31
Raman Research Institute, Bangalore SSVEP Based BCI System Krishna Prasad S Visiting Student (June 2013 – February 2014) An Internship report on
R a m a n R e s e a r c h I n s t i t u t e , B a n g a l o r e
SSVEP Based BCI System Krishna Prasad S Visiting Student (June 2013 – February 2014)
08 Fall
An Internship report on
2
Contents
EXECUTIVE SUMMARY ................................................................................................................................................... 3 1. ACKNOWLEDGEMENTS ............................................................................................................................................. 4 2. INTRODUCTION AND OVERVIEW .......................................................................................................................... 5 2.1 WHAT IS BCI? ............................................................................................................................................................................ 5 2.2 WHY BCI? ................................................................................................................................................................................... 5 2.3 INTRODUCTION TO SSVEP ...................................................................................................................................................... 5 2.4 BCI AT RRI ................................................................................................................................................................................. 6
3. AN OVERVIEW OF THE RRI-‐BCI SYSTEM ............................................................................................................. 6 3.1 BLOCK-‐DIAGRAM OF THE COMPLETE SSVEP BASED BCI SYSTEM .................................................................................. 6
4. THE RRI-‐BCI SYSTEM IN DETAIL ........................................................................................................................ 7 4.1 THE ALTERNATE VISUAL INPUT DEVICE (AVID) ................................................................................................................ 7 4.2 THE MINIATURIZED RRI EEG AMPLIFIER (BIO-‐AMP V2) ............................................................................................... 8 4.2.1 Operating Characteristics .............................................................................................................................................. 8 4.2.2 Utility ....................................................................................................................................................................................... 9 4.2.3 Components Used ............................................................................................................................................................... 9 4.2.4 Distinct Circuits present .................................................................................................................................................. 9 4.2.5 Simulation Software and Parameters ....................................................................................................................... 9 4.2.6 Simulation Circuits .......................................................................................................................................................... 10 4.2.7 Simulation Output ........................................................................................................................................................... 14 4.2.8 PCB Schematic and Layout Design ........................................................................................................................... 15 4.2.9 How they look .................................................................................................................................................................... 18 4.2.10 Testing and Results ....................................................................................................................................................... 18 4.3.11 Planned Enhancements .............................................................................................................................................. 19
4.3 DIGITIZER .................................................................................................................................................................................. 19 4.4 SIGNAL PROCESSOR ................................................................................................................................................................. 21 4.4.1 Objective ............................................................................................................................................................................... 21 4.4.2 Theory ................................................................................................................................................................................... 21 4.3.3 Implementing the Matched Filter ............................................................................................................................. 22
5. TAKEAWAYS .......................................................................................................................................................... 24 6. APPENDIX ............................................................................................................................................................... 25 6.1 THE AVID ARDUINO CODE FOR 3 LEDS ............................................................................................................................. 25 6.2 4-‐CHANNEL ARDUINO DIGITIZER ......................................................................................................................................... 27 6.3 MATCHED FILTER PROGRAM ON ARDUINO ......................................................................................................................... 27
3
Executive Summary My internship at Raman Research Institute, Bangalore under the Visiting Students Program was to build a Brain-‐‑Computer Interface (BCI) System using Steady State Visually Evoked Potentials (SSVEP), the details of which are covered in this report. My 9-‐‑month long internship at RRI under the BCI group headed by Dr. HemaRamachandran and Dr. Ramesh Balasubramaniam involved working in some key areas of the project as listed below:
i. Visual Input Device: An ArduinoTM powered hand-‐‑held LED stimulator panel capable of firing 4 independent LEDs at different frequencies.
ii. EEG Amplifier: A 2-‐‑channel EEG amplifier, which was simulated first using NI Multisim 13.0 and then fabricated using Cadence Allegro PCB suite. The EEG Amplifier comprises of a 50Hz Notch Filter, a Low Pass Filter at 200Hz and a Variable Gain Stage in addition to Instrumentation Amplifiers.
iii. Digitizer: A 4-‐‑channel digitizer operating at 256Hz was developed on ArduinoTM Platform.
iv. Signal Processor: A Matched Filter was implemented on Java and then on ArduinoTM (2-‐‑Channels), which now can sample and look for the signal frequency at the same time.
I acquired many new technical skills throughout my work term. The internship was helpful in brushing up my basics in Analog Electronics and Signal Processing. Besides, I got hands-‐‑on experience in fabricating Printed Circuit Boards and populating them with Surface-‐‑Mount devices. Additionally, I got introduced to the area of research and how to approach it. Most importantly, the work experience was very valuable which included good fellowship, cooperative teamwork, accepting responsibilities, how to make and meet deadlines, manage time and most importantly, deal with situations even when desired results are not forthcoming. Additionally, the poster and the talk delivered during the In-‐‑House 2014 event at RRI improved my presentation skills and greatly boosted my confidence. This report concludes with my overall impressions of my work experience.
