Full text
“Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.” Materials In this study, we employed one of Tolife’s [1] daily-life monitoring devices: the Smart Mattress Cover System (SMCS), a compact, non-invasive, and scalable solution for real-world sleep monitoring. The SMCS comprises two components: the Smart Mattress Cover (SMC) and the Bedroom Box Hub (BBH). Here, we present the dataset only related to the SMC’s validation. The SMC is a compact textile-based mattress cover comprising (Fig. 1): (i) a pressure matrix (PM) sensor, (ii) two 3D accelerometers (ACCs), (iii) a microcontroller Unit (MCU) with a custom front-end. The SMC is placed in the thoracic subject’s area: on the mattress, under the pillow line, beneath the bedsheet (Fig. 1a). Its raw data enables the extraction of heart rate (HR), breathing rate (BR), body movements and bed occupancy status. Specifically, BR can be estimated from PM data by tracking mattress pressure variations induced by chest volume changes during respiration; HR can be derived from ACC data via ballistocardiographic analysis of heart-induced vibrations; body movements and bed occupancy can be detected using both PM and ACC signals. The design of the SMC is detailed in our previous work [2]. The pressure matrix (PM) consists of a custom, compact, low-density textile-based pressure sensor array comprising 40 sensing elements (SE), arranged in a 4×10 grid as shown in Fig. 1b. The SE raw data are sampled at 8 Hz. The ACCs integrated into the SMC are LSM6DSL units (STMicroelectronics, Switzerland), positioned at the edges of the PM and referred to as 𝐴𝐶𝐶1 and 𝐴𝐶𝐶2 (Fig. 1a, c). ACC raw data are sampled at 80 Hz, with axis orientation illustrated in Fig. 1c. Reference measurements for HR and BR were obtained using the Shimmer3 ECG unit (Shimmer Research, Ireland), worn on the chest with a 4-lead thoracic band configuration. Regarding movement (MOV) reference, we relied on the timestamps recorded within the experimental protocol, therefore it is not provided. (a) (b) (c) Figure 1: (a) The Smart Mattress Cover (SMC) includes: (i) the pressure matrix (PM, green), (ii) two 3D accelerometers (ACCs, orange) positioned at PM’s edge, and (iii) a microcontroller unit (MCU, red). SMC’s bed positioning: over the mattress, under the bed sheet, near the pillow, aligned with the subject’s thorax. (b) The pressure matrix (PM) sensing elements (SE) arranged as a 4x10 grid and their associated indexes. (c) The ACCs axis orientation: x, y, z respectively the subject’s longitudinal, transverse, and sagittal axis.
2 “Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.” Acquisition protocol We conducted experimental protocols with 11 healthy participants (aged 23–58 years; weight: 52–106 kg; height: 163–191 cm), each completing three dedicated protocols targeting: (i) HR, (ii) BR, and (iii) movement and bed occupancy detection (MOV). The HR protocol was designed to estimate HR under still conditions, simulating sleep scenarios; the BR protocol included various respiratory patterns—slow, normal, and fast; the MOV protocol aimed to discriminate among off-bed (OFF), on-bed static (S), and on-bed movement (M) phases, with movements performed at different frequencies (slow, medium, fast). In all protocols, participants were instructed to lie in three randomly ordered positions: supine, prone, and lateral (randomly assigned to the left or right side for each trial). (i) HR protocol: Each position was recorded in three independent trials, with participants lying still for 5 minutes in each trial. (ii) BR protocol: All positions were acquired sequentially within a single trial, with 30-second transitions between them. For each position, the breathing sequence included: 1 minute of normal breathing, 2 minutes of slow breathing, 1 minute of normal breathing, 30 seconds of fast breathing, and a final 1 minute of normal breathing. No specific respiratory rate was imposed; participants were instructed to interpret the breathing pace naturally. (iii) MOV protocol: All positions were acquired within a single continuous trial. Participants alternated between OFF, S and M across five main phases. In P1, the subject stood beside the bed for 30 seconds. In P2, they lay on the bed in a supine position. Phase P3 consisted of a structured sequence of transitions: supine (S), movement (M), lateral (S), movement (M), prone (S), and movement back to supine (M). This was followed by P4 (bed exit) and P5 (30 seconds standing). Phase P3 was repeated three times with varying durations for S and M: (30 s, 10 s), (15 s, 7 s), and (5 s, 5 s), and each repetition was performed twice. See Fig. 2 for a detailed overview. Figure 2: On/off bed status and movement (MOV) protocol description: structured in five phases (P1–P5), subjects alternated between on-bed static (S) and on-bed movement (M) periods. Phase P3 included three repetitions, each performed twice with varying durations of (S, M) intervals to simulate slow, medium, and fast transitions.
