Adaptive Non-Stationary Cardiac Signals Identification using an Augmented MLP Network

Adaptive or recursive learning technique using neural-network as the black-model has been a subject of interest for more than a decade. In this paper hybrid form recursive training algorithms, which combines both linear and
nonlinear orientation of weights, is being used to model or identify ElectroCardioGraphy (ECG) signals. Modeling
or representing a signal of a system (in this case biomedical system), can be viewed as a filtering technique which intends to reduce white noise or uncorrelated noise, where modeled data is used for proceeding tasks such as
feature extractions and classification techniques.
Recursive hybrid training techniques is the choice to consider when learning a non-stationary system which
changes for a given prescribed range. It will be also an ideal case when dealing with ECG signals where the pattern
of signals varies as it depends on the condition of patience at very short frame of time.In this paper the recursive learning algorithms is being tested on an Augmented a Multilayer- Perceptron (MLP) or also known as Direct-Link MLP (DMLP) networks. Variants of recursive hybrid neural learning is being applied on Direct Link MLP (DMLP) network to identify ECG signals and their performance when compared to different MLP network structure.