An adaptive filter for steady-state evoked potentials - Engineering in Medicine and Biology Society, 1993. Proceedings of the 15th Annual International Co

We modified the Adaptive Noise Cancellation procedure of Widrow et al. (11 for use as an adaptive band-pass filter to process Steady-State Evoked Potentials (SSEPs). The adative filter yields an approsimate 10 dB improvement in Signal to Noise Ratio over the Discrete Fourier Transform. Introduction Steady-state evoked potentials (SSEPs) are comprised of narrow-band signals buried in broad-band noise. They are usually processed by some form of narrow-band filtering. Widrow and ccFworkers [l] suggested that the Adaptive Noise Canceller (ANC) could be used as an adaptive narrow-band filter. The A N C is a correlation filter which can be used to separate a signal that is correlated with a known reference from unknown additive noise. If the reference signal is sinusoidal, the filter converges to a narrow, band-pass filter centered on the reference frequency. Adaptive band-pass filter The raw data y(k) consists of the SSEP, combined with EEG and other noise sources. The SSEP is itself comprised of a series of response components at exact integer multiples of the stimulator frequency. Because the frequencies of the harmonics are known, one can select appropriate reference signals. For the second harmonic, the two reference signals are selected as where k = 0,1,2, ... N , fo is the stimulus frequency and the f, is the sampling rate. The noise canceller produces estimates of the weights h,(C) and h, (k ) by minimizing the exponentially weighted squared estimation error E,, using the RLS algorithm [2] where e ( k ) = y(k) G ( k ) = y(k) ( h , ( k ) * rl(k) Ih,(k) * r z ( k ) ) . The forgetting factor X is positive and less than one. The performance indes En emphasizes the most recent observations and exponentially ignores the older ones. The amplitude and phase of the SSEP (second harmonic) are estimated as