Foetal ECG extraction using non‐linear adaptive noise canceller with multiple primary channels

Abstract

A non-linear multi-sensory adaptive noise canceller (ANC, MsANC) with both multi-primary and multi-reference channels is proposed for foetal electrocardiogram (FECG) extraction. The primary channels are connected by a linear combiner (LC) whose output serves as a primary signal for the whole MsANC. A finite impulse response (FIR) filter or a non-linear filter [a Volterra filter, or an FIR filter plus a functional link artificial neural network (FLANN), or an FIR filter plus a generalised FLANN] is placed in each reference channel to approximate the linear and non-linear mappings between the chest maternal ECG (ANC reference signal) and the abdominal ECG (ANC primary signal). The LC connecting the primary channels is updated by a constrained recursive least square algorithm, while the linear and non-linear filters placed in reference channels are updated in a least mean square sense. Two real datasets derived from the Physionet non-invasive FECG database as well as the DaISy database are used to show the effectiveness of the proposed MsANC. Experimental results have revealed that the proposed MsANC results in considerable performance improvement as the number of primary channels is increased in comparison with existing ANCs with a single primary channel.