Home Pages of Ted Clancy
POSSIBLE RESEARCH PROJECTS
The following is a sample
of projects related to my research which I would like to see initiated
or continued.
They are listed in no particular order of importance, and are representative
of the type of project work in which I am interested.
Many variations and combinations of these projects are possible.
Please contact me if you are interested in any of the projects or the sort
of work that they represent.
Wearable Wireless Medical & Health Sensors
We are designing a wireless electromyogram
(EMG) sensor for use in myoelectric prosthesis control. The aim is
to make the sensor package as small as possible, but battery-powered
for as long as possible. The work includes some analog design of
the EMG transduction and conditioning electronics; real-time signal processing
on a small, low-powered microprocessor; transmission to a
base station (currently using the Bluetooth protocol); then back-end
processing of the received signal for control of a prosthetic hand
and/or wrist.
Motor Unit Decomposition and Spatial Filtering
When muscles contract, individual groups of
muscle fibers emit repeated, non-periodic bursts of electrical activity.
These electrical bursts travel down the muscle fiber at speeds of 2-6 m/s,
and are known as action potentials. With available electronic recording
apparatus and sampling rates, it is possible to follow an action potential
as it traverses the muscle, by observing from several recording sites along
the path, and/or as it travels past one recording site. At any one
time, many muscle fiber groups (known as motor units) can be simultaneously
active. In order to understand how the body controls muscular activation,
in both health and pathology, it is necessary to decompose the electrical
activity of several motor units into their constituent activations.
This process is known as motor unit decomposition. Most decomposition
is achieved by inserting needle electrodes into the muscle. These
electrodes preferentially record the activity of only a few motor units
in their immediate vicinity, simplifying decomposition. Recently,
arrays of electrodes have been placed on the skin surface, and arranged
to have the necessary spatial selectivity to also see only the activity
of a few dominant motor units.
I am now beginning activity in the development
of such surface recording arrays (and also participate in studies that
utilize the needle recordings). I am interested in developing signal
processing/ pattern recognition algorithms to perform motor unit decomposition.
This work should expand upon existing algorithms available in the literature,
and take advantage of the ongoing advances in computing power.
A related interest is in spatial filtering of the surface array data.
Certain linear combinations of the electrical activity from the individual
array elements are more selective of the electrical signal from a distinct
volume of the muscle. More selective "spatially filtered" recordings
are easier to decompose. Thus, I am also interested in the design of
these spatial filters. This work may include electrophysiologic modeling,
as well as signal processing.
Applications of this work include: clinical
evaluation of neuromuscular function (particularly in children, who
do not well tolerate existing needle examinations); and
scientific/ergonomic studies of muscle function in health and disease.
Electromyogram Amplitude Estimation
The electromyogram (EMG), or electrical activity of skeletal muscle,
can be modeled as a Gaussian random process plus noise. The process is zero
mean with a covariance matrix that is constant except for a scaling factor.
The scaling factor increases with muscular effort. Thus, EMG can
be modeled as a zero-mean, unit-variance, quasi-stationary Gaussian random process
multiplied by an amplitude. The amplitude is time-varying, and thus
may be better described as a modulating factor. EMG amplitude estimation
is the problem of estimating the modulating factor (which is the time-varying
standard deviation of the process) as a function of time (and generally
in the presence of noise).
Improved Adaptive Whitening and Adaptive Noise Attenuation of Electromyograms
Because the sampled EMG are correlated, direct
estimation of the standard deviation by the classical statistics estimation
formula is inappropriate (the classical formula requires that the EMG samples
be independent). Hence, the time sample should be orthogonalized,
or whitened. Recently, an
adaptive whitening method was developed.
This method adapts the whitening filter as a function of the amplitude
by accounting for measurement noise in the recorded EMG. At each
amplitude level, the relative strength of the noise (which is assumed to
be stationary) differs. At low EMG amplitudes, the relative strength
of background noise is large. At high EMG amplitudes, the relative
strength of background noise is small. An adaptive Wiener filter
was developed to attenuate the noise.
