Atrial Fibrillation Analysis using Bessel Kernel based Time Frequency Distribution Technique

Abstract

We propose a Bessel kernel based time frequency distribution technique for identification and tracking of spectrum of Atrial Fibrillation (AF) in ECG. The algorithm shows a good frequency resolution and a low RMS error even when the noise dominates the signal which is critical for detecting the low amplitude AF activity within the ECG. In comparison with other time frequency distributions, the Bessel kernel reduces cross terms between frequencies in the multi-component ECG signal. Superiority of the Bessel kernel method over the short time Fourier transform (STFT) is demonstrated using a frequency modulated sinusoidal model and using real AF data. At low signal to noise levels the Bessel distribution outperforms the STFT and at an SNR of -5dB the RMS error is reduced from 1.8Hz to 0.8Hz. Also it achieves a frequency resolution of 0.5Hz at an SNR of 0dB which is four times better than that of the STFT.