A QR decomposition based RLS algorithm with forgetting factor for adaptation of AR EEG features

Electroencephalography (EEG) is an effective modality being used to develop brain computer interface (BCI) systems, but the EEG signal is predominantly marked with low signal-to-noise ratio. Thus, extracting information of interest from EEG data becomes quite a challenging task. Brain activity is usually modeled as an adaptive autoregressive (AAR) model and various adaptive algorithms are implemented to obtain its parameters. Here, QR decomposition based recursive least squares (QR-RLS) algorithm with forgetting factor is derived for AAR parameter estimation and realized to track the EEG features with numerical stability. EEG features from motor imagery datasets are extracted by the proposed method and classified with Linear Discriminate Analysis. For further validation, a comparative analysis is obtained between the presented methodology and already existing EEG processing methods, i.e., Least Mean Squares, RLS and conventional QR-RLS. Results indicate that EEG features extracted from the proposed algorithm provide better classification accuracy than other adaptive algorithms. Thus, promising towards development of reliable and proficient brain computer interface systems.