A Novel Wireless Brain-Computer Interface System using Steady-State Visual Evoked Potentials

Size: px

Start display at page:

Download "A Novel Wireless Brain-Computer Interface System using Steady-State Visual Evoked Potentials"

Arabella Ross
5 years ago
Views:

1 A Novel Wireless Brai-Computer Iterface System usig Steady-State Visual Evoed Potetials Zhegpig Wu 1, Jiafeg Wag 1, Logqig Dai 1, HuaWei 2,Chu Che 1, Xiagyu Yag 3,Ziai Zhou 3 1. School of Egieerig Educatio, SaJiag Uiversity, , Chia 2.Brai Medical Techology Co.Ltd of Jiagsu, , Chia 3. Istitute of Life Scieces, Southeast Uiversity, Najig , Chia Abstract - Existig brai-computer iterfaces are geerally expesive, limited ad heavy. I order to solve the problems, we developed a ovel wireless brai-computer iterface based o Steady-State Visual Evoed Potetials(SSVEP), ad developed algorithms for EEG feature extractio. The EEG acquisitio device is powered by lithium batteries ad the sigals are trasferred wirelessly to host for data processig. The sigals was pre-treated usig Butterworth filter based o MATLAB, I additio, this paper combied with the caoical correlatio aalysis ad wavelet trasform time-frequecy dual characteristics, puts forward the feature extractio algorithm WT CCA, the processed filtered data usig our proposed caoical correlatio aalysis method based o wavelet trasform. The experimetal results from 5 differet test attedees showed that the proposed iterface was able to extract useful features ad the feature sigals which was processed by a host computer could be correctly (>90%) recogized by outer computers. The proposed iterface is ecoomical ad portable, ad importat to the developmet of the brai cotrol techology. It also helps disabled to better utilize the techology. Keywords -SSVEP; BCI; feature extractio; Wavelet Trasform; CCA I. INTRODUCTION The Brai-Computer Iterface(BCI) is a system to achieve the goal of cotrollig the equipmet by cosciousess which is o eed to establish a commuicatio eviromet by covetioal brai output chael betwee huma or other aimals with the exteral eviromet[1].germa europhysiologist Has Berger is the first oe to discovered the existece of Electroecephalogram (EEG) sigals, ad laid the foudatio for the developmet of BCI. After that, researchers have bee worig o how to use the brai electrical sigal to cotrol the exteral device[2].philip Keedy ad Roy Baay of Emory Uiversity i the Uited States first implated the iterface for obtaiig high-quality EEG sigals i Johy Ray, a patiet with a sydrome of braistem stroe-iduced loc-i sydrome, allowig Ray to survive for log eough ad to lear The use of brai sigals to cotrol the computer cursor[3]. The brai-computer iterface ca be divided ito two types: active ad passive, active based o impotet motio imagiatio, ad passive maily based o P300 sigal ad SSVEP (Steady-State Visual Evoed Potetial, SSVEP) sigal to achieve BCI Fuctio. SSVEP is a respose to a cotiuous ad stimulatig frequecy (frequecy or frequecy doublig) i the visual cortex of the brai whe the huma body is subjected to a costat frequecy of visual stimuli. The wireless BCI system developed i this paper is composed of three parts: a portable multi-chael radio sigal collector, a visual evoed program, ad a brai electrical sigal processig program. Multi-chael wireless EEG sigal acquisitio device is optimized ad desiged based o the prelimiary research, which maes EEG sigal acquisitio device is more portable ad easy to use [4].SSVEP-based programmig ca focus the user's attetio o a certai visual stimulus to produce a specific EEG modality, ad the combie Caoical Correlatio Aalysis (CCA) ad Wavelet Trasform (WT) for feature extractio, A feature extractio algorithm based o CCA-WT is proposed, which the average accuracy improved by early 8.3% compared with the most commoly used CCA correlatio coefficiet feature extractio algorithm. II. SYSTEM STRUCTURE DESIGN System bloc diagram show i Figure 1, which cotais four modules, the master cotrol system, visual stimulator, EEG sigal acquisitio, sigal processig ad istructio output. DOI /IJSSST.a ISSN: x olie, prit

