Detection of Respiratory Sinus Arrhythmia by Using Adaptive Filter

A method for detecting respiratory sinus arrhythmia (RSA) by using an adaptive filter is proposed. RSA is a correlative component to breath which is comprised in a heart rate fluctuation. Since the spectrum of the breath changes with respiratory rate, methods that assume statistical stationarity, e.g. FFT, cannot detect RSA precisely under free breath. So we have adopted an adaptive filter which has two inputs. One is a primary input to which an interval between successive R-wave in ECG is applied. The other is a reference input to which a respiratory signal, such as chest movement or respiratory flow rate, is applied. This filter extracts correlated component to the reference signal in the primary input. A remarkable merit of an adaptive filter is a tracking capability, such as the characteristics are followed to input signal, even if power spectrum of the signal changes slowly with time. In this paper, a least mean square algorithm is used to control the filter because of its simplicity. There are two parameters to be decided for the adaptive filter. One is dimension n, and the other is convergent parameter µ. These parameters must be optimized, because they affect to tracking speed and error. A small µ causes slow tracking characteristics with small error. A large µ brings fast tracking speed with large error. As a result of simulation, we can conclude that the optimal µl is obtained by 1/(‘power of reference input’ x70x w). To test the filter characteristics, change of RSA during exercise of bicycle ergometer and that with posture characteristics were measured under free breath. The results of experiments have shown that the proposed method is effective for the estimation of RSA.