Effects of input frequency content and signal-to-noise ratio on the parametric estimation of surface EMG-torque dynamics

The dynamic relationship between surface EMG (sEMG) and torque can be estimated from data acquired while subjects voluntarily modulate joint torque. We have shown that for such data, the input (EMG) contains a feedback component from the output (torque) and so accurate estimates of the dynamics require the use of closed-loop identification algorithms. Moreover, this approach has several other limitations since the input is controlled indirectly and so the frequency content and signal-to-noise ratio cannot be controlled. This paper investigates how these factors influence the accuracy of estimates. This was studied using experimental sEMG recorded from healthy human subjects for tasks with different modulation rates. Box-Jenkin (BJ) method was used for identification. Results showed that input frequency content had little effect on estimates of gain and natural frequency but had strong effect on damping factor estimates. It was demonstrated that to accurately estimate the damping factor, the command signal switching rate must be less than 2s. It was also shown that random errors increased with noise level but was limited to 10% of the parameters true value for highest noise level tested. To summarize, simulation study of this work showed that voluntary modulation paradigm can accurately identify sEMG-torque dynamics.