Sensor Basic Test

Fig. 3- 26: Diagram of sensor basic test

After the circuitry and program system comparison, we verify how good performance dry sensor has as possible as it can. Thus, we design the verification procedure as the diagram of Fig. 3-26. The participant was placed day sensor and wet sensor with about 2-cm distance. Both of sensors simultaneously monitored EEG activity, and transmitted the signal to two inputs of 16-channel mobile and wireless circuitry. The circuitry sampled signal with 125Hz of rate, and then carried it to back-end program via wireless communication interface. In analysis process, we

extracted one channel data measured by dry sensor and the other channel measured by wet sensor to compare the performance in time and frequency domain respectively.

In addition, we follow above experiment using verify the circuitry and program, and include close eye for one of actions to verify the performance of dry sensor monitoring alpha wave in this experiment. We expect to check this dry sensor whether eliminating the difference of measurement quality for many subjects or not, and furthermore, in which frequency band the measurement quality is the best and how good it is are the issues we concern.

3.3.1 Participant

Three participants are invited to go in this experiment. Two males and one female are about 20-23 years old. One of them has the thinner and general number of hairs in head, one has general number of hairs in head, and the other has the thicker and less number of hairs in head. The participants owns different characteristic of hair to each other, hence it is good for us to test the reliability of our system.

3.3.2 Experiment Procedure and Presentation

In this experiment, we assigned the participant to behave four kinds of actions which are “Blink”, “Close”, “Tooth”, and “Normal”. “Blink” action, which is inducing EOG signal, verifies the lower frequency band about 0.2 - 5 Hz. “Close”

action, which is inducing alpha wave easily, verifies the medium frequency band about 8 – 13 Hz. “Tooth” action, which is referred to EMG signal, verifies the higher frequency band about 10 – 25 Hz. “Normal” action verifies frequency band about 0.2 – 60 Hz. In this procedure of experiment, we required the participant to blink once per second within occurring “Blink” command duration, to close eyes and relax within occurring “Close” command duration until the alert voice produced, to grind the molar with uninterrupted within occurring “Tooth” command duration, and to do

general action naturally within occurring “Normal” command duration. Among Normal state, EOG like eye-movement and EMG like muscle-movement at chin position are randomly appeared to cover clear EEG signal. Thus, EOG and EMG are ordinary considered no-use activity and even interfering more important EEG feature in fact. However, sensor verification in this experiment can use the characteristic of EOG and EMG to check lower and higher frequency bands respectively.

The experiment is going in general environment. Although most of cognitive experiments are executed in electromagnetic-shielded space, our system purpose to apply in daily life. Fig. 3-27 shows the space of experiment.

Fig. 3- 27: Experiment environment of sensor basic test

Fig. 3- 28: The view of sensor placement. (a) Ground, (b) Dry sensor and wet sensor, (c) Reference

Beginning of experiment, the participant sat on a chair motionlessly and was asked to keep facing on the front screen and followed occurred commands by presentation. The experiment includes four sections, each of about 12-miniute duration. Each section includes thirteen trials, each of seventy-second duration. The participant can take a rest among sections. Fig. 3-29 shows procedure of presentation in one trial.

Fig. 3- 29: The procedure of sensor basic test in one trial

Referencing to 10-20 system (Guideline for Standard Electrode Position Nomenclature, 2006), we choose the edge and center positions (Fig. 3-28, 3-30) to place dry sensor. The forefront side is FP1. The most left side is T3. The top side is CZ. The most behind side is OZ. In addition, the participant is also placed an electrode behind ear as reference potential and an electrode on G as ground potential.

All the dry sensors are pressed by hands of the other person (operator), because we do not find the better fixation to make the least motion artifact than manual pressing yet.

The Operator has to force down the dry sensor stably to avoid the movement between sensor and skin as possible.

