CN104007823B - Frequency phase hybrid decoding brain-computer interface method based on point-pass filtering and frequency phase hybrid decoding brain-computer interface device based on point-pass filtering - Google Patents

Abstract

The invention discloses a frequency phase hybrid decoding brain-computer interface method based on point-pass filtering and a frequency phase hybrid decoding brain-computer interface device based on point-pass filtering, and relates to the field of digital signal processing. The method comprises the following steps that a coefficient g(n) is used for filtering SSVEP (Steady-State Visual Evoked Potential) signals to obtain y(n); the point-pass filtering output signals y(n) are subjected to FFT (Fast Fourier Transformation) to obtain a final spectral analysis result Y(k); phase values phi1 and phi2 are respectively read from peak spectrums Y(m1) and Y(m2), and in addition, the corrected phase values phi<^>1 and phi<^>2 are respectively and further worked out; f1 and f2 are used as frequency decoded output, and (phi<^>1-phi<^>2)-phic is used as phase decoded output. The device comprises an analog-to-digital converter, a DSP (Digital Signal Processing) device and an output driving and display module, wherein collected signals x(t) are sampled through the analog-to-digital converter to obtain a sample sequence x(n), the sample sequence x(n) enters the DSP device in a parallel digital input form, and through internal processing of the DSP device, parameter estimation of the signals are obtained; then, commands sent by subjects are displayed through the output driving and display module, and finally, external equipment responds the corresponding commands.

Description

Frequency plot hybrid decoding brain-machine interface method based on a pass filter and its device

Technical field

The present invention relates to digital processing field, more particularly, to a kind of frequency plot hybrid decoding based on a pass filter Brain-machine interface method and its device are and in particular to adopt the Steady State Visual Evoked Potential of optional frequency to swash in brain-computer interface device When encouraging source, accurately leach signal and carry out target identification by the controlled point bandpass filter of center frequency point.

Background technology

In order to realize and the exchanging and control of external environment condition^[1], improve the quality of living, brain-computer interface (brain-computer Interface, is abbreviated as bci)^[2]Arise at the historic moment.Brain-computer interface does not rely on the normal output channel of brain, but in human brain A kind of direct information interchange control passage is set up and computer or other electronic equipments between^[3].By extracting EEG signals Feature, and will identify that brain instruction or information transmission give controlled external equipment, finally can complete brain external The direct control of portion's equipment.

Research shows, when outer bound pair brain visual cortex is carried out more than 6hz flicker excitation, by extracting occipital region brain telecommunications Number (eeg) will obtain Steady State Visual Evoked Potential^[4].Due to Steady State Visual Evoked Potential (steady-state visual Evoked potential (ssvep)-based ssvep) there is the non-property invaded, system configuration is simple, the training time is short and its The advantages such as high information transferring rate, are quickly grown based on the brain-computer interface of ssvep in recent years.

Obviously, ssvep signal contains abundant pathology, physiology and psychographic information, is medical diagnosis, brain mind and recognizes The important tool of scientific research activity such as know.But reality EEG signals with electrode collection at brain scalp are very faint (only micro- Volt level), also mainly it is subject to two kinds of interference: one kind is multiple artefact compositions (as Hz noise, eye electricity and Muscle artifacts etc.), and this is done Disturb the extraction being highly detrimental to pure EEG signals and analysis^[5,6]；Another kind is when experimenter watches certain excitation block notice attentively not During concentration, because indicator screen assumes the excitation block of multiple different frequencies or phase place simultaneously, these interference excitation blocks are to vision god Through producing larger interference.Therefore, filter any of the above doing for realizing accurately target identification it is necessary to introduce pretreatment measure Disturb.

But most ssvep-bci system of early stage ignores the pretreatment operation to interference.For example document [7,8] is direct Using canonical correlation analysis (cononical correlation analysisi, cca), different ssevp driving frequencies are carried out Target identification, but cca needs extra reference signal just to can achieve；For another example document [9] directly adopts fft method to ssvep Signal working frequency and phase decoding are although noise and useful pumping signal can be distinguished by fft to a certain extent on the frequency axis Come, but the intrinsic spectrum leakage effect of fft can reduce discrimination, lead to follow-up signal to decode not accurate.