2 WIFI Sigal acquisitio Mai cotrol system Data aalysis Cotrol Feedbac Stimulatig iterface Istructio output Figure 1. System Structure. A. Visual Stimulator Visual stimulator is based o LabVIEW developmet platform. I the case of less stimulatio, relative to the high requiremets of LED o the hardware circuit ad the excessive volume of CRT, CD display oly eed to operate through the software, the operatio is simple, more coducive to the use of the user. Its software desig bloc diagram is show i figure 2,based o the 60Hz refresh characteristics of CD display. Image ca flicer at specified frequecy by adjustig the time parameters, a sigle stimulus program to start the cycle ad the ed is realized by a NOT gate fuctio, while the shift register ad while loop call esured the cotiuity of the cycle. The stimulus iterface is composed of four idetical blac circles. Oly eed to adjust the cotrol parameters to chage the frequecy of the flicer, the whole operatio iterface is simple ad easy to cotrol. B. EEG collector The self-developed wireless EEG sigal collector has the characteristics of coveiet use, abudat fuctio ad excellet performace. The size is oly 56.7mm 50.8mm 19mm, which is easy to carry. Collected by the electrode collector to the EEG sigal by the aalog-digital coversio, the wireless output set to the host computer, the PC program for sigal aalysis ad processig. Figure 3 depicts the bloc diagram of the collector, icludig the power supply module, aalog frot-ed module, the mai cotrol system module ad the WIFI wireless trasmissio module. lithium Power supply system battery EEG Cap Coverter ADS1299 Aalog Frot Ed STM32 MIcroprocessor WIFI Traslatio Figure 3. EEG Acquisitio Hardware Structure Figure 2. Bloc Diagram Of Stimulator Wireless EEG collector uses a built-i rechargeable lithium battery-powered, exteral USB iterface to charge. The use of Lier Techology's high-performace LTC4065 charge maagemet chip cosistig of chargig circuit ca exted the service life of lithium batteries, lithium batteries ca slow dow the agig process ad the low utilizatio rate caused by the degradatio of capacity. Power is the use of LTC2941 battery detectio circuit chip builds the battery gauge circuit through the I2C iterface coected to the host computer software o the real-time display of electricity usage, so as to avoid the AC sigal to brig the power frequecy iterferece ca also be portable for the DOI /IJSSST.a ISSN: x olie, prit

Users. Part of the positive ad egative aalog circuit with Liear Techology Corporatio's low oise liear regulator device (LDO) LT1763 ad egative output switchig regulator chip LT1614.

3 Users. Part of the positive ad egative aalog circuit with Liear Techology Corporatio's low oise liear regulator device (LDO) LT1763 ad egative output switchig regulator chip LT1614. Both chips allow the use of small iductors ad capacitors, ot oly greatly saves the circuit space, reducig costs, ad has a very low ripple, oly 1mV or so. Figure 4. ADS1299 iteral schematic diagram As the EEG sigal wea (0.5uV-100uV) characteristics, the traditioal EEG sigal acquisitio frot-ed usually i each chael is equipped with amplifyig circuit ad filter circuit, leadig to large collector volume, ati-iterferece ability is wea, due to magificatio Large (thousads of times) liely to cause saturatio. I view of the above shortcomigs, this simulatio frot-ed module with TI's iteral productio gai amplifier (PGA) chip ADS1299, its iteral priciple show i Figure 4. The ADS1299 is a multichael differetial iput, sychroous 24-bit samplig, delta-sigma (Δ-Σ) aalog-to-digital coverter (ADC), ad a high-impedace, high iput impedace ad high commo mode rejectio ratio. ADC), which has a built-i programmable gai amplifier (Programmable Gai Amplifier, PGA), adjustable PGA sigal ca easily determie the magificatio. The system has strog ati-iterferece ability ad is suitable for use i various complicated eviromets. The iput gai ad referece bias ca be flexibly cofigured via the MCU. The microprocessor adopts STM32F405R as the mai cotrol chip, which has high performace, low power cosumptio, fast computig power i various medical ad hadheld devices are widely used, the use of SPI commuicatio protocol ca be coveiet ad flexible wireless trasmissio chip ad ADS1299 Of the register cofiguratio, whe receivig the ADS1299 real-time data uploaded to the host computer. To meet the eeds of the collector fuctio, the collector of the ew storage fuctio, the data ca be stored for real-time lie of data aalysis. Wireless trasmissio module is produced by TI's CC3200, it is TI's idustry's first built-i WIFI MCU, small size, low power cosumptio, its support for b / g / protocol stadards ad the iteral itegratio of TCP / IP protocol stac ca be real-time two-way commuicatio with the host computer, the collected EEG sigal to the host computer for real-time processig feedbac. Through the test, the collector has very low distortio ad iput oise. The mai parameters are show i Table I. TABLE I. COLLECTOR PARAMETERS C. Waveform Display Iterface Parameter Frequecy respose Iput oise Commo mode iput impedace Commo mode rejectio ratio Precisio Samplig Rate The system waveform display iterface is also prepared by the LabVIEW software. As with the visual stimulator Parameter Measured value 0.3~70Hz(-3dB) <1.0uVp-p <100MΩ >110dB 24bit 256SPS~1KSPS software eviromet described above, the display iterface is show i Figure 5,EEG waveform data ca be displayed 8 chaels. DOI /IJSSST.a ISSN: x olie, prit

