This dataset contains cardiovascular data recorded during progressive exsanguination in a porcine model of hemorrhage. Both wearable and catheter-based sensors were used to capture cardiovascular function; the wearable system contained a fusion of ECG, SCG, and PPG sensors while the catheter-based system was comprised of pressure catheters in the aortic arch, femoral artery, and right and left atria via a Swan-Ganz catheter.


Experimental Protocol

This protocol included 6 Yorkshire swine (3 castrated male, 3 female, Age: 114–-150 days, Weight: 51.5-–71.4 kg), each of which passed a health assessment examination but were not subject to other exclusion criteria. Anesthesia was induced in the animal with xylazine and telazol and maintained with inhaled isoflurane during mechanical ventilation. Intravenous heparin was administered as needed to prevent coagulation of blood during the protocol. Before the induction of hypovolemia, a blood sample was taken to assess baseline plasma absorption. Following this baseline sample, Evans Blue dye was administered for blood volume estimation. After waiting several minutes to allow for even distribution of the dye, a second blood sample was taken to measure plasma volume. In this method, plasma volume is used along with hematocrit to estimate total blood volume. For one animal in the protocol (Pig 4), atropine was administered to raise the starting heart rate and blood pressure due to critically low values.

Hypovolemia was induced by draining blood through an arterial line at four levels of blood volume loss (7%, 14%, 21%, and 28%) as determined by the estimated total blood volume from the Evans Blue dye protocol. After draining passively through the arterial line, the blood was stored in a sterile container. Following each level of blood loss, exsanguination was paused for approximately 5-10 minutes to allow the cardiovascular system to stabilize. If cardiovascular collapse occurred once a level was reached, as defined by a 20% drop in mean aortic pressure from baseline after stabilization, exsanguination was terminated. Note that cardiovascular collapse was reached at different blood volume levels for each animal: Pigs 1, 3, and 4 reached 21% blood volume loss; Pigs 2 and 6 reached 28% blood volume loss; and Pig 5 reached 14% blood volume loss before the experimental protocol was terminated.


Signals from wearable sensors were continuously recorded using a BIOPAC MP160 data acquisition system (BIOPAC Systems, Inc., Goleta, California, USA) with a sampling frequency of 2 kHz. Electrocardiogram (ECG) signals were captured using a three-lead system of adhesive-backed Ag/AgCl electrodes placed in Einthoven Lead II configuration, which interfaced with a BIOPAC ECG100C amplifier. Reflectance-mode photoplethysmogram (PPG) was captured with a BIOPAC TSD270A transreflectance transducer, which interfaced with a BIOPAC OXY200 veterinary pulse oximeter. The transducer was placed over the femoral artery on either the right or left caudal limb, contralateral to inducer placement. Seismocardiogram (SCG) signals were captured using an ADXL354 accelerometer (Analog Devices, Inc., Norwood, Massachusetts, USA) placed on the mid-sternum, interfacing with a BIOPAC HLT100C transducer interface module.

Aortic root pressure was captured by inserting a fluid-filled catheter through a vascular introducer in the right carotid artery, fed through to the aortic root. Femoral artery pressure was obtained directly from an introducer placed on either the left or right femoral artery depending on accessibility. Right and left atrial pressures were captured with a Swan-Ganz catheter with proximal and distal monitoring ports inserted in either the right or left femoral vein. Left atrial pressure was inferred via PCWP captured using an Edwards 131F7 Swan-Ganz catheter (Edwards Lifesciences Corp, Irvine, California, USA). The vascular introducers were connected via pressure monitoring lines to ADInstruments MLT0670 pressure transducers (ADInstruments Inc., Colorado Springs, Colorado, USA). Data from the catheters were continuously recorded with an ADInstruments Powerlab 8/35 acquisition system sampling at 2 kHz.


