as a system for detecting whether a person is asleep.

In an implementation form of the second aspect, the cardiac functions sensor comprises a photoplethysmographic (PPG) sensor. In an implementation form of the second aspect, the system comprises an accelerometer configured to generate the acceleration signal. Optionally, the accelerometer may be a sensor system that comprise multiple elementary sensors. The accelerometer may be configured to measure (e.g. non-invasively) measure acceleration of the person for generating the acceleration signal.

In an implementation form of the second aspect, the system is a device wearable by the person.

The device may be wearable for example at a finger, wrist, hand and/or arm of the person. The device may be for example a watch (e.g. smart watch), a ring or clothing (e.g. band, wristband, bracelet etc.) wearable by the person. According to an implementation form, the apparatus of the system may be part of or may be a first device and the one or more optional sensors of the system described above may be part of a second device. The first device may be a portable device or user end device, such as a smart phone, tablet computer, laptop etc. The first device may be a stationary device, such as a desktop computer. The second device may be a watch, smart watch, clothing etc. The second device may be wearable by the person.

The above description of the apparatus according to the first aspect is correspondingly valid for the system of the second aspect. The description of the system according to the second aspect is correspondingly valid for the apparatus according to the first aspect.

The system of the second aspect and its implementation forms and optional features achieve the same advantages as the apparatus of the first aspect and its respective implementation forms and respective optional features.

In order to achieve the system according to the second aspect of the disclosure, some or all of the implementation forms and optional features of the second aspect, as described above, may be combined with each other.

A third aspect of the disclosure provides a method for determining whether a person is asleep. The method comprises receiving a heart signal and/or a temperature signal. The heart signal is correlated with a heart rate of the person and/or with a blood volume variation of the person. The temperature signal is correlated with a temperature of the person. Further the method comprises determining, based on an amplitude of the heart signal and/or based on a variable regarding amplitude variation of the temperature signal, whether the person is asleep or not. In other words, the heart signal is correlated with cardiac functions of the person. The description of the apparatus of the first aspect is correspondingly valid for the method of the third aspect.

In an implementation form of the third aspect, the heart signal is a photoplethysmographic (PPG) signal.

In an implementation form of the third aspect, the method comprises computing or determining the amplitude of the heart signal by computing an average amplitude of the heart signal for subsequent time windows.

In an implementation form of the third aspect, the method comprises determining that a signal quality of the heart signal during a time window of the time windows is equal to or greater than a threshold for the signal quality in case, during the time window, an amplitude variation of the heart signal and/or a variation of temporal distance between peaks of the heart signal is within normal physiological limits for cardiac functions. Further, the method may comprise computing the average amplitude of the heart signal only for time windows for which the heart signal has a signal quality equal to or greater than the threshold for the signal quality.

In an implementation form of the third aspect, the method comprises determining that the person falls asleep in case the amplitude of the heart signal increases above a first threshold for the amplitude of the heart signal. Further, the method may comprise determining that the person awakes in case the amplitude of the heart signal decreases below a second threshold for the amplitude of the heart signal.

In an implementation form of the third aspect, the variable related to the amplitude variation of the temperature signal indicates at least one of the amplitude variation of the temperature signal, a frequency of the amplitude variation of the temperature signal, and a rate of the amplitude variation of the temperature signal.

In an implementation form of the third aspect, the method comprises computing, based on the temperature signal, the variable related to the amplitude variation of the temperature signal by computing at least one of a sum of at least two absolute values of temperature signal derivative over a sliding time window, a Fourier transform of the temperature signal for a time window, and a maximum and minimum of the temperature signal over a sliding time window or a fixed step time window.

In an implementation form of the third aspect, the method comprises determining that the person falls asleep in case the variable related to the amplitude variation of the temperature signal decreases below a first threshold for the variable. Further, the method may comprise determining that the person awakes in case the variable related to the amplitude variation of the temperature signal increases above a second threshold for the variable.

In an implementation form of the third aspect, the method comprises determining, based on at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal, whether the person is asleep or not, using a trained machine learning model. The trained machine learning model is trained based on training data comprising a plurality of data sets. Each data set of the plurality of data sets comprises a sleep state variable indicating whether the person is asleep or not at a respective time, in association with the at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal at the respective time.

In an implementation form of the third aspect, the method comprises receiving an acceleration signal correlated with an acceleration of the person. Further, the method may comprise computing, based on the acceleration signal, a degree of activity of the person. Furthermore, the method may comprise determining, based on the degree of activity in addition to at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal, whether the person is asleep or not.

In an implementation form of the third aspect, the method comprises determining that the person falls asleep in case the degree of activity decreases below a first threshold for the degree of activity. Further, the method may comprise determining that the person awakes in case the degree of activity increases above a second threshold for the degree of activity.

In an implementation form of the third aspect, the method comprises receiving a temperature signal being correlated with a temperature of the person. Further, the method may comprise determining, based on an amplitude of the temperature signal in addition to at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal, whether the person is asleep or not.

In an implementation form of the third aspect, the method comprises determining that the person falls asleep in case the amplitude of the temperature signal increases above a first threshold for the amplitude of the temperature signal. Further, the method may comprise determining that the person awakes in case the amplitude of the temperature signal decreases below a second threshold for the amplitude of the temperature signal.

In an implementation form of the third aspect, the method comprises determining, based on at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal and based on at least one of the degree of activity and the amplitude of the temperature signal, whether the person is asleep or not, using a trained machine learning model. The trained machine learning model is trained based on training data comprising a plurality of data sets. Each data set of the plurality of data sets comprises a sleep state variable indicating whether the person is asleep or not at a respective time, in association with the at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal at the respective time, and in association with the at least one of the degree of activity and the amplitude of the temperature signal at the respective time.

