GB2619418A

GB2619418A - Nap monitoring device

Info

Publication number: GB2619418A
Application number: GB2308973.3A
Authority: GB
Inventors: Praxiteles Speel Stanley; Mahdi Sammy; Grys David-Benjamin; Bertrand Ambre
Original assignee: Somvai Ltd
Current assignee: Somvai Ltd
Priority date: 2023-06-15
Filing date: 2023-06-15
Publication date: 2023-12-06
Anticipated expiration: 2043-06-15
Also published as: GB2619418B

Abstract

A method of monitoring a nap of a user comprising: emitting a radio detection and ranging signal with a radar module and detecting reflection of the radio signal from the user with the radar module. sending a signal from the radar module to a processor; at the processor, determining an optimal time delay from onset of sleep until sending a wake-up signal to an alarm. The optimal nap length based on the detected reflection from the user, wherein the optimal time delay from the onset of the nap is before or after a sleep stage. The sleep stage may be REM stage or NREM-3 stage, sleep or deep sleep. The radar may also detect the environment of the user and the processor using the signal to identify the surrounding environment and location of the user such as a bed, chair etc. The radar may use frequency modulated electromagnetic wave or pulsed electromagnetic wave.

Description

Nap monitoring device

Field of the invention

The invention relates to a system and method for monitoring and optimising naps using radar technology.

Background

A short period of sleep during day time, also called a nap, can provide an improvement in energy and alertness. If a person sleeps too long, on the other hand, that person may feel groggy and disoriented. Without the right timing of waking up from a nap, a user may feel more tired after a nap than before the nap. The effect of sleep depends on the length of sleep and the moment of waking up in relation to the stages of sleep. A typical night-time sleep cycle lasts about 90 minutes and includes several stages, including light sleep, deep sleep, and REM sleep.

A 20-minute day-time nap may be beneficial for improving alertness and productivity without causing drowsiness, as it usually involves light sleep and one does not enter the deeper stages of sleep. A 90-minute nap may allow a person to complete a full sleep cycle, and may be beneficial for improving memory and creativity. Sleep during a time period of less than 3 hours may be referred to as a nap. However, the actual timing of sleep cycles and the perceived response to a sleep cycle may be personal and varies depending on the time of the day and other external factors. Technology able to monitor and control naps will therefore have significant benefits.

Statement of invention

According to a first aspect of the invention, there is provided a method of monitoring a nap of a user, the method comprising: emitting a radio detection and ranging, radar, signal with a radar module detecting reflection of the radio signal from the user with the radar module, sending a signal from the radar module to a processor; at the processor, determining an optimal time delay from onset of the nap until sending a wake-up signal to an alarm based on the detected reflection from the user, wherein the optimal time delay from the onset of the nap is before or after a sleep stage.

The determining an optimal time delay may be based on determining onset of sleep of the user based on the detected reflection of the radio signal. The sleep stage may comprise: Stage 3 non-rapid eye movement, NREM-3 stage, or rapid eye movement, REM, sleep, or deep sleep. The nap may in particular be defined as day-time sleep, and/or a period of sleep during the day with a duration of less than 3 hours.

The determining an optimal time delay based on the detected reflection from the user may comprise analysing the detected reflection from the user to determine one or more of: respiratory rate, heart rate and movement.

The method may further comprise detecting reflection of the radio signal from the environment of the user with the radar module. The determining an optimal time delay based on the detected reflection from the environment may comprise: determining with a trained machine learning algorithm an object in which the user is placed, and optionally determining whether the object is a bed.

The method may further comprise receiving an input from the user of a time constraint for sending the wakeup signal. The determining of the optimal time delay may further be based on the time of the day.

The method may further comprise activating the alarm to wake up the user on receiving the wakeup signal at the alarm. The alarm may be one or more of: an audible alarm, a tactile alarm or a visual alarm.

The method may further comprise receiving feedback from the user after waking up, using the feedback in a learning loop to adapt the optimal time delay for subsequent use.

The determining an optimal time delay may comprise: determining breathing rate and movement data based on said detecting, determining a deviation of the data from prior data by normalising the determined breathing rate and movement data over prior breathing rate and movement data, inputting the data into a deterministic or statistical model, optionally a logistic regression model, to determine the time of nap onset, determining said optimal time delay depending on the user's input constraint, a determined surrounding environment, and prior feedback from the user.

The determining an optimal time delay may comprise: determining breathing rate and movement data based on said detecting, determining a deviation of the data from prior data by normalising the determined breathing rate and movement data over prior breathing rate and movement data, inputting the data into a deterministic or statistical model, optionally a logistic regression model, to determine the time of nap onset, determining time to N3REM onset, determining a probability that the time to N3REM onset is below a static threshold; and determining whether that probability is greater than a dynamic threshold, whereby the dynamic threshold is a function of the user's input constraint, the surrounding environment, and prior feedback from the user; and when the probability is greater than the dynamic threshold sending a wakeup signal to an alarm.

The determining an optimal time delay may comprise: determining breathing rate, heart rate and movement data based on said detecting, determining a deviation of the data from prior data by normalising the determined breathing rate, heart rate and movement data over prior data, inputting the data into a deterministic or statistical model, optionally a logistic regression model, to determine the time of nap onset, determining said optimal time delay depending on the user's input constraint, a determined surrounding environment, phase of the nap, and prior feedback from the user.

The determining an optimal time delay may comprise: using a trained machine-learning model to determine the optimal time delay, wherein the machine-learning model has been trained based on prior data comprising: movement, breathing rate, the user's environment and user constraints.