Also there is scholar to be studied to removing the noise jamming in EEG signals at present, such as adopt bandstop filter^[10]、 Comb filter^[11], bandpass filter^[12], wavelet de-noising^[13,14]Etc. method.Document [10] is intended to disappear using bandstop filter Except 50hz Hz noise, but when being realized using fir digital filter, because technology is limited, its wave filter only enables with resistance Transmission characteristic (stopband is set to 45 to 55hz), and cannot realize being fine to the trap of 50hz single-frequency point, this is inevitably right Useful brain electricity composition around 50hz causes to damage；Comb filter is applied to brain-computer interface signal and locates in advance by document [11] proposition Reason link, but comb filter equally exists the wide shortcoming of passband, and can only remove limited specific frequency interference it is impossible to Realize removing the interference of optional frequency composition；In document [12], author adopts 4 rank Butterworth band logical iir wave filters to remove Noise jamming, the transmission passband of this wave filter is 29hz～35hz, therefore can remove the interference outside this passband, and retains centre frequency Driving frequency composition for 32hz, it is apparent that, this filter passband is still wide, exists non-linear additionally, due to iir wave filter The inherent shortcoming of phase place, therefore easily cause additive phase distortion it is difficult to ensure phase extraction precision.For document [13,14] Based on the noise-reduction method of wavelet decomposition, because wavelet field noise reduction needs to realize in multi-level decomposition, step is comparatively laborious, and every fraction What solution was observed is still certain frequency band, rather than certain frequency, and ssvep is using single-frequency point excitation, therefore based on small echo fall The ssevp target identification made an uproar still suffers from room for promotion.

Therefore, for improving the performance of brain machine interface system and the accuracy rate of feature recognition, its filter preprocessing process is very Important, based on the defect analysis to existing pretreatment, for ensureing coding information that is pure, quick, extracting ssvep without distortion, So that control command is recognized accurately, filter preprocessing measure should meet as far as possible following some: (1) filter bandwidht is narrower, leaches Signal then cleaner, follow-up decoding precision is then higher；(2) for the ssvep-bcis comprising phase information coding, pre- place Reason must assure that wave filter has linear phase in itself, and otherwise, the phase value recording is then inaccurate, badly influences follow-up solution Code precision；(3) disturb pre- removing measure step of should trying one's best simple.

Content of the invention

The invention provides a kind of frequency plot hybrid decoding brain-machine interface method based on a pass filter and its device, this Invention eliminates has interference produced by frequency deviation because of various noises and excitation, meets high accuracy and identifies ssvep order Demand, described below:

A kind of frequency plot hybrid decoding brain-machine interface method based on a pass filter, the method comprising the steps of:

(1) n point fft is done to the ssvep signal of input, obtain preliminary analysis of spectrum result x (k), k=0,1 ..., n-1, and Fft spectral line at search highest two, records the spectrum peak position k=m of every cluster spectral line respectively₁With k=m₂；

(2) from driving frequency table f set in advance₁, f₂To f_nMiddle select respectively and m₁δ f, m₂Immediate 2 frequencies of δ f Rate is it is assumed that be f₁With f₂, determine parameter lambda₁With λ₂, δ f is frequency resolution；

(3) utilize m₁With λ₁Value, m₂With λ₂Value and convolution window values w_cN (), configured length is that the point of 2n-1 leads to fir wave filter Coefficient g (n)；

(4) with coefficient g (n), ssvep signal is filtered obtaining y (n)；

(5) fft is done to pass filter output signal y (n), obtain final analysis of spectrum result y (k), k=0,1 ..., n- 1；

(6) compose y (m from peak value₁) and y (m₂) on read phase value respectivelyWithAnd after calculating correction further respectively Phase valueWith

(7) with f₁、f₂Decode output as frequency,Export as phase decoding, whereinFor setting in advance Fixed.

Described parameter lambda₁With λ₂For:

${\mathrm{\&lambda;}}_1=\frac{f_1}{\mathrm{\&delta;f}}-m_1,{\mathrm{\&lambda;}}_2=\frac{f_2}{\mathrm{\&delta;f}}-m_2.$

Described coefficient g (n) is:

With

A kind of frequency plot hybrid decoding brain-computer interface device based on a pass filter, comprising: analog-to-digital conversion device, dsp device Part and output driving and its display module.

Signal x (t) collecting is obtained sample sequence x (n) through the sampling of described analog-to-digital conversion device, defeated with Parallel Digital The form entering enters described dsp device, through the inter-process of described dsp device, obtains the parameter Estimation of signal；Relend and help institute State the order that output driving and its display module show that experimenter sends, the corresponding order of last external-device response.

Frequency plot hybrid decoding brain-machine interface method based on a pass filter proposed by the present invention and its device, if application In Practical Project field and clinical medicine domain, following beneficial effect can be produced:

The point bandpass filter Parameter adjustable section of first present invention, disclosure satisfy that the filtering demands of various frequencies；

Known to priori, true frequency is likely located at frequency axis any place, or there is frequency deviation, and this wave filter passes through Setting m, λ determine point bandpass filter parameter；

For multifrequency composition, existing transfer function can be substituted into by arranging multiple m, λ.