The method of the third aspect and its implementation forms and optional features achieve the same advantages as the apparatus of the first aspect and its respective implementation forms and respective optional features.

In order to achieve the method according to the third aspect of the disclosure, some or all of the implementation forms and optional features of the third aspect, as described above, may be combined with each other.

A fourth aspect of the disclosure provides a computer program comprising program code for performing when implemented on a processor, a method according to the third aspect or any of its implementation forms.

A fifth aspect of the disclosure provides a computer program comprising a program code for performing the method according to the third aspect or any of its implementation forms.

An sixth aspect of the disclosure provides a computer comprising a memory and a processor, which are configured to store and execute program code to perform the method according to the third aspect or any of its implementation forms.

A seventh aspect of the disclosure provides a non-transitory storage medium storing executable program code which, when executed by a processor, causes the method according to the third aspect or any of its implementation forms to be performed.

An eighth aspect of the disclosure provides a computer readable storage medium storing executable program code which, when executed by a processor, causes the method according to the third aspect or any of its implementation forms to be performed. The computer program of the fourth aspect, the computer program of the fifth aspect, the computer of the sixth aspect, the non-transitory storage medium of the seventh aspect and the computer readable storage medium of the eighth aspect each achieve the same advantages as the apparatus of the first aspect and its respective implementation forms and respective optional features.

It has to be noted that all devices, elements, units and means described in the present application could be implemented in software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity which performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements, or any kind of combination thereof.

BRIEF DESCRIPTION OF DRAWINGS

The above described aspects and implementation forms of the present disclosure will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which

Figure 1 shows an apparatus for determining whether a person is asleep according to an embodiment of the present disclosure (cf. Figure 1A) and a system according to an embodiment of the present disclosure (cf. Figure IB);

Figure 2 shows an apparatus for determining whether a person is asleep according to an embodiment of the present disclosure (cf. Figure 2A) and a system according to an embodiment of the present disclosure (cf. Figure 2B);

Figure 3 shows a method for determining whether a person is asleep according to two embodiments of the present disclosure;

Figure 4 shows an example of an amplitude of a heart signal for describing how it may be determined that a person falls asleep according to an embodiment of the present disclosure;

Figure 5 shows an example of an amplitude of a heart signal for describing how it may be determined that a person awakes according to an embodiment of the present disclosure;

Figure 6 shows an example of a temperature signal in case a person falls asleep and a processing of the temperature signal according to an embodiment of the present disclosure;

Figure 7 shows an example of a temperature signal in case a person awakes and a processing of the temperature signal according to an embodiment of the present disclosure; and Figure 8 shows the graphs of Figures 4 and 6 together with a graph showing degree of activity of a person and a graph showing a sleep state variable indicating whether the person is asleep or not.

In the Figures, corresponding elements are labeled with the same reference sign.

DETAILED DESCRIPTION OF EMBODIMENTS

Figure 1 shows an apparatus for determining whether a person is asleep according to an embodiment of the present disclosure (cf. Figure 1A) and a system according to an embodiment of the present disclosure (cf. Figure IB). The apparatus of Figure 1A is an example of the apparatus according to the first aspect as described above. The system of Figure IB is an example of the system according to the second aspect as described above.

As indicated in Figure 1A, according to an alternative, the apparatus 1 for determining whether a person is asleep, may be configured to receive a signal SI indicating operation of heart, wherein the signal SI indicating operation of heart is correlated with cardiac functions of the person. The signal SI indicating operation of heart is correlated with a heart rate of the person and blood volume variations (of the person) at a measurement location. The apparatus 1 may be configured to determine, based on an amplitude of the signal S 1 indicating operation of heart whether the person is asleep or not. Optionally, the signal S 1 indicating operation of heart is a photoplethysmographic (PPG) signal. The signal SI indicating operation of heart may be any other known heart signal. The apparatus 1 may be configured to compute, based on the signal SI indicating operation of heart, the amplitude of the heart signal.

According to a further alternative, the apparatus 1 for determining whether a person is asleep may be configured to receive a temperature signal S2, wherein the temperature signal S2 is correlated with a temperature of the person. The apparatus 1 may be configured to determine, based on a variable regarding amplitude variation of the temperature signal S2, whether the person is asleep or not. The variable related to the amplitude variation of the temperature signal S2 may indicate at least one of the following: the amplitude variation of the temperature signal S2, a frequency of the amplitude variation of the temperature signal S2, and a rate of the amplitude variation of the temperature signal S2. The apparatus 1 may be configured to compute, based on the temperature signal S2, the variable related to the amplitude variation of the temperature signal S2. For computing the variable, the apparatus 1 may be configured to compute a sum of at least two absolute values of temperature signal derivative over a sliding time window. In addition or alternatively, the apparatus 1 may be configured to compute a Fourier transform of the temperature signal S2 for a time window. In addition or alternatively, the apparatus 1 may be configured to compute a maximum and minimum of the temperature signal S2 over a sliding time window or a fixed step time window. According to a further alternative, the apparatus 1 for determining whether a person is asleep may be configured to receive the signal S 1 indicating operation of heart and the temperature signal S2, and determine, based on the amplitude of the signal SI indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2, whether the person is asleep or not.

The apparatus may be or may comprise a processing circuitry (not shown in Figure l).The processing circuitry is configured to perform, conduct or initiate the various operations of the apparatus described herein for determining whether the person is asleep or not. The processing circuitry may be configured to perform, conduct or initiate the various operations of the apparatus according to the first aspect, as described above. The processing circuitry may comprise hardware and/or may be controlled by software. The hardware may comprise analog circuitry or digital circuitry, or both analog and digital circuitry. The digital circuitry may comprise components such as application-specific integrated circuits (ASICs), field-programmable arrays (FPGAs), digital signal processors (DSPs), or multi-purpose processors.