According to a second aspect of the invention, there is provided a non-invasive system suitable for monitoring and controlling a nap of a user, the system comprising: a radio detection and ranging, radar, module arranged to emit a radio signal and detect reflection of the radio signal from the user, and further to detect reflection of the radio signal by an environment of the user; a processor arranged to receive an output signal from the radar module, wherein the processor is arranged to determine an optimal time delay from onset of the nap until sending a wake-up signal to an alarm based on the detected reflection from the user, wherein the optimal time delay from the onset of the nap is before or after a sleep stage.

The sleep stage may comprise: Stage 3 non-rapid eye movement, Stage NREM-3, or rapid eye movement, REM, sleep, or deep sleep, and wherein said nap may be day-time sleep, and/or wherein said nap may comprise a period of sleep with a duration of less than 3 hours.

The frequency of the radio signal may be 24GHz or 60GHz, or may be centred around these frequencies. The radio signal may be a frequency-modulated continuous electromagnetic wave, or a pulsed electromagnetic wave.

The system may further be arranged to analyse the detected reflection from the user to determine one or more of: respiratory rate, heart rate and movement, for determining the optimal time delay based on the detected reflection from the user comprises.

The system may further be arranged to detect reflection of the radio signal from the environment of the user with the radar module, and optionally further be arranged to determine with a trained machine learning algorithm an object in which the user is placed, and optionally to determine whether the object is a bed.

The system may further comprise a user input interface, the interface may be suitable for receiving an input from the user of a time constraint for sending the wakeup signal.

The system may further comprise an alarm for waking up the user, wherein the alarm is optionally one or more of: an audible alarm, a tactile alarm or a visual alarm.

The system may further comprise an output interface for providing information to a user, for example a time constraint set by the user, time left until the alarm is activated, or user inputs received.

Figures Some embodiments of the invention will now be described by way of example only and with reference to the accompanying drawings, in which: Fig. 1 is a schematic drawing of a user and radar system; Fig. 2 is a timeline of a process; Fig. 3 is a flowchart of method steps; Fig. 4 is a schematic overview of a system; Figs. 4a to 4d are schematic representations of algorithms; and Fig 5 is a flowchart of a method.

Specific description

Short periods of sleep, also referred to as naps, are a distinct form of rest that occur during the daytime hours and are characterised by a different pattern of brain activity than night-time sleep. Unlike night-time sleep, naps are typically shorter in duration, lasting between 10 and 45 minutes, although longer naps are also possible. Naps do not typically involve, or may aim to avoid, the full range of sleep stages, including rapid-eyemovement (REM) sleep, and for shorter naps, often also avoid stage 3 non-rapid-eyemovement (NREM-3) sleep. Instead, naps are defined by specific brain-wave patterns, and may be characterised by a reduction in alpha waves and an increase in slow-wave activity, which is generally less intense than seen during night-time sleep. As naps are shorter than a full-night's sleep, the detectable brain wave patterns are often shorter and do not have the same proportion of each sleep stage. The inventors have developed an algorithm for optimising day-time naps, rather than night-time sleep. The algorithm determines the optimal wake-up time for an individual by considering multiple input factors, for example sleep modality, time-to-sleep onset, circadian rhythm, and the surrounding environment.

The inventors have realised that non-invasive radio detection and ranging (radar) technology can be used to monitor and control sleep, in particular naps of 3 hours or less. A radar module is provided and arranged to emit a radar signal, and subsequently detect the reflection of the radar signal from a user and optionally also the environment of the user. The reflection from the user can be used to detect respiratory rate, heart rate and movement, and radar technology is sufficiently sensitive to detect such micro-movements of users. The reflected signal can be analysed by an algorithm installed on a processor to determine when the user falls asleep. It is also possible to detect the different stages of sleep, such as light sleep, REM sleep, NREM-3 sleep, and deep sleep. The user's movement and breathing patterns that are associated with different stages of sleep can be detected. For example, during REM sleep, the body's muscles are typically more relaxed, and there may be more movement in the user's eyes and chest due to increased breathing and heart rate variability. The reflected signal from the environment can be used to detect the object in which the user rests, such as a chair, at a desk, on the floor, or in bed. The environment signal can also be used to detect whether the user is in a bedroom, in an office, or another space such as a public space. The environment may be recognisable by a trained machine learning model, and/or recognise specifically the object the user rests in.

Fig. 1 illustrates schematically a user 101, resting on a desk 102. A radar module 103 emits a signal 104 and detects a return signal 105 reflected by the user 101 and the desk 102. The radar module 103 is coupled to a processor 106 for processing the return signal and determining the optimal wake-up time, and the processor is coupled to an alarm 107 for generating a wake-up signal.

The combined data from detected reflections of the radar signal from the user and the environment are used to confirm that a user is taking a nap, and that the user is not intending to sleep for a longer period of time. If the user is not in bed but in a chair or at a desk, it is more likely that the user intends to take a nap than if the user is in bed. In addition, other factors can be taken into account such as the input data from the user (i.e. time constraints), and time of day. The time of the day can be used because during daytime the user is more likely to take a nap than at night time. At night time, the device may be disabled if the device is switched on continuously otherwise.

The device may also be activated by the user only when they intend to take a nap during the day, while otherwise the device is not in an active state. When the user intends to take a nap, he or she activates the device, and enters as input data of any time constraints. Input data from the user such as a scheduled meeting can be used as well to confirm that a user is taking a nap, as well as wake a user up in time before the meeting. A computer algorithm is used to decide the optimal time to wake the user up, depending on multiple inputs, whilst adhering to any time-constraints input by the user.