By adjusting this two parameters it is possible to rapid configuration goes out the point bandpass filter that any dot frequency is passed through, relax Restriction to frequency, also ensure that the pure property of desired signal.

Second present invention goes jamming countermeasure based on a pass filter, can remove and there is frequency deviation because of various noises and excitation Produced interference, meets the order demand that high accuracy identifies ssvep；

3rd can accurately leach pumping signal in the situation of various frequency shift (FS)s, objectively relax to driving frequency Require, increased target number of blocks.

Brief description

Fig. 1 is the basic composition block diagram of brain machine interface system；

Fig. 2 is the flow chart of the frequency plot hybrid decoding brain-machine interface method based on a pass filter；

Fig. 3 is no frequency deviation time point bandpass filter filter factor and transmission curve；

Fig. 4 is waveform compares figure before and after filtering during no frequency deviation；

Fig. 5 is to there is frequency deviation time point bandpass filter filter factor and transmission curve；

Fig. 6 is to there is waveform compares figure before and after filtering during frequency deviation；

Fig. 7 is two object block excitation display devices；

Fig. 8 is certain experimenter's oz electrode ssvep signal filtering cross-reference figure；

Fig. 9 is the hardware enforcement figure of the frequency plot hybrid decoding brain-computer interface device based on a pass filter；

Figure 10 is the internal processes flow chart of dsp device.

Specific embodiment

For making the object, technical solutions and advantages of the present invention clearer, below embodiment of the present invention is made further Ground describes in detail.

Based on the problems of the prior art, the present invention, with the ssvep-bci device of frequency plot hybrid coding as platform, carries Go out the controlled point filtering approach of center frequency point, by a pass filter is carried out to ssvep signal, leach excitation frequency exactly Signal, effectively filters out other all interference outer of driving frequency composition so that the pumping signal of any frequency deviation all can be used as target frequency Rate, eventually passes and the signal after pre-filtering is done with phase spectrum correction, object block is recognized accurately.It is somebody's turn to do the brain machine based on a pass filter Interface arrangement relaxes the requirement to target excitation frequency, improves target frequency number, is also improved by this pretreatment operation The offline accuracy rate of system, has very high value.

Brain machine interface system shown in Fig. 1 is broadly divided into four parts:

(1) the various Evoked ptential generation excitations comprising different frequency and first phase information (i.e. coding information) are produced by extraneous Signal；

(2) in cortical electrode or scalp electrode, electroneurographic signal is acquired, and carries out multichannel amplification, filtering and a/ D changes, and completes to pre-process；

(3) feature extraction and control command generate: utilize signal transacting and algorithm for pattern recognition, extract Evoked ptential Characteristic information, and classified, decoded and changed, produced the control command corresponding with nervous activity pattern；

(4) external environment condition and equipment are manipulated using the control command producing.

Need to emphasize, pretreatment occupies an important position in above step, is remaining with use only in pre-treatment step While ssvep signal, thoroughly filter various interference, the follow-up feature extraction of guarantee, decoding and control command are more accurate Really.And present invention introduces centre frequency controlled point bandpass filter serve indispensable effect in this link.

101: n point fft is done to the ssvep signal inputting and (makes sampling rate be f_s, then corresponding frequency resolution is δ f =f_s/ n), obtain preliminary analysis of spectrum result x (k), k=0,1 ..., n-1, and search for fft spectral line at highest two, record respectively The spectrum peak position k=m of every cluster spectral line₁With k=m₂；

102: according to driving frequency table f set in advance₁, f₂To f_n, select respectively and m from driving frequency table₁δ f, m₂ Immediate 2 frequencies of δ f are it is assumed that be f₁With f₂(or be f₁With f₃), determine parameter lambda₁With λ₂；

${\mathrm{\&lambda;}}_1=\frac{f_1}{\mathrm{\&delta;f}}-m_1,{\mathrm{\&lambda;}}_2=\frac{f_2}{\mathrm{\&delta;f}}-m_2---\left(1\right)$

103: utilize m₁With λ₁Value, m₂With λ₂Value and convolution window values w set in advance_cN (), configured length is the point of 2n-1 Coefficient g (n) of logical fir wave filter is

$g\left(n\right)=\frac{{2w}_c\left(n\right)}{n}\cos \left[\frac{2\mathrm{\&pi;}}{n}(m_1-{\mathrm{\&lambda;}}_1)n\right]+\frac{{2w}_c\left(n\right)}{n}\cos \left[\frac{2\mathrm{\&pi;}}{n}(m_2-{\mathrm{\&lambda;}}_2)n\right],n\&element;[-n+1,n-1]---\left(2\right)$

104: with point bandpass filter g (n) having configured, ssvep signal is filtered obtaining y (n), completes to eliminate interference Pretreatment；

105: fft is done to pass filter output signal y (n), obtains final analysis of spectrum result y (k), k=0,1 ..., n- 1；

106: compose y (m from peak value₁) and y (m₂) on read phase value respectivelyWithAnd calculate the phase after correction further Place valueWith

107: with f₁、f₂Decode output as frequency,Export (wherein as phase decodingFor being previously set Fixed value).

