1. Introduction
Evolution in the field of Digital signal processing over the years have made an absolute impact on biomedical research and technology [2]. Biomedical signals in their fundamental form are a potential source of information [2]. Meaningful information can be extracted using biomedical signal processing [3], widely utilised for the purpose of diagnoses, observe and treat diseases and abnormalities [2]. This signal processing technology have made it possible for scientists and engineers to create a system that can be controlled using brain signals, this communication system is called brain-computer interface [4]. With the understanding of brain function, the advent of BCI technology and by acknowledging the needs and potentials of the physically challenged (Wolpaw et al., 2002). BCI researchers have focused on developing control and communication technology for people with a severe motor disorder such as myasthenia gravis, spinal muscular atrophy, and amputees (Wolpaw et al., 2002). BCI detects the pattern of the ongoing brain activity associated with user intention and translates it into a meaningful control command [6]. The intent of the user is determined using electrophysiological signals such as EEG, EMG, and ECOG. [1]. Electroencephalogram (EEG) invented by German psychiatrist Hans Berger is commonly used technique. Brain signals contain different frequencies such as beta, alpha (mu), theta, delta, and gamma waves which are related to various activities and activates part of the brain integrated with the particular activity. These signals can be recorded using invasive or non-invasive methods of recording brain signals. Signals are converted in real time into commands (Wolpaw et al., 2002) that should be decoded successfully by the machines that could be a robot, a computer or a robotic arm. For designing a BCI system that can decode the user's intention precisely, signals processing is essential. The signals collected are often noisy and indecorous to be used in a BCI system. Digital signal processing techniques elucidate signal making it suitable for a BCI design. A working BCI system has four stages that are: extracting brain signals, the second is pre-processing which refers to the removal of noise from the signal like fluctuation in signal due to eye movement or any other. In Pre-processing features essential depending on a particular application are also extracted, then the signal is classified mostly using algorithms. There are a plethora of classification algorithm and methods available example few most commonly used techniques are support vector machine, adaptive neural network, and neural network. The last stage is the interfacing with the application which could be anything from a computer to a wheelchair to a robotic arm; figure 1 shows the four steps.
This thesis focus on scrutinising the brain recorded (EEG) signals and determine whether a BCI can discriminate between event-related synchronisation (ERS) and desynchronization (ERD) of alpha and beta rhythms during left-right hand movement [1]. The EEG signals for voluntary left-right hand movement were recorded using non-invasive emotive epoc plus headset; the signals were examined and processed using EEGLAB and MATLAB. Hand motion detection is an issue of interest in robotics, computer vision and other applications [10]. [draft]
Fig 1: Brain Computer Interface (Using High-Frequency Electroencephalogram in Visual and Auditory-Based Brain-Computer Interface Designs by Cota Navin Gupta and Ramaswamy Palaniappan-(July- 2012) e-contact! Available at http://econtact.ca/14_2/gupta-palaniappan_interfacedesign.html)
Literature review
Human Brain
Physicist Michio Kaku once said 'human brain has hundred billion neurones and each neurone connected to ten thousand other neurones. Sitting on your shoulder is the most complicated object in the universe' WordPress ann com, https://wordpressannscom.wordpress.com/ (accessed August 31, 2016). Most of the human action involves motor function from walking to gesturing, but even simple action for example movement of the hand or picking up a ball is a complex movement to study [8]. Hence for accurate analysis of the brain signals it is important to have an understanding of the brain [1].
Fig 2: Motor cortex in human brain (image taken from Wikipedia)
Brain area involved in motor function (planning, control, and voluntary movement) is motor cortex; it is in the region of the cerebral cortex [Wikipedia]. Motor cortex works like a system which has various parts (areas of the brain) shown in figure 3 each performing a particular task.
Primary motor cortex
It plays a vital role in motor function as it produces neural impulse that controls the execution of movement [8]. Primary motor cortex constitutes every part of the body as shown in fig 3. Known that left region of the brain controls the right side of the body, and right region/hemisphere controls the left side, signals from primary motor cortex cross the body's midline to activate muscles on opposite site [8].
Fig 3: Motor Homunculus
The figure shows that primary motor cortex constitutes every part of the body somatotopically represented, the tongue is next to face which is next to hand. Hands occupy large of region primary motor cortex which signifies its control over hands [8].