In a particular example, the user activates the device and sets a time-constraint. For example, the user activates the device after lunch at 2pm and sets a time constraint of a latest wake-up time of 3 pm. The device subsequently detects the user with a pattern recognition algorithm, and infers biometric data such as breathing rate, movement, and heart-rate. The onset of the nap is determined as the point where one or more of the biometric data changes sufficiently when compared to the biometric data immediately after activating the device. The biometric data can also be compared to historic data of the user entered prior to using the device, as factors like resting heart rate and movement vary from person to person. Once the breathing rate decreases beyond a certain point, and the heart rate slows down compared to the user's resting heart rate, for example, the device determines that the user has fallen asleep. The algorithm will then determine the optimal wake-up time delay from the onset of sleep. If the user selects an option of short sleep, the alarm will optimally wake the user up after 20 to 30 min from the onset of sleep, unless the time-constraint requires an earlier wake-up time. The algorithm is arranged to (directly or indirectly) avoid waking the user up in an unfavourable stage of sleep, such as stage 3 non-REM sleep, to prevent the user from feeling drowsy after taking a nap. The multiple inputs can be weighted in the algorithm.

By way of illustration, some specific scenarios are provided, although the invention is not limited to these scenarios. A user input of a time constraint is received by the algorithm, indicating the maximum time the user wants to nap for.

In scenario A, the received time constraint is 40 min. The detected time to fall asleep is 15 min, timed from the moment the algorithm is activated and the time constraint is set. The algorithm determines that the optimal wake-up time is 25-30 min. The algorithm aims to wake the user up just before entering stage 3 NREM sleep. The user's movement and respiratory rate is detected, and in this scenario the onset of sleep is detected, but apart from light sleep no further sleep stages are detected because it is the aim to avoid any further sleep stages.

In scenario B, the received time constraint is also 40 min, but the user falls asleep after min. The optimal wakeup time is now determined to be 20 to 25 min. User fell asleep quickly, so they would likely want to be woken up with plenty of time to spare before their appointment.

In scenario C, the received time constraint is 100 min, and the user falls asleep after 20 min. The wake-up time is determined to be 80-90 min. The wake-up time is determined to be after the end of the first full sleep cycle to avoid waking during stage 3 non-REM sleep. The algorithm is able to detect stage 3 non-REM sleep based on the returned radar signal, and the method avoids waking up during the stage 3 non-REM sleep.

However, some users may find that waking up at this stage of the sleep cycle makes them still feel tired as they are being awoken during a sleep transition period. As a result, the algorithm may use feedback from the specific user and wake the user up after just 40-45 minutes in this scenario. The 40-45 minutes are calculated as 20 minutes it took them to fall asleep plus the algorithm's estimate of the time up to the moment the user is expected to enter deep sleep.

In scenario D, the received time constraint is 100 min, and the user falls asleep after 5 min. The wake-up signal is sent after 90 to 100 min. The wake-up time is just after the first full sleep cycle, allowing the user to complete a whole sleep cycle.

Fig. 2 illustrates a schematic timeline of an algorithm to implement scenario A or B, whereby at time T1 the onset of sleep is detected by the radar signal and algorithm. Time delay TDelta is the estimate of the optimal time for the user to wake up, in view of the user specified constraint indicated by the dashed line. The dashed line is before a subsequent stage of N3-REM at time T2 and a later stage of REM onset. Given that the user constraint is set prior to the (expected) onset of N3-REM sleep, it is not necessary for the system to detect this stage of sleep. The actual wake-up signal may be generated before the dashed line, or at the dashed line, but not after the dashed line. Each of the times has some uncertainty associated with it, as indicated by the Gaussian curves around the lines T1, T2 and REM onset. The uncertainty arises from person-to-person variations in sleep-cycles, as well as algorithmic estimation error. The transition between sleep stages is also gradual, so a precise determination of the onset of a sleep stage is neither possible nor required. An accuracy of 2 to 3 min will be sufficient.

Fig. 3 illustrates a sequence of method steps of the algorithm. The order of some of the steps, in particular 301 to 304 can be varied, and steps 301 to 304 can be carried out independent of each other, or as alternatives to each other, and can be provided in parallel to the estimation step. Reference is made to subsystems, which are described in more detail with reference to Fig. 4 after the description of Fig. 3. At step 301, a vital signs detection subsystem provides a breathing rate and movement data, whereby noise is removed. Optionally, heart rate can also be detected and provided as an input. At step 302, a nap feedback subsystem provides previously entered user feedback from wake-onset, together with an optional estimate of the user's circadian rhythm. At step 303, an environment classifier provides a current time of the day and information about the surrounding environment, such as the object the user rests in or the room. An algorithm based on a trained neural network is most suitable for recognising objects, and the skilled person will be able to select the appropriate algorithm. Other algorithms that use factors such as range, noise and movement may also be suitable. At step 304, a nap feedback system provides user specified constraints, as discussed previously. At step 305, a linear or non-linear function of the input of step 301 is used to determine the onset of sleep. At step 306, the output of step 305 is combined with the outputs of steps 302, 303 and 304 to calculate with a linear or non-linear function the optimal wake-up time. The final step 307 is the generating of a wake-up signal at the estimated optimal time. In case of a short nap, this will be before the onset of N3-REM, as illustrated in Fig. 2.