The apparatus may further comprise memory circuitry (associated with the processing circuitry, optionally being part of the processing circuitry), which stores one or more instruction(s) that can be executed by the processing circuity, optionally under control of the software (not shown in Figure 1). For instance, the memory circuitry may comprise a non-transitory storage medium storing executable software code which, when executed by the processing circuitry, causes the processing circuitry to perform, conduct or initiate the operations or methods described herein. The non-transitory storage medium may store executable software code which, when executed by the processing circuitry, causes the processing circuitry to perform, conduct or initiate the method according to the third aspect described herein.

For further details on the apparatus of Figure 1A reference is made to the above description of the apparatus according to the first aspect of the present disclosure.

As indicated in Figure IB, the system 2 comprises the apparatus 1 of Figure 1A as described above. According to an alternative, in case the apparatus 1 uses the heart signal SI for determining a sleeping state of the person, the system 2 may further comprise a cardiac functions sensor 3 configured to generate the signal S 1 indicating operation of heart. The cardiac functions sensor 3 is configured to provide the signal SI indicating operation of heart to the apparatus 1, as shown in Figure IB. The cardiac functions sensor 3 may be or may comprise a photoplethysmographic (PPG) sensor. The cardiac functions sensor 3 may be differently implemented. According to a further alternative, in case the apparatus 1 uses the temperature signal S2 for determining a sleeping state of the person, the system 2 may further comprise a temperature sensor 4 configured to generate the temperature signal S2. The temperature sensor 4 is configured to provide the temperature signal S2 to the apparatus 1, as shown in Figure IB. According to a further alternative, in case the apparatus 1 uses the signal SI indicating operation of heart and the temperature signal S2 for determining a sleeping state of the person, the system 2 may further comprise the cardiac functions sensor 3 and the temperature sensor 4.

Optionally, the system 2 is a device wearable by the person. The device 2 may be wearable for example at a finger, wrist, hand and/or arm of the person. The device 2 may be for example a watch (e.g. smart watch), a ring or clothing (e.g. band, wristband, bracelet etc.) wearable by the person. According to an embodiment, the apparatus 1 of the system 2 may part of or may be a first device and the one or more sensors 3 and 4 of the system 2 described above may be part of a second device. The first device may be a portable device or user end device, such as a smart phone, tablet computer, laptop etc. The first device may be a stationary device, such as a desktop computer. The second device may be a device wearable by the person, such as a watch, smart watch, clothing etc.

For further details on the system of Figure IB reference is made to the above description of the system according to the second aspect of the present disclosure.

Figure 2 shows an apparatus for determining whether a person is asleep according to an embodiment of the present disclosure (cf. Figure 2A) and a system according to an embodiment of the present disclosure (cf. Figure 2B). The apparatus 1 of Figure 2 A corresponds to the apparatus 1 of Figure 1A with additional optional features and the system 2 of Figure 2B corresponds to the system 2 of Figure IB with additional optional features. Therefore, the above description of the apparatus 1 of Figure 1A is valid for the apparatus 1 of Figure 2A and the above description of the system 2 of Figure IB is valid for the system 2 of Figure 2B. In the following, mainly the additional optional features with regard to Figure 1 are described.

As indicated in Figure 2A, the apparatus 1 may use, in addition to the signal S 1 indicating operation of heart and/or the temperature signal S2, an acceleration signal S3 correlated with an acceleration of the person for determining whether the person is asleep or not. In this case, the apparatus 1 may be configured to receive the acceleration signal S3 and compute, based on the acceleration signal S3, a degree of activity of the person. Further, the apparatus 1 may be configured to determine, based on the degree of activity in addition to at least one of the amplitude of the signal S 1 indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2, whether the person is asleep or not.

In addition or alternatively, the apparatus 1 may be configured to receive the temperature signal S2 and the apparatus 1 may be configured to determine, based on an amplitude of the temperature signal S2 in addition to at least one of the amplitude of the signal S 1 indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2, whether the person is asleep or not.

Thus, according to an alternative, the apparatus may be configured to determine, based on the degree of activity (computed using the acceleration signal S3) in addition to at least one of the amplitude of the signal SI indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2, whether the person is asleep or not. According to a further alternative, the apparatus 1 may be configured to determine, based on the amplitude of the temperature signal S2 in addition to at least one of the amplitude of the signal S 1 indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2, whether the person is asleep or not. According to a further alternative, the apparatus 1 may be configured to determine, based on the degree of activity and the amplitude of the temperature signal S2 in addition to at least one of the amplitude of the signal S 1 indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2, whether the person is asleep or not.

In other words, the apparatus 1 of Figure 2A may be configured to determine, based on at least one of the amplitude of the signal S 1 indicating operation of heart and the variable related to the amplitude variation of the temperature signal S2 and based on at least one of the degree of activity and the amplitude of the temperature signal S2, whether the person is asleep or not.

For further details on the apparatus 1 of Figure 2A reference is made to the above description of the apparatus 1 of Figure 1A and to the above description of the apparatus according to the first aspect of the present disclosure.

As indicated in Figure 2B, the system 2 comprises the apparatus 1 of Figure 2A as described above. In case the apparatus 1 uses the acceleration signal S3 (in addition to the signal SI indicating operation of heart and/or the temperature signal S2) for determining a sleeping state of the person, the system 2 may further comprise an accelerometer 5 configured to generate the acceleration signal S3. The accelerometer 5 is configured to provide the acceleration signal S3 to the apparatus 1, as shown in Figure 2B. In case the apparatus 1 is configured to determine, based on the amplitude of the temperature signal S2 in addition to the amplitude of the signal SI indicating operation of heart and/or the variable related to the amplitude variation of the temperature signal S2, the system 2 comprises the temperature sensor S2.