Need to point out: the point bandpass filter of step 103 design, all can accurately extract that is to say, that working as ssvep to optional frequency Driving frequency exists during frequency deviation and also can remove out-of-band interference on the premise of retaining excitation composition, is convenient to follow-up decoding and mesh Mark identification process.

Experiment

Emulation experiment

In this emulation experiment, burst is { x (n)=a cos (ω₁n+θ₁)+b·cos(ω₂n+θ₂)+w (n), n= 0,1 ..., l-1 }, wherein a=1, θ₁=50 °, b=1, θ₂=80 °, l=1200.W (n) is the additive white noise randomly generating, It is used for simulating the ambient noise in ssvep signal.In this experiment, frequency resolution δ ω=2 π/n, n are the length of frequency vector h Degree, here makes n=64.Lead to characteristic for the deep point that this wave filter is described, carry out from no frequency deviation with terms of having frequency deviation two below Checking.

(1) there is not frequency shift (FS) situation

It is ω in the case of being somebody's turn to do₁、ω₂It is the situation of the integral multiple of frequency resolution δ ω, ω might as well be made₁=2 δ ω, ω₂ =7 δ ω, Fig. 3 (a), Fig. 3 (b) put coefficient and the transfer curve of bandpass filter in the case of sets forth this kind.Fig. 4 (a), Fig. 4 (b) respectively with this bandpass filter filtering before and after signal waveform.

From figure 3, it can be seen that point pass filter transfer curve passes through ω just₁=2 δ ω, ω₂The frequency of=7 δ ω.Meet Desired point leads to transmission characteristic.

As can be seen from Figure 4, with, after the filtering of this bandpass filter, the noise of waveform is substantially suppressed, and signal is more pure. The fft of n=64 point is done respectively to the waveform of Fig. 4 (a), Fig. 4 (b), and at k=2 and k=7, reads phase value, such as table 1 respectively Shown:

Table 1 no frequency deviation when phase place reading value

In table 1, the ideal value of two phase places is respectively 50 ° and 80 ° it can be seen that directly producing from fft result of spectrum analysis Phase error be 46.1212 ° -50 °=- 3.878 °, 88.9883 ° -80 °=8.9883 °, and the pretreatment through the present invention The phase error producing is 50.4629 ° -50 °=0.4629 °, 79.5854 ° -80 °=- 0.4146 °.The filter of this explanation present invention Ripple pretreatment serves noise suppressed effect, surveys phase error and is reduced.There is not frequency deviation because this situation belongs to, thus directly Although fft surveys phase method there is error, less, still can accept.

(2) there is frequency shift (FS) situation

This situation might as well set ω₀=2.3 δ ω, ω₁=6.8 δ ω, using method proposed by the present invention to signal at Reason.Fig. 5 (a), Fig. 5 (b) put coefficient and the transfer curve of bandpass filter in the case of sets forth this kind.Fig. 6 (a), Fig. 6 (b) respectively with this bandpass filter filtering before and after signal waveform.

Can be seen that from Fig. 5 (b), it is possible to find even if there is frequency shift (FS), the point bandpass filter tradition that the present invention is configured is bent Line also can pass through ω just completely₀=2.3 δ ω, ω₁Amplitude at=6.8 δ ω is 1 frequency, and eliminates other bands Outer interference.

As can be seen from Figure 6, even if signal has frequency deviation, after the filtering of this bandpass filter, still can get pure signal, The noise of waveform is equally substantially suppressed.The fft of n=64 point is done respectively to the waveform of Fig. 6 (a), Fig. 6 (b), and in k=2 and At k=7, read phase value respectively, as shown in table 2:

There is phase place reading value during frequency deviation in table 2

By table 2 it is seen that, when there is frequency deviation, due to spectrum leakage and interference noise presence lead to Direct Phase read Can produce very big error (survey phase error and be respectively 110.5895 ° -50 °=60.5895 °, -121.7470 ° -80 °=- 221.7470 °) it is impossible to meet the demand of clinical practice.And the pretreatment through the present invention, phase error then become very little (survey Phase error is respectively 49.0613 ° -50 °=- 0.9387 °, 79.4096 ° -80 °=- 0.5904 °), substantially it is negligible.