4
1. Acknowledgements Firstly, I am extremely thankful to my guides: Dr. Hema Ramachandran and Dr. Ramesh Balasubramaniam for providing me an opportunity to contribute and for their valuable guidance. I am extremely grateful to Mrs. Sujatha. S of RAL (and the BCI Team), without whom, I cannot imagine how my internship term at RRI would have been. It was indeed a pleasure working with Mr. James Matthew of LAMP, for his helpful technical inputs and signal processing expertise. I am also thankful to RAL Staff, especially Mr. K.B. Raghavendra Rao, Mr. Somashekar, Mrs. Arasi and Mr. Ibrahim of the RRI Workshop for their constant guidance and support. Also, I am very thankful to my 'ʹgang'ʹ of friends from the A&A and LAMP group for making my stay at RRI and in Bangalore a pleasant and a memorable one.
5
2. Introduction and Overview
2.1 What is BCI? A brain computer interface (BCI) is a new way of communication between humans and computers. It utilizes a very uncommon, but on the other hand probably the most direct way of access to the intentions of a person. The will / intention of the person gets communicated directly to the computer from the origin of the intention itself – the brain. With BCI, a person ideally does not have to make use of the common output pathways of peripheral nerves and muscles. Typically the input to a BCI system uses the electrical brainwaves of the person, measured with the well-‐‑known electroencephalography (EEG), which was primarily used for clinical purposes only in the past. This EEG is then amplified and fed into a computing device, which processes them with appropriate algorithms to give the person a new kind of communication channel.
2.2 Why BCI? The main motivation behind BCI research all over the world is to restore independence and mobility for people with paralysis or limb loss. A BCI-‐‑system provides a completely new output pathway and this is perhaps the only way a person can express herself if he/she suffers from disorders like amyotrophic lateral sclerosis (ALS), brainstem stroke, brain or spinal cord injury or other diseases which impair the function of the common output pathways, responsible for the control of muscles.For a quadriplegic, something as basic as controlling a computer cursor via mental commands would represent a revolutionary improvement in quality of life.
2.3 Introduction to SSVEP At RRI, the aim is to build a BCI system utilizing a neurological phenomenon called steady state visual evoked potential (SSVEP) to control action devices for the disabled, like a wheelchair and a robotic arm. A visually evoked potential (VEP) is an electrical potential-‐‑difference, which can be derived from the scalp after a visual stimulus, for example a flashlight. VEPs after stimuli with a frequency < 3.5 Hz are called “transient” VEPs. If the stimulation frequency is > 3.5 Hz they are called “steady state” VEPs because the individual responses overlap and result in a quasi-‐‑sinusoid oscillation with the same frequency as the stimulus. The goal is to detect this frequency reliably with high accuracy and also to detect when the frequency is not present, thus when the person does not look at the stimulus.
6
2.4 BCI at RRI The motivation behind carrying-‐‑out BCI Research at RRI is to engineer individual components required for building a complete SSVEP based BCI system indigenously so that the cost of an end-‐‑to-‐‑end BCI based system like a Wheelchair comes down drastically, making it more affordable to people of all economic strata.
3. An Overview of the RRI-‐‑BCI System
3.1 Block-‐‑Diagram of the complete SSVEP based BCI System
As shown, in the SSVEP based BCI system being built at RRI the following are being engineered in-‐‑house:
a. RRI Visual Input Device b. RRI Electrodes c. RRI EEG Amplifier d. ArduinoTM based Digitizer & Signal Processor e. RRI Action Devices: Wheelchair and Robotic Arm
RRI Visual Input Device
RRI Cup Electrodes
EEG Amplifier
Digitizer & Signal Processor
RRI Action Devices
Figure 1. The RRI BCI System Block Diagram
7
4. The RRI-‐‑BCI System in Detail My contribution to the RRI BCI system has been mainly in the ArduinoTM based Digitizer and Signal Processor part and in the Development and Testing of the Miniaturized EEG Amplifier (popularly known as BioAmpV2) together with Mrs. Sujatha S of RAL, although we also made a mini, hand-‐‑held, ArduinoTM powered Visual input device (popularly known as the Alternate-‐‑VID). In this section, every component of the system, I have contributed to, is discussed in detail.
4.1 The Alternate Visual Input Device (aVID) Although a full-‐‑fledged Visual Input Device is being made by Mr. Somashekar and his students – Ms. Pavithra, Mr. Nikhil and Mr. Chandan, the necessity for a mini, portable visual input device was inevitable. Prior to the aVID, a cardboard box with a set of 3-‐‑LEDs, powered using a signal generator was used as a stimulus for the SSVEP Experiments. But since it required a dedicated signal generator just for blinking a set of LEDs and since some spurious signal was also detected when the subject came in physical contact with the box, a compact, plastic cased, programmable, visual stimulus was certainly needed. The aVID was designed with the following considerations:
a. No spurious signal pick-‐‑up when the subject comes in contact with the stimulator b. To accommodate more than 1 LED blinking at different frequencies simultaneously c. To enable SSVEP testing with small although fixed distance of separation between
the LEDs d. Ability to change the blinking frequencies of the LEDs
The aVID was then designed, within just a day by Mrs. Sujatha and I, on a general-‐‑purpose circuit board with four LEDs at top, bottom, left and right respectively as shown. This board was designed to be an add-‐‑on shield that would get fixed atop an ArduinoTM Uno board. Thus the presence of a microcontroller makes it programmable to any desired frequency. Later, with Mr. Ibrahim’s help, the Shield and the controller was covered with a plastic case to avoid spurious signal pick-‐‑up.