For further details on the system of Figure 2B reference is made to the above description of the system of Figure IB and to the above description of the system according to the second aspect of the present disclosure. Figure 3 shows a method for determining whether a person is asleep according to two embodiments of the present disclosure. The method of Figure 3A and the method of Figure 3B each are an example of the method according to the third aspect of the disclosure as described above. The apparatus 1 of Figures 1 and 2 may be configured to perform the methods of Figures 3A and 3B for determining whether the person is asleep or not. The above description of Figures 1 and 2 may be correspondingly valid for the methods of Figures 3A and 3B.

In step 30 of the method of Figure 3A for determining whether a person is asleep or not, a heart signal and/or a temperature signal may be received or obtained. The heart signal is correlated with the heart rate of the person and blood volume variations at a measurement location on the person (e.g. at a finger, wrist, hand and/or arm). The temperature signal is correlated with a temperature of the person. In addition, in step 30 of the method of Figure 3A, an acceleration signal correlated with an acceleration of the person may optionally be received or obtained. In step 31 following step 30, the method comprises determining, based on an amplitude of the heart signal and/or a variable regarding amplitude variation of the temperature signal and optionally based on an amplitude of the temperature signal and/or a degree of activity of the person computed based on the optional acceleration signal, whether the person is asleep or not.

The step 300 of the method of Figure 3B corresponds to the step 30 of the method of Figure 3A. In other words, in step 300 sensor measurements may be received or obtained. Therefore, the above description of the step 30 of the method of Figure 3A is valid for the step 300 of the method of Figure 3B. In the step 301 following step 300 an amplitude of the heart signal may be computed based on the heart signal in case the heart signal is received or obtained in step 300. In addition or alternatively, in step 301 following step 300 a variable regarding amplitude variation of the temperature signal may be computed based on the temperature signal in case the temperature signal is received or obtained in step 300. As outlined already above, the variable related to the amplitude variation of the temperature signal may indicate at least one of the following: the amplitude variation of the temperature signal, a frequency of the amplitude variation of the temperature signal, and a rate of the amplitude variation of the temperature signal. For example, for computing, based on the temperature signal, the variable related to the amplitude variation of the temperature signal, a sum of at least two absolute values of temperature signal derivative over a sliding time window may be computed. In addition or alternatively, a Fourier transform of the temperature signal for a time window may be computed. In addition or alternatively, a maximum and minimum of the temperature signal over a sliding time window or a fixed step time window may be computed. Optionally, in step 301 following step 300, a degree of activity of the person may be computed based on the acceleration signal, in case the acceleration signal is received or obtained in step 300. Thus, in the step 301 a feature extraction based on sensor measurements may be performed. In the step 302 following step 301, it may be determined, based on at least one of the amplitude of the heart signal and the variable regarding amplitude variation of the temperature signal and optionally based on at least one of the degree of activity of the person and an amplitude of the temperature signal, whether the person is asleep or not, using a trained machine learning model. In other words, in step 302, the amplitude of the heart signal and/or the variable related to the amplitude variation of the temperature signal and optionally the degree of activity of the person and/or the amplitude of the temperature signal may be provided as input(s) to a trained machine learning model (i.e. input to the machine learning model) in order to determine whether the person is asleep or not. For this, the trained machine learning model may compute or provide, based on the aforementioned input(s), a sleep state variable as an output, wherein the sleep state variable indicates whether the person is asleep or not. In other words, the trained machine learning model may compute for a respective time, based on the aforementioned input(s) at the respective time, the sleep state variable indicating whether the person is asleep or not at the respective time.

The trained machine learning model is trained based on training data comprising a plurality of data sets, wherein each data set of the plurality of data sets comprises the sleep state variable indicating whether the person is asleep or not at a respective time in association with the aforementioned input(s) at the respective time. Thus, each data set of the plurality of data sets may comprise the sleep state variable for a respective time in association with at least one of the amplitude of the heart signal and the variable related to the amplitude variation of the temperature signal at the respective time and optionally in association with at least one of the degree of activity of the person and an amplitude of the temperature signal at the respective time. The sleep state variable of each data set is a ground truth.

The sleep state variable may be a binary variable, wherein one value of the binary variable indicates a sleeping state (the person is asleep) and the other value of the binary variable indicates an awake state (the person is not asleep but awake). Alternatively, the sleep state variable may be a percentage value, wherein a percentage value above or below a percentage threshold indicates the sleeping state and a percentage value below or above, respectively, the percentage threshold indicates the awake state. The sleep state variable may be differently implemented.

The machine learning model used in the method of Figure 3B may be any known machine learning model, such as decision tree model, a random forest model, a neural network model, a deep neural network model or a hidden markov model etc. In the following description, it is assumed only by way of example that the trained machine learning model used in step 302 of the method of Figure 3B is a trained decision tree model. The description with regard to the decision tree model is correspondingly valid for any other machine learning model, i.e. in case another machine learning model is used. In the following table, the signals that may be received or obtained in step 300 and the input(s) for the trained machine learning model (obtainable from the aforementioned signals) for determining whether the person is asleep or not are shown. In addition, the behavior of each input at sleep onset (i.e. when the person falls asleep) and at sleep offset (i.e. when the person awakes) are shown.