Comprehensive above do not have frequency deviation and there are two kinds of conditions of frequency deviation feelings and can be seen that, (1), in terms of removing interference, uses the present invention The wave filter of design is all feasible in terms of removing interference, and can complete the accurate extraction to multifrequency frequency content simultaneously. The center frequency point of the filter transfer curve of two kinds of situations all falls exactly in the frequency position being previously set, and attenuation outside a channel Hurry up.Contrast the waveform of signal before and after filtering it can be seen that waveform all substantially eliminates noise after filtering, and waveform all becomes pure； (2), in terms of phase-measurement accuracy, after the frequency deviation value of present invention wave filter is to phasing, surveys phase error and all survey phase than direct fft Significantly reduce.

Thus, the advantage disturbed and precise phase is extracted can be completed under any offset frequency situation based on the present invention, for It is meant that the requirement to driving frequency can be relaxed significantly so that optional frequency all can be used as excitation for ssvep-bcis Frequency, provides theories integration for increasing incentives target number.

Ssvep surveys and mutually tests

(1) experimental provision

In this ssvep-bci system, survey and mutually test the frequency plot hybrid coding platform built by University of Macao, and Carry out interference using proposed by the present invention pass filter and its spectrum means for correcting to eliminate and phase property extraction.Set by experiment Sampling rate f_s=600hz, one 22 inches of experiment needs, refreshing frequency 120hz, screen resolution are 1680 × 1050 display Device, as shown in Figure 7.

Ssvep excitation is to be obtained by doing frequency dividing to the line scan signals of this display.In experiment, indicator screen is divided into The two incentives target blocks in left and right, each object block is made up of two sub-piecemeals；During coding, give this two sub-piecemeals with different Flicker frequency and phase place, during decoding, differentiate, by detecting the corresponding phase difference of each pair object block, the target that experimenter watches attentively.

In this survey is mutually tested, filtered process and correct after ssvep measurement phase place (be designated as) it is not ssvep's Excitation phase (is designated as).It is true that there is the reaction time of human brain between ssvep excitation and response, this brain response is prolonged The delay phase varying with each individual can accordingly be produced lateFollowing relation is met between this three

Because decoding is according to excitation phaseRather than measurement phase placeIt is thus desirable to further determining that delay phase

Present invention introduces double frequency ssevp incentive program, by obtaining two driving frequencies f₁With f₂Correction after measurement phase Potential differenceTo identify incentives target.This phase difference is

In formula (5), when driving frequency f₁And f₂Very close to when, delay phase difference is negligible, therefore target identification when Then can use measurement phase difference to replace actual phase difference, easily identify target excitation block.But in this ssvep surveys and mutually tests, Two test frequencies f chosen₁And f₂It is respectively 10 frequency dividings and 11 frequency dividings of refreshing frequency 120hz, that is, 12hz is (at the beginning of including two It is mutually 0 ° and 180 ° of excitation) and 10.9hz (two first phases of inclusion are 0 ° of excitation), its corresponding target block driving frequency and mesh Mark block frequency coding table is as shown in table 3

Table 3 half-court mixed excitation phase code parameters of display table

Obviously selected two frequencies of table 3 still suffer from certain intervals.Therefore the delay phase in formula (5) is poor Can not ignore, need to determine by multiple clinical trial.Many experiments find, the at this moment delay phase difference of two driving frequencies (it is designated as) be relatively fixed as 36 °, therefore practical application when need to measurement phase difference on the basis of deduct 36 ° Object block could more accurately be identified.

(2) experimentation and result

Three experimenters (s1, s2, s3) are entered with row energization induce, 13 electrode positions of collection (po3, po5, po7, poz, Po4, po6, po8, p1, o1, oz and o2) the ssvep signal that produces, signal connects usb interface note by Electroencephalo signal amplifier Record, sampling rate is 600hz.This experiment collection EEG signals is divided into 5 wheels, and each wheel comprises 10 collections.Encourage every time Therapy lasted 8 seconds, in whole process, experimenter requires and focuses on, and wherein encouraging first 2 seconds is time, latter 6 seconds Then require experimenter to watch corresponding flicker excitation block as requested attentively, watch order attentively according to ' 1212121212 ' alternately.

In this is surveyed and mutually tests, it is broadly divided into following step:

Step1: ssvep signal is respectively adopted a pass filter device with no filtering with two kinds of different scheme working frequencies with Phase estimation, obtains two groups of surveys and is mutually worth

Step2: ask for the measurement phase difference value estimating in step oneIt is used for replacing actual phase difference

Step3: substitute into (5) and ask for phase difference value

Step4: with the criterion shown in formula (6), the phase difference value that step3 calculates is differentiated, to determine object block p；

In formula (6), m is number of targets, c_kFor preferable cluster centre accordingly.Only need during the classification therefore judging this object block From r₁To r_mIn find out maximum p and (seek r_kValue is closest to k value corresponding when 1), the object block label as being identified.

