Estimation of spectral components of auditory steady-state response using least squares and phase compensation for objective response detectors

Auditory Steady-State Responses (ASSR) are evoked potentials useful for estimating hearing thresholds. The ASSR are manifested in the electroencephalogram (EEG) and their presence is verified by statistical methods, usually in the frequency domain, with techniques called Objective Response Detectors (ORD). The standard method of implementing ORD involves dividing the EEG signal into epochs and estimating the spectral components of each epoch at the frequencies of stimulation using the discrete Fourier transform (DFT). One of the disadvantages of estimating spectral components by DFT is that it is necessary to respect the coherent sampling criterion, which reduces considerably the possible choices of epoch length and frequencies of stimulation to apply ORD techniques. This work proposes the least squares method with phase compensation as an alternative technique to DFT. In this work, the ORD technique used was the magnitude-squared coherence (MSC). Our results showed that the highest detection rates occurred slightly above the frequencies of stimulation, which indicates a small calibration error in the dataset. The second analysis was to verify how the performance of the MSC behaves when varying the epoch length. This analysis allowed us to verify that, in real data, the performance of the MSC worsens when using either very small or very large epoch lengths. Thus, in conclusion, the method proposed in this work allowed performing analysis with an ORD technique varying the epoch length and the analyzed frequencies, which was not possible using the DFT.

Automated summary

Auditory Steady-State Responses (ASSR) are evoked potentials used for estimating hearing thresholds. This study proposes an alternative technique to the discrete Fourier transform (DFT) called the least squares method with phase compensation. Results showed a small calibration error in the dataset and demonstrated the performance degradation of the magnitude-squared coherence (MSC) when using either very small or very large epoch lengths in real data. The proposed method allows for analysis with varying epoch lengths and frequencies, which was not possible with DFT.