With regard to the above table, an increase of the amplitude of the heart signal may indicate that the person falls asleep and a decrease of the amplitude of the heart signal may indicate that the person awakes. For example, it may be determined that the person falls asleep in case the amplitude of the heart signal increases above a first threshold for the amplitude of the heart signal. It may be determined that the person awakes in case the amplitude of the heart signal decreases below a second threshold for the amplitude of the heart signal. The aforementioned first threshold may be equal to or greater than the aforementioned second threshold.

A decrease of the variable related to the amplitude variation of the temperature signal may indicate that the person falls asleep and an increase of that variable may indicate that the person awakes. For example, it may be determined that the person falls asleep in case that variable decreases below a first threshold for that variable. It may be determined that the person awakes in case that variable increases above a second threshold for the variable. The aforementioned first threshold may be equal to or smaller than the aforementioned second threshold.

An increase of the amplitude of the temperature signal may indicate that the person falls asleep and a decrease of the amplitude of the temperature signal may indicate that the person awakes. For example, it may be determined that the person falls asleep in case the amplitude of the temperature signal increases above a first threshold for the amplitude of temperature signal. It may be determined that the person awakes in case the amplitude of the temperature signal decreases below a second threshold for the amplitude of temperature signal. The aforementioned first threshold may be equal to or greater than the aforementioned second threshold.

A decrease of the degree of activity of the person may indicate that the person falls asleep and an increase of degree of activity of the person may indicate that the person awakes. For example, it may be determined that the person falls asleep in case the degree of activity of the person decreases below a first threshold for the degree of activity of the person. It may be determined that the person awakes in case the degree of activity of the person increases above a second threshold for the degree of activity of the person. The aforementioned first threshold may be equal to or smaller than the aforementioned second threshold.

The trained machine learning model used in step 302 of the method of Figure 3B may be trained to combine the above information on the behavior of the different possible inputs indicated in the above table and described above for determining, based on the aforementioned input(s), whether the person is asleep or not. This improves the determining and, thus, contributes to overcoming the problem of erroneously determining an inactive person being awake as being asleep.

In step 302, the amplitude of the heart signal and/or the variable related to the amplitude variation of the temperature signal may be input as input(s) to the trained decision tree model. Optional, the amplitude of temperature signal and/or the degree of activity of the person may additionally be input as further input(s) to the trained decision tree model. The trained decision tree model outputs, based on the provided input(s), a classification result that is either awake or asleep. The classification result may be provided in the form of the aforementioned sleep state variable indicating whether the person is asleep or not. Classification may be done at predefined intervals or windows. Values of the aforementioned signals in current window and in neighboring windows may be used to compute the respective input(s) for that window. The trained decision tree model is trained with training data comprising multiple data sets, wherein each data set comprises the input(s) that are input to the decision tree model in association with the correct classification result (ground truth).

In an optional step 303 following the step 302 of the method of Figure 3B, the classification results output by the trained machine learning model may optionally be further processed with a morphological filter. That is, the output of the trained machine learning model may be further processed with a morphological filter. The morphological filter may comprise a closing operation and opening operation according to the following equation

Ye (y 1 * Lp) ° Lp In this equation “yi” is a binary signal coming from a classifier, “L_p” is a window length in minutes and “y_e” is the binary signal output of morphological filtering. The morphological closing operation is represented by the

The morphological opening operation is presented by the

operation, it can be represented by the morphological erosion © and dilation © operations as yi ° L_p = (yi Q L_p) © L_p.

The purpose of the morphological filtering is to take out short spurious state changes from the output of the machine learning model.

Thus, the output of the step 303 is a filtered output of the trained machine learning model indicating whether the person is asleep or not. For example, the output of step 303 may be a filtered sleep state variable indicting whether the person is asleep or not. Irrespective of whether the optional filtering step 303 is present or not, at the end the method of Figure 3B outputs a sleep/awake status indicating whether the person is asleep or awake.

For further details on the methods of Figure 3 A and 3B reference is made to the above description of the method according to the third aspect of the present disclosure.

Figure 4 shows an example of an amplitude of a heart signal for describing how it may be determined that a person falls asleep according to an embodiment of the present disclosure.

The bottom graph of Figure 4 shows the amplitude of the heart signal, which is a(the heart signal being correlated with the blood volume variations at a measurement location on the person). The measurement location on the person may be for example at a finger, wrist, hand and/or arm of the person. The amplitude of the heart signal may be computed or obtained from the heart signal, as described above. The y-axis indicates the amplitude of the heart signal in arbitrary units. The x-axis indicates the time in minutes (min). The heart signal shown in Figure 4 is a PPG signal. This is only by way of example and does not limit the present disclosure. The two graphs on the top of Figure 4 each show a filtered version of the heart signal (e.g. filtered by a high pass filter) for different time points, wherein the y-axis indicates the amplitude of the filtered PPG signal and the x-axis the time in minutes.

The graph on the top left of Figure 4 shows the filtered PPG signal when the person is awake and inactive. The graph on the top right of Figure 4 shows the filtered PPG signal when the person is sleeping (i.e. the person is asleep). Thus, at around the time of 50 minutes the person is awake and inactive and at around the time of 112 minutes the person is asleep. Therefore, an increase of the amplitude of the heart signal (the amplitude shown in the top right graph is greater than the amplitude shown in the top left graph) may indicated that the person is asleep.

The amplitude of the PPG signal shown in Figure 4 that is correlated with the blood volume variations at a measurement location on the person (at a finger, wrist, hand and/or arm), may be computed from a high pass filtered PPG signal. For example, at first all PPG signal peaks may be searched from the high pass filtered PPG signal. Then subsequent peak amplitudes may be compared to each other. The normal cardiac functions of the person produces peaks that have similar (very similar) peak amplitude and the duration between peaks varies only within physiological heart rate variability limits. Peaks caused by motion artefacts (due to motion of the person) on the other hand usually have a lot of amplitude variation. This amplitude variation may be used to determine whether the signal quality is good enough to compute an average PPG signal variance due to the heart rate.