For the correctness embodying institute of the present invention extracting method directly perceived, true ssvep signal filtering cross-reference figure is given below. , primary signal and filtered signal are respectively as follows: taking the ssvep signal of certain experimenter's oz electrode as a example

By Fig. 8 it is seen that, the ssvep signal before filtering has mixed many interference signals, these be derived from extraneous and itself Interference undoubtedly influence whether follow-up decoding process；And after passing point pass filter, signal pure many, only driving frequency (12hz and 10.9hz) corresponding useful signal is retained, and this pretreatment measure will necessarily improve the degree of accuracy of decoding.

Experiment carries out frequency plot knowledge with two schemes (direct fft method and point filtering approach) to ssvep signal respectively Not and sort out (classification is two classes (excitation block 1 and 2), m=2), table 4 is given and different experimenters is added with different length hanning window The target identification accuracy rate obtaining.

The target identification accuracy rate of table 4 different window length difference experimenter

In table 4, p_fft represents passing point bandpass filter bearing calibration, and fft is direct method.It is seen that, target identification Accuracy rate is not only relevant with window length, also with whether passing point pass filter has much relations, accuracy rate after filtering apparently higher than Directly fft method, Average Accuracy exceeds more than 10%.

Choose a length of 4 seconds of window below, by number of targets m be set to 4 it is assumed that corresponding phase place be respectively 0 °, 90 °, 180 °, 270 °, again ssvep signal is processed with two methods, the accuracy rate result of contrast target identification is as shown in the table

The estimation phase place of the different experimenter of table 5 and feature recognition r_k(means standard deviation)

As shown in Table 5, identification incentives target block only need to find out r₁, r₂, r₃And r₄In closest to 1 value.

(1) for driving frequency, the direct fft for 10.9hz surveys phase situation, i.e. the situation of no any pretreatment, due to swashing Encourage the integral multiple that frequency 10.9hz is not frequency resolution, from table 5 the 2nd rowData it may be clearly seen that to should Can there is very big deviation (preferable first phase value is 0 °, and actual survey is mutually worth deviation maximum close to 90 °) in the survey phase average of frequency.

(2) for driving frequency, the introducing point pass filter for 10.9hz pre-processes the survey phase situation of line phase correction of going forward side by side, Although driving frequency 10.9hz is not the integral multiple of frequency resolution, from table 5 the 2nd rowData it may be clearly seen that To the survey phase average of frequency less deviation should be only existed (preferable first phase value is 0 °, and actual mutually value deviation of surveying does not surpass substantially Cross 30 °).

(3) for 12hz, phase situation is surveyed for driving frequency, because driving frequency 12hz is exactly the integral multiple of resolution ratio, that is, There is not frequency deviation, therefore error comes solely from noise, rather than itself frequency deviation factor, thus from table 5 the 1st rowData It can be seen that the survey phase effect of two schemes is about the same, and (preferable average is 0 ° and 180 °, and surveyed phase place is all equal in this two ideals Value small range distribution nearby).

(4) arrange from table 5 the 1stWith the 2nd rowIt is also seen that the standard deviation of its survey phase data is all attached at 36 ° Nearly small range distribution, demonstrates and sets in advanceThe correctness of brain time delay phase difference.

(5) observe classification results r in table 5_k(k=1,2,3,4, m=4) corresponding p value (in table 5, the often maximum r of row_kValue It is marked with shade, corresponding k value is target p), because in this m=4 imaginary object block, corresponding to the mesh of k=2 The object block of mark block and k=4 assumes that, only corresponds to the object block (corresponding to the object block of j=1) of k=1 and k=3 Object block (corresponding to the object block of j=2) is only possible excitation, therefore p value during correct target identification also should be limited to p=1 With two kinds of situations of p=3, and its error probability is big compared to during m=2.As can be seen from Table 5, phase decoder feelings are surveyed for direct fft Condition, because the excitation for 10.9hz for the frequency has frequency deviation, adds the noise jamming outside band, therefore the knot of almost all of target identification Fruit is all wrong, and p value is all detected as the result of p=2 or p=4, loses clinical reference value.And with present invention introduces Binding site pass filter and the method for phasing, p value is all detected as the correct result of p=1 or p=3, i.e. corresponding maximum detection Decision value r_kIt is meant that accuracy reaches 100% all more than 0.95, therefore compared to direct fft method, the inventive method is more Reliable.