Other areas involved in motor function are:
1. Supplementary Motor Area (SMA)
Its purpose is planning of action and coordination movement including both side of the body, for example, hold a glass with both the hands. It is in front of primary motor cortex and above premotor area [8].
2. Posterior parietal cortex
It converts the visual information into a motor command; it is considered to be involved in a certain facet of motor planning. It sends information to the premotor cortex and supplementary motor area [8].
3. Premotor cortex
In a motor function, it is responsible for the sensory guidance of movement, preparation of action (movement) or spatial guidance of reaching (Wikipedia).
Brainwaves
Billions of neurones firing together in the brain are the genesis of thoughts, action, and behaviour. This firing of neurones together in synchronisation produces brain waves, electrodes placed on the scalp are used to record brain waves; (EEG) Electroencephalogram. BCI can employ profusion of brainwaves depending on the application [11], brain waves classified as:
1. Slow cortical potentials also are known as infra brain waves having a frequency less than 0.5Hz. Slow in speed, hence specialised equipment used in measuring SCP because of their slow nature [12]. SCP plays a critical role in brain timing and function, because of slow nature; not a lot is known about SCP [9]. Figure 4 shows the average of slow cortical potentials in inhabiting and average form.
Fig 4: Slow cortical potential (figure taken from [12])
2. Delta brain waves (0.5 to 3.5 Hz). They are slow in frequency, deeply penetrating that has high amplitude and mostly found in adults in deep meditative state and dreamless sleep [11] and [9].
Fig 5: Delta wave (Wikipedia)
3. Theta brain waves (3.5 to 8Hz) mostly occurs in deep meditation and sleep; it acts as a gateway to learning and memory [9]. In theta waves a person is in a state that is beyond consciousness; vivid imagery and intuition [9].
Fig 6: Theta waves(Wikipedia)
4. Alpha brain waves (8 to 12Hz) often regarded as resting state of the brain; it aids mental coordination, alertness, and mind-body integration [9]. Researchers have found that smoking marijuana rises alpha power in the brain [11].
Fig 7: Alpha brain waves
Beta brain waves (12 to 30) Hz they are dominant in wakes state, decision-making, problem-solving, judgment or any mental task [9]. Beta waves further divided into; low beta 12-15Hz, beta or beta2 (15-22Hz) refers to high engagement or figuring out something and high beta 22-30Hz indicating complex thought or excitement [9].
Fig 8: Beta brain waves
Gamma brain waves range from 30 to 42Hz indicates the processing of information from different areas of the brain, they are the fastest moving waves [9]. They were considered as brain noise but was later found to be present in a state of universal love and is often considered to be related to consciousness and spirituality [9].
Fig 9: Gamma brain waves
Electroencephalogram (EEG)
Many techniques have been invented for measuring brain signals like functional magnetic resonance imaging (fMRI), Functional Near-Infrared Spectroscopy (fNIRS), single-photon emission tomography (SPET) but these techniques are slow, indirect and have a low resolution [11]. Another method is (MEG) magnetoencephalography which works on the principle that every electric charge produces a magnetic field and uses this phenomenon to detect impulses produced by neurones [16]. Its spatial and temporal resolution is excellent.
The most commonly used technique is EEG (electroencephalogram) which measures the electrical activity of the neurones, analysed by metal electrodes attached to the scalp [15] as shown in figure 10. EEG is used to measure brain's electrical activity which can further indicate certain brain disorder [13]:
' seizure disorders like epilepsy.
' Head injury.
' Dementia
' Sleep disorder
Fig 10: Electroencephalogram (http://www.saintlukeshealthsystem.org/health-library/electroencephalogram-eeg)
Signal has mostly low frequencies around 10Hz alpha rhythm and below, its amplitude fluctuates between 5-50mv [14]. This activity of neurones is distributed all over the scalp which is non-stationary [14]. Only a large number of active neurones can generate electrical signals detectable on the head surface, the current produced by the neural activity penetrates through the skin, skull and several other layers where weak signals on the scalp are amplified [15]. EEG is desirable considering its speed; as it can record complex neural activity patterns within seconds a stimulus has been introduced and can measure strength and position of the activity, but its spatial resolution is less as compared to MRI scan [15].