The example of Fig. 3 makes use of linear or non-linear functions of inputs in order to calculate an optimal nap wake-up time. The function can be a conditional statement, for example. Alternatively, a more complicated function such as a learning loop can be used, whereby the learning is based on user feedback whether the wake-up time was too early, too late, or correct. Four example functions are provided for the optimal nap wake-up time inference.

Algorithm 008a: Nap Onset Reference Timing Algorithm The nap onset reference timing algorithm is suitable for predicting the optimal wake-up time in the case of a short nap (e.g. scenario A or B). The method involves precisely estimating the time of nap onset, and subsequently estimating an optimal wake-up time from nap-onset that aims to wake up the user just before they would enter N3-REM sleep. As external input, the inferred nap-onset time, user-input time constraints, and optionally prior user feedback and the user's environment are required.

Fig. 4a illustrates a learning loop that provides a nonlinear function that is used to calculate an optimal wake-up time from a nap using the nap onset reference timing algorithm. The denoised breathing rate and movement data, V(t), is normalised over a historical period of prior data to form a continuous-time z-score (or equivalent), V'(t). The z-score provides an indication of deviation of breathing rate and movement data from historical normality. Other statistical measures of deviation or deviation from standardised distributions may be used. This time-deviation is input into a deterministic or statistical model (for example, the logistic regression model depicted in Fig. 4a) to estimate the nap onset, Tl. In the case of a statistical model, the model parameters, b, are learned from historical data. In the case of a deterministic model, the model parameters are directly input from curated factors.

An optimal delay, TDelta, from nap-onset is estimated as a function of: the user's input constraints, C; the surrounding environment, E; and the prior feedback of a user, F. The optimal wake-up-time estimate, TOpt* , is thus obtained by adding the optimal delay estimate to the nap-onset time. The user's time-constraint is viewed as a hard-limit by the algorithm; there is no circumstance whereby this is exceeded. The detected environment's contribution (i.e. time of day and surrounding environment) may increase or decrease the estimated optimal time delay. For example, if the time is near midday and the surrounding environment is an office, this will reduce the estimated time as far as possible in order to aim to reduce the likelihood of an abrupt wakeup as well as reduce the impact of mid-afternoon drowsiness.

The user-specific feedback, F, may also be a contributing factor to the optimal wakeup time calculation. For example, if a user gives feedback to the device that they felt drowsy after being woken up, the device will re-calibrate its estimation to attempt to adjust for this feedback, further optimising the wakeup time. Online methods can be used to provide a continuous feedback loop, such as Bayesian Optimisation.

Bayesian Optimisation uses a probabilistic model to capture uncertainty about the unknown function that connects the wake-up time to the user's level of wakefulness. It then makes decisions by balancing exploration (waking the user at different times to learn more about their reaction) and exploitation (choosing the wake-up time that is currently estimated to be optimal). This ensures an ongoing adaptation of the system to the user's specific needs, which becomes more accurate and reliable over time as more feedback is collected.

The algorithm thus provides a way to dynamically and intelligently calculate optimal wake-up times for naps based on multiple factors including historical data, user constraints, environmental factors, and user feedback, thereby optimising the nap benefits and minimising potential negative effects like post-nap grogginess. Ultimately, the aim is to maximise the user's alertness and overall wellbeing by finding the most suitable wake-up time from a nap.

The Nap Onset Reference Timing Algorithm is focused on estimating nap onset to calculate the optimal wake-up time. This mechanism, coupled with the consideration of user constraints and environmental factors, forms a uniquely targeted approach for short nap scenarios. When integrated with the optional feedback loop, it can offer increased adaptability and personalisation for each user's specific nap patterns.

Algorithm 008b: N3-REM Reference Timing Algorithm The N3-REM reference timing algorithm is suitable for predicting the optimal wake-up time in the case of a short nap (e.g. scenario A or B), as well as a long nap (e.g. scenario D). The method involves continually estimating the time to N3-REM onset, and subsequently waking the user up just before this onset. The inferred nap-onset time, user-input time constraints, continual measurements of movement and breathing rate, and optionally the user's environment and feedback are utilised to make this prediction.

Fig. 4b illustrates a learning loop that provides a nonlinear function that is used to calculate an optimal wake-up time from a nap using the nap onset reference timing algorithm. The nap-onset time T1 is determined in an identical manner to algorithm 008a.

A separate statistical model estimates whether the time to N3-REM onset, TN3, is less than a static threshold S (e.g. 1 minute). It does so by outputting a probability p(TN3 < 5). The statistical model's parameters may be learned from historical data.

Logistic regression models are depicted in this case, however there are other suitable probabilistic statistical models that may be used. Deterministic models may also be used, that aim to directly calculate this probability from measured values.

The model's output probability, p(TN3 < 6), is expected to increase as N3-REM nears.

In the case of a short nap, the user is woken up when the model's probability is greater than a dynamic threshold, w.The dynamic threshold is a function of: the user's input constraints, C; the surrounding environment, E; and the prior feedback of a user, F. For example, if the user time-constraint is strict, this will have the effect of lowering the dynamic threshold such that the algorithm does not risk having to fall-back to waking the user up after the maximum time set by the user. The dynamic threshold is also affected by the surrounding environment and time of day. For instance, if the environmental condition is an office setting and the time is near midday, this would lead to a lowering of the dynamic threshold to decrease the probability of abrupt awakenings and mitigate potential mid-afternoon drowsiness. User feedback may also be incorporated similarly to the outline in algorithm 008a. The first instance whereby the model's probability is greater than this dynamic threshold is the estimate of the optimal wake-up time, denoted by TOpt*.

