Faculty Advisor or Committee Member

Edward A. Clancy, Advisor

Faculty Advisor or Committee Member

Xinming Huang, Committee Member

Faculty Advisor or Committee Member

Paolo Bonato, Committee Member




This dissertation of various aspects of electromyogram (EMG: muscle electrical activity) signal processing is comprised of two projects in which I was the lead investigator and two team projects in which I participated. The first investigator-led project was a study of reconstructing continuous EMG discharge rates from neural impulses. Related methods for calculating neural firing rates in other contexts were adapted and applied to the intramuscular motor unit action potential train firing rate. Statistical results based on simulation and clinical data suggest that performances of spline-based methods are superior to conventional filter-based methods in the absence of decomposition error, but they unacceptably degrade in the presence of even the smallest decomposition errors present in real EMG data, which is typically around 3-5%. Optimal parameters for each method are found, and with normal decomposition error rates, ranks of these methods with their optimal parameters are given. Overall, Hanning filtering and Berger methods exhibit consistent and significant advantages over other methods. In the second investigator-led project, the technique of signal whitening was applied prior to motion classification of upper limb surface EMG signals previously collected from the forearm muscles of intact and amputee subjects. The motions classified consisted of 11 hand and wrist actions pertaining to prosthesis control. Theoretical models and experimental data showed that whitening increased EMG signal bandwidth by 65-75% and the coefficients of variation of temporal features computed from the EMG were reduced. As a result, a consistent classification accuracy improvement of 3-5% was observed for all subjects at small analysis durations (< 100 ms). In the first team-based project, advanced modeling methods of the constant posture EMG-torque relationship about the elbow were studied: whitened and multi-channel EMG signals, training set duration, regularized model parameter estimation and nonlinear models. Combined, these methods reduced error to less than a quarter of standard techniques. In the second team-based project, a study related biceps-triceps surface EMG to elbow torque at seven joint angles during constant-posture contractions. Models accounting for co-contraction estimated that individual flexion muscle torques were much higher than models that did not account for co-contraction.


Worcester Polytechnic Institute

Degree Name



Electrical & Computer Engineering

Project Type


Date Accepted





signal processing, myoelectric, EMG