Other methods of recording brain's signals similar to EEG exist like:
' ECOG – it works on the same principle as EEG, but the electrodes are implanted surgically in the brain which enhances its spatial resolution compared to EEG but is more expensive and dangerous [16].
' Inter cortical electrode ' Electrodes of small proportion placed on specific areas of the cortex which acquire signal generated by a group of neurones, this method has a best spatial resolution, but also requires surgery [16].
Brain-computer Interface (BCI)
The brain-computer interface is a system that allows communication between a human and a digital device using the brain's electrical activity (EEG signals) [16]. Antithetical to the traditional input devices which require muscular activity (like a pen, keyboards, mobile phones, etc.) a BCI acquires its input from human brain in the form of waves produced in different areas of the brain, convert that into a command that can control the digital device [11]. To be effective, a BCI must modify itself on the users EEG and must reorganise to short and long-term changes in these characteristics [17]. BCI is a revolutionary technology having many amazing applications especially for disabled people or those affected by motor disability such as:
1. A new technique of playing games using brain and a broad scope of virtual reality.
2. Helping disabled people to interact with computational devices [11].
3. Elevate understanding of human brain and its neural network.
4. Its main application is in prosthetics for example an artificial robotic arm but contributes to other applications such as virtual keyboard, vehicle control [16].
5. It helps to elevate the quality of life for people with motor disabilities, partially or fully paralysed or suffer from locked-in syndrome [16] like a wheelchair that operates using brain signals. Emotion and feeling recognition applications.
It is working principle same as the human brain and central nervous system; having a brain and spinal cord connected. Its primary function is to process and consolidate the input stimuli and give impulse back to the muscles or glands causing automatic and volunteer action [11]. Practical usage of BCI using EEG suffers many challenges in real life such as:
' Signals extraction from the brains is an arduous task which often resulting in weak, noisy signals and requiring amplification [11]. Not all toolkits used contain amplifiers making signals undesirable to use [11].
' Data transfer rate of a BCI is very low; this low data transfer rate makes BCI suffer from the problem of fast response and accurate control [11].
' Low data transfer rate and weak signal increase the possibility of the percentage of error in the signals, and as the signals are highly capricious in nature therefore expected error rate is high [11].
' Signals classified in a BCI sometimes due to inappropriate classification can change the characteristic or quality of the signal, the signal might suffer from interference [11].
For the steady recording of EEG signals from a particular area of the brain international 10/20 system or standard 10/20 system is internationally used to describe the electrode position on the scalp [18] and [11].
A BCI system contains four steps which are signal recording (data acquisition), preprocessing, classification and application interface which can be a computer, wheelchair, neuroprosthetics or robot [11] figure 11.
Fig 11: A functional BCI system [20]
Data Acquisition
Brain signals which are data and can be acquired by some methods, invasive or non-invasive. There are several techniques used in BCI system to extract data which is classified as shown in figure 12.
Fig 12: Types of BCI ( image taken from [11])
Invasive BCI
Here special kind devices are used to record brain signal known as invasive BCI; further divided into single unit acquiring a signal from the single area of brain cells and multi-unit acquiring from multiple areas [11]. These devices have to be implanted surgically in the human brain (figure 12) resulting in the best quality of brain signals, ECOG electrocorticogram as mentioned earlier obtain signals using this method [11]. This method provides good quality signals but is expensive and dangerous as it can damage the tissues in the brain [11].
Fig 13: Invasive BCI technique ( image from Independent frequencies may explain memory recall-science node- 27 February 2013 by Amber Harmon)
Partially Invasive BCI
Another type of BCI is partially invasive in which are placed inside the skull on the brain [11] as shown in figure 14.
Fig 14: partially invasive BCI [11]
Non-invasive BCI
Non-invasive BCI has weaker signals as compared to the invasive or partially invasive methods, but it is a low cost and most safest type BCI [11]. The weaker signals are due to the human skull; signals are extracted using electrodes placed on the scalp [11] using EEG. Some of the new devices have a better temporal resolution as more electrodes are used to extract signals from the brain [11]. Figure 15 shows a wireless EEG headset used in extracting brain signals; other non-invasive BCI are fMRI, SPECT, PET.
