Frequency Domain Analysis of ECG Signals using Auto-Associative Neural Networks

In this paper, we utilize the frequency domain representation of electrocardiogram (ECG) signals for the training of auto-associative neural networks. Since ECG signals, when taken over shorter duration, are almost periodic; so they can be considered to be short term stationary. We use the harmonics of individual ECG beats, as input to train the auto-associative neural network (AANN). For the purpose of analysis, a five-layer AANN model is used to capture the feature vector distribution. We provide the training results of AANNs for ECG beats from two different ECG databases. Similar experiments are performed using discrete fourier transform (DFT) coefficients of ECG beats, instead of harmonics. The results and relevant observations are reported and discussed.