Digital filter with confidence input
Axel Heim, Martin Hoch · 2015
An extension to standard digital finite input response low-pass filters is proposed, allowing full suppression of noise peaks at the expense of gradually decreasing broadband noise-suppression performance. Each input sample is associated with a confidence value, where zero-confidence samples must not contribute to the filter output. Sample confidence is assumed to be known by other means, e.g., a peak-noise detector. Filter coefficients are dynamically adapted, depending on the confidence values in the filter buffer, gradually turning the filter into a selective arithmetic mean filter with an increasing number of low-confidence inputs, while preserving the filter's optimum broadband noise-suppression performance when input samples have full confidence.