Information Technology Reference
In-Depth Information
also been proposed including amplifier designs [50, 117, 52], analog-to-digital con-
version (A/D) [51], and voltage multiplier designs [113].
13.8.1 Closed-Loop Cardiac Pacemaker Technology
Some research in cardiac pacemaker technology has sought to modify stimulus
parameters in response to measured neural activity. Moreover, this notion of au-
tonomous regulation is similar in principal to adaptive, autonomous, or closed-loop
deep brain stimulation (DBS).
The current standard for signal processing in cardiac pacemaking still consists
of a simple band-pass filter with adaptive threshold detection [6, 103, 128]. How-
ever, new methods have been proposed that also include nonlinear filtering, wavelet
analysis, and linear regression as well as threshold detection [86,128,42]. For exam-
ple, Rodrigues et al. [128] implement filter banks (wavelets) with linear regression
and threshold techniques in an IC design for detecting “R-waves” in cardiograms.
In particular, given an input waveform x
(
)
n
and wavelet filter H , the output of the
wavelet decomposition is
T H
(
)=
(
)
.
(13.10)
y
n
x
n
Next, the “decision signal” is computed as
T H
H T H
) 1 H T x
T
(
n
)=
x
(
n
)
(
(
n
) .
(13.11)
Finally, the detection of the R-wave is considered positive if for some
β >
0
and maximum decision signal T max , T
(
n
) β
T max . Furthermore, complexity of the
algorithm is O
, while the circuit design reported in [128] requires 6 multiplica-
tions and 45 summations per iteration and achieves a performance of roughly 99%
correct detection and less than 1% false alarm.
(
N
)
13.8.2 Brain-to-Computer Interface
The first reported brain-to-computer interface (BCI) employing an adaptive algo-
rithm and feedback was reported by Tzanakou et al. [150, 105, 38] where pixels on
a screen were modified by the ALOPEX algorithm [62] to excite particular neurons
(or receptive fields) in the visual pathway of a frog brain. Recently, BCI methods
have been reported for detecting intended movements of primates. These include
linear methods such as the “population vector” algorithm [48], finite impulse re-
sponse (FIR) filters [132], Kalman filtering [161], nonlinear methods such as neural
networks (NN) including time-delay NN's (TDNN) [155], gamma models [49] and
recurrent NN's [132], and probabilistic approaches such as Bayesian inference [46].
Moreover, the nonlinear methods tend to achieve more accurate results at the ex-
pense of computational complexity.
Search WWH ::




Custom Search