A Constant Q Transform based approach for robust EEG spectral analysis

Robust and accurate spectral analysis for electroencephalogram (EEG) signal becomes more and more essential and crucial in psychological and neural studies. The traditional fixed precision Fourier Transform (FT) cannot well reveal the subtle difference and characteristics among individuals. In this paper, a Constant Q Transform (CQT) based approach for EEG spectral analysis was proposed in order to match each person´s features and emphasis the spectrum deviation among different capacitive activities. Firstly, the subject-independent features were extracted based on frequency structure of EEG signal. Then, the adaptive bandwidth was calculated. After that, the CQT based spectrum were acquired using these adaptive frequencies. We experiment with analog data to evaluate the performance of CQT based process and also provide detail comparison in real EEG dataset of eyes closed and eyes open. From the results, two improvements are made based on psychology basis: (a) CQT based unequal-interval spectral analysis method can not only avoid unbalance samples among EEG bands, but also reduce the calculation; (b) the classification ratio has been improved by the CQT-SA algorithm. The CQT based spectral analysis method can become an universal spectral analysis tool for EEG study.