Therefore, for computing the amplitude of the PPG signal the average amplitude caused by the cardiac functions within subsequent time windows may be computed. This allows discarding parts of the PPG signal that represent data disturbed by noise (i.e. disturbed by motion of the person) and using only the parts of the PPG signal that represent data where the amplitude change is due to the cardiac functions and, thus, data correlated with the cardiac functions. An advantage of using the amplitude of the PPG signal for determining a sleeping state is that the average amplitude of a PPG signal is much easier to compute than for instance heart rate variability (HRV) values. Thus, using the amplitude of the PPG signal provides a reliable algorithm operation (e.g. more reliable than an algorithm operation using HRV values).

Figure 5 shows an example of an amplitude of a heart signal for describing how it may be determined that a person awakes according to an embodiment of the present disclosure.

Like the two top graphs of Figure 4, the two graphs of Figure 5 each show a filtered version of the heart signal (e.g. filtered by a high pass filter) correlated with blood volume variations at a measurement location (e.g. at a finger, wrist, hand and/or arm of the person) for different time points, wherein the y- axis indicates the amplitude of the filtered PPG signal and the x-axis indicates the time in minutes. The heart signal shown in Figure 5 is a PPG signal. This is only by way of example and does not limit the present disclosure.

The graphs of Figure 5 show how the amplitude of the heart signal changes when the user wakes up. According to the left graph of Figure 5, e.g. at around the time of 464 minutes, the person is asleep and according to the right graph of Figure 5, e.g. at around the time of 526 minutes, the person is awake. Thus, the left graph of Figure 5 shows the filtered PPG signal when the person is asleep before the person awakes. The right graph of Figure 5 shows the filtered PPG signal after the person has woken up and e.g. is doing morning activities. Therefore, a decrease of the amplitude of the heart signal (the amplitude shown in the left graph is greater than the amplitude shown in the right graph) may indicated that the person is awake. Based on the amplitude of the heart signal (e.g. PPG signal) shown in the two top graphs of Figure 4 and the two graphs of Figure 5 a first threshold and second threshold for the amplitude of the heart signal may be determined, wherein in case the amplitude is above the first threshold the person is asleep and in case the amplitude is below the second threshold the person is awake. The first threshold may be equal to or greater than the second threshold.

Figure 6 shows an example of a temperature signal in case a person falls asleep and a processing of the temperature signal according to an embodiment of the present disclosure.

The top graph of Figure 6 shows an example of a temperature signal correlated with a skin temperature of the person for a time when the person is awake and inactive and a time when the person is asleep (e.g. in the evening). That is, Figure 6 shows an example of the temperature signal when a person falls asleep. The temperature signal may be generated by a temperature sensor that measures the skin temperature at an extremity of the person, e.g. at a finger, wrist, hand and/or arm. The bottom graph of Figure 6 shows an example of the variable related to the amplitude variation of the temperature signal over time, wherein the variable is computed, based on the temperature signal (shown in the top graph), by computing a sum of at least two (N > 2) absolute values of temperature signal derivative over a sliding time window. The y-axis of the top graph of Figure 6 indicates the amplitude of the temperature signal in °C and the y-axis of the bottom graph of Figure 6 indicates the value of the variable related to the amplitude variation of the temperature . The x-axis of the top and bottom graph of Figure 6 indicates the time in minutes.

As shown in Figure 6, the x-axis (time axis) is divided into four parts. Part 1 corresponds to a time period during which the person is inactive (e.g. motionlessly watching a movie on a tablet). Part 2 corresponds to a time period during which the person is preparing to sleep. Part 3 correspond to a time period during which the person falls asleep (e.g. lying in bed waiting for sleep). Part 4 corresponds to a time period during which the person is asleep. Thus, an increase of the amplitude of the temperature signal (shown during Part 3 of the top graph of Figure 6) may indicate that the person falls asleep. Further, a decrease of the variable related to the amplitude variation of the temperature signal (shown during Part 3 of the bottom graph of Figure 6) may indicate that the person falls asleep.

Figure 7 shows an example of a temperature signal in case a person awakes and a processing of the temperature signal according to an embodiment of the present disclosure.

The top graph of Figure 7 shows an example of a temperature signal correlated with a skin temperature of the person for a time when the person is asleep and a time when the person is awake (e.g. in the morning). That is, Figure 7 shows an example of the temperature signal when a person awakes. The temperature signal may be generated by a temperature sensor that measures the skin temperature at an extremity of the person, e.g. at a finger, wrist, hand and/or arm. The bottom graph of Figure 7 shows an example of the variable related to the amplitude variation of the temperature signal over time, wherein the variable is computed, based on the temperature signal (shown in the top graph), by computing a sum of at least two (N > 2) absolute values of temperature signal derivative over a sliding time window. The y-axis of the top graph of Figure 7 indicates the amplitude of the temperature signal in °C and the y-axis of the bottom graph of Figure 7 indicates the value of the variable related to the amplitude variation of the temperature signal. The x-axis of the top and bottom graph of Figure 7 indicates the time in minutes.

As shown in Figure 7, the x-axis (time axis) is divided into two parts. Part 1 of Figure 7 corresponds to a time period during which the person is sleeping (i.e. the person is asleep). Part 2 of Figure 7 corresponds to a time period during which the person awakes and, thus, is awake. Thus, a decrease of the amplitude of the temperature signal (shown during Part 2 of the top graph of Figure 7) may indicate that the person awakes. Further, an increase of the variable related to the amplitude variation of the temperature signal (shown during Part 2 of the bottom graph of Figure 7) may indicate that the person awakes. The term “wake up” may be used as a synonyms for the term “awake”.

