Biomedical Engineering Reference
In-Depth Information
an action is completed. BCIs require subject training through biofeedback, and they display a low
bandwidth for effective communication (15-25 bits/min) [
77
], which hinders the speed at which
tasks can be accomplished. However, because of their noninvasiveness, BCIs have already been
tested with success in paraplegics and locked-in patients. Several groups all over the world have
demonstrated working versions of EEG-based interfaces [
77
], and a system software standard has
been proposed. There is a biannual BCI meeting and even a BCI competition to evaluate the prog-
ress of algorithms in this area. A special issue of the
IEEE Transactions on Biomedical Engineering
also surveyed the state of the art in its June 2004 issue.
Trajectory control BMIs use brain activity to control directly the path of an external actuator
in three-dimensional space, mimicking the role of the motor system when it controls, for instance,
the trajectory of a hand movement. The control of trajectories is a common problem encountered
by engineers (aircraft stability, gun control, etc.); therefore, feedback controllers in the area of physi-
ological motor control are common. An example block diagram of a simplified motor system is
given in Figure
1.7
[
78
]. The technical control problem here, at least in a first approach, does not
lend itself to classify brain signals, but requires the BMI to generate a trajectory directly from brain
activity. Here, the motor command generator could be the neural representation of the dynamics
of the controlled object, whereas the motor command is the series of signals sent through the
motor neurons, brainstem, spinal cord, and muscle fibers. In real-world situations, the controllers
are subject to perturbations (loads, noise, moving objects) that can introduce errors in the original
commands. To overcome these discrepancies, sensory systems can be used to modify the motor
program. The two sensory modalities described here, visual and proprioceptive, operate on very
different time scales. Nevertheless, because timing is critical for internal adaptation of this system,
it has been difficult so far to extract from the EEG the motor command signatures with sufficient
spatiotemporal resolution. Therefore, invasive techniques utilizing directly neuronal firings or LFPs
have been utilized. In the last decade, the introduction of new methods for recording and analyz-
Noise
Desired
State
Observed
State
Motor
Command
Motor
Command
Generator
Controlled
Object
Sensory System
Visual
Proprioceptive
FIgURE 1.7:
Elements of a feedback control system.