In the scenario of a long nap, the user is awakened when the statistical model's probability exceeds a dynamic threshold or after the completion of a full sleep cycle, whichever is earlier. The dynamic threshold here is also a function of time constraints set by the user, the surrounding environment, and any previous user feedback. If the calculated wake-up time extends beyond the user-set maximum time limit or is inconsistent with the user's historical sleep patterns, the system will default to a pre-determined fallback wake-up time. This fallback time is tailored based on the users preferred wake-up time, their historical average wake-up time, or any other user-defined parameter.

The N3-REM Reference Timing Algorithm introduces a novel concept of estimating the time to N3-REM onset and waking the user just prior to this stage. The ability to adapt to both short and long nap scenarios with an intelligent dynamic threshold, responsive to time-constraints, environmental factors and user feedback, makes this algorithm uniquely suited for personalised optimal nap timing.

Algorithm 008c: Multimodal Nap-Phase Inference Timing Algorithm: The Multimodal Nap-Phase Inference Timing Algorithm enhances the accuracy and personalisation of nap timing by integrating a comprehensive set of physiological parameters, inferred nap-onset time, user-input time constraints, and environmental factors. It is distinctive in its ability to infer not just the nap onset but also the current phase of the nap. Based on these inferences and the consideration of the other inputs, the algorithm determines the optimal wake-up time, aiming to wake the user during lighter nap phases when they are likely to feel more refreshed.

This algorithm starts by estimating the nap onset, similar to Algorithms 008a and 008b.

However, it extends the data inputs by incorporating heart rate, functions of movement, and functions of breathing rate. These physiological parameters are denoised and normalised over a historical period.

Fig. 4c illustrates a learning loop that provides a nonlinear function that is used to calculate an optimal wake-up time from a nap using the multimodal nap-phase Inference timing algorithm. The nap-onset time T1 is determined in an identical manner to algorithm 008a.

The z-scores of the breathing rate and movement data, V'(t), are calculated identically to algorithm 008a. The inferred heart rate signal from the radar h(t), is combined with the previously defined signals in a nonlinear deterministic model f(r, Tl, h, E). This model directly estimates the phase of the nap at a given point in time, P", as a function of these parameters. For example, a user is likely to be in stage N3-REM when the heart rate is relatively low, breathing rate is regular and movement is limited or non-existent.

A second nonlinear function w(P*, C, E) estimates the optimal wakeup time T Opt* based on the estimated nap-phase, environment, and time constraints. This function aims to wake the user up before the onset of N3-REM (but taking into account the user's time constraints as in algorithm 008b), and so involves a thresholding function. This function analyses rate of change of the probability of a phase output by the previously defined model f.

A distinct feature of this algorithm is its ability to infer the current nap phase based on the multimodal physiological data. By identifying transitions between nap phases, the algorithm is able to determine the optimal wake-up time, aiming to wake the user during lighter nap phases when they are likely to feel more refreshed.

The optimal wake-up time from the nap is then determined by a function of the user's input constraints, environmental conditions, physiological data, and inferred nap phase.

As with the previous algorithms, the user's constraints are considered as a strict limit.

Environmental factors such as the time of day and location also influence the optimal wake-up time. The resulting prediction is thus tailored to the user's unique nap patterns and needs, thereby maximising the benefits of their nap.

Unique Contribution: The Multimodal Nap-Phase Inference Timing Algorithm stands out in its integration of a comprehensive set of physiological parameters, resulting in a more accurate and personalised estimation of the nap phase. By linking this estimation to the calculation of the optimal wake-up time, this algorithm maximises the benefits of the nap by aiming to wake up the user during lighter nap phases.

Algorithm 008d: Machine-Learning Direct Nap Wakeup Algorithm The Machine-Learning Direct Nap Wakeup Algorithm uses a nap-specific, data-driven approach to determining optimal nap wakeup times. It uses specialised machine learning models, such as recurrent neural networks or convolutional neural networks, that have demonstrated effectiveness in handling time-series data like movement, heart rate, and breathing rate.

These models are trained on a combination of user-input constraints, environmental factors, and physiological parameters, particularly emphasising granular historical movement data. The algorithm distinguishes itself by its focus on movement data, acknowledging the role it plays in distinguishing nap patterns from extended, overnight sleep behaviours.

The model inputs, including movement data, are denoised and normalised over a historical period to create robust datasets. Further considerations include the users environmental context and constraints, such as the nap setting and timing, and any specified user constraints regarding wakeup times. The multi-dimensional feature space formed by these variables is used to train the machine learning model. The model is customised according to the quality and quantity of available data and computational resources.

Fig. 4d illustrates this architecture, whereby a machine-learning model is first trained on data points containing historical data, input at the time of nap onset T1. Each data point contains the current (normalised) movement and breathing data v, as well as historical measurements of the normalised movement and breathing data v_i. Each data point also contains information about the current environment, E, as well as the user's time constraints, C, which is as defined in previous algorithms.

During model training, for i training data points, the model is directly provided with optimal wakeup times TOpti. During the model inference stage, TOpt* is predicted directly by the model, where the current normalised readings are input at the time of nap onset.

Differing from conventional sleep-staging techniques, this algorithm directly predicts optimal wakeup times TOpt from nap onset, instead of determining nap stages and using these as a basis for wakeup time predictions. The algorithm learns to predict these wakeup times based on the observed data and the user's feedback regarding their alertness and well-being after a nap. This methodology allows the algorithm to learn nuanced patterns specific to napping, delivering more personalised and accurate predictions through a nap-specific target variable.