While the initial adaptive Weiner filter was
succesful, several improvements to it are in order. This project
will follow-up the previous work and develop the necessary improvements.
These improvements include: determining the maximum frequency out
to which whitening is useful, developing the adaptive Wiener filter as
a general noise attenuator even when the amplitude of the EMG is not needed,
and altering the adaptive Wiener filter to require less calibration and
fewer coefficients (the existing algorithm is very robust, but very inefficient
in its use of filter coefficients).
Generalized Non-Stationary Smoother for EMG Amplitude Estimation
Because EMG can be thought of as a random process, smoothing the amplitude can decrease the noise in the amplitude estimate. However, the decreased noise comes at the cost of increased bias in the estimate. Therefore, optimal estimators have been developed that vary the amount of smoothing depending on the dynamics of the amplitude. When the amplitude is changing rapidly, little smoothing should be used. When the amplitude is changing slowly, much smoothing should be used. In the past, adaptive smoothers which adapt the amount of smoothing on a sample-by-sample basis have been developed. Unfortunately, both simulation and experimental studies indicate that the most common adaptive smoothing strategy, varying the cut-off frequency of a low-pass filter, most often does not improve the amplitude estimate.
In this project, a new concept for non-stationary
processing of the EMG will be considered. Instead of considering
only a one-dimensional problem (i.e., the location of a cut-off frequency),
the complete shape of a linear filter will be studied. In other words,
the original works asked how to optimize the location of the cut-off frequency
of an averaging filter in order to form the best estimate. This project
will attempt to ask how to optimize a complete linear filter in order to
form the best estimate. A more generalized smoother will result.
Applications of this work include:
the control of powered prosthetics; stroke rehabilitation biotherapy;
ergonomic assessment of work activities and performance.
EMG-Torque Modeling, EMG-Impedance Modeling
One application of EMG amplitude is estimation of joint torque.
Typically, EMG from both agonist and antagonist muscles about a joint are
monitored and then related to the net torque about the joint. This
work involves black box modeling (a.k.a. system identification). Thus
far, we have been working with constant-posture, nonfatiguing, (attempted)
flexion-extension about the elbow. We have looked at (quasi-) constant
torque and dynamically varying torque conditions. We would like to
continue these studies and move towards progressively more complex situations,
including: evaluation at multiple angles, evaluation in more
dynamic conditions, more degrees of freedom (e.g., adding supination-pronation
to flexion-extension), more complex joints (e.g., shoulder, wrist), multiple
joints, and more muscles (and muscle sites) monitored.
A related interest is the relationship
between EMG amplitude and the mechanical impedance produced by a limb.
Presently, we are investigating the linearized mechanical impedance
(characterized as a second-order system with stiffness, viscosity and
inertia) about the elbow joint. Our long-term goal is to be able
to use EMG amplitude to estimate joint impedance.
Applications include: assessment
of work activities and performance; musculoskeletal modeling; and
control of powered prostheses.
Interactive, Self-Study Teaching Modules for EMG
Recently, Roberto Merletti and Phil Parker edited a new, graduate-level
anthology of topics related to EMG signal processing. Their text,
titled "Electromyography: Physiology, Engineering, and Non-Invasive
Applications" (Wiley-IEEE Press, 2004) contains 18 chapters of material
from basic electromygraphic physiology to EMG signal processing. The
text is an excellent background for work in my field. (The text
does NOT include exercises.) Unfortunately, this field may be too
narrow in topic for this text to be the basis of a regular, lecture-based
university course.
Thus, I am interested in the development of
an electronic self-study course based around this book. This project
would investigate ways to design such a self-study and/or produce
the self-study for one or more chapters. Any one project would likely
not provide self-study materials for all 18 chapters. However, each
chapter is stand-alone, as each was written by a different set of authors.
This type of self-study would be the basis for myself to train future
students in the field and would be made available to other researchers around
the world.
![[L(SP)**2]](image_style/lsp2.jpg "L(SP)**2 Home Page")
![[WPI]](image_style/wpi_logohome.gif)