Identification of Constant-Posture EMG–Torque Relationship About the Elbow Using Nonlinear Dynamic Models

The surface electromyogram (EMG) from biceps and triceps muscles of 33 subjects was related to elbow torque, contrasting EMG amplitude (EMGσ) estimation processors, linear/nonlinear model structures, and system identification techniques. Torque estimation was improved by 1) advanced EMGσ processors (i.e., whitened, multiple-channel signals); 2) longer duration training sets (52 s versus 26 s); and 3) determination of model parameters via pseudoinverse and ridge regression methods. Dynamic, nonlinear parametric models that included second- or third-degree polynomial functions of EMGσ outperformed linear models and Hammerstein/Weiner models. A minimum error of 4.65 ± 3.6% maximum voluntary contraction (MVC) flexion was attained using a third-degree polynomial, 28th-order dynamic model, with model parameters determined using the pseudoinverse method with tolerance 5.6 × 10-3 on 52 s of four-channel whitened EMG data. Similar performance (4.67 ± 3.7% MVC flexion error) was realized using a second-degree, 18th-order ridge regression model with ridge parameter 50.1.