H s N 1 1 s p c (1) I the formula, is the cut-off frequecy, N is c order, p is the ormalized pole, 1 2 1 j 2 2N p e (2) Figure 5.EEG Sigal Acquisitio PC. III. SSVEP FEATURE EXTRACTION ALGORITHM A.

This paper uses the Butterworth filter ba to filter the received EEG sigal.

4 H s N 1 1 s p c (1) I the formula, is the cut-off frequecy, N is c order, p is the ormalized pole, j 2 2N p e (2) Figure 5.EEG Sigal Acquisitio PC. III. SSVEP FEATURE EXTRACTION ALGORITHM A. Sigal Preprocessig Due to evirometal oise ad their ow iterferece (such as eye, EMG), the sigal must be pre-processed to elimiate the iterferece sigal. This paper uses the Butterworth filter ba to filter the received EEG sigal. The filter has a maximum flat amplitude respose i the passbad, ad its frequecy respose model is Selectig the appropriate cutoff frequecy ad order of the filter to filter out uwated EEG sigal. The frequecy of EEG processed i this paper is maily 10Hz. I cojuctio SSVEP huma brai occipital evoed EEG frequecy. The stimulus frequecy was 4Hz, 8.57Hz, 15Hz, 20Hz. But i the actual process mixed with other frequecy iterferece. I order to esure the existece of harmoics, the pass bad is set to 40 Hz,which will ot oly remove uwated sigals but also ca guaratee the presece of harmoics. Filter effect compariso chart is show i Fig. 6. B. Sigal Feature Extractio ad Processig As a commoly used digital sigal aalysis ad processig methods, wavelet pacet algorithm ca decompose the sigal accordig to frequecy bad. Accordig to the characteristics of the sigal ad aalysis requiremets to match the best correspodig frequecy bad ad sigal spectrum, the recostructio formula of wavelet pacet decompositio is Figure 6.Before ad after filterig the sigal. j/2 j f t dj, 2 u 2 t j, SZ (3) I the formula, j decompose the umber of layers for u the wavelet pacet, is: DOI /IJSSST.a ISSN: x olie, prit