(a) (b) (c)
Figure 2. The 3D model of the aVID Shield
a. Front View b. Right-‐‑side view c. Left-‐‑side view
8
The pin numbers to which the LEDs are connected are displayed in Figure 2(a). Programmable Timers are used to get the LEDs blinking at the desired frequency. The ArduinoTM code for the Visual Input device is included in the Appendix of this report. The fabricated aVID looks like this:
4.2 The Miniaturized RRI EEG Amplifier (Bio-‐‑Amp V2) The Human brain is enclosed within a highly protected cage -‐‑ the Skull and the signals that are tapped non-‐‑invasively (from the scalp) are highly attenuated, which explains the need for an EEG amplifier. This amplifier is responsible for amplifying such faint brain signals collected non-‐‑invasively from electrodes attached to the subject'ʹs scalp, preserving signals of frequencies within the operating range while rejecting unwanted frequencies. Following the successful fabrication and testing of the Bio-‐‑Amp V1 by Mrs. Sujatha and integration with the 50Hz Notch and the High Pass Filter board by Mr. Sashi Raj (Visiting Student: April 2013 – August 2013), it was decided to integrate and miniaturize these separate entities into just one PCB with SMD components. Following this decision, the work for fabrication the miniaturized EEG Amplifier began with simulation using National Instruments Multisim v13.0.
4.2.1 Operating Characteristics Operating Frequency Range 1 -‐‑ 125 Hz Gain 7000 -‐‑ 15000 Notch 50 Hz Input Voltage Levels 1-‐‑100µμV Output Voltage Levels Typically, 0.5-‐‑2.0 V (peak to peak) DC Offset in output ~ 2.5 V
Figure 3. The aVID
9
4.2.2 Utility The amplified output of the brain signals will be subjected to various analyses working on digital logic, which require signals as data (bits). This conversion will be possible with the help of Analog to Digital Converter (ADC) circuits. The RRI EEG Amplifier aims at amplifying the brain signals to levels usable by an ADC module, implemented on ArduinoTM Platform.
4.2.3 Components Used S.No Component Description Package Quantity 1 INA114 Instrumentation Amplifier SOL-‐‑16 2 Nos. 2 LM7805C Voltage Regulator (5V) SOT-‐‑223 1 No. 3 1N5819 Schottky Barrier Rectifier SOD-‐‑123 4 AP358 Low Power Dual Op-‐‑Amps SOP-‐‑8L 5 BAV199 Dual Diode (Low Leakage) SOT-‐‑23 6 PGB1010603 ESD Suppressor 0603 7 TMV 0505S DC-‐‑DC Converter 1 No. 8 3.5mm Jack Input / Output Connectors SMD 7 Nos.
4.2.4 Distinct Circuits present S.No Circuit Remarks No. present 1 Power Supply Circuit Consumes 9V DC from Battery and supplies
5V, 2.5V and GND terminals 1 No.
2 Instrumentation Amplifier Circuit
Input from 3.5MM jack is supplied to Instrumentation amplifier present in each channel to improve CMRR and Gain by a factor of 12.36
1 per channel
3 50Hz Notch Output of Instrumentation Amp is stripped of 50Hz hum. Gain is Unity.
1 per channel
4 Low Pass Filter (1-‐‑125 Hz)
Rejects signals of Frequencies > 125Hz; Unity Gain
1 per channel
5 Gain Stage Improves the Gain of the incoming signal to a final variable factor of 7000 -‐‑ 15000, making it usable for Analog-‐‑Digital Converter modules
1 per channel
4.2.5 Simulation Software and Parameters National Instruments MultiSim v13.0 (Trial Version) was used for simulating the Bio-‐‑signal Amplifier Circuit. AC Analysis was carried out and the frequency response was obtained to satisfy the requirements. Throughout this simulation, TLC277 Dual Op-‐‑Amp is used in place
10
of AP358. The simulation is performed for a single channel. Throughout the document, the X and Y axes in the output graphs are in logarithmic and decibel scales respectively.
4.2.6 Simulation Circuits A. Power Supply Circuit
The power supply circuit consists of a 5V input, which is assumed to have come from a 9V Battery / Voltage Regulator / DC-‐‑DC Converter circuit. The output of the power supply circuit provides three terminals: Vcc, VGND and AGND. AGND is Actual Ground or negative terminal of the battery. VGND is a virtual/lifted ground that caters as the 0V level for all components in the board. Since Op-‐‑Amps in the board require a negative power supply, AGND acts as a negative supply source (-‐‑Vcc) while VGND acts as 0V source (GND).