Signal Pre-Processing

All signals were filtered with finite impulse response band-pass filters with Kaiser window, both in the forward and reverse directions to offset phase shift. Cutoff frequencies were 0.5–-40Hz for ECG and 1-40Hz for SCG. Only the dorso-ventral component of the SCG acceleration signal was used in this study. PPG signals, along with all four catheter-based pressure signals, were filtered with cutoffs at 0.5-10Hz. After filtering, data from all signals were heartbeat-separated using ECG R-peaks. The signal segments were then abbreviated to a length of 1,000 samples (500 ms) to enable more uniform analysis; however, due to the long left ventricular ejection time of Pig 3, a length of 1,500 samples (750 ms) is provided for this subject.


Using the Dataset

This dataset contains a separate .mat file for each of the 6 animal subjects in the protocol. The variables "scg" and "ppg" contain R-peak-separated signals from the SCG and PPG respectively during the protocol. The variables "aortic", "femoral", "rightAtrium", and "wedge" contain the R-peak-separated pressure waveforms from the catheters placed in the aortic root, femoral artery, right atrium, and left atrium (wedge pressure) respectivley. Each of these variables is a struct, with each of its fields representing a different level of blood volume loss. The field "B1" corresponds to the baseline level (pre-exsanguination); "L1", "L2", "L3", and "L4" correspond to the 7%, 14%, 21%, and 28% drop in blood volume respectively. Thus, the data in each field represents the heartbeat-separated signals collected during each blood volume level. The data has been selected such that periods of active draining of blood have been removed, such that the provided data reflects the heartbeat-separated signals during the resting period between blood-draws. The data is formatted in columnwise matrices, with the columns arranged in sequention order such that the first column is the first heartbeat and the last row is the last heartbeat.


The indices of ECG R-peaks are provided as a vector as well during each blood volume level, such that each element in the vector corresponds to its respective column in the provided column matrices. The unit of these values is in miliseconds, staring from t = 0 (onset of baseline recording).


Data are collected before and after percutaneous transluminal angiography (PTA) for dialysis patients.

Each sample is labeled as a-b-before.wav or a-b-after.wav and the associated txt, where a is the patient id and b is the location id.

The first position was the arterial-venous junction,  and the second point was 3 cm from the first position along the vein.

 The distances between the adjacent positions were also about 3 cm.



Each voice sample is stored as a .WAV file, which is then pre-processed for acoustic analysis using the specan function from the WarbleR R package. Specan measures 22 acoustic parameters on acoustic signals for which the start and end times are provided.

The output from the pre-processed WAV files were saved into a CSV file, containing 3168 rows and 21 columns (20 columns for each feature and one label column for the classification of male or female).


[17-APR-2020: WE ARE STILL UPLOADING THE DATASET, PLEASE WAIT UNTIL IT IS COMPLETED] -The dataset comprises a set of 11 different actions performed by 17 subjects that is created for multimodal fall detection. Five types of falls and six daily activities were considered in the experiment. Data collection comes from five wearable sensors, one brainwave helmet sensor, six infrared sensors around the room and two RGB-cameras. Three attempts per action were recorded. The dataset contains raw signals as well as three windowing-based feature sets.


We will upload the instructions in the following days.


This dataset is a collection of brainwave EEG signals from eight subjects. The data is collected in a lab controlled environment under a specific visualization experiment. 


Synergistic prostheses enable the coordinated movement of the human-prosthetic arm, as required by activities of daily living. This is achieved by coupling the motion of the prosthesis to the human command, such as residual limb movement in motion-based interfaces. Previous studies demonstrated that developing human-prosthetic synergies in joint-space must consider individual motor behaviour and the intended task to be performed, requiring personalisation and task calibration.


Task-space synergy comparison data-set for the experiments performed in 2019-2020.


  • Processed: Processed data from MATLAB in ".mat" format. Organised by session and subject.
  • Raw: Raw time-series data gathered from sensors in ".csv" format. Each file represents a trial where a subject performed a reaching task. Organised by subject, modality and session. Anonymised subject information is included in a ".json" file.
    • Columns of the time-series files represent the different data gathered.
    • Rows of the time-series files represent the values at the given time "t".
  • Scripts: MATLAB scripts used to process and plot data. See ProcessAndUpdateSubjectData for data processing steps.