5 u2l t hul t u2l 1 t gul t (4) I the formula, h, g is a scalig fuctio ( x) ad wavelet fuctios ( x) correspodig to the filter fuctio. Wavelet pacet algorithm will recostructed EEG sigal ito SSVEP specific frequecy poit, for the frequecy poit selected i this paper, the umber of decompositio layers is selected accordig to the samplig rate, for example 256SPSSamplig Rate, Selectio of sigal wavelet pacet decompositio layer 6,recostructs the SSVEP specific frequecy sigal, ad the do correlatio with the referece sigal CCA algorithm. The Caoical Correlatio Aalysis (CCA) algorithm for simple ad efficiet computatio of correlatios i SSVEP sigals ca be combied with multi-chael characteristics to meet the 8-chael sigal acquisitio used i this paper. I additio, CCA ca sigificatly improve the sigal-to-oise ratio durig sigal processig. CCA is a algorithm to study the correlatio betwee multiple variables ad multiple variables. The basic idea is to fid the correlatio coefficiet ρ of two sets of variables X ad Y o the whole Ad to maximize it. The multi-chael SSVEP sigal is defied as, x f 1 x f2 X... x f m (5) I the formula, is the chael umber, ad x f is m the f bad sigal recostructed by the origial SSVEP m EEG sigal through wavelet pacet decompositio. The correspodig referece sigal matrix is established as: si 2 f1 si 2 f2 Y... si 2 fm (6) The the maximum correlatio coefficiet is obtaied. max corr X, Y I the formula, cov, Y, ad cov X, Y X Y (7) X Y is the covariace of X ad ad X are the stadard deviatios of Y ad Y respectively. Because the CCA algorithm is based o the recogitio of the SSVEP frequecy, the recogitio effect is iflueced by the iterferece sigal i the SSVEP. I this paper, the characteristics of CCA algorithm ad wavelet pacet decompositio of time-frequecy characteristics, proposed WT-CCA algorithm for the feature extractio ad processig of EEG sigals. The WT-CCA algorithm proposed i this paper ca decompose the low-frequecy ad high-frequecy parts. Ad get the same frequecy bad legth, it ca get a higher resolutio to avoid iterferece i the referece sigal, effectively improve the sigal recogitio rate. The process is show i Figure 7 below. X Wavelet pacet decompositio EEG sigal based o SSVEP EEG sigal of each frequecy bad Calculate correlatio with CCA Referece sigal Figure 7. WT-CCA idetificatio method DOI /IJSSST.a ISSN: x olie, prit

6 The SSVEP sigal is decomposed by wavelet pacet to get the recostructed sigal. The maximum correlatio coefficiet of X ad Y of each chael is calculated accordig to the CCA algorithm,ad the compariso to judge the frequecy of SSVEP stimulatio sigal f m. IV. SYSTEM TEST The test selected five metally well-tested voluteers, the test frequecy is set to 5Hz. I order to esure the quality of EEG sigals, the metal state of the collected persos is good, the eyesight is ormal, the scalp is refreshed, the body ad mid are i the relaxed state, ad the iterferece of other factors is reduced(such as eye, EMG ad so o).collectig the bacgroud light should ot be too bright, the time should ot be too log, so as ot to cause the test subject's discomfort. Durig the preparatio phase, the 8-chael sigal collector is attached to the head of the perso to be tested ad properly adjusted to esure good cotact. The distace betwee the eyes ad the LCD is about 60cm, i the impedace test to debug each chael (the geeral impedace is ot higher tha 1000Ω).SSVEP evoed potetial occurred maily i the huma brai occipital, accordig to the "10-20 Iteratioal Stadard Lead", will collect the electrodes of the EEG sigal placed i the Fp1, Fp2,F3,F4,Cz,Pz,O1,O2 positio. Referece electrode placed i the body behid the earlobe, groudig electrode Fz ad the Fzz middle positio. Figure 8. Shows The Distributio Of Lead Durig the experimet, the acquisitio iterval was about 3 miutes. After all the data is completed, the system will automatically record the collected data ad upload to the host computer for data feature extractio ad classificatio. Table II shows the recogitio rate of WT-CCA algorithm to target frequecy characteristics. We ca see that the recogitio rate of this system ad algorithm to SSVEP is very high, reachig 90%.The same test subjects to differet frequecies of the target stimuli have differet resposes, ad for the same stimulus frequecy, differet testers have differet resposes, which also reflects the differeces betwee idividuals, ad the biological body of differet Cosistet view. Test umber TABLE II. WT-CCA TARGET FREQUENCY FEATURE RECOGNITION ACCURACY RATE Target frequecy 4Hz 8.75Hz 15Hz 20Hz % 92.4% 90.2% 88.5% % 90.1% 92.1% 91.8% % 91.6% 91.7% 89.7% % 90.3% 90.7% 90.3% % 89.7% 94.3% 92.4% Mea value 91.5% 90.8% 91.8% 90.5% The i the same experimetal eviromet, the CCA algorithm is compared with the WT-CCA algorithm proposed by the system. As show i Table III, for the idetificatio of SSVEP sigals, the recogitio rate of the WT-CCA algorithm is sigificatly higher tha the CCA algorithm. TABLE III. algorithm COMPARISON OF THE CORRECT RATE OF TARGET FREQUENCY FEATURE RECOGNITION Target frequecy 4Hz 8.75Hz 15Hz 20Hz mea value CCA 79.5% 81.3% 85.3% 84.3% 82.4% WT-CCA 91.5% 90.8% 91.8% 90.5% 91.1% DOI /IJSSST.a ISSN: x olie, prit

