Understanding the effect of window length and overlap for assessing sEMG in dynamic fatiguing contractions: a non-linear dimensionality reduction and clustering

Research output: Contribution to journalArticlepeer-review


  • Carlos De la Fuente
  • Jose Ignacio Priego-Quesada
  • Alejandro Weinstein
  • Oscar Valencia
  • Marcos Kunzler
  • Alvarez-Ruf Joel
  • Felipe Carpes

Colleges, School and Institutes

External organisations

  • University of Valencia
  • Universidad de Valparaiso
  • Universidad de Los Andes
  • Universidad Metropolitana de Ciencias de la Educacion
  • Pontificia Universidad Católica de Chile
  • Universidade Federal do Pampa
  • Clinica MEDS, Santiago, Chile
  • Universidad del Desarrollo


The Short-Time Fourier transform (STFT) is a helpful tool to identify muscle fatigue with clinical and sports applications. However, the choice of STFT parameters may affect the estimation of myoelectrical manifestations of fatigue. Here, we determine the effect of window length and overlap selections on the frequency slope and the coefficient of variation from EMG spectrum features in fatiguing contractions. We also determine whether STFT parameters affect the relationship between frequency slopes and task failure. Eighty-eight healthy adult men performed one-leg heel-rise until exhaustion. A factorial design with a window length of 50, 100, 250, 500, and 1000 ms with 0, 25, 50, 75, and 90% of overlap was used. The frequency slope was non-linearly fitted as a task failure function, followed by a dimensionality reduction and clustering analysis. The STFT parameters elicited five patterns. A small window length produced a higher slope frequency for the peak frequency (p < 0.001). The contrary was found for the mean and median frequency (p < 0.001). A larger window length elicited a higher slope frequency for the mean and peak frequencies. The largest frequency slope and dispersion was found for a window length of 50 ms without overlap using peak frequency. A combination of 250 ms with 50% of overlap reduced the dispersion both for peak, median, and mean frequency, but decreased the slope frequency. Therefore, the selection of STFT parameters during dynamic contractions should be accompanied by a mechanical measure of the task failure, and its parameters should be adjusted according to the experiment’s requirements.

Bibliographic note

Funding Information: CD was supported by the De Luca Foundation and Delsys Inc. through Delsys’ donation initiative 2020. FPC is supported by a CNPq research fellowship. AW was supported by grant BASAL FB0008. OV was supported by Fondo de Ayuda a la Investigación, Universidad de los Andes, Santiago, Chile (FAI: INV-IN-2017-01). Publisher Copyright: © 2021 Elsevier Ltd


Original languageEnglish
Article number110598
Number of pages7
JournalJournal of Biomechanics
Early online date29 Jun 2021
Publication statusPublished - 26 Aug 2021


  • Electromyography, Fatigue, Fourier, Gastrocnemius medialis, Methods, Muscle activation