11
B. Instrumentation Amplifier Stage The input (differential mode) signals from electrodes are sent to the Instrumentation Amplifier for Common Mode noise reduction and amplification and through Dual Diodes and ESD suppressors. The instrumentation amplifier chip INA114 amplifies the input by a factor of 12.36. The output of this stage is fed to the 50Hz notch filter for eliminating the 50Hz hum seen in the signals because of power lines.
As seen in the response below, the output signal (red) is 12.36 times the input, represented in Decibel scale ( 20 log (12.36) = 21.8 ). The red cursor indicates the gain at -‐‑3dB point, which is 18.8 dB at 3.5 kHz.
12
C. 50Hz Notch filter As the name goes, the notch filter rejects 50Hz waves from the signal. The value 50 Hz is determined by the RC Combination. The gain is Unity.
The plot below shows the response of the notch filter. At 50 Hz, the attenuation is 30 dB. The -‐‑3dB bandwidth is approximately 12 Hz (FH -‐‑ FL) where FL is 40 Hz and FH is 62 Hz.
13
D. Low Pass Filter The Unity Gain Low Pass filter is a 3-‐‑stage active filter intended to cut-‐‑off at 125Hz.
The -‐‑3 dB point of the low pass filter, as shown below, is situated at 70 Hz and it continues to attenuate the signals at frequencies beyond 70Hz.
E. Gain Stage The gain stage improves the gain of the input signal by a factor of 103(Vo/Vin) to make it usable by ADC stages that may follow the Bio-‐‑Amp. Typically the output signals have a peak-‐‑to-‐‑peak magnitude of 0.5 to 2V. The R&C values determine the characteristics of the outputwaveform.
10k 8.2µμ
14
The gain of this stage as seen in the circuit is variable. The plot below shows VOUT = 55 dB, which is 1000 times the input as seen in the circuit ( 20 log (10M/10.1k) ).
4.2.7 Simulation Output The output of GainStage1 (Vo/Vin), the final stage in the Bio-‐‑Amp is plotted on y-‐‑axis with Frequency on x-‐‑axis, in logarthmic scale. As shown, the gain begins to roll-‐‑off completely after 125Hz and is also notched at 50Hz. The gain in the operating range is shown as 9500, which is actually variable from 7000 to 15000.
15
4.2.8 PCB Schematic and Layout Design Following the simulation, PCB schematic was designed to accommodate 2 channels in a 2-‐‑Layer Bio-‐‑Amp card of 10cm x 5cm dimension. The PCB schematic was done with Cadence Allegro Design Entry CIS and PCB Layout was done with Cadence Allegro PCB Editor Suite v16.6. PCB Design, Layout and Populationwere done in-‐‑house while fabrication was done by 3rd party vendors. The schematic for PCB was done for two channels. SMD packages of ICs have been used along with resistors and capacitors of 0805 package. The Gerber file circuits of different layers of the final PCB Layout of the Bio-‐‑Amp board that was sent to the manufacturer has been presented. The Gerber files are of RS-‐‑274X format. Layout Design Data Etch Track Parameters: Minimum Maximum Etch Track width (all nets) 0.5 mm 0.6 mm VCC, VCC1, VGND line width
0.75 mm 1 mm
Min neck width (all nets) 0.38 mm VCC, VCC1, VGND min neck width 0.43 mm Max Neck length (all) 20mm Spacing Parameters Line to line spacing 0.38 mm Line to Thru pin spacing 0.38 mm Line to SMD pin spacing 0.38 mm Thru pin to Thru Via spacing 0.40 mm SMD pin to Thru Via spacing 0.40 mm Drill dimensions VIA Drill diameter 0.6096 mm VIA Regular pad diameter 1 mm Thru Pin Drill diameter 0.9 mm Thru Pin Regular pad diameter 1.8 mm
All other parameters that are not mentioned above have their respective default values, given by the software.
16
A. Etch-‐‑Top
B. Etch-‐‑Bottom
C. Soldermask-‐‑Top
17
D. Soldermask-‐‑Bottom
E. Silkscreen-‐‑Top
E. Silkscreen-‐‑Bottom
18
4.2.9 How they look
4.2.10 Testing and Results Bio-‐‑Amp v2 was extensively tested and its performance was constantly compared with Bio-‐‑Amp v1. It was found that the Bio-‐‑Amp v2 was equipped with a simpler Notch filter, which does not sufficiently remove the 50Hz pickup. The presence of this power-‐‑line interference impairs the SSVEP signal quality. The reason for this insufficiency in Notching of the 50Hz signal may be attributed to the simulation results which although clearly shows considerable performance, at very low voltage levels (~ < 10uV), the attenuation of 50Hz pickup is insufficient. Some of the plots of the Bio-‐‑Amp v2 are presented below. Apart from the signal, frequency levels of 43 Hz, 50 Hz, 56 Hz and 100 Hz signals are almost always seen.