Fig. 3- 30: Positions of dry sensor and wet sensor in sensor basic test

3.3.3 Method of Analysis

The method of analysis is mainly parted two sections which are time-domain analysis and frequency-domain analysis. Before analysis beginning, we have to preprocess raw data recorded in TXT file. The preprocessing step is only one, which is rejecting no-use data. Due to dry sensor needing not conductive gels to maintain the stable impedance between sensor and skin in the brain, dry sensor has the opportunity to contact badly with the skin. As the following Fig. 3-31, we reject this data owning to voltage saturation and high impedance which is just like floating. The plunger of dry sensor is flexible in travel distance. It is possible to make capacitance effects that operator forces down the sensor in travel distance. If the sensor is not stably fixing on the skin site, there is floating data recorded by our system.

Fig. 3- 31: No-use data for floating case and saturation case

Fig. 3-32 shows the procedure diagram of analysis. First, compute all data temporal correlation for the data measured by dry and wet sensor in certain position.

Next, divide the whole data to four parts toward corresponding action type. On the other word, the whole data combining four actions EEG data, and we part it to

“Blink”, “Close”, “Tooth”, and “Normal” data. After time-domain analysis, in frequency domain, we analyze the FFT correlations result for each trial data. This

method finally gets eight results for each trial data induced by the same action: (1) temporal correlations for all time, (2) FFT correlation for all time, (3) 0-30-Hz FFT correlation, (4) 30-60-Hz FFT correlation, (5) FFT correlation of Delta band, and (6) FFT correlation of Theta band, (7) FFT correlation of Alpha band, and (8) FFT correlation of Beta band. Here, we just get the results of one trial. By computing all the trial data, we average the data of all the trial and get the final eight correlations for the same action.

The data got from different positions pass through average process. The record EEG data of three participants are all computed in this analysis process. In the end of analysis, the results considering different subjects and different positions are presented in the following session.

Fig. 3- 32: Diagram of analysis method in sensor basic test

3.3.4 Experiment Results

A. Results after Averaging All Subjects and All Positions

Fig. 3- 33: Overall results of sensor basic test

Table 17: Analysis results of Blink in sensor basic test

Blink Time FFT 0-30Hz 30-60Hz Delta Theta Alpha Beta Subject 1 50.9 65.7 69.3 80.9 47.9 78.5 76.0 77.7 Subject 2 73.9 80.4 84.4 87.0 70.5 87.4 81.2 77.4 Subject 3 59.9 87.5 88.2 65.5 83.4 83.9 66.5 54.2

Average 61.5 77.9 80.6 77.8 67.2 83.2 74.6 69.8

Table 18: Analysis results of Close in sensor basic test

Close Time FFT 0-30Hz 30-60Hz Delta Theta Alpha Beta Subject 1 46.4 69.0 74.5 77.3 86.9 53.5 91.3 90.4 Subject 2 59.2 80.7 80.8 87.7 87.1 59.0 97.6 79.6 Subject 3 37.8 88.2 90.0 73.5 91.7 48.5 82.8 80.5

Average 47.8 79.3 81.4 79.5 88.6 53.7 90.6 83.5

Table 19: Analysis results of Tooth in sensor basic test

Tooth Time FFT 0-30Hz 30-60Hz Delta Theta Alpha Beta Subject 1 48.5 56.3 65.0 44.7 57.9 64.0 60.0 58.9 Subject 2 46.0 83.0 88.1 58.5 81.6 73.2 48.2 55.2 Subject 3 39.2 67.8 76.4 57.6 72.7 42.1 30.5 53.0 Average 44.6 69.0 76.5 53.6 70.7 59.8 46.3 55.7

Table 20: Analysis results of Normal in sensor basic test

Normal Time FFT 0-30Hz 30-60Hz Delta Theta Alpha Beta Subject 1 54 67.7 75.3 74.6 67.5 58.9 79.5 77.7 Subject 2 68.8 87.7 91.4 89.7 88.3 74.3 56.6 62.2 Subject 3 52.4 90.3 90.9 65.0 85.9 69.8 70.4 59.2 Average 58.4 81.9 85.9 76.5 80.6 67.6 68.8 66.4