Fig 15: Non-Invasive BCI- emotiv headset [19]
Pre-processing
The second stage of BCI is signal processing; the extracted EEG data (raw data) has noise because of several reasons like neck movement, eye movement, or breathing during the EEG recording. Signal undergoes filtering for noise cancellation also at this stage Features in the signal that are of interest are extracted using algorithm; there are various algorithms available for feature extraction ( wavelet transform, common spatial pattern, etc.). Features of interest depend on the final application. This stage makes the signal desirable for classification.
Classification
Next stage in a BCI is a classification of the signal which makes the signal features in an order recognisable by the application [16], classification algorithms used for this purpose like support vector machine or neural network. There are two ways of classification linear and non-linear, but an efficient algorithm should be utilised which can adapt itself to users given a better performance [16].
Application
BCI's critical contribution focused on prosthetics but due to radically new technology it now has two major fields; medical field ( rehabilitation ) and other entertainment [16]. People with motor disabilities and locked in syndromes lose the ability of voluntary movement, and communication becomes utterly impossible for them ( even though they are cognitively well) [16]. BCI application which can use these patient's brain signal and stimulate a keyboard [16] may help them communicate with the world.
A patient's neurophysiological signals can be represented as image sound or vibration using neurofeedback technique; there are used to motivate patients to control brain activity abating diseases like epilepsy and hyperactivity [16].
In 'Composing SMS and Ringtones by thought ' Brain Computer Interface application' by Pranav Bhardwaj, Gaurav Bhateja, Bharati Vidyapeeth a BCI system was designed where text message (SMS) can be typed or a call can be placed using EEG signals.
Signals used in EEG-based BCI
EEG activity can be examined and measured in both time domain and frequency domain, in time domain voltage against time and frequency domain frequency against power or voltage [1]. Hence both can be used for EEG-based BCI (Wolpaw et al., 2002). Studies have shown that controlling certain EEG features can be learned [1] hence it makes EEG-based BCI more desirable [1]. There are five signals commonly used for BCI which are as follows:
1. Visual Evoked Potentials (aka Visual Evoked response)- it refers to electrical potential initiated by light or visual stimulus [21]. These are extracted from the scalp over visual cortex, 'determining the user's eye gaze direction depending on user's ability to control gaze direction' [1]. This indicates that EEG can fetch accurate information about motor output; making it superior to other techniques for accessing gaze direction [1].
Fig 16: VEP [21]
2. P300 potentials- Picton TW in The P300 wave of the human event-related potential-1992 Oct (J Clin Neurophysiol.) said that 'The P300 wave is a positive deflection in the human event-related potential' can be considered as infrequent. A sudden evoke in EEG peak at 300ms, it occurs when a subject is engaged in sleuthing a given target [22] as shown in the figure below. P300 refers to positive peak at 300ms which occurs at parietal lobe [1].
Fig 17: P300 potentials (Neurophysiological functioning of occasional and heavy cannabis users during THC intoxication. -Theunissen EL, Kauert GF, Toennes SW, Moeller MR, Sambeth A, Blanchard MM, Ramaekers JG – Psychopharmacology (Berl.) (2012), Open Access Biomedical Image Search Engine)
3. Event-related potentials (ERP-N400)- They are time-related voltage fluctuations with the physical or mental occurrence, can be measured using EEG by the mean filtering and averaging [23]. Considered as negative deflection 400ms after introducing a stimulus over central parietal [1].
Fig 18: ERP (https://faculty.newpaltz.edu/giordanagrossi/index.php/lab-of-brain-cognition/event-related-potentials/)
4. Mu and Beta Rhythms- 8-14Hz EEG is often displayed in wake people when not engaging in any movement or processing stimulus input over a sensory motor area of the brain [1]. Any movement or even preparation of movement causes desynchronised mu and beta rhythms are known as ERD (event related desynchronises) [24]. When the movement stops then mu and beta rhythms increases which are known as ERS (event related synchronises), ERD and ERS does not require actual movement they occur with imaged movement as well [24]. Mu and Beta rhythms are used in BCI as they are associated with cortical areas directly in connection with brains motor output channels [24]. [draft]
Get an image from original readings
5. Slow Cortical Potentials- These are a low-frequency change in the cortex, these potential changes occur in 0.5-10sec therefore called as a slow cortical potential [1]. Movement and other cortical activation cause negative slow cortical potentials; similarly, positive SCP's is associated with reduced cortical activity [1].
