Home Pages of Ted Clancy

Quick links:

If interested, read more detailed descriptions below and/or see a list of possible research projects.

Wearable Wireless Medical & Health Sensors

    We have several emerging projects in which we are designing wireless medical/health sensors that are wearable. In one project, we are designing a small prototype wireless electromyogram (EMG) sensor. While such a sensor can be used generally, we are tailoring this sensor specifically for use as the control input to myoelectric prostheses—initially hand and hand-wrist prostheses. We have other emerging project interests in other biosignals related to physiological monitoring and various assistive devices.
    These wearable sensors are an excellent fit to the field of Electrical and Computer Engineering (ECE). Typically, the front end of the wireless node requires analog design to interface and transduce the physiological signal, the signal is then acquired and processed on a low-power microcomputer (requiring fundamental knowledge in computer engineering and signal processing), the signal is transmitted wirelessly to a base station (or "The Cloud"), and then more advanced signal processing/machine learning is applied to interpret the data. The skills required for development of these sensors span most of the field of ECE. Dependending on the specific project, of course, there are opportunities for work that spans these various sub-disciplinary areas or for work that "dives deep" into any one area.

Electromyogram (EMG) Amplitude Estimation

    The electromyogram, or EMG, is the electrical activity produced by skeletal muscles during contraction.  When individual motor units (motor units are the smallest functional grouping of muscle fibers) contract, they repetitively emit a short burst of electrical activity known as a motor unit action potential.  The time between successive bursts is somewhat random for each motor unit.  When several motor units are active (the timing of the electrical burst between distinct motor units is mostly uncorrelated), a random interference pattern of electrical activity results.  Observed at the skin surface by conventional bipolar electrodes, the interference pattern can be modeled as a zero-mean stochastic process.  To modulate muscle tension, the number of active motor units is modulated and/or the average firing rate of active motor units is modulated.  In either case, the standard deviation of the interference pattern is altered (it is increased by an increasing number of active motor units and/or an increase in the firing rate of individual motor units).  Thus, the standard deviation of the interference pattern, generally referred to as the EMG amplitude (EMGamp), is a measure of muscular activation level.
    Because EMG amplitude is, by definition, a parameter of a random process, its value can only be estimated from a sample of an EMG.  The work that collaborators and I have been involved in is in the development of improved techniques for estimating EMG amplitude.  This work has involved the iterative process of (1) using existing theory and experimental results to develop an improved stochastic mathematic model of EMG, (2) based on the improved model, determine mathematically optimal methods for estimating EMG amplitude (as well as determining an objective quality measure for the new methods), (3) implementing and experimentally testing the new estimation technique, and (4) adding these results to the existing theory.
    In general, EMG amplitude estimation from an EMG sample requires six processing steps: (1) noise rejection/filtering, (2) whitening, (3) multiple-channel combination, (4) detection, (5) smoothing, and (6) relinearization.  A more complete description of these steps, as well as a review of the present state-of-the-art can be found within the EMG Amplitude Estimation toolbox.  Development of EMG amplitude estimation techniques is a continuing research interest.

Relating Surface EMG Amplitude to Joint Torque

    EMG is essentially a by-product of muscle contraction.  The development of muscular tension is the main purpose of contraction.  Hence, it is logical to try to relate the electrical activity of muscle to its mechanical activity.  In general, as the number of active motor units is increased and/or the average firing rate of active motor units is increased, both EMG amplitude and total muscle tension increases. However, the relationship is dynamic and, depending on the resolution desired, may also need to be treated as non-linear.
    It would be ideal, therefore, to relate surface EMG amplitude to the tension (or, force) produced by individual muscles.  However, there are at least two fundamental difficulties in doing so.  First, classical EMG recorded at the surface of the skin can contain "cross-talk".  That is, EMG from muscles other than that which the experimenter intends to record may be included in the signal.  Cross-talk is a difficult problem with no immediate, simple solution.  Second, relating EMG to individual muscle tension requires independent verification via direct mechanical measurement of individual muscle tension.  At present, there is no practical (and perhaps not any) method for reliably making such measurements.
    Because of these fundamental limitations, colleagues and I have focused our efforts on relating surface EMG amplitude to joint torque.  First, cross-talk, if it exists, may not be as problematic.  Certain cross-talk contributions are automatically removed from the estimated torque, even if they can not be removed from the individual muscle tension contributions.  Second, net torque about a joint can, in many cases, be reasonably (or even easily) verified via direct mechanical measurement.
    To date, we have examined the EMG-torque relationship in two simple cases for constant-posture, nonfatiguing contractions about the elbow.  In the first case, slowly force-varying contractions (essentially no dynamics) were examined and the EMG-torque relationship modeled as a polynomial relationship.  A degree three polynomial (non-linear model) was found to best describe the relationship.  Also, advanced EMG amplitude estimation algorithms were compared to classical algorithms.  The advanced algorithms gave lower EMG-torque errors.  In our second study case, force-varying contractions were recorded.  Linear, dynamic ("black box") system identification techniques were used to relate EMG amplitude to torque.  Again, results indicated that advanced EMG amplitude estimation leads to lower EMG-torque errors.  Relating EMG amplitude to joint torque is a continuing research interest.