Survey phase study from above-mentioned emulation experiment and actual ssvep all to can be seen that, proposed by the present invention pass filter correction side Method can be in the case of any frequency deviation, and filtering interference signals well obtain clean EEG signals, then by being corrected Quickly and accurately extract the phase information of signal, therefore the technology of the present invention has in the engineering in medicine application of actual brain-computer interface Very wide application prospect.

Referring to Fig. 9, included based on the frequency plot hybrid decoding brain-computer interface device of a pass filter: analog-to-digital conversion device, dsp Device and output driving and its display module, signal x (t) collecting is obtained sample through a/d (analog-to-digital conversion device) sampling Sequence x (n), enters dsp device in the form of Parallel Digital input, processes through the internal algorithm of dsp device, obtains signal Parameter Estimation；Relend and help output driving and its display module to show the order that experimenter sends, last external-device response corresponds to Order.

Wherein, the dsp device of Figure 10 is core devices, during Signal parameter estimation, completes following major function:

(1) call core algorithm, the parameter Estimation completing receipt signal is processed；

(2) signal is carried out correcting after a pass filter, phase estimation result is substituted into discriminate, carries out object block identification And send the corresponding command export in real time to drive and display module.

It may be noted that due to employing digitized method of estimation, thus determine the complexity of Fig. 8 system, real-time levels Principal element with stability is not the periphery connection of dsp device in Fig. 9, but what dsp internal program memory was stored Kernel estimation algorithm.

The internal processes flow process of dsp device is as shown in Figure 10.

" the controlled point pass filter of center frequency point " that proposed this kernel estimation algorithm is implanted dsp device by the present invention Interior, high accuracy, low complex degree, the identification of efficient incentives target are completed based on this.

Figure 10 flow process is divided into several steps as follows:

(1) need first to require (free indoor activity as paralytic etc.) according to concrete application, the sampling of setting signal The points n and number of times i of duplicate measurements, and setting accuracy requires according to specific needs.This step is to propose specifically in terms of engineering Demand, so that follow-up process is targetedly processed.

(2) and then, cpu main controller reads sampled data from i/o port, enters internal ram.

(3) follow-up " DC processing ", is the impact in order to eliminate the flip-flop in measured signal.Otherwise, direct current The presence of composition, can reduce certainty of measurement.Flip-flop is easy to measure it is only necessary to the mean value calculating sampling point can get.

(4) it is filtered being the most crucial part of dsp algorithm by the processing procedure of Fig. 2 present invention, after running this algorithm, that is, Can get clean ssvep signal, on this basis phase place is made an estimate.

(5) judge whether the inventive method meets clinical demand, if being unsatisfactory for, program returns, and again sets as requested Sample frequency carries out next round phase measurement and sorts out identification.

(6) until identification target is correct, correct control command can be sent.Repeat above measurement process i time.

(7) exported to outside display drive device by the output bus of dsp, command instruction is passed to external equipment.As Control the switch of TV and adjust platform, control the wind-speed gear of electric fan, control moving forward and backward of wheelchair.

It may be noted that realizing so that whole parameter estimation operation becomes more flexible due to employing dsp, can be according to signal The concrete condition of the various components being comprised, changes the inner parameter setting of algorithm by flexible in programming, such as sampling number n, adopts Sample rate f_sDeng.

Bibliography

[1] Wu little Pei, reviews and prospects [j] .2012 of Song Junke, Guo Xiaojing, et al. brain-computer interface technology,

[2]wolpaw j r,birbaumer n,mcfarland d j,et al.brain–computer interfaces for communication and control[j].clinical neurophysiology,2002,113 (6):767-91.

[3]cheng m,gao x,gao s,et al.design and implementation of a brain- computer interface with high transfer rates[j].biomedical engineering,ieee transactions on,2002,49(10):1181-6.

[4]middendorf m,mcmillan g,calhoun g,et al.brain-computer interfaces based on the steady-state visual-evoked response[j].ieee transactions on rehabilitation engineering,2000,8(2):211-4.

[5] Xie Songyun, application study [c] in EEG signals denoising for Zhang Zhenzhong, Zhang Weiping, the et al.ica method [j]. Chinese biomedical engineering progress 2007 Chinese biomedical engineering joint Annual Conference collection of thesis (first volume), 2007,

[6]garcia-molina g,zhu d.optimal spatial filtering for the steady state visual evoked potential:bci application；proceedings of the neural engineering(ner),2011 5th international ieee/embs conference on,f,2011[c] .ieee.

[7]bin g,gao x,yan z,et al.an online multi-channel ssvep-based brain– computer interface using a canonical correlation analysis method[j].journal of neural engineering,2009,6(4):046002.

[8]li y,bin g,gao x,et al.analysis of phase coding ssvep based on canonical correlation analysis(cca)；proceedings of the neural engineering (ner),20115th international ieee/embs conference on,f,2011[c].ieee.