Figure 4. EEG Amplifier V1 and V2 together
Figure 5. EEG Amplifier V2 Board
Figure 6. 13Hz signal pickup after 30 seconds of data acquisition with aVID stimulator.
19
While the signal is picked up in most of the cases, during some trials, heavy noise pickup at all frequencies < 40 Hz is also seen sometimes. While the cause for this remains unknown, one of the suggested solutions is to eliminate the DC in the output so that the entire signal is amplified to the levels of 0-‐‑5 V and input to the digitizer for better signal detection.
4.3.11 Planned Enhancements Since the signal is measured with AGND in the output side, instead of VGND, an additional DC voltage of ~ 2.5V is always present in the output. Since this does not contribute to signal detection much, apart from lifting the signal from falling below 0 V, a Buck-‐‑off circuit is being implemented, at the time of writing this report, to avoid this quantization noise being introduced because of the presence of this DC Voltage.
4.3 Digitizer The analog signals obtained from the amplifier are set to be around 2-‐‑4 V (by varying the gain) making it viable for the 10-‐‑bit ADC in Arduino to sample the analog signals. Since the operating frequency range of the SSVEP Experiments is less than 40 Hz, a sampling frequency of 256 Hz was more than sufficient for faithfully digitizing & reproducing the signal. Program for sampling and digitizing the analog signals was written, making use of one of the inbuilt timers that is invoked 256 times a second. The code can be found in the appendix of the report.
Figure 7. 17Hz signal pickup after 30 seconds of data acquisition with aVID stimulator.
Figure 8. 17Hz signal amongst heavy noise pickup.
20
Pure tones from signal generator were fed into the Arduino Digitizer and the time-‐‑domain and frequency domain spectra were analyzed. Two channels were simultaneously tested with 10 Hz and 20 Hz tones respectively and were sampled at 256 Hz. The plots are presented below. This program was later modified to simultaneously sample 4 channels at 256 Hz and at a data transfer rate of 115200 baud.
Figure 9. Pure tones of 20 Hz & 10 Hz: Their time domain and frequency domain responses. The last picture is a small section of the time-‐‑domain response of the super-‐‑imposed tones.
21
4.4 Signal Processor
4.4.1 Objective The ultimate aim of digitization was to extract useful information out of the signals. Since EEG data obtained from SSVEP experiments contains signals of the frequency-‐‑coded LEDs, it is important to extract this frequency information. A typical approach to decipher the frequency of the signals is to employ Fast Fourier Transform (FFT). But since implementing an FFT algorithm on embedded microcomputers may be computationally intensive, an alternate approach suggested by Dr. Ramesh was to implement a Matched Filter for frequency determination.
4.4.2 Theory In Matched filter, the unknown incoming signal is multiplied with Sine and Cosine components of a known frequency. Following the multiplication, the signals are time averaged over one full time period, which eliminates all undesired frequencies. These individual, time averaged signals are then squared and resulting in a correlation co-‐‑efficient. Since the signal is multiplied and time averaged with both the orthogonal components, phase dependency is eliminated. What is obtained is a correlation coefficient, whose strength is high when the signal frequency matches with frequency of the multiplied signal. This is used to decide the threshold and eventually in decision making. Let us consider that the incoming signal is represented as
f(t)=Σ (An sin wnt + Bncoswnt )
Let the desired frequency that is being looked for bewo. In correlation matched filtering, the incoming signal f(t) is multiplied with the Sine and Cosine components of woand time averaged from 0 to π.
S = ∫ f(t) sin (wot) dt from 0 to π = ∫ A sin(wt) sin (wot) dt + ∫ B cos(wt) sin (wot) dt
= Aπ/2 if w=wo ; or else S < Aπ/2
C = ∫ f(t) cos (wot) dt from 0 to π = ∫ A sin(wt) cos (wot) dt + ∫ B cos(wt) cos (wot) dt = Bπ/2 if w=wo ; or else C < Bπ/2
Correlation of a signal f(t) to a frequencywo is given by,
22
Correlation Coefficient = ∫ f(t) ejwtdt = F(wo) = C (wo) + j S(wo)[since ejwt = cos (wt) + jsin (wt)]
∫ f(t) sin (wot) dt= S(wo) ∫ f(t) cos(wot) dt= C(wo)
Therefore, Correlation Co-‐‑efficient,
F(wo) = √ [ C2(wo) + S2(wo)] = √ (π2/4)[A2 +B2] This value, F(wo) is the correlation co-‐‑efficient that is computed in real-‐‑time to check if the frequency of the incoming signal matches any one of the stimulus frequencies used in the SSVEP experiment.
4.3.3 Implementing the Matched Filter Matched filter was first simulated as a Proof-‐‑of-‐‑Concept on MATLAB, followed by its programmatic implementation on Java. Initially, pure tones generated from the signal generator were used to check the correctness of the algorithm implemented. Following this, the Matched filter for 2 channels was implemented on Arduino Uno, which could then sample and process the signals by itself. From then on, since I got involved with the fabrication of the RRI Bio-‐‑Amp v2, enhancing the signal processing code on Arduino to multiple-‐‑channels and developing a MATLAB GUI was taken care of Mr. James Mathew (JRF, LAMP). The flowchart of the Matched filter program on Arduino is presented below.
