An adaptive filter for distorted line-frequency noise

Line-frequency noise often accompanies biomedical signals from the points where they are sensed. Since this noise often lies in the frequency range of the signal of interest (ie: in a diagnostic-bandwidth ECG) noise removal using conventional frequency filtering techniques can remove important signal components. Adaptive filters based upon prediction of the noise waveform and its subsequent subtraction offer an attractive alternative to conventional techniques. Such filters may assume that the noise is sinusoidal or they may learn the actual shape of the noise waveform. We have designed an adaptive filter which learns and tracks the shape of the noise waveform and its amplitude. The filter borrows techniques from time-ensemble averaging but requires no synchronous trigger. It adapts to slow changes in noise amplitude, harmonic content, and phase (small frequency errors) by sampling the signal at a rate corresponding to an integer multiple of the noise frequency and passing the samples through a two-dimensional, non-recursive waveform filter. The filter has been implemented as a quick and compact assembly-language program of the NSC800 microprocessor.