Project Details

Sleep is characterized by a cyclical alternation of 5 phases divided into two macro-moments:

• NRem phase (Non rapid eye movement) or quiet sleep;
• Rem phase (Rapid eye movement) or active sleep.

After falling asleep, one progressively transitions from the awake phase to the following ones until arriving, after about 60-90 minutes from falling asleep, to the REM phase, which lasts about 15 minutes.

NRem phase. It is the first macro-phase of sleep, it lasts about 75% of total sleep, in which there are no rapid eye movements. In turn, it consists of 4 stages lasting from a minimum of 5 minutes to a maximum of 15 minutes. In the transition from one of these periods to another, sleep gradually deepens:

• Fall asleep: this is the phase that gradually allows the body to pass from the awake state to sleep and is characterized by the lowering of the body temperature, the partial relaxation of the muscles and the slowing of the heartbeat. Furthermore, eye movements are not rapid, a sign that brain activity slowly decreases.
• Light sleep: heart rate continues to slow down, thus preparing the body to enter the actual sleep stage in which eye movement is almost completely absent. We are therefore completely asleep, the muscles are completely relaxed and the breathing is very deep.
• Deep sleep: eye movement is very slow and body temperature drops further. In this phase, the body restores its metabolic reserves by regenerating itself. If you were awakened at this time, you would feel a feeling of confusion and disorientation. This stage is characterized by brain waves typical of dreamless sleep and are closely linked to involuntary / unconscious activities of the organism (heartbeat, digestion) and are responsible for deeper relaxation.

Rem phase. In this phase, brain activity awakens with an alternation of waves (Theta, Alpha and Beta), while the eyes begin to move rapidly. Dreams occur in this stage of active sleep, accompanied by a gradual but sensitive increase in blood flow, breathing and brain activity. In fact, from the monitoring of subjects in the rem phase, the electroencephalogram shows how the brain is as active as in a waking state, so much so that oxygen and glucose are consumed as if one were awake. Furthermore, one of the characteristics of this phase is that the muscles of the legs and arms are paralyzed. This phenomenon of muscular “atony” has the function of preventing rash and unpredictable movements, possibly generated by the dream.

Rem and NRem phases alternate about 4-6 times during the night for a total duration of an hour and a half, resulting in the so-called sleep cycles. In addition to an adequate duration, in order for sleep to be regenerating and restful, this cyclical structure characterized by the alternation of phases must be respected.

Limits

Compared to what has been said, however, the limits of this tool must be considered. In fact, sleep is a complex phenomenon that requires the skills of specialized figures (specialists in Sleep Medicine) and the carrying out of Polysomnography. The latter, specifically, is the only exam capable of recognizing the macro-structure of sleep through:

• the study of brain activity (EEG);
• eye movements;
• muscle tone;
• EKG (electrocardiography);
• breathing and respiratory noise (for the analysis of OSAS and sleep apnea);
• O2 saturation;
• limbs movement and body position.

Our smart alarm device implements an algorithm essentially based on limbs movements, according to research presented in many papers, like [1].

The smart alarm device (also called Central Device or Alarm Clock) is powered by an Arduino Nano BLE which includes a BLE (Bluetooth Low Energy) 5.0 module. Its task is to receive information on the user's sleep status and to sound the alarm at the correct time. If the alarm clock fails to connect to the peripheral device, it will implement the basic functionality of a standard alarm clock.

Users can set the alarm time through the user interface, consisting of the 4x4 keypad. In addition, some ceramic capacitors were added on the row pins to delete external interferences, which had brought unpredicted behaviors to the keys during the testing phase. As shown in the figure below, by pressing the C key, the user can set the deadline (aka target time) for the alarm to be played: if this time is reached, the user will be woken up, no matter the sleep stage he’s in. When the alarm is ringing, the user can turn it off using the D button. Furthermore, the user can turn ON or OFF the alarm functionality with an A button press and change the current time by clicking the B button; this option can be very useful when the time shifts from legal to solar and vice versa.