Relating Surface EMG Amplitude to Joint Impedance

    Joint impedance refers to the relationship between perturbations in joint angle to perturbations in joint torque.  That is, if joint posture is disturbed by applying an external torque about the joint, what will be the resulting change in joint angle (ignoring adjustments due to feedback)?  Typically, one can decrease the influence of perturbations by voluntarily increasing impedance via co-activation of antagonists muscles acting about the joint.  Increased impedance is beneficial in many circumstances, as it serves to reject inevitable external disturbances.  However, there is a metabolic cost to maintaining co-activation, and co-activation increases the overall workload on the muscles.  In many work-related tasks, constant/frequency muscle activation may be related to cummulative stress disorders.  Thus, we are interested in understanding what mechanical impedance is generated by various individuals as they perform tasks.
    Typically, impedance is measured by applying a transient force, and then measuring the resulting angle perturbation.  In general, joint impedance is highly non-linear, depending on the type of force transient applied and the muscle conditions (e.g., joint angle, level of background muscle activation, fatigue state of the muscle).  However, for small random perturbations applied at one (non-fatigued) muscle condition, the impedance relationship is well represented as a second-order linear system.  Thus, impedance can be characterized by its stiffness, viscosity and inertia (which are a function of the "operating point," or muscle condition).  For accurate measurement of impedance, the joint must be perturbed.
    In many cases (e.g., ergonomic analysis), an approximation of the joint impedance would be acceptable, but only if the impedance could be estimated without perturbing the joint.  In this manner, a worker's impedance could be measure while performing a task — and not interrupted by the perturbations.  Here, again, surface EMG amplitude may be useful.  Since modulation of impedance is achieved by modulating muscular effort, we should be able to develop an EMG-impedance relationship in much the same way that prior researchers have developed an EMG-torque relationship.  At a given operating point, we can measure the agonist and antagonist EMG amplitude, and the corresponding joint impedance.  We can then relate the EMG amplitudes to joint impedance.  We are currently involving in preliminary work to investigate this relation for conditions which satisfy the linear second-order model assumptions.

High Resolution EMG Arrays

    As noted above, a typical surface EMG signal records the interference pattern derived from many contributing motor units.  If arrays of small-diameter (≤ 2mm), tightly-spaced (≤ 8mm) EMG electrodes are used to measure the EMG signal, then weighted combinations of the recorded signals produce a spatially filtered EMG channel that can have sufficient resolution to identify individual motor unit action potentials.  There is active research into identifying optimal spatial filters, electrode array configurations and methods for using the arrays signals to decompose the EMG signal.

Other EMG-Related Interests

Depending on the opportunity, I'm interested in many other areas of EMG-related research, including:

Other Biosignal Interests

Other areas of past (and perhaps future) research include:

Sample Size Selection and Train-Test Set Usage in System Identification

    In order to develop parameterized models (e.g., EMG-torque models), data are required both to train (fit) a model and then separately to test the model.  In many cases, the sample sizes required for training and testing sets are not easily determined, particularly when iterative model development occurs.  In addition, some model development schemes encourage the mixing of training and testing sets.  I am interested in investigating model development strategies from collected data sets.

Links to Collaborating Researchers (Past and Present)

Collaborator Collaboration Topics
Paolo Bonato (Spaulding Hosp/Harvard) EMG signal processing, rehabilitation, prosthetics
Richard J. Cohen (MIT) ECG signal processing (electrical-mechanical alternans)
Kristin Farry (Intelligenta) EMG signal processing, myoelectric control of prosthesis
Gary Kamen (UMass-Amherst) MU decomposition and interpretation
Neville Hogan (MIT) EMG signal processing, myoelectric control of prosthesis
Kevin McGill (Palo Alto VA) EMG decomposition, EMGlab
Roberto Merletti (Politecnico di Torino) EMG signal processing, instrumentation
Denis Rancourt (U. of Sherbrooke) EMG signal processing, EMG-torque/impedance modeling
Maintained by ted@wpi.edu
[L(SP)**2]      [WPI]