Weighted averaging of steady-state responses

Objective: To compare weighted averaging and artifact-rejection to normal averaging in the detection of steady-state responses. Methods: Multiple steady-state responses were evoked by auditory stimuli modulated at rates between 78 and 95 Hz. The responses were evaluated after recording periods of 3, 6 and 10 min, using 5 averaging protocols: (1) normal averaging; (2) sample-weighted averaging; (3) noise-weighted averaging; (4) amplitude-based artifact-rejection; and (5) percentage-based artifact rejection. The responses were analyzed in the frequency domain and the signal-to-noise ratio was estimated by comparing the signals at the modulation-frequencies to the noise at adjacent frequencies. Results: Weighted averaging gave the best signal-to-noise ratios. Artifact-rejection was better than normal averaging but not as good as weighted averaging. Responses that were not significant with normal averaging became significant with weighted averaging much more frequently than vice versa. False alarm rates did not significantly differ among the protocols. The advantage of weighted averaging was especially evident when stimuli were presented at lower intensities or when smaller amounts (e.g. only 3 or 6 min) of data were evaluated. Weighted averaging was most effective when the background noise levels were variable. Weighted averaging underestimated the amplitude of the responses by about 2%. Conclusion: Weighted averaging should be used instead of normal averaging for detecting steady-state responses