Generate & store look-‐‑up tables of stimulus
frequencies
TIMER Loop @ 256 Hz Sample data and Map to 8 bits
Multiply incoming data f(t) sin wt & f(t) cos wt
Accumulate multiplied data until
one T
After one T, determine the
average Square and add Sine & Cosine averages
Check correlation coef_icient threshold & make decision
Drive Wheelchair
23
Later, moving average was also implemented where the final output of the correlation coefficient would be the average of the correlation co-‐‑efficients of the previous 4 cycles. Since after one T, the correlation co-‐‑efficient is calculated and a decision is made, the data can be discarded and the operation begins on fresh set of data. The complexity increases with multiple channels and multiple frequencies to be checked with each channel. The code was extensively tested with Arduino Uno and Arduino Mega by Mr. James Mathew and upto 7 individual frequencies could be checked with 2 channels on Arduino Mega. Since sampling and computation are done simultaneously, sampling rate of not more than 256 Hz is used. Program for the Matched filter implemented on Arduino is included in the Appendix.
Figure 10. Correlation co-‐‑efficient plot indicating the presence of 10 Hz signal with Gaussian Noise
24
5. Takeaways The internship provided me the opportunity to work on an end-‐‑to-‐‑end instrumentation system for the BCI SSVEP Experiment. It helped me in improving my technical skills and enhanced my knowledge in new areas. Specifically,
• I gained new knowledge about working with Bioelectronics. Various specific aspects of working with electronics for Bio-‐‑systems is one of the major takeaways from my internship.
• Gained hands-‐‑on experience in the one of the most coveted research areas: ‘Brain-‐‑computer Interface’. This helped me explore the current research in this domain and realize the possibilities.
• Working on Analog Electronics helped me brush up all the basic knowledge obtained during my Undergraduate degree and more importantly, put all that knowledge to good use. Additionally, I got to learn working with latest software and tools used in the industry.
• Working on some of the core modules of the BCI system made me feel responsible and
committed and also kept me motivated especially when desired results were not forth-‐‑coming.
• My programming skills were extensively used in Bio-‐‑signal processing, a domain on which I had never worked before.
• My participation in In-‐‑House 2014 event gave me the opportunity to represent the BCI
team amidst other eminent scientists, professors and students, besides contributing to my presentation skills.
25
6. Appendix
6.1 The aVID Arduino Code for 3 LEDs //http://www.instructables.com/id/Arduino-Timer-Interrupts/ //This code will enable all three arduino timer interrupts. //timer0 will interrupt at 2kHz (and then arithmetically divide to get 17 Hz) //timer1 will interrupt at 13Hz //timer2 will interrupt at 15Hz //storage variables boolean toggle0 = 0; boolean toggle1 = 0; boolean toggle2 = 0; int timer2_count=0; int timer0_count=0; void setup(){ //set pins as outputs pinMode(8, OUTPUT); pinMode(10, OUTPUT); pinMode(12, OUTPUT); cli();//stop interrupts //set timer0 interrupt at 2kHz TCCR0A = 0;// set entire TCCR2A register to 0 TCCR0B = 0;// same for TCCR2B TCNT0 = 0;//initialize counter value to 0 // set compare match register for 2khz increments OCR0A = 123;// = (16*10^6) / (1024*100) - 1 (must be <256) // turn on CTC mode TCCR0A |= (1 << WGM01); // Set CS02 and CS00 bits for 1024 prescaler TCCR0B |= (1 << CS01) | (1 << CS00); // enable timer compare interrupt TIMSK0 |= (1 << OCIE0A); //set timer1 interrupt at 13Hz TCCR1A = 0;// set entire TCCR1A register to 0 TCCR1B = 0;// same for TCCR1B TCNT1 = 0;//initialize counter value to 0 // set compare match register for 13Hz increments //OCR1A = 15624;// = (16*10^6) / (1*1024) - 1 (must be <65536) OCR1A = 600;// = (16*10^6) / (13*1024) - 1 (must be <65536) // turn on CTC mode TCCR1B |= (1 << WGM12); // Set CS12 and CS10 bits for 1024 prescaler TCCR1B |= (1 << CS12) | (1 << CS10); // enable timer compare interrupt TIMSK1 |= (1 << OCIE1A); //set timer2 interrupt at 15Hz TCCR2A = 0;// set entire TCCR2A register to 0 TCCR2B = 0;// same for TCCR2B TCNT2 = 0;//initialize counter value to 0 // set compare match register for 15Hz increments OCR2A = 248;// = (16*10^6) / (1005*64) - 1 (must be <256) // turn on CTC mode TCCR2A |= (1 << WGM21); // Set CS21 bit for 8 prescaler