Ear-EEG recording collects brain signals from electrodes placed in the ear canal. Compared with existing scalp-EEG,  ear-EEG is more wearable and user-comfortable compared with existing scalp-EEG.


** Please note that this is under construction, and instruction is still being updated **




6 adults ( 2 males/ 4 females, age:22-28) participated in this experiment. The subjects were first given information about the study and then signed an informed consent form. The study was approved by the ethics committee at the  City  University of  Hong  Kong(Reference number:  2-25-201602_01).


Hardware and Software

We recorded the scalp-EEG using the a Neuroscan Quick Cap (Model C190) . Ear-EEG were recorded simultaneously with scalp-EEG. The 8 ear electrodes placed at the front and back ear canal (labeled as xF,  xB), and two upper and bottom positions in the concha (labeled as xOU and xOD). All ear and scalp electrodes were referenced to a  scalp  REF electrode.  The scalp  GRD  electrode  was  used as a  ground reference. The signals were sampled at 1000 Hz then filtered with a  bandpass filter between  0.5  Hz and  100  Hz together with a  notch filter to suppress the line noises.  The recording amplifier was SynAmps2,  and  Curry  7  was used for real-time data monitoring and collecting.


Experimental design

Subjects were seated  in front of a computer monitor. A fixation cross presented in  the center of  the monitor for 3s, followed by an arrow pseudo-randomly pointing to  the  right  or  left for 4s. During  the  4  s  arrow presentation, subjects needed to imagine and grasp the left or right hand according  to  the arrow direction. A short  warning beep was played  2  s  after the cross onset to call the subjects. 


Data Records

The data and the metadata from 6 subjects are stored in the IEEE Dataport. Note that Subject 1-4 completed 10 blocks of trials, subject 6 finished  only  5  blocks.  Each  block contained 16  trials.  In our dataset, each folder contain individual dataset from one subject.  For each individual dataset, there were four type of files (.dat, .rs3, .ceo, .dap). All four files were needed for EEGLAB and MNE package processing.  Each individual dataset contains the raw EEG data from 122 channels (from scale EEG recording), 8 channels (from ear EEG recording), and 1 channels (REF electrode). 

Individual dataset of subject 1,5,6 has different sub-datasets. The index indicates the time order of that sub-dataset (motor1, then followed by motor2, motor3, motor 4 etc).  While Individual dataset of subject 2,3,4 has one main dataset.

Each dataset has timestamps for epoch extraction. Two event labels marked the start of the arrow, which indicated the start of subject hand grasping (event number 1: left hand; event number 2: right hand).


One subject, five different movements, four levels of motor imagery data.The sampling rate is 25Hz, a total of 33,000 lines.


The CLAS (Cognitive Load, Affect and Stress) dataset was conceived as a freelyaccessible repository which is purposely developed to support research on the automated assessment of certain states of mind and the emotional condition of a person.


1.      Database structure (

The database is organized in 4 folders:

·         Answers – answers of the questions in the interactive tasks (Math problems, Logic problems and the Stroop test) for each person.

·         Block_details – metadata for each block (1 block per task) for every participant.

·         Data – raw signal recordings for the individual participants.

·         Documentation – accompanying documents.


When using the CLAS dataset, please cite:

     Markova, V., Ganchev, T., Kalinkov, K. (2019). CLAS: A Database for Cognitive Load, Affect and Stress Recognition, in Proceedings of the International Conference on Biomedical Innovations and Applications, (BIA-2019), art. no. 8967457,  DOI: 10.1109/BIA48344.2019.8967457. Available on-line:


This dataset provides the ECG signals recorded in ambulatory (moving) conditions of subjects. The ambulatory ECG (A-ECG) data acquired with two different recorders viz. Biopac MP36 Acquisition system and a self-developed wearable ECG recorder are made available. Total 10 subjects' (with avg. age of 27 years, 1 female and 9 males) ECG signals with four body movements- Left & Right arm up/down, Sitting down & standing up and Waist twist are uploaded.

An EEG signals dataset is also provided here.


Please contact me at: for how to use the dataset and further discussion.