B. The Best Trial Data in Blink Action

Fig. 3- 34: The best performance of Blink at FP1

Fig. 3- 35: The best performance of Blink at T3 (up), CZ (mid), OZ (down)

C. The Best Trial Data in Close Action

Fig. 3- 37: The best performance of Close at OZ

D. The Best Trial Data in Tooth Action

Fig. 3- 38: The best performance of Tooth at FP1

Fig. 3- 39: The best performance of Tooth at T3 (up), CZ (mid), OZ (down)

E. The Best Trial Data in Normal Action

Fig. 3- 40: The best performance of Normal at FP1 (up), T3 (mid), CZ (down)

Fig. 3- 41: The best performance of Normal at OZ

F. The Worst Trial Data

Fig. 3- 42: The worst performance during to 60-Hz Electromagnetic Interference

Fig. 3- 43: The worst performance during to motion artifact

Fig. 3- 44: The worst performance during to DC shifting

3.3.5 Discussion

A. Sensor Test for All Subjects and All Positions

As Fig. 3-33 shown, the Time correlations for four actions are all less than 65%.

The highest one is Blink during to the high amplitude of EOG being distributed the whole head more and less. As following we discuss results toward four actions respectively.

Blink: Time correlation is the highest one which is 61.5%, but what the reason that it can’t be higher is. According to FFT correlations, we find the poor 30-60-Hz FFT correlation of 77.8%, so the whole FFT correlation is only 77.9% (because 0-30-Hz FFT correlation is 80.6%). In addition, in 0-30-Hz frequency band, we find Theta band has the best performance about 83.2%, and Delta band has the worst performance about 67.2%. In fact, Beta band has poor performance which is just a little higher than Delta band about 69.8%.

Close: Time correlation is the highest one which is 47.8%, but what the reason that it can’t be higher is. According to FFT correlations, we find the poor 30-60-Hz FFT correlation of 79.5%, so the whole FFT correlation is only 79.3% (because 0-30-Hz FFT correlation is 81.4%). In addition, in 0-30-Hz frequency band, we find Alpha band has the best performance about 90.6%, and Theta band has the worst performance about 53.7%.

Tooth: Time correlation is the highest one which is 44.6%, but what the reason that it can’t be higher is. According to FFT correlations, we find the poor 30-60-Hz FFT correlation of 53.6%, so the whole FFT correlation is only 69% (because 0-30-Hz FFT correlation is 76.5%). In addition, in 0-30-Hz frequency band, we find Delta band has the best performance about 70.7%, and Alpha band has the worst performance about 46.3%.

Normal: Time correlation is the highest one which is 58.4%, but what the reason that it can’t be higher is. According to FFT correlations, we find the poor 30-60-Hz FFT correlation of 76.5%, so the whole FFT correlation is only 81.9% (because 0-30-Hz FFT correlation is 85.9%). In addition, in 0-30-Hz frequency band, we find Delta band has the best performance about 80.6%, and Beta band has the worst performance about 66.4%. In fact, Theta band and Alpha band also have poor

respectively.

In conclusion, during to there are different frequency bands with the best and the worst performances toward four kinds of actions, we demonstrate that dry sensor can not measure the best or the worst performance in certain frequency band. In fact, the best and worst results are caused by the behaved action more. From Table 17 to Table 20, we recognize the best performance in “Blink” is owing to three subjects consistently presenting the best, and the worst one in “Blink” is owing to two subjects consistently presenting the worse; the best performance in “Close” is owing to two subjects consistently presenting the best, and the worst one in “Close” is owing to three subjects consistently presenting the worse; the best performance in “Tooth” is owing to two subjects consistently presenting the best, and the worst one in “Tooth” is owing to two subjects consistently presenting the worse. However, there is not any consistent performance in “Normal”. Thus, we know that the best and the worst performances of comparing between dry sensor and wet sensor in different behaviors are as a result of many subjects rather than single subject.

