Adaptive FIR filter use for signal noise cancelling

Noise reduction, together with signal modelling, classification and prediction, forms a basic problem in many engineering, biomedical and econometric systems. The paper summarizes basic adaptive methods based upon the application of artificial neural networks simplified in many cases to an adaptive linear element only. Adaptive noise cancelling algorithms by such finite impulse response (FIR) filters are studied in the first part of the paper both for the case that a reference signal is known and for reference-free structures. Resulting signals are compared with those obtained by noise rejection in the wavelet transform domain either based upon signal decomposition and reconstruction using properly chosen threshold levels for signal modification or incorporating adaptive FIR filtering. Signals preprocessed in this way are then used for their segmentation and classification to verify the effect of signal denoising. Methods presented in the paper are used for simulated signals at first and then applied to the processing of real signals representing physiological signals and technological time series.