The 20x4 LCD display normally shows current date, time and notifies if the alarm time is set (when it is, a bell symbol is shown). As the user navigates in the menus, the LCD also displays the status information related to the specific operation. There’s also a Real Time Clock (RTC), which gives the time data using an internal oscillator and a 3V battery to keep the RTC running when the general system is turned off.

The speaker plays a custom alarm using an offline made note-to-frequency conversion via software; moreover, an amplifier is needed in this operation, because the output current from the Arduino nano pins is not high enough. For this reason, we introduced an amplification stage composed of one NPN bipolar transistor in common emitter configuration and the bias resistors.

The external case is made with recycled wood panels and the internal layout is optimized and kept tidy using custom-length wires (e.g., the ethernet cable used for the keypad connections). To obtain the night sleep trend and to plot the measurements, an additional feature was added: the data can be conveyed to the serial output and processed by the external Matlab script.

The wearable device is powered by an Arduino Nano 33 IoT board and a 9V rechargeable battery, both housed in a 3D-printed plastic case. The Nano 33 IoT board includes the Inertial Measurement Unit (IMU) and the BLE module, which enables wireless data transmission from the wearable to the Alarm Clock.

The IMU module provides instant acceleration values in 3-axis dimension (x,y,z), composing a 3D vector. As explained in the following sections, these acceleration values are locally processed through arithmetical operations (here, one of the main targets is measurement noise reduction [1]), then they are transmitted to the Central Device. To notify the user's sleep state, a flag called "SNREM" is also sent to the alarm clock.

An HR sensor (MAX30102) is present in the schematics below, as it was involved in early test phases; check section Future works for more details on the matter.

Sleep stage recognition is possible through a simple evaluation of the md value:

\(T_{max}\) and \(T_{min}\) being the maximum and the minimum threshold. If the md value is between these two, then a non-REM stage is detected.

Threshold computation. The two thresholds are determined as follows:

• \(T_{max}\,\): it is set as the maximum md value calculated inside a time window. It is important to note that whenever a \(\text{md}>T_{max}\) is evaluated, this \(\text{md}\) value instantly becomes the new \(T_{max}\) value.
• \(T_{min}\,\): it is the minimum Tmax value met across all time windows.

The complete algorithm is shown in the Figure: the Tmax_history array is used to store all \(T_{max}\) values, to find the minimum between them and assign it to \(T_{min}\).

The central device constantly compares the current time to the alarm time set by the user (aka set time), in order to enter into the “alarm window” when necessary. In this window, the alarm is played as the first SNREM value is received by the central device. If no SNREM value is collected, the window ends with the alarm time set by the user: at that moment, the user must be awakened even if he is in a REM sleep state. The tests were conducted activating the alarm window 15 mins or 30 mins before the set time.

Results

Light sleep detection. Since the acceleration-based algorithm can only detect light-sleep phases (SNREM), only those are considered when evaluating the experimental results of the Smart Alarm device. The outcomes are in line with those of popular sleep-stage detecting apps available on mobile stores, the majority of which perform a microphone-based analysis. These comparisons are shown in the following pictures: the continous line depicts the sleep trend according to a popular mobile app, while the diamonds point out the time instants in which Non-Rem statuses (SNREM, i.e. light sleep stages) where detected by the Smart Alarm.

Alarm function. As a consequence of the reliable light sleep detection results, it can be noticed that the patient is awakened as a non-Rem status is observed in the last 30 minutes before the set time.

Future works

To further improve the light sleep phase recognition capabilities and extend it to full sleep stage detection, heart rate (HR) data can be added to the algorithm. During the early phases of development, the implementation of an HR sensor was experimented, but it was ultimately left out due to poor measure reliability. Indeed, sudden drops in accuracy were exposed when the HR sensor was subjected to long test cycles and motion. Since sleep periods usually last more than a few hours, the feature was dropped as final decision.

As shown in the picture below, the HR results in short time periods are acceptable. When one HR sample is acquired by the sensor, it is stored in the appropriate array. As said array is filled up, an ascending order sorting algorithm is performed, then the median is computed. In earlier tests the mean value was determined, though this solution didn’t provide enough protection from outliers and measure inconstancy.