B. The Similarities between Dry Sensor and Wet Sensor

As Fig. 3-34 to Fig. 3-41 shown, we verify dry sensor measuring the same EEG feature with wet sensor.

Blink: In Time domain, dry sensor monitor EOG signal at FP1 position as wet sensor. Moreover, there are a little attenuated EOG at not only T3 position but CZ position. Because of CZ being behind the head, there is almost no EOG signal measured at CZ position. In frequency domain, the signal of low frequency is more obviously occurred in activity stage (A) than non-activity stage (NA).

Close: In time domain, there is existing alpha wave in every position. In frequency domain, the power of Alpha band is enhanced in activity stage, and the highest power which is over 40dB occurred in CZ position.

Tooth: In time domain, EMG high-frequency signal distribute FP1, T3 and CZ positions. In frequency domain, there is large power enhanced within the frequency band over 10Hz.

Normal: In frequency band, the power distribution in non-activity stage is pretty similar to in activity stage because of the participant assigned to behave nature action (the same as in “baseline” stage) in “Normal”.

C. The Differences between Dry Sensor and Wet Sensor

Although dry sensor presents the same measured EEG patterns as wet sensor, the drawback of dry sensor is still existent to make the poor performance compared to wet sensor. We sort out three conditions of bad measurement quality:

(1) 60-Hz Electromagnetic Interference

Fig. 3-42 displays that dry sensor measures more 60-Hz noise from natural environment. It may mean dry sensor is more sensitive than wet sensor. The ability of resisting high-frequency noise for Dry sensor is worse than wet sensor.

(2) Motion artifact

Fig. 3-43 shows when the participant grinding the teeth, dry sensor moves a little offset related original skin site, which is referred to motion artifact. Hence, the mechanism for fixation is relatively important toward dry sensor.

(3) DC shifting during to unequal force pressed

Dry sensor is forced down by hands of operator, so there is inevitable that a little manually shaking happened within the process of experiment. Thus, the record data combines a little manual negligence such as DC shifting (Fig. 3-44).

3.4 Sensor Test in Oddball Task

In this session, we illustrate our procedure of oddball task, the participants, and experiment results. By last experiment, we get some sensor performance information, but the performance of Event-Related Potential (ERP) seems more important. Owning to ERP being a very tiny EEG activity, it needs to average more trials not only for reducing noise but for enlarging event-related potential. For the purpose of verifying whether our system can monitor tiny signal even ERP or not, we follow the same system verification procedure diagram as Fig. 3-40 and re-design an Oddball experiment. Finally, in analysis process, we compare the performance between dry sensor and wet sensor.

3.4.1 Participant

There are ten participants with 6 males and 4 females (19 – 23 ages). Most of then are normal number of hairs, two people are fewer number of hairs, and three people are larger number of hairs.

3.4.2 Experiment Procedure and Presentation

The oddball task combines normal stimulus and target stimulus. The participant was asked to click button when target stimulus occurred. The target only presented in 75-ms duration, and the following 1925-ms duration is for waiting next stimulus. Due to past researcher indicating that the ratio of normal stimulus to target stimulus is 8 to 2. We designed the target to be randomly occurred, total number of times is 5-time less than non-target (normal) stimulus in oddball task. When one session of experiment beginning, the screen presents a letter first and show “Please put down the button for non-X” to tell the participant what normal letter “X” is in this session. The normal letter, which is possible to be A, B, C, D or E, is random produced by presentation program in each session (Fig. 3-45).

Fig. 3- 45: Procedure of oddball experiment

Beginning of experiment, the participant sat on a chair motionlessly and was asked to keep facing on the front screen and followed occurred commands by presentation. The experiment includes four sections, each of about 10-miniute duration. Each section includes 300 trials, each of 2-second duration. Among one session, there are 240 trials with normal stimulus and 60 trials with target stimulus.