7 V. CONCLUSIONS The mai goal of this research is to realize the desig of the wireless brai computer iterface system based o SSVEP. A. The desig of LCD visual stimulator based o LabVIEW, accordig to the characteristics of 60Hz display set frequecy, through software cotrol stimulus source umber ad frequecy, which is a characteristic differet from other stimulator. B. The idepedet research ad developmet of EEG sigal acquisitio system successfully collected 8 chaels of EEG sigals. C. The proposed WT-CCA algorithm ca feature extractio ad classificatio of EEG sigals.after provig, the WT-CCA algorithm is proved to be superior to CCA algorithm i recogitio ad classificatio, ad the recogitio rate is over 90%. Acowledgmets Project supported by the Natural Sciece Foudatio of the Jiagsu Higher Educatio Istitutios of Chia (15KJB310016); Project supported by the Prospective Joit Research of Sciece ad Techology of Jiagsu (BY ). REFERENCE [1] Hwag H J, et al. Developmet of a SSVEP-based BCI spellig system adoptig a QWERTY-style LED eyboard[j]. Joural of eurosciece methods, 2012, 208(1): [2] Hwag H J, et al. A ew dual-frequecy stimulatio method to icrease the umber of visual stimuli for multi-class SSVEP-based brai computer iterface (BCI)[J]. brai research, 2013, 1515: [3] Tello R, et al. Performace improvemets for avigatio of a robotic wheelchair based o SSVEP-BCI[M]//XII SBAI SimpósioBrasileiro de AutomaçaoIteligete [4] Yi E, et al. A dyamically optimized SSVEP brai computer iterface (BCI) speller[j]. IEEE Trasactios o Biomedical Egieerig, 2015, 62(6): doi: /TBME [5] Feradez-Vargas J, et al. Assisted closed-loop optimizatio of SSVEP-BCI efficiecy[j]. Closig the Loop Aroud Neural Systems, 2014: 391. [6] Shyu K K, et al. Adaptive SSVEP-based BCI system with frequecy ad pulse duty-cycle stimuli tuig desig[j]. IEEE Trasactios o Neural Systems ad Rehabilitatio Egieerig, 2013, 21(5): [7] Hsu H T, et al. Evaluate the Feasibility of Usig Frotal SSVEP to Implemet a SSVEP-Based BCI i Youg, Elderly ad ALS Groups[J]. IEEE Trasactios o Neural Systems ad Rehabilitatio Egieerig, 2016, 24(5): [8] Ng K.B., et al. Effect of competig stimuli o SSVEP-based BCI[J]. IEEE Trasactios o Egieerig i Medicie ad Biology Society, 2011, 8(2): [9] Sutter E.E.. The brai respose iterface: commuicatio through visually-iduced electrical brai resposes[j]. Joural of Microcomputer Applicatios, 1992, 15(1): [10] Muller-PutzG.R., et al. Steady-state somatosesory evoed potetials: suitable brai sigals for brai-computer iterfaces[j]. IEEE Trasactios o Neural Systems ad Rehabilitatio Egieerig, 2006, 14(1): [11] DayagZ, et al. Brai-Robot Iteractio System with Portable Brai Sigal Collector[J]. Joural of Electroic Measuremet ad Istrumetatio, 2016,30(5): (i Chiese). [12] Lopez M A, et al. Sigal Processig ad Perceptros i a Auditory Based Brai-Computer Iterface[C]//Hybrid Itelliget Systems, HIS'08. Eighth Iteratioal Coferece o. IEEE, 2008: [13] Muller S M T, et al. Robotic wheelchair commaded by people with disabilities usig low/high-frequecy ssvep-based BCI[C]//World Cogress o Medical Physics ad Biomedical Egieerig, Jue 7-12, 2015, Toroto, Caada. Spriger Iteratioal Publishig, 2015: DOI /IJSSST.a ISSN: x olie, prit

Tehrani N Journal of Scientific and Engineering Research, 2018, 5(7):1-7

Tehrani N Journal of Scientific and Engineering Research, 2018, 5(7):1-7 Available olie www.jsaer.com, 2018, 5(7):1-7 Research Article ISSN: 2394-2630 CODEN(USA): JSERBR 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38