[9]jia c,gao x,hong b,et al.frequency and phase mixed coding in ssvep-based brain--computer interface[j].biomedical engineering,ieee transactions on,2011,58(1):200-6.

[10] Zhao Lun .erp experiment textbook [j]. Tianjin publishing house of Academy of Social Sciences, 2004,

[11]materka a,byczuk m.using comb filter to enhance ssvep for bci applications；proceedings of the advances in medical,signal and information processing,2006medsip2006iet 3rd international conference on,f,2006[c].iet.

[12]chang h-c,lee p-l,lo m-t,et al.independence of amplitude- frequency and phase calibrations in an ssvep-based bci using stepping delay flickering sequences[j].neural systems and rehabilitation engineering,ieee transactions on,2012,20(3):305-12.

[13]tu y,huang g,hung y s,et al.single-trial detection of visual evoked potentials by common spatial patterns and wavelet filtering for brain- computer interface；proceedings of the engineering in medicine and biology society(embc),201335th annual international conference of the ieee,f,2013[c] .ieee.

[14] Luo Zhizeng, Cao Ming. EEG signals artefact filtering algorithm [j] based on maximum signal to noise ratio blind source separating. electronics Journal, 2012,39 (12): 2926-31.

To the model of each device in addition to doing specified otherwise, the model of other devices is not limited the embodiment of the present invention, As long as the device of above-mentioned functions can be completed.

It will be appreciated by those skilled in the art that accompanying drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention Sequence number is for illustration only, does not represent the quality of embodiment.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all spirit in the present invention and Within principle, any modification, equivalent substitution and improvement made etc., should be included within the scope of the present invention.

Claims (5)

1. a kind of frequency plot hybrid decoding brain-machine interface method based on a pass filter is it is characterised in that methods described includes Following steps:

(1) n point fft is done to the ssvep signal of input, obtain preliminary analysis of spectrum result x (k), k=0,1 ..., n-1, and search for Fft spectral line at highest two, records the spectrum peak position k=m of every cluster spectral line respectively₁With k=m₂；

(2) from driving frequency table f set in advance₁, f₂To f_nMiddle select respectively and m₁δ f, m₂Immediate 2 frequencies of δ f, false It is set to f₁With f₂, determine parameter lambda₁With λ₂, δ f is frequency resolution；

(3) utilize m₁With λ₁Value, m₂With λ₂Value and convolution window values w_c(n), configured length be 2n-1 point lead to fir wave filter be Number g (n)；

(4) with coefficient g (n), ssvep signal is filtered obtaining y (n)；

(5) fft is done to pass filter output signal y (n), obtain final analysis of spectrum result y (k), k=0,1 ..., n-1；

(6) compose y (m from peak value₁) and y (m₂) on read phase value respectivelyWithAnd calculate the phase place after correction respectively further ValueWith

(7) with f₁、f₂Decode output as frequency,Export as phase decoding, whereinFor presetting.

2. a kind of frequency plot hybrid decoding brain-machine interface method based on a pass filter according to claim 1, it is special Levy and be, described parameter lambda₁With λ₂For:

${\mathrm{\&lambda;}}_1=\frac{f_1}{\mathrm{\&delta;}f}-m_1,{\mathrm{\&lambda;}}_2=\frac{f_2}{\mathrm{\&delta;}f}-m_2.$

3. a kind of frequency plot hybrid decoding brain-machine interface method based on a pass filter according to claim 1, it is special Levy and be, described coefficient g (n) is:

$\begin{array}{c}g\left(n\right)=\frac{2w_c\left(n\right)}{n}\cos \left[\frac{2\mathrm{\&pi;}}{n}(m_1-{\mathrm{\&lambda;}}_1)n\right]+\frac{2w_c\left(n\right)}{n}\cos \left[\frac{2\mathrm{\&pi;}}{n}(m_2-{\mathrm{\&lambda;}}_2)n\right]\\ n\&element;[-n+1,n-1]\end{array}.$

4. a kind of frequency plot hybrid decoding brain-machine interface method based on a pass filter according to claim 1, it is special Levy and be, the phase value after described correctionWithFor

With

5. a kind of for implementing the frequency plot hybrid decoding brain-machine interface method based on a pass filter described in claim 1 Interface arrangement, comprising: analog-to-digital conversion device, dsp device and output driving and its display module it is characterised in that

Signal x (t) collecting is obtained sample sequence x (n) through the sampling of described analog-to-digital conversion device, with Parallel Digital input Form enters described dsp device, through the inter-process of described dsp device, obtains the parameter Estimation of signal；Again by described defeated Go out to drive and its display module shows the order that sends of experimenter, the corresponding order of last external-device response.