The participant can take a rest among sections. This experiment may be executed for one hour. Fig. shows procedure of presentation in 18 trials with random-stimulus.

Referencing to 10-20 system (Guideline for Standard Electrode Position Nomenclature, 2006), we choose CZ position (Fig. 3-46) to place dry sensor. All the dry sensors are pressed by hands of the other person. All channels are linked to 16-channel EEG device, and the program record EEG data.

Fig. 3- 46: Positions of dry sensor and wet sensor in oddball task

3.4.3 Method of Analysis

As following steps describe the method of analyzing ERP in oddball task using EEGLAB toolbox [25]:

(1) Low-pass filter: Dry sensor record the frequency combined 0.2 – 125 Hz signals.

Only P300 is the EEG data of interest. To filter high-frequency signal is good for us to view the pattern of P300. Hence, first step is to filter for remaining low 0.2 – 30 Hz EEG signal.

(2) Separate the epoch for data of the same event: To separate “Normal epoch” for normal stimulus event and “Oddball epoch” for target stimulus event, both of them are remained interval of [-0.5 1.5], in which [-0.5 0] is baseline, is one step of ERP preprocessing.

(3) Reject no-use data: According to last experiment of sensor basic test, we get the information about the measuring effects of dry sensor like DC-shifting, floating, and voltage saturation. They are all no-use data for computing ERP, so we reject them early. Fig. 3-47 shows what kind of data we reject. In most researches, EOG is also rejected, but it is difficult to reject EOG here during to only monitoring CZ position in which EOG is difficultly distinguished within EEG data.

Fig. 3- 47: No-use data for floating case, saturation case, and DC shifting

(4) Compute ERP: EEG data, which is separated to the same interval related event, pass through averaged more and more trial data not only to enhance event-related potential but to decrease the influence of noise. In this step, “Normal epoch” and

“Oddball epoch” respectively average all trial data. Thus, the averaged data are corresponding Normal ERP and Oddball ERP.

3.4.4 Experiment Results

Fig. 3- 48: ERP of s01 (left-up), s02 (right-up), s03 (left-down), s04 (right-down) in oddball task

Fig. 3- 49: ERP of s05 (left-up), s06 (right-up), s07 (left-mid), s08 (right-mid), s09 (left-down), s10 (right-down) in oddball task

3.4.5 Discussion

A. P300 of Single Stimulus

This experiment presents a stimulus per 2 seconds, each of in about 75-ms duration. No matter using dry sensor or wet electrode, there is measured P300 by our multi-channels mobile and wireless system. From Fig. 3-48 to Fig. 3-49, the figures show event-related potential of single stimulus for all participants. In baseline interval, there is no potential occurred. After stimulus, which may be normal stimulus or target stimulus, triggered, ERP simultaneously occurs in 300ms – 500ms. Hence, we get a result of ten participants all inducing P300 by single stimulus. On the other words, all ten participants own the same P300 tendency for single stimulus on CZ position.

B. P300 of Oddball Task

In this discussion about P300 through Oddball task, we have to respectively discuss “Normal” and “Oddball” state for ten participants initially. It is noteworthy that the ERP pattern in this experiment is a little different from standard ERP pattern possibly owing to EOG signal being not rejected.

Subject 1 (s01): Dry sensor measured the larger amplitude of P300 in “Oddball” than

“Normal”, and we can recognize that P300 in “Oddball” is a little later than in

“Normal” (the P300 latency in “Oddball” is higher than in “Normal”). However, wet sensor measured the higher latency in “Oddball” and the very similar amplitude in

“Oddball” and “Normal”.

Subject 2 (s02): Dry and wet sensors both measured the higher amplitude of

Subject 2 (s02): Dry and wet sensors both measured the higher amplitude of