Significance of Mu-Beta rhythms and Event Related Synchronises -Desynchronises
Existing corroboration indicates that Mu rhythms and other alpha rhythms are independent phenomena due to difference in source of generation, bilateral coherence sensitivity to sensory events, frequency, and power [29]. Mu synchronises and desynchronises disturbs the processing of sensory motor in the front parietal part of the brain, they are sensitive to cognitive effect and does not represents an idle brain state [29]. These rhythms are considered to be have dynamical properties, humans can learn to control these rhythms which according to research plays critical part in processing information like perception and transforming 'seeing' and 'hearing' into doing [29].
Mu rhythms also known as Rolandic, sensorimotor, central rhythm and did not receive the kind of attention as other EEG rhythms until recently, it was thought to be infrequent and only existed in small percentage [29]. Mu rhythms are present in most adults, these rhythms occur in EEG from frequencies 8-13Hz and 15-25Hz for briefly for 0.5 to 2 seconds [29]. Where 8-10Hz can be regarded as lower Mu rhythm and 10-12Hz can be considered upper Mu rhythm [1]. Mu and Beta rhythm can be recorded over sensory motor area with peaks around 10Hz to 20Hz [1].
Study of motor movement is of interest (like hand movement in this thesis) as it leads to more interesting study of motor imagery movement that is imagination of movement. During motor imagery or perception of movement (hand movement for example) activates the almost the same cortical area which is involved in actual motor movement (hand movement) [1] refer to figure 21. Both mu and beta rhythms are reactive to imaginary and even observation of movement [1].
Motor movement or preparation of movement causes desynchronises (decrease) of mu and beta rhythms which is known as Event Related Desynchronises (ERD), the degree of mu and beta suppressed during a movement can be expressed the percentage of peak power value at rest which shows an average decay of about 61% (SD =25) [29]. After the movement there is an increase in mu and beta rhythm known as Event Related Synchronises, this inverse relation is believed to be due to spatial scale and frequency band of cortical synchrony [29]. Synchrony that is present in alpha like rhythms are because of activation of small portion of neurones in cortical area which may give rise to EEG signal, while the inactive neurones maintain low metabolism, thus this type of synchrony is associated with decayed brain metabolism [29]. Rhythmic activation of EEG can be induced by external events (ERD), these 10Hz oscillation (10Hz ERD in alpha band [1]) in EEG can be measured in time and space [29]. 10Hz ERD represents information processing, preparation of movement and attentive state of brain. ERD can be recorded 2 seconds before the movement, size and magnitude of ERD refers to the amount (size) of neurones involved in a particular movement or activity [1 and 29]. Magnitude of ERD depends of the task complexity for example movement of hand and movement of finger [1].
Fig 20: Represents ERD and ERS. Figure from: http://cse.iitk.ac.in/users/se367/13/submissions/psinha/project/proposal.html
Fig 21(a): ERD curve recorded over sensory motor cortex, C3 left motor cortex and C4 right motor cortex. Figure from (Pfurtscheller et al., 2000a)
Fig 21(b): Represents ERD and ERS during actual movement at 10-12Hz frequency. Figures 21(a) and (b) represents comparison between ERD and ERS at sensory motor cortex of the brain during imagery movement and actual movement. The two figures show that ERD and ERS induced during the imagery and actual movement have no difference because of which most BCI's use ERD and ERS due to motor imagery and do not require actual movement [1] such BCI's are mostly used in neuro prosthetics. Image taken from [1].
Hand mu and beta rhythms are considered the most important mu rhythm and can be recorded over the sensory motor cortex (on the scalp).
It is interesting that often alpha waves are considered similar to mu waves because Mu rhythms frequency 8-14Hz overlap or matches the frequency of alpha rhythm, but both are different [29]. The Mu rhythm represents processing of sensorimotor in the front parietal part whereas the alpha rhythm shows visual processing in occipital networks of the brain [29]. Alpha shows desynchronises when visual stimulus is introduced or during opening and closing of eyes, while the mu rhythm is desynchronized during actual or motor imagery movement [29]. The frequency of the mu rhythm is high (mean=10.5Hz) in comparison to alpha (mean=9.6Hz) that is what is sometimes referred to as low alpha and high alpha but some other research had shown different values [29].