The Machine-Learning Direct Nap Wakeup Algorithm incorporates a feedback loop for ongoing learning and refinement of its model. As such, the algorithm improves over time at predicting wakeup times that align with the user's needs and preferences, based on the accumulation of more feedback and data.

By focusing on direct prediction and particularly highlighting the significance of movement data, this algorithm stands distinctively apart from traditional sleep-staging models, optimising the benefits of napping for the user. Its adaptability to evolving user patterns ensures its continued relevance and effectiveness in enhancing the users alertness and overall well-being.

Unique Contribution: The Machine-Learning Direct Nap Wakeup Algorithm provides a novel approach to nap timing by utilising machine learning models to directly predict optimal wakeup times, rather than through sleep staging. The emphasis on movement data and the ability to adapt and learn from user feedback make this a uniquely personalised approach to nap optimisation.

Nap Optimisation System: Fig. 4 illustrates a nap optimisation system. The system can be implemented as a single device, whereby all method steps are carried out by that device. Alternatively, the system can be a distributed system comprising a network of devices. Although the radar sensor will be a device local to the user, the algorithm can also be implemented in a remote device or in a cloud-based system. The method can also be implemented in a mobile user device, such as a smart phone which has radar functionality, and the user interface can be a local application downloaded for the specific purpose.

The system comprises a radar sensor 2, which is functionally coupled to the nap state detection system including a vital signs detection subsystem 4 and an environment classifier 5. The radar sensor carries out the function of transmitting and receiving a radio signal for contactless measurement of the user's breathing rate and movement, and potentially heart rate, depending on the sensitivity of the radar system. The sensor operates by using electromagnetic waves to detect objects and collect information about their position, distance, angle, speed, and absolute and relative movement. The radar itself only collects a signal, and the breathing rate and movement data are extracted from this signal using signal processing functions or machine learning algorithms. The nap optimisation system can thus monitor the user's physiological data without physical contact, ensuring a comfortable and non-intrusive experience.

The radar sensor 2 can utilise frequency-modulated continuous-wave (FMCW) or pulsed radar technologies, both of which have their own advantages and trade-offs. The FMCW radar system or sensor emits a continuous and usually repeating frequency-modulated signal which is typically a sawtooth or a triangular frequency sweep. Moving or stationary objects reflect this incident signal which is then received by the FMCW radar and converted into an intermediate (IF-signal) signal. By processing the IF signal (e.g. extraction and analysis of amplitude, phase or frequency components), ranging and other information about objects can be derived. Generally, a FMCW system may offer a high signal-to-noise ratio (SNR) and dynamic range (DR), making it suitable for detecting subtle changes in breathing rate and movement. On the other hand, pulsed radar systems or sensors transmit short electromagnetic pulses (single pulse or pulse train) and receive signals that are reflected by stationary and moving objects. Similar to the FMCW radar, the received signal is converted into an IF or baseband signal. Because of the short duration of the pulses, often stretch processing techniques are required to process the received signals and improve SNR. In many cases, pulsed radar sensors may offer a lower dynamic range and worse SNR.

Two example radar operating frequency bands are located (approximately) at 24 GHz and 60 GHz radar sensors. The first example is a 24 GHz radar sensor which operates at a lower frequency and bandwidth and has a longer wavelength compared to the 60 GHz radar. It typically provides a longer detection range as it is less sensitive to attenuation caused by atmospheric conditions, such as humidity or rain, and better penetrates certain materials. However, because of the smaller available bandwidth it may have a lower range resolution than the 60 GHz radar, which could affect the accuracy of breathing rate, heart rate, and movement measurements. The smaller available bandwidth at systems using 24 GHz when compared to 60 GHz causes the lower range resolution, more than the different centre frequency. The 24 GHz radar sensor is generally less expensive and consumes less power, making it more suitable for cost-sensitive applications.

The 60 GHz radar sensor operates at a higher frequency and has a shorter wavelength, potentially providing a higher range resolution, and smaller antenna sizes when compared to the 24 GHz example. This type of radar sensor is particularly effective at detecting small movements and changes in the user's breathing rate. However, it may have a shorter detection range and be more sensitive to atmospheric attenuation compared to the 24 GHz radar sensor. The 60 GHz radar sensor typically has a higher cost and power consumption, but its higher resolution may be more appropriate for applications requiring high precision.

The nap optimisation system comprises a user input interface 7 arranged to enable users to enter data based on their specific needs and preferences. The user input interface is functionally coupled to the nap state detection module 3 and the nap wake-up-time algorithm 8. The user input interface allows the user to set the maximum time available for a nap. This functionality enables users to tailor their napping duration to fit their time schedule and constraints, ensuring that their rest period is optimised within the available time frame.

Users can manually input the maximum time they have available for a nap using the interface. This input can be as a touch screen, a dial, or buttons on the device, depending on the design parameters and dimensions of the nap optimisation device. By providing a maximum nap duration, users can ensure that the device's algorithms optimise the nap experience, in particular the wake-up time, within the set time constraints.

The user input interface offers an automatic mode that adjusts the maximum nap time based on the user's prior nap data, personal preferences, and environmental factors. The automatic mode uses the nap state detection and nap wake up time algorithms to determine an optimal nap duration that considers the user's circadian rhythm, sleep history, and other factors, such as ambient noise or temperature. This feature provides a more personalised and adaptive napping experience for users who prefer not to set a specific nap duration manually.