3 “Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.” Folders structure The dataset is organized in two main folders: one containing data acquired from the SMC, and the other containing reference measurements. Within the reference folder, two subfolders are provided: HR and BR. A detailed overview of the folder structure is shown in Fig. 3. The HR folder includes 11 subfolders, one per participant. Each participant’s folder contains three subfolders (P1, P2, P3), corresponding to the three postural conditions used in the HR protocol. Each posture folder includes the reference signals: breath.csv for breathing and ecg.csv for cardiac activity. The sequence of postures followed by each subject is reported in the file position_HR.csv, where P1, P2, and P3 indicate the posture associated with each trial. Similarly, the BR folder includes 11 subfolders (one per participant), each containing the corresponding breath.csv and ecg.csv files. The file position_BR.csv provides the posture sequence adopted during the BR protocol, where P1, P2, and P3 indicate the first, second, and third phases of the protocol, respectively. For MOV, we only provide the posture sequence adopted during the protocol in the file position_MOV.csv, where P1, P2, and P3 indicate respectively the first, second, and third posture performed. Figure 3: Reference folder structure. The SMC folder contains three subfolders: HR, BR, and MOV. The detailed structure is illustrated in Fig. 4. The HR folder includes 11 subfolders, each corresponding to a participant. Within each participant’s folder, three subfolders (P1, P2, P3) represent the different postural conditions. Each of these contains the raw data files: acc1.csv and acc2.csv respectively for 𝐴𝐶𝐶1 and 𝐴𝐶𝐶2, and mat.csv for the PM data. The BR and MOV folders follow the same structure: each includes 11 participant-specific subfolders, each containing the corresponding acc1.csv, acc2.csv, and mat.csv files.
4 “Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.” Figure 4: Smart Mattress Cover (SMC) folder structure. Reference data – HR File name: ecg.csv Feature Name Units Timestamp ms Ecg mV Reference data – BR File name: breath.csv Feature Name Units Timestamp ms Breath adim
5 “Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.” SMC data – 𝑨𝑪𝑪𝟏,𝑨𝑪𝑪𝟐 File name: acc_1.csv, acc_2.csv Feature Name Units Timestamp ms x mg y mg z mg SMC data – PM File name: mat.csv Feature Name Units Description Timestamp ms Time unit from microcontroller. A1 LSB The 𝐴𝑖 with i = 1:40, correspond to the 40 sensing elements of the pressure matrix (PM) described above and indexed in Fig. 1b. Each 𝐴𝑖 contains raw data sampled by the 12-bit analog-to-digital converter (ADC) used. A2 LSB … ... A40 LSB REFERENCES [1] «TOLIFE | AI and smart sensing for COPD | Horizon Europe project», Tolife Project. Consultato: 20 febbraio 2024. [Online]. Disponibile su: https://www.tolife-project.eu/ [2] C. Marinai, L. Arcarisi, F. Bossi, P. Bufano, F. D. Rienzo, E. Melissa, G. Rho, M. Zanoletti, A. Greco, M. Laurino, C. Vallati, N. Carbonaro, and A. Tognetti, “Smart mattress cover for unobtrusive monitoring of sleep-quality correlates in real-life,” 2024, IEEE Sensor Conference.