Figure 8 shows the graphs of Figures 4 and 6 together with a graph showing degree of activity of a person and a graph showing a sleep state variable indicating whether the person is asleep or not.

The graph (A) of Figure 8 shows a degree of activity (activity level) of the person over time. Thus, the y-axis of the graph (A) indicates the degree of activity of the person. The graph (B) of Figure 8 corresponds to the top graph of Figure 6 and the graph (C) of Figure 8 corresponds to the bottom graph of Figure 6. Therefore, for describing the graphs (B) and (C) of Figure 8 reference is made to the above description of Figure 6. The graph (D) of Figure 8 shows an amplitude of a heart signal correlated with blood volume variations at a measurement location on the person (e.g. at a finger, wrist, hand and/or arm). The heart signal is a PPG signal, which is only by way of example. The y-axis of the graph (D) of Figure 8 indicates the amplitude of the heart signal (PPG signal). The graph (D) of Figure 8 corresponds to the bottom graph of Figure 4. Therefore, the description of Figure 4 is correspondingly valid for the graph (D) of Figure 8. The bottom graph (E) of Figure 8 shows a sleep state variable over time, wherein the sleep state variable indicates whether a person is asleep or not (i.e. asleep or awake). The y-axis of the bottom graph (E) indicates the value of the sleep state variable. According to Figure 8, the sleep state variable may be a binary variable, wherein a low value (zero value, “0”) of the sleep state variable indicates that the person is asleep and a high value (one value, “1”) of the sleep state variable indicates that the person is awake (not asleep). This may be vice versa, i.e. the low value may indicate the awake state and the high value may indicate the sleeping state (not shown in Figure 8). The solid line of graph (E) shows the ground truth, i.e. whether the person is actually awake or asleep. The dashed line of graph (E) shows the sleep state variable generated when performing the methods for determining whether a person is asleep according to the present disclosure as described above. The dashed dotted line of graph (E) shows the sleep state variable generated when only using the degree of activity of the person for determining whether the person is asleep or not. The x-axis of each graph of Figure 8 indicates the time in minutes.

As shown in Figure 8, the x-axis (time axis) of the graphs is divided into four parts. Part 1 corresponds to a time period during which the person is inactive (e.g. motionlessly watching a movie on a tablet). Part 2 corresponds to a time period during which the person is preparing to sleep. Part 3 correspond to a time period during which the person falls asleep (e.g. lying in bed waiting for sleep). Part 4 corresponds to a time period during which the person is asleep.

As may be derived from Figure 8, a decrease of the degree of activity of the person (shown during Part 3 of graph (A) of Figure 8) may indicate that the person falls asleep. Further, an increase of the amplitude of the temperature signal (shown during Part 3 of graph (B) of Figure 8) may indicate that the person falls asleep. Furthermore, a decrease of the variable related to the amplitude variation of the temperature signal (shown during Part 3 of graph (C) of Figure 8) may indicate that the person falls asleep. Moreover, an increase of the amplitude of the heart signal (during Part 3 of graph (D) of Figure 8, this is indicated in the two top graphs of Figure 4) may indicate that the person falls asleep.

As shown in graph (E) with the dashed dotted line indicating a sleep state determination based on only the degree of activity of the person, such a determination may erroneously determine the person being asleep. Namely, as shown in graph (A), already during Part 1 and Part 2 the degree of activity of the person changes. As a result, the sleep state variable indicated by the dashed dotted line of Figure (E) erroneously changes during the time period of Part 1 from the high value (indicating that the person is awake) to the low value (indicating that the person is sleeping) because the degree of activity decreases during that time period. Further, the sleep state variable indicated by the dashed dotted line of Figure (E) erroneously changes during the time period of Part 2 from the low value to the high value and back again to the low value because the degree of activity first increases and then decreases again during that time period. Thus, determining whether the person is asleep or not using only the degree of activity may yield wrong results because an awake state of the person, in which the person is inactive (only little or no motion), may be erroneously determined, based on only the degree of activity of the person, as the person being asleep. As outlined above, during parts 1 and 2 the person is still awake (this is indicated in graph (E) by the solid line indicating the ground truth).

According to the present disclosure, this problem may be overcome by using the amplitude of the heart signal and/or the variable regarding amplitude variation of the temperature signal for determining whether the person is asleep or not. Additionally, according to the present disclosure, the degree of activity and/or the amplitude of the temperature signal may optionally be used for determining whether the person is asleep or not. As a result, the sleep state variable generated by the methods of the present disclosure for determining whether the person is asleep or not indicated by the dashed line in graph (E) of Figure 8 only changes from the high value (indicating that the person is awake) to the low value (indicating that the person is asleep) at the beginning of Part 4. The dashed line of graph (E) does not change due to a change of the degree of activity of the person during the time period of Parts 1 and 2.

The present disclosure has been described in conjunction with various embodiments as examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed subject-matter, from the studies of the drawings, this disclosure and the independent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Claims

1. An apparatus (1) for determining whether a person is asleep, wherein the apparatus (1) is configured to receive a heart signal (SI) and/or a temperature signal (S2), the heart signal (SI) being a signal correlated with a heart rate and/or with a blood volume variation of the person, the temperature signal (S2) being a signal correlated with a temperature of the person; and determine, based on an amplitude of the heart signal (SI) and/or based on a variable related to an amplitude variation of the temperature signal (S2), whether the person is asleep or not.

2. The apparatus (1) according to claim 1, wherein the heart signal (SI) is a photoplethysmographic, PPG, signal.

3. The apparatus (1) according to claim 1 or 2, wherein the apparatus (1) is configured to compute the amplitude of the heart signal (SI) by computing an average amplitude of the heart signal (SI) for subsequent time windows.