The nap state detection module 3 is arranged to monitor and process the user's physiological data and environmental conditions input by the radar sensor 2. The nap state detection module 3 comprises three subsystems: the vital signs detection subsystem 4, the environment classifier 5, and the nap feedback subsystem 6. The subsystems may be separate systems at a functional level, while all being implemented on the same device and processor, or they may be different physical devices.

The vital signs detection subsystem 4 uses the output from the radar sensor 2 to estimate the users vital signs, including breathing rate, movement, and optionally heart-rate during the nap. The collected data is processed and analysed to identify the user's current sleep state and passed on to the nap wake up time algorithm to determine the most appropriate time for waking up within the set maximum nap time.

The environment classifier 5 is arranged to assess the user's immediate surroundings to detect the type of resting environment, such as a bed, desk, sofa, or reclining chair. This information is used to adapt the nap optimisation algorithms according to the specific environment, enabling the device to better predict the user's nap states and provide a more personalised and adaptive napping experience.

The nap feedback subsystem 6 is arranged to provide the user with feedback based on their napping experience. This feedback can include recommendations for improving future naps, such as adjusting the nap duration, environmental conditions, or nap timing based on the user's sleep data and preferences. The feedback is delivered to the user through an appropriate interface, such as a mobile app, a web-based dashboard, or the device's display.

The three subsystems of the nap state detection system work together to monitor and process relevant information received both from the user and the radar sensor and send output data to the nap optimisation algorithm and providing feedback to users.

The nap wake-up-time algorithm 8 is arranged to determine the optimal wake-up time for users during a nap, based on input from the nap state detection system 3. Unlike sleep staging or smart sleep algorithms tailored for night-time sleep, the nap wake-up time algorithm is adapted to determine the wake-up time based on the unique characteristics of naps.

The algorithm analyses data obtained from the vital signs detection subsystem 4 and the environment classifier 5, taking into account the user's sleep state, physiological data, and the specific resting environment. By processing this information, the algorithm predicts the optimal wake-up time to ensure that the user wakes up feeling refreshed and alert. The algorithm has been described above. Specific and alternative implementations of suitable algorithms illustrated as insets 8a to 8d include: a nap-onset reference timing algorithm 8a arranged to determine the time of sleep onset; an N3-REM reference timing algorithm 8b, arranged to determine the time of onset of N3-REM sleep; a multimodal nap-phase inference algorithm 8c arranged to infer the physiological state of the user and changes to that state based on the detected breathing, movement and optionally heart rate; and finally a machine-learning direct nap wakeup algorithm 8d, which is an algorithm for directly predicting the wake-up time based on previous use of the system, the nap state detection system's signals, and feedback received from the user. An algorithm based on a combination of one or more of algorithms 8a to 8d may also be used.

An output interface 9 is provided for presenting information and feedback to the user.-The interface may be a display to present information like the constraints entered by the user, and basic information like an on/off state. An integrated display on the nap optimisation device can provide users with direct access to information, including the optimal wake-up time determined by the smart alarm, recommendations for nap improvements, and historical napping data. The display can also visually indicate the approaching wake-up time.

The output interface may also include the alarm functionality discussed previously. An audible alarm will be a different structural aspect of the output interface than the display, but functionally they are grouped together. The alarm may be an audible alarm, a tactile alarm, or a visual alarm. A tactile alarm would be beneficial for users with impaired hearing, but would require a physical contact with the user's body, for example by way of a vibrating pad or vibrations of a smartphone, while otherwise the system is a non-invasive system based on the radar detection. A visual alarm may be a bright light source arranged to generate a series of bright light flashes arranged to alert a user even when they have their eyes closed, but this is less practical than an audible alarm.

A dedicated mobile app may be provided, which is compatible with smartphones or tablets, to allows users to access their napping information, feedback, and smart alarm settings through a familiar and portable platform. The app can send notifications, vibrations, or gentle audio cues to rouse the user at the optimal wake-up time. The wake-up function implemented in the smartphone can be provided in addition to, or instead of, an alarm function at the system.

The device can provide smart alarm functionality through built-in speakers or a compatible audio output, using gradually increasing volume levels, soothing tones, or nature sounds to gently awaken the user at the most appropriate time, as determined by the nap wake up time algorithm.

The Output Interface serves as the communication link between the nap optimisation device and the user, delivering information, feedback, and smart alarm functionality in an accessible and user-friendly manner.

The system further comprises a storage 10. The storage is a computer memory for saving data such as user specific settings and software for running the algorithm. The system further comprises a processor arranged to run execute instructions for running the algorithms, classifier and determination steps described previously.

Fig 5 illustrates the main steps of the method described above, comprising: S1, emitting a radar signal; S2, detecting a reflection of the radar signal; S3, sending a signal from the radar module to a processor; S4 determining an optimal time delay from onset of sleep for sending a wakeup signal.

Although the invention has been described in terms of preferred embodiments as set forth above, it should be understood that these embodiments are illustrative only and that the claims are not limited to those embodiments. Those skilled in the art will be able to make modifications and alternatives in view of the disclosure which are contemplated as falling within the scope of the appended claims. Each feature disclosed or illustrated in the present specification may be incorporated in the invention, whether alone or in any appropriate combination with any other feature disclosed or illustrated herein.