Methods and Material
Subjects
Four subjects all male participated in the experiment, all the subjects were right-handed with no previous experience of EEG recording. Subjects aged between 22 to 26 years, all were healthy and suffered no brain disorder (example epilepsy or dementia). All subjects were informed about the procedure that their identity will be confidential, regarded as subjects and their EEG data can be shared with researchers outside the university.
Experiment and EEG Recording
There were four sessions for each subject ( participants), each had to perform 3 task in every session. The tasks were right-hand movement, left-hand movement and no movement at all.
Three tasks referred to as Subject 1-R for right-hand movement of subject 1, subject 1-L for left-hand movement and subject 1-N for no movement. During the experiment, subject sat on a chair and was asked to perform hand movements, first the right-hand movement than the left-hand movement and last no movement. Each hand trial lasted approximately 20 seconds recorded using a clock watch where the subjects were instructed to have no movement for the first 5sec, a rapid hand movement from 5sec to 15sec and again no movement in the last 5sec.
Subjects were requested not to perform any other movement such as leg movement or eye movement which could have affected the EEG recording. The time frame of 20 seconds can be divided into three parts as shown in figure 20 below:
Fig 20: Timing of one trial of experiment; Refers to the Mu and Beta rhythms synchronises and desynchronosis with the movement of hand which effects the output signal from the scalp ( sensory motor cortex). Desynchronises in mu and beta rhythms occurs between 5-15secs when the hand movement occurs.
EEG signals were recorded in EDF format with a sampling rate of 128Hz[draft].
Wireless EEG Headset
Emotiv EPOC plus headset is used in recording the brain signals in this project. The headset contains 16 electrodes; 14 of the electrode detects signals and two electrodes are reference electrodes, placed by standard 10/20 system see figure 21(a) and (b).
Fig 21(a) Emotiv EPOC plus [26]
The headset comes with a saline solution to wet the electrodes which are considered better than the traditional sticky gels [https://www.emotiv.com/epoc/] The headset sends the data wirelessly using bluetooth technology and operates at a frequency of 802.11 (2.4GHz) [25]. Emotiv uses software called testbench where EEG signals are displayed in real time and recorded; this software has additional features like 'gyro' which shows a change in signals when performing some activity. 'Expressive' for various facial expressions, 'Affective' for the level of engagement, excitement or meditative and the training of subject like push, pull or rotate there is another functionality called 'cognitive' [25]. This headset has been used in numerous applications; control video games, motorised skateboard research in the field of neuropathy [26].
Fig 21 (b): Two-dimensional plot of the 14 electrodes on the head (brain) produced in EEGLAB
EEGLAB-MATLAB
In this thesis, EEGLAB is used for the purpose of signal processing. EEGLAB is an Open source interactive Matlab toolbox by Swartz Center for Computational Neuroscience (SCCN Lab), used for analysis and processing of electrophysiological signals like EEG, MEG, etc. [27] and [1].
EEGLAB enables reading of data channel locations, data information and importing and exporting of files in different formats: ASCII, EDF (European data format), neuroscan, EGI, BDF and biosemi [28]. EEGLAB offers epoc extraction, baseline removal, data resampling and epoch extraction time locked to specified experimental events from continuous or epoched data [28]. EEGLAB also allows users to remove artifacts, epochs, and channels; it also provides users to plot the scalp map, channels in time-frequency domain [28].
EEGLAB stores all the data related to a particular recording like channel locations, epochs, events, sampling time and acquisition parameter as a single structure or file called EEG dataset [28]. Epochs and Events can be imported in several formats as mentioned above, epoch and events information then can be edited [28], and epochs can be extracted using event information. As ERP (event-related potential) forms an important part of the analysis of electrophysiological signals [28], hence EEGLAB offers several features to study ERP like 'channel ERP image', compare ERP and ERP map series, an extra plugin called ERPLAB can be downloaded into the existing EEGLAB for more features. There are three layers of EEGLAB functions in which the first (top layer) layer allows yours to use a graphic interface to study the data; the second layer allows to customize data processing using history script, and data pop functions [1].
The signals recorded by the emotive headset were in EDF format; which is compatible with the EEGLAB. All the pre-processing that is the removal of artifacts, application of FIR filter and notch filter is done using EEGLAB. Some data processing extensions were added to EEGLAB like AAR (automatic artifact removal) which also offers different methods for artifact removal like LMS regression, RLS regression, but here BSS (Blind Source Separation) was used. [DRAFT]