Quantitative representation of electromyographic patterns generated during human locomotion.

The problem addressed is the representation of electromyographic (EMG) profiles in an economical manner so that two tasks can be undertaken: (1) discovery of the different types of patterns in a population using pattern analysis techniques and (2) integration of the information in the EMG profiles with kinematic and dynamic information. The three signal processing techniques used to extract representative features of the EMG profiles are summarized. These techniques are based on the Tauberian approximation, Fourier series, and Karhunen-Loeve expansion and were chosen because they emphasize some characteristic of the gait EMG. All are designed to reduce the dimensionality of the linear envelope of the EMG so that interpretability can be enhanced and subsequent quantitative analyses can be performed.