1 introduction
Epilepsy’s diagnosis main dependence clinical medical history, the electroencephalogram inspection may take one extremely valuable auxiliary diagnosis method. Statistics indicated that 80% about epilepsy patients have the definite brain electricity to be unusual, but has 5~20% about epilepsy patient electroencephalogram performance to be normal. Especially to the clinical diagnosis difficult non-typical epileptic paroxysm, each kind of heterogeneous epilepsy and concealment epilepsy, the electroencephalogram inspection’s importance is more prominent, even is playing decisive role [1].
Brain electricity (EEG) is ultra Gauss or the Asian Gauss signal, usually includes the noise, the false mark and the crosstalk. Usually, the brain electricity activity is divided as a whole 4 frequency band ingredients (Beta, Alpha, Theta and Delta and so on rhythms), these ingredient’s frequency is very low (in 0.5~40 Hz scope). But the clinical analysis indicated when epilepsy patient morbidity as sees take 3 Hz thorn slow synthesis wave. In other words, in the brain electricity the meaningful ingredient basically is the low-frequency signal. This means that we may will mix the repeat through the wavelet decomposition after the brain electricity high-frequency component filtration restructure again, thus filtration noise and false mark. Through the research epilepsy patient’s brain electrical signal, is helpful the decision which in the medicine choice, the dosage adjustment and the medicine stops using, is helpful treats case’s designation in the surgical operation, is helpful distinction which gets sick in epilepsy and other paroxysmic illness.
This article selects based on TI Corporation’s TMS320C54X the series DSP chip develops the platform. With the aid of the DSP fast data processing’s merit, carries on the wavelet transformation to epilepsy brain electrical signal, then filters the small criterion (high frequency) the ingredient, retains the great size (low frequency) the ingredient, finally after processing the signal carries on the reconstruction again. Realizes flow as shown in Figure 1.
2 separate wavelet transformation algorithm
A separate wavelet transformation’s unprecedented achievement was S.Mallat in 1989 fast algorithm 11 Mallat which proposed in the multi-resolution analysis’s foundation algorithm [2]. The Mallat algorithm is equal in the wavelet analysis’s function to fast Fournier transformation (FFT) in the Fourier’s analysis function, he symbolizes that the wavelet analysis has stepped onto the broad application domain. The Mallat algorithm is called the tower system algorithm, he carries on by wavelet filter H, G and h, g to the signal decomposes and restructures [3]. The decomposition algorithm is:
In the formula, t is the discrete time series number, t=1,2,…, N; f(t) is the primary signal; j is the layer or the wavelet criterion, j=1,2,…, J, J=log2N; H, G is in the time domain wavelet decomposition filter, in fact is the filter coefficient; Aj is signal f(t) approaches the part in jth (i.e. low frequency ingredient) the wavelet coefficient; Dj is signal f(t) in the jth detail part (i.e. high-frequency unit) wavelet coefficient.
Type (1) meaning is: Supposes discrete signal f(t) which examines is A. Signal, signal f(t) in the 2j criterion (jth) the approximate part, namely low frequency part’s wavelet coefficient Aj is through the 2j-1 criterion (the j-1 level) approaches the part wavelet coefficient Aj-1 and the filter H convolution, will then convolute the result separates a sampling to obtain; But signal f(t) in the 2j criterion (jth) the detail part, namely high-frequency unit’s wavelet coefficient Dj is through the 2j-1 criterion (the j-1 level) compels to resemble the part the wavelet coefficient and the decomposition filter G convolution, will then convolute the result separates a sampling to obtain.
Through type (1) decomposition, (or on jth) signal f(t) decomposes on each criterion 2j into approximate part wavelet coefficient Aj (on low frequency innertube) and detail part wavelet coefficient D, (on high frequency innertube).
The restructuring algorithm is:
In the formula, j is the decomposition layer, if decomposes the topmost story is the decomposition depth is J, then j=J-1, J-2,…,1,0; h, g is in the time domain wavelet restructuring filter, in fact is the filter coefficient.
Type (2) meaning is: Signal f(t) in the 2j criterion (jth) the approximate part’s wavelet coefficient, namely low frequency part’s wavelet coefficient Aj is after the 2j 1 criterion (the j 1 level) approaches the part wavelet coefficient Aj 1 to separate the spot inserts zero after the restructuring filter h convolution as well as the 2j 1 criterion (the j 1 level) detail part’s wavelet coefficient Dj 1 separates the spot inserts zero with the restructuring filter g convolution, then the summation obtains. Is unceasingly redundant this process, until the 2° criterion, obtains the restructuring signal.
3 wavelet transformation’s DSP realizes
3.1 brain electrical signals in CCS 2.2 on input-output
CCS 2.2 (Code Composer Studio) is one kind which promotes by TI Corporation in view of standard TMS320 debugging connection integrated development environment (IDE), uses the CCS integrated development environment, the user may complete work link [4] and so on project definition, procedure edition, translation link, debugging and data analysis. We carry on decimal base’s floating number with two sexadecimal numbers the expression, uses the C language to realize.
Inducts again using in CCS File->Load Data hexadecimal system’s data to the DSP corresponding memory.
In turn, after the DSP processing data derives File->Save using the CCS data by the text document form preservation, uses the C language to carry on the data counter transformation again, carries on two sexadecimal numbers transforms decimal base’s floating number.
In which result array is decimal base’s floating point, the origin array is hexadecimal system’s floating point.
3.2 core assembly program introduction
The following is realizes 32 floating point multiplication part assembly program by 16 fixed-point multiplications:
3.3 experimental results and analysis
Figure 2(a) is treats the processing brain electrical signal, after wavelet decomposition (b)~(f) is in turn all levels of approaches the oscillogram, (g)~(k) in turn is the corresponding detail oscillogram. Filters out j=3 the detail profile is Figure 2(i), carries on wavelet restructuring, after again obtains Figure 2(1), discovered that after the original map 2(a) and restructuring chart 2(1) cannot see the obvious difference nearly.
4 conclusions
Carries on the wavelet decomposition using the wavelet transformation’s Mallat algorithm to epilepsy patient’s brain electrical signal, retains the brain electricity the source signaling message, the high frequency noise filtration, favors further analyzes [5]. This article has used the DSP fast data processing merit, uses performance-to-price ratio high fixed point TMS320C54x DSP to carry on floating-point data processing, finally indicated that the processing method is feasible, the effect is obvious, in the article introduced the method has certain theory and the practical application value.