Claims

CLAIMS: 1. A method of monitoring a nap of a user, the method comprising: emitting a radio detection and ranging, radar, signal with a radar module detecting reflection of the radio signal from the user with the radar module, sending a signal from the radar module to a processor; at the processor, determining an optimal time delay from onset of the nap until sending a wake-up signal to an alarm based on the detected reflection from the user, wherein the optimal time delay from the onset of the nap is before or after a sleep stage.
2. The method of claim 1, wherein the determining an optimal time delay is based on determining onset of sleep of the user based on the detected reflection of the radio signal.
3. The method of claim 1, wherein said sleep stage comprises: Stage 3 non-rapid eye movement, NREM-3 stage, or rapid eye movement, REM, sleep, or deep sleep.
4. The method of any one of the preceding claims, wherein said nap is day-time sleep, and/or wherein said nap comprises a period of sleep with a duration of less than 3 hours.
5. The method of any one of the preceding claims, wherein the determining an optimal time delay based on the detected reflection from the user comprises analysing the detected reflection from the user to determine one or more of: respiratory rate, heart rate and movement.
6. The method of any one of the preceding claims, further comprising detecting reflection of the radio signal from the environment of the user with the radar module.
7. The method of claim 6, wherein the determining an optimal time delay based on the detected reflection from the environment comprises: determining with a trained machine learning algorithm an object in which the user is placed, and optionally determining whether the object is a bed.
8. The method of any one of the preceding claims, further comprising receiving an input from the user of a time constraint for sending the wakeup signal.
9. The method of any one of the preceding claims, wherein the determining of the optimal time delay is further based on the time of the day.
10. The method of any one of the preceding claims, further comprising activating the alarm to wake up the user on receiving the wakeup signal at the alarm.
11. The method of any one of the preceding claims, wherein the alarm is one or more of: an audible alarm, a tactile alarm or a visual alarm.
12. The method of any one of the preceding claims, further comprising receiving feedback from the user after waking up, using the feedback in a learning loop to adapt the optimal time delay for subsequent use.
13. The method of claim 1, wherein the determining an optimal time delay comprises: determining breathing rate and movement data based on said detecting, determining a deviation of the data from prior data by normalising the determined breathing rate and movement data over prior breathing rate and movement data, inputting the data into a deterministic or statistical model, optionally a logistic regression model, to determine the time of nap onset, determining said optimal time delay depending on the user's input constraint, a determined surrounding environment, and prior feedback from the user.
14. The method of claim 1, wherein the determining an optimal time delay comprises: determining breathing rate and movement data based on said detecting, determining a deviation of the data from prior data by normalising the determined breathing rate and movement data over prior breathing rate and movement data, inputting the data into a deterministic or statistical model, optionally a logistic regression model, to determine the time of nap onset, determining time to N3REM onset, determining a probability that the time to N3REM onset is below a static threshold; and determining whether that probability is greater than a dynamic threshold, whereby the dynamic threshold is a function of the user's input constraint, the surrounding environment, and prior feedback from the user; and when the probability is greater than the dynamic threshold sending a wakeup signal to an alarm.
15. The method of claim 1, wherein the determining an optimal time delay comprises: determining breathing rate, heart rate and movement data based on said detecting, determining a deviation of the data from prior data by normalising the determined breathing rate, heart rate and movement data over prior data, inputting the data into a deterministic or statistical model, optionally a logistic regression model, to determine the time of nap onset, determining said optimal time delay depending on the user's input constraint, a determined surrounding environment, phase of the nap, and prior feedback from the user.
16. The method of claim 1, wherein the determining an optimal time delay comprises: using a trained machine-learning model to determine the optimal time delay, wherein the machine-learning model has been trained based on prior data comprising: movement, breathing rate, the user's environment and user constraints.
17. A non-invasive system suitable for monitoring and controlling a nap of a user, the system comprising: a radio detection and ranging, radar, module arranged to emit a radio signal and detect reflection of the radio signal from the user, and further to detect reflection of the radio signal by an environment of the user; a processor arranged to receive an output signal from the radar module, wherein the processor is arranged to determine an optimal time delay from onset of the nap until sending a wake-up signal to an alarm based on the detected reflection from the user, wherein the optimal time delay from the onset of the nap is before or after a sleep stage.
18. The system of claim 17, wherein said sleep stage comprises: Stage 3 non-rapid eye movement, Stage NREM-3, or rapid eye movement, REM, sleep, or deep sleep, and wherein said nap is day-time sleep, and/or wherein said nap comprises a period of sleep with a duration of less than 3 hours.
19. The system of claim 17 or 18, wherein the frequency of the radio signal is 24GHz or 60GHz.
20. The system of any one of claims 17 to 19, wherein the radio signal is a frequency-modulated continuous electromagnetic wave, or a pulsed electromagnetic wave.
21. The system of any one of claims 17 to 20, wherein the system is further arranged to analyse the detected reflection from the user to determine one or more of: respiratory rate, heart rate and movement, for determining the optimal time delay based on the detected reflection from the user comprises.
22. The system of any one of claims 17 to 21, further arranged to detect reflection of the radio signal from the environment of the user with the radar module, and optionally further arranged to determine with a trained machine learning algorithm an object in which the user is placed, and optionally determining whether the object is a bed.
23. The system of any one of claims 17 to 22, further comprising a user input interface, the interface suitable for receiving an input from the user of a time constraint for sending the wakeup signal.
24. The system of any one of claims 17 to 23, further comprising an alarm for waking up the user, wherein the alarm is optionally one or more of: an audible alarm, a tactile alarm or a visual alarm.
25. The system of any one of claims 17 to 24, further comprising an output interface for providing information to a user.