Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
11. METHODS
The general idea is to make the projection of the signal
x(t) on one or several elementary functions vi(t) which
depend on one or several parameters. In this way we obtain
a set of coefficients ci(t) which are then compared to
reference parameters relative to the expected events.
Decompositionof the signal is obtained using :
Y2i(t)=-'Y(-)
2'
2'
x(t) = C(ci(t)xvi(t))
(synthesis)
i=O
Wavelets Decomposition
Wavelets analysis is obtain in this way :
t
ci(t)= x*y~2i(t)=jx(T).v2i(t-T).dT
0
111. SIMULATIONS
The previous methods have been applied to a set of test
signals composed of a number n of events corrupted by an
additive gausian noise. To compare the results given by
these methods we consider the following criteria : miss
detection, false alarm rate and robustness to model
deviations. The elementary event is constructed according
to a model of K complexesproposed by Bremer [6] :
12
Matched Filter
1--
tl
Parameters ai, bi, ri, Ti, i = l..nof the event Ei, i = l..n
are obtained according to :
C;(t) = x*pi(Ti - t) = ~ x ( T pi
) . (T + t - Ti) .dT
where * denotes the convolution product and Ti is the
duration of the event pi (t) to be detected.
a, =(aref --)+6.rvli
2
~
1,
1218
ri = (rref - -)
2
; b, =(b,f --)+6.rv2;
50
I
:Wavelets
;
20
8 lo
2
.-
__-
0.1
---0.2
- _ _ _ _ a
0.3
0.4
*_-- -
-.--
0.6
0.7
e-*
0.5
0.8
0.9
1 . 5 1 .
-1.5'
0
:-:
TN?al:
200
400
600
800
*++*++
+++++----f+++*+++-+-+-+~=~~~~~~~~-++
25
.,
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
.I
UM
400
600
8M)
1'333
12M
1400
IMI
1830 2wO
Nutnbero f points
r* I C
I
*--
0.5
1.0
1.5
2.0
2.5
3.0
- - : Orthonormal Functions
:Wavelets
0.5
* -
1.0
15
2.0
Standard variation of addited noise
2.5
3.0
V. CONCLUSION
The proposed methods are very efficient for the
detection of events buried in noise, even when the SNR
(Signal to Noise Ratio) is very small. But when the events
are deformed, very few false alarms are encountered. In
detection problems, robustness to model deviations is
generally an advantage, but for an application like Kcomplex detection in sleep EEG signals, these methods are
unable to distinguish events which the patterns are very
close. For this reason these methods are unsuited in Slow
Wave Sleep (S.W.S.) where appear delta waves but are
very efficient in sleep stages 1 or 2.
REFERENCES
[ 13 H. BERGER, "Uber Elektroenkephalogram des
menschen," Arch. Psychiat. Neverkunden, vol. 87, pp.
527-570,1929.
121 D. SAUTER, "Recognition of K-complex in sleep
EEG using a fuzzy c-means algorithm", 15th Ann. Int.
Conf. IEEE Eng. Med. & Biol. Soc., San Diego, 1993.
[3] M. BARKAT, Signal Detection and Estimation. Artech
House, New York, 1991.
[4] S. MALLAT, "Singularity Detection and Processing
with Wavelets," IEEE Trans. Information Theory, vol. 38,
p ~617-643,1992.
.
[5] S. MALLAT, "Characterization of Signals from
Multiscale Edges," IEEE Trans. on Pattern Analysis and
Machine Int., vol. 14, pp. 710-732, 1992.
161 G. BREMER, "Automatic detection of the K-complex
in sleep EEG," IEEE Trans. Biomed. Eng., vol. 17, pp.
314-323,1970.
1219