26
TCCR2B |= (1 << CS22); // enable timer compare interrupt TIMSK2 |= (1 << OCIE2A); sei();//allow interrupts }//end setup // Timer-Interrupt Service routine to toggle LED at pin 8 ISR(TIMER0_COMPA_vect){ if(timer0_count==58) { if (toggle0){ digitalWrite(8,HIGH); toggle0 = 0; } else{ digitalWrite(8,LOW); toggle0 = 1; } timer0_count=0; } else {timer0_count++;} } // Timer-Interrupt Service routine to toggle LED at pin 10 ISR(TIMER1_COMPA_vect){ if (toggle1){ digitalWrite(10,HIGH); toggle1 = 0; } else{ digitalWrite(10,LOW); toggle1 = 1; } } // Timer-Interrupt Service routine to toggle LED at pin 12 ISR(TIMER2_COMPA_vect){ if(timer2_count==32) { if (toggle2){ digitalWrite(12,HIGH); toggle2 = 0; } else{ digitalWrite(12,LOW); toggle2 = 1; } timer2_count=0; } else {timer2_count++;} } void loop(){ //do other things here }
27
6.2 4-‐‑Channel Arduino Digitizer
//Arduino 4-Channel Digitizer void setup(){ Serial.begin(115200); cli();//stop interrupts //set timer1 interrupt at 256Hz TCCR1A = 0;// set entire TCCR1A register to 0 TCCR1B = 0;// same for TCCR1B TCNT1 = 0;//initialize counter value to 0 // set compare match register for 1hz increments OCR1A = 60;// = (16*10^6) / (100*1024) - 1 (must be <65536) // turn on CTC mode TCCR1B |= (1 << WGM12); // Set CS12 and CS10 bits for 1024 prescaler TCCR1B |= (1 << CS12) | (1 << CS10); // enable timer compare interrupt TIMSK1 |= (1 << OCIE1A); sei();//allow interrupts }//end setup //Timer-1 interrupt service routine //Samples analog signals from A0, A1, A2 and A3, downconverts to 8bits and serially sends it with start and stop characters (A and Z) ISR(TIMER1_COMPA_vect){ intval=0; Serial.write("A"); val = map(analogRead(A0), 0, 1024, 0, 255); Serial.write(val); val = map(analogRead(A1), 0, 1024, 0, 255); Serial.write(val); val = map(analogRead(A2), 0, 1024, 0, 255); Serial.write(val); val = map(analogRead(A3), 0, 1024, 0, 255); Serial.write(val); Serial.write("Z"); } void loop(){ //do other things here }
6.3 Matched Filter program on Arduino //This is a 1-Channel Matched filter that can check 4 individual frequencies // 1 Channel 4 MF Correlation Values #define N 24 #include<Math.h> int v1=0; int v2=0; int v3=0; int v4=0; int V=0; int data1 = -1; double LUT_sin1[24];
28
double LUT_cos1[24]; double LUT_sin2[18]; double LUT_cos2[18]; double LUT_sin3[16]; double LUT_cos3[16]; double LUT_sin4[12]; double LUT_cos4[12]; double mul_ch1_sin1[24]; double mul_ch1_cos1[24]; double mul_ch1_sin2[18]; double mul_ch1_cos2[18]; double mul_ch1_sin3[16]; double mul_ch1_cos3[16]; double mul_ch1_sin4[12]; double mul_ch1_cos4[12]; int f1 = 12;//Matched frequency to be checked int f2 = 16;//Matched frequency to be checked int f3 = 18;//Matched frequency to be checked int f4 = 24;//Matched frequency to be checked int Fs = 256; int k1=0; int k2=0; int k3=0; int k4=0; double accum1_ch1 = 0.0; double accum2_ch1 = 0.0; double accum3_ch1 = 0.0; double accum4_ch1 = 0.0; double CorrCoef1_ch1=0.0; double CorrCoef2_ch1=0.0; double CorrCoef3_ch1=0.0; double CorrCoef4_ch1=0.0; double S1_ch1=0; double C1_ch1=0; double S2_ch1=0; double C2_ch1=0; double S3_ch1=0; double C3_ch1=0; double S4_ch1=0; double C4_ch1=0; // Populating Look-up tables at required frequencies void setup() { Serial.begin(115200); pinMode(12, OUTPUT); //for LED control for(int i=0;i<24;i=i++) { LUT_sin1[i] = sin(2*3.14159*f1*i/Fs); LUT_cos1[i] = cos(2*3.14159*f1*i/Fs); Serial.print(" --LUT-sine- "); Serial.print(LUT_sin1[i],4); Serial.println("**********"); } for(int i=0;i<18;i=i++) { LUT_sin2[i] = sin(2*3.14159*f2*i/Fs); LUT_cos2[i] = cos(2*3.14159*f2*i/Fs); } for(int i=0;i<16;i=i++) {