4. The apparatus (1) according to claim 3, wherein the apparatus (1) is configured to determine that a signal quality of the heart signal (SI) during a time window of the time windows is equal to or greater than a threshold for the signal quality in case, during the time window, an amplitude variation of the heart signal (SI) and/or a variation of temporal distance between peaks of the heart signal (SI) is within normal physiological limits for cardiac functions, compute the average amplitude of the heart signal (SI) only for time windows for which the heart signal (SI) has a signal quality equal to or greater than the threshold for the signal quality.

5. The apparatus (1) according to any one of the previous claims, wherein the apparatus (1) is configured to determine that the person falls asleep in case the amplitude of the heart signal (SI) increases above a first threshold for the amplitude of the heart signal (SI) , and the person awakes in case the amplitude of the heart signal (SI) decreases below a second threshold for the amplitude of the heart signal (SI) .

6. The apparatus (1) according to any one of the previous claims, wherein the variable related to the amplitude variation of the temperature signal (S2) indicates at least one of the amplitude variation of the temperature signal (S2),

34 a frequency of the amplitude variation of the temperature signal (S2), and a rate of the amplitude variation of the temperature signal (S2). The apparatus (1) according to any one of the previous claims, wherein the apparatus (1) is configured to compute, based on the temperature signal (S2), the variable related to the amplitude variation of the temperature signal (S2) by computing at least one of a sum of at least two absolute values of temperature signal derivative over a sliding time window, a Fourier transform of the temperature signal (S2) for a time window, and a maximum and minimum of the temperature signal (S2) over a sliding time window or a fixed step time window. The apparatus (1) according to any one of the previous claims, wherein the apparatus (1) is configured to determine that the person falls asleep in case the variable related to the amplitude variation of the temperature signal (S2) decreases below a first threshold for the variable, and the person awakes in case the variable related to the amplitude variation of the temperature signal (S2) increases above a second threshold for the variable. The apparatus (1) according to any one of the previous claims, wherein the apparatus (1) is configured to determine, based on at least one of the amplitude of the heart signal (SI) and the variable related to the amplitude variation of the temperature signal (S2), whether the person is asleep or not, using a trained machine learning model, wherein the trained machine learning model is trained based on training data comprising a plurality of data sets, wherein each data set of the plurality of data sets comprises a sleep state variable indicating whether the person is asleep or not at a respective time, in association with the at least one of the amplitude of the heart signal (SI) and the variable related to the amplitude variation of the temperature signal (S2) at the respective time. The apparatus (1) according to any one of the previous claims, wherein the apparatus (1) is configured to receive an acceleration signal (S3) correlated with an acceleration of the person, compute, based on the acceleration signal (S3), a degree of activity of the person, determine, based on the degree of activity in addition to at least one of the amplitude of the heart signal (SI) and the variable related to the amplitude variation of the temperature signal (S2), whether the person is asleep or not.

35 The apparatus (1) according to claim 10, wherein the apparatus (1) is configured to determine that the person falls asleep in case the degree of activity decreases below a first threshold for the degree of activity, and the person awakes in case the degree of activity increases above a second threshold for the degree of activity. The apparatus (1) according to any one of the previous claims, wherein the apparatus (1) is configured to receive a temperature signal (S2) being correlated with a temperature of the person, and determine, based on an amplitude of the temperature signal (S2) in addition to at least one of the amplitude of the heart signal (SI) and the variable related to the amplitude variation of the temperature signal (S2), whether the person is asleep or not. The apparatus (1) according to claim 12, wherein the apparatus (1) is configured to determine that the person falls asleep in case the amplitude of the temperature signal (S2) increases above a first threshold for the amplitude of the temperature signal (S2), and the person awakes in case the amplitude of the temperature signal (S2) decreases below a second threshold for the amplitude of the temperature signal (S2). The apparatus (1) according to any one of claims 10 to 13, wherein the apparatus (1) is configured to determine, based on at least one of the amplitude of the heart signal (SI) and the variable related to the amplitude variation of the temperature signal (S2) and based on at least one of the degree of activity and the amplitude of the temperature signal (S2), whether the person is asleep or not, using a trained machine learning model, wherein the trained machine learning model is trained based on training data comprising a plurality of data sets, wherein each data set of the plurality of data sets comprises a sleep state variable indicating whether the person is asleep or not at a respective time, in association with the at least one of the amplitude of the heart signal (SI) and the variable related to the amplitude variation of the temperature signal (S2) at the respective time, and in association with the at least one of the degree of activity and the amplitude of the temperature signal (S2) at the respective time. A system (2) for determining whether a person is asleep, the system (2) comprising the apparatus (1) according to any one of the previous claims, and a cardiac functions sensor (3) configured to generate the heart signal (SI) and/or a temperature sensor (4) configured to generate the temperature signal (S2). The system (2) according to claim 15, wherein the cardiac functions sensor (3) comprises a photoplethysmographic, PPG, sensor. The system (2) according to claim 15 or 16, wherein the system comprises an accelerometer (5) configured to generate the acceleration signal (S3). The system (2) according to any one of claims 15 to 17, wherein the system (2) is a device wearable by the person. A method for determining whether a person is asleep, wherein the method comprises receiving (30) a heart signal (SI) and/or a temperature signal, the heart signal (SI) being a signal correlated with a heart rate of the person and/or with a blood volume variation, the temperature signal being correlated with a temperature of the person; and determining (31), based on an amplitude of the heart signal (SI) and/or based on a variable related to an amplitude variation of the temperature signal (S2), whether the person is asleep or not.