29
LUT_sin3[i] = sin(2*3.14159*f3*i/Fs); LUT_cos3[i] = cos(2*3.14159*f3*i/Fs); } for(int i=0;i<12;i=i++) { LUT_sin4[i] = sin(2*3.14159*f4*i/Fs); LUT_cos4[i] = cos(2*3.14159*f4*i/Fs); } cli();//stop interrupts //set timer1 interrupt at 256Hz TCCR1A = 0;// set entire TCCR1A register to 0 TCCR1B = 0;// same for TCCR1B TCNT1 = 0;//initialize counter value to 0 OCR1A = 60;// = (16*10^6) / (Fs*1024) - 1 (must be <65536) TCCR1B |= (1 << WGM12); TCCR1B |= (1 << CS12) | (1 << CS10); TIMSK1 |= (1 << OCIE1A); sei();//allow interrupts } // Correlation co-efficients are computed here void loop() { if(data1!=-1) { double d1 = data1; data1 = -1; V=V+1; //************************************************************************** //1st frequency check******************************************************* mul_ch1_sin1[v1] = LUT_sin1[v1]*d1; mul_ch1_cos1[v1] = LUT_cos1[v1]*d1; if (v1==23) { for(int i=0;i<24;i++) { S1_ch1 = S1_ch1 + mul_ch1_sin1[i] ; C1_ch1 = C1_ch1 + mul_ch1_cos1[i] ; } k1=k1+1;v1=-1; } if(k1==12) { accum1_ch1=(S1_ch1*S1_ch1 + C1_ch1*C1_ch1); CorrCoef1_ch1=accum1_ch1/1000; k1=0;S1_ch1=0;C1_ch1=0; } //*************************************************************************** //2nd frequency check******************************************************** mul_ch1_sin2[v2] = LUT_sin2[v2]*d1; mul_ch1_cos2[v2] = LUT_cos2[v2]*d1; if (v2==17) { for(int i=0;i<18;i++) { S2_ch1 = S2_ch1 + mul_ch1_sin2[i] ; C2_ch1 = C2_ch1 + mul_ch1_cos2[i] ;
30
} k2=k2+1;v2=-1; } if(k2==16) { accum2_ch1=(S2_ch1*S2_ch1 + C2_ch1*C2_ch1); CorrCoef2_ch1=accum2_ch1/1000; k2=0;S2_ch1=0;C2_ch1=0; } //************************************************************************** //3rd frequency check****************************************************** mul_ch1_sin3[v3] = LUT_sin3[v3]*d1; mul_ch1_cos3[v3] = LUT_cos3[v3]*d1; if (v3==15) { for(int i=0;i<16;i++) { S3_ch1 = S3_ch1 + mul_ch1_sin3[i] ; C3_ch1 = C3_ch1 + mul_ch1_cos3[i] ; } k3=k3+1;v3=-1; } if(k3==18) { accum3_ch1=(S3_ch1*S3_ch1 + C3_ch1*C3_ch1); CorrCoef3_ch1=accum3_ch1/1000; k3=0;S3_ch1=0;C3_ch1=0; } //******************************************************************************* //4th frequency check************************************************************ mul_ch1_sin4[v4] = LUT_sin4[v4]*d1; mul_ch1_cos4[v4] = LUT_cos4[v4]*d1; if (v4==11) { for(int i=0;i<12;i++) { S4_ch1 = S4_ch1 + mul_ch1_sin4[i] ; C4_ch1 = C4_ch1 + mul_ch1_cos4[i] ; } k4=k4+1;v4=-1; } if(k4==24) { accum4_ch1=(S4_ch1*S4_ch1 + C4_ch1*C4_ch1); CorrCoef4_ch1=accum4_ch1/1000; k4=0;S4_ch1=0;C4_ch1=0; } //******************************************************************************** // Serial.println(d1); //Control LED if(CorrCoef1_ch1>50) { digitalWrite(12,HIGH); digitalWrite(11,LOW); digitalWrite(10,LOW);} else if(CorrCoef2_ch1>50) {digitalWrite(11,HIGH); digitalWrite(12,LOW); digitalWrite(10,LOW);} else if(CorrCoef3_ch1>50)
31
{digitalWrite(10,HIGH); digitalWrite(11,LOW); digitalWrite(12,LOW);} else {digitalWrite(10,LOW); digitalWrite(11,LOW); digitalWrite(12,LOW);} if(V==288) { Serial.println(CorrCoef1_ch1); Serial.println(CorrCoef2_ch1); Serial.println(CorrCoef3_ch1); Serial.println(CorrCoef4_ch1); V=0;} v1++; v2++; v3++; v4++; } } //Timer to sample & digitize data at Fs = 256Hz ISR(TIMER1_COMPA_vect){ data1 = map(analogRead(A0), 0,1024, 0, 255);}
![3D Geological Modeling under Extremely Complex Geological ... · 3D geological modeling Ref. [4]. In 1994, 3D geological modeling technology was firstly put up by Simon W Houlding](https://static.documents.pub/doc/80x56/5e77a1369d577a7b9f489d8a/3d-geological-modeling-under-extremely-complex-geological-3d-geological-modeling.jpg)
